hugetlbfs: avoid taking i_mutex from hugetlbfs_read()
Aneesh Kumar K.V [Wed, 21 Mar 2012 23:34:08 +0000 (16:34 -0700)]
Taking i_mutex in hugetlbfs_read() can result in deadlock with mmap as
explained below

 Thread A:
  read() on hugetlbfs
   hugetlbfs_read() called
    i_mutex grabbed
     hugetlbfs_read_actor() called
      __copy_to_user() called
       page fault is triggered
 Thread B, sharing address space with A:
  mmap() the same file
   ->mmap_sem is grabbed on task_B->mm->mmap_sem
    hugetlbfs_file_mmap() is called
     attempt to grab ->i_mutex and block waiting for A to give it up
 Thread A:
  pagefault handled blocked on attempt to grab task_A->mm->mmap_sem,
 which happens to be the same thing as task_B->mm->mmap_sem.  Block waiting
 for B to give it up.

AFAIU the i_mutex locking was added to hugetlbfs_read() as per
http://lkml.indiana.edu/hypermail/linux/kernel/0707.2/3066.html to take
care of the race between truncate and read.  This patch fixes this by
looking at page->mapping under lock_page() (find_lock_page()) to ensure
that the inode didn't get truncated in the range during a parallel read.

Ideally we can extend the patch to make sure we don't increase i_size in
mmap.  But that will break userspace, because applications will now have
to use truncate(2) to increase i_size in hugetlbfs.

Based on the original patch from Hillf Danton.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Hillf Danton <dhillf@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@kernel.org> [everything after 2007 :)]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/hugetlbfs/inode.c

index b7bc786..19654cf 100644 (file)
@@ -245,17 +245,10 @@ static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
        loff_t isize;
        ssize_t retval = 0;
 
-       mutex_lock(&inode->i_mutex);
-
        /* validate length */
        if (len == 0)
                goto out;
 
-       isize = i_size_read(inode);
-       if (!isize)
-               goto out;
-
-       end_index = (isize - 1) >> huge_page_shift(h);
        for (;;) {
                struct page *page;
                unsigned long nr, ret;
@@ -263,18 +256,21 @@ static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 
                /* nr is the maximum number of bytes to copy from this page */
                nr = huge_page_size(h);
+               isize = i_size_read(inode);
+               if (!isize)
+                       goto out;
+               end_index = (isize - 1) >> huge_page_shift(h);
                if (index >= end_index) {
                        if (index > end_index)
                                goto out;
                        nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
-                       if (nr <= offset) {
+                       if (nr <= offset)
                                goto out;
-                       }
                }
                nr = nr - offset;
 
                /* Find the page */
-               page = find_get_page(mapping, index);
+               page = find_lock_page(mapping, index);
                if (unlikely(page == NULL)) {
                        /*
                         * We have a HOLE, zero out the user-buffer for the
@@ -286,17 +282,18 @@ static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
                        else
                                ra = 0;
                } else {
+                       unlock_page(page);
+
                        /*
                         * We have the page, copy it to user space buffer.
                         */
                        ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
                        ret = ra;
+                       page_cache_release(page);
                }
                if (ra < 0) {
                        if (retval == 0)
                                retval = ra;
-                       if (page)
-                               page_cache_release(page);
                        goto out;
                }
 
@@ -306,16 +303,12 @@ static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
                index += offset >> huge_page_shift(h);
                offset &= ~huge_page_mask(h);
 
-               if (page)
-                       page_cache_release(page);
-
                /* short read or no more work */
                if ((ret != nr) || (len == 0))
                        break;
        }
 out:
        *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
-       mutex_unlock(&inode->i_mutex);
        return retval;
 }