xref: /openbmc/linux/arch/um/os-Linux/skas/process.c (revision 32981ea5)
1 /*
2  * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
3  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
4  * Licensed under the GPL
5  */
6 
7 #include <stdlib.h>
8 #include <unistd.h>
9 #include <sched.h>
10 #include <errno.h>
11 #include <string.h>
12 #include <sys/mman.h>
13 #include <sys/wait.h>
14 #include <asm/unistd.h>
15 #include <as-layout.h>
16 #include <init.h>
17 #include <kern_util.h>
18 #include <mem.h>
19 #include <os.h>
20 #include <ptrace_user.h>
21 #include <registers.h>
22 #include <skas.h>
23 #include <sysdep/stub.h>
24 
25 int is_skas_winch(int pid, int fd, void *data)
26 {
27 	return pid == getpgrp();
28 }
29 
30 static int ptrace_dump_regs(int pid)
31 {
32 	unsigned long regs[MAX_REG_NR];
33 	int i;
34 
35 	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
36 		return -errno;
37 
38 	printk(UM_KERN_ERR "Stub registers -\n");
39 	for (i = 0; i < ARRAY_SIZE(regs); i++)
40 		printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
41 
42 	return 0;
43 }
44 
45 /*
46  * Signals that are OK to receive in the stub - we'll just continue it.
47  * SIGWINCH will happen when UML is inside a detached screen.
48  */
49 #define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
50 
51 /* Signals that the stub will finish with - anything else is an error */
52 #define STUB_DONE_MASK (1 << SIGTRAP)
53 
54 void wait_stub_done(int pid)
55 {
56 	int n, status, err;
57 
58 	while (1) {
59 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
60 		if ((n < 0) || !WIFSTOPPED(status))
61 			goto bad_wait;
62 
63 		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
64 			break;
65 
66 		err = ptrace(PTRACE_CONT, pid, 0, 0);
67 		if (err) {
68 			printk(UM_KERN_ERR "wait_stub_done : continue failed, "
69 			       "errno = %d\n", errno);
70 			fatal_sigsegv();
71 		}
72 	}
73 
74 	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
75 		return;
76 
77 bad_wait:
78 	err = ptrace_dump_regs(pid);
79 	if (err)
80 		printk(UM_KERN_ERR "Failed to get registers from stub, "
81 		       "errno = %d\n", -err);
82 	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
83 	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
84 	       status);
85 	fatal_sigsegv();
86 }
87 
88 extern unsigned long current_stub_stack(void);
89 
90 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
91 {
92 	int err;
93 	unsigned long fpregs[FP_SIZE];
94 
95 	err = get_fp_registers(pid, fpregs);
96 	if (err < 0) {
97 		printk(UM_KERN_ERR "save_fp_registers returned %d\n",
98 		       err);
99 		fatal_sigsegv();
100 	}
101 	err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
102 	if (err) {
103 		printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
104 		       "errno = %d\n", pid, errno);
105 		fatal_sigsegv();
106 	}
107 	wait_stub_done(pid);
108 
109 	/*
110 	 * faultinfo is prepared by the stub-segv-handler at start of
111 	 * the stub stack page. We just have to copy it.
112 	 */
113 	memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
114 
115 	err = put_fp_registers(pid, fpregs);
116 	if (err < 0) {
117 		printk(UM_KERN_ERR "put_fp_registers returned %d\n",
118 		       err);
119 		fatal_sigsegv();
120 	}
121 }
122 
123 static void handle_segv(int pid, struct uml_pt_regs * regs)
124 {
125 	get_skas_faultinfo(pid, &regs->faultinfo);
126 	segv(regs->faultinfo, 0, 1, NULL);
127 }
128 
129 /*
130  * To use the same value of using_sysemu as the caller, ask it that value
131  * (in local_using_sysemu
132  */
133 static void handle_trap(int pid, struct uml_pt_regs *regs,
134 			int local_using_sysemu)
135 {
136 	int err, status;
137 
138 	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
139 		fatal_sigsegv();
140 
141 	if (!local_using_sysemu)
142 	{
143 		err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
144 			     __NR_getpid);
145 		if (err < 0) {
146 			printk(UM_KERN_ERR "handle_trap - nullifying syscall "
147 			       "failed, errno = %d\n", errno);
148 			fatal_sigsegv();
149 		}
150 
151 		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
152 		if (err < 0) {
153 			printk(UM_KERN_ERR "handle_trap - continuing to end of "
154 			       "syscall failed, errno = %d\n", errno);
155 			fatal_sigsegv();
156 		}
157 
158 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
159 		if ((err < 0) || !WIFSTOPPED(status) ||
160 		    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
161 			err = ptrace_dump_regs(pid);
162 			if (err)
163 				printk(UM_KERN_ERR "Failed to get registers "
164 				       "from process, errno = %d\n", -err);
165 			printk(UM_KERN_ERR "handle_trap - failed to wait at "
166 			       "end of syscall, errno = %d, status = %d\n",
167 			       errno, status);
168 			fatal_sigsegv();
169 		}
170 	}
171 
172 	handle_syscall(regs);
173 }
174 
175 extern char __syscall_stub_start[];
176 
177 static int userspace_tramp(void *stack)
178 {
179 	void *addr;
180 	int fd;
181 	unsigned long long offset;
182 
183 	ptrace(PTRACE_TRACEME, 0, 0, 0);
184 
185 	signal(SIGTERM, SIG_DFL);
186 	signal(SIGWINCH, SIG_IGN);
187 
188 	/*
189 	 * This has a pte, but it can't be mapped in with the usual
190 	 * tlb_flush mechanism because this is part of that mechanism
191 	 */
192 	fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
193 	addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
194 		      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
195 	if (addr == MAP_FAILED) {
196 		printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
197 		       "errno = %d\n", STUB_CODE, errno);
198 		exit(1);
199 	}
200 
201 	if (stack != NULL) {
202 		fd = phys_mapping(to_phys(stack), &offset);
203 		addr = mmap((void *) STUB_DATA,
204 			    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
205 			    MAP_FIXED | MAP_SHARED, fd, offset);
206 		if (addr == MAP_FAILED) {
207 			printk(UM_KERN_ERR "mapping segfault stack "
208 			       "at 0x%lx failed, errno = %d\n",
209 			       STUB_DATA, errno);
210 			exit(1);
211 		}
212 	}
213 	if (stack != NULL) {
214 		struct sigaction sa;
215 
216 		unsigned long v = STUB_CODE +
217 				  (unsigned long) stub_segv_handler -
218 				  (unsigned long) __syscall_stub_start;
219 
220 		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
221 		sigemptyset(&sa.sa_mask);
222 		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
223 		sa.sa_sigaction = (void *) v;
224 		sa.sa_restorer = NULL;
225 		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
226 			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
227 			       "handler failed - errno = %d\n", errno);
228 			exit(1);
229 		}
230 	}
231 
232 	kill(os_getpid(), SIGSTOP);
233 	return 0;
234 }
235 
236 /* Each element set once, and only accessed by a single processor anyway */
237 #undef NR_CPUS
238 #define NR_CPUS 1
239 int userspace_pid[NR_CPUS];
240 
241 int start_userspace(unsigned long stub_stack)
242 {
243 	void *stack;
244 	unsigned long sp;
245 	int pid, status, n, flags, err;
246 
247 	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
248 		     PROT_READ | PROT_WRITE | PROT_EXEC,
249 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
250 	if (stack == MAP_FAILED) {
251 		err = -errno;
252 		printk(UM_KERN_ERR "start_userspace : mmap failed, "
253 		       "errno = %d\n", errno);
254 		return err;
255 	}
256 
257 	sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
258 
259 	flags = CLONE_FILES | SIGCHLD;
260 
261 	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
262 	if (pid < 0) {
263 		err = -errno;
264 		printk(UM_KERN_ERR "start_userspace : clone failed, "
265 		       "errno = %d\n", errno);
266 		return err;
267 	}
268 
269 	do {
270 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
271 		if (n < 0) {
272 			err = -errno;
273 			printk(UM_KERN_ERR "start_userspace : wait failed, "
274 			       "errno = %d\n", errno);
275 			goto out_kill;
276 		}
277 	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
278 
279 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
280 		err = -EINVAL;
281 		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
282 		       "status = %d\n", status);
283 		goto out_kill;
284 	}
285 
286 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
287 		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
288 		err = -errno;
289 		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
290 		       "failed, errno = %d\n", errno);
291 		goto out_kill;
292 	}
293 
294 	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
295 		err = -errno;
296 		printk(UM_KERN_ERR "start_userspace : munmap failed, "
297 		       "errno = %d\n", errno);
298 		goto out_kill;
299 	}
300 
301 	return pid;
302 
303  out_kill:
304 	os_kill_ptraced_process(pid, 1);
305 	return err;
306 }
307 
308 void userspace(struct uml_pt_regs *regs)
309 {
310 	int err, status, op, pid = userspace_pid[0];
311 	/* To prevent races if using_sysemu changes under us.*/
312 	int local_using_sysemu;
313 	siginfo_t si;
314 
315 	/* Handle any immediate reschedules or signals */
316 	interrupt_end();
317 
318 	while (1) {
319 
320 		/*
321 		 * This can legitimately fail if the process loads a
322 		 * bogus value into a segment register.  It will
323 		 * segfault and PTRACE_GETREGS will read that value
324 		 * out of the process.  However, PTRACE_SETREGS will
325 		 * fail.  In this case, there is nothing to do but
326 		 * just kill the process.
327 		 */
328 		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
329 			fatal_sigsegv();
330 
331 		if (put_fp_registers(pid, regs->fp))
332 			fatal_sigsegv();
333 
334 		/* Now we set local_using_sysemu to be used for one loop */
335 		local_using_sysemu = get_using_sysemu();
336 
337 		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
338 					     singlestepping(NULL));
339 
340 		if (ptrace(op, pid, 0, 0)) {
341 			printk(UM_KERN_ERR "userspace - ptrace continue "
342 			       "failed, op = %d, errno = %d\n", op, errno);
343 			fatal_sigsegv();
344 		}
345 
346 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
347 		if (err < 0) {
348 			printk(UM_KERN_ERR "userspace - wait failed, "
349 			       "errno = %d\n", errno);
350 			fatal_sigsegv();
351 		}
352 
353 		regs->is_user = 1;
354 		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
355 			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
356 			       "errno = %d\n", errno);
357 			fatal_sigsegv();
358 		}
359 
360 		if (get_fp_registers(pid, regs->fp)) {
361 			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
362 			       "errno = %d\n", errno);
363 			fatal_sigsegv();
364 		}
365 
366 		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
367 
368 		if (WIFSTOPPED(status)) {
369 			int sig = WSTOPSIG(status);
370 
371 			ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
372 
373 			switch (sig) {
374 			case SIGSEGV:
375 				if (PTRACE_FULL_FAULTINFO) {
376 					get_skas_faultinfo(pid,
377 							   &regs->faultinfo);
378 					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
379 							     regs);
380 				}
381 				else handle_segv(pid, regs);
382 				break;
383 			case SIGTRAP + 0x80:
384 			        handle_trap(pid, regs, local_using_sysemu);
385 				break;
386 			case SIGTRAP:
387 				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
388 				break;
389 			case SIGALRM:
390 				break;
391 			case SIGIO:
392 			case SIGILL:
393 			case SIGBUS:
394 			case SIGFPE:
395 			case SIGWINCH:
396 				block_signals();
397 				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
398 				unblock_signals();
399 				break;
400 			default:
401 				printk(UM_KERN_ERR "userspace - child stopped "
402 				       "with signal %d\n", sig);
403 				fatal_sigsegv();
404 			}
405 			pid = userspace_pid[0];
406 			interrupt_end();
407 
408 			/* Avoid -ERESTARTSYS handling in host */
409 			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
410 				PT_SYSCALL_NR(regs->gp) = -1;
411 		}
412 	}
413 }
414 
415 static unsigned long thread_regs[MAX_REG_NR];
416 static unsigned long thread_fp_regs[FP_SIZE];
417 
418 static int __init init_thread_regs(void)
419 {
420 	get_safe_registers(thread_regs, thread_fp_regs);
421 	/* Set parent's instruction pointer to start of clone-stub */
422 	thread_regs[REGS_IP_INDEX] = STUB_CODE +
423 				(unsigned long) stub_clone_handler -
424 				(unsigned long) __syscall_stub_start;
425 	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
426 		sizeof(void *);
427 #ifdef __SIGNAL_FRAMESIZE
428 	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
429 #endif
430 	return 0;
431 }
432 
433 __initcall(init_thread_regs);
434 
435 int copy_context_skas0(unsigned long new_stack, int pid)
436 {
437 	int err;
438 	unsigned long current_stack = current_stub_stack();
439 	struct stub_data *data = (struct stub_data *) current_stack;
440 	struct stub_data *child_data = (struct stub_data *) new_stack;
441 	unsigned long long new_offset;
442 	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
443 
444 	/*
445 	 * prepare offset and fd of child's stack as argument for parent's
446 	 * and child's mmap2 calls
447 	 */
448 	*data = ((struct stub_data) {
449 			.offset	= MMAP_OFFSET(new_offset),
450 			.fd     = new_fd
451 	});
452 
453 	err = ptrace_setregs(pid, thread_regs);
454 	if (err < 0) {
455 		err = -errno;
456 		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
457 		       "failed, pid = %d, errno = %d\n", pid, -err);
458 		return err;
459 	}
460 
461 	err = put_fp_registers(pid, thread_fp_regs);
462 	if (err < 0) {
463 		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
464 		       "failed, pid = %d, err = %d\n", pid, err);
465 		return err;
466 	}
467 
468 	/* set a well known return code for detection of child write failure */
469 	child_data->err = 12345678;
470 
471 	/*
472 	 * Wait, until parent has finished its work: read child's pid from
473 	 * parent's stack, and check, if bad result.
474 	 */
475 	err = ptrace(PTRACE_CONT, pid, 0, 0);
476 	if (err) {
477 		err = -errno;
478 		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
479 		       "errno = %d\n", pid, errno);
480 		return err;
481 	}
482 
483 	wait_stub_done(pid);
484 
485 	pid = data->err;
486 	if (pid < 0) {
487 		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
488 		       "error %d\n", -pid);
489 		return pid;
490 	}
491 
492 	/*
493 	 * Wait, until child has finished too: read child's result from
494 	 * child's stack and check it.
495 	 */
496 	wait_stub_done(pid);
497 	if (child_data->err != STUB_DATA) {
498 		printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
499 		       "error %ld\n", child_data->err);
500 		err = child_data->err;
501 		goto out_kill;
502 	}
503 
504 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
505 		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
506 		err = -errno;
507 		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
508 		       "failed, errno = %d\n", errno);
509 		goto out_kill;
510 	}
511 
512 	return pid;
513 
514  out_kill:
515 	os_kill_ptraced_process(pid, 1);
516 	return err;
517 }
518 
519 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
520 {
521 	(*buf)[0].JB_IP = (unsigned long) handler;
522 	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
523 		sizeof(void *);
524 }
525 
526 #define INIT_JMP_NEW_THREAD 0
527 #define INIT_JMP_CALLBACK 1
528 #define INIT_JMP_HALT 2
529 #define INIT_JMP_REBOOT 3
530 
531 void switch_threads(jmp_buf *me, jmp_buf *you)
532 {
533 	if (UML_SETJMP(me) == 0)
534 		UML_LONGJMP(you, 1);
535 }
536 
537 static jmp_buf initial_jmpbuf;
538 
539 /* XXX Make these percpu */
540 static void (*cb_proc)(void *arg);
541 static void *cb_arg;
542 static jmp_buf *cb_back;
543 
544 int start_idle_thread(void *stack, jmp_buf *switch_buf)
545 {
546 	int n;
547 
548 	set_handler(SIGWINCH);
549 
550 	/*
551 	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
552 	 * and restore signals, with the possible side-effect of
553 	 * trying to handle any signals which came when they were
554 	 * blocked, which can't be done on this stack.
555 	 * Signals must be blocked when jumping back here and restored
556 	 * after returning to the jumper.
557 	 */
558 	n = setjmp(initial_jmpbuf);
559 	switch (n) {
560 	case INIT_JMP_NEW_THREAD:
561 		(*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
562 		(*switch_buf)[0].JB_SP = (unsigned long) stack +
563 			UM_THREAD_SIZE - sizeof(void *);
564 		break;
565 	case INIT_JMP_CALLBACK:
566 		(*cb_proc)(cb_arg);
567 		longjmp(*cb_back, 1);
568 		break;
569 	case INIT_JMP_HALT:
570 		kmalloc_ok = 0;
571 		return 0;
572 	case INIT_JMP_REBOOT:
573 		kmalloc_ok = 0;
574 		return 1;
575 	default:
576 		printk(UM_KERN_ERR "Bad sigsetjmp return in "
577 		       "start_idle_thread - %d\n", n);
578 		fatal_sigsegv();
579 	}
580 	longjmp(*switch_buf, 1);
581 }
582 
583 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
584 {
585 	jmp_buf here;
586 
587 	cb_proc = proc;
588 	cb_arg = arg;
589 	cb_back = &here;
590 
591 	block_signals();
592 	if (UML_SETJMP(&here) == 0)
593 		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
594 	unblock_signals();
595 
596 	cb_proc = NULL;
597 	cb_arg = NULL;
598 	cb_back = NULL;
599 }
600 
601 void halt_skas(void)
602 {
603 	block_signals();
604 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
605 }
606 
607 void reboot_skas(void)
608 {
609 	block_signals();
610 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
611 }
612 
613 void __switch_mm(struct mm_id *mm_idp)
614 {
615 	userspace_pid[0] = mm_idp->u.pid;
616 }
617