1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
4 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
15 #include <asm/unistd.h>
16 #include <as-layout.h>
18 #include <kern_util.h>
21 #include <ptrace_user.h>
22 #include <registers.h>
24 #include <sysdep/stub.h>
25 #include <linux/threads.h>
27 int is_skas_winch(int pid, int fd, void *data)
29 return pid == getpgrp();
32 static const char *ptrace_reg_name(int idx)
34 #define R(n) case HOST_##n: return #n
57 #elif defined(__i386__)
80 static int ptrace_dump_regs(int pid)
82 unsigned long regs[MAX_REG_NR];
85 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
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);
92 printk(UM_KERN_ERR "\t%s\t(%2d): %lx\n", regname, i, regs[i]);
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.
102 #define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
104 /* Signals that the stub will finish with - anything else is an error */
105 #define STUB_DONE_MASK (1 << SIGTRAP)
107 void wait_stub_done(int pid)
112 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
113 if ((n < 0) || !WIFSTOPPED(status))
116 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
119 err = ptrace(PTRACE_CONT, pid, 0, 0);
121 printk(UM_KERN_ERR "wait_stub_done : continue failed, "
122 "errno = %d\n", errno);
127 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
131 err = ptrace_dump_regs(pid);
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,
141 extern unsigned long current_stub_stack(void);
143 static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
147 err = get_fp_registers(pid, aux_fp_regs);
149 printk(UM_KERN_ERR "save_fp_registers returned %d\n",
153 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
155 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
156 "errno = %d\n", pid, errno);
162 * faultinfo is prepared by the stub_segv_handler at start of
163 * the stub stack page. We just have to copy it.
165 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
167 err = put_fp_registers(pid, aux_fp_regs);
169 printk(UM_KERN_ERR "put_fp_registers returned %d\n",
175 static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
177 get_skas_faultinfo(pid, ®s->faultinfo, aux_fp_regs);
178 segv(regs->faultinfo, 0, 1, NULL);
182 * To use the same value of using_sysemu as the caller, ask it that value
183 * (in local_using_sysemu
185 static void handle_trap(int pid, struct uml_pt_regs *regs,
186 int local_using_sysemu)
190 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
193 if (!local_using_sysemu)
195 err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
198 printk(UM_KERN_ERR "handle_trap - nullifying syscall "
199 "failed, errno = %d\n", errno);
203 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
205 printk(UM_KERN_ERR "handle_trap - continuing to end of "
206 "syscall failed, errno = %d\n", errno);
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);
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",
224 handle_syscall(regs);
227 extern char __syscall_stub_start[];
230 * userspace_tramp() - userspace trampoline
231 * @stack: pointer to the new userspace stack page, can be NULL, if? FIXME:
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.
242 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
244 static int userspace_tramp(void *stack)
248 unsigned long long offset;
250 ptrace(PTRACE_TRACEME, 0, 0, 0);
252 signal(SIGTERM, SIG_DFL);
253 signal(SIGWINCH, SIG_IGN);
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);
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",
279 unsigned long v = STUB_CODE +
280 (unsigned long) stub_segv_handler -
281 (unsigned long) __syscall_stub_start;
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);
295 kill(os_getpid(), SIGSTOP);
299 int userspace_pid[NR_CPUS];
300 int kill_userspace_mm[NR_CPUS];
303 * start_userspace() - prepare a new userspace process
304 * @stub_stack: pointer to the stub stack. Can be NULL, if? FIXME:
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.
309 * Return: When positive: the process id of the new userspace process,
310 * when negative: an error number.
311 * FIXME: can PIDs become negative?!
313 int start_userspace(unsigned long stub_stack)
317 int pid, status, n, flags, err;
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) {
325 printk(UM_KERN_ERR "start_userspace : mmap failed, "
326 "errno = %d\n", errno);
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;
333 flags = CLONE_FILES | SIGCHLD;
335 /* clone into new userspace process */
336 pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
339 printk(UM_KERN_ERR "start_userspace : clone failed, "
340 "errno = %d\n", errno);
345 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
348 printk(UM_KERN_ERR "start_userspace : wait failed, "
349 "errno = %d\n", errno);
352 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
354 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
356 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
357 "status = %d\n", status);
361 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
362 (void *) PTRACE_O_TRACESYSGOOD) < 0) {
364 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
365 "failed, errno = %d\n", errno);
369 if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
371 printk(UM_KERN_ERR "start_userspace : munmap failed, "
372 "errno = %d\n", errno);
379 os_kill_ptraced_process(pid, 1);
383 void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
385 int err, status, op, pid = userspace_pid[0];
386 /* To prevent races if using_sysemu changes under us.*/
387 int local_using_sysemu;
390 /* Handle any immediate reschedules or signals */
394 if (kill_userspace_mm[0])
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.
405 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
406 printk(UM_KERN_ERR "userspace - ptrace set regs "
407 "failed, errno = %d\n", errno);
411 if (put_fp_registers(pid, regs->fp)) {
412 printk(UM_KERN_ERR "userspace - ptrace set fp regs "
413 "failed, errno = %d\n", errno);
417 /* Now we set local_using_sysemu to be used for one loop */
418 local_using_sysemu = get_using_sysemu();
420 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
421 singlestepping(NULL));
423 if (ptrace(op, pid, 0, 0)) {
424 printk(UM_KERN_ERR "userspace - ptrace continue "
425 "failed, op = %d, errno = %d\n", op, errno);
429 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
431 printk(UM_KERN_ERR "userspace - wait failed, "
432 "errno = %d\n", errno);
437 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
438 printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
439 "errno = %d\n", errno);
443 if (get_fp_registers(pid, regs->fp)) {
444 printk(UM_KERN_ERR "userspace - get_fp_registers failed, "
445 "errno = %d\n", errno);
449 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
451 if (WIFSTOPPED(status)) {
452 int sig = WSTOPSIG(status);
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.
465 ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
471 if (PTRACE_FULL_FAULTINFO) {
472 get_skas_faultinfo(pid,
473 ®s->faultinfo, aux_fp_regs);
474 (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
477 else handle_segv(pid, regs, aux_fp_regs);
480 handle_trap(pid, regs, local_using_sysemu);
483 relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
492 block_signals_trace();
493 (*sig_info[sig])(sig, (struct siginfo *)&si, regs);
494 unblock_signals_trace();
497 printk(UM_KERN_ERR "userspace - child stopped "
498 "with signal %d\n", sig);
501 pid = userspace_pid[0];
504 /* Avoid -ERESTARTSYS handling in host */
505 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
506 PT_SYSCALL_NR(regs->gp) = -1;
511 static unsigned long thread_regs[MAX_REG_NR];
512 static unsigned long thread_fp_regs[FP_SIZE];
514 static int __init init_thread_regs(void)
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 -
523 #ifdef __SIGNAL_FRAMESIZE
524 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
529 __initcall(init_thread_regs);
531 int copy_context_skas0(unsigned long new_stack, int pid)
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);
541 * prepare offset and fd of child's stack as argument for parent's
542 * and child's mmap2 calls
544 *data = ((struct stub_data) {
545 .offset = MMAP_OFFSET(new_offset),
547 .parent_err = -ESRCH,
551 *child_data = ((struct stub_data) {
555 err = ptrace_setregs(pid, thread_regs);
558 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
559 "failed, pid = %d, errno = %d\n", pid, -err);
563 err = put_fp_registers(pid, thread_fp_regs);
565 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
566 "failed, pid = %d, err = %d\n", pid, err);
571 * Wait, until parent has finished its work: read child's pid from
572 * parent's stack, and check, if bad result.
574 err = ptrace(PTRACE_CONT, pid, 0, 0);
577 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
578 "errno = %d\n", pid, errno);
584 pid = data->parent_err;
586 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
592 * Wait, until child has finished too: read child's result from
593 * child's stack and check it.
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;
603 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
604 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
606 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
607 "failed, errno = %d\n", errno);
614 os_kill_ptraced_process(pid, 1);
618 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
620 (*buf)[0].JB_IP = (unsigned long) handler;
621 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
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
630 void switch_threads(jmp_buf *me, jmp_buf *you)
632 if (UML_SETJMP(me) == 0)
636 static jmp_buf initial_jmpbuf;
638 /* XXX Make these percpu */
639 static void (*cb_proc)(void *arg);
641 static jmp_buf *cb_back;
643 int start_idle_thread(void *stack, jmp_buf *switch_buf)
647 set_handler(SIGWINCH);
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.
657 n = setjmp(initial_jmpbuf);
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 *);
664 case INIT_JMP_CALLBACK:
666 longjmp(*cb_back, 1);
671 case INIT_JMP_REBOOT:
675 printk(UM_KERN_ERR "Bad sigsetjmp return in "
676 "start_idle_thread - %d\n", n);
679 longjmp(*switch_buf, 1);
682 printk(UM_KERN_ERR "impossible long jump!");
687 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
695 block_signals_trace();
696 if (UML_SETJMP(&here) == 0)
697 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
698 unblock_signals_trace();
707 block_signals_trace();
708 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
711 static bool noreboot;
713 static int __init noreboot_cmd_param(char *str, int *add)
719 __uml_setup("noreboot", noreboot_cmd_param,
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"
725 void reboot_skas(void)
727 block_signals_trace();
728 UML_LONGJMP(&initial_jmpbuf, noreboot ? INIT_JMP_HALT : INIT_JMP_REBOOT);
731 void __switch_mm(struct mm_id *mm_idp)
733 userspace_pid[0] = mm_idp->u.pid;
734 kill_userspace_mm[0] = mm_idp->kill;