linux/arch/um/os-Linux/skas/process.c

729 lines
18 KiB
C
Raw Normal View History

/*
* Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
* Licensed under the GPL
*/
#include <stdlib.h>
#include <unistd.h>
#include <sched.h>
#include <errno.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <asm/unistd.h>
#include "as-layout.h"
#include "chan_user.h"
#include "kern_constants.h"
#include "kern_util.h"
#include "mem.h"
#include "os.h"
#include "process.h"
#include "proc_mm.h"
#include "ptrace_user.h"
#include "registers.h"
#include "skas.h"
#include "skas_ptrace.h"
#include "user.h"
#include "sysdep/stub.h"
int is_skas_winch(int pid, int fd, void *data)
{
if (pid != getpgrp())
return 0;
register_winch_irq(-1, fd, -1, data, 0);
return 1;
}
static int ptrace_dump_regs(int pid)
{
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
unsigned long regs[MAX_REG_NR];
int i;
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
return -errno;
printk(UM_KERN_ERR "Stub registers -\n");
for (i = 0; i < ARRAY_SIZE(regs); i++)
printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
return 0;
}
/*
* Signals that are OK to receive in the stub - we'll just continue it.
* SIGWINCH will happen when UML is inside a detached screen.
*/
#define STUB_SIG_MASK (1 << SIGVTALRM)
/* Signals that the stub will finish with - anything else is an error */
#define STUB_DONE_MASK (1 << SIGTRAP)
void wait_stub_done(int pid)
{
int n, status, err;
while (1) {
CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
if ((n < 0) || !WIFSTOPPED(status))
goto bad_wait;
if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
break;
err = ptrace(PTRACE_CONT, pid, 0, 0);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err) {
printk(UM_KERN_ERR "wait_stub_done : continue failed, "
"errno = %d\n", errno);
fatal_sigsegv();
}
}
if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
return;
bad_wait:
err = ptrace_dump_regs(pid);
if (err)
printk(UM_KERN_ERR "Failed to get registers from stub, "
"errno = %d\n", -err);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
"pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
status);
fatal_sigsegv();
}
extern unsigned long current_stub_stack(void);
void get_skas_faultinfo(int pid, struct faultinfo * fi)
{
int err;
if (ptrace_faultinfo) {
err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err) {
printk(UM_KERN_ERR "get_skas_faultinfo - "
"PTRACE_FAULTINFO failed, errno = %d\n", errno);
fatal_sigsegv();
}
/* Special handling for i386, which has different structs */
if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
sizeof(struct faultinfo) -
sizeof(struct ptrace_faultinfo));
}
else {
uml: fix FP register corruption Commit ee3d9bd4de1ed93d2a7ee41c331ed30a1c7b8acd ("uml: simplify SIGSEGV handling"), while greatly simplifying the kernel SIGSEGV handler that runs in the process address space, introduced a bug which corrupts FP state in the process. Previously, the SIGSEGV handler called the sigreturn system call by hand - it couldn't return through the restorer provided to it because that could try to call the libc restorer which likely wouldn't exist in the process address space. So, it blocked off some signals, including SIGUSR1, on entry to the SIGSEGV handler, queued a SIGUSR1 to itself, and invoked sigreturn. The SIGUSR1 was delivered, and was visible to the UML kernel after sigreturn finished. The commit eliminated the signal masking and the call to sigreturn. The handler simply hits itself with a SIGTRAP to let the UML kernel know that it is finished. UML then restores the process registers, which effectively longjmps the process out of the signal handler, skipping sigreturn's restoring of register state and the signal mask. The bug is that the host apparently sets used_fp to 0 when it saves the process FP state in the sigcontext on the process signal stack. Thus, when the process is longjmped out of the handler, its FP state is corrupt because it wasn't saved on the context switch to the UML kernel. This manifested itself as sleep hanging. For some reason, sleep uses floating point in order to calculate the sleep interval. When a page fault corrupts its FP state, it is faked into essentially sleeping forever. This patch saves the FP state before entering the SIGSEGV handler and restores it afterwards. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-24 00:23:49 +01:00
unsigned long fpregs[FP_SIZE];
err = get_fp_registers(pid, fpregs);
if (err < 0) {
printk(UM_KERN_ERR "save_fp_registers returned %d\n",
err);
fatal_sigsegv();
}
err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err) {
printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
"errno = %d\n", pid, errno);
fatal_sigsegv();
}
wait_stub_done(pid);
/*
* faultinfo is prepared by the stub-segv-handler at start of
* the stub stack page. We just have to copy it.
*/
memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
uml: fix FP register corruption Commit ee3d9bd4de1ed93d2a7ee41c331ed30a1c7b8acd ("uml: simplify SIGSEGV handling"), while greatly simplifying the kernel SIGSEGV handler that runs in the process address space, introduced a bug which corrupts FP state in the process. Previously, the SIGSEGV handler called the sigreturn system call by hand - it couldn't return through the restorer provided to it because that could try to call the libc restorer which likely wouldn't exist in the process address space. So, it blocked off some signals, including SIGUSR1, on entry to the SIGSEGV handler, queued a SIGUSR1 to itself, and invoked sigreturn. The SIGUSR1 was delivered, and was visible to the UML kernel after sigreturn finished. The commit eliminated the signal masking and the call to sigreturn. The handler simply hits itself with a SIGTRAP to let the UML kernel know that it is finished. UML then restores the process registers, which effectively longjmps the process out of the signal handler, skipping sigreturn's restoring of register state and the signal mask. The bug is that the host apparently sets used_fp to 0 when it saves the process FP state in the sigcontext on the process signal stack. Thus, when the process is longjmped out of the handler, its FP state is corrupt because it wasn't saved on the context switch to the UML kernel. This manifested itself as sleep hanging. For some reason, sleep uses floating point in order to calculate the sleep interval. When a page fault corrupts its FP state, it is faked into essentially sleeping forever. This patch saves the FP state before entering the SIGSEGV handler and restores it afterwards. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-24 00:23:49 +01:00
err = put_fp_registers(pid, fpregs);
if (err < 0) {
printk(UM_KERN_ERR "put_fp_registers returned %d\n",
err);
fatal_sigsegv();
}
}
}
static void handle_segv(int pid, struct uml_pt_regs * regs)
{
get_skas_faultinfo(pid, &regs->faultinfo);
segv(regs->faultinfo, 0, 1, NULL);
}
/*
* To use the same value of using_sysemu as the caller, ask it that value
* (in local_using_sysemu
*/
static void handle_trap(int pid, struct uml_pt_regs *regs,
int local_using_sysemu)
{
int err, status;
if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
fatal_sigsegv();
/* Mark this as a syscall */
UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
if (!local_using_sysemu)
{
err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
__NR_getpid);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err < 0) {
printk(UM_KERN_ERR "handle_trap - nullifying syscall "
"failed, errno = %d\n", errno);
fatal_sigsegv();
}
err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err < 0) {
printk(UM_KERN_ERR "handle_trap - continuing to end of "
"syscall failed, errno = %d\n", errno);
fatal_sigsegv();
}
CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
if ((err < 0) || !WIFSTOPPED(status) ||
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
(WSTOPSIG(status) != SIGTRAP + 0x80)) {
err = ptrace_dump_regs(pid);
if (err)
printk(UM_KERN_ERR "Failed to get registers "
"from process, errno = %d\n", -err);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
printk(UM_KERN_ERR "handle_trap - failed to wait at "
"end of syscall, errno = %d, status = %d\n",
errno, status);
fatal_sigsegv();
}
}
handle_syscall(regs);
}
extern int __syscall_stub_start;
static int userspace_tramp(void *stack)
{
void *addr;
int err;
ptrace(PTRACE_TRACEME, 0, 0, 0);
signal(SIGTERM, SIG_DFL);
signal(SIGWINCH, SIG_IGN);
err = set_interval();
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err) {
printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
"errno = %d\n", err);
exit(1);
}
if (!proc_mm) {
/*
* This has a pte, but it can't be mapped in with the usual
* tlb_flush mechanism because this is part of that mechanism
*/
int fd;
unsigned long long offset;
fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
if (addr == MAP_FAILED) {
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
"errno = %d\n", STUB_CODE, errno);
exit(1);
}
if (stack != NULL) {
fd = phys_mapping(to_phys(stack), &offset);
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
addr = mmap((void *) STUB_DATA,
UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
MAP_FIXED | MAP_SHARED, fd, offset);
if (addr == MAP_FAILED) {
printk(UM_KERN_ERR "mapping segfault stack "
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
"at 0x%lx failed, errno = %d\n",
STUB_DATA, errno);
exit(1);
}
}
}
if (!ptrace_faultinfo && (stack != NULL)) {
struct sigaction sa;
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
unsigned long v = STUB_CODE +
(unsigned long) stub_segv_handler -
(unsigned long) &__syscall_stub_start;
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
sigemptyset(&sa.sa_mask);
sa.sa_flags = SA_ONSTACK | SA_NODEFER;
sa.sa_handler = (void *) v;
sa.sa_restorer = NULL;
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (sigaction(SIGSEGV, &sa, NULL) < 0) {
printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
"handler failed - errno = %d\n", errno);
exit(1);
}
}
kill(os_getpid(), SIGSTOP);
return 0;
}
/* Each element set once, and only accessed by a single processor anyway */
#undef NR_CPUS
#define NR_CPUS 1
int userspace_pid[NR_CPUS];
int start_userspace(unsigned long stub_stack)
{
void *stack;
unsigned long sp;
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
int pid, status, n, flags, err;
stack = mmap(NULL, UM_KERN_PAGE_SIZE,
PROT_READ | PROT_WRITE | PROT_EXEC,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (stack == MAP_FAILED) {
err = -errno;
printk(UM_KERN_ERR "start_userspace : mmap failed, "
"errno = %d\n", errno);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
return err;
}
sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
flags = CLONE_FILES;
if (proc_mm)
flags |= CLONE_VM;
else
flags |= SIGCHLD;
pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (pid < 0) {
err = -errno;
printk(UM_KERN_ERR "start_userspace : clone failed, "
"errno = %d\n", errno);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
return err;
}
do {
CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (n < 0) {
err = -errno;
printk(UM_KERN_ERR "start_userspace : wait failed, "
"errno = %d\n", errno);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
goto out_kill;
}
} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
err = -EINVAL;
printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
"status = %d\n", status);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
goto out_kill;
}
if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
(void *) PTRACE_O_TRACESYSGOOD) < 0) {
err = -errno;
printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
"failed, errno = %d\n", errno);
goto out_kill;
}
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
err = -errno;
printk(UM_KERN_ERR "start_userspace : munmap failed, "
"errno = %d\n", errno);
goto out_kill;
}
return pid;
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
out_kill:
os_kill_ptraced_process(pid, 1);
return err;
}
void userspace(struct uml_pt_regs *regs)
{
struct itimerval timer;
unsigned long long nsecs, now;
int err, status, op, pid = userspace_pid[0];
/* To prevent races if using_sysemu changes under us.*/
int local_using_sysemu;
if (getitimer(ITIMER_VIRTUAL, &timer))
printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno);
nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
timer.it_value.tv_usec * UM_NSEC_PER_USEC;
nsecs += os_nsecs();
while (1) {
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
/*
* This can legitimately fail if the process loads a
* bogus value into a segment register. It will
* segfault and PTRACE_GETREGS will read that value
* out of the process. However, PTRACE_SETREGS will
* fail. In this case, there is nothing to do but
* just kill the process.
*/
if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
fatal_sigsegv();
/* Now we set local_using_sysemu to be used for one loop */
local_using_sysemu = get_using_sysemu();
op = SELECT_PTRACE_OPERATION(local_using_sysemu,
singlestepping(NULL));
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (ptrace(op, pid, 0, 0)) {
printk(UM_KERN_ERR "userspace - ptrace continue "
"failed, op = %d, errno = %d\n", op, errno);
fatal_sigsegv();
}
CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err < 0) {
printk(UM_KERN_ERR "userspace - wait failed, "
"errno = %d\n", errno);
fatal_sigsegv();
}
regs->is_user = 1;
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
"errno = %d\n", errno);
fatal_sigsegv();
}
UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
if (WIFSTOPPED(status)) {
int sig = WSTOPSIG(status);
switch (sig) {
case SIGSEGV:
if (PTRACE_FULL_FAULTINFO ||
!ptrace_faultinfo) {
get_skas_faultinfo(pid,
&regs->faultinfo);
(*sig_info[SIGSEGV])(SIGSEGV, regs);
}
else handle_segv(pid, regs);
break;
case SIGTRAP + 0x80:
handle_trap(pid, regs, local_using_sysemu);
break;
case SIGTRAP:
relay_signal(SIGTRAP, regs);
break;
case SIGVTALRM:
now = os_nsecs();
if (now < nsecs)
break;
block_signals();
(*sig_info[sig])(sig, regs);
unblock_signals();
nsecs = timer.it_value.tv_sec *
UM_NSEC_PER_SEC +
timer.it_value.tv_usec *
UM_NSEC_PER_USEC;
nsecs += os_nsecs();
break;
case SIGIO:
case SIGILL:
case SIGBUS:
case SIGFPE:
case SIGWINCH:
block_signals();
(*sig_info[sig])(sig, regs);
unblock_signals();
break;
default:
printk(UM_KERN_ERR "userspace - child stopped "
"with signal %d\n", sig);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
fatal_sigsegv();
}
pid = userspace_pid[0];
interrupt_end();
/* Avoid -ERESTARTSYS handling in host */
if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
PT_SYSCALL_NR(regs->gp) = -1;
}
}
}
static unsigned long thread_regs[MAX_REG_NR];
static int __init init_thread_regs(void)
{
get_safe_registers(thread_regs);
/* Set parent's instruction pointer to start of clone-stub */
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
thread_regs[REGS_IP_INDEX] = STUB_CODE +
(unsigned long) stub_clone_handler -
(unsigned long) &__syscall_stub_start;
uml: fix stub address calculations The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the cold. I'd rather not duplicate the four known host vmsplit cases for each of these symbols. I'd also like to calculate them based on the highest userspace address. The Kconfig language seems not to allow calculation of hex constants, so I moved this to as-layout.h. CONFIG_STUB_CODE, CONFIG_STUB_DATA, and CONFIG_STUB_START are now gone. In their place are STUB_CODE, STUB_DATA, and STUB_START in as-layout.h. i386 and x86_64 seem to differ as to whether an unadorned constant is an int or a long, so I cast them to unsigned long so they can be printed consistently. However, they are also used in stub.S, where C types don't work so well. So, there are ASM_ versions of these constants for use in stub.S. I also ifdef-ed the non-asm-friendly portion of as-layout.h. With this in place, most of the rest of this patch is changing CONFIG_STUB_* to STUB_*, except in stub.S, where they are changed to ASM_STUB_*. defconfig has the old symbols deleted. I also print these addresses out in case there is any problem mapping them on the host. The two stub.S files had some trailing whitespace, so that is cleaned up here. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 10:27:33 +02:00
thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
sizeof(void *);
#ifdef __SIGNAL_FRAMESIZE
thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
#endif
return 0;
}
__initcall(init_thread_regs);
int copy_context_skas0(unsigned long new_stack, int pid)
{
struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
int err;
unsigned long current_stack = current_stub_stack();
struct stub_data *data = (struct stub_data *) current_stack;
struct stub_data *child_data = (struct stub_data *) new_stack;
unsigned long long new_offset;
int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
/*
* prepare offset and fd of child's stack as argument for parent's
* and child's mmap2 calls
*/
*data = ((struct stub_data) { .offset = MMAP_OFFSET(new_offset),
.fd = new_fd,
.timer = ((struct itimerval)
{ .it_value = tv,
.it_interval = tv }) });
err = ptrace_setregs(pid, thread_regs);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err < 0) {
err = -errno;
printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
"failed, pid = %d, errno = %d\n", pid, -err);
return err;
}
/* set a well known return code for detection of child write failure */
child_data->err = 12345678;
/*
* Wait, until parent has finished its work: read child's pid from
* parent's stack, and check, if bad result.
*/
err = ptrace(PTRACE_CONT, pid, 0, 0);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err) {
err = -errno;
printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
"errno = %d\n", pid, errno);
return err;
}
wait_stub_done(pid);
pid = data->err;
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (pid < 0) {
printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
"error %d\n", -pid);
return pid;
}
/*
* Wait, until child has finished too: read child's result from
* child's stack and check it.
*/
wait_stub_done(pid);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (child_data->err != STUB_DATA) {
printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
"error %ld\n", child_data->err);
err = child_data->err;
goto out_kill;
}
if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
(void *)PTRACE_O_TRACESYSGOOD) < 0) {
err = -errno;
printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
"failed, errno = %d\n", errno);
goto out_kill;
}
return pid;
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
out_kill:
os_kill_ptraced_process(pid, 1);
return err;
}
/*
* This is used only, if stub pages are needed, while proc_mm is
* available. Opening /proc/mm creates a new mm_context, which lacks
* the stub-pages. Thus, we map them using /proc/mm-fd
*/
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
int map_stub_pages(int fd, unsigned long code, unsigned long data,
unsigned long stack)
{
struct proc_mm_op mmop;
int n;
unsigned long long code_offset;
int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
&code_offset);
mmop = ((struct proc_mm_op) { .op = MM_MMAP,
.u =
{ .mmap =
{ .addr = code,
.len = UM_KERN_PAGE_SIZE,
.prot = PROT_EXEC,
.flags = MAP_FIXED | MAP_PRIVATE,
.fd = code_fd,
.offset = code_offset
} } });
CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
if (n != sizeof(mmop)) {
n = errno;
printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
"offset = %llx\n", code, code_fd,
(unsigned long long) code_offset);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
"failed, err = %d\n", n);
return -n;
}
if (stack) {
unsigned long long map_offset;
int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
mmop = ((struct proc_mm_op)
{ .op = MM_MMAP,
.u =
{ .mmap =
{ .addr = data,
.len = UM_KERN_PAGE_SIZE,
.prot = PROT_READ | PROT_WRITE,
.flags = MAP_FIXED | MAP_SHARED,
.fd = map_fd,
.offset = map_offset
} } });
CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (n != sizeof(mmop)) {
n = errno;
printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
"data failed, err = %d\n", n);
return -n;
}
}
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
return 0;
}
[PATCH] uml: thread creation tidying fork on UML has always somewhat subtle. The underlying cause has been the need to initialize a stack for the new process. The only portable way to initialize a new stack is to set it as the alternate signal stack and take a signal. The signal handler does whatever initialization is needed and jumps back to the original stack, where the fork processing is finished. The basic context switching mechanism is a jmp_buf for each process. You switch to a new process by longjmping to its jmp_buf. Now that UML has its own implementation of setjmp and longjmp, and I can poke around inside a jmp_buf without fear that libc will change the structure, a much simpler mechanism is possible. The jmpbuf can simply be initialized by hand. This eliminates - the need to set up and remove the alternate signal stack sending and handling a signal the signal blocking needed around the stack switching, since there is no stack switching setting up the jmp_buf needed to jump back to the original stack after the new one is set up In addition, since jmp_buf is now defined by UML, and not by libc, it can be embedded in the thread struct. This makes it unnecessary to have it exist on the stack, where it used to be. It also simplifies interfaces, since the switch jmp_buf used to be a void * inside the thread struct, and functions which took it as an argument needed to define a jmp_buf variable and assign it from the void *. Signed-off-by: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 10:50:40 +02:00
void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
{
[PATCH] uml: thread creation tidying fork on UML has always somewhat subtle. The underlying cause has been the need to initialize a stack for the new process. The only portable way to initialize a new stack is to set it as the alternate signal stack and take a signal. The signal handler does whatever initialization is needed and jumps back to the original stack, where the fork processing is finished. The basic context switching mechanism is a jmp_buf for each process. You switch to a new process by longjmping to its jmp_buf. Now that UML has its own implementation of setjmp and longjmp, and I can poke around inside a jmp_buf without fear that libc will change the structure, a much simpler mechanism is possible. The jmpbuf can simply be initialized by hand. This eliminates - the need to set up and remove the alternate signal stack sending and handling a signal the signal blocking needed around the stack switching, since there is no stack switching setting up the jmp_buf needed to jump back to the original stack after the new one is set up In addition, since jmp_buf is now defined by UML, and not by libc, it can be embedded in the thread struct. This makes it unnecessary to have it exist on the stack, where it used to be. It also simplifies interfaces, since the switch jmp_buf used to be a void * inside the thread struct, and functions which took it as an argument needed to define a jmp_buf variable and assign it from the void *. Signed-off-by: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 10:50:40 +02:00
(*buf)[0].JB_IP = (unsigned long) handler;
(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
sizeof(void *);
}
#define INIT_JMP_NEW_THREAD 0
[PATCH] uml: thread creation tidying fork on UML has always somewhat subtle. The underlying cause has been the need to initialize a stack for the new process. The only portable way to initialize a new stack is to set it as the alternate signal stack and take a signal. The signal handler does whatever initialization is needed and jumps back to the original stack, where the fork processing is finished. The basic context switching mechanism is a jmp_buf for each process. You switch to a new process by longjmping to its jmp_buf. Now that UML has its own implementation of setjmp and longjmp, and I can poke around inside a jmp_buf without fear that libc will change the structure, a much simpler mechanism is possible. The jmpbuf can simply be initialized by hand. This eliminates - the need to set up and remove the alternate signal stack sending and handling a signal the signal blocking needed around the stack switching, since there is no stack switching setting up the jmp_buf needed to jump back to the original stack after the new one is set up In addition, since jmp_buf is now defined by UML, and not by libc, it can be embedded in the thread struct. This makes it unnecessary to have it exist on the stack, where it used to be. It also simplifies interfaces, since the switch jmp_buf used to be a void * inside the thread struct, and functions which took it as an argument needed to define a jmp_buf variable and assign it from the void *. Signed-off-by: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 10:50:40 +02:00
#define INIT_JMP_CALLBACK 1
#define INIT_JMP_HALT 2
#define INIT_JMP_REBOOT 3
[PATCH] uml: thread creation tidying fork on UML has always somewhat subtle. The underlying cause has been the need to initialize a stack for the new process. The only portable way to initialize a new stack is to set it as the alternate signal stack and take a signal. The signal handler does whatever initialization is needed and jumps back to the original stack, where the fork processing is finished. The basic context switching mechanism is a jmp_buf for each process. You switch to a new process by longjmping to its jmp_buf. Now that UML has its own implementation of setjmp and longjmp, and I can poke around inside a jmp_buf without fear that libc will change the structure, a much simpler mechanism is possible. The jmpbuf can simply be initialized by hand. This eliminates - the need to set up and remove the alternate signal stack sending and handling a signal the signal blocking needed around the stack switching, since there is no stack switching setting up the jmp_buf needed to jump back to the original stack after the new one is set up In addition, since jmp_buf is now defined by UML, and not by libc, it can be embedded in the thread struct. This makes it unnecessary to have it exist on the stack, where it used to be. It also simplifies interfaces, since the switch jmp_buf used to be a void * inside the thread struct, and functions which took it as an argument needed to define a jmp_buf variable and assign it from the void *. Signed-off-by: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 10:50:40 +02:00
void switch_threads(jmp_buf *me, jmp_buf *you)
{
if (UML_SETJMP(me) == 0)
[PATCH] uml: thread creation tidying fork on UML has always somewhat subtle. The underlying cause has been the need to initialize a stack for the new process. The only portable way to initialize a new stack is to set it as the alternate signal stack and take a signal. The signal handler does whatever initialization is needed and jumps back to the original stack, where the fork processing is finished. The basic context switching mechanism is a jmp_buf for each process. You switch to a new process by longjmping to its jmp_buf. Now that UML has its own implementation of setjmp and longjmp, and I can poke around inside a jmp_buf without fear that libc will change the structure, a much simpler mechanism is possible. The jmpbuf can simply be initialized by hand. This eliminates - the need to set up and remove the alternate signal stack sending and handling a signal the signal blocking needed around the stack switching, since there is no stack switching setting up the jmp_buf needed to jump back to the original stack after the new one is set up In addition, since jmp_buf is now defined by UML, and not by libc, it can be embedded in the thread struct. This makes it unnecessary to have it exist on the stack, where it used to be. It also simplifies interfaces, since the switch jmp_buf used to be a void * inside the thread struct, and functions which took it as an argument needed to define a jmp_buf variable and assign it from the void *. Signed-off-by: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 10:50:40 +02:00
UML_LONGJMP(you, 1);
}
static jmp_buf initial_jmpbuf;
/* XXX Make these percpu */
static void (*cb_proc)(void *arg);
static void *cb_arg;
static jmp_buf *cb_back;
[PATCH] uml: thread creation tidying fork on UML has always somewhat subtle. The underlying cause has been the need to initialize a stack for the new process. The only portable way to initialize a new stack is to set it as the alternate signal stack and take a signal. The signal handler does whatever initialization is needed and jumps back to the original stack, where the fork processing is finished. The basic context switching mechanism is a jmp_buf for each process. You switch to a new process by longjmping to its jmp_buf. Now that UML has its own implementation of setjmp and longjmp, and I can poke around inside a jmp_buf without fear that libc will change the structure, a much simpler mechanism is possible. The jmpbuf can simply be initialized by hand. This eliminates - the need to set up and remove the alternate signal stack sending and handling a signal the signal blocking needed around the stack switching, since there is no stack switching setting up the jmp_buf needed to jump back to the original stack after the new one is set up In addition, since jmp_buf is now defined by UML, and not by libc, it can be embedded in the thread struct. This makes it unnecessary to have it exist on the stack, where it used to be. It also simplifies interfaces, since the switch jmp_buf used to be a void * inside the thread struct, and functions which took it as an argument needed to define a jmp_buf variable and assign it from the void *. Signed-off-by: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 10:50:40 +02:00
int start_idle_thread(void *stack, jmp_buf *switch_buf)
{
int n;
set_handler(SIGWINCH, (__sighandler_t) sig_handler,
SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGVTALRM, -1);
/*
* Can't use UML_SETJMP or UML_LONGJMP here because they save
* and restore signals, with the possible side-effect of
* trying to handle any signals which came when they were
* blocked, which can't be done on this stack.
* Signals must be blocked when jumping back here and restored
* after returning to the jumper.
*/
n = setjmp(initial_jmpbuf);
switch (n) {
case INIT_JMP_NEW_THREAD:
[PATCH] uml: thread creation tidying fork on UML has always somewhat subtle. The underlying cause has been the need to initialize a stack for the new process. The only portable way to initialize a new stack is to set it as the alternate signal stack and take a signal. The signal handler does whatever initialization is needed and jumps back to the original stack, where the fork processing is finished. The basic context switching mechanism is a jmp_buf for each process. You switch to a new process by longjmping to its jmp_buf. Now that UML has its own implementation of setjmp and longjmp, and I can poke around inside a jmp_buf without fear that libc will change the structure, a much simpler mechanism is possible. The jmpbuf can simply be initialized by hand. This eliminates - the need to set up and remove the alternate signal stack sending and handling a signal the signal blocking needed around the stack switching, since there is no stack switching setting up the jmp_buf needed to jump back to the original stack after the new one is set up In addition, since jmp_buf is now defined by UML, and not by libc, it can be embedded in the thread struct. This makes it unnecessary to have it exist on the stack, where it used to be. It also simplifies interfaces, since the switch jmp_buf used to be a void * inside the thread struct, and functions which took it as an argument needed to define a jmp_buf variable and assign it from the void *. Signed-off-by: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 10:50:40 +02:00
(*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
(*switch_buf)[0].JB_SP = (unsigned long) stack +
UM_THREAD_SIZE - sizeof(void *);
break;
case INIT_JMP_CALLBACK:
(*cb_proc)(cb_arg);
longjmp(*cb_back, 1);
break;
case INIT_JMP_HALT:
kmalloc_ok = 0;
return 0;
case INIT_JMP_REBOOT:
kmalloc_ok = 0;
return 1;
default:
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
printk(UM_KERN_ERR "Bad sigsetjmp return in "
"start_idle_thread - %d\n", n);
fatal_sigsegv();
}
longjmp(*switch_buf, 1);
}
void initial_thread_cb_skas(void (*proc)(void *), void *arg)
{
jmp_buf here;
cb_proc = proc;
cb_arg = arg;
cb_back = &here;
block_signals();
if (UML_SETJMP(&here) == 0)
UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
unblock_signals();
cb_proc = NULL;
cb_arg = NULL;
cb_back = NULL;
}
void halt_skas(void)
{
block_signals();
UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
}
void reboot_skas(void)
{
block_signals();
UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
}
void __switch_mm(struct mm_id *mm_idp)
{
int err;
/* FIXME: need cpu pid in __switch_mm */
if (proc_mm) {
err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
mm_idp->u.mm_fd);
uml: kill processes instead of panicing kernel UML was panicing in the case of failures of libc calls which shouldn't happen. This is an overreaction since a failure from libc doesn't normally mean that kernel data structures are in an unknown state. Instead, the current process should just be killed if there is no way to recover. The case that prompted this was a failure of PTRACE_SETREGS restoring the same state that was read by PTRACE_GETREGS. It appears that when a process tries to load a bogus value into a segment register, it segfaults (as expected) and the value is actually loaded and is seen by PTRACE_GETREGS (not expected). This case is fixed by forcing a fatal SIGSEGV on the process so that it immediately dies. fatal_sigsegv was added for this purpose. It was declared as noreturn, so in order to pursuade gcc that it actually does not return, I added a call to os_dump_core (and declared it noreturn) so that I get a core file if somehow the process survives. All other calls in arch/um/os-Linux/skas/process.c got the same treatment, with failures causing the process to die instead of a kernel panic, with some exceptions. userspace_tramp exits with status 1 if anything goes wrong there. That will cause start_userspace to return an error. copy_context_skas0 and map_stub_pages also now return errors instead of panicing. Callers of thes functions were changed to check for errors and do something appropriate. Usually that's to return an error to their callers. check_skas3_ptrace_faultinfo just exits since that's too early to do anything else. save_registers, restore_registers, and init_registers now return status instead of panicing on failure, with their callers doing something appropriate. There were also duplicate declarations of save_registers and restore_registers in os.h - these are gone. I noticed and fixed up some whitespace damage. Signed-off-by: Jeff Dike <jdike@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 07:30:58 +01:00
if (err) {
printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
"failed, errno = %d\n", errno);
fatal_sigsegv();
}
}
else userspace_pid[0] = mm_idp->u.pid;
}