nfsd41: Backchannel: Implement cb_recall over NFSv4.1
[safe/jmp/linux-2.6] / mm / mmap.c
index 3b3ed0b..34579b2 100644 (file)
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -28,6 +28,7 @@
 #include <linux/mempolicy.h>
 #include <linux/rmap.h>
 #include <linux/mmu_notifier.h>
+#include <linux/perf_counter.h>
 
 #include <asm/uaccess.h>
 #include <asm/cacheflush.h>
@@ -85,7 +86,10 @@ EXPORT_SYMBOL(vm_get_page_prot);
 int sysctl_overcommit_memory = OVERCOMMIT_GUESS;  /* heuristic overcommit */
 int sysctl_overcommit_ratio = 50;      /* default is 50% */
 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
-atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
+struct percpu_counter vm_committed_as;
+
+/* amount of vm to protect from userspace access */
+unsigned long mmap_min_addr = CONFIG_DEFAULT_MMAP_MIN_ADDR;
 
 /*
  * Check that a process has enough memory to allocate a new virtual
@@ -179,11 +183,7 @@ 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);
@@ -919,7 +919,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
        struct inode *inode;
        unsigned int vm_flags;
        int error;
-       int accountable = 1;
        unsigned long reqprot = prot;
 
        /*
@@ -1020,8 +1019,6 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
                                        return -EPERM;
                                vm_flags &= ~VM_MAYEXEC;
                        }
-                       if (is_file_hugepages(file))
-                               accountable = 0;
 
                        if (!file->f_op || !file->f_op->mmap)
                                return -ENODEV;
@@ -1057,8 +1054,7 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
        if (error)
                return error;
 
-       return mmap_region(file, addr, len, flags, vm_flags, pgoff,
-                          accountable);
+       return mmap_region(file, addr, len, flags, vm_flags, pgoff);
 }
 EXPORT_SYMBOL(do_mmap_pgoff);
 
@@ -1096,17 +1092,23 @@ int vma_wants_writenotify(struct vm_area_struct *vma)
 
 /*
  * We account for memory if it's a private writeable mapping,
- * and VM_NORESERVE wasn't set.
+ * not hugepages and VM_NORESERVE wasn't set.
  */
-static inline int accountable_mapping(unsigned int vm_flags)
+static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
 {
+       /*
+        * hugetlb has its own accounting separate from the core VM
+        * VM_HUGETLB may not be set yet so we cannot check for that flag.
+        */
+       if (file && is_file_hugepages(file))
+               return 0;
+
        return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
 }
 
 unsigned long mmap_region(struct file *file, unsigned long addr,
                          unsigned long len, unsigned long flags,
-                         unsigned int vm_flags, unsigned long pgoff,
-                         int accountable)
+                         unsigned int vm_flags, unsigned long pgoff)
 {
        struct mm_struct *mm = current->mm;
        struct vm_area_struct *vma, *prev;
@@ -1132,18 +1134,22 @@ munmap_back:
 
        /*
         * Set 'VM_NORESERVE' if we should not account for the
-        * memory use of this mapping. We only honor MAP_NORESERVE
-        * if we're allowed to overcommit memory.
+        * memory use of this mapping.
         */
-       if ((flags & MAP_NORESERVE) && sysctl_overcommit_memory != OVERCOMMIT_NEVER)
-               vm_flags |= VM_NORESERVE;
-       if (!accountable)
-               vm_flags |= VM_NORESERVE;
+       if ((flags & MAP_NORESERVE)) {
+               /* We honor MAP_NORESERVE if allowed to overcommit */
+               if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
+                       vm_flags |= VM_NORESERVE;
+
+               /* hugetlb applies strict overcommit unless MAP_NORESERVE */
+               if (file && is_file_hugepages(file))
+                       vm_flags |= VM_NORESERVE;
+       }
 
        /*
         * Private writable mapping: check memory availability
         */
-       if (accountable_mapping(vm_flags)) {
+       if (accountable_mapping(file, vm_flags)) {
                charged = len >> PAGE_SHIFT;
                if (security_vm_enough_memory(charged))
                        return -ENOMEM;
@@ -1217,6 +1223,8 @@ munmap_back:
        if (correct_wcount)
                atomic_inc(&inode->i_writecount);
 out:
+       perf_counter_mmap(vma);
+
        mm->total_vm += len >> PAGE_SHIFT;
        vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
        if (vm_flags & VM_LOCKED) {
@@ -1569,7 +1577,7 @@ static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, uns
         * Overcommit..  This must be the final test, as it will
         * update security statistics.
         */
-       if (security_vm_enough_memory(grow))
+       if (security_vm_enough_memory_mm(mm, grow))
                return -ENOMEM;
 
        /* Ok, everything looks good - let it rip */
@@ -2082,12 +2090,8 @@ void exit_mmap(struct mm_struct *mm)
        unsigned long end;
 
        /* mm's last user has gone, and its about to be pulled down */
-       arch_exit_mmap(mm);
        mmu_notifier_release(mm);
 
-       if (!mm->mmap)  /* Can happen if dup_mmap() received an OOM */
-               return;
-
        if (mm->locked_vm) {
                vma = mm->mmap;
                while (vma) {
@@ -2096,7 +2100,13 @@ void exit_mmap(struct mm_struct *mm)
                        vma = vma->vm_next;
                }
        }
+
+       arch_exit_mmap(mm);
+
        vma = mm->mmap;
+       if (!vma)       /* Can happen if dup_mmap() received an OOM */
+               return;
+
        lru_add_drain();
        flush_cache_mm(mm);
        tlb = tlb_gather_mmu(mm, 1);
@@ -2301,6 +2311,8 @@ int install_special_mapping(struct mm_struct *mm,
 
        mm->total_vm += len >> PAGE_SHIFT;
 
+       perf_counter_mmap(vma);
+
        return 0;
 }
 
@@ -2473,7 +2485,8 @@ void mm_drop_all_locks(struct mm_struct *mm)
  */
 void __init mmap_init(void)
 {
-       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);
 }