xref: /linux/arch/um/os-Linux/skas/process.c (revision 59376fb2)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
4  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
5  */
6 
7 #include <stdlib.h>
8 #include <stdbool.h>
9 #include <unistd.h>
10 #include <sched.h>
11 #include <errno.h>
12 #include <string.h>
13 #include <sys/mman.h>
14 #include <sys/wait.h>
15 #include <asm/unistd.h>
16 #include <as-layout.h>
17 #include <init.h>
18 #include <kern_util.h>
19 #include <mem.h>
20 #include <os.h>
21 #include <ptrace_user.h>
22 #include <registers.h>
23 #include <skas.h>
24 #include <sysdep/stub.h>
25 #include <linux/threads.h>
26 #include <timetravel.h>
27 #include "../internal.h"
28 
is_skas_winch(int pid,int fd,void * data)29 int is_skas_winch(int pid, int fd, void *data)
30 {
31 	return pid == getpgrp();
32 }
33 
ptrace_reg_name(int idx)34 static const char *ptrace_reg_name(int idx)
35 {
36 #define R(n) case HOST_##n: return #n
37 
38 	switch (idx) {
39 #ifdef __x86_64__
40 	R(BX);
41 	R(CX);
42 	R(DI);
43 	R(SI);
44 	R(DX);
45 	R(BP);
46 	R(AX);
47 	R(R8);
48 	R(R9);
49 	R(R10);
50 	R(R11);
51 	R(R12);
52 	R(R13);
53 	R(R14);
54 	R(R15);
55 	R(ORIG_AX);
56 	R(CS);
57 	R(SS);
58 	R(EFLAGS);
59 #elif defined(__i386__)
60 	R(IP);
61 	R(SP);
62 	R(EFLAGS);
63 	R(AX);
64 	R(BX);
65 	R(CX);
66 	R(DX);
67 	R(SI);
68 	R(DI);
69 	R(BP);
70 	R(CS);
71 	R(SS);
72 	R(DS);
73 	R(FS);
74 	R(ES);
75 	R(GS);
76 	R(ORIG_AX);
77 #endif
78 	}
79 	return "";
80 }
81 
ptrace_dump_regs(int pid)82 static int ptrace_dump_regs(int pid)
83 {
84 	unsigned long regs[MAX_REG_NR];
85 	int i;
86 
87 	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
88 		return -errno;
89 
90 	printk(UM_KERN_ERR "Stub registers -\n");
91 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
92 		const char *regname = ptrace_reg_name(i);
93 
94 		printk(UM_KERN_ERR "\t%s\t(%2d): %lx\n", regname, i, regs[i]);
95 	}
96 
97 	return 0;
98 }
99 
100 /*
101  * Signals that are OK to receive in the stub - we'll just continue it.
102  * SIGWINCH will happen when UML is inside a detached screen.
103  */
104 #define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
105 
106 /* Signals that the stub will finish with - anything else is an error */
107 #define STUB_DONE_MASK (1 << SIGTRAP)
108 
wait_stub_done(int pid)109 void wait_stub_done(int pid)
110 {
111 	int n, status, err;
112 
113 	while (1) {
114 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
115 		if ((n < 0) || !WIFSTOPPED(status))
116 			goto bad_wait;
117 
118 		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
119 			break;
120 
121 		err = ptrace(PTRACE_CONT, pid, 0, 0);
122 		if (err) {
123 			printk(UM_KERN_ERR "%s : continue failed, errno = %d\n",
124 			       __func__, errno);
125 			fatal_sigsegv();
126 		}
127 	}
128 
129 	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
130 		return;
131 
132 bad_wait:
133 	err = ptrace_dump_regs(pid);
134 	if (err)
135 		printk(UM_KERN_ERR "Failed to get registers from stub, errno = %d\n",
136 		       -err);
137 	printk(UM_KERN_ERR "%s : failed to wait for SIGTRAP, pid = %d, n = %d, errno = %d, status = 0x%x\n",
138 	       __func__, pid, n, errno, status);
139 	fatal_sigsegv();
140 }
141 
142 extern unsigned long current_stub_stack(void);
143 
get_skas_faultinfo(int pid,struct faultinfo * fi,unsigned long * aux_fp_regs)144 static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
145 {
146 	int err;
147 
148 	err = get_fp_registers(pid, aux_fp_regs);
149 	if (err < 0) {
150 		printk(UM_KERN_ERR "save_fp_registers returned %d\n",
151 		       err);
152 		fatal_sigsegv();
153 	}
154 	err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
155 	if (err) {
156 		printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
157 		       "errno = %d\n", pid, errno);
158 		fatal_sigsegv();
159 	}
160 	wait_stub_done(pid);
161 
162 	/*
163 	 * faultinfo is prepared by the stub_segv_handler at start of
164 	 * the stub stack page. We just have to copy it.
165 	 */
166 	memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
167 
168 	err = put_fp_registers(pid, aux_fp_regs);
169 	if (err < 0) {
170 		printk(UM_KERN_ERR "put_fp_registers returned %d\n",
171 		       err);
172 		fatal_sigsegv();
173 	}
174 }
175 
handle_segv(int pid,struct uml_pt_regs * regs,unsigned long * aux_fp_regs)176 static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
177 {
178 	get_skas_faultinfo(pid, &regs->faultinfo, aux_fp_regs);
179 	segv(regs->faultinfo, 0, 1, NULL);
180 }
181 
handle_trap(int pid,struct uml_pt_regs * regs)182 static void handle_trap(int pid, struct uml_pt_regs *regs)
183 {
184 	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
185 		fatal_sigsegv();
186 
187 	handle_syscall(regs);
188 }
189 
190 extern char __syscall_stub_start[];
191 
192 /**
193  * userspace_tramp() - userspace trampoline
194  * @stack:	pointer to the new userspace stack page
195  *
196  * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
197  * This function will run on a temporary stack page.
198  * It ptrace()'es itself, then
199  * Two pages are mapped into the userspace address space:
200  * - STUB_CODE (with EXEC), which contains the skas stub code
201  * - STUB_DATA (with R/W), which contains a data page that is used to transfer certain data between the UML userspace process and the UML kernel.
202  * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
203  * And last the process stops itself to give control to the UML kernel for this userspace process.
204  *
205  * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
206  */
userspace_tramp(void * stack)207 static int userspace_tramp(void *stack)
208 {
209 	struct sigaction sa;
210 	void *addr;
211 	int fd;
212 	unsigned long long offset;
213 	unsigned long segv_handler = STUB_CODE +
214 				     (unsigned long) stub_segv_handler -
215 				     (unsigned long) __syscall_stub_start;
216 
217 	ptrace(PTRACE_TRACEME, 0, 0, 0);
218 
219 	signal(SIGTERM, SIG_DFL);
220 	signal(SIGWINCH, SIG_IGN);
221 
222 	fd = phys_mapping(uml_to_phys(__syscall_stub_start), &offset);
223 	addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
224 		      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
225 	if (addr == MAP_FAILED) {
226 		os_info("mapping mmap stub at 0x%lx failed, errno = %d\n",
227 			STUB_CODE, errno);
228 		exit(1);
229 	}
230 
231 	fd = phys_mapping(uml_to_phys(stack), &offset);
232 	addr = mmap((void *) STUB_DATA,
233 		    STUB_DATA_PAGES * UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
234 		    MAP_FIXED | MAP_SHARED, fd, offset);
235 	if (addr == MAP_FAILED) {
236 		os_info("mapping segfault stack at 0x%lx failed, errno = %d\n",
237 			STUB_DATA, errno);
238 		exit(1);
239 	}
240 
241 	set_sigstack((void *) STUB_DATA, STUB_DATA_PAGES * UM_KERN_PAGE_SIZE);
242 	sigemptyset(&sa.sa_mask);
243 	sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
244 	sa.sa_sigaction = (void *) segv_handler;
245 	sa.sa_restorer = NULL;
246 	if (sigaction(SIGSEGV, &sa, NULL) < 0) {
247 		os_info("%s - setting SIGSEGV handler failed - errno = %d\n",
248 			__func__, errno);
249 		exit(1);
250 	}
251 
252 	kill(os_getpid(), SIGSTOP);
253 	return 0;
254 }
255 
256 int userspace_pid[NR_CPUS];
257 
258 /**
259  * start_userspace() - prepare a new userspace process
260  * @stub_stack:	pointer to the stub stack.
261  *
262  * Setups a new temporary stack page that is used while userspace_tramp() runs
263  * Clones the kernel process into a new userspace process, with FDs only.
264  *
265  * Return: When positive: the process id of the new userspace process,
266  *         when negative: an error number.
267  * FIXME: can PIDs become negative?!
268  */
start_userspace(unsigned long stub_stack)269 int start_userspace(unsigned long stub_stack)
270 {
271 	void *stack;
272 	unsigned long sp;
273 	int pid, status, n, flags, err;
274 
275 	/* setup a temporary stack page */
276 	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
277 		     PROT_READ | PROT_WRITE | PROT_EXEC,
278 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
279 	if (stack == MAP_FAILED) {
280 		err = -errno;
281 		printk(UM_KERN_ERR "%s : mmap failed, errno = %d\n",
282 		       __func__, errno);
283 		return err;
284 	}
285 
286 	/* set stack pointer to the end of the stack page, so it can grow downwards */
287 	sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
288 
289 	flags = CLONE_FILES | SIGCHLD;
290 
291 	/* clone into new userspace process */
292 	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
293 	if (pid < 0) {
294 		err = -errno;
295 		printk(UM_KERN_ERR "%s : clone failed, errno = %d\n",
296 		       __func__, errno);
297 		return err;
298 	}
299 
300 	do {
301 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
302 		if (n < 0) {
303 			err = -errno;
304 			printk(UM_KERN_ERR "%s : wait failed, errno = %d\n",
305 			       __func__, errno);
306 			goto out_kill;
307 		}
308 	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
309 
310 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
311 		err = -EINVAL;
312 		printk(UM_KERN_ERR "%s : expected SIGSTOP, got status = %d\n",
313 		       __func__, status);
314 		goto out_kill;
315 	}
316 
317 	if (ptrace(PTRACE_SETOPTIONS, pid, NULL,
318 		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
319 		err = -errno;
320 		printk(UM_KERN_ERR "%s : PTRACE_SETOPTIONS failed, errno = %d\n",
321 		       __func__, errno);
322 		goto out_kill;
323 	}
324 
325 	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
326 		err = -errno;
327 		printk(UM_KERN_ERR "%s : munmap failed, errno = %d\n",
328 		       __func__, errno);
329 		goto out_kill;
330 	}
331 
332 	return pid;
333 
334  out_kill:
335 	os_kill_ptraced_process(pid, 1);
336 	return err;
337 }
338 
userspace(struct uml_pt_regs * regs,unsigned long * aux_fp_regs)339 void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
340 {
341 	int err, status, op, pid = userspace_pid[0];
342 	siginfo_t si;
343 
344 	/* Handle any immediate reschedules or signals */
345 	interrupt_end();
346 
347 	while (1) {
348 		time_travel_print_bc_msg();
349 
350 		current_mm_sync();
351 
352 		/* Flush out any pending syscalls */
353 		err = syscall_stub_flush(current_mm_id());
354 		if (err) {
355 			if (err == -ENOMEM)
356 				report_enomem();
357 
358 			printk(UM_KERN_ERR "%s - Error flushing stub syscalls: %d",
359 				__func__, -err);
360 			fatal_sigsegv();
361 		}
362 
363 		/*
364 		 * This can legitimately fail if the process loads a
365 		 * bogus value into a segment register.  It will
366 		 * segfault and PTRACE_GETREGS will read that value
367 		 * out of the process.  However, PTRACE_SETREGS will
368 		 * fail.  In this case, there is nothing to do but
369 		 * just kill the process.
370 		 */
371 		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
372 			printk(UM_KERN_ERR "%s - ptrace set regs failed, errno = %d\n",
373 			       __func__, errno);
374 			fatal_sigsegv();
375 		}
376 
377 		if (put_fp_registers(pid, regs->fp)) {
378 			printk(UM_KERN_ERR "%s - ptrace set fp regs failed, errno = %d\n",
379 			       __func__, errno);
380 			fatal_sigsegv();
381 		}
382 
383 		if (singlestepping())
384 			op = PTRACE_SYSEMU_SINGLESTEP;
385 		else
386 			op = PTRACE_SYSEMU;
387 
388 		if (ptrace(op, pid, 0, 0)) {
389 			printk(UM_KERN_ERR "%s - ptrace continue failed, op = %d, errno = %d\n",
390 			       __func__, op, errno);
391 			fatal_sigsegv();
392 		}
393 
394 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
395 		if (err < 0) {
396 			printk(UM_KERN_ERR "%s - wait failed, errno = %d\n",
397 			       __func__, errno);
398 			fatal_sigsegv();
399 		}
400 
401 		regs->is_user = 1;
402 		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
403 			printk(UM_KERN_ERR "%s - PTRACE_GETREGS failed, errno = %d\n",
404 			       __func__, errno);
405 			fatal_sigsegv();
406 		}
407 
408 		if (get_fp_registers(pid, regs->fp)) {
409 			printk(UM_KERN_ERR "%s -  get_fp_registers failed, errno = %d\n",
410 			       __func__, errno);
411 			fatal_sigsegv();
412 		}
413 
414 		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
415 
416 		if (WIFSTOPPED(status)) {
417 			int sig = WSTOPSIG(status);
418 
419 			/* These signal handlers need the si argument.
420 			 * The SIGIO and SIGALARM handlers which constitute the
421 			 * majority of invocations, do not use it.
422 			 */
423 			switch (sig) {
424 			case SIGSEGV:
425 			case SIGTRAP:
426 			case SIGILL:
427 			case SIGBUS:
428 			case SIGFPE:
429 			case SIGWINCH:
430 				ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
431 				break;
432 			}
433 
434 			switch (sig) {
435 			case SIGSEGV:
436 				if (PTRACE_FULL_FAULTINFO) {
437 					get_skas_faultinfo(pid,
438 							   &regs->faultinfo, aux_fp_regs);
439 					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
440 							     regs);
441 				}
442 				else handle_segv(pid, regs, aux_fp_regs);
443 				break;
444 			case SIGTRAP + 0x80:
445 				handle_trap(pid, regs);
446 				break;
447 			case SIGTRAP:
448 				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
449 				break;
450 			case SIGALRM:
451 				break;
452 			case SIGIO:
453 			case SIGILL:
454 			case SIGBUS:
455 			case SIGFPE:
456 			case SIGWINCH:
457 				block_signals_trace();
458 				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
459 				unblock_signals_trace();
460 				break;
461 			default:
462 				printk(UM_KERN_ERR "%s - child stopped with signal %d\n",
463 				       __func__, sig);
464 				fatal_sigsegv();
465 			}
466 			pid = userspace_pid[0];
467 			interrupt_end();
468 
469 			/* Avoid -ERESTARTSYS handling in host */
470 			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
471 				PT_SYSCALL_NR(regs->gp) = -1;
472 		}
473 	}
474 }
475 
new_thread(void * stack,jmp_buf * buf,void (* handler)(void))476 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
477 {
478 	(*buf)[0].JB_IP = (unsigned long) handler;
479 	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
480 		sizeof(void *);
481 }
482 
483 #define INIT_JMP_NEW_THREAD 0
484 #define INIT_JMP_CALLBACK 1
485 #define INIT_JMP_HALT 2
486 #define INIT_JMP_REBOOT 3
487 
switch_threads(jmp_buf * me,jmp_buf * you)488 void switch_threads(jmp_buf *me, jmp_buf *you)
489 {
490 	if (UML_SETJMP(me) == 0)
491 		UML_LONGJMP(you, 1);
492 }
493 
494 static jmp_buf initial_jmpbuf;
495 
496 /* XXX Make these percpu */
497 static void (*cb_proc)(void *arg);
498 static void *cb_arg;
499 static jmp_buf *cb_back;
500 
start_idle_thread(void * stack,jmp_buf * switch_buf)501 int start_idle_thread(void *stack, jmp_buf *switch_buf)
502 {
503 	int n;
504 
505 	set_handler(SIGWINCH);
506 
507 	/*
508 	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
509 	 * and restore signals, with the possible side-effect of
510 	 * trying to handle any signals which came when they were
511 	 * blocked, which can't be done on this stack.
512 	 * Signals must be blocked when jumping back here and restored
513 	 * after returning to the jumper.
514 	 */
515 	n = setjmp(initial_jmpbuf);
516 	switch (n) {
517 	case INIT_JMP_NEW_THREAD:
518 		(*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
519 		(*switch_buf)[0].JB_SP = (unsigned long) stack +
520 			UM_THREAD_SIZE - sizeof(void *);
521 		break;
522 	case INIT_JMP_CALLBACK:
523 		(*cb_proc)(cb_arg);
524 		longjmp(*cb_back, 1);
525 		break;
526 	case INIT_JMP_HALT:
527 		kmalloc_ok = 0;
528 		return 0;
529 	case INIT_JMP_REBOOT:
530 		kmalloc_ok = 0;
531 		return 1;
532 	default:
533 		printk(UM_KERN_ERR "Bad sigsetjmp return in %s - %d\n",
534 		       __func__, n);
535 		fatal_sigsegv();
536 	}
537 	longjmp(*switch_buf, 1);
538 
539 	/* unreachable */
540 	printk(UM_KERN_ERR "impossible long jump!");
541 	fatal_sigsegv();
542 	return 0;
543 }
544 
initial_thread_cb_skas(void (* proc)(void *),void * arg)545 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
546 {
547 	jmp_buf here;
548 
549 	cb_proc = proc;
550 	cb_arg = arg;
551 	cb_back = &here;
552 
553 	block_signals_trace();
554 	if (UML_SETJMP(&here) == 0)
555 		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
556 	unblock_signals_trace();
557 
558 	cb_proc = NULL;
559 	cb_arg = NULL;
560 	cb_back = NULL;
561 }
562 
halt_skas(void)563 void halt_skas(void)
564 {
565 	block_signals_trace();
566 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
567 }
568 
569 static bool noreboot;
570 
noreboot_cmd_param(char * str,int * add)571 static int __init noreboot_cmd_param(char *str, int *add)
572 {
573 	noreboot = true;
574 	return 0;
575 }
576 
577 __uml_setup("noreboot", noreboot_cmd_param,
578 "noreboot\n"
579 "    Rather than rebooting, exit always, akin to QEMU's -no-reboot option.\n"
580 "    This is useful if you're using CONFIG_PANIC_TIMEOUT in order to catch\n"
581 "    crashes in CI\n");
582 
reboot_skas(void)583 void reboot_skas(void)
584 {
585 	block_signals_trace();
586 	UML_LONGJMP(&initial_jmpbuf, noreboot ? INIT_JMP_HALT : INIT_JMP_REBOOT);
587 }
588 
__switch_mm(struct mm_id * mm_idp)589 void __switch_mm(struct mm_id *mm_idp)
590 {
591 	userspace_pid[0] = mm_idp->pid;
592 }
593