MIPS: JZ4740: Fix GCC 4.6.0 build error.
[linux-2.6.git] / mm / nommu.c
index a6e8ccf..c4c542c 100644 (file)
@@ -10,7 +10,7 @@
  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
- *  Copyright (c) 2007-2008 Paul Mundt <lethal@linux-sh.org>
+ *  Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
  */
 
 #include <linux/module.h>
 #include <linux/personality.h>
 #include <linux/security.h>
 #include <linux/syscalls.h>
+#include <linux/audit.h>
 
 #include <asm/uaccess.h>
 #include <asm/tlb.h>
 #include <asm/tlbflush.h>
+#include <asm/mmu_context.h>
 #include "internal.h"
 
-static inline __attribute__((format(printf, 1, 2)))
-void no_printk(const char *fmt, ...)
-{
-}
-
 #if 0
 #define kenter(FMT, ...) \
        printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
@@ -56,20 +53,19 @@ void no_printk(const char *fmt, ...)
        no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
 #endif
 
-#include "internal.h"
-
 void *high_memory;
 struct page *mem_map;
 unsigned long max_mapnr;
 unsigned long num_physpages;
-atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
+unsigned long highest_memmap_pfn;
+struct percpu_counter vm_committed_as;
 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
 int sysctl_overcommit_ratio = 50; /* default is 50% */
 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
-int sysctl_nr_trim_pages = 1; /* page trimming behaviour */
+int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
 int heap_stack_gap = 0;
 
-atomic_t mmap_pages_allocated;
+atomic_long_t mmap_pages_allocated;
 
 EXPORT_SYMBOL(mem_map);
 EXPORT_SYMBOL(num_physpages);
@@ -79,50 +75,10 @@ static struct kmem_cache *vm_region_jar;
 struct rb_root nommu_region_tree = RB_ROOT;
 DECLARE_RWSEM(nommu_region_sem);
 
-struct vm_operations_struct generic_file_vm_ops = {
+const struct vm_operations_struct generic_file_vm_ops = {
 };
 
 /*
- * Handle all mappings that got truncated by a "truncate()"
- * system call.
- *
- * NOTE! We have to be ready to update the memory sharing
- * between the file and the memory map for a potential last
- * incomplete page.  Ugly, but necessary.
- */
-int vmtruncate(struct inode *inode, loff_t offset)
-{
-       struct address_space *mapping = inode->i_mapping;
-       unsigned long limit;
-
-       if (inode->i_size < offset)
-               goto do_expand;
-       i_size_write(inode, offset);
-
-       truncate_inode_pages(mapping, offset);
-       goto out_truncate;
-
-do_expand:
-       limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
-       if (limit != RLIM_INFINITY && offset > limit)
-               goto out_sig;
-       if (offset > inode->i_sb->s_maxbytes)
-               goto out;
-       i_size_write(inode, offset);
-
-out_truncate:
-       if (inode->i_op->truncate)
-               inode->i_op->truncate(inode);
-       return 0;
-out_sig:
-       send_sig(SIGXFSZ, current, 0);
-out:
-       return -EFBIG;
-}
-
-EXPORT_SYMBOL(vmtruncate);
-
-/*
  * Return the total memory allocated for this pointer, not
  * just what the caller asked for.
  *
@@ -149,6 +105,20 @@ unsigned int kobjsize(const void *objp)
                return ksize(objp);
 
        /*
+        * If it's not a compound page, see if we have a matching VMA
+        * region. This test is intentionally done in reverse order,
+        * so if there's no VMA, we still fall through and hand back
+        * PAGE_SIZE for 0-order pages.
+        */
+       if (!PageCompound(page)) {
+               struct vm_area_struct *vma;
+
+               vma = find_vma(current->mm, (unsigned long)objp);
+               if (vma)
+                       return vma->vm_end - vma->vm_start;
+       }
+
+       /*
         * The ksize() function is only guaranteed to work for pointers
         * returned by kmalloc(). So handle arbitrary pointers here.
         */
@@ -156,30 +126,30 @@ unsigned int kobjsize(const void *objp)
 }
 
 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
-                    unsigned long start, int len, int flags,
-               struct page **pages, struct vm_area_struct **vmas)
+                    unsigned long start, int nr_pages, unsigned int foll_flags,
+                    struct page **pages, struct vm_area_struct **vmas,
+                    int *retry)
 {
        struct vm_area_struct *vma;
        unsigned long vm_flags;
        int i;
-       int write = !!(flags & GUP_FLAGS_WRITE);
-       int force = !!(flags & GUP_FLAGS_FORCE);
-       int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
 
        /* calculate required read or write permissions.
-        * - if 'force' is set, we only require the "MAY" flags.
+        * If FOLL_FORCE is set, we only require the "MAY" flags.
         */
-       vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
-       vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
+       vm_flags  = (foll_flags & FOLL_WRITE) ?
+                       (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
+       vm_flags &= (foll_flags & FOLL_FORCE) ?
+                       (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 
-       for (i = 0; i < len; i++) {
+       for (i = 0; i < nr_pages; i++) {
                vma = find_vma(mm, start);
                if (!vma)
                        goto finish_or_fault;
 
                /* protect what we can, including chardevs */
-               if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
-                   (!ignore && !(vm_flags & vma->vm_flags)))
+               if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
+                   !(vm_flags & vma->vm_flags))
                        goto finish_or_fault;
 
                if (pages) {
@@ -189,7 +159,7 @@ int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
                }
                if (vmas)
                        vmas[i] = vma;
-               start += PAGE_SIZE;
+               start = (start + PAGE_SIZE) & PAGE_MASK;
        }
 
        return i;
@@ -198,7 +168,6 @@ finish_or_fault:
        return i ? : -EFAULT;
 }
 
-
 /*
  * get a list of pages in an address range belonging to the specified process
  * and indicate the VMA that covers each page
@@ -207,22 +176,42 @@ finish_or_fault:
  * - don't permit access to VMAs that don't support it, such as I/O mappings
  */
 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
-       unsigned long start, int len, int write, int force,
+       unsigned long start, int nr_pages, int write, int force,
        struct page **pages, struct vm_area_struct **vmas)
 {
        int flags = 0;
 
        if (write)
-               flags |= GUP_FLAGS_WRITE;
+               flags |= FOLL_WRITE;
        if (force)
-               flags |= GUP_FLAGS_FORCE;
+               flags |= FOLL_FORCE;
 
-       return __get_user_pages(tsk, mm,
-                               start, len, flags,
-                               pages, vmas);
+       return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
+                               NULL);
 }
 EXPORT_SYMBOL(get_user_pages);
 
+/**
+ * follow_pfn - look up PFN at a user virtual address
+ * @vma: memory mapping
+ * @address: user virtual address
+ * @pfn: location to store found PFN
+ *
+ * Only IO mappings and raw PFN mappings are allowed.
+ *
+ * Returns zero and the pfn at @pfn on success, -ve otherwise.
+ */
+int follow_pfn(struct vm_area_struct *vma, unsigned long address,
+       unsigned long *pfn)
+{
+       if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
+               return -EINVAL;
+
+       *pfn = address >> PAGE_SHIFT;
+       return 0;
+}
+EXPORT_SYMBOL(follow_pfn);
+
 DEFINE_RWLOCK(vmlist_lock);
 struct vm_struct *vmlist;
 
@@ -307,12 +296,60 @@ void *vmalloc(unsigned long size)
 }
 EXPORT_SYMBOL(vmalloc);
 
+/*
+ *     vzalloc - allocate virtually continguos memory with zero fill
+ *
+ *     @size:          allocation size
+ *
+ *     Allocate enough pages to cover @size from the page level
+ *     allocator and map them into continguos kernel virtual space.
+ *     The memory allocated is set to zero.
+ *
+ *     For tight control over page level allocator and protection flags
+ *     use __vmalloc() instead.
+ */
+void *vzalloc(unsigned long size)
+{
+       return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
+                       PAGE_KERNEL);
+}
+EXPORT_SYMBOL(vzalloc);
+
+/**
+ * vmalloc_node - allocate memory on a specific node
+ * @size:      allocation size
+ * @node:      numa node
+ *
+ * Allocate enough pages to cover @size from the page level
+ * allocator and map them into contiguous kernel virtual space.
+ *
+ * For tight control over page level allocator and protection flags
+ * use __vmalloc() instead.
+ */
 void *vmalloc_node(unsigned long size, int node)
 {
        return vmalloc(size);
 }
 EXPORT_SYMBOL(vmalloc_node);
 
+/**
+ * vzalloc_node - allocate memory on a specific node with zero fill
+ * @size:      allocation size
+ * @node:      numa node
+ *
+ * Allocate enough pages to cover @size from the page level
+ * allocator and map them into contiguous kernel virtual space.
+ * The memory allocated is set to zero.
+ *
+ * For tight control over page level allocator and protection flags
+ * use __vmalloc() instead.
+ */
+void *vzalloc_node(unsigned long size, int node)
+{
+       return vzalloc(size);
+}
+EXPORT_SYMBOL(vzalloc_node);
+
 #ifndef PAGE_KERNEL_EXEC
 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 #endif
@@ -380,6 +417,24 @@ void vunmap(const void *addr)
 }
 EXPORT_SYMBOL(vunmap);
 
+void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
+{
+       BUG();
+       return NULL;
+}
+EXPORT_SYMBOL(vm_map_ram);
+
+void vm_unmap_ram(const void *mem, unsigned int count)
+{
+       BUG();
+}
+EXPORT_SYMBOL(vm_unmap_ram);
+
+void vm_unmap_aliases(void)
+{
+}
+EXPORT_SYMBOL_GPL(vm_unmap_aliases);
+
 /*
  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
  * have one.
@@ -388,6 +443,31 @@ void  __attribute__((weak)) vmalloc_sync_all(void)
 {
 }
 
+/**
+ *     alloc_vm_area - allocate a range of kernel address space
+ *     @size:          size of the area
+ *
+ *     Returns:        NULL on failure, vm_struct on success
+ *
+ *     This function reserves a range of kernel address space, and
+ *     allocates pagetables to map that range.  No actual mappings
+ *     are created.  If the kernel address space is not shared
+ *     between processes, it syncs the pagetable across all
+ *     processes.
+ */
+struct vm_struct *alloc_vm_area(size_t size)
+{
+       BUG();
+       return NULL;
+}
+EXPORT_SYMBOL_GPL(alloc_vm_area);
+
+void free_vm_area(struct vm_struct *area)
+{
+       BUG();
+}
+EXPORT_SYMBOL_GPL(free_vm_area);
+
 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
                   struct page *page)
 {
@@ -402,7 +482,7 @@ EXPORT_SYMBOL(vm_insert_page);
  *  to a regular file.  in this case, the unmapping will need
  *  to invoke file system routines that need the global lock.
  */
-asmlinkage unsigned long sys_brk(unsigned long brk)
+SYSCALL_DEFINE1(brk, unsigned long, brk)
 {
        struct mm_struct *mm = current->mm;
 
@@ -423,6 +503,7 @@ asmlinkage unsigned long sys_brk(unsigned long brk)
        /*
         * Ok, looks good - let it rip.
         */
+       flush_icache_range(mm->brk, brk);
        return mm->brk = brk;
 }
 
@@ -431,12 +512,11 @@ asmlinkage unsigned long sys_brk(unsigned long brk)
  */
 void __init mmap_init(void)
 {
-       vm_region_jar = kmem_cache_create("vm_region_jar",
-                                         sizeof(struct vm_region), 0,
-                                         SLAB_PANIC, NULL);
-       vm_area_cachep = kmem_cache_create("vm_area_struct",
-                                          sizeof(struct vm_area_struct), 0,
-                                          SLAB_PANIC, NULL);
+       int ret;
+
+       ret = percpu_counter_init(&vm_committed_as, 0);
+       VM_BUG_ON(ret);
+       vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 }
 
 /*
@@ -454,27 +534,24 @@ static noinline void validate_nommu_regions(void)
                return;
 
        last = rb_entry(lastp, struct vm_region, vm_rb);
-       if (unlikely(last->vm_end <= last->vm_start))
-               BUG();
-       if (unlikely(last->vm_top < last->vm_end))
-               BUG();
+       BUG_ON(unlikely(last->vm_end <= last->vm_start));
+       BUG_ON(unlikely(last->vm_top < last->vm_end));
 
        while ((p = rb_next(lastp))) {
                region = rb_entry(p, struct vm_region, vm_rb);
                last = rb_entry(lastp, struct vm_region, vm_rb);
 
-               if (unlikely(region->vm_end <= region->vm_start))
-                       BUG();
-               if (unlikely(region->vm_top < region->vm_end))
-                       BUG();
-               if (unlikely(region->vm_start < last->vm_top))
-                       BUG();
+               BUG_ON(unlikely(region->vm_end <= region->vm_start));
+               BUG_ON(unlikely(region->vm_top < region->vm_end));
+               BUG_ON(unlikely(region->vm_start < last->vm_top));
 
                lastp = p;
        }
 }
 #else
-#define validate_nommu_regions() do {} while(0)
+static void validate_nommu_regions(void)
+{
+}
 #endif
 
 /*
@@ -487,8 +564,6 @@ static void add_nommu_region(struct vm_region *region)
 
        validate_nommu_regions();
 
-       BUG_ON(region->vm_start & ~PAGE_MASK);
-
        parent = NULL;
        p = &nommu_region_tree.rb_node;
        while (*p) {
@@ -531,27 +606,28 @@ static void free_page_series(unsigned long from, unsigned long to)
                struct page *page = virt_to_page(from);
 
                kdebug("- free %lx", from);
-               atomic_dec(&mmap_pages_allocated);
+               atomic_long_dec(&mmap_pages_allocated);
                if (page_count(page) != 1)
-                       kdebug("free page %p [%d]", page, page_count(page));
+                       kdebug("free page %p: refcount not one: %d",
+                              page, page_count(page));
                put_page(page);
        }
 }
 
 /*
  * release a reference to a region
- * - the caller must hold the region semaphore, which this releases
+ * - the caller must hold the region semaphore for writing, which this releases
  * - the region may not have been added to the tree yet, in which case vm_top
  *   will equal vm_start
  */
 static void __put_nommu_region(struct vm_region *region)
        __releases(nommu_region_sem)
 {
-       kenter("%p{%d}", region, atomic_read(&region->vm_usage));
+       kenter("%p{%d}", region, region->vm_usage);
 
        BUG_ON(!nommu_region_tree.rb_node);
 
-       if (atomic_dec_and_test(&region->vm_usage)) {
+       if (--region->vm_usage == 0) {
                if (region->vm_top > region->vm_start)
                        delete_nommu_region(region);
                up_write(&nommu_region_sem);
@@ -581,6 +657,22 @@ static void put_nommu_region(struct vm_region *region)
 }
 
 /*
+ * update protection on a vma
+ */
+static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
+{
+#ifdef CONFIG_MPU
+       struct mm_struct *mm = vma->vm_mm;
+       long start = vma->vm_start & PAGE_MASK;
+       while (start < vma->vm_end) {
+               protect_page(mm, start, flags);
+               start += PAGE_SIZE;
+       }
+       update_protections(mm);
+#endif
+}
+
+/*
  * add a VMA into a process's mm_struct in the appropriate place in the list
  * and tree and add to the address space's page tree also if not an anonymous
  * page
@@ -588,7 +680,7 @@ static void put_nommu_region(struct vm_region *region)
  */
 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 {
-       struct vm_area_struct *pvma, **pp;
+       struct vm_area_struct *pvma, **pp, *next;
        struct address_space *mapping;
        struct rb_node **p, *parent;
 
@@ -599,6 +691,8 @@ static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
        mm->map_count++;
        vma->vm_mm = mm;
 
+       protect_vma(vma, vma->vm_flags);
+
        /* add the VMA to the mapping */
        if (vma->vm_file) {
                mapping = vma->vm_file->f_mapping;
@@ -646,8 +740,11 @@ static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
                        break;
        }
 
-       vma->vm_next = *pp;
+       next = *pp;
        *pp = vma;
+       vma->vm_next = next;
+       if (next)
+               next->vm_prev = vma;
 }
 
 /*
@@ -661,6 +758,8 @@ static void delete_vma_from_mm(struct vm_area_struct *vma)
 
        kenter("%p", vma);
 
+       protect_vma(vma, 0);
+
        mm->map_count--;
        if (mm->mmap_cache == vma)
                mm->mmap_cache = NULL;
@@ -802,7 +901,7 @@ static int validate_mmap_request(struct file *file,
        int ret;
 
        /* do the simple checks first */
-       if (flags & MAP_FIXED || addr) {
+       if (flags & MAP_FIXED) {
                printk(KERN_DEBUG
                       "%d: Can't do fixed-address/overlay mmap of RAM\n",
                       current->pid);
@@ -873,6 +972,10 @@ static int validate_mmap_request(struct file *file,
                if (!file->f_op->read)
                        capabilities &= ~BDI_CAP_MAP_COPY;
 
+               /* The file shall have been opened with read permission. */
+               if (!(file->f_mode & FMODE_READ))
+                       return -EACCES;
+
                if (flags & MAP_SHARED) {
                        /* do checks for writing, appending and locking */
                        if ((prot & PROT_WRITE) &&
@@ -889,14 +992,6 @@ static int validate_mmap_request(struct file *file,
                        if (!(capabilities & BDI_CAP_MAP_DIRECT))
                                return -ENODEV;
 
-                       if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
-                           ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
-                           ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
-                           ) {
-                               printk("MAP_SHARED not completely supported on !MMU\n");
-                               return -EINVAL;
-                       }
-
                        /* we mustn't privatise shared mappings */
                        capabilities &= ~BDI_CAP_MAP_COPY;
                }
@@ -912,6 +1007,20 @@ static int validate_mmap_request(struct file *file,
                                capabilities &= ~BDI_CAP_MAP_DIRECT;
                }
 
+               if (capabilities & BDI_CAP_MAP_DIRECT) {
+                       if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
+                           ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
+                           ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
+                           ) {
+                               capabilities &= ~BDI_CAP_MAP_DIRECT;
+                               if (flags & MAP_SHARED) {
+                                       printk(KERN_WARNING
+                                              "MAP_SHARED not completely supported on !MMU\n");
+                                       return -EINVAL;
+                               }
+                       }
+               }
+
                /* handle executable mappings and implied executable
                 * mappings */
                if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
@@ -967,22 +1076,20 @@ static unsigned long determine_vm_flags(struct file *file,
        unsigned long vm_flags;
 
        vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
-       vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
        /* vm_flags |= mm->def_flags; */
 
        if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
                /* attempt to share read-only copies of mapped file chunks */
+               vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
                if (file && !(prot & PROT_WRITE))
                        vm_flags |= VM_MAYSHARE;
-       }
-       else {
+       } else {
                /* overlay a shareable mapping on the backing device or inode
                 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
                 * romfs/cramfs */
+               vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
                if (flags & MAP_SHARED)
-                       vm_flags |= VM_MAYSHARE | VM_SHARED;
-               else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
-                       vm_flags |= VM_MAYSHARE;
+                       vm_flags |= VM_SHARED;
        }
 
        /* refuse to let anyone share private mappings with this process if
@@ -1006,15 +1113,14 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
        ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
        if (ret == 0) {
                vma->vm_region->vm_top = vma->vm_region->vm_end;
-               return ret;
+               return 0;
        }
        if (ret != -ENOSYS)
                return ret;
 
-       /* getting an ENOSYS error indicates that direct mmap isn't
-        * possible (as opposed to tried but failed) so we'll fall
-        * through to making a private copy of the data and mapping
-        * that if we can */
+       /* getting -ENOSYS indicates that direct mmap isn't possible (as
+        * opposed to tried but failed) so we can only give a suitable error as
+        * it's not possible to make a private copy if MAP_SHARED was given */
        return -ENODEV;
 }
 
@@ -1023,7 +1129,8 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
  */
 static int do_mmap_private(struct vm_area_struct *vma,
                           struct vm_region *region,
-                          unsigned long len)
+                          unsigned long len,
+                          unsigned long capabilities)
 {
        struct page *pages;
        unsigned long total, point, n, rlen;
@@ -1034,13 +1141,13 @@ static int do_mmap_private(struct vm_area_struct *vma,
         * shared mappings on devices or memory
         * - VM_MAYSHARE will be set if it may attempt to share
         */
-       if (vma->vm_file) {
+       if (capabilities & BDI_CAP_MAP_DIRECT) {
                ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
                if (ret == 0) {
                        /* shouldn't return success if we're not sharing */
                        BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
                        vma->vm_region->vm_top = vma->vm_region->vm_end;
-                       return ret;
+                       return 0;
                }
                if (ret != -ENOSYS)
                        return ret;
@@ -1064,7 +1171,7 @@ static int do_mmap_private(struct vm_area_struct *vma,
                goto enomem;
 
        total = 1 << order;
-       atomic_add(total, &mmap_pages_allocated);
+       atomic_long_add(total, &mmap_pages_allocated);
 
        point = rlen >> PAGE_SHIFT;
 
@@ -1075,7 +1182,7 @@ static int do_mmap_private(struct vm_area_struct *vma,
                        order = ilog2(total - point);
                        n = 1 << order;
                        kdebug("shave %lu/%lu @%lu", n, total - point, total);
-                       atomic_sub(n, &mmap_pages_allocated);
+                       atomic_long_sub(n, &mmap_pages_allocated);
                        total -= n;
                        set_page_refcounted(pages + total);
                        __free_pages(pages + total, order);
@@ -1114,9 +1221,6 @@ static int do_mmap_private(struct vm_area_struct *vma,
                if (ret < rlen)
                        memset(base + ret, 0, rlen - ret);
 
-       } else {
-               /* if it's an anonymous mapping, then just clear it */
-               memset(base, 0, rlen);
        }
 
        return 0;
@@ -1129,8 +1233,8 @@ error_free:
        return ret;
 
 enomem:
-       printk("Allocation of length %lu from process %d failed\n",
-              len, current->pid);
+       printk("Allocation of length %lu from process %d (%s) failed\n",
+              len, current->pid, current->comm);
        show_free_areas();
        return -ENOMEM;
 }
@@ -1153,9 +1257,6 @@ unsigned long do_mmap_pgoff(struct file *file,
 
        kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
 
-       if (!(flags & MAP_FIXED))
-               addr = round_hint_to_min(addr);
-
        /* decide whether we should attempt the mapping, and if so what sort of
         * mapping */
        ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
@@ -1165,6 +1266,9 @@ unsigned long do_mmap_pgoff(struct file *file,
                return ret;
        }
 
+       /* we ignore the address hint */
+       addr = 0;
+
        /* we've determined that we can make the mapping, now translate what we
         * now know into VMA flags */
        vm_flags = determine_vm_flags(file, prot, flags, capabilities);
@@ -1178,11 +1282,11 @@ unsigned long do_mmap_pgoff(struct file *file,
        if (!vma)
                goto error_getting_vma;
 
-       atomic_set(&region->vm_usage, 1);
+       region->vm_usage = 1;
        region->vm_flags = vm_flags;
        region->vm_pgoff = pgoff;
 
-       INIT_LIST_HEAD(&vma->anon_vma_node);
+       INIT_LIST_HEAD(&vma->anon_vma_chain);
        vma->vm_flags = vm_flags;
        vma->vm_pgoff = pgoff;
 
@@ -1245,7 +1349,7 @@ unsigned long do_mmap_pgoff(struct file *file,
                        }
 
                        /* we've found a region we can share */
-                       atomic_inc(&pregion->vm_usage);
+                       pregion->vm_usage++;
                        vma->vm_region = pregion;
                        start = pregion->vm_start;
                        start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
@@ -1262,7 +1366,7 @@ unsigned long do_mmap_pgoff(struct file *file,
                                        vma->vm_region = NULL;
                                        vma->vm_start = 0;
                                        vma->vm_end = 0;
-                                       atomic_dec(&pregion->vm_usage);
+                                       pregion->vm_usage--;
                                        pregion = NULL;
                                        goto error_just_free;
                                }
@@ -1278,7 +1382,7 @@ unsigned long do_mmap_pgoff(struct file *file,
                 * - this is the hook for quasi-memory character devices to
                 *   tell us the location of a shared mapping
                 */
-               if (file && file->f_op->get_unmapped_area) {
+               if (capabilities & BDI_CAP_MAP_DIRECT) {
                        addr = file->f_op->get_unmapped_area(file, addr, len,
                                                             pgoff, flags);
                        if (IS_ERR((void *) addr)) {
@@ -1303,16 +1407,22 @@ unsigned long do_mmap_pgoff(struct file *file,
 
        vma->vm_region = region;
 
-       /* set up the mapping */
+       /* set up the mapping
+        * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
+        */
        if (file && vma->vm_flags & VM_SHARED)
                ret = do_mmap_shared_file(vma);
        else
-               ret = do_mmap_private(vma, region, len);
+               ret = do_mmap_private(vma, region, len, capabilities);
        if (ret < 0)
-               goto error_put_region;
-
+               goto error_just_free;
        add_nommu_region(region);
 
+       /* clear anonymous mappings that don't ask for uninitialized data */
+       if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
+               memset((void *)region->vm_start, 0,
+                      region->vm_end - region->vm_start);
+
        /* okay... we have a mapping; now we have to register it */
        result = vma->vm_start;
 
@@ -1321,33 +1431,26 @@ unsigned long do_mmap_pgoff(struct file *file,
 share:
        add_vma_to_mm(current->mm, vma);
 
-       up_write(&nommu_region_sem);
+       /* we flush the region from the icache only when the first executable
+        * mapping of it is made  */
+       if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
+               flush_icache_range(region->vm_start, region->vm_end);
+               region->vm_icache_flushed = true;
+       }
 
-       if (prot & PROT_EXEC)
-               flush_icache_range(result, result + len);
+       up_write(&nommu_region_sem);
 
        kleave(" = %lx", result);
        return result;
 
-error_put_region:
-       __put_nommu_region(region);
-       if (vma) {
-               if (vma->vm_file) {
-                       fput(vma->vm_file);
-                       if (vma->vm_flags & VM_EXECUTABLE)
-                               removed_exe_file_vma(vma->vm_mm);
-               }
-               kmem_cache_free(vm_area_cachep, vma);
-       }
-       kleave(" = %d [pr]", ret);
-       return ret;
-
 error_just_free:
        up_write(&nommu_region_sem);
 error:
-       fput(region->vm_file);
+       if (region->vm_file)
+               fput(region->vm_file);
        kmem_cache_free(vm_region_jar, region);
-       fput(vma->vm_file);
+       if (vma->vm_file)
+               fput(vma->vm_file);
        if (vma->vm_flags & VM_EXECUTABLE)
                removed_exe_file_vma(vma->vm_mm);
        kmem_cache_free(vm_area_cachep, vma);
@@ -1377,6 +1480,56 @@ error_getting_region:
 }
 EXPORT_SYMBOL(do_mmap_pgoff);
 
+SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
+               unsigned long, prot, unsigned long, flags,
+               unsigned long, fd, unsigned long, pgoff)
+{
+       struct file *file = NULL;
+       unsigned long retval = -EBADF;
+
+       audit_mmap_fd(fd, flags);
+       if (!(flags & MAP_ANONYMOUS)) {
+               file = fget(fd);
+               if (!file)
+                       goto out;
+       }
+
+       flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
+
+       down_write(&current->mm->mmap_sem);
+       retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
+       up_write(&current->mm->mmap_sem);
+
+       if (file)
+               fput(file);
+out:
+       return retval;
+}
+
+#ifdef __ARCH_WANT_SYS_OLD_MMAP
+struct mmap_arg_struct {
+       unsigned long addr;
+       unsigned long len;
+       unsigned long prot;
+       unsigned long flags;
+       unsigned long fd;
+       unsigned long offset;
+};
+
+SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
+{
+       struct mmap_arg_struct a;
+
+       if (copy_from_user(&a, arg, sizeof(a)))
+               return -EFAULT;
+       if (a.offset & ~PAGE_MASK)
+               return -EINVAL;
+
+       return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
+                             a.offset >> PAGE_SHIFT);
+}
+#endif /* __ARCH_WANT_SYS_OLD_MMAP */
+
 /*
  * split a vma into two pieces at address 'addr', a new vma is allocated either
  * for the first part or the tail.
@@ -1390,10 +1543,9 @@ int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
 
        kenter("");
 
-       /* we're only permitted to split anonymous regions that have a single
-        * owner */
-       if (vma->vm_file ||
-           atomic_read(&vma->vm_region->vm_usage) != 1)
+       /* we're only permitted to split anonymous regions (these should have
+        * only a single usage on the region) */
+       if (vma->vm_file)
                return -ENOMEM;
 
        if (mm->map_count >= sysctl_max_map_count)
@@ -1467,7 +1619,7 @@ static int shrink_vma(struct mm_struct *mm,
 
        /* cut the backing region down to size */
        region = vma->vm_region;
-       BUG_ON(atomic_read(&region->vm_usage) != 1);
+       BUG_ON(region->vm_usage != 1);
 
        down_write(&nommu_region_sem);
        delete_nommu_region(region);
@@ -1504,10 +1656,15 @@ int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
        /* find the first potentially overlapping VMA */
        vma = find_vma(mm, start);
        if (!vma) {
-               printk(KERN_WARNING
-                      "munmap of memory not mmapped by process %d (%s):"
-                      " 0x%lx-0x%lx\n",
-                      current->pid, current->comm, start, start + len - 1);
+               static int limit = 0;
+               if (limit < 5) {
+                       printk(KERN_WARNING
+                              "munmap of memory not mmapped by process %d"
+                              " (%s): 0x%lx-0x%lx\n",
+                              current->pid, current->comm,
+                              start, start + len - 1);
+                       limit++;
+               }
                return -EINVAL;
        }
 
@@ -1559,7 +1716,7 @@ erase_whole_vma:
 }
 EXPORT_SYMBOL(do_munmap);
 
-asmlinkage long sys_munmap(unsigned long addr, size_t len)
+SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
 {
        int ret;
        struct mm_struct *mm = current->mm;
@@ -1588,6 +1745,7 @@ void exit_mmap(struct mm_struct *mm)
                mm->mmap = vma->vm_next;
                delete_vma_from_mm(vma);
                delete_vma(mm, vma);
+               cond_resched();
        }
 
        kleave("");
@@ -1643,10 +1801,9 @@ unsigned long do_mremap(unsigned long addr,
 }
 EXPORT_SYMBOL(do_mremap);
 
-asmlinkage
-unsigned long sys_mremap(unsigned long addr,
-                        unsigned long old_len, unsigned long new_len,
-                        unsigned long flags, unsigned long new_addr)
+SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
+               unsigned long, new_len, unsigned long, flags,
+               unsigned long, new_addr)
 {
        unsigned long ret;
 
@@ -1685,10 +1842,6 @@ int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 }
 EXPORT_SYMBOL(remap_vmalloc_range);
 
-void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
-{
-}
-
 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
        unsigned long len, unsigned long pgoff, unsigned long flags)
 {
@@ -1707,27 +1860,6 @@ void unmap_mapping_range(struct address_space *mapping,
 EXPORT_SYMBOL(unmap_mapping_range);
 
 /*
- * ask for an unmapped area at which to create a mapping on a file
- */
-unsigned long get_unmapped_area(struct file *file, unsigned long addr,
-                               unsigned long len, unsigned long pgoff,
-                               unsigned long flags)
-{
-       unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
-                                 unsigned long, unsigned long);
-
-       get_area = current->mm->get_unmapped_area;
-       if (file && file->f_op && file->f_op->get_unmapped_area)
-               get_area = file->f_op->get_unmapped_area;
-
-       if (!get_area)
-               return -ENOSYS;
-
-       return get_area(file, addr, len, pgoff, flags);
-}
-EXPORT_SYMBOL(get_unmapped_area);
-
-/*
  * Check that a process has enough memory to allocate a new virtual
  * mapping. 0 means there is enough memory for the allocation to
  * succeed and -ENOMEM implies there is not.
@@ -1818,19 +1950,16 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
        if (mm)
                allowed -= mm->total_vm / 32;
 
-       /*
-        * cast `allowed' as a signed long because vm_committed_space
-        * sometimes has a negative value
-        */
-       if (atomic_long_read(&vm_committed_space) < (long)allowed)
+       if (percpu_counter_read_positive(&vm_committed_as) < allowed)
                return 0;
+
 error:
        vm_unacct_memory(pages);
 
        return -ENOMEM;
 }
 
-int in_gate_area_no_task(unsigned long addr)
+int in_gate_area_no_mm(unsigned long addr)
 {
        return 0;
 }
@@ -1842,21 +1971,10 @@ int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 }
 EXPORT_SYMBOL(filemap_fault);
 
-/*
- * Access another process' address space.
- * - source/target buffer must be kernel space
- */
-int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
+static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
+               unsigned long addr, void *buf, int len, int write)
 {
        struct vm_area_struct *vma;
-       struct mm_struct *mm;
-
-       if (addr + len < addr)
-               return 0;
-
-       mm = get_task_mm(tsk);
-       if (!mm)
-               return 0;
 
        down_read(&mm->mmap_sem);
 
@@ -1869,9 +1987,11 @@ int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, in
 
                /* only read or write mappings where it is permitted */
                if (write && vma->vm_flags & VM_MAYWRITE)
-                       len -= copy_to_user((void *) addr, buf, len);
+                       copy_to_user_page(vma, NULL, addr,
+                                        (void *) addr, buf, len);
                else if (!write && vma->vm_flags & VM_MAYREAD)
-                       len -= copy_from_user(buf, (void *) addr, len);
+                       copy_from_user_page(vma, NULL, addr,
+                                           buf, (void *) addr, len);
                else
                        len = 0;
        } else {
@@ -1879,6 +1999,105 @@ int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, in
        }
 
        up_read(&mm->mmap_sem);
+
+       return len;
+}
+
+/**
+ * @access_remote_vm - access another process' address space
+ * @mm:                the mm_struct of the target address space
+ * @addr:      start address to access
+ * @buf:       source or destination buffer
+ * @len:       number of bytes to transfer
+ * @write:     whether the access is a write
+ *
+ * The caller must hold a reference on @mm.
+ */
+int access_remote_vm(struct mm_struct *mm, unsigned long addr,
+               void *buf, int len, int write)
+{
+       return __access_remote_vm(NULL, mm, addr, buf, len, write);
+}
+
+/*
+ * Access another process' address space.
+ * - source/target buffer must be kernel space
+ */
+int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
+{
+       struct mm_struct *mm;
+
+       if (addr + len < addr)
+               return 0;
+
+       mm = get_task_mm(tsk);
+       if (!mm)
+               return 0;
+
+       len = __access_remote_vm(tsk, mm, addr, buf, len, write);
+
        mmput(mm);
        return len;
 }
+
+/**
+ * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
+ * @inode: The inode to check
+ * @size: The current filesize of the inode
+ * @newsize: The proposed filesize of the inode
+ *
+ * Check the shared mappings on an inode on behalf of a shrinking truncate to
+ * make sure that that any outstanding VMAs aren't broken and then shrink the
+ * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
+ * automatically grant mappings that are too large.
+ */
+int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
+                               size_t newsize)
+{
+       struct vm_area_struct *vma;
+       struct prio_tree_iter iter;
+       struct vm_region *region;
+       pgoff_t low, high;
+       size_t r_size, r_top;
+
+       low = newsize >> PAGE_SHIFT;
+       high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
+
+       down_write(&nommu_region_sem);
+
+       /* search for VMAs that fall within the dead zone */
+       vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
+                             low, high) {
+               /* found one - only interested if it's shared out of the page
+                * cache */
+               if (vma->vm_flags & VM_SHARED) {
+                       up_write(&nommu_region_sem);
+                       return -ETXTBSY; /* not quite true, but near enough */
+               }
+       }
+
+       /* reduce any regions that overlap the dead zone - if in existence,
+        * these will be pointed to by VMAs that don't overlap the dead zone
+        *
+        * we don't check for any regions that start beyond the EOF as there
+        * shouldn't be any
+        */
+       vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
+                             0, ULONG_MAX) {
+               if (!(vma->vm_flags & VM_SHARED))
+                       continue;
+
+               region = vma->vm_region;
+               r_size = region->vm_top - region->vm_start;
+               r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
+
+               if (r_top > newsize) {
+                       region->vm_top -= r_top - newsize;
+                       if (region->vm_end > region->vm_top)
+                               region->vm_end = region->vm_top;
+               }
+       }
+
+       up_write(&nommu_region_sem);
+       return 0;
+}