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
46 static 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
52 static inline int is_ia32_frame(struct ksignal *ksig)
53 {
54 return IS_ENABLED(CONFIG_X86_32) || is_ia32_compat_frame(ksig);
55 }
56
57 static 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 * Set up a signal frame.
65 */
66
67 /* x86 ABI requires 16-byte alignment */
68 #define FRAME_ALIGNMENT 16UL
69
70 #define MAX_FRAME_PADDING (FRAME_ALIGNMENT - 1)
71
72 /*
73 * Determine which stack to use..
74 */
75 void __user *
76 get_sigframe(struct ksignal *ksig, struct pt_regs *regs, size_t frame_size,
77 void __user **fpstate)
78 {
79 struct k_sigaction *ka = &ksig->ka;
80 int ia32_frame = is_ia32_frame(ksig);
81 /* Default to using normal stack */
82 bool nested_altstack = on_sig_stack(regs->sp);
83 bool entering_altstack = false;
84 unsigned long math_size = 0;
85 unsigned long sp = regs->sp;
86 unsigned long buf_fx = 0;
87 u32 pkru = read_pkru();
88
89 /* redzone */
90 if (!ia32_frame)
91 sp -= 128;
92
93 /* This is the X/Open sanctioned signal stack switching. */
94 if (ka->sa.sa_flags & SA_ONSTACK) {
95 /*
96 * This checks nested_altstack via sas_ss_flags(). Sensible
97 * programs use SS_AUTODISARM, which disables that check, and
98 * programs that don't use SS_AUTODISARM get compatible.
99 */
100 if (sas_ss_flags(sp) == 0) {
101 sp = current->sas_ss_sp + current->sas_ss_size;
102 entering_altstack = true;
103 }
104 } else if (ia32_frame &&
105 !nested_altstack &&
106 regs->ss != __USER_DS &&
107 !(ka->sa.sa_flags & SA_RESTORER) &&
108 ka->sa.sa_restorer) {
109 /* This is the legacy signal stack switching. */
110 sp = (unsigned long) ka->sa.sa_restorer;
111 entering_altstack = true;
112 }
113
114 sp = fpu__alloc_mathframe(sp, ia32_frame, &buf_fx, &math_size);
115 *fpstate = (void __user *)sp;
116
117 sp -= frame_size;
118
119 if (ia32_frame)
120 /*
121 * Align the stack pointer according to the i386 ABI,
122 * i.e. so that on function entry ((sp + 4) & 15) == 0.
123 */
124 sp = ((sp + 4) & -FRAME_ALIGNMENT) - 4;
125 else
126 sp = round_down(sp, FRAME_ALIGNMENT) - 8;
127
128 /*
129 * If we are on the alternate signal stack and would overflow it, don't.
130 * Return an always-bogus address instead so we will die with SIGSEGV.
131 */
132 if (unlikely((nested_altstack || entering_altstack) &&
133 !__on_sig_stack(sp))) {
134
135 if (show_unhandled_signals && printk_ratelimit())
136 pr_info("%s[%d] overflowed sigaltstack\n",
137 current->comm, task_pid_nr(current));
138
139 return (void __user *)-1L;
140 }
141
142 /* save i387 and extended state */
143 if (!copy_fpstate_to_sigframe(*fpstate, (void __user *)buf_fx, math_size, pkru))
144 return (void __user *)-1L;
145
146 return (void __user *)sp;
147 }
148
149 /*
150 * There are four different struct types for signal frame: sigframe_ia32,
151 * rt_sigframe_ia32, rt_sigframe_x32, and rt_sigframe. Use the worst case
152 * -- the largest size. It means the size for 64-bit apps is a bit more
153 * than needed, but this keeps the code simple.
154 */
155 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
156 # define MAX_FRAME_SIGINFO_UCTXT_SIZE sizeof(struct sigframe_ia32)
157 #else
158 # define MAX_FRAME_SIGINFO_UCTXT_SIZE sizeof(struct rt_sigframe)
159 #endif
160
161 /*
162 * The FP state frame contains an XSAVE buffer which must be 64-byte aligned.
163 * If a signal frame starts at an unaligned address, extra space is required.
164 * This is the max alignment padding, conservatively.
165 */
166 #define MAX_XSAVE_PADDING 63UL
167
168 /*
169 * The frame data is composed of the following areas and laid out as:
170 *
171 * -------------------------
172 * | alignment padding |
173 * -------------------------
174 * | (f)xsave frame |
175 * -------------------------
176 * | fsave header |
177 * -------------------------
178 * | alignment padding |
179 * -------------------------
180 * | siginfo + ucontext |
181 * -------------------------
182 */
183
184 /* max_frame_size tells userspace the worst case signal stack size. */
185 static unsigned long __ro_after_init max_frame_size;
186 static unsigned int __ro_after_init fpu_default_state_size;
187
188 static int __init init_sigframe_size(void)
189 {
190 fpu_default_state_size = fpu__get_fpstate_size();
191
192 max_frame_size = MAX_FRAME_SIGINFO_UCTXT_SIZE + MAX_FRAME_PADDING;
193
194 max_frame_size += fpu_default_state_size + MAX_XSAVE_PADDING;
195
196 /* Userspace expects an aligned size. */
197 max_frame_size = round_up(max_frame_size, FRAME_ALIGNMENT);
198
199 pr_info("max sigframe size: %lu\n", max_frame_size);
200 return 0;
201 }
202 early_initcall(init_sigframe_size);
203
204 unsigned long get_sigframe_size(void)
205 {
206 return max_frame_size;
207 }
208
209 static int
210 setup_rt_frame(struct ksignal *ksig, struct pt_regs *regs)
211 {
212 /* Perform fixup for the pre-signal frame. */
213 rseq_signal_deliver(ksig, regs);
214
215 /* Set up the stack frame */
216 if (is_ia32_frame(ksig)) {
217 if (ksig->ka.sa.sa_flags & SA_SIGINFO)
218 return ia32_setup_rt_frame(ksig, regs);
219 else
220 return ia32_setup_frame(ksig, regs);
221 } else if (is_x32_frame(ksig)) {
222 return x32_setup_rt_frame(ksig, regs);
223 } else {
224 return x64_setup_rt_frame(ksig, regs);
225 }
226 }
227
228 static void
229 handle_signal(struct ksignal *ksig, struct pt_regs *regs)
230 {
231 bool stepping, failed;
232 struct fpu *fpu = ¤t->thread.fpu;
233
234 if (v8086_mode(regs))
235 save_v86_state((struct kernel_vm86_regs *) regs, VM86_SIGNAL);
236
237 /* Are we from a system call? */
238 if (syscall_get_nr(current, regs) != -1) {
239 /* If so, check system call restarting.. */
240 switch (syscall_get_error(current, regs)) {
241 case -ERESTART_RESTARTBLOCK:
242 case -ERESTARTNOHAND:
243 regs->ax = -EINTR;
244 break;
245
246 case -ERESTARTSYS:
247 if (!(ksig->ka.sa.sa_flags & SA_RESTART)) {
248 regs->ax = -EINTR;
249 break;
250 }
251 fallthrough;
252 case -ERESTARTNOINTR:
253 regs->ax = regs->orig_ax;
254 regs->ip -= 2;
255 break;
256 }
257 }
258
259 /*
260 * If TF is set due to a debugger (TIF_FORCED_TF), clear TF now
261 * so that register information in the sigcontext is correct and
262 * then notify the tracer before entering the signal handler.
263 */
264 stepping = test_thread_flag(TIF_SINGLESTEP);
265 if (stepping)
266 user_disable_single_step(current);
267
268 failed = (setup_rt_frame(ksig, regs) < 0);
269 if (!failed) {
270 /*
271 * Clear the direction flag as per the ABI for function entry.
272 *
273 * Clear RF when entering the signal handler, because
274 * it might disable possible debug exception from the
275 * signal handler.
276 *
277 * Clear TF for the case when it wasn't set by debugger to
278 * avoid the recursive send_sigtrap() in SIGTRAP handler.
279 */
280 regs->flags &= ~(X86_EFLAGS_DF|X86_EFLAGS_RF|X86_EFLAGS_TF);
281 /*
282 * Ensure the signal handler starts with the new fpu state.
283 */
284 fpu__clear_user_states(fpu);
285 }
286 signal_setup_done(failed, ksig, stepping);
287 }
288
289 static inline unsigned long get_nr_restart_syscall(const struct pt_regs *regs)
290 {
291 #ifdef CONFIG_IA32_EMULATION
292 if (current->restart_block.arch_data & TS_COMPAT)
293 return __NR_ia32_restart_syscall;
294 #endif
295 #ifdef CONFIG_X86_X32_ABI
296 return __NR_restart_syscall | (regs->orig_ax & __X32_SYSCALL_BIT);
297 #else
298 return __NR_restart_syscall;
299 #endif
300 }
301
302 /*
303 * Note that 'init' is a special process: it doesn't get signals it doesn't
304 * want to handle. Thus you cannot kill init even with a SIGKILL even by
305 * mistake.
306 */
307 void arch_do_signal_or_restart(struct pt_regs *regs)
308 {
309 struct ksignal ksig;
310
311 if (get_signal(&ksig)) {
312 /* Whee! Actually deliver the signal. */
313 handle_signal(&ksig, regs);
314 return;
315 }
316
317 /* Did we come from a system call? */
318 if (syscall_get_nr(current, regs) != -1) {
319 /* Restart the system call - no handlers present */
320 switch (syscall_get_error(current, regs)) {
321 case -ERESTARTNOHAND:
322 case -ERESTARTSYS:
323 case -ERESTARTNOINTR:
324 regs->ax = regs->orig_ax;
325 regs->ip -= 2;
326 break;
327
328 case -ERESTART_RESTARTBLOCK:
329 regs->ax = get_nr_restart_syscall(regs);
330 regs->ip -= 2;
331 break;
332 }
333 }
334
335 /*
336 * If there's no signal to deliver, we just put the saved sigmask
337 * back.
338 */
339 restore_saved_sigmask();
340 }
341
342 void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
343 {
344 struct task_struct *me = current;
345
346 if (show_unhandled_signals && printk_ratelimit()) {
347 printk("%s"
348 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
349 task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
350 me->comm, me->pid, where, frame,
351 regs->ip, regs->sp, regs->orig_ax);
352 print_vma_addr(KERN_CONT " in ", regs->ip);
353 pr_cont("\n");
354 }
355
356 force_sig(SIGSEGV);
357 }
358
359 #ifdef CONFIG_DYNAMIC_SIGFRAME
360 #ifdef CONFIG_STRICT_SIGALTSTACK_SIZE
361 static bool strict_sigaltstack_size __ro_after_init = true;
362 #else
363 static bool strict_sigaltstack_size __ro_after_init = false;
364 #endif
365
366 static int __init strict_sas_size(char *arg)
367 {
368 return kstrtobool(arg, &strict_sigaltstack_size) == 0;
369 }
370 __setup("strict_sas_size", strict_sas_size);
371
372 /*
373 * MINSIGSTKSZ is 2048 and can't be changed despite the fact that AVX512
374 * exceeds that size already. As such programs might never use the
375 * sigaltstack they just continued to work. While always checking against
376 * the real size would be correct, this might be considered a regression.
377 *
378 * Therefore avoid the sanity check, unless enforced by kernel
379 * configuration or command line option.
380 *
381 * When dynamic FPU features are supported, the check is also enforced when
382 * the task has permissions to use dynamic features. Tasks which have no
383 * permission are checked against the size of the non-dynamic feature set
384 * if strict checking is enabled. This avoids forcing all tasks on the
385 * system to allocate large sigaltstacks even if they are never going
386 * to use a dynamic feature. As this is serialized via sighand::siglock
387 * any permission request for a dynamic feature either happened already
388 * or will see the newly install sigaltstack size in the permission checks.
389 */
390 bool sigaltstack_size_valid(size_t ss_size)
391 {
392 unsigned long fsize = max_frame_size - fpu_default_state_size;
393 u64 mask;
394
395 lockdep_assert_held(¤t->sighand->siglock);
396
397 if (!fpu_state_size_dynamic() && !strict_sigaltstack_size)
398 return true;
399
400 fsize += current->group_leader->thread.fpu.perm.__user_state_size;
401 if (likely(ss_size > fsize))
402 return true;
403
404 if (strict_sigaltstack_size)
405 return ss_size > fsize;
406
407 mask = current->group_leader->thread.fpu.perm.__state_perm;
408 if (mask & XFEATURE_MASK_USER_DYNAMIC)
409 return ss_size > fsize;
410
411 return true;
412 }
413 #endif /* CONFIG_DYNAMIC_SIGFRAME */
414
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.