CGroup API files: make CGROUP_DEBUG default to off
[safe/jmp/linux-2.6] / mm / mmap.c
index 9e685b9..677d184 100644 (file)
--- a/mm/mmap.c
+++ b/mm/mmap.c
 #define arch_mmap_check(addr, len, flags)      (0)
 #endif
 
+#ifndef arch_rebalance_pgtables
+#define arch_rebalance_pgtables(addr, len)             (addr)
+#endif
+
 static void unmap_region(struct mm_struct *mm,
                struct vm_area_struct *vma, struct vm_area_struct *prev,
                unsigned long start, unsigned long end);
@@ -181,8 +185,6 @@ error:
        return -ENOMEM;
 }
 
-EXPORT_SYMBOL(__vm_enough_memory);
-
 /*
  * Requires inode->i_mapping->i_mmap_lock
  */
@@ -230,7 +232,7 @@ static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
                vma->vm_ops->close(vma);
        if (vma->vm_file)
                fput(vma->vm_file);
-       mpol_free(vma_policy(vma));
+       mpol_put(vma_policy(vma));
        kmem_cache_free(vm_area_cachep, vma);
        return next;
 }
@@ -243,7 +245,7 @@ asmlinkage unsigned long sys_brk(unsigned long brk)
 
        down_write(&mm->mmap_sem);
 
-       if (brk < mm->end_code)
+       if (brk < mm->start_brk)
                goto out;
 
        /*
@@ -253,7 +255,8 @@ asmlinkage unsigned long sys_brk(unsigned long brk)
         * not page aligned -Ram Gupta
         */
        rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
-       if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
+       if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
+                       (mm->end_data - mm->start_data) > rlim)
                goto out;
 
        newbrk = PAGE_ALIGN(brk);
@@ -623,7 +626,7 @@ again:                      remove_next = 1 + (end > next->vm_end);
                if (file)
                        fput(file);
                mm->map_count--;
-               mpol_free(vma_policy(next));
+               mpol_put(vma_policy(next));
                kmem_cache_free(vm_area_cachep, next);
                /*
                 * In mprotect's case 6 (see comments on vma_merge),
@@ -914,6 +917,9 @@ unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
        if (!len)
                return -EINVAL;
 
+       if (!(flags & MAP_FIXED))
+               addr = round_hint_to_min(addr);
+
        error = arch_mmap_check(addr, len, flags);
        if (error)
                return error;
@@ -1050,8 +1056,7 @@ int vma_wants_writenotify(struct vm_area_struct *vma)
 
        /* The open routine did something to the protections already? */
        if (pgprot_val(vma->vm_page_prot) !=
-           pgprot_val(protection_map[vm_flags &
-                   (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]))
+           pgprot_val(vm_get_page_prot(vm_flags)))
                return 0;
 
        /* Specialty mapping? */
@@ -1063,7 +1068,6 @@ int vma_wants_writenotify(struct vm_area_struct *vma)
                mapping_cap_account_dirty(vma->vm_file->f_mapping);
 }
 
-
 unsigned long mmap_region(struct file *file, unsigned long addr,
                          unsigned long len, unsigned long flags,
                          unsigned int vm_flags, unsigned long pgoff,
@@ -1132,8 +1136,7 @@ munmap_back:
        vma->vm_start = addr;
        vma->vm_end = addr + len;
        vma->vm_flags = vm_flags;
-       vma->vm_page_prot = protection_map[vm_flags &
-                               (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
+       vma->vm_page_prot = vm_get_page_prot(vm_flags);
        vma->vm_pgoff = pgoff;
 
        if (file) {
@@ -1175,25 +1178,22 @@ munmap_back:
        vm_flags = vma->vm_flags;
 
        if (vma_wants_writenotify(vma))
-               vma->vm_page_prot =
-                       protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC)];
+               vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
 
-       if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
+       if (file && vma_merge(mm, prev, addr, vma->vm_end,
                        vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
-               file = vma->vm_file;
-               vma_link(mm, vma, prev, rb_link, rb_parent);
-               if (correct_wcount)
-                       atomic_inc(&inode->i_writecount);
-       } else {
-               if (file) {
-                       if (correct_wcount)
-                               atomic_inc(&inode->i_writecount);
-                       fput(file);
-               }
-               mpol_free(vma_policy(vma));
+               mpol_put(vma_policy(vma));
                kmem_cache_free(vm_area_cachep, vma);
+               fput(file);
+       } else {
+               vma_link(mm, vma, prev, rb_link, rb_parent);
+               file = vma->vm_file;
        }
-out:   
+
+       /* Once vma denies write, undo our temporary denial count */
+       if (correct_wcount)
+               atomic_inc(&inode->i_writecount);
+out:
        mm->total_vm += len >> PAGE_SHIFT;
        vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
        if (vm_flags & VM_LOCKED) {
@@ -1425,7 +1425,7 @@ get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
        if (addr & ~PAGE_MASK)
                return -EINVAL;
 
-       return addr;
+       return arch_rebalance_pgtables(addr, len);
 }
 
 EXPORT_SYMBOL(get_unmapped_area);
@@ -1620,6 +1620,12 @@ static inline int expand_downwards(struct vm_area_struct *vma,
         */
        if (unlikely(anon_vma_prepare(vma)))
                return -ENOMEM;
+
+       address &= PAGE_MASK;
+       error = security_file_mmap(NULL, 0, 0, 0, address, 1);
+       if (error)
+               return error;
+
        anon_vma_lock(vma);
 
        /*
@@ -1627,8 +1633,6 @@ static inline int expand_downwards(struct vm_area_struct *vma,
         * is required to hold the mmap_sem in read mode.  We need the
         * anon_vma lock to serialize against concurrent expand_stacks.
         */
-       address &= PAGE_MASK;
-       error = 0;
 
        /* Somebody else might have raced and expanded it already */
        if (address < vma->vm_start) {
@@ -1806,7 +1810,7 @@ int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
                new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
        }
 
-       pol = mpol_copy(vma_policy(vma));
+       pol = mpol_dup(vma_policy(vma));
        if (IS_ERR(pol)) {
                kmem_cache_free(vm_area_cachep, new);
                return PTR_ERR(pol);
@@ -1939,6 +1943,10 @@ unsigned long do_brk(unsigned long addr, unsigned long len)
        if (is_hugepage_only_range(mm, addr, len))
                return -EINVAL;
 
+       error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
+       if (error)
+               return error;
+
        flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
 
        error = arch_mmap_check(addr, len, flags);
@@ -2004,8 +2012,7 @@ unsigned long do_brk(unsigned long addr, unsigned long len)
        vma->vm_end = addr + len;
        vma->vm_pgoff = pgoff;
        vma->vm_flags = flags;
-       vma->vm_page_prot = protection_map[flags &
-                               (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
+       vma->vm_page_prot = vm_get_page_prot(flags);
        vma_link(mm, vma, prev, rb_link, rb_parent);
 out:
        mm->total_vm += len >> PAGE_SHIFT;
@@ -2119,7 +2126,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
                new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
                if (new_vma) {
                        *new_vma = *vma;
-                       pol = mpol_copy(vma_policy(vma));
+                       pol = mpol_dup(vma_policy(vma));
                        if (IS_ERR(pol)) {
                                kmem_cache_free(vm_area_cachep, new_vma);
                                return NULL;
@@ -2155,24 +2162,31 @@ int may_expand_vm(struct mm_struct *mm, unsigned long npages)
 }
 
 
-static struct page *special_mapping_nopage(struct vm_area_struct *vma,
-                                          unsigned long address, int *type)
+static int special_mapping_fault(struct vm_area_struct *vma,
+                               struct vm_fault *vmf)
 {
+       pgoff_t pgoff;
        struct page **pages;
 
-       BUG_ON(address < vma->vm_start || address >= vma->vm_end);
+       /*
+        * special mappings have no vm_file, and in that case, the mm
+        * uses vm_pgoff internally. So we have to subtract it from here.
+        * We are allowed to do this because we are the mm; do not copy
+        * this code into drivers!
+        */
+       pgoff = vmf->pgoff - vma->vm_pgoff;
 
-       address -= vma->vm_start;
-       for (pages = vma->vm_private_data; address > 0 && *pages; ++pages)
-               address -= PAGE_SIZE;
+       for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
+               pgoff--;
 
        if (*pages) {
                struct page *page = *pages;
                get_page(page);
-               return page;
+               vmf->page = page;
+               return 0;
        }
 
-       return NOPAGE_SIGBUS;
+       return VM_FAULT_SIGBUS;
 }
 
 /*
@@ -2184,7 +2198,7 @@ static void special_mapping_close(struct vm_area_struct *vma)
 
 static struct vm_operations_struct special_mapping_vmops = {
        .close = special_mapping_close,
-       .nopage = special_mapping_nopage,
+       .fault = special_mapping_fault,
 };
 
 /*
@@ -2210,8 +2224,8 @@ int install_special_mapping(struct mm_struct *mm,
        vma->vm_start = addr;
        vma->vm_end = addr + len;
 
-       vma->vm_flags = vm_flags | mm->def_flags;
-       vma->vm_page_prot = protection_map[vma->vm_flags & 7];
+       vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
+       vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 
        vma->vm_ops = &special_mapping_vmops;
        vma->vm_private_data = pages;