linux/arch/um/kernel/tlb.c

536 lines
12 KiB
C
Raw Normal View History

/*
* Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
* Licensed under the GPL
*/
uml: header untangling Untangle UML headers somewhat and add some includes where they were needed explicitly, but gotten accidentally via some other header. arch/um/include/um_uaccess.h loses asm/fixmap.h because it uses no fixmap stuff and gains elf.h, because it needs FIXADDR_USER_*, and archsetjmp.h, because it needs jmp_buf. pmd_alloc_one is uninlined because it needs mm_struct, and that's inconvenient to provide in asm-um/pgtable-3level.h. elf_core_copy_fpregs is also uninlined from elf-i386.h and elf-x86_64.h, which duplicated the code anyway, to arch/um/kernel/process.c, so that the reference to current_thread doesn't pull sched.h or anything related into asm/elf.h. arch/um/sys-i386/ldt.c, arch/um/kernel/tlb.c and arch/um/kernel/skas/uaccess.c got sched.h because they dereference task_structs. Its includes of linux and asm headers got turned from "" to <>. arch/um/sys-i386/bug.c gets asm/errno.h because it needs errno constants. asm/elf-i386 gets asm/user.h because it needs user_regs_struct. asm/fixmap.h gets page.h because it needs PAGE_SIZE and PAGE_MASK and system.h for BUG_ON. asm/pgtable doesn't need sched.h. asm/processor-generic.h defined mm_segment_t, but didn't use it. So, that definition is moved to uaccess.h, which defines a bunch of mm_segment_t-related stuff. thread_info.h uses mm_segment_t, and includes uaccess.h, which causes a recursion. So, the definition is placed above the include of thread_info. in uaccess.h. thread_info.h also gets page.h because it needs PAGE_SIZE. ObCheckpatchViolationJustification - I'm not adding a typedef; I'm moving mm_segment_t from one place to another. 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 06:30:53 +00:00
#include <linux/mm.h>
#include <linux/sched.h>
#include <asm/pgtable.h>
#include <asm/tlbflush.h>
#include "as-layout.h"
#include "mem_user.h"
#include "os.h"
#include "skas.h"
#include "tlb.h"
struct host_vm_change {
struct host_vm_op {
enum { NONE, MMAP, MUNMAP, MPROTECT } type;
union {
struct {
unsigned long addr;
unsigned long len;
unsigned int prot;
int fd;
__u64 offset;
} mmap;
struct {
unsigned long addr;
unsigned long len;
} munmap;
struct {
unsigned long addr;
unsigned long len;
unsigned int prot;
} mprotect;
} u;
} ops[1];
int index;
struct mm_id *id;
void *data;
int force;
};
#define INIT_HVC(mm, force) \
((struct host_vm_change) \
{ .ops = { { .type = NONE } }, \
.id = &mm->context.id, \
.data = NULL, \
.index = 0, \
.force = force })
static int do_ops(struct host_vm_change *hvc, int end,
int finished)
{
struct host_vm_op *op;
int i, ret = 0;
for (i = 0; i < end && !ret; i++) {
op = &hvc->ops[i];
switch (op->type) {
case MMAP:
ret = map(hvc->id, op->u.mmap.addr, op->u.mmap.len,
op->u.mmap.prot, op->u.mmap.fd,
op->u.mmap.offset, finished, &hvc->data);
break;
case MUNMAP:
ret = unmap(hvc->id, op->u.munmap.addr,
op->u.munmap.len, finished, &hvc->data);
break;
case MPROTECT:
ret = protect(hvc->id, op->u.mprotect.addr,
op->u.mprotect.len, op->u.mprotect.prot,
finished, &hvc->data);
break;
default:
printk(KERN_ERR "Unknown op type %d in do_ops\n",
op->type);
break;
}
}
return ret;
}
static int add_mmap(unsigned long virt, unsigned long phys, unsigned long len,
unsigned int prot, struct host_vm_change *hvc)
{
__u64 offset;
struct host_vm_op *last;
int fd, ret = 0;
fd = phys_mapping(phys, &offset);
if (hvc->index != 0) {
last = &hvc->ops[hvc->index - 1];
if ((last->type == MMAP) &&
(last->u.mmap.addr + last->u.mmap.len == virt) &&
(last->u.mmap.prot == prot) && (last->u.mmap.fd == fd) &&
(last->u.mmap.offset + last->u.mmap.len == offset)) {
last->u.mmap.len += len;
return 0;
}
}
if (hvc->index == ARRAY_SIZE(hvc->ops)) {
ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
hvc->index = 0;
}
hvc->ops[hvc->index++] = ((struct host_vm_op)
{ .type = MMAP,
.u = { .mmap = { .addr = virt,
.len = len,
.prot = prot,
.fd = fd,
.offset = offset }
} });
return ret;
}
static int add_munmap(unsigned long addr, unsigned long len,
struct host_vm_change *hvc)
{
struct host_vm_op *last;
int ret = 0;
if (hvc->index != 0) {
last = &hvc->ops[hvc->index - 1];
if ((last->type == MUNMAP) &&
(last->u.munmap.addr + last->u.mmap.len == addr)) {
last->u.munmap.len += len;
return 0;
}
}
if (hvc->index == ARRAY_SIZE(hvc->ops)) {
ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
hvc->index = 0;
}
hvc->ops[hvc->index++] = ((struct host_vm_op)
{ .type = MUNMAP,
.u = { .munmap = { .addr = addr,
.len = len } } });
return ret;
}
static int add_mprotect(unsigned long addr, unsigned long len,
unsigned int prot, struct host_vm_change *hvc)
{
struct host_vm_op *last;
int ret = 0;
if (hvc->index != 0) {
last = &hvc->ops[hvc->index - 1];
if ((last->type == MPROTECT) &&
(last->u.mprotect.addr + last->u.mprotect.len == addr) &&
(last->u.mprotect.prot == prot)) {
last->u.mprotect.len += len;
return 0;
}
}
if (hvc->index == ARRAY_SIZE(hvc->ops)) {
ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
hvc->index = 0;
}
hvc->ops[hvc->index++] = ((struct host_vm_op)
{ .type = MPROTECT,
.u = { .mprotect = { .addr = addr,
.len = len,
.prot = prot } } });
return ret;
}
#define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
unsigned long end,
struct host_vm_change *hvc)
{
pte_t *pte;
int r, w, x, prot, ret = 0;
pte = pte_offset_kernel(pmd, addr);
do {
uml: cover stubs with a VMA Give the stubs a VMA. This allows the removal of a truly nasty kludge to make sure that mm->nr_ptes was correct in exit_mmap. The underlying problem was always that the stubs, which have ptes, and thus allocated a page table, weren't covered by a VMA. This patch fixes that by using install_special_mapping in arch_dup_mmap and activate_context to create the VMA. The stubs have to be moved, since shift_arg_pages seems to assume that the stack is the only VMA present at that point during exec, and uses vma_adjust to fiddle its VMA. However, that extends the stub VMA by the amount removed from the stack VMA. To avoid this problem, the stubs were moved to a different fixed location at the start of the address space. The init_stub_pte calls were moved from init_new_context to arch_dup_mmap because I was occasionally seeing arch_dup_mmap not being called, causing exit_mmap to die. Rather than figure out what was really happening, I decided it was cleaner to just move the calls so that there's no doubt that both the pte and VMA creation happen, no matter what. arch_exit_mmap is used to clear the stub ptes at exit time. The STUB_* constants in as-layout.h no longer depend on UM_TASK_SIZE, that that definition is removed, along with the comments complaining about gcc. Because the stubs are no longer at the top of the address space, some care is needed while flushing TLBs. update_pte_range checks for addresses in the stub range and skips them. flush_thread now issues two unmaps, one for the range before STUB_START and one for the range after STUB_END. 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 06:31:01 +00:00
if ((addr >= STUB_START) && (addr < STUB_END))
continue;
r = pte_read(*pte);
w = pte_write(*pte);
x = pte_exec(*pte);
if (!pte_young(*pte)) {
r = 0;
w = 0;
} else if (!pte_dirty(*pte))
w = 0;
prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
(x ? UM_PROT_EXEC : 0));
if (hvc->force || pte_newpage(*pte)) {
if (pte_present(*pte))
ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
PAGE_SIZE, prot, hvc);
else
ret = add_munmap(addr, PAGE_SIZE, hvc);
} else if (pte_newprot(*pte))
ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
*pte = pte_mkuptodate(*pte);
} while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
return ret;
}
static inline int update_pmd_range(pud_t *pud, unsigned long addr,
unsigned long end,
struct host_vm_change *hvc)
{
pmd_t *pmd;
unsigned long next;
int ret = 0;
pmd = pmd_offset(pud, addr);
do {
next = pmd_addr_end(addr, end);
if (!pmd_present(*pmd)) {
if (hvc->force || pmd_newpage(*pmd)) {
ret = add_munmap(addr, next - addr, hvc);
pmd_mkuptodate(*pmd);
}
}
else ret = update_pte_range(pmd, addr, next, hvc);
} while (pmd++, addr = next, ((addr < end) && !ret));
return ret;
}
static inline int update_pud_range(pgd_t *pgd, unsigned long addr,
unsigned long end,
struct host_vm_change *hvc)
{
pud_t *pud;
unsigned long next;
int ret = 0;
pud = pud_offset(pgd, addr);
do {
next = pud_addr_end(addr, end);
if (!pud_present(*pud)) {
if (hvc->force || pud_newpage(*pud)) {
ret = add_munmap(addr, next - addr, hvc);
pud_mkuptodate(*pud);
}
}
else ret = update_pmd_range(pud, addr, next, hvc);
} while (pud++, addr = next, ((addr < end) && !ret));
return ret;
}
void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
unsigned long end_addr, int force)
{
pgd_t *pgd;
struct host_vm_change hvc;
unsigned long addr = start_addr, next;
int ret = 0;
hvc = INIT_HVC(mm, force);
pgd = pgd_offset(mm, addr);
do {
next = pgd_addr_end(addr, end_addr);
if (!pgd_present(*pgd)) {
if (force || pgd_newpage(*pgd)) {
ret = add_munmap(addr, next - addr, &hvc);
pgd_mkuptodate(*pgd);
}
}
else ret = update_pud_range(pgd, addr, next, &hvc);
} while (pgd++, addr = next, ((addr < end_addr) && !ret));
if (!ret)
ret = do_ops(&hvc, hvc.index, 1);
/* This is not an else because ret is modified above */
if (ret) {
printk(KERN_ERR "fix_range_common: failed, killing current "
"process\n");
force_sig(SIGKILL, current);
}
}
int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
{
struct mm_struct *mm;
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
unsigned long addr, last;
int updated = 0, err;
mm = &init_mm;
for (addr = start; addr < end;) {
pgd = pgd_offset(mm, addr);
if (!pgd_present(*pgd)) {
last = ADD_ROUND(addr, PGDIR_SIZE);
if (last > end)
last = end;
if (pgd_newpage(*pgd)) {
updated = 1;
err = os_unmap_memory((void *) addr,
last - addr);
if (err < 0)
panic("munmap failed, errno = %d\n",
-err);
}
addr = last;
continue;
}
pud = pud_offset(pgd, addr);
if (!pud_present(*pud)) {
last = ADD_ROUND(addr, PUD_SIZE);
if (last > end)
last = end;
if (pud_newpage(*pud)) {
updated = 1;
err = os_unmap_memory((void *) addr,
last - addr);
if (err < 0)
panic("munmap failed, errno = %d\n",
-err);
}
addr = last;
continue;
}
pmd = pmd_offset(pud, addr);
if (!pmd_present(*pmd)) {
last = ADD_ROUND(addr, PMD_SIZE);
if (last > end)
last = end;
if (pmd_newpage(*pmd)) {
updated = 1;
err = os_unmap_memory((void *) addr,
last - addr);
if (err < 0)
panic("munmap failed, errno = %d\n",
-err);
}
addr = last;
continue;
}
pte = pte_offset_kernel(pmd, addr);
if (!pte_present(*pte) || pte_newpage(*pte)) {
updated = 1;
err = os_unmap_memory((void *) addr,
PAGE_SIZE);
if (err < 0)
panic("munmap failed, errno = %d\n",
-err);
if (pte_present(*pte))
map_memory(addr,
pte_val(*pte) & PAGE_MASK,
PAGE_SIZE, 1, 1, 1);
}
else if (pte_newprot(*pte)) {
updated = 1;
os_protect_memory((void *) addr, PAGE_SIZE, 1, 1, 1);
}
addr += PAGE_SIZE;
}
return updated;
}
void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
{
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
struct mm_struct *mm = vma->vm_mm;
void *flush = NULL;
int r, w, x, prot, err = 0;
struct mm_id *mm_id;
address &= PAGE_MASK;
pgd = pgd_offset(mm, address);
if (!pgd_present(*pgd))
goto kill;
pud = pud_offset(pgd, address);
if (!pud_present(*pud))
goto kill;
pmd = pmd_offset(pud, address);
if (!pmd_present(*pmd))
goto kill;
pte = pte_offset_kernel(pmd, address);
r = pte_read(*pte);
w = pte_write(*pte);
x = pte_exec(*pte);
if (!pte_young(*pte)) {
r = 0;
w = 0;
} else if (!pte_dirty(*pte)) {
w = 0;
}
mm_id = &mm->context.id;
prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
(x ? UM_PROT_EXEC : 0));
if (pte_newpage(*pte)) {
if (pte_present(*pte)) {
unsigned long long offset;
int fd;
fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
1, &flush);
}
else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
}
else if (pte_newprot(*pte))
err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
if (err)
goto kill;
*pte = pte_mkuptodate(*pte);
return;
kill:
printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
force_sig(SIGKILL, current);
}
pgd_t *pgd_offset_proc(struct mm_struct *mm, unsigned long address)
{
return pgd_offset(mm, address);
}
pud_t *pud_offset_proc(pgd_t *pgd, unsigned long address)
{
return pud_offset(pgd, address);
}
pmd_t *pmd_offset_proc(pud_t *pud, unsigned long address)
{
return pmd_offset(pud, address);
}
pte_t *pte_offset_proc(pmd_t *pmd, unsigned long address)
{
return pte_offset_kernel(pmd, address);
}
pte_t *addr_pte(struct task_struct *task, unsigned long addr)
{
pgd_t *pgd = pgd_offset(task->mm, addr);
pud_t *pud = pud_offset(pgd, addr);
pmd_t *pmd = pmd_offset(pud, addr);
return pte_offset_map(pmd, addr);
}
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
void flush_tlb_all(void)
{
flush_tlb_mm(current->mm);
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
}
void flush_tlb_kernel_range(unsigned long start, unsigned long end)
{
flush_tlb_kernel_range_common(start, end);
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
}
void flush_tlb_kernel_vm(void)
{
flush_tlb_kernel_range_common(start_vm, end_vm);
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
}
void __flush_tlb_one(unsigned long addr)
{
flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
}
static void fix_range(struct mm_struct *mm, unsigned long start_addr,
unsigned long end_addr, int force)
{
fix_range_common(mm, start_addr, end_addr, force);
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
}
void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
unsigned long end)
{
if (vma->vm_mm == NULL)
flush_tlb_kernel_range_common(start, end);
else fix_range(vma->vm_mm, start, end, 0);
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
}
void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
unsigned long end)
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
{
/*
* Don't bother flushing if this address space is about to be
* destroyed.
*/
if (atomic_read(&mm->mm_users) == 0)
return;
fix_range(mm, start, end, 0);
}
void flush_tlb_mm(struct mm_struct *mm)
{
struct vm_area_struct *vma = mm->mmap;
while (vma != NULL) {
fix_range(mm, vma->vm_start, vma->vm_end, 0);
vma = vma->vm_next;
}
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
}
void force_flush_all(void)
{
struct mm_struct *mm = current->mm;
struct vm_area_struct *vma = mm->mmap;
while (vma != NULL) {
fix_range(mm, vma->vm_start, vma->vm_end, 1);
vma = vma->vm_next;
}
[PATCH] uml: skas0 - separate kernel address space on stock hosts UML has had two modes of operation - an insecure, slow mode (tt mode) in which the kernel is mapped into every process address space which requires no host kernel modifications, and a secure, faster mode (skas mode) in which the UML kernel is in a separate host address space, which requires a patch to the host kernel. This patch implements something very close to skas mode for hosts which don't support skas - I'm calling this skas0. It provides the security of the skas host patch, and some of the performance gains. The two main things that are provided by the skas patch, /proc/mm and PTRACE_FAULTINFO, are implemented in a way that require no host patch. For the remote address space changing stuff (mmap, munmap, and mprotect), we set aside two pages in the process above its stack, one of which contains a little bit of code which can call mmap et al. To update the address space, the system call information (system call number and arguments) are written to the stub page above the code. The %esp is set to the beginning of the data, the %eip is set the the start of the stub, and it repeatedly pops the information into its registers and makes the system call until it sees a system call number of zero. This is to amortize the cost of the context switch across multiple address space updates. When the updates are done, it SIGSTOPs itself, and the kernel process continues what it was doing. For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the child, and let it handle segfaults rather than nullifying them. The handler is in the same page as the mmap stub. The second page is used as the stack. The handler reads cr2 and err from the sigcontext, sticks them at the base of the stack in a faultinfo struct, and SIGSTOPs itself. The kernel then reads the faultinfo and handles the fault. A complication on x86_64 is that this involves resetting the registers to the segfault values when the process is inside the kill system call. This breaks on x86_64 because %rcx will contain %rip because you tell SYSRET where to return to by putting the value in %rcx. So, this corrupts $rcx on return from the segfault. To work around this, I added an arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces the child back through the sigreturn. This causes %rcx to be restored by sigreturn and avoids the corruption. Ultimately, I think I will replace this with the trick of having it send itself a blocked signal which will be unblocked by the sigreturn. This will allow it to be stopped just after the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of PTRACE_SYSCALLing it through sigreturn. This runs on a stock host, so theoretically (and hopefully), tt mode isn't needed any more. We need to make sure that this is better in every way than tt mode, though. I'm concerned about the speed of address space updates and page fault handling, since they involve extra round-trips to the child. We can amortize the round-trip cost for large address space updates by writing all of the operations to the data page and having the child execute them all at the same time. This will help fork and exec, but not page faults, since they involve only one page. I can't think of any way to help page faults, except to add something like PTRACE_FAULTINFO to the host. There is PTRACE_SIGINFO, but UML doesn't use siginfo for SIGSEGV (or anything else) because there isn't enough information in the siginfo struct to handle page faults (the faulting operation type is missing). Adding that would make PTRACE_SIGINFO a usable equivalent to PTRACE_FAULTINFO. As for the code itself: - The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S. It is put in its own section of the binary along with stub_segv_handler in arch/um/kernel/skas/process.c. This is manipulated with run_syscall_stub in arch/um/kernel/skas/mem_user.c. syscall_stub will execute any system call at all, but it's only used for mmap, munmap, and mprotect. - The x86_64 stub calls sigreturn by hand rather than allowing the normal sigreturn to happen, because the normal sigreturn is a SA_RESTORER in UML's address space provided by libc. Needless to say, this is not available in the child's address space. Also, it does a couple of odd pops before that which restore the stack to the state it was in at the time the signal handler was called. - There is a new field in the arch mmu_context, which is now a union. This is the pid to be manipulated rather than the /proc/mm file descriptor. Code which deals with this now checks proc_mm to see whether it should use the usual skas code or the new code. - userspace_tramp is now used to create a new host process for every UML process, rather than one per UML processor. It checks proc_mm and ptrace_faultinfo to decide whether to map in the pages above its stack. - start_userspace now makes CLONE_VM conditional on proc_mm since we need separate address spaces now. - switch_mm_skas now just sets userspace_pid[0] to the new pid rather than PTRACE_SWITCH_MM. There is an addition to userspace which updates its idea of the pid being manipulated each time around the loop. This is important on exec, when the pid will change underneath userspace(). - The stub page has a pte, but it can't be mapped in using tlb_flush because it is part of tlb_flush. This is why it's required for it to be mapped in by userspace_tramp. Other random things: - The stub section in uml.lds.S is page aligned. This page is written out to the backing vm file in setup_physmem because it is mapped from there into user processes. - There's some confusion with TASK_SIZE now that there are a couple of extra pages that the process can't use. TASK_SIZE is considered by the elf code to be the usable process memory, which is reasonable, so it is decreased by two pages. This confuses the definition of USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down of the uneven division. So we round it to the nearest PGDIR_SIZE rather than the lower one. - I added a missing PT_SYSCALL_ARG6_OFFSET macro. - um_mmu.h was made into a userspace-usable file. - proc_mm and ptrace_faultinfo are globals which say whether the host supports these features. - There is a bad interaction between the mm.nr_ptes check at the end of exit_mmap, stack randomization, and skas0. exit_mmap will stop freeing pages at the PGDIR_SIZE boundary after the last vma. If the stack isn't on the last page table page, the last pte page won't be freed, as it should be since the stub ptes are there, and exit_mmap will BUG because there is an unfreed page. To get around this, TASK_SIZE is set to the next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the calls to init_stub_pte. This ensures that we know the process stack (and all other process mappings) will be below the top page table page, and thus we know that mm->nr_ptes will be one too many, and can be decremented. Things that need fixing: - We may need better assurrences that the stub code is PIC. - The stub pte is set up in init_new_context_skas. - alloc_pgdir is probably the right place. Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-08 00:56:49 +00:00
}