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
27int is_skas_winch(int pid, int fd, void *data)
28{
29	return pid == getpgrp();
30}
31
32static int ptrace_dump_regs(int pid)
33{
34	unsigned long regs[MAX_REG_NR];
35	int i;
36
37	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
38		return -errno;
39
40	printk(UM_KERN_ERR "Stub registers -\n");
41	for (i = 0; i < ARRAY_SIZE(regs); i++)
42		printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
43
44	return 0;
45}
46
47/*
48 * Signals that are OK to receive in the stub - we'll just continue it.
49 * SIGWINCH will happen when UML is inside a detached screen.
50 */
51#define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
52
53/* Signals that the stub will finish with - anything else is an error */
54#define STUB_DONE_MASK (1 << SIGTRAP)
55
56void wait_stub_done(int pid)
57{
58	int n, status, err;
59
60	while (1) {
61		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
62		if ((n < 0) || !WIFSTOPPED(status))
63			goto bad_wait;
64
65		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
66			break;
67
68		err = ptrace(PTRACE_CONT, pid, 0, 0);
69		if (err) {
70			printk(UM_KERN_ERR "wait_stub_done : continue failed, "
71			       "errno = %d\n", errno);
72			fatal_sigsegv();
73		}
74	}
75
76	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
77		return;
78
79bad_wait:
80	err = ptrace_dump_regs(pid);
81	if (err)
82		printk(UM_KERN_ERR "Failed to get registers from stub, "
83		       "errno = %d\n", -err);
84	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
85	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
86	       status);
87	fatal_sigsegv();
88}
89
90extern unsigned long current_stub_stack(void);
91
92static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
93{
94	int err;
95
96	err = get_fp_registers(pid, aux_fp_regs);
97	if (err < 0) {
98		printk(UM_KERN_ERR "save_fp_registers returned %d\n",
99		       err);
100		fatal_sigsegv();
101	}
102	err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
103	if (err) {
104		printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
105		       "errno = %d\n", pid, errno);
106		fatal_sigsegv();
107	}
108	wait_stub_done(pid);
109
110	/*
111	 * faultinfo is prepared by the stub_segv_handler at start of
112	 * the stub stack page. We just have to copy it.
113	 */
114	memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
115
116	err = put_fp_registers(pid, aux_fp_regs);
117	if (err < 0) {
118		printk(UM_KERN_ERR "put_fp_registers returned %d\n",
119		       err);
120		fatal_sigsegv();
121	}
122}
123
124static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
125{
126	get_skas_faultinfo(pid, &regs->faultinfo, aux_fp_regs);
127	segv(regs->faultinfo, 0, 1, NULL);
128}
129
130/*
131 * To use the same value of using_sysemu as the caller, ask it that value
132 * (in local_using_sysemu
133 */
134static void handle_trap(int pid, struct uml_pt_regs *regs,
135			int local_using_sysemu)
136{
137	int err, status;
138
139	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
140		fatal_sigsegv();
141
142	if (!local_using_sysemu)
143	{
144		err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
145			     __NR_getpid);
146		if (err < 0) {
147			printk(UM_KERN_ERR "handle_trap - nullifying syscall "
148			       "failed, errno = %d\n", errno);
149			fatal_sigsegv();
150		}
151
152		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
153		if (err < 0) {
154			printk(UM_KERN_ERR "handle_trap - continuing to end of "
155			       "syscall failed, errno = %d\n", errno);
156			fatal_sigsegv();
157		}
158
159		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
160		if ((err < 0) || !WIFSTOPPED(status) ||
161		    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
162			err = ptrace_dump_regs(pid);
163			if (err)
164				printk(UM_KERN_ERR "Failed to get registers "
165				       "from process, errno = %d\n", -err);
166			printk(UM_KERN_ERR "handle_trap - failed to wait at "
167			       "end of syscall, errno = %d, status = %d\n",
168			       errno, status);
169			fatal_sigsegv();
170		}
171	}
172
173	handle_syscall(regs);
174}
175
176extern char __syscall_stub_start[];
177
178/**
179 * userspace_tramp() - userspace trampoline
180 * @stack:	pointer to the new userspace stack page, can be NULL, if? FIXME:
181 *
182 * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
183 * This function will run on a temporary stack page.
184 * It ptrace()'es itself, then
185 * Two pages are mapped into the userspace address space:
186 * - STUB_CODE (with EXEC), which contains the skas stub code
187 * - 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.
188 * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
189 * And last the process stops itself to give control to the UML kernel for this userspace process.
190 *
191 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
192 */
193static int userspace_tramp(void *stack)
194{
195	void *addr;
196	int fd;
197	unsigned long long offset;
198
199	ptrace(PTRACE_TRACEME, 0, 0, 0);
200
201	signal(SIGTERM, SIG_DFL);
202	signal(SIGWINCH, SIG_IGN);
203
204	/*
205	 * This has a pte, but it can't be mapped in with the usual
206	 * tlb_flush mechanism because this is part of that mechanism
207	 */
208	fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
209	addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
210		      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
211	if (addr == MAP_FAILED) {
212		printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
213		       "errno = %d\n", STUB_CODE, errno);
214		exit(1);
215	}
216
217	if (stack != NULL) {
218		fd = phys_mapping(to_phys(stack), &offset);
219		addr = mmap((void *) STUB_DATA,
220			    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
221			    MAP_FIXED | MAP_SHARED, fd, offset);
222		if (addr == MAP_FAILED) {
223			printk(UM_KERN_ERR "mapping segfault stack "
224			       "at 0x%lx failed, errno = %d\n",
225			       STUB_DATA, errno);
226			exit(1);
227		}
228	}
229	if (stack != NULL) {
230		struct sigaction sa;
231
232		unsigned long v = STUB_CODE +
233				  (unsigned long) stub_segv_handler -
234				  (unsigned long) __syscall_stub_start;
235
236		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
237		sigemptyset(&sa.sa_mask);
238		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
239		sa.sa_sigaction = (void *) v;
240		sa.sa_restorer = NULL;
241		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
242			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
243			       "handler failed - errno = %d\n", errno);
244			exit(1);
245		}
246	}
247
248	kill(os_getpid(), SIGSTOP);
249	return 0;
250}
251
252int userspace_pid[NR_CPUS];
253int kill_userspace_mm[NR_CPUS];
254
255/**
256 * start_userspace() - prepare a new userspace process
257 * @stub_stack:	pointer to the stub stack. Can be NULL, if? FIXME:
258 *
259 * Setups a new temporary stack page that is used while userspace_tramp() runs
260 * Clones the kernel process into a new userspace process, with FDs only.
261 *
262 * Return: When positive: the process id of the new userspace process,
263 *         when negative: an error number.
264 * FIXME: can PIDs become negative?!
265 */
266int start_userspace(unsigned long stub_stack)
267{
268	void *stack;
269	unsigned long sp;
270	int pid, status, n, flags, err;
271
272	/* setup a temporary stack page */
273	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
274		     PROT_READ | PROT_WRITE | PROT_EXEC,
275		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
276	if (stack == MAP_FAILED) {
277		err = -errno;
278		printk(UM_KERN_ERR "start_userspace : mmap failed, "
279		       "errno = %d\n", errno);
280		return err;
281	}
282
283	/* set stack pointer to the end of the stack page, so it can grow downwards */
284	sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
285
286	flags = CLONE_FILES | SIGCHLD;
287
288	/* clone into new userspace process */
289	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
290	if (pid < 0) {
291		err = -errno;
292		printk(UM_KERN_ERR "start_userspace : clone failed, "
293		       "errno = %d\n", errno);
294		return err;
295	}
296
297	do {
298		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
299		if (n < 0) {
300			err = -errno;
301			printk(UM_KERN_ERR "start_userspace : wait failed, "
302			       "errno = %d\n", errno);
303			goto out_kill;
304		}
305	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
306
307	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
308		err = -EINVAL;
309		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
310		       "status = %d\n", status);
311		goto out_kill;
312	}
313
314	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
315		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
316		err = -errno;
317		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
318		       "failed, errno = %d\n", errno);
319		goto out_kill;
320	}
321
322	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
323		err = -errno;
324		printk(UM_KERN_ERR "start_userspace : munmap failed, "
325		       "errno = %d\n", errno);
326		goto out_kill;
327	}
328
329	return pid;
330
331 out_kill:
332	os_kill_ptraced_process(pid, 1);
333	return err;
334}
335
336void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
337{
338	int err, status, op, pid = userspace_pid[0];
339	/* To prevent races if using_sysemu changes under us.*/
340	int local_using_sysemu;
341	siginfo_t si;
342
343	/* Handle any immediate reschedules or signals */
344	interrupt_end();
345
346	while (1) {
347		if (kill_userspace_mm[0])
348			fatal_sigsegv();
349
350		/*
351		 * This can legitimately fail if the process loads a
352		 * bogus value into a segment register.  It will
353		 * segfault and PTRACE_GETREGS will read that value
354		 * out of the process.  However, PTRACE_SETREGS will
355		 * fail.  In this case, there is nothing to do but
356		 * just kill the process.
357		 */
358		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
359			printk(UM_KERN_ERR "userspace - ptrace set regs "
360			       "failed, errno = %d\n", errno);
361			fatal_sigsegv();
362		}
363
364		if (put_fp_registers(pid, regs->fp)) {
365			printk(UM_KERN_ERR "userspace - ptrace set fp regs "
366			       "failed, errno = %d\n", errno);
367			fatal_sigsegv();
368		}
369
370		/* Now we set local_using_sysemu to be used for one loop */
371		local_using_sysemu = get_using_sysemu();
372
373		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
374					     singlestepping(NULL));
375
376		if (ptrace(op, pid, 0, 0)) {
377			printk(UM_KERN_ERR "userspace - ptrace continue "
378			       "failed, op = %d, errno = %d\n", op, errno);
379			fatal_sigsegv();
380		}
381
382		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
383		if (err < 0) {
384			printk(UM_KERN_ERR "userspace - wait failed, "
385			       "errno = %d\n", errno);
386			fatal_sigsegv();
387		}
388
389		regs->is_user = 1;
390		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
391			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
392			       "errno = %d\n", errno);
393			fatal_sigsegv();
394		}
395
396		if (get_fp_registers(pid, regs->fp)) {
397			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
398			       "errno = %d\n", errno);
399			fatal_sigsegv();
400		}
401
402		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
403
404		if (WIFSTOPPED(status)) {
405			int sig = WSTOPSIG(status);
406
407			ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
408
409			switch (sig) {
410			case SIGSEGV:
411				if (PTRACE_FULL_FAULTINFO) {
412					get_skas_faultinfo(pid,
413							   &regs->faultinfo, aux_fp_regs);
414					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
415							     regs);
416				}
417				else handle_segv(pid, regs, aux_fp_regs);
418				break;
419			case SIGTRAP + 0x80:
420			        handle_trap(pid, regs, local_using_sysemu);
421				break;
422			case SIGTRAP:
423				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
424				break;
425			case SIGALRM:
426				break;
427			case SIGIO:
428			case SIGILL:
429			case SIGBUS:
430			case SIGFPE:
431			case SIGWINCH:
432				block_signals_trace();
433				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
434				unblock_signals_trace();
435				break;
436			default:
437				printk(UM_KERN_ERR "userspace - child stopped "
438				       "with signal %d\n", sig);
439				fatal_sigsegv();
440			}
441			pid = userspace_pid[0];
442			interrupt_end();
443
444			/* Avoid -ERESTARTSYS handling in host */
445			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
446				PT_SYSCALL_NR(regs->gp) = -1;
447		}
448	}
449}
450
451static unsigned long thread_regs[MAX_REG_NR];
452static unsigned long thread_fp_regs[FP_SIZE];
453
454static int __init init_thread_regs(void)
455{
456	get_safe_registers(thread_regs, thread_fp_regs);
457	/* Set parent's instruction pointer to start of clone-stub */
458	thread_regs[REGS_IP_INDEX] = STUB_CODE +
459				(unsigned long) stub_clone_handler -
460				(unsigned long) __syscall_stub_start;
461	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
462		sizeof(void *);
463#ifdef __SIGNAL_FRAMESIZE
464	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
465#endif
466	return 0;
467}
468
469__initcall(init_thread_regs);
470
471int copy_context_skas0(unsigned long new_stack, int pid)
472{
473	int err;
474	unsigned long current_stack = current_stub_stack();
475	struct stub_data *data = (struct stub_data *) current_stack;
476	struct stub_data *child_data = (struct stub_data *) new_stack;
477	unsigned long long new_offset;
478	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
479
480	/*
481	 * prepare offset and fd of child's stack as argument for parent's
482	 * and child's mmap2 calls
483	 */
484	*data = ((struct stub_data) {
485			.offset	= MMAP_OFFSET(new_offset),
486			.fd     = new_fd
487	});
488
489	err = ptrace_setregs(pid, thread_regs);
490	if (err < 0) {
491		err = -errno;
492		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
493		       "failed, pid = %d, errno = %d\n", pid, -err);
494		return err;
495	}
496
497	err = put_fp_registers(pid, thread_fp_regs);
498	if (err < 0) {
499		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
500		       "failed, pid = %d, err = %d\n", pid, err);
501		return err;
502	}
503
504	/* set a well known return code for detection of child write failure */
505	child_data->err = 12345678;
506
507	/*
508	 * Wait, until parent has finished its work: read child's pid from
509	 * parent's stack, and check, if bad result.
510	 */
511	err = ptrace(PTRACE_CONT, pid, 0, 0);
512	if (err) {
513		err = -errno;
514		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
515		       "errno = %d\n", pid, errno);
516		return err;
517	}
518
519	wait_stub_done(pid);
520
521	pid = data->err;
522	if (pid < 0) {
523		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
524		       "error %d\n", -pid);
525		return pid;
526	}
527
528	/*
529	 * Wait, until child has finished too: read child's result from
530	 * child's stack and check it.
531	 */
532	wait_stub_done(pid);
533	if (child_data->err != STUB_DATA) {
534		printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
535		       "error %ld\n", child_data->err);
536		err = child_data->err;
537		goto out_kill;
538	}
539
540	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
541		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
542		err = -errno;
543		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
544		       "failed, errno = %d\n", errno);
545		goto out_kill;
546	}
547
548	return pid;
549
550 out_kill:
551	os_kill_ptraced_process(pid, 1);
552	return err;
553}
554
555void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
556{
557	(*buf)[0].JB_IP = (unsigned long) handler;
558	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
559		sizeof(void *);
560}
561
562#define INIT_JMP_NEW_THREAD 0
563#define INIT_JMP_CALLBACK 1
564#define INIT_JMP_HALT 2
565#define INIT_JMP_REBOOT 3
566
567void switch_threads(jmp_buf *me, jmp_buf *you)
568{
569	if (UML_SETJMP(me) == 0)
570		UML_LONGJMP(you, 1);
571}
572
573static jmp_buf initial_jmpbuf;
574
575/* XXX Make these percpu */
576static void (*cb_proc)(void *arg);
577static void *cb_arg;
578static jmp_buf *cb_back;
579
580int start_idle_thread(void *stack, jmp_buf *switch_buf)
581{
582	int n;
583
584	set_handler(SIGWINCH);
585
586	/*
587	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
588	 * and restore signals, with the possible side-effect of
589	 * trying to handle any signals which came when they were
590	 * blocked, which can't be done on this stack.
591	 * Signals must be blocked when jumping back here and restored
592	 * after returning to the jumper.
593	 */
594	n = setjmp(initial_jmpbuf);
595	switch (n) {
596	case INIT_JMP_NEW_THREAD:
597		(*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
598		(*switch_buf)[0].JB_SP = (unsigned long) stack +
599			UM_THREAD_SIZE - sizeof(void *);
600		break;
601	case INIT_JMP_CALLBACK:
602		(*cb_proc)(cb_arg);
603		longjmp(*cb_back, 1);
604		break;
605	case INIT_JMP_HALT:
606		kmalloc_ok = 0;
607		return 0;
608	case INIT_JMP_REBOOT:
609		kmalloc_ok = 0;
610		return 1;
611	default:
612		printk(UM_KERN_ERR "Bad sigsetjmp return in "
613		       "start_idle_thread - %d\n", n);
614		fatal_sigsegv();
615	}
616	longjmp(*switch_buf, 1);
617
618	/* unreachable */
619	printk(UM_KERN_ERR "impossible long jump!");
620	fatal_sigsegv();
621	return 0;
622}
623
624void initial_thread_cb_skas(void (*proc)(void *), void *arg)
625{
626	jmp_buf here;
627
628	cb_proc = proc;
629	cb_arg = arg;
630	cb_back = &here;
631
632	block_signals_trace();
633	if (UML_SETJMP(&here) == 0)
634		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
635	unblock_signals_trace();
636
637	cb_proc = NULL;
638	cb_arg = NULL;
639	cb_back = NULL;
640}
641
642void halt_skas(void)
643{
644	block_signals_trace();
645	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
646}
647
648static bool noreboot;
649
650static int __init noreboot_cmd_param(char *str, int *add)
651{
652	noreboot = true;
653	return 0;
654}
655
656__uml_setup("noreboot", noreboot_cmd_param,
657"noreboot\n"
658"    Rather than rebooting, exit always, akin to QEMU's -no-reboot option.\n"
659"    This is useful if you're using CONFIG_PANIC_TIMEOUT in order to catch\n"
660"    crashes in CI\n");
661
662void reboot_skas(void)
663{
664	block_signals_trace();
665	UML_LONGJMP(&initial_jmpbuf, noreboot ? INIT_JMP_HALT : INIT_JMP_REBOOT);
666}
667
668void __switch_mm(struct mm_id *mm_idp)
669{
670	userspace_pid[0] = mm_idp->u.pid;
671	kill_userspace_mm[0] = mm_idp->kill;
672}
673