Loading...
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 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
80static 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
107void 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 "wait_stub_done : continue failed, "
122 "errno = %d\n", errno);
123 fatal_sigsegv();
124 }
125 }
126
127 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
128 return;
129
130bad_wait:
131 err = ptrace_dump_regs(pid);
132 if (err)
133 printk(UM_KERN_ERR "Failed to get registers from stub, "
134 "errno = %d\n", -err);
135 printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
136 "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
137 status);
138 fatal_sigsegv();
139}
140
141extern unsigned long current_stub_stack(void);
142
143static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
144{
145 int err;
146
147 err = get_fp_registers(pid, aux_fp_regs);
148 if (err < 0) {
149 printk(UM_KERN_ERR "save_fp_registers returned %d\n",
150 err);
151 fatal_sigsegv();
152 }
153 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
154 if (err) {
155 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
156 "errno = %d\n", pid, errno);
157 fatal_sigsegv();
158 }
159 wait_stub_done(pid);
160
161 /*
162 * faultinfo is prepared by the stub_segv_handler at start of
163 * the stub stack page. We just have to copy it.
164 */
165 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
166
167 err = put_fp_registers(pid, aux_fp_regs);
168 if (err < 0) {
169 printk(UM_KERN_ERR "put_fp_registers returned %d\n",
170 err);
171 fatal_sigsegv();
172 }
173}
174
175static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
176{
177 get_skas_faultinfo(pid, ®s->faultinfo, aux_fp_regs);
178 segv(regs->faultinfo, 0, 1, NULL);
179}
180
181/*
182 * To use the same value of using_sysemu as the caller, ask it that value
183 * (in local_using_sysemu
184 */
185static void handle_trap(int pid, struct uml_pt_regs *regs,
186 int local_using_sysemu)
187{
188 int err, status;
189
190 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
191 fatal_sigsegv();
192
193 if (!local_using_sysemu)
194 {
195 err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
196 __NR_getpid);
197 if (err < 0) {
198 printk(UM_KERN_ERR "handle_trap - nullifying syscall "
199 "failed, errno = %d\n", errno);
200 fatal_sigsegv();
201 }
202
203 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
204 if (err < 0) {
205 printk(UM_KERN_ERR "handle_trap - continuing to end of "
206 "syscall failed, errno = %d\n", errno);
207 fatal_sigsegv();
208 }
209
210 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
211 if ((err < 0) || !WIFSTOPPED(status) ||
212 (WSTOPSIG(status) != SIGTRAP + 0x80)) {
213 err = ptrace_dump_regs(pid);
214 if (err)
215 printk(UM_KERN_ERR "Failed to get registers "
216 "from process, errno = %d\n", -err);
217 printk(UM_KERN_ERR "handle_trap - failed to wait at "
218 "end of syscall, errno = %d, status = %d\n",
219 errno, status);
220 fatal_sigsegv();
221 }
222 }
223
224 handle_syscall(regs);
225}
226
227extern char __syscall_stub_start[];
228
229/**
230 * userspace_tramp() - userspace trampoline
231 * @stack: pointer to the new userspace stack page, can be NULL, if? FIXME:
232 *
233 * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
234 * This function will run on a temporary stack page.
235 * It ptrace()'es itself, then
236 * Two pages are mapped into the userspace address space:
237 * - STUB_CODE (with EXEC), which contains the skas stub code
238 * - 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.
239 * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
240 * And last the process stops itself to give control to the UML kernel for this userspace process.
241 *
242 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
243 */
244static int userspace_tramp(void *stack)
245{
246 void *addr;
247 int fd;
248 unsigned long long offset;
249
250 ptrace(PTRACE_TRACEME, 0, 0, 0);
251
252 signal(SIGTERM, SIG_DFL);
253 signal(SIGWINCH, SIG_IGN);
254
255 fd = phys_mapping(uml_to_phys(__syscall_stub_start), &offset);
256 addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
257 PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
258 if (addr == MAP_FAILED) {
259 printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
260 "errno = %d\n", STUB_CODE, errno);
261 exit(1);
262 }
263
264 if (stack != NULL) {
265 fd = phys_mapping(uml_to_phys(stack), &offset);
266 addr = mmap((void *) STUB_DATA,
267 UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
268 MAP_FIXED | MAP_SHARED, fd, offset);
269 if (addr == MAP_FAILED) {
270 printk(UM_KERN_ERR "mapping segfault stack "
271 "at 0x%lx failed, errno = %d\n",
272 STUB_DATA, errno);
273 exit(1);
274 }
275 }
276 if (stack != NULL) {
277 struct sigaction sa;
278
279 unsigned long v = STUB_CODE +
280 (unsigned long) stub_segv_handler -
281 (unsigned long) __syscall_stub_start;
282
283 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
284 sigemptyset(&sa.sa_mask);
285 sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
286 sa.sa_sigaction = (void *) v;
287 sa.sa_restorer = NULL;
288 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
289 printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
290 "handler failed - errno = %d\n", errno);
291 exit(1);
292 }
293 }
294
295 kill(os_getpid(), SIGSTOP);
296 return 0;
297}
298
299int userspace_pid[NR_CPUS];
300int kill_userspace_mm[NR_CPUS];
301
302/**
303 * start_userspace() - prepare a new userspace process
304 * @stub_stack: pointer to the stub stack. Can be NULL, if? FIXME:
305 *
306 * Setups a new temporary stack page that is used while userspace_tramp() runs
307 * Clones the kernel process into a new userspace process, with FDs only.
308 *
309 * Return: When positive: the process id of the new userspace process,
310 * when negative: an error number.
311 * FIXME: can PIDs become negative?!
312 */
313int start_userspace(unsigned long stub_stack)
314{
315 void *stack;
316 unsigned long sp;
317 int pid, status, n, flags, err;
318
319 /* setup a temporary stack page */
320 stack = mmap(NULL, UM_KERN_PAGE_SIZE,
321 PROT_READ | PROT_WRITE | PROT_EXEC,
322 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
323 if (stack == MAP_FAILED) {
324 err = -errno;
325 printk(UM_KERN_ERR "start_userspace : mmap failed, "
326 "errno = %d\n", errno);
327 return err;
328 }
329
330 /* set stack pointer to the end of the stack page, so it can grow downwards */
331 sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
332
333 flags = CLONE_FILES | SIGCHLD;
334
335 /* clone into new userspace process */
336 pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
337 if (pid < 0) {
338 err = -errno;
339 printk(UM_KERN_ERR "start_userspace : clone failed, "
340 "errno = %d\n", errno);
341 return err;
342 }
343
344 do {
345 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
346 if (n < 0) {
347 err = -errno;
348 printk(UM_KERN_ERR "start_userspace : wait failed, "
349 "errno = %d\n", errno);
350 goto out_kill;
351 }
352 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
353
354 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
355 err = -EINVAL;
356 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
357 "status = %d\n", status);
358 goto out_kill;
359 }
360
361 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
362 (void *) PTRACE_O_TRACESYSGOOD) < 0) {
363 err = -errno;
364 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
365 "failed, errno = %d\n", errno);
366 goto out_kill;
367 }
368
369 if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
370 err = -errno;
371 printk(UM_KERN_ERR "start_userspace : munmap failed, "
372 "errno = %d\n", errno);
373 goto out_kill;
374 }
375
376 return pid;
377
378 out_kill:
379 os_kill_ptraced_process(pid, 1);
380 return err;
381}
382
383void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
384{
385 int err, status, op, pid = userspace_pid[0];
386 /* To prevent races if using_sysemu changes under us.*/
387 int local_using_sysemu;
388 siginfo_t si;
389
390 /* Handle any immediate reschedules or signals */
391 interrupt_end();
392
393 while (1) {
394 if (kill_userspace_mm[0])
395 fatal_sigsegv();
396
397 /*
398 * This can legitimately fail if the process loads a
399 * bogus value into a segment register. It will
400 * segfault and PTRACE_GETREGS will read that value
401 * out of the process. However, PTRACE_SETREGS will
402 * fail. In this case, there is nothing to do but
403 * just kill the process.
404 */
405 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
406 printk(UM_KERN_ERR "userspace - ptrace set regs "
407 "failed, errno = %d\n", errno);
408 fatal_sigsegv();
409 }
410
411 if (put_fp_registers(pid, regs->fp)) {
412 printk(UM_KERN_ERR "userspace - ptrace set fp regs "
413 "failed, errno = %d\n", errno);
414 fatal_sigsegv();
415 }
416
417 /* Now we set local_using_sysemu to be used for one loop */
418 local_using_sysemu = get_using_sysemu();
419
420 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
421 singlestepping(NULL));
422
423 if (ptrace(op, pid, 0, 0)) {
424 printk(UM_KERN_ERR "userspace - ptrace continue "
425 "failed, op = %d, errno = %d\n", op, errno);
426 fatal_sigsegv();
427 }
428
429 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
430 if (err < 0) {
431 printk(UM_KERN_ERR "userspace - wait failed, "
432 "errno = %d\n", errno);
433 fatal_sigsegv();
434 }
435
436 regs->is_user = 1;
437 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
438 printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
439 "errno = %d\n", errno);
440 fatal_sigsegv();
441 }
442
443 if (get_fp_registers(pid, regs->fp)) {
444 printk(UM_KERN_ERR "userspace - get_fp_registers failed, "
445 "errno = %d\n", errno);
446 fatal_sigsegv();
447 }
448
449 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
450
451 if (WIFSTOPPED(status)) {
452 int sig = WSTOPSIG(status);
453
454 /* These signal handlers need the si argument.
455 * The SIGIO and SIGALARM handlers which constitute the
456 * majority of invocations, do not use it.
457 */
458 switch (sig) {
459 case SIGSEGV:
460 case SIGTRAP:
461 case SIGILL:
462 case SIGBUS:
463 case SIGFPE:
464 case SIGWINCH:
465 ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
466 break;
467 }
468
469 switch (sig) {
470 case SIGSEGV:
471 if (PTRACE_FULL_FAULTINFO) {
472 get_skas_faultinfo(pid,
473 ®s->faultinfo, aux_fp_regs);
474 (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
475 regs);
476 }
477 else handle_segv(pid, regs, aux_fp_regs);
478 break;
479 case SIGTRAP + 0x80:
480 handle_trap(pid, regs, local_using_sysemu);
481 break;
482 case SIGTRAP:
483 relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
484 break;
485 case SIGALRM:
486 break;
487 case SIGIO:
488 case SIGILL:
489 case SIGBUS:
490 case SIGFPE:
491 case SIGWINCH:
492 block_signals_trace();
493 (*sig_info[sig])(sig, (struct siginfo *)&si, regs);
494 unblock_signals_trace();
495 break;
496 default:
497 printk(UM_KERN_ERR "userspace - child stopped "
498 "with signal %d\n", sig);
499 fatal_sigsegv();
500 }
501 pid = userspace_pid[0];
502 interrupt_end();
503
504 /* Avoid -ERESTARTSYS handling in host */
505 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
506 PT_SYSCALL_NR(regs->gp) = -1;
507 }
508 }
509}
510
511static unsigned long thread_regs[MAX_REG_NR];
512static unsigned long thread_fp_regs[FP_SIZE];
513
514static int __init init_thread_regs(void)
515{
516 get_safe_registers(thread_regs, thread_fp_regs);
517 /* Set parent's instruction pointer to start of clone-stub */
518 thread_regs[REGS_IP_INDEX] = STUB_CODE +
519 (unsigned long) stub_clone_handler -
520 (unsigned long) __syscall_stub_start;
521 thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
522 sizeof(void *);
523#ifdef __SIGNAL_FRAMESIZE
524 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
525#endif
526 return 0;
527}
528
529__initcall(init_thread_regs);
530
531int copy_context_skas0(unsigned long new_stack, int pid)
532{
533 int err;
534 unsigned long current_stack = current_stub_stack();
535 struct stub_data *data = (struct stub_data *) current_stack;
536 struct stub_data *child_data = (struct stub_data *) new_stack;
537 unsigned long long new_offset;
538 int new_fd = phys_mapping(uml_to_phys((void *)new_stack), &new_offset);
539
540 /*
541 * prepare offset and fd of child's stack as argument for parent's
542 * and child's mmap2 calls
543 */
544 *data = ((struct stub_data) {
545 .offset = MMAP_OFFSET(new_offset),
546 .fd = new_fd,
547 .parent_err = -ESRCH,
548 .child_err = 0,
549 });
550
551 *child_data = ((struct stub_data) {
552 .child_err = -ESRCH,
553 });
554
555 err = ptrace_setregs(pid, thread_regs);
556 if (err < 0) {
557 err = -errno;
558 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
559 "failed, pid = %d, errno = %d\n", pid, -err);
560 return err;
561 }
562
563 err = put_fp_registers(pid, thread_fp_regs);
564 if (err < 0) {
565 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
566 "failed, pid = %d, err = %d\n", pid, err);
567 return err;
568 }
569
570 /*
571 * Wait, until parent has finished its work: read child's pid from
572 * parent's stack, and check, if bad result.
573 */
574 err = ptrace(PTRACE_CONT, pid, 0, 0);
575 if (err) {
576 err = -errno;
577 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
578 "errno = %d\n", pid, errno);
579 return err;
580 }
581
582 wait_stub_done(pid);
583
584 pid = data->parent_err;
585 if (pid < 0) {
586 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
587 "error %d\n", -pid);
588 return pid;
589 }
590
591 /*
592 * Wait, until child has finished too: read child's result from
593 * child's stack and check it.
594 */
595 wait_stub_done(pid);
596 if (child_data->child_err != STUB_DATA) {
597 printk(UM_KERN_ERR "copy_context_skas0 - stub-child %d reports "
598 "error %ld\n", pid, data->child_err);
599 err = data->child_err;
600 goto out_kill;
601 }
602
603 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
604 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
605 err = -errno;
606 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
607 "failed, errno = %d\n", errno);
608 goto out_kill;
609 }
610
611 return pid;
612
613 out_kill:
614 os_kill_ptraced_process(pid, 1);
615 return err;
616}
617
618void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
619{
620 (*buf)[0].JB_IP = (unsigned long) handler;
621 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
622 sizeof(void *);
623}
624
625#define INIT_JMP_NEW_THREAD 0
626#define INIT_JMP_CALLBACK 1
627#define INIT_JMP_HALT 2
628#define INIT_JMP_REBOOT 3
629
630void switch_threads(jmp_buf *me, jmp_buf *you)
631{
632 if (UML_SETJMP(me) == 0)
633 UML_LONGJMP(you, 1);
634}
635
636static jmp_buf initial_jmpbuf;
637
638/* XXX Make these percpu */
639static void (*cb_proc)(void *arg);
640static void *cb_arg;
641static jmp_buf *cb_back;
642
643int start_idle_thread(void *stack, jmp_buf *switch_buf)
644{
645 int n;
646
647 set_handler(SIGWINCH);
648
649 /*
650 * Can't use UML_SETJMP or UML_LONGJMP here because they save
651 * and restore signals, with the possible side-effect of
652 * trying to handle any signals which came when they were
653 * blocked, which can't be done on this stack.
654 * Signals must be blocked when jumping back here and restored
655 * after returning to the jumper.
656 */
657 n = setjmp(initial_jmpbuf);
658 switch (n) {
659 case INIT_JMP_NEW_THREAD:
660 (*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
661 (*switch_buf)[0].JB_SP = (unsigned long) stack +
662 UM_THREAD_SIZE - sizeof(void *);
663 break;
664 case INIT_JMP_CALLBACK:
665 (*cb_proc)(cb_arg);
666 longjmp(*cb_back, 1);
667 break;
668 case INIT_JMP_HALT:
669 kmalloc_ok = 0;
670 return 0;
671 case INIT_JMP_REBOOT:
672 kmalloc_ok = 0;
673 return 1;
674 default:
675 printk(UM_KERN_ERR "Bad sigsetjmp return in "
676 "start_idle_thread - %d\n", n);
677 fatal_sigsegv();
678 }
679 longjmp(*switch_buf, 1);
680
681 /* unreachable */
682 printk(UM_KERN_ERR "impossible long jump!");
683 fatal_sigsegv();
684 return 0;
685}
686
687void initial_thread_cb_skas(void (*proc)(void *), void *arg)
688{
689 jmp_buf here;
690
691 cb_proc = proc;
692 cb_arg = arg;
693 cb_back = &here;
694
695 block_signals_trace();
696 if (UML_SETJMP(&here) == 0)
697 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
698 unblock_signals_trace();
699
700 cb_proc = NULL;
701 cb_arg = NULL;
702 cb_back = NULL;
703}
704
705void halt_skas(void)
706{
707 block_signals_trace();
708 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
709}
710
711static bool noreboot;
712
713static int __init noreboot_cmd_param(char *str, int *add)
714{
715 noreboot = true;
716 return 0;
717}
718
719__uml_setup("noreboot", noreboot_cmd_param,
720"noreboot\n"
721" Rather than rebooting, exit always, akin to QEMU's -no-reboot option.\n"
722" This is useful if you're using CONFIG_PANIC_TIMEOUT in order to catch\n"
723" crashes in CI\n");
724
725void reboot_skas(void)
726{
727 block_signals_trace();
728 UML_LONGJMP(&initial_jmpbuf, noreboot ? INIT_JMP_HALT : INIT_JMP_REBOOT);
729}
730
731void __switch_mm(struct mm_id *mm_idp)
732{
733 userspace_pid[0] = mm_idp->u.pid;
734 kill_userspace_mm[0] = mm_idp->kill;
735}
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 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
80static 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
107void 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
130bad_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
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, ®s->faultinfo, aux_fp_regs);
177 segv(regs->faultinfo, 0, 1, NULL);
178}
179
180static void handle_trap(int pid, struct uml_pt_regs *regs)
181{
182 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
183 fatal_sigsegv();
184
185 handle_syscall(regs);
186}
187
188extern char __syscall_stub_start[];
189
190/**
191 * userspace_tramp() - userspace trampoline
192 * @stack: pointer to the new userspace stack page
193 *
194 * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
195 * This function will run on a temporary stack page.
196 * It ptrace()'es itself, then
197 * Two pages are mapped into the userspace address space:
198 * - STUB_CODE (with EXEC), which contains the skas stub code
199 * - 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.
200 * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
201 * And last the process stops itself to give control to the UML kernel for this userspace process.
202 *
203 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
204 */
205static int userspace_tramp(void *stack)
206{
207 struct sigaction sa;
208 void *addr;
209 int fd;
210 unsigned long long offset;
211 unsigned long segv_handler = STUB_CODE +
212 (unsigned long) stub_segv_handler -
213 (unsigned long) __syscall_stub_start;
214
215 ptrace(PTRACE_TRACEME, 0, 0, 0);
216
217 signal(SIGTERM, SIG_DFL);
218 signal(SIGWINCH, SIG_IGN);
219
220 fd = phys_mapping(uml_to_phys(__syscall_stub_start), &offset);
221 addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
222 PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
223 if (addr == MAP_FAILED) {
224 os_info("mapping mmap stub at 0x%lx failed, errno = %d\n",
225 STUB_CODE, errno);
226 exit(1);
227 }
228
229 fd = phys_mapping(uml_to_phys(stack), &offset);
230 addr = mmap((void *) STUB_DATA,
231 STUB_DATA_PAGES * UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
232 MAP_FIXED | MAP_SHARED, fd, offset);
233 if (addr == MAP_FAILED) {
234 os_info("mapping segfault stack at 0x%lx failed, errno = %d\n",
235 STUB_DATA, errno);
236 exit(1);
237 }
238
239 set_sigstack((void *) STUB_DATA, STUB_DATA_PAGES * UM_KERN_PAGE_SIZE);
240 sigemptyset(&sa.sa_mask);
241 sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
242 sa.sa_sigaction = (void *) segv_handler;
243 sa.sa_restorer = NULL;
244 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
245 os_info("%s - setting SIGSEGV handler failed - errno = %d\n",
246 __func__, errno);
247 exit(1);
248 }
249
250 kill(os_getpid(), SIGSTOP);
251 return 0;
252}
253
254int userspace_pid[NR_CPUS];
255int kill_userspace_mm[NR_CPUS];
256
257/**
258 * start_userspace() - prepare a new userspace process
259 * @stub_stack: pointer to the stub stack.
260 *
261 * Setups a new temporary stack page that is used while userspace_tramp() runs
262 * Clones the kernel process into a new userspace process, with FDs only.
263 *
264 * Return: When positive: the process id of the new userspace process,
265 * when negative: an error number.
266 * FIXME: can PIDs become negative?!
267 */
268int start_userspace(unsigned long stub_stack)
269{
270 void *stack;
271 unsigned long sp;
272 int pid, status, n, flags, err;
273
274 /* setup a temporary stack page */
275 stack = mmap(NULL, UM_KERN_PAGE_SIZE,
276 PROT_READ | PROT_WRITE | PROT_EXEC,
277 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
278 if (stack == MAP_FAILED) {
279 err = -errno;
280 printk(UM_KERN_ERR "%s : mmap failed, errno = %d\n",
281 __func__, errno);
282 return err;
283 }
284
285 /* set stack pointer to the end of the stack page, so it can grow downwards */
286 sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
287
288 flags = CLONE_FILES | SIGCHLD;
289
290 /* clone into new userspace process */
291 pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
292 if (pid < 0) {
293 err = -errno;
294 printk(UM_KERN_ERR "%s : clone failed, errno = %d\n",
295 __func__, errno);
296 return err;
297 }
298
299 do {
300 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
301 if (n < 0) {
302 err = -errno;
303 printk(UM_KERN_ERR "%s : wait failed, errno = %d\n",
304 __func__, errno);
305 goto out_kill;
306 }
307 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
308
309 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
310 err = -EINVAL;
311 printk(UM_KERN_ERR "%s : expected SIGSTOP, got status = %d\n",
312 __func__, status);
313 goto out_kill;
314 }
315
316 if (ptrace(PTRACE_SETOPTIONS, pid, NULL,
317 (void *) PTRACE_O_TRACESYSGOOD) < 0) {
318 err = -errno;
319 printk(UM_KERN_ERR "%s : PTRACE_SETOPTIONS failed, errno = %d\n",
320 __func__, errno);
321 goto out_kill;
322 }
323
324 if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
325 err = -errno;
326 printk(UM_KERN_ERR "%s : munmap failed, errno = %d\n",
327 __func__, errno);
328 goto out_kill;
329 }
330
331 return pid;
332
333 out_kill:
334 os_kill_ptraced_process(pid, 1);
335 return err;
336}
337
338void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
339{
340 int err, status, op, pid = userspace_pid[0];
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 "%s - ptrace set regs failed, errno = %d\n",
360 __func__, errno);
361 fatal_sigsegv();
362 }
363
364 if (put_fp_registers(pid, regs->fp)) {
365 printk(UM_KERN_ERR "%s - ptrace set fp regs failed, errno = %d\n",
366 __func__, errno);
367 fatal_sigsegv();
368 }
369
370 if (singlestepping())
371 op = PTRACE_SYSEMU_SINGLESTEP;
372 else
373 op = PTRACE_SYSEMU;
374
375 if (ptrace(op, pid, 0, 0)) {
376 printk(UM_KERN_ERR "%s - ptrace continue failed, op = %d, errno = %d\n",
377 __func__, op, errno);
378 fatal_sigsegv();
379 }
380
381 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
382 if (err < 0) {
383 printk(UM_KERN_ERR "%s - wait failed, errno = %d\n",
384 __func__, errno);
385 fatal_sigsegv();
386 }
387
388 regs->is_user = 1;
389 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
390 printk(UM_KERN_ERR "%s - PTRACE_GETREGS failed, errno = %d\n",
391 __func__, errno);
392 fatal_sigsegv();
393 }
394
395 if (get_fp_registers(pid, regs->fp)) {
396 printk(UM_KERN_ERR "%s - get_fp_registers failed, errno = %d\n",
397 __func__, errno);
398 fatal_sigsegv();
399 }
400
401 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
402
403 if (WIFSTOPPED(status)) {
404 int sig = WSTOPSIG(status);
405
406 /* These signal handlers need the si argument.
407 * The SIGIO and SIGALARM handlers which constitute the
408 * majority of invocations, do not use it.
409 */
410 switch (sig) {
411 case SIGSEGV:
412 case SIGTRAP:
413 case SIGILL:
414 case SIGBUS:
415 case SIGFPE:
416 case SIGWINCH:
417 ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
418 break;
419 }
420
421 switch (sig) {
422 case SIGSEGV:
423 if (PTRACE_FULL_FAULTINFO) {
424 get_skas_faultinfo(pid,
425 ®s->faultinfo, aux_fp_regs);
426 (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
427 regs);
428 }
429 else handle_segv(pid, regs, aux_fp_regs);
430 break;
431 case SIGTRAP + 0x80:
432 handle_trap(pid, regs);
433 break;
434 case SIGTRAP:
435 relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
436 break;
437 case SIGALRM:
438 break;
439 case SIGIO:
440 case SIGILL:
441 case SIGBUS:
442 case SIGFPE:
443 case SIGWINCH:
444 block_signals_trace();
445 (*sig_info[sig])(sig, (struct siginfo *)&si, regs);
446 unblock_signals_trace();
447 break;
448 default:
449 printk(UM_KERN_ERR "%s - child stopped with signal %d\n",
450 __func__, sig);
451 fatal_sigsegv();
452 }
453 pid = userspace_pid[0];
454 interrupt_end();
455
456 /* Avoid -ERESTARTSYS handling in host */
457 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
458 PT_SYSCALL_NR(regs->gp) = -1;
459 }
460 }
461}
462
463static unsigned long thread_regs[MAX_REG_NR];
464static unsigned long thread_fp_regs[FP_SIZE];
465
466static int __init init_thread_regs(void)
467{
468 get_safe_registers(thread_regs, thread_fp_regs);
469 /* Set parent's instruction pointer to start of clone-stub */
470 thread_regs[REGS_IP_INDEX] = STUB_CODE +
471 (unsigned long) stub_clone_handler -
472 (unsigned long) __syscall_stub_start;
473 thread_regs[REGS_SP_INDEX] = STUB_DATA + STUB_DATA_PAGES * UM_KERN_PAGE_SIZE -
474 sizeof(void *);
475#ifdef __SIGNAL_FRAMESIZE
476 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
477#endif
478 return 0;
479}
480
481__initcall(init_thread_regs);
482
483int copy_context_skas0(unsigned long new_stack, int pid)
484{
485 int err;
486 unsigned long current_stack = current_stub_stack();
487 struct stub_data *data = (struct stub_data *) current_stack;
488 struct stub_data *child_data = (struct stub_data *) new_stack;
489 unsigned long long new_offset;
490 int new_fd = phys_mapping(uml_to_phys((void *)new_stack), &new_offset);
491
492 /*
493 * prepare offset and fd of child's stack as argument for parent's
494 * and child's mmap2 calls
495 */
496 *data = ((struct stub_data) {
497 .offset = MMAP_OFFSET(new_offset),
498 .fd = new_fd,
499 .parent_err = -ESRCH,
500 .child_err = 0,
501 });
502
503 *child_data = ((struct stub_data) {
504 .child_err = -ESRCH,
505 });
506
507 err = ptrace_setregs(pid, thread_regs);
508 if (err < 0) {
509 err = -errno;
510 printk(UM_KERN_ERR "%s : PTRACE_SETREGS failed, pid = %d, errno = %d\n",
511 __func__, pid, -err);
512 return err;
513 }
514
515 err = put_fp_registers(pid, thread_fp_regs);
516 if (err < 0) {
517 printk(UM_KERN_ERR "%s : put_fp_registers failed, pid = %d, err = %d\n",
518 __func__, pid, err);
519 return err;
520 }
521
522 /*
523 * Wait, until parent has finished its work: read child's pid from
524 * parent's stack, and check, if bad result.
525 */
526 err = ptrace(PTRACE_CONT, pid, 0, 0);
527 if (err) {
528 err = -errno;
529 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, errno = %d\n",
530 pid, errno);
531 return err;
532 }
533
534 wait_stub_done(pid);
535
536 pid = data->parent_err;
537 if (pid < 0) {
538 printk(UM_KERN_ERR "%s - stub-parent reports error %d\n",
539 __func__, -pid);
540 return pid;
541 }
542
543 /*
544 * Wait, until child has finished too: read child's result from
545 * child's stack and check it.
546 */
547 wait_stub_done(pid);
548 if (child_data->child_err != STUB_DATA) {
549 printk(UM_KERN_ERR "%s - stub-child %d reports error %ld\n",
550 __func__, pid, data->child_err);
551 err = data->child_err;
552 goto out_kill;
553 }
554
555 if (ptrace(PTRACE_SETOPTIONS, pid, NULL,
556 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
557 err = -errno;
558 printk(UM_KERN_ERR "%s : PTRACE_SETOPTIONS failed, errno = %d\n",
559 __func__, errno);
560 goto out_kill;
561 }
562
563 return pid;
564
565 out_kill:
566 os_kill_ptraced_process(pid, 1);
567 return err;
568}
569
570void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
571{
572 (*buf)[0].JB_IP = (unsigned long) handler;
573 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
574 sizeof(void *);
575}
576
577#define INIT_JMP_NEW_THREAD 0
578#define INIT_JMP_CALLBACK 1
579#define INIT_JMP_HALT 2
580#define INIT_JMP_REBOOT 3
581
582void switch_threads(jmp_buf *me, jmp_buf *you)
583{
584 if (UML_SETJMP(me) == 0)
585 UML_LONGJMP(you, 1);
586}
587
588static jmp_buf initial_jmpbuf;
589
590/* XXX Make these percpu */
591static void (*cb_proc)(void *arg);
592static void *cb_arg;
593static jmp_buf *cb_back;
594
595int start_idle_thread(void *stack, jmp_buf *switch_buf)
596{
597 int n;
598
599 set_handler(SIGWINCH);
600
601 /*
602 * Can't use UML_SETJMP or UML_LONGJMP here because they save
603 * and restore signals, with the possible side-effect of
604 * trying to handle any signals which came when they were
605 * blocked, which can't be done on this stack.
606 * Signals must be blocked when jumping back here and restored
607 * after returning to the jumper.
608 */
609 n = setjmp(initial_jmpbuf);
610 switch (n) {
611 case INIT_JMP_NEW_THREAD:
612 (*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
613 (*switch_buf)[0].JB_SP = (unsigned long) stack +
614 UM_THREAD_SIZE - sizeof(void *);
615 break;
616 case INIT_JMP_CALLBACK:
617 (*cb_proc)(cb_arg);
618 longjmp(*cb_back, 1);
619 break;
620 case INIT_JMP_HALT:
621 kmalloc_ok = 0;
622 return 0;
623 case INIT_JMP_REBOOT:
624 kmalloc_ok = 0;
625 return 1;
626 default:
627 printk(UM_KERN_ERR "Bad sigsetjmp return in %s - %d\n",
628 __func__, n);
629 fatal_sigsegv();
630 }
631 longjmp(*switch_buf, 1);
632
633 /* unreachable */
634 printk(UM_KERN_ERR "impossible long jump!");
635 fatal_sigsegv();
636 return 0;
637}
638
639void initial_thread_cb_skas(void (*proc)(void *), void *arg)
640{
641 jmp_buf here;
642
643 cb_proc = proc;
644 cb_arg = arg;
645 cb_back = &here;
646
647 block_signals_trace();
648 if (UML_SETJMP(&here) == 0)
649 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
650 unblock_signals_trace();
651
652 cb_proc = NULL;
653 cb_arg = NULL;
654 cb_back = NULL;
655}
656
657void halt_skas(void)
658{
659 block_signals_trace();
660 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
661}
662
663static bool noreboot;
664
665static int __init noreboot_cmd_param(char *str, int *add)
666{
667 noreboot = true;
668 return 0;
669}
670
671__uml_setup("noreboot", noreboot_cmd_param,
672"noreboot\n"
673" Rather than rebooting, exit always, akin to QEMU's -no-reboot option.\n"
674" This is useful if you're using CONFIG_PANIC_TIMEOUT in order to catch\n"
675" crashes in CI\n");
676
677void reboot_skas(void)
678{
679 block_signals_trace();
680 UML_LONGJMP(&initial_jmpbuf, noreboot ? INIT_JMP_HALT : INIT_JMP_REBOOT);
681}
682
683void __switch_mm(struct mm_id *mm_idp)
684{
685 userspace_pid[0] = mm_idp->u.pid;
686 kill_userspace_mm[0] = mm_idp->kill;
687}