mm: move mm_populate()-related code to mm/gup.c
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Tue, 14 Apr 2015 22:44:45 +0000 (15:44 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 14 Apr 2015 23:49:00 +0000 (16:49 -0700)
It's odd that we have populate_vma_page_range() and __mm_populate() in
mm/mlock.c.  It's implementation of generic memory population and mlocking
is one of possible side effect, if VM_LOCKED is set.

__get_user_pages() is core of the implementation.  Let's move the code
into mm/gup.c.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/gup.c
mm/mlock.c

index 1b114ba..ca7b607 100644 (file)
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -818,6 +818,124 @@ long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 }
 EXPORT_SYMBOL(get_user_pages);
 
+/**
+ * populate_vma_page_range() -  populate a range of pages in the vma.
+ * @vma:   target vma
+ * @start: start address
+ * @end:   end address
+ * @nonblocking:
+ *
+ * This takes care of mlocking the pages too if VM_LOCKED is set.
+ *
+ * return 0 on success, negative error code on error.
+ *
+ * vma->vm_mm->mmap_sem must be held.
+ *
+ * If @nonblocking is NULL, it may be held for read or write and will
+ * be unperturbed.
+ *
+ * If @nonblocking is non-NULL, it must held for read only and may be
+ * released.  If it's released, *@nonblocking will be set to 0.
+ */
+long populate_vma_page_range(struct vm_area_struct *vma,
+               unsigned long start, unsigned long end, int *nonblocking)
+{
+       struct mm_struct *mm = vma->vm_mm;
+       unsigned long nr_pages = (end - start) / PAGE_SIZE;
+       int gup_flags;
+
+       VM_BUG_ON(start & ~PAGE_MASK);
+       VM_BUG_ON(end   & ~PAGE_MASK);
+       VM_BUG_ON_VMA(start < vma->vm_start, vma);
+       VM_BUG_ON_VMA(end   > vma->vm_end, vma);
+       VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
+
+       gup_flags = FOLL_TOUCH | FOLL_POPULATE;
+       /*
+        * We want to touch writable mappings with a write fault in order
+        * to break COW, except for shared mappings because these don't COW
+        * and we would not want to dirty them for nothing.
+        */
+       if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
+               gup_flags |= FOLL_WRITE;
+
+       /*
+        * We want mlock to succeed for regions that have any permissions
+        * other than PROT_NONE.
+        */
+       if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
+               gup_flags |= FOLL_FORCE;
+
+       /*
+        * We made sure addr is within a VMA, so the following will
+        * not result in a stack expansion that recurses back here.
+        */
+       return __get_user_pages(current, mm, start, nr_pages, gup_flags,
+                               NULL, NULL, nonblocking);
+}
+
+/*
+ * __mm_populate - populate and/or mlock pages within a range of address space.
+ *
+ * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
+ * flags. VMAs must be already marked with the desired vm_flags, and
+ * mmap_sem must not be held.
+ */
+int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
+{
+       struct mm_struct *mm = current->mm;
+       unsigned long end, nstart, nend;
+       struct vm_area_struct *vma = NULL;
+       int locked = 0;
+       long ret = 0;
+
+       VM_BUG_ON(start & ~PAGE_MASK);
+       VM_BUG_ON(len != PAGE_ALIGN(len));
+       end = start + len;
+
+       for (nstart = start; nstart < end; nstart = nend) {
+               /*
+                * We want to fault in pages for [nstart; end) address range.
+                * Find first corresponding VMA.
+                */
+               if (!locked) {
+                       locked = 1;
+                       down_read(&mm->mmap_sem);
+                       vma = find_vma(mm, nstart);
+               } else if (nstart >= vma->vm_end)
+                       vma = vma->vm_next;
+               if (!vma || vma->vm_start >= end)
+                       break;
+               /*
+                * Set [nstart; nend) to intersection of desired address
+                * range with the first VMA. Also, skip undesirable VMA types.
+                */
+               nend = min(end, vma->vm_end);
+               if (vma->vm_flags & (VM_IO | VM_PFNMAP))
+                       continue;
+               if (nstart < vma->vm_start)
+                       nstart = vma->vm_start;
+               /*
+                * Now fault in a range of pages. populate_vma_page_range()
+                * double checks the vma flags, so that it won't mlock pages
+                * if the vma was already munlocked.
+                */
+               ret = populate_vma_page_range(vma, nstart, nend, &locked);
+               if (ret < 0) {
+                       if (ignore_errors) {
+                               ret = 0;
+                               continue;       /* continue at next VMA */
+                       }
+                       break;
+               }
+               nend = nstart + ret * PAGE_SIZE;
+               ret = 0;
+       }
+       if (locked)
+               up_read(&mm->mmap_sem);
+       return ret;     /* 0 or negative error code */
+}
+
 /**
  * get_dump_page() - pin user page in memory while writing it to core dump
  * @addr: user address
index 0214263..6fd2cf1 100644 (file)
@@ -205,62 +205,6 @@ out:
        return nr_pages - 1;
 }
 
-/**
- * populate_vma_page_range() -  populate a range of pages in the vma.
- * @vma:   target vma
- * @start: start address
- * @end:   end address
- * @nonblocking:
- *
- * This takes care of mlocking the pages too if VM_LOCKED is set.
- *
- * return 0 on success, negative error code on error.
- *
- * vma->vm_mm->mmap_sem must be held.
- *
- * If @nonblocking is NULL, it may be held for read or write and will
- * be unperturbed.
- *
- * If @nonblocking is non-NULL, it must held for read only and may be
- * released.  If it's released, *@nonblocking will be set to 0.
- */
-long populate_vma_page_range(struct vm_area_struct *vma,
-               unsigned long start, unsigned long end, int *nonblocking)
-{
-       struct mm_struct *mm = vma->vm_mm;
-       unsigned long nr_pages = (end - start) / PAGE_SIZE;
-       int gup_flags;
-
-       VM_BUG_ON(start & ~PAGE_MASK);
-       VM_BUG_ON(end   & ~PAGE_MASK);
-       VM_BUG_ON_VMA(start < vma->vm_start, vma);
-       VM_BUG_ON_VMA(end   > vma->vm_end, vma);
-       VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
-
-       gup_flags = FOLL_TOUCH | FOLL_POPULATE;
-       /*
-        * We want to touch writable mappings with a write fault in order
-        * to break COW, except for shared mappings because these don't COW
-        * and we would not want to dirty them for nothing.
-        */
-       if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
-               gup_flags |= FOLL_WRITE;
-
-       /*
-        * We want mlock to succeed for regions that have any permissions
-        * other than PROT_NONE.
-        */
-       if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
-               gup_flags |= FOLL_FORCE;
-
-       /*
-        * We made sure addr is within a VMA, so the following will
-        * not result in a stack expansion that recurses back here.
-        */
-       return __get_user_pages(current, mm, start, nr_pages, gup_flags,
-                               NULL, NULL, nonblocking);
-}
-
 /*
  * convert get_user_pages() return value to posix mlock() error
  */
@@ -660,68 +604,6 @@ static int do_mlock(unsigned long start, size_t len, int on)
        return error;
 }
 
-/*
- * __mm_populate - populate and/or mlock pages within a range of address space.
- *
- * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
- * flags. VMAs must be already marked with the desired vm_flags, and
- * mmap_sem must not be held.
- */
-int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
-{
-       struct mm_struct *mm = current->mm;
-       unsigned long end, nstart, nend;
-       struct vm_area_struct *vma = NULL;
-       int locked = 0;
-       long ret = 0;
-
-       VM_BUG_ON(start & ~PAGE_MASK);
-       VM_BUG_ON(len != PAGE_ALIGN(len));
-       end = start + len;
-
-       for (nstart = start; nstart < end; nstart = nend) {
-               /*
-                * We want to fault in pages for [nstart; end) address range.
-                * Find first corresponding VMA.
-                */
-               if (!locked) {
-                       locked = 1;
-                       down_read(&mm->mmap_sem);
-                       vma = find_vma(mm, nstart);
-               } else if (nstart >= vma->vm_end)
-                       vma = vma->vm_next;
-               if (!vma || vma->vm_start >= end)
-                       break;
-               /*
-                * Set [nstart; nend) to intersection of desired address
-                * range with the first VMA. Also, skip undesirable VMA types.
-                */
-               nend = min(end, vma->vm_end);
-               if (vma->vm_flags & (VM_IO | VM_PFNMAP))
-                       continue;
-               if (nstart < vma->vm_start)
-                       nstart = vma->vm_start;
-               /*
-                * Now fault in a range of pages. populate_vma_page_range()
-                * double checks the vma flags, so that it won't mlock pages
-                * if the vma was already munlocked.
-                */
-               ret = populate_vma_page_range(vma, nstart, nend, &locked);
-               if (ret < 0) {
-                       if (ignore_errors) {
-                               ret = 0;
-                               continue;       /* continue at next VMA */
-                       }
-                       break;
-               }
-               nend = nstart + ret * PAGE_SIZE;
-               ret = 0;
-       }
-       if (locked)
-               up_read(&mm->mmap_sem);
-       return ret;     /* 0 or negative error code */
-}
-
 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
 {
        unsigned long locked;