vserver 2.0 rc7
[linux-2.6.git] / arch / x86_64 / ia32 / syscall32.c
index 399ff49..01d8db1 100644 (file)
@@ -9,6 +9,7 @@
 #include <linux/gfp.h>
 #include <linux/init.h>
 #include <linux/stringify.h>
+#include <linux/security.h>
 #include <asm/proto.h>
 #include <asm/tlbflush.h>
 #include <asm/ia32_unistd.h>
@@ -30,51 +31,57 @@ extern int sysctl_vsyscall32;
 char *syscall32_page; 
 static int use_sysenter = -1;
 
-/*
- * Map the 32bit vsyscall page on demand.
- *
- * RED-PEN: This knows too much about high level VM.
- *
- * Alternative would be to generate a vma with appropriate backing options
- * and let it be handled by generic VM.
- */
-int __map_syscall32(struct mm_struct *mm, unsigned long address)
-{ 
-       pgd_t *pgd;
-       pud_t *pud;
-       pte_t *pte;
-       pmd_t *pmd;
-       int err = -ENOMEM;
-
-       spin_lock(&mm->page_table_lock); 
-       pgd = pgd_offset(mm, address);
-       pud = pud_alloc(mm, pgd, address);
-       if (pud) {
-               pmd = pmd_alloc(mm, pud, address);
-               if (pmd && (pte = pte_alloc_map(mm, pmd, address)) != NULL) {
-                       if (pte_none(*pte)) {
-                               set_pte(pte,
-                                       mk_pte(virt_to_page(syscall32_page),
-                                              PAGE_KERNEL_VSYSCALL32));
-                       }
-                       /* Flush only the local CPU. Other CPUs taking a fault
-                          will just end up here again
-                          This probably not needed and just paranoia. */
-                       __flush_tlb_one(address);
-                       err = 0;
-               }
-       }
-       spin_unlock(&mm->page_table_lock);
-       return err;
+static struct page *
+syscall32_nopage(struct vm_area_struct *vma, unsigned long adr, int *type)
+{
+       struct page *p = virt_to_page(adr - vma->vm_start + syscall32_page);
+       get_page(p);
+       return p;
 }
 
-int map_syscall32(struct mm_struct *mm, unsigned long address)
+/* Prevent VMA merging */
+static void syscall32_vma_close(struct vm_area_struct *vma)
 {
-       int err;
-       down_read(&mm->mmap_sem);
-       err = __map_syscall32(mm, address);
-       up_read(&mm->mmap_sem);
-       return err;
+}
+
+static struct vm_operations_struct syscall32_vm_ops = {
+       .close = syscall32_vma_close,
+       .nopage = syscall32_nopage,
+};
+
+struct linux_binprm;
+
+/* Setup a VMA at program startup for the vsyscall page */
+int syscall32_setup_pages(struct linux_binprm *bprm, int exstack)
+{
+       int npages = (VSYSCALL32_END - VSYSCALL32_BASE) >> PAGE_SHIFT;
+       struct vm_area_struct *vma;
+       struct mm_struct *mm = current->mm;
+
+       vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
+       if (!vma)
+               return -ENOMEM;
+       if (security_vm_enough_memory(npages)) {
+               kmem_cache_free(vm_area_cachep, vma);
+               return -ENOMEM;
+       }
+
+       memset(vma, 0, sizeof(struct vm_area_struct));
+       /* Could randomize here */
+       vma->vm_start = VSYSCALL32_BASE;
+       vma->vm_end = VSYSCALL32_END;
+       /* MAYWRITE to allow gdb to COW and set breakpoints */
+       vma->vm_flags = VM_READ|VM_EXEC|VM_MAYREAD|VM_MAYEXEC|VM_MAYEXEC|VM_MAYWRITE;
+       vma->vm_flags |= mm->def_flags;
+       vma->vm_page_prot = protection_map[vma->vm_flags & 7];
+       vma->vm_ops = &syscall32_vm_ops;
+       vma->vm_mm = mm;
+
+       down_write(&mm->mmap_sem);
+       insert_vm_struct(mm, vma);
+       mm->total_vm += npages;
+       up_write(&mm->mmap_sem);
+       return 0;
 }
 
 static int __init init_syscall32(void)
@@ -82,7 +89,6 @@ static int __init init_syscall32(void)
        syscall32_page = (void *)get_zeroed_page(GFP_KERNEL); 
        if (!syscall32_page) 
                panic("Cannot allocate syscall32 page"); 
-       SetPageReserved(virt_to_page(syscall32_page));
        if (use_sysenter > 0) {
                memcpy(syscall32_page, syscall32_sysenter,
                       syscall32_sysenter_end - syscall32_sysenter);