Linux Audio

Check our new training course

Loading...
v5.14.15
  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 <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#include <linux/threads.h>
 25
 26int is_skas_winch(int pid, int fd, void *data)
 27{
 28	return pid == getpgrp();
 29}
 30
 31static const char *ptrace_reg_name(int idx)
 32{
 33#define R(n) case HOST_##n: return #n
 34
 35	switch (idx) {
 36#ifdef __x86_64__
 37	R(BX);
 38	R(CX);
 39	R(DI);
 40	R(SI);
 41	R(DX);
 42	R(BP);
 43	R(AX);
 44	R(R8);
 45	R(R9);
 46	R(R10);
 47	R(R11);
 48	R(R12);
 49	R(R13);
 50	R(R14);
 51	R(R15);
 52	R(ORIG_AX);
 53	R(CS);
 54	R(SS);
 55	R(EFLAGS);
 56#elif defined(__i386__)
 57	R(IP);
 58	R(SP);
 59	R(EFLAGS);
 60	R(AX);
 61	R(BX);
 62	R(CX);
 63	R(DX);
 64	R(SI);
 65	R(DI);
 66	R(BP);
 67	R(CS);
 68	R(SS);
 69	R(DS);
 70	R(FS);
 71	R(ES);
 72	R(GS);
 73	R(ORIG_AX);
 74#endif
 75	}
 76	return "";
 77}
 78
 79static int ptrace_dump_regs(int pid)
 80{
 81	unsigned long regs[MAX_REG_NR];
 82	int i;
 83
 84	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
 85		return -errno;
 86
 87	printk(UM_KERN_ERR "Stub registers -\n");
 88	for (i = 0; i < ARRAY_SIZE(regs); i++) {
 89		const char *regname = ptrace_reg_name(i);
 90
 91		printk(UM_KERN_ERR "\t%s\t(%2d): %lx\n", regname, i, regs[i]);
 92	}
 93
 94	return 0;
 95}
 96
 97/*
 98 * Signals that are OK to receive in the stub - we'll just continue it.
 99 * SIGWINCH will happen when UML is inside a detached screen.
100 */
101#define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
102
103/* Signals that the stub will finish with - anything else is an error */
104#define STUB_DONE_MASK (1 << SIGTRAP)
105
106void wait_stub_done(int pid)
107{
108	int n, status, err;
109
110	while (1) {
111		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
112		if ((n < 0) || !WIFSTOPPED(status))
113			goto bad_wait;
114
115		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
116			break;
117
118		err = ptrace(PTRACE_CONT, pid, 0, 0);
119		if (err) {
120			printk(UM_KERN_ERR "wait_stub_done : continue failed, "
121			       "errno = %d\n", errno);
122			fatal_sigsegv();
123		}
124	}
125
126	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
127		return;
128
129bad_wait:
130	err = ptrace_dump_regs(pid);
131	if (err)
132		printk(UM_KERN_ERR "Failed to get registers from stub, "
133		       "errno = %d\n", -err);
134	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
135	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
136	       status);
137	fatal_sigsegv();
138}
139
140extern unsigned long current_stub_stack(void);
141
142static 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
174static 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 */
184static 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 "handle_trap - nullifying syscall "
198			       "failed, errno = %d\n", errno);
199			fatal_sigsegv();
200		}
201
202		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
203		if (err < 0) {
204			printk(UM_KERN_ERR "handle_trap - continuing to end of "
205			       "syscall failed, errno = %d\n", 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 "
215				       "from process, errno = %d\n", -err);
216			printk(UM_KERN_ERR "handle_trap - failed to wait at "
217			       "end of syscall, errno = %d, status = %d\n",
218			       errno, status);
219			fatal_sigsegv();
220		}
221	}
222
223	handle_syscall(regs);
224}
225
226extern char __syscall_stub_start[];
227
228/**
229 * userspace_tramp() - userspace trampoline
230 * @stack:	pointer to the new userspace stack page, can be NULL, if? FIXME:
231 *
232 * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
233 * This function will run on a temporary stack page.
234 * It ptrace()'es itself, then
235 * Two pages are mapped into the userspace address space:
236 * - STUB_CODE (with EXEC), which contains the skas stub code
237 * - 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.
238 * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
239 * And last the process stops itself to give control to the UML kernel for this userspace process.
240 *
241 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
242 */
243static int userspace_tramp(void *stack)
244{
245	void *addr;
246	int fd;
247	unsigned long long offset;
248
249	ptrace(PTRACE_TRACEME, 0, 0, 0);
250
251	signal(SIGTERM, SIG_DFL);
252	signal(SIGWINCH, SIG_IGN);
253
 
 
 
 
254	fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
255	addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
256		      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
257	if (addr == MAP_FAILED) {
258		printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
259		       "errno = %d\n", STUB_CODE, errno);
260		exit(1);
261	}
262
263	if (stack != NULL) {
264		fd = phys_mapping(to_phys(stack), &offset);
265		addr = mmap((void *) STUB_DATA,
266			    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
267			    MAP_FIXED | MAP_SHARED, fd, offset);
268		if (addr == MAP_FAILED) {
269			printk(UM_KERN_ERR "mapping segfault stack "
270			       "at 0x%lx failed, errno = %d\n",
271			       STUB_DATA, errno);
272			exit(1);
273		}
274	}
275	if (stack != NULL) {
276		struct sigaction sa;
277
278		unsigned long v = STUB_CODE +
279				  (unsigned long) stub_segv_handler -
280				  (unsigned long) __syscall_stub_start;
281
282		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
283		sigemptyset(&sa.sa_mask);
284		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
285		sa.sa_sigaction = (void *) v;
286		sa.sa_restorer = NULL;
287		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
288			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
289			       "handler failed - errno = %d\n", errno);
290			exit(1);
291		}
292	}
293
294	kill(os_getpid(), SIGSTOP);
295	return 0;
296}
297
298int userspace_pid[NR_CPUS];
299int kill_userspace_mm[NR_CPUS];
300
301/**
302 * start_userspace() - prepare a new userspace process
303 * @stub_stack:	pointer to the stub stack. Can be NULL, if? FIXME:
304 *
305 * Setups a new temporary stack page that is used while userspace_tramp() runs
306 * Clones the kernel process into a new userspace process, with FDs only.
307 *
308 * Return: When positive: the process id of the new userspace process,
309 *         when negative: an error number.
310 * FIXME: can PIDs become negative?!
311 */
312int start_userspace(unsigned long stub_stack)
313{
314	void *stack;
315	unsigned long sp;
316	int pid, status, n, flags, err;
317
318	/* setup a temporary stack page */
319	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
320		     PROT_READ | PROT_WRITE | PROT_EXEC,
321		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
322	if (stack == MAP_FAILED) {
323		err = -errno;
324		printk(UM_KERN_ERR "start_userspace : mmap failed, "
325		       "errno = %d\n", errno);
326		return err;
327	}
328
329	/* set stack pointer to the end of the stack page, so it can grow downwards */
330	sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
331
332	flags = CLONE_FILES | SIGCHLD;
333
334	/* clone into new userspace process */
335	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
336	if (pid < 0) {
337		err = -errno;
338		printk(UM_KERN_ERR "start_userspace : clone failed, "
339		       "errno = %d\n", errno);
340		return err;
341	}
342
343	do {
344		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
345		if (n < 0) {
346			err = -errno;
347			printk(UM_KERN_ERR "start_userspace : wait failed, "
348			       "errno = %d\n", errno);
349			goto out_kill;
350		}
351	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
352
353	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
354		err = -EINVAL;
355		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
356		       "status = %d\n", status);
357		goto out_kill;
358	}
359
360	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
361		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
362		err = -errno;
363		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
364		       "failed, errno = %d\n", errno);
365		goto out_kill;
366	}
367
368	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
369		err = -errno;
370		printk(UM_KERN_ERR "start_userspace : munmap failed, "
371		       "errno = %d\n", errno);
372		goto out_kill;
373	}
374
375	return pid;
376
377 out_kill:
378	os_kill_ptraced_process(pid, 1);
379	return err;
380}
381
382void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
383{
384	int err, status, op, pid = userspace_pid[0];
385	/* To prevent races if using_sysemu changes under us.*/
386	int local_using_sysemu;
387	siginfo_t si;
388
389	/* Handle any immediate reschedules or signals */
390	interrupt_end();
391
392	while (1) {
393		if (kill_userspace_mm[0])
394			fatal_sigsegv();
395
396		/*
397		 * This can legitimately fail if the process loads a
398		 * bogus value into a segment register.  It will
399		 * segfault and PTRACE_GETREGS will read that value
400		 * out of the process.  However, PTRACE_SETREGS will
401		 * fail.  In this case, there is nothing to do but
402		 * just kill the process.
403		 */
404		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
405			printk(UM_KERN_ERR "userspace - ptrace set regs "
406			       "failed, errno = %d\n", errno);
407			fatal_sigsegv();
408		}
409
410		if (put_fp_registers(pid, regs->fp)) {
411			printk(UM_KERN_ERR "userspace - ptrace set fp regs "
412			       "failed, errno = %d\n", errno);
413			fatal_sigsegv();
414		}
415
416		/* Now we set local_using_sysemu to be used for one loop */
417		local_using_sysemu = get_using_sysemu();
418
419		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
420					     singlestepping(NULL));
421
422		if (ptrace(op, pid, 0, 0)) {
423			printk(UM_KERN_ERR "userspace - ptrace continue "
424			       "failed, op = %d, errno = %d\n", op, errno);
425			fatal_sigsegv();
426		}
427
428		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
429		if (err < 0) {
430			printk(UM_KERN_ERR "userspace - wait failed, "
431			       "errno = %d\n", errno);
432			fatal_sigsegv();
433		}
434
435		regs->is_user = 1;
436		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
437			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
438			       "errno = %d\n", errno);
439			fatal_sigsegv();
440		}
441
442		if (get_fp_registers(pid, regs->fp)) {
443			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
444			       "errno = %d\n", errno);
445			fatal_sigsegv();
446		}
447
448		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
449
450		if (WIFSTOPPED(status)) {
451			int sig = WSTOPSIG(status);
452
453			/* These signal handlers need the si argument.
454			 * The SIGIO and SIGALARM handlers which constitute the
455			 * majority of invocations, do not use it.
456			 */
457			switch (sig) {
458			case SIGSEGV:
459			case SIGTRAP:
460			case SIGILL:
461			case SIGBUS:
462			case SIGFPE:
463			case SIGWINCH:
464				ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
465				break;
466			}
467
468			switch (sig) {
469			case SIGSEGV:
470				if (PTRACE_FULL_FAULTINFO) {
471					get_skas_faultinfo(pid,
472							   &regs->faultinfo, aux_fp_regs);
473					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
474							     regs);
475				}
476				else handle_segv(pid, regs, aux_fp_regs);
477				break;
478			case SIGTRAP + 0x80:
479			        handle_trap(pid, regs, local_using_sysemu);
480				break;
481			case SIGTRAP:
482				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
483				break;
484			case SIGALRM:
485				break;
486			case SIGIO:
487			case SIGILL:
488			case SIGBUS:
489			case SIGFPE:
490			case SIGWINCH:
491				block_signals_trace();
492				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
493				unblock_signals_trace();
494				break;
495			default:
496				printk(UM_KERN_ERR "userspace - child stopped "
497				       "with signal %d\n", sig);
498				fatal_sigsegv();
499			}
500			pid = userspace_pid[0];
501			interrupt_end();
502
503			/* Avoid -ERESTARTSYS handling in host */
504			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
505				PT_SYSCALL_NR(regs->gp) = -1;
506		}
507	}
508}
509
510static unsigned long thread_regs[MAX_REG_NR];
511static unsigned long thread_fp_regs[FP_SIZE];
512
513static int __init init_thread_regs(void)
514{
515	get_safe_registers(thread_regs, thread_fp_regs);
516	/* Set parent's instruction pointer to start of clone-stub */
517	thread_regs[REGS_IP_INDEX] = STUB_CODE +
518				(unsigned long) stub_clone_handler -
519				(unsigned long) __syscall_stub_start;
520	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
521		sizeof(void *);
522#ifdef __SIGNAL_FRAMESIZE
523	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
524#endif
525	return 0;
526}
527
528__initcall(init_thread_regs);
529
530int copy_context_skas0(unsigned long new_stack, int pid)
531{
532	int err;
533	unsigned long current_stack = current_stub_stack();
534	struct stub_data *data = (struct stub_data *) current_stack;
535	struct stub_data *child_data = (struct stub_data *) new_stack;
536	unsigned long long new_offset;
537	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
538
539	/*
540	 * prepare offset and fd of child's stack as argument for parent's
541	 * and child's mmap2 calls
542	 */
543	*data = ((struct stub_data) {
544		.offset	= MMAP_OFFSET(new_offset),
545		.fd     = new_fd,
546		.parent_err = -ESRCH,
547		.child_err = 0,
548	});
549
550	*child_data = ((struct stub_data) {
551		.child_err = -ESRCH,
552	});
553
554	err = ptrace_setregs(pid, thread_regs);
555	if (err < 0) {
556		err = -errno;
557		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
558		       "failed, pid = %d, errno = %d\n", pid, -err);
559		return err;
560	}
561
562	err = put_fp_registers(pid, thread_fp_regs);
563	if (err < 0) {
564		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
565		       "failed, pid = %d, err = %d\n", pid, err);
566		return err;
567	}
568
 
 
 
569	/*
570	 * Wait, until parent has finished its work: read child's pid from
571	 * parent's stack, and check, if bad result.
572	 */
573	err = ptrace(PTRACE_CONT, pid, 0, 0);
574	if (err) {
575		err = -errno;
576		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
577		       "errno = %d\n", pid, errno);
578		return err;
579	}
580
581	wait_stub_done(pid);
582
583	pid = data->parent_err;
584	if (pid < 0) {
585		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
586		       "error %d\n", -pid);
587		return pid;
588	}
589
590	/*
591	 * Wait, until child has finished too: read child's result from
592	 * child's stack and check it.
593	 */
594	wait_stub_done(pid);
595	if (child_data->child_err != STUB_DATA) {
596		printk(UM_KERN_ERR "copy_context_skas0 - stub-child %d reports "
597		       "error %ld\n", pid, data->child_err);
598		err = data->child_err;
599		goto out_kill;
600	}
601
602	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
603		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
604		err = -errno;
605		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
606		       "failed, errno = %d\n", errno);
607		goto out_kill;
608	}
609
610	return pid;
611
612 out_kill:
613	os_kill_ptraced_process(pid, 1);
614	return err;
615}
616
617void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
618{
619	(*buf)[0].JB_IP = (unsigned long) handler;
620	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
621		sizeof(void *);
622}
623
624#define INIT_JMP_NEW_THREAD 0
625#define INIT_JMP_CALLBACK 1
626#define INIT_JMP_HALT 2
627#define INIT_JMP_REBOOT 3
628
629void switch_threads(jmp_buf *me, jmp_buf *you)
630{
631	if (UML_SETJMP(me) == 0)
632		UML_LONGJMP(you, 1);
633}
634
635static jmp_buf initial_jmpbuf;
636
637/* XXX Make these percpu */
638static void (*cb_proc)(void *arg);
639static void *cb_arg;
640static jmp_buf *cb_back;
641
642int start_idle_thread(void *stack, jmp_buf *switch_buf)
643{
644	int n;
645
646	set_handler(SIGWINCH);
647
648	/*
649	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
650	 * and restore signals, with the possible side-effect of
651	 * trying to handle any signals which came when they were
652	 * blocked, which can't be done on this stack.
653	 * Signals must be blocked when jumping back here and restored
654	 * after returning to the jumper.
655	 */
656	n = setjmp(initial_jmpbuf);
657	switch (n) {
658	case INIT_JMP_NEW_THREAD:
659		(*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
660		(*switch_buf)[0].JB_SP = (unsigned long) stack +
661			UM_THREAD_SIZE - sizeof(void *);
662		break;
663	case INIT_JMP_CALLBACK:
664		(*cb_proc)(cb_arg);
665		longjmp(*cb_back, 1);
666		break;
667	case INIT_JMP_HALT:
668		kmalloc_ok = 0;
669		return 0;
670	case INIT_JMP_REBOOT:
671		kmalloc_ok = 0;
672		return 1;
673	default:
674		printk(UM_KERN_ERR "Bad sigsetjmp return in "
675		       "start_idle_thread - %d\n", n);
676		fatal_sigsegv();
677	}
678	longjmp(*switch_buf, 1);
679
680	/* unreachable */
681	printk(UM_KERN_ERR "impossible long jump!");
682	fatal_sigsegv();
683	return 0;
684}
685
686void initial_thread_cb_skas(void (*proc)(void *), void *arg)
687{
688	jmp_buf here;
689
690	cb_proc = proc;
691	cb_arg = arg;
692	cb_back = &here;
693
694	block_signals_trace();
695	if (UML_SETJMP(&here) == 0)
696		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
697	unblock_signals_trace();
698
699	cb_proc = NULL;
700	cb_arg = NULL;
701	cb_back = NULL;
702}
703
704void halt_skas(void)
705{
706	block_signals_trace();
707	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
708}
709
710void reboot_skas(void)
711{
712	block_signals_trace();
713	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
714}
715
716void __switch_mm(struct mm_id *mm_idp)
717{
718	userspace_pid[0] = mm_idp->u.pid;
719	kill_userspace_mm[0] = mm_idp->kill;
720}
v4.17
 
  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#include <linux/threads.h>
 25
 26int is_skas_winch(int pid, int fd, void *data)
 27{
 28	return pid == getpgrp();
 29}
 30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 31static int ptrace_dump_regs(int pid)
 32{
 33	unsigned long regs[MAX_REG_NR];
 34	int i;
 35
 36	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
 37		return -errno;
 38
 39	printk(UM_KERN_ERR "Stub registers -\n");
 40	for (i = 0; i < ARRAY_SIZE(regs); i++)
 41		printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
 
 
 
 42
 43	return 0;
 44}
 45
 46/*
 47 * Signals that are OK to receive in the stub - we'll just continue it.
 48 * SIGWINCH will happen when UML is inside a detached screen.
 49 */
 50#define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
 51
 52/* Signals that the stub will finish with - anything else is an error */
 53#define STUB_DONE_MASK (1 << SIGTRAP)
 54
 55void wait_stub_done(int pid)
 56{
 57	int n, status, err;
 58
 59	while (1) {
 60		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
 61		if ((n < 0) || !WIFSTOPPED(status))
 62			goto bad_wait;
 63
 64		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
 65			break;
 66
 67		err = ptrace(PTRACE_CONT, pid, 0, 0);
 68		if (err) {
 69			printk(UM_KERN_ERR "wait_stub_done : continue failed, "
 70			       "errno = %d\n", errno);
 71			fatal_sigsegv();
 72		}
 73	}
 74
 75	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
 76		return;
 77
 78bad_wait:
 79	err = ptrace_dump_regs(pid);
 80	if (err)
 81		printk(UM_KERN_ERR "Failed to get registers from stub, "
 82		       "errno = %d\n", -err);
 83	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
 84	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
 85	       status);
 86	fatal_sigsegv();
 87}
 88
 89extern unsigned long current_stub_stack(void);
 90
 91static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
 92{
 93	int err;
 94
 95	err = get_fp_registers(pid, aux_fp_regs);
 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, aux_fp_regs);
116	if (err < 0) {
117		printk(UM_KERN_ERR "put_fp_registers returned %d\n",
118		       err);
119		fatal_sigsegv();
120	}
121}
122
123static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
124{
125	get_skas_faultinfo(pid, &regs->faultinfo, aux_fp_regs);
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 */
133static 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
175extern char __syscall_stub_start[];
176
177/**
178 * userspace_tramp() - userspace trampoline
179 * @stack:	pointer to the new userspace stack page, can be NULL, if? FIXME:
180 *
181 * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
182 * This function will run on a temporary stack page.
183 * It ptrace()'es itself, then
184 * Two pages are mapped into the userspace address space:
185 * - STUB_CODE (with EXEC), which contains the skas stub code
186 * - 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.
187 * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
188 * And last the process stops itself to give control to the UML kernel for this userspace process.
189 *
190 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
191 */
192static int userspace_tramp(void *stack)
193{
194	void *addr;
195	int fd;
196	unsigned long long offset;
197
198	ptrace(PTRACE_TRACEME, 0, 0, 0);
199
200	signal(SIGTERM, SIG_DFL);
201	signal(SIGWINCH, SIG_IGN);
202
203	/*
204	 * This has a pte, but it can't be mapped in with the usual
205	 * tlb_flush mechanism because this is part of that mechanism
206	 */
207	fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
208	addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
209		      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
210	if (addr == MAP_FAILED) {
211		printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
212		       "errno = %d\n", STUB_CODE, errno);
213		exit(1);
214	}
215
216	if (stack != NULL) {
217		fd = phys_mapping(to_phys(stack), &offset);
218		addr = mmap((void *) STUB_DATA,
219			    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
220			    MAP_FIXED | MAP_SHARED, fd, offset);
221		if (addr == MAP_FAILED) {
222			printk(UM_KERN_ERR "mapping segfault stack "
223			       "at 0x%lx failed, errno = %d\n",
224			       STUB_DATA, errno);
225			exit(1);
226		}
227	}
228	if (stack != NULL) {
229		struct sigaction sa;
230
231		unsigned long v = STUB_CODE +
232				  (unsigned long) stub_segv_handler -
233				  (unsigned long) __syscall_stub_start;
234
235		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
236		sigemptyset(&sa.sa_mask);
237		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
238		sa.sa_sigaction = (void *) v;
239		sa.sa_restorer = NULL;
240		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
241			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
242			       "handler failed - errno = %d\n", errno);
243			exit(1);
244		}
245	}
246
247	kill(os_getpid(), SIGSTOP);
248	return 0;
249}
250
251int userspace_pid[NR_CPUS];
 
252
253/**
254 * start_userspace() - prepare a new userspace process
255 * @stub_stack:	pointer to the stub stack. Can be NULL, if? FIXME:
256 *
257 * Setups a new temporary stack page that is used while userspace_tramp() runs
258 * Clones the kernel process into a new userspace process, with FDs only.
259 *
260 * Return: When positive: the process id of the new userspace process,
261 *         when negative: an error number.
262 * FIXME: can PIDs become negative?!
263 */
264int start_userspace(unsigned long stub_stack)
265{
266	void *stack;
267	unsigned long sp;
268	int pid, status, n, flags, err;
269
270	/* setup a temporary stack page */
271	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
272		     PROT_READ | PROT_WRITE | PROT_EXEC,
273		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
274	if (stack == MAP_FAILED) {
275		err = -errno;
276		printk(UM_KERN_ERR "start_userspace : mmap failed, "
277		       "errno = %d\n", errno);
278		return err;
279	}
280
281	/* set stack pointer to the end of the stack page, so it can grow downwards */
282	sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
283
284	flags = CLONE_FILES | SIGCHLD;
285
286	/* clone into new userspace process */
287	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
288	if (pid < 0) {
289		err = -errno;
290		printk(UM_KERN_ERR "start_userspace : clone failed, "
291		       "errno = %d\n", errno);
292		return err;
293	}
294
295	do {
296		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
297		if (n < 0) {
298			err = -errno;
299			printk(UM_KERN_ERR "start_userspace : wait failed, "
300			       "errno = %d\n", errno);
301			goto out_kill;
302		}
303	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
304
305	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
306		err = -EINVAL;
307		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
308		       "status = %d\n", status);
309		goto out_kill;
310	}
311
312	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
313		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
314		err = -errno;
315		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
316		       "failed, errno = %d\n", errno);
317		goto out_kill;
318	}
319
320	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
321		err = -errno;
322		printk(UM_KERN_ERR "start_userspace : munmap failed, "
323		       "errno = %d\n", errno);
324		goto out_kill;
325	}
326
327	return pid;
328
329 out_kill:
330	os_kill_ptraced_process(pid, 1);
331	return err;
332}
333
334void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
335{
336	int err, status, op, pid = userspace_pid[0];
337	/* To prevent races if using_sysemu changes under us.*/
338	int local_using_sysemu;
339	siginfo_t si;
340
341	/* Handle any immediate reschedules or signals */
342	interrupt_end();
343
344	while (1) {
 
 
345
346		/*
347		 * This can legitimately fail if the process loads a
348		 * bogus value into a segment register.  It will
349		 * segfault and PTRACE_GETREGS will read that value
350		 * out of the process.  However, PTRACE_SETREGS will
351		 * fail.  In this case, there is nothing to do but
352		 * just kill the process.
353		 */
354		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
355			printk(UM_KERN_ERR "userspace - ptrace set regs "
356			       "failed, errno = %d\n", errno);
357			fatal_sigsegv();
358		}
359
360		if (put_fp_registers(pid, regs->fp)) {
361			printk(UM_KERN_ERR "userspace - ptrace set fp regs "
362			       "failed, errno = %d\n", errno);
363			fatal_sigsegv();
364		}
365
366		/* Now we set local_using_sysemu to be used for one loop */
367		local_using_sysemu = get_using_sysemu();
368
369		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
370					     singlestepping(NULL));
371
372		if (ptrace(op, pid, 0, 0)) {
373			printk(UM_KERN_ERR "userspace - ptrace continue "
374			       "failed, op = %d, errno = %d\n", op, errno);
375			fatal_sigsegv();
376		}
377
378		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
379		if (err < 0) {
380			printk(UM_KERN_ERR "userspace - wait failed, "
381			       "errno = %d\n", errno);
382			fatal_sigsegv();
383		}
384
385		regs->is_user = 1;
386		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
387			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
388			       "errno = %d\n", errno);
389			fatal_sigsegv();
390		}
391
392		if (get_fp_registers(pid, regs->fp)) {
393			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
394			       "errno = %d\n", errno);
395			fatal_sigsegv();
396		}
397
398		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
399
400		if (WIFSTOPPED(status)) {
401			int sig = WSTOPSIG(status);
402
403			ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
 
 
 
 
 
 
 
 
 
 
 
 
 
404
405			switch (sig) {
406			case SIGSEGV:
407				if (PTRACE_FULL_FAULTINFO) {
408					get_skas_faultinfo(pid,
409							   &regs->faultinfo, aux_fp_regs);
410					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
411							     regs);
412				}
413				else handle_segv(pid, regs, aux_fp_regs);
414				break;
415			case SIGTRAP + 0x80:
416			        handle_trap(pid, regs, local_using_sysemu);
417				break;
418			case SIGTRAP:
419				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
420				break;
421			case SIGALRM:
422				break;
423			case SIGIO:
424			case SIGILL:
425			case SIGBUS:
426			case SIGFPE:
427			case SIGWINCH:
428				block_signals();
429				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
430				unblock_signals();
431				break;
432			default:
433				printk(UM_KERN_ERR "userspace - child stopped "
434				       "with signal %d\n", sig);
435				fatal_sigsegv();
436			}
437			pid = userspace_pid[0];
438			interrupt_end();
439
440			/* Avoid -ERESTARTSYS handling in host */
441			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
442				PT_SYSCALL_NR(regs->gp) = -1;
443		}
444	}
445}
446
447static unsigned long thread_regs[MAX_REG_NR];
448static unsigned long thread_fp_regs[FP_SIZE];
449
450static int __init init_thread_regs(void)
451{
452	get_safe_registers(thread_regs, thread_fp_regs);
453	/* Set parent's instruction pointer to start of clone-stub */
454	thread_regs[REGS_IP_INDEX] = STUB_CODE +
455				(unsigned long) stub_clone_handler -
456				(unsigned long) __syscall_stub_start;
457	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
458		sizeof(void *);
459#ifdef __SIGNAL_FRAMESIZE
460	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
461#endif
462	return 0;
463}
464
465__initcall(init_thread_regs);
466
467int copy_context_skas0(unsigned long new_stack, int pid)
468{
469	int err;
470	unsigned long current_stack = current_stub_stack();
471	struct stub_data *data = (struct stub_data *) current_stack;
472	struct stub_data *child_data = (struct stub_data *) new_stack;
473	unsigned long long new_offset;
474	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
475
476	/*
477	 * prepare offset and fd of child's stack as argument for parent's
478	 * and child's mmap2 calls
479	 */
480	*data = ((struct stub_data) {
481			.offset	= MMAP_OFFSET(new_offset),
482			.fd     = new_fd
 
 
 
 
 
 
483	});
484
485	err = ptrace_setregs(pid, thread_regs);
486	if (err < 0) {
487		err = -errno;
488		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
489		       "failed, pid = %d, errno = %d\n", pid, -err);
490		return err;
491	}
492
493	err = put_fp_registers(pid, thread_fp_regs);
494	if (err < 0) {
495		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
496		       "failed, pid = %d, err = %d\n", pid, err);
497		return err;
498	}
499
500	/* set a well known return code for detection of child write failure */
501	child_data->err = 12345678;
502
503	/*
504	 * Wait, until parent has finished its work: read child's pid from
505	 * parent's stack, and check, if bad result.
506	 */
507	err = ptrace(PTRACE_CONT, pid, 0, 0);
508	if (err) {
509		err = -errno;
510		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
511		       "errno = %d\n", pid, errno);
512		return err;
513	}
514
515	wait_stub_done(pid);
516
517	pid = data->err;
518	if (pid < 0) {
519		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
520		       "error %d\n", -pid);
521		return pid;
522	}
523
524	/*
525	 * Wait, until child has finished too: read child's result from
526	 * child's stack and check it.
527	 */
528	wait_stub_done(pid);
529	if (child_data->err != STUB_DATA) {
530		printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
531		       "error %ld\n", child_data->err);
532		err = child_data->err;
533		goto out_kill;
534	}
535
536	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
537		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
538		err = -errno;
539		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
540		       "failed, errno = %d\n", errno);
541		goto out_kill;
542	}
543
544	return pid;
545
546 out_kill:
547	os_kill_ptraced_process(pid, 1);
548	return err;
549}
550
551void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
552{
553	(*buf)[0].JB_IP = (unsigned long) handler;
554	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
555		sizeof(void *);
556}
557
558#define INIT_JMP_NEW_THREAD 0
559#define INIT_JMP_CALLBACK 1
560#define INIT_JMP_HALT 2
561#define INIT_JMP_REBOOT 3
562
563void switch_threads(jmp_buf *me, jmp_buf *you)
564{
565	if (UML_SETJMP(me) == 0)
566		UML_LONGJMP(you, 1);
567}
568
569static jmp_buf initial_jmpbuf;
570
571/* XXX Make these percpu */
572static void (*cb_proc)(void *arg);
573static void *cb_arg;
574static jmp_buf *cb_back;
575
576int start_idle_thread(void *stack, jmp_buf *switch_buf)
577{
578	int n;
579
580	set_handler(SIGWINCH);
581
582	/*
583	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
584	 * and restore signals, with the possible side-effect of
585	 * trying to handle any signals which came when they were
586	 * blocked, which can't be done on this stack.
587	 * Signals must be blocked when jumping back here and restored
588	 * after returning to the jumper.
589	 */
590	n = setjmp(initial_jmpbuf);
591	switch (n) {
592	case INIT_JMP_NEW_THREAD:
593		(*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
594		(*switch_buf)[0].JB_SP = (unsigned long) stack +
595			UM_THREAD_SIZE - sizeof(void *);
596		break;
597	case INIT_JMP_CALLBACK:
598		(*cb_proc)(cb_arg);
599		longjmp(*cb_back, 1);
600		break;
601	case INIT_JMP_HALT:
602		kmalloc_ok = 0;
603		return 0;
604	case INIT_JMP_REBOOT:
605		kmalloc_ok = 0;
606		return 1;
607	default:
608		printk(UM_KERN_ERR "Bad sigsetjmp return in "
609		       "start_idle_thread - %d\n", n);
610		fatal_sigsegv();
611	}
612	longjmp(*switch_buf, 1);
 
 
 
 
 
613}
614
615void initial_thread_cb_skas(void (*proc)(void *), void *arg)
616{
617	jmp_buf here;
618
619	cb_proc = proc;
620	cb_arg = arg;
621	cb_back = &here;
622
623	block_signals();
624	if (UML_SETJMP(&here) == 0)
625		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
626	unblock_signals();
627
628	cb_proc = NULL;
629	cb_arg = NULL;
630	cb_back = NULL;
631}
632
633void halt_skas(void)
634{
635	block_signals();
636	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
637}
638
639void reboot_skas(void)
640{
641	block_signals();
642	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
643}
644
645void __switch_mm(struct mm_id *mm_idp)
646{
647	userspace_pid[0] = mm_idp->u.pid;
 
648}