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
14 #include <asm/unistd.h>
15 #include <as-layout.h>
17 #include <kern_util.h>
20 #include <ptrace_user.h>
21 #include <registers.h>
23 #include <sysdep/stub.h>
24 #include <linux/threads.h>
26 int is_skas_winch(int pid, int fd, void *data)
28 return pid == getpgrp();
31 static int ptrace_dump_regs(int pid)
33 unsigned long regs[MAX_REG_NR];
36 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
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]);
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.
50 #define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
52 /* Signals that the stub will finish with - anything else is an error */
53 #define STUB_DONE_MASK (1 << SIGTRAP)
55 void wait_stub_done(int pid)
60 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
61 if ((n < 0) || !WIFSTOPPED(status))
64 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
67 err = ptrace(PTRACE_CONT, pid, 0, 0);
69 printk(UM_KERN_ERR "wait_stub_done : continue failed, "
70 "errno = %d\n", errno);
75 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
79 err = ptrace_dump_regs(pid);
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,
89 extern unsigned long current_stub_stack(void);
91 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
94 unsigned long fpregs[FP_SIZE];
96 err = get_fp_registers(pid, fpregs);
98 printk(UM_KERN_ERR "save_fp_registers returned %d\n",
102 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
104 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
105 "errno = %d\n", pid, errno);
111 * faultinfo is prepared by the stub_segv_handler at start of
112 * the stub stack page. We just have to copy it.
114 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
116 err = put_fp_registers(pid, fpregs);
118 printk(UM_KERN_ERR "put_fp_registers returned %d\n",
124 static void handle_segv(int pid, struct uml_pt_regs * regs)
126 get_skas_faultinfo(pid, ®s->faultinfo);
127 segv(regs->faultinfo, 0, 1, NULL);
131 * To use the same value of using_sysemu as the caller, ask it that value
132 * (in local_using_sysemu
134 static void handle_trap(int pid, struct uml_pt_regs *regs,
135 int local_using_sysemu)
139 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
142 if (!local_using_sysemu)
144 err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
147 printk(UM_KERN_ERR "handle_trap - nullifying syscall "
148 "failed, errno = %d\n", errno);
152 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
154 printk(UM_KERN_ERR "handle_trap - continuing to end of "
155 "syscall failed, errno = %d\n", errno);
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);
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",
173 handle_syscall(regs);
176 extern char __syscall_stub_start[];
179 * userspace_tramp() - userspace trampoline
180 * @stack: pointer to the new userspace stack page, can be NULL, if? FIXME:
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.
191 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
193 static int userspace_tramp(void *stack)
197 unsigned long long offset;
199 ptrace(PTRACE_TRACEME, 0, 0, 0);
201 signal(SIGTERM, SIG_DFL);
202 signal(SIGWINCH, SIG_IGN);
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
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);
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",
232 unsigned long v = STUB_CODE +
233 (unsigned long) stub_segv_handler -
234 (unsigned long) __syscall_stub_start;
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);
248 kill(os_getpid(), SIGSTOP);
252 int userspace_pid[NR_CPUS];
255 * start_userspace() - prepare a new userspace process
256 * @stub_stack: pointer to the stub stack. Can be NULL, if? FIXME:
258 * Setups a new temporary stack page that is used while userspace_tramp() runs
259 * Clones the kernel process into a new userspace process, with FDs only.
261 * Return: When positive: the process id of the new userspace process,
262 * when negative: an error number.
263 * FIXME: can PIDs become negative?!
265 int start_userspace(unsigned long stub_stack)
269 int pid, status, n, flags, err;
271 /* setup a temporary stack page */
272 stack = mmap(NULL, UM_KERN_PAGE_SIZE,
273 PROT_READ | PROT_WRITE | PROT_EXEC,
274 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
275 if (stack == MAP_FAILED) {
277 printk(UM_KERN_ERR "start_userspace : mmap failed, "
278 "errno = %d\n", errno);
282 /* set stack pointer to the end of the stack page, so it can grow downwards */
283 sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
285 flags = CLONE_FILES | SIGCHLD;
287 /* clone into new userspace process */
288 pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
291 printk(UM_KERN_ERR "start_userspace : clone failed, "
292 "errno = %d\n", errno);
297 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
300 printk(UM_KERN_ERR "start_userspace : wait failed, "
301 "errno = %d\n", errno);
304 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
306 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
308 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
309 "status = %d\n", status);
313 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
314 (void *) PTRACE_O_TRACESYSGOOD) < 0) {
316 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
317 "failed, errno = %d\n", errno);
321 if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
323 printk(UM_KERN_ERR "start_userspace : munmap failed, "
324 "errno = %d\n", errno);
331 os_kill_ptraced_process(pid, 1);
335 void userspace(struct uml_pt_regs *regs)
337 int err, status, op, pid = userspace_pid[0];
338 /* To prevent races if using_sysemu changes under us.*/
339 int local_using_sysemu;
342 /* Handle any immediate reschedules or signals */
348 * This can legitimately fail if the process loads a
349 * bogus value into a segment register. It will
350 * segfault and PTRACE_GETREGS will read that value
351 * out of the process. However, PTRACE_SETREGS will
352 * fail. In this case, there is nothing to do but
353 * just kill the process.
355 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
356 printk(UM_KERN_ERR "userspace - ptrace set regs "
357 "failed, errno = %d\n", errno);
361 if (put_fp_registers(pid, regs->fp)) {
362 printk(UM_KERN_ERR "userspace - ptrace set fp regs "
363 "failed, errno = %d\n", errno);
367 /* Now we set local_using_sysemu to be used for one loop */
368 local_using_sysemu = get_using_sysemu();
370 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
371 singlestepping(NULL));
373 if (ptrace(op, pid, 0, 0)) {
374 printk(UM_KERN_ERR "userspace - ptrace continue "
375 "failed, op = %d, errno = %d\n", op, errno);
379 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
381 printk(UM_KERN_ERR "userspace - wait failed, "
382 "errno = %d\n", errno);
387 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
388 printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
389 "errno = %d\n", errno);
393 if (get_fp_registers(pid, regs->fp)) {
394 printk(UM_KERN_ERR "userspace - get_fp_registers failed, "
395 "errno = %d\n", errno);
399 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
401 if (WIFSTOPPED(status)) {
402 int sig = WSTOPSIG(status);
404 ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
408 if (PTRACE_FULL_FAULTINFO) {
409 get_skas_faultinfo(pid,
411 (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
414 else handle_segv(pid, regs);
417 handle_trap(pid, regs, local_using_sysemu);
420 relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
430 (*sig_info[sig])(sig, (struct siginfo *)&si, regs);
434 printk(UM_KERN_ERR "userspace - child stopped "
435 "with signal %d\n", sig);
438 pid = userspace_pid[0];
441 /* Avoid -ERESTARTSYS handling in host */
442 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
443 PT_SYSCALL_NR(regs->gp) = -1;
448 static unsigned long thread_regs[MAX_REG_NR];
449 static unsigned long thread_fp_regs[FP_SIZE];
451 static int __init init_thread_regs(void)
453 get_safe_registers(thread_regs, thread_fp_regs);
454 /* Set parent's instruction pointer to start of clone-stub */
455 thread_regs[REGS_IP_INDEX] = STUB_CODE +
456 (unsigned long) stub_clone_handler -
457 (unsigned long) __syscall_stub_start;
458 thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
460 #ifdef __SIGNAL_FRAMESIZE
461 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
466 __initcall(init_thread_regs);
468 int copy_context_skas0(unsigned long new_stack, int pid)
471 unsigned long current_stack = current_stub_stack();
472 struct stub_data *data = (struct stub_data *) current_stack;
473 struct stub_data *child_data = (struct stub_data *) new_stack;
474 unsigned long long new_offset;
475 int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
478 * prepare offset and fd of child's stack as argument for parent's
479 * and child's mmap2 calls
481 *data = ((struct stub_data) {
482 .offset = MMAP_OFFSET(new_offset),
486 err = ptrace_setregs(pid, thread_regs);
489 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
490 "failed, pid = %d, errno = %d\n", pid, -err);
494 err = put_fp_registers(pid, thread_fp_regs);
496 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
497 "failed, pid = %d, err = %d\n", pid, err);
501 /* set a well known return code for detection of child write failure */
502 child_data->err = 12345678;
505 * Wait, until parent has finished its work: read child's pid from
506 * parent's stack, and check, if bad result.
508 err = ptrace(PTRACE_CONT, pid, 0, 0);
511 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
512 "errno = %d\n", pid, errno);
520 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
526 * Wait, until child has finished too: read child's result from
527 * child's stack and check it.
530 if (child_data->err != STUB_DATA) {
531 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
532 "error %ld\n", child_data->err);
533 err = child_data->err;
537 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
538 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
540 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
541 "failed, errno = %d\n", errno);
548 os_kill_ptraced_process(pid, 1);
552 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
554 (*buf)[0].JB_IP = (unsigned long) handler;
555 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
559 #define INIT_JMP_NEW_THREAD 0
560 #define INIT_JMP_CALLBACK 1
561 #define INIT_JMP_HALT 2
562 #define INIT_JMP_REBOOT 3
564 void switch_threads(jmp_buf *me, jmp_buf *you)
566 if (UML_SETJMP(me) == 0)
570 static jmp_buf initial_jmpbuf;
572 /* XXX Make these percpu */
573 static void (*cb_proc)(void *arg);
575 static jmp_buf *cb_back;
577 int start_idle_thread(void *stack, jmp_buf *switch_buf)
581 set_handler(SIGWINCH);
584 * Can't use UML_SETJMP or UML_LONGJMP here because they save
585 * and restore signals, with the possible side-effect of
586 * trying to handle any signals which came when they were
587 * blocked, which can't be done on this stack.
588 * Signals must be blocked when jumping back here and restored
589 * after returning to the jumper.
591 n = setjmp(initial_jmpbuf);
593 case INIT_JMP_NEW_THREAD:
594 (*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
595 (*switch_buf)[0].JB_SP = (unsigned long) stack +
596 UM_THREAD_SIZE - sizeof(void *);
598 case INIT_JMP_CALLBACK:
600 longjmp(*cb_back, 1);
605 case INIT_JMP_REBOOT:
609 printk(UM_KERN_ERR "Bad sigsetjmp return in "
610 "start_idle_thread - %d\n", n);
613 longjmp(*switch_buf, 1);
616 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
625 if (UML_SETJMP(&here) == 0)
626 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
637 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
640 void reboot_skas(void)
643 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
646 void __switch_mm(struct mm_id *mm_idp)
648 userspace_pid[0] = mm_idp->u.pid;