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 
is_skas_winch(int pid,int fd,void * data)27 int is_skas_winch(int pid, int fd, void *data)
28 {
29 	return pid == getpgrp();
30 }
31 
ptrace_reg_name(int idx)32 static const char *ptrace_reg_name(int idx)
33 {
34 #define R(n) case HOST_##n: return #n
35 
36 	switch (idx) {
37 #ifdef __x86_64__
38 	R(BX);
39 	R(CX);
40 	R(DI);
41 	R(SI);
42 	R(DX);
43 	R(BP);
44 	R(AX);
45 	R(R8);
46 	R(R9);
47 	R(R10);
48 	R(R11);
49 	R(R12);
50 	R(R13);
51 	R(R14);
52 	R(R15);
53 	R(ORIG_AX);
54 	R(CS);
55 	R(SS);
56 	R(EFLAGS);
57 #elif defined(__i386__)
58 	R(IP);
59 	R(SP);
60 	R(EFLAGS);
61 	R(AX);
62 	R(BX);
63 	R(CX);
64 	R(DX);
65 	R(SI);
66 	R(DI);
67 	R(BP);
68 	R(CS);
69 	R(SS);
70 	R(DS);
71 	R(FS);
72 	R(ES);
73 	R(GS);
74 	R(ORIG_AX);
75 #endif
76 	}
77 	return "";
78 }
79 
ptrace_dump_regs(int pid)80 static int ptrace_dump_regs(int pid)
81 {
82 	unsigned long regs[MAX_REG_NR];
83 	int i;
84 
85 	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
86 		return -errno;
87 
88 	printk(UM_KERN_ERR "Stub registers -\n");
89 	for (i = 0; i < ARRAY_SIZE(regs); i++) {
90 		const char *regname = ptrace_reg_name(i);
91 
92 		printk(UM_KERN_ERR "\t%s\t(%2d): %lx\n", regname, i, regs[i]);
93 	}
94 
95 	return 0;
96 }
97 
98 /*
99  * Signals that are OK to receive in the stub - we'll just continue it.
100  * SIGWINCH will happen when UML is inside a detached screen.
101  */
102 #define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
103 
104 /* Signals that the stub will finish with - anything else is an error */
105 #define STUB_DONE_MASK (1 << SIGTRAP)
106 
wait_stub_done(int pid)107 void wait_stub_done(int pid)
108 {
109 	int n, status, err;
110 
111 	while (1) {
112 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
113 		if ((n < 0) || !WIFSTOPPED(status))
114 			goto bad_wait;
115 
116 		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
117 			break;
118 
119 		err = ptrace(PTRACE_CONT, pid, 0, 0);
120 		if (err) {
121 			printk(UM_KERN_ERR "%s : continue failed, errno = %d\n",
122 			       __func__, errno);
123 			fatal_sigsegv();
124 		}
125 	}
126 
127 	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
128 		return;
129 
130 bad_wait:
131 	err = ptrace_dump_regs(pid);
132 	if (err)
133 		printk(UM_KERN_ERR "Failed to get registers from stub, errno = %d\n",
134 		       -err);
135 	printk(UM_KERN_ERR "%s : failed to wait for SIGTRAP, pid = %d, n = %d, errno = %d, status = 0x%x\n",
136 	       __func__, pid, n, errno, status);
137 	fatal_sigsegv();
138 }
139 
140 extern unsigned long current_stub_stack(void);
141 
get_skas_faultinfo(int pid,struct faultinfo * fi,unsigned long * aux_fp_regs)142 static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
143 {
144 	int err;
145 
146 	err = get_fp_registers(pid, aux_fp_regs);
147 	if (err < 0) {
148 		printk(UM_KERN_ERR "save_fp_registers returned %d\n",
149 		       err);
150 		fatal_sigsegv();
151 	}
152 	err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
153 	if (err) {
154 		printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
155 		       "errno = %d\n", pid, errno);
156 		fatal_sigsegv();
157 	}
158 	wait_stub_done(pid);
159 
160 	/*
161 	 * faultinfo is prepared by the stub_segv_handler at start of
162 	 * the stub stack page. We just have to copy it.
163 	 */
164 	memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
165 
166 	err = put_fp_registers(pid, aux_fp_regs);
167 	if (err < 0) {
168 		printk(UM_KERN_ERR "put_fp_registers returned %d\n",
169 		       err);
170 		fatal_sigsegv();
171 	}
172 }
173 
handle_segv(int pid,struct uml_pt_regs * regs,unsigned long * aux_fp_regs)174 static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
175 {
176 	get_skas_faultinfo(pid, &regs->faultinfo, aux_fp_regs);
177 	segv(regs->faultinfo, 0, 1, NULL);
178 }
179 
180 /*
181  * To use the same value of using_sysemu as the caller, ask it that value
182  * (in local_using_sysemu
183  */
handle_trap(int pid,struct uml_pt_regs * regs,int local_using_sysemu)184 static void handle_trap(int pid, struct uml_pt_regs *regs,
185 			int local_using_sysemu)
186 {
187 	int err, status;
188 
189 	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
190 		fatal_sigsegv();
191 
192 	if (!local_using_sysemu)
193 	{
194 		err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
195 			     __NR_getpid);
196 		if (err < 0) {
197 			printk(UM_KERN_ERR "%s - nullifying syscall failed, errno = %d\n",
198 			       __func__, errno);
199 			fatal_sigsegv();
200 		}
201 
202 		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
203 		if (err < 0) {
204 			printk(UM_KERN_ERR "%s - continuing to end of syscall failed, errno = %d\n",
205 			       __func__, errno);
206 			fatal_sigsegv();
207 		}
208 
209 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
210 		if ((err < 0) || !WIFSTOPPED(status) ||
211 		    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
212 			err = ptrace_dump_regs(pid);
213 			if (err)
214 				printk(UM_KERN_ERR "Failed to get registers from process, errno = %d\n",
215 				       -err);
216 			printk(UM_KERN_ERR "%s - failed to wait at end of syscall, errno = %d, status = %d\n",
217 			       __func__, errno, status);
218 			fatal_sigsegv();
219 		}
220 	}
221 
222 	handle_syscall(regs);
223 }
224 
225 extern char __syscall_stub_start[];
226 
227 /**
228  * userspace_tramp() - userspace trampoline
229  * @stack:	pointer to the new userspace stack page, can be NULL, if? FIXME:
230  *
231  * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
232  * This function will run on a temporary stack page.
233  * It ptrace()'es itself, then
234  * Two pages are mapped into the userspace address space:
235  * - STUB_CODE (with EXEC), which contains the skas stub code
236  * - 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.
237  * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
238  * And last the process stops itself to give control to the UML kernel for this userspace process.
239  *
240  * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
241  */
userspace_tramp(void * stack)242 static int userspace_tramp(void *stack)
243 {
244 	void *addr;
245 	int fd;
246 	unsigned long long offset;
247 
248 	ptrace(PTRACE_TRACEME, 0, 0, 0);
249 
250 	signal(SIGTERM, SIG_DFL);
251 	signal(SIGWINCH, SIG_IGN);
252 
253 	fd = phys_mapping(uml_to_phys(__syscall_stub_start), &offset);
254 	addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
255 		      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
256 	if (addr == MAP_FAILED) {
257 		printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, errno = %d\n",
258 		       STUB_CODE, errno);
259 		exit(1);
260 	}
261 
262 	if (stack != NULL) {
263 		fd = phys_mapping(uml_to_phys(stack), &offset);
264 		addr = mmap((void *) STUB_DATA,
265 			    STUB_DATA_PAGES * UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
266 			    MAP_FIXED | MAP_SHARED, fd, offset);
267 		if (addr == MAP_FAILED) {
268 			printk(UM_KERN_ERR "mapping segfault stack at 0x%lx failed, errno = %d\n",
269 			       STUB_DATA, errno);
270 			exit(1);
271 		}
272 	}
273 	if (stack != NULL) {
274 		struct sigaction sa;
275 
276 		unsigned long v = STUB_CODE +
277 				  (unsigned long) stub_segv_handler -
278 				  (unsigned long) __syscall_stub_start;
279 
280 		set_sigstack((void *) STUB_DATA, STUB_DATA_PAGES * UM_KERN_PAGE_SIZE);
281 		sigemptyset(&sa.sa_mask);
282 		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
283 		sa.sa_sigaction = (void *) v;
284 		sa.sa_restorer = NULL;
285 		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
286 			printk(UM_KERN_ERR "%s - setting SIGSEGV handler failed - errno = %d\n",
287 			       __func__, errno);
288 			exit(1);
289 		}
290 	}
291 
292 	kill(os_getpid(), SIGSTOP);
293 	return 0;
294 }
295 
296 int userspace_pid[NR_CPUS];
297 int kill_userspace_mm[NR_CPUS];
298 
299 /**
300  * start_userspace() - prepare a new userspace process
301  * @stub_stack:	pointer to the stub stack. Can be NULL, if? FIXME:
302  *
303  * Setups a new temporary stack page that is used while userspace_tramp() runs
304  * Clones the kernel process into a new userspace process, with FDs only.
305  *
306  * Return: When positive: the process id of the new userspace process,
307  *         when negative: an error number.
308  * FIXME: can PIDs become negative?!
309  */
start_userspace(unsigned long stub_stack)310 int start_userspace(unsigned long stub_stack)
311 {
312 	void *stack;
313 	unsigned long sp;
314 	int pid, status, n, flags, err;
315 
316 	/* setup a temporary stack page */
317 	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
318 		     PROT_READ | PROT_WRITE | PROT_EXEC,
319 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
320 	if (stack == MAP_FAILED) {
321 		err = -errno;
322 		printk(UM_KERN_ERR "%s : mmap failed, errno = %d\n",
323 		       __func__, errno);
324 		return err;
325 	}
326 
327 	/* set stack pointer to the end of the stack page, so it can grow downwards */
328 	sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
329 
330 	flags = CLONE_FILES | SIGCHLD;
331 
332 	/* clone into new userspace process */
333 	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
334 	if (pid < 0) {
335 		err = -errno;
336 		printk(UM_KERN_ERR "%s : clone failed, errno = %d\n",
337 		       __func__, errno);
338 		return err;
339 	}
340 
341 	do {
342 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
343 		if (n < 0) {
344 			err = -errno;
345 			printk(UM_KERN_ERR "%s : wait failed, errno = %d\n",
346 			       __func__, errno);
347 			goto out_kill;
348 		}
349 	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
350 
351 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
352 		err = -EINVAL;
353 		printk(UM_KERN_ERR "%s : expected SIGSTOP, got status = %d\n",
354 		       __func__, status);
355 		goto out_kill;
356 	}
357 
358 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
359 		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
360 		err = -errno;
361 		printk(UM_KERN_ERR "%s : PTRACE_OLDSETOPTIONS failed, errno = %d\n",
362 		       __func__, errno);
363 		goto out_kill;
364 	}
365 
366 	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
367 		err = -errno;
368 		printk(UM_KERN_ERR "%s : munmap failed, errno = %d\n",
369 		       __func__, errno);
370 		goto out_kill;
371 	}
372 
373 	return pid;
374 
375  out_kill:
376 	os_kill_ptraced_process(pid, 1);
377 	return err;
378 }
379 
userspace(struct uml_pt_regs * regs,unsigned long * aux_fp_regs)380 void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
381 {
382 	int err, status, op, pid = userspace_pid[0];
383 	/* To prevent races if using_sysemu changes under us.*/
384 	int local_using_sysemu;
385 	siginfo_t si;
386 
387 	/* Handle any immediate reschedules or signals */
388 	interrupt_end();
389 
390 	while (1) {
391 		if (kill_userspace_mm[0])
392 			fatal_sigsegv();
393 
394 		/*
395 		 * This can legitimately fail if the process loads a
396 		 * bogus value into a segment register.  It will
397 		 * segfault and PTRACE_GETREGS will read that value
398 		 * out of the process.  However, PTRACE_SETREGS will
399 		 * fail.  In this case, there is nothing to do but
400 		 * just kill the process.
401 		 */
402 		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
403 			printk(UM_KERN_ERR "%s - ptrace set regs failed, errno = %d\n",
404 			       __func__, errno);
405 			fatal_sigsegv();
406 		}
407 
408 		if (put_fp_registers(pid, regs->fp)) {
409 			printk(UM_KERN_ERR "%s - ptrace set fp regs failed, errno = %d\n",
410 			       __func__, errno);
411 			fatal_sigsegv();
412 		}
413 
414 		/* Now we set local_using_sysemu to be used for one loop */
415 		local_using_sysemu = get_using_sysemu();
416 
417 		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
418 					     singlestepping(NULL));
419 
420 		if (ptrace(op, pid, 0, 0)) {
421 			printk(UM_KERN_ERR "%s - ptrace continue failed, op = %d, errno = %d\n",
422 			       __func__, op, errno);
423 			fatal_sigsegv();
424 		}
425 
426 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
427 		if (err < 0) {
428 			printk(UM_KERN_ERR "%s - wait failed, errno = %d\n",
429 			       __func__, errno);
430 			fatal_sigsegv();
431 		}
432 
433 		regs->is_user = 1;
434 		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
435 			printk(UM_KERN_ERR "%s - PTRACE_GETREGS failed, errno = %d\n",
436 			       __func__, errno);
437 			fatal_sigsegv();
438 		}
439 
440 		if (get_fp_registers(pid, regs->fp)) {
441 			printk(UM_KERN_ERR "%s -  get_fp_registers failed, errno = %d\n",
442 			       __func__, errno);
443 			fatal_sigsegv();
444 		}
445 
446 		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
447 
448 		if (WIFSTOPPED(status)) {
449 			int sig = WSTOPSIG(status);
450 
451 			/* These signal handlers need the si argument.
452 			 * The SIGIO and SIGALARM handlers which constitute the
453 			 * majority of invocations, do not use it.
454 			 */
455 			switch (sig) {
456 			case SIGSEGV:
457 			case SIGTRAP:
458 			case SIGILL:
459 			case SIGBUS:
460 			case SIGFPE:
461 			case SIGWINCH:
462 				ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
463 				break;
464 			}
465 
466 			switch (sig) {
467 			case SIGSEGV:
468 				if (PTRACE_FULL_FAULTINFO) {
469 					get_skas_faultinfo(pid,
470 							   &regs->faultinfo, aux_fp_regs);
471 					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
472 							     regs);
473 				}
474 				else handle_segv(pid, regs, aux_fp_regs);
475 				break;
476 			case SIGTRAP + 0x80:
477 			        handle_trap(pid, regs, local_using_sysemu);
478 				break;
479 			case SIGTRAP:
480 				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
481 				break;
482 			case SIGALRM:
483 				break;
484 			case SIGIO:
485 			case SIGILL:
486 			case SIGBUS:
487 			case SIGFPE:
488 			case SIGWINCH:
489 				block_signals_trace();
490 				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
491 				unblock_signals_trace();
492 				break;
493 			default:
494 				printk(UM_KERN_ERR "%s - child stopped with signal %d\n",
495 				       __func__, sig);
496 				fatal_sigsegv();
497 			}
498 			pid = userspace_pid[0];
499 			interrupt_end();
500 
501 			/* Avoid -ERESTARTSYS handling in host */
502 			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
503 				PT_SYSCALL_NR(regs->gp) = -1;
504 		}
505 	}
506 }
507 
508 static unsigned long thread_regs[MAX_REG_NR];
509 static unsigned long thread_fp_regs[FP_SIZE];
510 
init_thread_regs(void)511 static int __init init_thread_regs(void)
512 {
513 	get_safe_registers(thread_regs, thread_fp_regs);
514 	/* Set parent's instruction pointer to start of clone-stub */
515 	thread_regs[REGS_IP_INDEX] = STUB_CODE +
516 				(unsigned long) stub_clone_handler -
517 				(unsigned long) __syscall_stub_start;
518 	thread_regs[REGS_SP_INDEX] = STUB_DATA + STUB_DATA_PAGES * UM_KERN_PAGE_SIZE -
519 		sizeof(void *);
520 #ifdef __SIGNAL_FRAMESIZE
521 	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
522 #endif
523 	return 0;
524 }
525 
526 __initcall(init_thread_regs);
527 
copy_context_skas0(unsigned long new_stack,int pid)528 int copy_context_skas0(unsigned long new_stack, int pid)
529 {
530 	int err;
531 	unsigned long current_stack = current_stub_stack();
532 	struct stub_data *data = (struct stub_data *) current_stack;
533 	struct stub_data *child_data = (struct stub_data *) new_stack;
534 	unsigned long long new_offset;
535 	int new_fd = phys_mapping(uml_to_phys((void *)new_stack), &new_offset);
536 
537 	/*
538 	 * prepare offset and fd of child's stack as argument for parent's
539 	 * and child's mmap2 calls
540 	 */
541 	*data = ((struct stub_data) {
542 		.offset	= MMAP_OFFSET(new_offset),
543 		.fd     = new_fd,
544 		.parent_err = -ESRCH,
545 		.child_err = 0,
546 	});
547 
548 	*child_data = ((struct stub_data) {
549 		.child_err = -ESRCH,
550 	});
551 
552 	err = ptrace_setregs(pid, thread_regs);
553 	if (err < 0) {
554 		err = -errno;
555 		printk(UM_KERN_ERR "%s : PTRACE_SETREGS failed, pid = %d, errno = %d\n",
556 		      __func__, pid, -err);
557 		return err;
558 	}
559 
560 	err = put_fp_registers(pid, thread_fp_regs);
561 	if (err < 0) {
562 		printk(UM_KERN_ERR "%s : put_fp_registers failed, pid = %d, err = %d\n",
563 		       __func__, pid, err);
564 		return err;
565 	}
566 
567 	/*
568 	 * Wait, until parent has finished its work: read child's pid from
569 	 * parent's stack, and check, if bad result.
570 	 */
571 	err = ptrace(PTRACE_CONT, pid, 0, 0);
572 	if (err) {
573 		err = -errno;
574 		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, errno = %d\n",
575 		       pid, errno);
576 		return err;
577 	}
578 
579 	wait_stub_done(pid);
580 
581 	pid = data->parent_err;
582 	if (pid < 0) {
583 		printk(UM_KERN_ERR "%s - stub-parent reports error %d\n",
584 		      __func__, -pid);
585 		return pid;
586 	}
587 
588 	/*
589 	 * Wait, until child has finished too: read child's result from
590 	 * child's stack and check it.
591 	 */
592 	wait_stub_done(pid);
593 	if (child_data->child_err != STUB_DATA) {
594 		printk(UM_KERN_ERR "%s - stub-child %d reports error %ld\n",
595 		       __func__, pid, data->child_err);
596 		err = data->child_err;
597 		goto out_kill;
598 	}
599 
600 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
601 		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
602 		err = -errno;
603 		printk(UM_KERN_ERR "%s : PTRACE_OLDSETOPTIONS failed, errno = %d\n",
604 		       __func__, errno);
605 		goto out_kill;
606 	}
607 
608 	return pid;
609 
610  out_kill:
611 	os_kill_ptraced_process(pid, 1);
612 	return err;
613 }
614 
new_thread(void * stack,jmp_buf * buf,void (* handler)(void))615 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
616 {
617 	(*buf)[0].JB_IP = (unsigned long) handler;
618 	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
619 		sizeof(void *);
620 }
621 
622 #define INIT_JMP_NEW_THREAD 0
623 #define INIT_JMP_CALLBACK 1
624 #define INIT_JMP_HALT 2
625 #define INIT_JMP_REBOOT 3
626 
switch_threads(jmp_buf * me,jmp_buf * you)627 void switch_threads(jmp_buf *me, jmp_buf *you)
628 {
629 	if (UML_SETJMP(me) == 0)
630 		UML_LONGJMP(you, 1);
631 }
632 
633 static jmp_buf initial_jmpbuf;
634 
635 /* XXX Make these percpu */
636 static void (*cb_proc)(void *arg);
637 static void *cb_arg;
638 static jmp_buf *cb_back;
639 
start_idle_thread(void * stack,jmp_buf * switch_buf)640 int start_idle_thread(void *stack, jmp_buf *switch_buf)
641 {
642 	int n;
643 
644 	set_handler(SIGWINCH);
645 
646 	/*
647 	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
648 	 * and restore signals, with the possible side-effect of
649 	 * trying to handle any signals which came when they were
650 	 * blocked, which can't be done on this stack.
651 	 * Signals must be blocked when jumping back here and restored
652 	 * after returning to the jumper.
653 	 */
654 	n = setjmp(initial_jmpbuf);
655 	switch (n) {
656 	case INIT_JMP_NEW_THREAD:
657 		(*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
658 		(*switch_buf)[0].JB_SP = (unsigned long) stack +
659 			UM_THREAD_SIZE - sizeof(void *);
660 		break;
661 	case INIT_JMP_CALLBACK:
662 		(*cb_proc)(cb_arg);
663 		longjmp(*cb_back, 1);
664 		break;
665 	case INIT_JMP_HALT:
666 		kmalloc_ok = 0;
667 		return 0;
668 	case INIT_JMP_REBOOT:
669 		kmalloc_ok = 0;
670 		return 1;
671 	default:
672 		printk(UM_KERN_ERR "Bad sigsetjmp return in %s - %d\n",
673 		       __func__, n);
674 		fatal_sigsegv();
675 	}
676 	longjmp(*switch_buf, 1);
677 
678 	/* unreachable */
679 	printk(UM_KERN_ERR "impossible long jump!");
680 	fatal_sigsegv();
681 	return 0;
682 }
683 
initial_thread_cb_skas(void (* proc)(void *),void * arg)684 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
685 {
686 	jmp_buf here;
687 
688 	cb_proc = proc;
689 	cb_arg = arg;
690 	cb_back = &here;
691 
692 	block_signals_trace();
693 	if (UML_SETJMP(&here) == 0)
694 		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
695 	unblock_signals_trace();
696 
697 	cb_proc = NULL;
698 	cb_arg = NULL;
699 	cb_back = NULL;
700 }
701 
halt_skas(void)702 void halt_skas(void)
703 {
704 	block_signals_trace();
705 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
706 }
707 
708 static bool noreboot;
709 
noreboot_cmd_param(char * str,int * add)710 static int __init noreboot_cmd_param(char *str, int *add)
711 {
712 	noreboot = true;
713 	return 0;
714 }
715 
716 __uml_setup("noreboot", noreboot_cmd_param,
717 "noreboot\n"
718 "    Rather than rebooting, exit always, akin to QEMU's -no-reboot option.\n"
719 "    This is useful if you're using CONFIG_PANIC_TIMEOUT in order to catch\n"
720 "    crashes in CI\n");
721 
reboot_skas(void)722 void reboot_skas(void)
723 {
724 	block_signals_trace();
725 	UML_LONGJMP(&initial_jmpbuf, noreboot ? INIT_JMP_HALT : INIT_JMP_REBOOT);
726 }
727 
__switch_mm(struct mm_id * mm_idp)728 void __switch_mm(struct mm_id *mm_idp)
729 {
730 	userspace_pid[0] = mm_idp->u.pid;
731 	kill_userspace_mm[0] = mm_idp->kill;
732 }
733