userfaultfd: avoid mmap_sem read recursion in mcopy_atomic
authorAndrea Arcangeli <aarcange@redhat.com>
Fri, 4 Sep 2015 22:47:08 +0000 (15:47 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 4 Sep 2015 23:54:41 +0000 (16:54 -0700)
If the rwsem starves writers it wasn't strictly a bug but lockdep
doesn't like it and this avoids depending on lowlevel implementation
details of the lock.

[akpm@linux-foundation.org: delete weird BUILD_BUG_ON()]
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Sanidhya Kashyap <sanidhya.gatech@gmail.com>
Cc: zhang.zhanghailiang@huawei.com
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: Peter Feiner <pfeiner@google.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Huangpeng (Peter)" <peter.huangpeng@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/userfaultfd.c

index c54c761609fc9c48b1a491e4a2ad66dbe10e3a86..77fee9325a5727825b9adcdac156b4e22247b6cf 100644 (file)
@@ -21,26 +21,39 @@ static int mcopy_atomic_pte(struct mm_struct *dst_mm,
                            pmd_t *dst_pmd,
                            struct vm_area_struct *dst_vma,
                            unsigned long dst_addr,
-                           unsigned long src_addr)
+                           unsigned long src_addr,
+                           struct page **pagep)
 {
        struct mem_cgroup *memcg;
        pte_t _dst_pte, *dst_pte;
        spinlock_t *ptl;
-       struct page *page;
        void *page_kaddr;
        int ret;
+       struct page *page;
 
-       ret = -ENOMEM;
-       page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, dst_vma, dst_addr);
-       if (!page)
-               goto out;
-
-       page_kaddr = kmap(page);
-       ret = -EFAULT;
-       if (copy_from_user(page_kaddr, (const void __user *) src_addr,
-                          PAGE_SIZE))
-               goto out_kunmap_release;
-       kunmap(page);
+       if (!*pagep) {
+               ret = -ENOMEM;
+               page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, dst_vma, dst_addr);
+               if (!page)
+                       goto out;
+
+               page_kaddr = kmap_atomic(page);
+               ret = copy_from_user(page_kaddr,
+                                    (const void __user *) src_addr,
+                                    PAGE_SIZE);
+               kunmap_atomic(page_kaddr);
+
+               /* fallback to copy_from_user outside mmap_sem */
+               if (unlikely(ret)) {
+                       ret = -EFAULT;
+                       *pagep = page;
+                       /* don't free the page */
+                       goto out;
+               }
+       } else {
+               page = *pagep;
+               *pagep = NULL;
+       }
 
        /*
         * The memory barrier inside __SetPageUptodate makes sure that
@@ -82,9 +95,6 @@ out_release_uncharge_unlock:
 out_release:
        page_cache_release(page);
        goto out;
-out_kunmap_release:
-       kunmap(page);
-       goto out_release;
 }
 
 static int mfill_zeropage_pte(struct mm_struct *dst_mm,
@@ -139,7 +149,8 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
        ssize_t err;
        pmd_t *dst_pmd;
        unsigned long src_addr, dst_addr;
-       long copied = 0;
+       long copied;
+       struct page *page;
 
        /*
         * Sanitize the command parameters:
@@ -151,6 +162,11 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
        BUG_ON(src_start + len <= src_start);
        BUG_ON(dst_start + len <= dst_start);
 
+       src_addr = src_start;
+       dst_addr = dst_start;
+       copied = 0;
+       page = NULL;
+retry:
        down_read(&dst_mm->mmap_sem);
 
        /*
@@ -160,10 +176,10 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
        err = -EINVAL;
        dst_vma = find_vma(dst_mm, dst_start);
        if (!dst_vma || (dst_vma->vm_flags & VM_SHARED))
-               goto out;
+               goto out_unlock;
        if (dst_start < dst_vma->vm_start ||
            dst_start + len > dst_vma->vm_end)
-               goto out;
+               goto out_unlock;
 
        /*
         * Be strict and only allow __mcopy_atomic on userfaultfd
@@ -175,14 +191,14 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
         * belonging to the userfaultfd and not syscalls.
         */
        if (!dst_vma->vm_userfaultfd_ctx.ctx)
-               goto out;
+               goto out_unlock;
 
        /*
         * FIXME: only allow copying on anonymous vmas, tmpfs should
         * be added.
         */
        if (dst_vma->vm_ops)
-               goto out;
+               goto out_unlock;
 
        /*
         * Ensure the dst_vma has a anon_vma or this page
@@ -191,12 +207,13 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
         */
        err = -ENOMEM;
        if (unlikely(anon_vma_prepare(dst_vma)))
-               goto out;
+               goto out_unlock;
 
-       for (src_addr = src_start, dst_addr = dst_start;
-            src_addr < src_start + len; ) {
+       while (src_addr < src_start + len) {
                pmd_t dst_pmdval;
+
                BUG_ON(dst_addr >= dst_start + len);
+
                dst_pmd = mm_alloc_pmd(dst_mm, dst_addr);
                if (unlikely(!dst_pmd)) {
                        err = -ENOMEM;
@@ -229,13 +246,32 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
 
                if (!zeropage)
                        err = mcopy_atomic_pte(dst_mm, dst_pmd, dst_vma,
-                                              dst_addr, src_addr);
+                                              dst_addr, src_addr, &page);
                else
                        err = mfill_zeropage_pte(dst_mm, dst_pmd, dst_vma,
                                                 dst_addr);
 
                cond_resched();
 
+               if (unlikely(err == -EFAULT)) {
+                       void *page_kaddr;
+
+                       up_read(&dst_mm->mmap_sem);
+                       BUG_ON(!page);
+
+                       page_kaddr = kmap(page);
+                       err = copy_from_user(page_kaddr,
+                                            (const void __user *) src_addr,
+                                            PAGE_SIZE);
+                       kunmap(page);
+                       if (unlikely(err)) {
+                               err = -EFAULT;
+                               goto out;
+                       }
+                       goto retry;
+               } else
+                       BUG_ON(page);
+
                if (!err) {
                        dst_addr += PAGE_SIZE;
                        src_addr += PAGE_SIZE;
@@ -248,8 +284,11 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
                        break;
        }
 
-out:
+out_unlock:
        up_read(&dst_mm->mmap_sem);
+out:
+       if (page)
+               page_cache_release(page);
        BUG_ON(copied < 0);
        BUG_ON(err > 0);
        BUG_ON(!copied && !err);