2 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
12 #include <sys/ptrace.h>
14 #include <asm/unistd.h>
15 #include "as-layout.h"
16 #include "chan_user.h"
17 #include "kern_constants.h"
18 #include "kern_util.h"
23 #include "ptrace_user.h"
24 #include "registers.h"
26 #include "skas_ptrace.h"
28 #include "sysdep/stub.h"
30 int is_skas_winch(int pid, int fd, void *data)
35 register_winch_irq(-1, fd, -1, data, 0);
39 static int ptrace_dump_regs(int pid)
41 unsigned long regs[MAX_REG_NR];
44 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
47 printk(UM_KERN_ERR "Stub registers -\n");
48 for (i = 0; i < ARRAY_SIZE(regs); i++)
49 printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
55 * Signals that are OK to receive in the stub - we'll just continue it.
56 * SIGWINCH will happen when UML is inside a detached screen.
58 #define STUB_SIG_MASK (1 << SIGVTALRM)
60 /* Signals that the stub will finish with - anything else is an error */
61 #define STUB_DONE_MASK (1 << SIGTRAP)
63 void wait_stub_done(int pid)
68 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
69 if ((n < 0) || !WIFSTOPPED(status))
72 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
75 err = ptrace(PTRACE_CONT, pid, 0, 0);
77 printk(UM_KERN_ERR "wait_stub_done : continue failed, "
78 "errno = %d\n", errno);
83 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
87 err = ptrace_dump_regs(pid);
89 printk(UM_KERN_ERR "Failed to get registers from stub, "
90 "errno = %d\n", -err);
91 printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
92 "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
97 extern unsigned long current_stub_stack(void);
99 void get_skas_faultinfo(int pid, struct faultinfo * fi)
103 if (ptrace_faultinfo) {
104 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
106 printk(UM_KERN_ERR "get_skas_faultinfo - "
107 "PTRACE_FAULTINFO failed, errno = %d\n", errno);
111 /* Special handling for i386, which has different structs */
112 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
113 memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
114 sizeof(struct faultinfo) -
115 sizeof(struct ptrace_faultinfo));
118 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
120 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
121 "errno = %d\n", pid, errno);
127 * faultinfo is prepared by the stub-segv-handler at start of
128 * the stub stack page. We just have to copy it.
130 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
134 static void handle_segv(int pid, struct uml_pt_regs * regs)
136 get_skas_faultinfo(pid, ®s->faultinfo);
137 segv(regs->faultinfo, 0, 1, NULL);
141 * To use the same value of using_sysemu as the caller, ask it that value
142 * (in local_using_sysemu
144 static void handle_trap(int pid, struct uml_pt_regs *regs,
145 int local_using_sysemu)
149 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
152 /* Mark this as a syscall */
153 UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
155 if (!local_using_sysemu)
157 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
160 printk(UM_KERN_ERR "handle_trap - nullifying syscall "
161 "failed, errno = %d\n", errno);
165 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
167 printk(UM_KERN_ERR "handle_trap - continuing to end of "
168 "syscall failed, errno = %d\n", errno);
172 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
173 if ((err < 0) || !WIFSTOPPED(status) ||
174 (WSTOPSIG(status) != SIGTRAP + 0x80)) {
175 err = ptrace_dump_regs(pid);
177 printk(UM_KERN_ERR "Failed to get registers "
178 "from process, errno = %d\n", -err);
179 printk(UM_KERN_ERR "handle_trap - failed to wait at "
180 "end of syscall, errno = %d, status = %d\n",
186 handle_syscall(regs);
189 extern int __syscall_stub_start;
191 static int userspace_tramp(void *stack)
196 ptrace(PTRACE_TRACEME, 0, 0, 0);
198 signal(SIGTERM, SIG_DFL);
199 signal(SIGWINCH, SIG_IGN);
200 err = set_interval();
202 printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
203 "errno = %d\n", err);
209 * This has a pte, but it can't be mapped in with the usual
210 * tlb_flush mechanism because this is part of that mechanism
213 unsigned long long offset;
214 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
215 addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
216 PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
217 if (addr == MAP_FAILED) {
218 printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
219 "errno = %d\n", STUB_CODE, errno);
224 fd = phys_mapping(to_phys(stack), &offset);
225 addr = mmap((void *) STUB_DATA,
226 UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
227 MAP_FIXED | MAP_SHARED, fd, offset);
228 if (addr == MAP_FAILED) {
229 printk(UM_KERN_ERR "mapping segfault stack "
230 "at 0x%lx failed, errno = %d\n",
236 if (!ptrace_faultinfo && (stack != NULL)) {
239 unsigned long v = STUB_CODE +
240 (unsigned long) stub_segv_handler -
241 (unsigned long) &__syscall_stub_start;
243 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
244 sigemptyset(&sa.sa_mask);
245 sa.sa_flags = SA_ONSTACK | SA_NODEFER;
246 sa.sa_handler = (void *) v;
247 sa.sa_restorer = NULL;
248 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
249 printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
250 "handler failed - errno = %d\n", errno);
255 kill(os_getpid(), SIGSTOP);
259 /* Each element set once, and only accessed by a single processor anyway */
262 int userspace_pid[NR_CPUS];
264 int start_userspace(unsigned long stub_stack)
268 int pid, status, n, flags, err;
270 stack = mmap(NULL, UM_KERN_PAGE_SIZE,
271 PROT_READ | PROT_WRITE | PROT_EXEC,
272 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
273 if (stack == MAP_FAILED) {
275 printk(UM_KERN_ERR "start_userspace : mmap failed, "
276 "errno = %d\n", errno);
280 sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
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) == SIGVTALRM));
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 struct itimerval timer;
338 unsigned long long nsecs, now;
339 int err, status, op, pid = userspace_pid[0];
340 /* To prevent races if using_sysemu changes under us.*/
341 int local_using_sysemu;
343 if (getitimer(ITIMER_VIRTUAL, &timer))
344 printk("Failed to get itimer, errno = %d\n", errno);
345 nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
346 timer.it_value.tv_usec * UM_NSEC_PER_USEC;
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.
358 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
361 /* Now we set local_using_sysemu to be used for one loop */
362 local_using_sysemu = get_using_sysemu();
364 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
365 singlestepping(NULL));
367 if (ptrace(op, pid, 0, 0)) {
368 printk(UM_KERN_ERR "userspace - ptrace continue "
369 "failed, op = %d, errno = %d\n", op, errno);
373 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
375 printk(UM_KERN_ERR "userspace - wait failed, "
376 "errno = %d\n", errno);
381 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
382 printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
383 "errno = %d\n", errno);
387 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
389 if (WIFSTOPPED(status)) {
390 int sig = WSTOPSIG(status);
393 if (PTRACE_FULL_FAULTINFO ||
395 get_skas_faultinfo(pid,
397 (*sig_info[SIGSEGV])(SIGSEGV, regs);
399 else handle_segv(pid, regs);
402 handle_trap(pid, regs, local_using_sysemu);
405 relay_signal(SIGTRAP, regs);
412 (*sig_info[sig])(sig, regs);
414 nsecs = timer.it_value.tv_sec *
416 timer.it_value.tv_usec *
426 (*sig_info[sig])(sig, regs);
430 printk(UM_KERN_ERR "userspace - child stopped "
431 "with signal %d\n", sig);
434 pid = userspace_pid[0];
437 /* Avoid -ERESTARTSYS handling in host */
438 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
439 PT_SYSCALL_NR(regs->gp) = -1;
444 static unsigned long thread_regs[MAX_REG_NR];
446 static int __init init_thread_regs(void)
448 get_safe_registers(thread_regs);
449 /* Set parent's instruction pointer to start of clone-stub */
450 thread_regs[REGS_IP_INDEX] = STUB_CODE +
451 (unsigned long) stub_clone_handler -
452 (unsigned long) &__syscall_stub_start;
453 thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
455 #ifdef __SIGNAL_FRAMESIZE
456 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
461 __initcall(init_thread_regs);
463 int copy_context_skas0(unsigned long new_stack, int pid)
465 struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
467 unsigned long current_stack = current_stub_stack();
468 struct stub_data *data = (struct stub_data *) current_stack;
469 struct stub_data *child_data = (struct stub_data *) new_stack;
470 unsigned long long new_offset;
471 int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
474 * prepare offset and fd of child's stack as argument for parent's
475 * and child's mmap2 calls
477 *data = ((struct stub_data) { .offset = MMAP_OFFSET(new_offset),
479 .timer = ((struct itimerval)
481 .it_interval = tv }) });
483 err = ptrace_setregs(pid, thread_regs);
486 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
487 "failed, pid = %d, errno = %d\n", pid, -err);
491 /* set a well known return code for detection of child write failure */
492 child_data->err = 12345678;
495 * Wait, until parent has finished its work: read child's pid from
496 * parent's stack, and check, if bad result.
498 err = ptrace(PTRACE_CONT, pid, 0, 0);
501 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
502 "errno = %d\n", pid, errno);
510 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
516 * Wait, until child has finished too: read child's result from
517 * child's stack and check it.
520 if (child_data->err != STUB_DATA) {
521 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
522 "error %ld\n", child_data->err);
523 err = child_data->err;
527 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
528 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
530 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
531 "failed, errno = %d\n", errno);
538 os_kill_ptraced_process(pid, 1);
543 * This is used only, if stub pages are needed, while proc_mm is
544 * available. Opening /proc/mm creates a new mm_context, which lacks
545 * the stub-pages. Thus, we map them using /proc/mm-fd
547 int map_stub_pages(int fd, unsigned long code, unsigned long data,
550 struct proc_mm_op mmop;
552 unsigned long long code_offset;
553 int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
556 mmop = ((struct proc_mm_op) { .op = MM_MMAP,
560 .len = UM_KERN_PAGE_SIZE,
562 .flags = MAP_FIXED | MAP_PRIVATE,
564 .offset = code_offset
566 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
567 if (n != sizeof(mmop)) {
569 printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
570 "offset = %llx\n", code, code_fd,
571 (unsigned long long) code_offset);
572 printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
573 "failed, err = %d\n", n);
578 unsigned long long map_offset;
579 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
580 mmop = ((struct proc_mm_op)
585 .len = UM_KERN_PAGE_SIZE,
586 .prot = PROT_READ | PROT_WRITE,
587 .flags = MAP_FIXED | MAP_SHARED,
591 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
592 if (n != sizeof(mmop)) {
594 printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
595 "data failed, err = %d\n", n);
603 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
605 (*buf)[0].JB_IP = (unsigned long) handler;
606 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
610 #define INIT_JMP_NEW_THREAD 0
611 #define INIT_JMP_CALLBACK 1
612 #define INIT_JMP_HALT 2
613 #define INIT_JMP_REBOOT 3
615 void switch_threads(jmp_buf *me, jmp_buf *you)
617 if (UML_SETJMP(me) == 0)
621 static jmp_buf initial_jmpbuf;
623 /* XXX Make these percpu */
624 static void (*cb_proc)(void *arg);
626 static jmp_buf *cb_back;
628 int start_idle_thread(void *stack, jmp_buf *switch_buf)
632 set_handler(SIGWINCH, (__sighandler_t) sig_handler,
633 SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGVTALRM, -1);
636 * Can't use UML_SETJMP or UML_LONGJMP here because they save
637 * and restore signals, with the possible side-effect of
638 * trying to handle any signals which came when they were
639 * blocked, which can't be done on this stack.
640 * Signals must be blocked when jumping back here and restored
641 * after returning to the jumper.
643 n = setjmp(initial_jmpbuf);
645 case INIT_JMP_NEW_THREAD:
646 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
647 (*switch_buf)[0].JB_SP = (unsigned long) stack +
648 UM_THREAD_SIZE - sizeof(void *);
650 case INIT_JMP_CALLBACK:
652 longjmp(*cb_back, 1);
657 case INIT_JMP_REBOOT:
661 printk(UM_KERN_ERR "Bad sigsetjmp return in "
662 "start_idle_thread - %d\n", n);
665 longjmp(*switch_buf, 1);
668 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
677 if (UML_SETJMP(&here) == 0)
678 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
689 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
692 void reboot_skas(void)
695 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
698 void __switch_mm(struct mm_id *mm_idp)
702 /* FIXME: need cpu pid in __switch_mm */
704 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
707 printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
708 "failed, errno = %d\n", errno);
712 else userspace_pid[0] = mm_idp->u.pid;