clarify get_user_pages() prototype
[linux-2.6.git] / mm / nommu.c
index aee0e1b..bf0cc76 100644 (file)
@@ -6,12 +6,14 @@
  *
  *  See Documentation/nommu-mmap.txt
  *
- *  Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
+ *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
  *  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) 2007-2009 Paul Mundt <lethal@linux-sh.org>
  */
 
+#include <linux/module.h>
 #include <linux/mm.h>
 #include <linux/mman.h>
 #include <linux/swap.h>
@@ -20,7 +22,7 @@
 #include <linux/pagemap.h>
 #include <linux/slab.h>
 #include <linux/vmalloc.h>
-#include <linux/ptrace.h>
+#include <linux/tracehook.h>
 #include <linux/blkdev.h>
 #include <linux/backing-dev.h>
 #include <linux/mount.h>
 #include <asm/uaccess.h>
 #include <asm/tlb.h>
 #include <asm/tlbflush.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__)
+#define kleave(FMT, ...) \
+       printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
+#define kdebug(FMT, ...) \
+       printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
+#else
+#define kenter(FMT, ...) \
+       no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
+#define kleave(FMT, ...) \
+       no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
+#define kdebug(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;
-unsigned long askedalloc, realalloc;
-atomic_t vm_committed_space = ATOMIC_INIT(0);
+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 = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
 int heap_stack_gap = 0;
 
+/* amount of vm to protect from userspace access */
+unsigned long mmap_min_addr = CONFIG_DEFAULT_MMAP_MIN_ADDR;
+
+atomic_long_t mmap_pages_allocated;
+
 EXPORT_SYMBOL(mem_map);
-EXPORT_SYMBOL(__vm_enough_memory);
 EXPORT_SYMBOL(num_physpages);
 
-/* list of shareable VMAs */
-struct rb_root nommu_vma_tree = RB_ROOT;
-DECLARE_RWSEM(nommu_vma_sem);
+/* list of mapped, potentially shareable regions */
+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 = {
 };
 
-EXPORT_SYMBOL(vfree);
-EXPORT_SYMBOL(vmalloc_to_page);
-EXPORT_SYMBOL(vmalloc_32);
-EXPORT_SYMBOL(vmap);
-EXPORT_SYMBOL(vunmap);
-
 /*
  * Handle all mappings that got truncated by a "truncate()"
  * system call.
@@ -89,7 +114,7 @@ do_expand:
        i_size_write(inode, offset);
 
 out_truncate:
-       if (inode->i_op && inode->i_op->truncate)
+       if (inode->i_op->truncate)
                inode->i_op->truncate(inode);
        return 0;
 out_sig:
@@ -110,32 +135,53 @@ unsigned int kobjsize(const void *objp)
 {
        struct page *page;
 
-       if (!objp || !((page = virt_to_page(objp))))
+       /*
+        * If the object we have should not have ksize performed on it,
+        * return size of 0
+        */
+       if (!objp || !virt_addr_valid(objp))
                return 0;
 
+       page = virt_to_head_page(objp);
+
+       /*
+        * If the allocator sets PageSlab, we know the pointer came from
+        * kmalloc().
+        */
        if (PageSlab(page))
                return ksize(objp);
 
-       BUG_ON(page->index < 0);
-       BUG_ON(page->index >= MAX_ORDER);
+       /*
+        * 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;
+       }
 
-       return (PAGE_SIZE << page->index);
+       /*
+        * The ksize() function is only guaranteed to work for pointers
+        * returned by kmalloc(). So handle arbitrary pointers here.
+        */
+       return PAGE_SIZE << compound_order(page);
 }
 
-/*
- * get a list of pages in an address range belonging to the specified process
- * and indicate the VMA that covers each page
- * - this is potentially dodgy as we may end incrementing the page count of a
- *   slab page or a secondary page from a compound page
- * - 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,
-       struct page **pages, struct vm_area_struct **vmas)
+int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
+                    unsigned long start, int nr_pages, int flags,
+                    struct page **pages, struct vm_area_struct **vmas)
 {
        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.
@@ -143,14 +189,14 @@ int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
        vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
        vm_flags &= 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) ||
-                   !(vm_flags & vma->vm_flags))
+                   (!ignore && !(vm_flags & vma->vm_flags)))
                        goto finish_or_fault;
 
                if (pages) {
@@ -169,34 +215,79 @@ 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
+ * - this is potentially dodgy as we may end incrementing the page count of a
+ *   slab page or a secondary page from a compound page
+ * - 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 nr_pages, int write, int force,
+       struct page **pages, struct vm_area_struct **vmas)
+{
+       int flags = 0;
+
+       if (write)
+               flags |= GUP_FLAGS_WRITE;
+       if (force)
+               flags |= GUP_FLAGS_FORCE;
+
+       return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
+}
 EXPORT_SYMBOL(get_user_pages);
 
 DEFINE_RWLOCK(vmlist_lock);
 struct vm_struct *vmlist;
 
-void vfree(void *addr)
+void vfree(const void *addr)
 {
        kfree(addr);
 }
+EXPORT_SYMBOL(vfree);
 
 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 {
        /*
-        * kmalloc doesn't like __GFP_HIGHMEM for some reason
+        *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
+        * returns only a logical address.
         */
        return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 }
+EXPORT_SYMBOL(__vmalloc);
+
+void *vmalloc_user(unsigned long size)
+{
+       void *ret;
+
+       ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
+                       PAGE_KERNEL);
+       if (ret) {
+               struct vm_area_struct *vma;
+
+               down_write(&current->mm->mmap_sem);
+               vma = find_vma(current->mm, (unsigned long)ret);
+               if (vma)
+                       vma->vm_flags |= VM_USERMAP;
+               up_write(&current->mm->mmap_sem);
+       }
+
+       return ret;
+}
+EXPORT_SYMBOL(vmalloc_user);
 
-struct page * vmalloc_to_page(void *addr)
+struct page *vmalloc_to_page(const void *addr)
 {
        return virt_to_page(addr);
 }
+EXPORT_SYMBOL(vmalloc_to_page);
 
-unsigned long vmalloc_to_pfn(void *addr)
+unsigned long vmalloc_to_pfn(const void *addr)
 {
        return page_to_pfn(virt_to_page(addr));
 }
-
+EXPORT_SYMBOL(vmalloc_to_pfn);
 
 long vread(char *buf, char *addr, unsigned long count)
 {
@@ -237,9 +328,29 @@ void *vmalloc_node(unsigned long size, int node)
 }
 EXPORT_SYMBOL(vmalloc_node);
 
-/*
- *     vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
+#ifndef PAGE_KERNEL_EXEC
+# define PAGE_KERNEL_EXEC PAGE_KERNEL
+#endif
+
+/**
+ *     vmalloc_exec  -  allocate virtually contiguous, executable memory
+ *     @size:          allocation size
+ *
+ *     Kernel-internal function to allocate enough pages to cover @size
+ *     the page level allocator and map them into contiguous and
+ *     executable kernel virtual space.
  *
+ *     For tight control over page level allocator and protection flags
+ *     use __vmalloc() instead.
+ */
+
+void *vmalloc_exec(unsigned long size)
+{
+       return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
+}
+
+/**
+ * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
  *     @size:          allocation size
  *
  *     Allocate enough 32bit PA addressable pages to cover @size from the
@@ -249,17 +360,58 @@ void *vmalloc_32(unsigned long size)
 {
        return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 }
+EXPORT_SYMBOL(vmalloc_32);
+
+/**
+ * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
+ *     @size:          allocation size
+ *
+ * The resulting memory area is 32bit addressable and zeroed so it can be
+ * mapped to userspace without leaking data.
+ *
+ * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
+ * remap_vmalloc_range() are permissible.
+ */
+void *vmalloc_32_user(unsigned long size)
+{
+       /*
+        * We'll have to sort out the ZONE_DMA bits for 64-bit,
+        * but for now this can simply use vmalloc_user() directly.
+        */
+       return vmalloc_user(size);
+}
+EXPORT_SYMBOL(vmalloc_32_user);
 
 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 {
        BUG();
        return NULL;
 }
+EXPORT_SYMBOL(vmap);
+
+void vunmap(const void *addr)
+{
+       BUG();
+}
+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 vunmap(void *addr)
+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
@@ -269,6 +421,13 @@ void  __attribute__((weak)) vmalloc_sync_all(void)
 {
 }
 
+int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
+                  struct page *page)
+{
+       return -EINVAL;
+}
+EXPORT_SYMBOL(vm_insert_page);
+
 /*
  *  sys_brk() for the most part doesn't need the global kernel
  *  lock, except when an application is doing something nasty
@@ -276,7 +435,7 @@ void  __attribute__((weak)) vmalloc_sync_all(void)
  *  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;
 
@@ -300,129 +459,173 @@ asmlinkage unsigned long sys_brk(unsigned long brk)
        return mm->brk = brk;
 }
 
-#ifdef DEBUG
-static void show_process_blocks(void)
+/*
+ * initialise the VMA and region record slabs
+ */
+void __init mmap_init(void)
 {
-       struct vm_list_struct *vml;
-
-       printk("Process blocks %d:", current->pid);
+       int ret;
 
-       for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
-               printk(" %p: %p", vml, vml->vma);
-               if (vml->vma)
-                       printk(" (%d @%lx #%d)",
-                              kobjsize((void *) vml->vma->vm_start),
-                              vml->vma->vm_start,
-                              atomic_read(&vml->vma->vm_usage));
-               printk(vml->next ? " ->" : ".\n");
-       }
+       ret = percpu_counter_init(&vm_committed_as, 0);
+       VM_BUG_ON(ret);
+       vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 }
-#endif /* DEBUG */
 
 /*
- * add a VMA into a process's mm_struct in the appropriate place in the list
- * - should be called with mm->mmap_sem held writelocked
+ * validate the region tree
+ * - the caller must hold the region lock
  */
-static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
+#ifdef CONFIG_DEBUG_NOMMU_REGIONS
+static noinline void validate_nommu_regions(void)
 {
-       struct vm_list_struct **ppv;
+       struct vm_region *region, *last;
+       struct rb_node *p, *lastp;
 
-       for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
-               if ((*ppv)->vma->vm_start > vml->vma->vm_start)
-                       break;
+       lastp = rb_first(&nommu_region_tree);
+       if (!lastp)
+               return;
+
+       last = rb_entry(lastp, struct vm_region, vm_rb);
+       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);
+
+               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));
 
-       vml->next = *ppv;
-       *ppv = vml;
+               lastp = p;
+       }
+}
+#else
+static void validate_nommu_regions(void)
+{
 }
+#endif
 
 /*
- * look up the first VMA in which addr resides, NULL if none
- * - should be called with mm->mmap_sem at least held readlocked
+ * add a region into the global tree
  */
-struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
+static void add_nommu_region(struct vm_region *region)
 {
-       struct vm_list_struct *loop, *vml;
+       struct vm_region *pregion;
+       struct rb_node **p, *parent;
 
-       /* search the vm_start ordered list */
-       vml = NULL;
-       for (loop = mm->context.vmlist; loop; loop = loop->next) {
-               if (loop->vma->vm_start > addr)
-                       break;
-               vml = loop;
+       validate_nommu_regions();
+
+       parent = NULL;
+       p = &nommu_region_tree.rb_node;
+       while (*p) {
+               parent = *p;
+               pregion = rb_entry(parent, struct vm_region, vm_rb);
+               if (region->vm_start < pregion->vm_start)
+                       p = &(*p)->rb_left;
+               else if (region->vm_start > pregion->vm_start)
+                       p = &(*p)->rb_right;
+               else if (pregion == region)
+                       return;
+               else
+                       BUG();
        }
 
-       if (vml && vml->vma->vm_end > addr)
-               return vml->vma;
+       rb_link_node(&region->vm_rb, parent, p);
+       rb_insert_color(&region->vm_rb, &nommu_region_tree);
 
-       return NULL;
+       validate_nommu_regions();
 }
-EXPORT_SYMBOL(find_vma);
 
 /*
- * find a VMA
- * - we don't extend stack VMAs under NOMMU conditions
+ * delete a region from the global tree
  */
-struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
+static void delete_nommu_region(struct vm_region *region)
 {
-       return find_vma(mm, addr);
-}
+       BUG_ON(!nommu_region_tree.rb_node);
 
-int expand_stack(struct vm_area_struct *vma, unsigned long address)
-{
-       return -ENOMEM;
+       validate_nommu_regions();
+       rb_erase(&region->vm_rb, &nommu_region_tree);
+       validate_nommu_regions();
 }
 
 /*
- * look up the first VMA exactly that exactly matches addr
- * - should be called with mm->mmap_sem at least held readlocked
+ * free a contiguous series of pages
  */
-static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
-                                                   unsigned long addr)
+static void free_page_series(unsigned long from, unsigned long to)
 {
-       struct vm_list_struct *vml;
-
-       /* search the vm_start ordered list */
-       for (vml = mm->context.vmlist; vml; vml = vml->next) {
-               if (vml->vma->vm_start == addr)
-                       return vml->vma;
-               if (vml->vma->vm_start > addr)
-                       break;
+       for (; from < to; from += PAGE_SIZE) {
+               struct page *page = virt_to_page(from);
+
+               kdebug("- free %lx", from);
+               atomic_long_dec(&mmap_pages_allocated);
+               if (page_count(page) != 1)
+                       kdebug("free page %p: refcount not one: %d",
+                              page, page_count(page));
+               put_page(page);
        }
-
-       return NULL;
 }
 
 /*
- * find a VMA in the global tree
+ * release a reference to a region
+ * - 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 inline struct vm_area_struct *find_nommu_vma(unsigned long start)
+static void __put_nommu_region(struct vm_region *region)
+       __releases(nommu_region_sem)
 {
-       struct vm_area_struct *vma;
-       struct rb_node *n = nommu_vma_tree.rb_node;
+       kenter("%p{%d}", region, atomic_read(&region->vm_usage));
 
-       while (n) {
-               vma = rb_entry(n, struct vm_area_struct, vm_rb);
+       BUG_ON(!nommu_region_tree.rb_node);
 
-               if (start < vma->vm_start)
-                       n = n->rb_left;
-               else if (start > vma->vm_start)
-                       n = n->rb_right;
-               else
-                       return vma;
+       if (atomic_dec_and_test(&region->vm_usage)) {
+               if (region->vm_top > region->vm_start)
+                       delete_nommu_region(region);
+               up_write(&nommu_region_sem);
+
+               if (region->vm_file)
+                       fput(region->vm_file);
+
+               /* IO memory and memory shared directly out of the pagecache
+                * from ramfs/tmpfs mustn't be released here */
+               if (region->vm_flags & VM_MAPPED_COPY) {
+                       kdebug("free series");
+                       free_page_series(region->vm_start, region->vm_top);
+               }
+               kmem_cache_free(vm_region_jar, region);
+       } else {
+               up_write(&nommu_region_sem);
        }
+}
 
-       return NULL;
+/*
+ * release a reference to a region
+ */
+static void put_nommu_region(struct vm_region *region)
+{
+       down_write(&nommu_region_sem);
+       __put_nommu_region(region);
 }
 
 /*
- * add a VMA in the global tree
+ * 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
+ * - should be called with mm->mmap_sem held writelocked
  */
-static void add_nommu_vma(struct vm_area_struct *vma)
+static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 {
-       struct vm_area_struct *pvma;
+       struct vm_area_struct *pvma, **pp;
        struct address_space *mapping;
-       struct rb_node **p = &nommu_vma_tree.rb_node;
-       struct rb_node *parent = NULL;
+       struct rb_node **p, *parent;
+
+       kenter(",%p", vma);
+
+       BUG_ON(!vma->vm_region);
+
+       mm->map_count++;
+       vma->vm_mm = mm;
 
        /* add the VMA to the mapping */
        if (vma->vm_file) {
@@ -433,42 +636,62 @@ static void add_nommu_vma(struct vm_area_struct *vma)
                flush_dcache_mmap_unlock(mapping);
        }
 
-       /* add the VMA to the master list */
+       /* add the VMA to the tree */
+       parent = NULL;
+       p = &mm->mm_rb.rb_node;
        while (*p) {
                parent = *p;
                pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 
-               if (vma->vm_start < pvma->vm_start) {
+               /* sort by: start addr, end addr, VMA struct addr in that order
+                * (the latter is necessary as we may get identical VMAs) */
+               if (vma->vm_start < pvma->vm_start)
                        p = &(*p)->rb_left;
-               }
-               else if (vma->vm_start > pvma->vm_start) {
+               else if (vma->vm_start > pvma->vm_start)
                        p = &(*p)->rb_right;
-               }
-               else {
-                       /* mappings are at the same address - this can only
-                        * happen for shared-mem chardevs and shared file
-                        * mappings backed by ramfs/tmpfs */
-                       BUG_ON(!(pvma->vm_flags & VM_SHARED));
-
-                       if (vma < pvma)
-                               p = &(*p)->rb_left;
-                       else if (vma > pvma)
-                               p = &(*p)->rb_right;
-                       else
-                               BUG();
-               }
+               else if (vma->vm_end < pvma->vm_end)
+                       p = &(*p)->rb_left;
+               else if (vma->vm_end > pvma->vm_end)
+                       p = &(*p)->rb_right;
+               else if (vma < pvma)
+                       p = &(*p)->rb_left;
+               else if (vma > pvma)
+                       p = &(*p)->rb_right;
+               else
+                       BUG();
        }
 
        rb_link_node(&vma->vm_rb, parent, p);
-       rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
+       rb_insert_color(&vma->vm_rb, &mm->mm_rb);
+
+       /* add VMA to the VMA list also */
+       for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
+               if (pvma->vm_start > vma->vm_start)
+                       break;
+               if (pvma->vm_start < vma->vm_start)
+                       continue;
+               if (pvma->vm_end < vma->vm_end)
+                       break;
+       }
+
+       vma->vm_next = *pp;
+       *pp = vma;
 }
 
 /*
- * delete a VMA from the global list
+ * delete a VMA from its owning mm_struct and address space
  */
-static void delete_nommu_vma(struct vm_area_struct *vma)
+static void delete_vma_from_mm(struct vm_area_struct *vma)
 {
+       struct vm_area_struct **pp;
        struct address_space *mapping;
+       struct mm_struct *mm = vma->vm_mm;
+
+       kenter("%p", vma);
+
+       mm->map_count--;
+       if (mm->mmap_cache == vma)
+               mm->mmap_cache = NULL;
 
        /* remove the VMA from the mapping */
        if (vma->vm_file) {
@@ -479,8 +702,115 @@ static void delete_nommu_vma(struct vm_area_struct *vma)
                flush_dcache_mmap_unlock(mapping);
        }
 
-       /* remove from the master list */
-       rb_erase(&vma->vm_rb, &nommu_vma_tree);
+       /* remove from the MM's tree and list */
+       rb_erase(&vma->vm_rb, &mm->mm_rb);
+       for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
+               if (*pp == vma) {
+                       *pp = vma->vm_next;
+                       break;
+               }
+       }
+
+       vma->vm_mm = NULL;
+}
+
+/*
+ * destroy a VMA record
+ */
+static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
+{
+       kenter("%p", vma);
+       if (vma->vm_ops && vma->vm_ops->close)
+               vma->vm_ops->close(vma);
+       if (vma->vm_file) {
+               fput(vma->vm_file);
+               if (vma->vm_flags & VM_EXECUTABLE)
+                       removed_exe_file_vma(mm);
+       }
+       put_nommu_region(vma->vm_region);
+       kmem_cache_free(vm_area_cachep, vma);
+}
+
+/*
+ * look up the first VMA in which addr resides, NULL if none
+ * - should be called with mm->mmap_sem at least held readlocked
+ */
+struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
+{
+       struct vm_area_struct *vma;
+       struct rb_node *n = mm->mm_rb.rb_node;
+
+       /* check the cache first */
+       vma = mm->mmap_cache;
+       if (vma && vma->vm_start <= addr && vma->vm_end > addr)
+               return vma;
+
+       /* trawl the tree (there may be multiple mappings in which addr
+        * resides) */
+       for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
+               vma = rb_entry(n, struct vm_area_struct, vm_rb);
+               if (vma->vm_start > addr)
+                       return NULL;
+               if (vma->vm_end > addr) {
+                       mm->mmap_cache = vma;
+                       return vma;
+               }
+       }
+
+       return NULL;
+}
+EXPORT_SYMBOL(find_vma);
+
+/*
+ * find a VMA
+ * - we don't extend stack VMAs under NOMMU conditions
+ */
+struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
+{
+       return find_vma(mm, addr);
+}
+
+/*
+ * expand a stack to a given address
+ * - not supported under NOMMU conditions
+ */
+int expand_stack(struct vm_area_struct *vma, unsigned long address)
+{
+       return -ENOMEM;
+}
+
+/*
+ * look up the first VMA exactly that exactly matches addr
+ * - should be called with mm->mmap_sem at least held readlocked
+ */
+static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
+                                            unsigned long addr,
+                                            unsigned long len)
+{
+       struct vm_area_struct *vma;
+       struct rb_node *n = mm->mm_rb.rb_node;
+       unsigned long end = addr + len;
+
+       /* check the cache first */
+       vma = mm->mmap_cache;
+       if (vma && vma->vm_start == addr && vma->vm_end == end)
+               return vma;
+
+       /* trawl the tree (there may be multiple mappings in which addr
+        * resides) */
+       for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
+               vma = rb_entry(n, struct vm_area_struct, vm_rb);
+               if (vma->vm_start < addr)
+                       continue;
+               if (vma->vm_start > addr)
+                       return NULL;
+               if (vma->vm_end == end) {
+                       mm->mmap_cache = vma;
+                       return vma;
+               }
+       }
+
+       return NULL;
 }
 
 /*
@@ -495,7 +825,7 @@ static int validate_mmap_request(struct file *file,
                                 unsigned long pgoff,
                                 unsigned long *_capabilities)
 {
-       unsigned long capabilities;
+       unsigned long capabilities, rlen;
        unsigned long reqprot = prot;
        int ret;
 
@@ -515,12 +845,12 @@ static int validate_mmap_request(struct file *file,
                return -EINVAL;
 
        /* Careful about overflows.. */
-       len = PAGE_ALIGN(len);
-       if (!len || len > TASK_SIZE)
+       rlen = PAGE_ALIGN(len);
+       if (!rlen || rlen > TASK_SIZE)
                return -ENOMEM;
 
        /* offset overflow? */
-       if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
+       if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
                return -EOVERFLOW;
 
        if (file) {
@@ -687,20 +1017,25 @@ static unsigned long determine_vm_flags(struct file *file,
         * it's being traced - otherwise breakpoints set in it may interfere
         * with another untraced process
         */
-       if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
+       if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
                vm_flags &= ~VM_MAYSHARE;
 
        return vm_flags;
 }
 
 /*
- * set up a shared mapping on a file
+ * set up a shared mapping on a file (the driver or filesystem provides and
+ * pins the storage)
  */
-static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
+static int do_mmap_shared_file(struct vm_area_struct *vma)
 {
        int ret;
 
        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;
+       }
        if (ret != -ENOSYS)
                return ret;
 
@@ -714,10 +1049,14 @@ static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 /*
  * set up a private mapping or an anonymous shared mapping
  */
-static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
+static int do_mmap_private(struct vm_area_struct *vma,
+                          struct vm_region *region,
+                          unsigned long len)
 {
+       struct page *pages;
+       unsigned long total, point, n, rlen;
        void *base;
-       int ret;
+       int ret, order;
 
        /* invoke the file's mapping function so that it can keep track of
         * shared mappings on devices or memory
@@ -725,34 +1064,63 @@ static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
         */
        if (vma->vm_file) {
                ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
-               if (ret != -ENOSYS) {
+               if (ret == 0) {
                        /* shouldn't return success if we're not sharing */
-                       BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
-                       return ret; /* success or a real error */
+                       BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
+                       vma->vm_region->vm_top = vma->vm_region->vm_end;
+                       return ret;
                }
+               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 try to
                 * make a private copy of the data and map that instead */
        }
 
+       rlen = PAGE_ALIGN(len);
+
        /* allocate some memory to hold the mapping
         * - note that this may not return a page-aligned address if the object
         *   we're allocating is smaller than a page
         */
-       base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
-       if (!base)
+       order = get_order(rlen);
+       kdebug("alloc order %d for %lx", order, len);
+
+       pages = alloc_pages(GFP_KERNEL, order);
+       if (!pages)
                goto enomem;
 
-       vma->vm_start = (unsigned long) base;
-       vma->vm_end = vma->vm_start + len;
-       vma->vm_flags |= VM_MAPPED_COPY;
+       total = 1 << order;
+       atomic_long_add(total, &mmap_pages_allocated);
+
+       point = rlen >> PAGE_SHIFT;
+
+       /* we allocated a power-of-2 sized page set, so we may want to trim off
+        * the excess */
+       if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
+               while (total > point) {
+                       order = ilog2(total - point);
+                       n = 1 << order;
+                       kdebug("shave %lu/%lu @%lu", n, total - point, total);
+                       atomic_long_sub(n, &mmap_pages_allocated);
+                       total -= n;
+                       set_page_refcounted(pages + total);
+                       __free_pages(pages + total, order);
+               }
+       }
+
+       for (point = 1; point < total; point++)
+               set_page_refcounted(&pages[point]);
 
-#ifdef WARN_ON_SLACK
-       if (len + WARN_ON_SLACK <= kobjsize(result))
-               printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
-                      len, current->pid, kobjsize(result) - len);
-#endif
+       base = page_address(pages);
+       region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
+       region->vm_start = (unsigned long) base;
+       region->vm_end   = region->vm_start + rlen;
+       region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
+
+       vma->vm_start = region->vm_start;
+       vma->vm_end   = region->vm_start + len;
 
        if (vma->vm_file) {
                /* read the contents of a file into the copy */
@@ -764,31 +1132,33 @@ static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 
                old_fs = get_fs();
                set_fs(KERNEL_DS);
-               ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
+               ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
                set_fs(old_fs);
 
                if (ret < 0)
                        goto error_free;
 
                /* clear the last little bit */
-               if (ret < len)
-                       memset(base + ret, 0, len - ret);
+               if (ret < rlen)
+                       memset(base + ret, 0, rlen - ret);
 
        } else {
                /* if it's an anonymous mapping, then just clear it */
-               memset(base, 0, len);
+               memset(base, 0, rlen);
        }
 
        return 0;
 
 error_free:
-       kfree(base);
-       vma->vm_start = 0;
+       free_page_series(region->vm_start, region->vm_end);
+       region->vm_start = vma->vm_start = 0;
+       region->vm_end   = vma->vm_end = 0;
+       region->vm_top   = 0;
        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;
 }
@@ -803,84 +1173,135 @@ unsigned long do_mmap_pgoff(struct file *file,
                            unsigned long flags,
                            unsigned long pgoff)
 {
-       struct vm_list_struct *vml = NULL;
-       struct vm_area_struct *vma = NULL;
+       struct vm_area_struct *vma;
+       struct vm_region *region;
        struct rb_node *rb;
-       unsigned long capabilities, vm_flags;
-       void *result;
+       unsigned long capabilities, vm_flags, result;
        int ret;
 
+       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,
                                    &capabilities);
-       if (ret < 0)
+       if (ret < 0) {
+               kleave(" = %d [val]", ret);
                return ret;
+       }
 
        /* 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);
 
-       /* we're going to need to record the mapping if it works */
-       vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
-       if (!vml)
-               goto error_getting_vml;
+       /* we're going to need to record the mapping */
+       region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
+       if (!region)
+               goto error_getting_region;
+
+       vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
+       if (!vma)
+               goto error_getting_vma;
+
+       atomic_set(&region->vm_usage, 1);
+       region->vm_flags = vm_flags;
+       region->vm_pgoff = pgoff;
+
+       INIT_LIST_HEAD(&vma->anon_vma_node);
+       vma->vm_flags = vm_flags;
+       vma->vm_pgoff = pgoff;
+
+       if (file) {
+               region->vm_file = file;
+               get_file(file);
+               vma->vm_file = file;
+               get_file(file);
+               if (vm_flags & VM_EXECUTABLE) {
+                       added_exe_file_vma(current->mm);
+                       vma->vm_mm = current->mm;
+               }
+       }
 
-       down_write(&nommu_vma_sem);
+       down_write(&nommu_region_sem);
 
-       /* if we want to share, we need to check for VMAs created by other
+       /* if we want to share, we need to check for regions created by other
         * mmap() calls that overlap with our proposed mapping
-        * - we can only share with an exact match on most regular files
+        * - we can only share with a superset match on most regular files
         * - shared mappings on character devices and memory backed files are
         *   permitted to overlap inexactly as far as we are concerned for in
         *   these cases, sharing is handled in the driver or filesystem rather
         *   than here
         */
        if (vm_flags & VM_MAYSHARE) {
-               unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
-               unsigned long vmpglen;
+               struct vm_region *pregion;
+               unsigned long pglen, rpglen, pgend, rpgend, start;
 
-               /* suppress VMA sharing for shared regions */
-               if (vm_flags & VM_SHARED &&
-                   capabilities & BDI_CAP_MAP_DIRECT)
-                       goto dont_share_VMAs;
+               pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
+               pgend = pgoff + pglen;
 
-               for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
-                       vma = rb_entry(rb, struct vm_area_struct, vm_rb);
+               for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
+                       pregion = rb_entry(rb, struct vm_region, vm_rb);
 
-                       if (!(vma->vm_flags & VM_MAYSHARE))
+                       if (!(pregion->vm_flags & VM_MAYSHARE))
                                continue;
 
                        /* search for overlapping mappings on the same file */
-                       if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
+                       if (pregion->vm_file->f_path.dentry->d_inode !=
+                           file->f_path.dentry->d_inode)
                                continue;
 
-                       if (vma->vm_pgoff >= pgoff + pglen)
+                       if (pregion->vm_pgoff >= pgend)
                                continue;
 
-                       vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
-                       vmpglen >>= PAGE_SHIFT;
-                       if (pgoff >= vma->vm_pgoff + vmpglen)
+                       rpglen = pregion->vm_end - pregion->vm_start;
+                       rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
+                       rpgend = pregion->vm_pgoff + rpglen;
+                       if (pgoff >= rpgend)
                                continue;
 
-                       /* handle inexactly overlapping matches between mappings */
-                       if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
+                       /* handle inexactly overlapping matches between
+                        * mappings */
+                       if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
+                           !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
+                               /* new mapping is not a subset of the region */
                                if (!(capabilities & BDI_CAP_MAP_DIRECT))
                                        goto sharing_violation;
                                continue;
                        }
 
-                       /* we've found a VMA we can share */
-                       atomic_inc(&vma->vm_usage);
-
-                       vml->vma = vma;
-                       result = (void *) vma->vm_start;
-                       goto shared;
+                       /* we've found a region we can share */
+                       atomic_inc(&pregion->vm_usage);
+                       vma->vm_region = pregion;
+                       start = pregion->vm_start;
+                       start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
+                       vma->vm_start = start;
+                       vma->vm_end = start + len;
+
+                       if (pregion->vm_flags & VM_MAPPED_COPY) {
+                               kdebug("share copy");
+                               vma->vm_flags |= VM_MAPPED_COPY;
+                       } else {
+                               kdebug("share mmap");
+                               ret = do_mmap_shared_file(vma);
+                               if (ret < 0) {
+                                       vma->vm_region = NULL;
+                                       vma->vm_start = 0;
+                                       vma->vm_end = 0;
+                                       atomic_dec(&pregion->vm_usage);
+                                       pregion = NULL;
+                                       goto error_just_free;
+                               }
+                       }
+                       fput(region->vm_file);
+                       kmem_cache_free(vm_region_jar, region);
+                       region = pregion;
+                       result = start;
+                       goto share;
                }
 
-       dont_share_VMAs:
-               vma = NULL;
-
                /* obtain the address at which to make a shared mapping
                 * - this is the hook for quasi-memory character devices to
                 *   tell us the location of a shared mapping
@@ -891,191 +1312,287 @@ unsigned long do_mmap_pgoff(struct file *file,
                        if (IS_ERR((void *) addr)) {
                                ret = addr;
                                if (ret != (unsigned long) -ENOSYS)
-                                       goto error;
+                                       goto error_just_free;
 
                                /* the driver refused to tell us where to site
                                 * the mapping so we'll have to attempt to copy
                                 * it */
                                ret = (unsigned long) -ENODEV;
                                if (!(capabilities & BDI_CAP_MAP_COPY))
-                                       goto error;
+                                       goto error_just_free;
 
                                capabilities &= ~BDI_CAP_MAP_DIRECT;
+                       } else {
+                               vma->vm_start = region->vm_start = addr;
+                               vma->vm_end = region->vm_end = addr + len;
                        }
                }
        }
 
-       /* we're going to need a VMA struct as well */
-       vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
-       if (!vma)
-               goto error_getting_vma;
-
-       INIT_LIST_HEAD(&vma->anon_vma_node);
-       atomic_set(&vma->vm_usage, 1);
-       if (file)
-               get_file(file);
-       vma->vm_file    = file;
-       vma->vm_flags   = vm_flags;
-       vma->vm_start   = addr;
-       vma->vm_end     = addr + len;
-       vma->vm_pgoff   = pgoff;
-
-       vml->vma = vma;
+       vma->vm_region = region;
 
        /* set up the mapping */
        if (file && vma->vm_flags & VM_SHARED)
-               ret = do_mmap_shared_file(vma, len);
+               ret = do_mmap_shared_file(vma);
        else
-               ret = do_mmap_private(vma, len);
+               ret = do_mmap_private(vma, region, len);
        if (ret < 0)
-               goto error;
+               goto error_put_region;
 
-       /* okay... we have a mapping; now we have to register it */
-       result = (void *) vma->vm_start;
-
-       if (vma->vm_flags & VM_MAPPED_COPY) {
-               realalloc += kobjsize(result);
-               askedalloc += len;
-       }
+       add_nommu_region(region);
 
-       realalloc += kobjsize(vma);
-       askedalloc += sizeof(*vma);
+       /* okay... we have a mapping; now we have to register it */
+       result = vma->vm_start;
 
        current->mm->total_vm += len >> PAGE_SHIFT;
 
-       add_nommu_vma(vma);
-
- shared:
-       realalloc += kobjsize(vml);
-       askedalloc += sizeof(*vml);
-
-       add_vma_to_mm(current->mm, vml);
+share:
+       add_vma_to_mm(current->mm, vma);
 
-       up_write(&nommu_vma_sem);
+       up_write(&nommu_region_sem);
 
        if (prot & PROT_EXEC)
-               flush_icache_range((unsigned long) result,
-                                  (unsigned long) result + len);
+               flush_icache_range(result, result + len);
 
-#ifdef DEBUG
-       printk("do_mmap:\n");
-       show_process_blocks();
-#endif
-
-       return (unsigned long) result;
+       kleave(" = %lx", result);
+       return result;
 
- error:
-       up_write(&nommu_vma_sem);
-       kfree(vml);
+error_put_region:
+       __put_nommu_region(region);
        if (vma) {
-               if (vma->vm_file)
+               if (vma->vm_file) {
                        fput(vma->vm_file);
-               kfree(vma);
+                       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;
 
- sharing_violation:
-       up_write(&nommu_vma_sem);
-       printk("Attempt to share mismatched mappings\n");
-       kfree(vml);
-       return -EINVAL;
+error_just_free:
+       up_write(&nommu_region_sem);
+error:
+       fput(region->vm_file);
+       kmem_cache_free(vm_region_jar, region);
+       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", ret);
+       return ret;
 
- error_getting_vma:
-       up_write(&nommu_vma_sem);
-       kfree(vml);
-       printk("Allocation of vma for %lu byte allocation from process %d failed\n",
+sharing_violation:
+       up_write(&nommu_region_sem);
+       printk(KERN_WARNING "Attempt to share mismatched mappings\n");
+       ret = -EINVAL;
+       goto error;
+
+error_getting_vma:
+       kmem_cache_free(vm_region_jar, region);
+       printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
+              " from process %d failed\n",
               len, current->pid);
        show_free_areas();
        return -ENOMEM;
 
- error_getting_vml:
-       printk("Allocation of vml for %lu byte allocation from process %d failed\n",
+error_getting_region:
+       printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
+              " from process %d failed\n",
               len, current->pid);
        show_free_areas();
        return -ENOMEM;
 }
+EXPORT_SYMBOL(do_mmap_pgoff);
 
 /*
- * handle mapping disposal for uClinux
+ * split a vma into two pieces at address 'addr', a new vma is allocated either
+ * for the first part or the tail.
  */
-static void put_vma(struct vm_area_struct *vma)
+int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
+             unsigned long addr, int new_below)
 {
-       if (vma) {
-               down_write(&nommu_vma_sem);
+       struct vm_area_struct *new;
+       struct vm_region *region;
+       unsigned long npages;
 
-               if (atomic_dec_and_test(&vma->vm_usage)) {
-                       delete_nommu_vma(vma);
+       kenter("");
 
-                       if (vma->vm_ops && vma->vm_ops->close)
-                               vma->vm_ops->close(vma);
+       /* 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)
+               return -ENOMEM;
 
-                       /* IO memory and memory shared directly out of the pagecache from
-                        * ramfs/tmpfs mustn't be released here */
-                       if (vma->vm_flags & VM_MAPPED_COPY) {
-                               realalloc -= kobjsize((void *) vma->vm_start);
-                               askedalloc -= vma->vm_end - vma->vm_start;
-                               kfree((void *) vma->vm_start);
-                       }
+       if (mm->map_count >= sysctl_max_map_count)
+               return -ENOMEM;
 
-                       realalloc -= kobjsize(vma);
-                       askedalloc -= sizeof(*vma);
+       region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
+       if (!region)
+               return -ENOMEM;
 
-                       if (vma->vm_file)
-                               fput(vma->vm_file);
-                       kfree(vma);
-               }
+       new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
+       if (!new) {
+               kmem_cache_free(vm_region_jar, region);
+               return -ENOMEM;
+       }
+
+       /* most fields are the same, copy all, and then fixup */
+       *new = *vma;
+       *region = *vma->vm_region;
+       new->vm_region = region;
+
+       npages = (addr - vma->vm_start) >> PAGE_SHIFT;
 
-               up_write(&nommu_vma_sem);
+       if (new_below) {
+               region->vm_top = region->vm_end = new->vm_end = addr;
+       } else {
+               region->vm_start = new->vm_start = addr;
+               region->vm_pgoff = new->vm_pgoff += npages;
        }
+
+       if (new->vm_ops && new->vm_ops->open)
+               new->vm_ops->open(new);
+
+       delete_vma_from_mm(vma);
+       down_write(&nommu_region_sem);
+       delete_nommu_region(vma->vm_region);
+       if (new_below) {
+               vma->vm_region->vm_start = vma->vm_start = addr;
+               vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
+       } else {
+               vma->vm_region->vm_end = vma->vm_end = addr;
+               vma->vm_region->vm_top = addr;
+       }
+       add_nommu_region(vma->vm_region);
+       add_nommu_region(new->vm_region);
+       up_write(&nommu_region_sem);
+       add_vma_to_mm(mm, vma);
+       add_vma_to_mm(mm, new);
+       return 0;
 }
 
 /*
- * release a mapping
- * - under NOMMU conditions the parameters must match exactly to the mapping to
- *   be removed
+ * shrink a VMA by removing the specified chunk from either the beginning or
+ * the end
  */
-int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
+static int shrink_vma(struct mm_struct *mm,
+                     struct vm_area_struct *vma,
+                     unsigned long from, unsigned long to)
 {
-       struct vm_list_struct *vml, **parent;
-       unsigned long end = addr + len;
+       struct vm_region *region;
 
-#ifdef DEBUG
-       printk("do_munmap:\n");
-#endif
+       kenter("");
 
-       for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
-               if ((*parent)->vma->vm_start > addr)
-                       break;
-               if ((*parent)->vma->vm_start == addr &&
-                   ((len == 0) || ((*parent)->vma->vm_end == end)))
-                       goto found;
+       /* adjust the VMA's pointers, which may reposition it in the MM's tree
+        * and list */
+       delete_vma_from_mm(vma);
+       if (from > vma->vm_start)
+               vma->vm_end = from;
+       else
+               vma->vm_start = to;
+       add_vma_to_mm(mm, vma);
+
+       /* cut the backing region down to size */
+       region = vma->vm_region;
+       BUG_ON(atomic_read(&region->vm_usage) != 1);
+
+       down_write(&nommu_region_sem);
+       delete_nommu_region(region);
+       if (from > region->vm_start) {
+               to = region->vm_top;
+               region->vm_top = region->vm_end = from;
+       } else {
+               region->vm_start = to;
        }
+       add_nommu_region(region);
+       up_write(&nommu_region_sem);
 
-       printk("munmap of non-mmaped memory by process %d (%s): %p\n",
-              current->pid, current->comm, (void *) addr);
-       return -EINVAL;
+       free_page_series(from, to);
+       return 0;
+}
 
- found:
-       vml = *parent;
+/*
+ * release a mapping
+ * - under NOMMU conditions the chunk to be unmapped must be backed by a single
+ *   VMA, though it need not cover the whole VMA
+ */
+int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
+{
+       struct vm_area_struct *vma;
+       struct rb_node *rb;
+       unsigned long end = start + len;
+       int ret;
 
-       put_vma(vml->vma);
+       kenter(",%lx,%zx", start, len);
 
-       *parent = vml->next;
-       realalloc -= kobjsize(vml);
-       askedalloc -= sizeof(*vml);
-       kfree(vml);
+       if (len == 0)
+               return -EINVAL;
 
-       update_hiwater_vm(mm);
-       mm->total_vm -= len >> PAGE_SHIFT;
+       /* find the first potentially overlapping VMA */
+       vma = find_vma(mm, start);
+       if (!vma) {
+               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;
+       }
 
-#ifdef DEBUG
-       show_process_blocks();
-#endif
+       /* we're allowed to split an anonymous VMA but not a file-backed one */
+       if (vma->vm_file) {
+               do {
+                       if (start > vma->vm_start) {
+                               kleave(" = -EINVAL [miss]");
+                               return -EINVAL;
+                       }
+                       if (end == vma->vm_end)
+                               goto erase_whole_vma;
+                       rb = rb_next(&vma->vm_rb);
+                       vma = rb_entry(rb, struct vm_area_struct, vm_rb);
+               } while (rb);
+               kleave(" = -EINVAL [split file]");
+               return -EINVAL;
+       } else {
+               /* the chunk must be a subset of the VMA found */
+               if (start == vma->vm_start && end == vma->vm_end)
+                       goto erase_whole_vma;
+               if (start < vma->vm_start || end > vma->vm_end) {
+                       kleave(" = -EINVAL [superset]");
+                       return -EINVAL;
+               }
+               if (start & ~PAGE_MASK) {
+                       kleave(" = -EINVAL [unaligned start]");
+                       return -EINVAL;
+               }
+               if (end != vma->vm_end && end & ~PAGE_MASK) {
+                       kleave(" = -EINVAL [unaligned split]");
+                       return -EINVAL;
+               }
+               if (start != vma->vm_start && end != vma->vm_end) {
+                       ret = split_vma(mm, vma, start, 1);
+                       if (ret < 0) {
+                               kleave(" = %d [split]", ret);
+                               return ret;
+                       }
+               }
+               return shrink_vma(mm, vma, start, end);
+       }
 
+erase_whole_vma:
+       delete_vma_from_mm(vma);
+       delete_vma(mm, vma);
+       kleave(" = 0");
        return 0;
 }
+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;
@@ -1087,32 +1604,26 @@ asmlinkage long sys_munmap(unsigned long addr, size_t len)
 }
 
 /*
- * Release all mappings
+ * release all the mappings made in a process's VM space
  */
-void exit_mmap(struct mm_struct * mm)
+void exit_mmap(struct mm_struct *mm)
 {
-       struct vm_list_struct *tmp;
-
-       if (mm) {
-#ifdef DEBUG
-               printk("Exit_mmap:\n");
-#endif
+       struct vm_area_struct *vma;
 
-               mm->total_vm = 0;
+       if (!mm)
+               return;
 
-               while ((tmp = mm->context.vmlist)) {
-                       mm->context.vmlist = tmp->next;
-                       put_vma(tmp->vma);
+       kenter("");
 
-                       realalloc -= kobjsize(tmp);
-                       askedalloc -= sizeof(*tmp);
-                       kfree(tmp);
-               }
+       mm->total_vm = 0;
 
-#ifdef DEBUG
-               show_process_blocks();
-#endif
+       while ((vma = mm->mmap)) {
+               mm->mmap = vma->vm_next;
+               delete_vma_from_mm(vma);
+               delete_vma(mm, vma);
        }
+
+       kleave("");
 }
 
 unsigned long do_brk(unsigned long addr, unsigned long len)
@@ -1125,8 +1636,8 @@ unsigned long do_brk(unsigned long addr, unsigned long len)
  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
  *
  * under NOMMU conditions, we only permit changing a mapping's size, and only
- * as long as it stays within the hole allocated by the kmalloc() call in
- * do_mmap_pgoff() and the block is not shareable
+ * as long as it stays within the region allocated by do_mmap_private() and the
+ * block is not shareable
  *
  * MREMAP_FIXED is not supported under NOMMU conditions
  */
@@ -1137,13 +1648,16 @@ unsigned long do_mremap(unsigned long addr,
        struct vm_area_struct *vma;
 
        /* insanity checks first */
-       if (new_len == 0)
+       if (old_len == 0 || new_len == 0)
                return (unsigned long) -EINVAL;
 
+       if (addr & ~PAGE_MASK)
+               return -EINVAL;
+
        if (flags & MREMAP_FIXED && new_addr != addr)
                return (unsigned long) -EINVAL;
 
-       vma = find_vma_exact(current->mm, addr);
+       vma = find_vma_exact(current->mm, addr, old_len);
        if (!vma)
                return (unsigned long) -EINVAL;
 
@@ -1153,21 +1667,18 @@ unsigned long do_mremap(unsigned long addr,
        if (vma->vm_flags & VM_MAYSHARE)
                return (unsigned long) -EPERM;
 
-       if (new_len > kobjsize((void *) addr))
+       if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
                return (unsigned long) -ENOMEM;
 
        /* all checks complete - do it */
        vma->vm_end = vma->vm_start + new_len;
-
-       askedalloc -= old_len;
-       askedalloc += new_len;
-
        return vma->vm_start;
 }
+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;
 
@@ -1191,6 +1702,21 @@ int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
 }
 EXPORT_SYMBOL(remap_pfn_range);
 
+int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
+                       unsigned long pgoff)
+{
+       unsigned int size = vma->vm_end - vma->vm_start;
+
+       if (!(vma->vm_flags & VM_USERMAP))
+               return -EINVAL;
+
+       vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
+       vma->vm_end = vma->vm_start + size;
+
+       return 0;
+}
+EXPORT_SYMBOL(remap_vmalloc_range);
+
 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 {
 }
@@ -1231,7 +1757,6 @@ unsigned long get_unmapped_area(struct file *file, unsigned long addr,
 
        return get_area(file, addr, len, pgoff, flags);
 }
-
 EXPORT_SYMBOL(get_unmapped_area);
 
 /*
@@ -1250,7 +1775,7 @@ EXPORT_SYMBOL(get_unmapped_area);
  * Note this is a helper function intended to be used by LSMs which
  * wish to use this logic.
  */
-int __vm_enough_memory(long pages, int cap_sys_admin)
+int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 {
        unsigned long free, allowed;
 
@@ -1322,14 +1847,12 @@ int __vm_enough_memory(long pages, int cap_sys_admin)
 
        /* Don't let a single process grow too big:
           leave 3% of the size of this process for other processes */
-       allowed -= current->mm->total_vm / 32;
+       if (mm)
+               allowed -= mm->total_vm / 32;
 
-       /*
-        * cast `allowed' as a signed long because vm_committed_space
-        * sometimes has a negative value
-        */
-       if (atomic_read(&vm_committed_space) < (long)allowed)
+       if (percpu_counter_read_positive(&vm_committed_as) < allowed)
                return 0;
+
 error:
        vm_unacct_memory(pages);
 
@@ -1341,11 +1864,12 @@ int in_gate_area_no_task(unsigned long addr)
        return 0;
 }
 
-struct page *filemap_fault(struct vm_area_struct *vma, struct fault_data *fdata)
+int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
        BUG();
-       return NULL;
+       return 0;
 }
+EXPORT_SYMBOL(filemap_fault);
 
 /*
  * Access another process' address space.