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