Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / mm / mprotect.c
index 5b438e1..05dc342 100644 (file)
 #include <linux/highmem.h>
 #include <linux/security.h>
 #include <linux/mempolicy.h>
+#include <linux/personality.h>
+#include <linux/syscalls.h>
 
 #include <asm/uaccess.h>
-#include <asm/pgalloc.h>
 #include <asm/pgtable.h>
+#include <asm/pgalloc.h>
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
 
-static inline void
-change_pte_range(pmd_t *pmd, unsigned long address,
-               unsigned long size, pgprot_t newprot)
+static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
+               unsigned long addr, unsigned long end, pgprot_t newprot)
 {
-       pte_t * pte;
-       unsigned long end;
-
-       if (pmd_none(*pmd))
-               return;
-       if (pmd_bad(*pmd)) {
-               pmd_ERROR(*pmd);
-               pmd_clear(pmd);
-               return;
-       }
-       pte = pte_offset_map(pmd, address);
-       address &= ~PMD_MASK;
-       end = address + size;
-       if (end > PMD_SIZE)
-               end = PMD_SIZE;
+       pte_t *pte;
+       spinlock_t *ptl;
+
+       pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
        do {
                if (pte_present(*pte)) {
-                       pte_t entry;
+                       pte_t ptent;
 
                        /* Avoid an SMP race with hardware updated dirty/clean
                         * bits by wiping the pte and then setting the new pte
                         * into place.
                         */
-                       entry = ptep_get_and_clear(pte);
-                       set_pte(pte, pte_modify(entry, newprot));
+                       ptent = pte_modify(ptep_get_and_clear(mm, addr, pte), newprot);
+                       set_pte_at(mm, addr, pte, ptent);
+                       lazy_mmu_prot_update(ptent);
                }
-               address += PAGE_SIZE;
-               pte++;
-       } while (address && (address < end));
-       pte_unmap(pte - 1);
+       } while (pte++, addr += PAGE_SIZE, addr != end);
+       pte_unmap_unlock(pte - 1, ptl);
 }
 
-static inline void
-change_pmd_range(pgd_t *pgd, unsigned long address,
-               unsigned long size, pgprot_t newprot)
+static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
+               unsigned long addr, unsigned long end, pgprot_t newprot)
 {
-       pmd_t * pmd;
-       unsigned long end;
-
-       if (pgd_none(*pgd))
-               return;
-       if (pgd_bad(*pgd)) {
-               pgd_ERROR(*pgd);
-               pgd_clear(pgd);
-               return;
-       }
-       pmd = pmd_offset(pgd, address);
-       address &= ~PGDIR_MASK;
-       end = address + size;
-       if (end > PGDIR_SIZE)
-               end = PGDIR_SIZE;
+       pmd_t *pmd;
+       unsigned long next;
+
+       pmd = pmd_offset(pud, addr);
        do {
-               change_pte_range(pmd, address, end - address, newprot);
-               address = (address + PMD_SIZE) & PMD_MASK;
-               pmd++;
-       } while (address && (address < end));
+               next = pmd_addr_end(addr, end);
+               if (pmd_none_or_clear_bad(pmd))
+                       continue;
+               change_pte_range(mm, pmd, addr, next, newprot);
+       } while (pmd++, addr = next, addr != end);
 }
 
-static void
-change_protection(struct vm_area_struct *vma, unsigned long start,
-               unsigned long end, pgprot_t newprot)
+static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
+               unsigned long addr, unsigned long end, pgprot_t newprot)
 {
-       pgd_t *dir;
-       unsigned long beg = start;
-
-       dir = pgd_offset(current->mm, start);
-       flush_cache_range(vma, beg, end);
-       if (start >= end)
-               BUG();
-       spin_lock(&current->mm->page_table_lock);
+       pud_t *pud;
+       unsigned long next;
+
+       pud = pud_offset(pgd, addr);
        do {
-               change_pmd_range(dir, start, end - start, newprot);
-               start = (start + PGDIR_SIZE) & PGDIR_MASK;
-               dir++;
-       } while (start && (start < end));
-       flush_tlb_range(vma, beg, end);
-       spin_unlock(&current->mm->page_table_lock);
-       return;
+               next = pud_addr_end(addr, end);
+               if (pud_none_or_clear_bad(pud))
+                       continue;
+               change_pmd_range(mm, pud, addr, next, newprot);
+       } while (pud++, addr = next, addr != end);
+}
+
+static void change_protection(struct vm_area_struct *vma,
+               unsigned long addr, unsigned long end, pgprot_t newprot)
+{
+       struct mm_struct *mm = vma->vm_mm;
+       pgd_t *pgd;
+       unsigned long next;
+       unsigned long start = addr;
+
+       BUG_ON(addr >= end);
+       pgd = pgd_offset(mm, addr);
+       flush_cache_range(vma, addr, end);
+       do {
+               next = pgd_addr_end(addr, end);
+               if (pgd_none_or_clear_bad(pgd))
+                       continue;
+               change_pud_range(mm, pgd, addr, next, newprot);
+       } while (pgd++, addr = next, addr != end);
+       flush_tlb_range(vma, start, end);
 }
 
 static int
 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
-       unsigned long start, unsigned long end, unsigned int newflags)
+       unsigned long start, unsigned long end, unsigned long newflags)
 {
-       struct mm_struct * mm = vma->vm_mm;
-       unsigned long charged = 0;
+       struct mm_struct *mm = vma->vm_mm;
+       unsigned long oldflags = vma->vm_flags;
+       long nrpages = (end - start) >> PAGE_SHIFT;
+       unsigned long charged = 0, old_end = vma->vm_end;
        pgprot_t newprot;
        pgoff_t pgoff;
        int error;
 
-       if (newflags == vma->vm_flags) {
+       if (newflags == oldflags) {
                *pprev = vma;
                return 0;
        }
@@ -132,8 +125,8 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
         * a MAP_NORESERVE private mapping to writable will now reserve.
         */
        if (newflags & VM_WRITE) {
-               if (!(vma->vm_flags & (VM_ACCOUNT|VM_WRITE|VM_SHARED|VM_HUGETLB))) {
-                       charged = (end - start) >> PAGE_SHIFT;
+               if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED))) {
+                       charged = nrpages;
                        if (security_vm_enough_memory(charged))
                                return -ENOMEM;
                        newflags |= VM_ACCOUNT;
@@ -153,16 +146,13 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
                goto success;
        }
 
+       *pprev = vma;
+
        if (start != vma->vm_start) {
                error = split_vma(mm, vma, start, 1);
                if (error)
                        goto fail;
        }
-       /*
-        * Unless it returns an error, this function always sets *pprev to
-        * the first vma for which vma->vm_end >= end.
-        */
-       *pprev = vma;
 
        if (end != vma->vm_end) {
                error = split_vma(mm, vma, end, 0);
@@ -177,7 +167,14 @@ success:
         */
        vma->vm_flags = newflags;
        vma->vm_page_prot = newprot;
-       change_protection(vma, start, end, newprot);
+       if (oldflags & VM_EXEC)
+               arch_remove_exec_range(current->mm, old_end);
+       if (is_vm_hugetlb_page(vma))
+               hugetlb_change_protection(vma, start, end, newprot);
+       else
+               change_protection(vma, start, end, newprot);
+       vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
+       vm_stat_account(mm, newflags, vma->vm_file, nrpages);
        return 0;
 
 fail:
@@ -188,7 +185,7 @@ fail:
 asmlinkage long
 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
 {
-       unsigned long vm_flags, nstart, end, tmp;
+       unsigned long vm_flags, nstart, end, tmp, reqprot;
        struct vm_area_struct *vma, *prev;
        int error = -EINVAL;
        const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
@@ -198,14 +195,22 @@ sys_mprotect(unsigned long start, size_t len, unsigned long prot)
 
        if (start & ~PAGE_MASK)
                return -EINVAL;
+       if (!len)
+               return 0;
        len = PAGE_ALIGN(len);
        end = start + len;
-       if (end < start)
+       if (end <= start)
                return -ENOMEM;
        if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
                return -EINVAL;
-       if (end == start)
-               return 0;
+
+       reqprot = prot;
+       /*
+        * Does the application expect PROT_READ to imply PROT_EXEC:
+        */
+       if (unlikely((prot & PROT_READ) &&
+                       (current->personality & READ_IMPLIES_EXEC)))
+               prot |= PROT_EXEC;
 
        vm_flags = calc_vm_prot_bits(prot);
 
@@ -237,23 +242,19 @@ sys_mprotect(unsigned long start, size_t len, unsigned long prot)
                prev = vma;
 
        for (nstart = start ; ; ) {
-               unsigned int newflags;
+               unsigned long newflags;
 
                /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
 
-               if (is_vm_hugetlb_page(vma)) {
-                       error = -EACCES;
-                       goto out;
-               }
-
                newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 
-               if ((newflags & ~(newflags >> 4)) & 0xf) {
+               /* newflags >> 4 shift VM_MAY% in place of VM_% */
+               if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
                        error = -EACCES;
                        goto out;
                }
 
-               error = security_file_mprotect(vma, prot);
+               error = security_file_mprotect(vma, reqprot, prot);
                if (error)
                        goto out;