[v5] vfs: fix page locking deadlocks when deduping files

Commit Message

Darrick J. Wong Aug. 16, 2019, 4:13 p.m. UTC

From: Darrick J. Wong <[e-mail]>

When dedupe wants to use the page cache to compare parts of two files
for dedupe, we must be very careful to handle locking correctly.  The
current code doesn't do this.  It must lock and unlock the page only
once if the two pages are the same, since the overlapping range check
doesn't catch this when blocksize < pagesize.  If the pages are distinct
but from the same file, we must observe page locking order and lock them
in order of increasing offset to avoid clashing with writeback locking.

Fixes: 876bec6f9bbfcb3 ("vfs: refactor clone/dedupe_file_range common functions")
Signed-off-by: Darrick J. Wong <[e-mail]>
Reviewed-by: Bill O'Donnell <[e-mail]>
Reviewed-by: Matthew Wilcox (Oracle) <[e-mail]>
---
v5: recheck pageuptodate
v4: drop the unnecessary page offset checks
v3: revalidate page after locking it
v2: provide an unlock helper
---
 fs/read_write.c |   49 +++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 41 insertions(+), 8 deletions(-)

Patch

diff --git a/fs/read_write.c b/fs/read_write.c
index 1f5088dec566..5bbf587f5bc1 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c
@@ -1811,10 +1811,7 @@  static int generic_remap_check_len(struct inode *inode_in,
 	return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
 }
 
-/*
- * Read a page's worth of file data into the page cache.  Return the page
- * locked.
- */
+/* Read a page's worth of file data into the page cache. */
 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
 {
 	struct page *page;
@@ -1826,10 +1823,32 @@  static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
 		put_page(page);
 		return ERR_PTR(-EIO);
 	}
-	lock_page(page);
 	return page;
 }
 
+/*
+ * Lock two pages, ensuring that we lock in offset order if the pages are from
+ * the same file.
+ */
+static void vfs_lock_two_pages(struct page *page1, struct page *page2)
+{
+	/* Always lock in order of increasing index. */
+	if (page1->index > page2->index)
+		swap(page1, page2);
+
+	lock_page(page1);
+	if (page1 != page2)
+		lock_page(page2);
+}
+
+/* Unlock two pages, being careful not to unlock the same page twice. */
+static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
+{
+	unlock_page(page1);
+	if (page1 != page2)
+		unlock_page(page2);
+}
+
 /*
  * Compare extents of two files to see if they are the same.
  * Caller must have locked both inodes to prevent write races.
@@ -1867,10 +1886,24 @@  static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
 		dest_page = vfs_dedupe_get_page(dest, destoff);
 		if (IS_ERR(dest_page)) {
 			error = PTR_ERR(dest_page);
-			unlock_page(src_page);
 			put_page(src_page);
 			goto out_error;
 		}
+
+		vfs_lock_two_pages(src_page, dest_page);
+
+		/*
+		 * Now that we've locked both pages, make sure they're still
+		 * mapped to the file data we're interested in.  If not,
+		 * someone is invalidating pages on us and we lose.
+		 */
+		if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
+		    src_page->mapping != src->i_mapping ||
+		    dest_page->mapping != dest->i_mapping) {
+			same = false;
+			goto unlock;
+		}
+
 		src_addr = kmap_atomic(src_page);
 		dest_addr = kmap_atomic(dest_page);
 
@@ -1882,8 +1915,8 @@  static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
 
 		kunmap_atomic(dest_addr);
 		kunmap_atomic(src_addr);
-		unlock_page(dest_page);
-		unlock_page(src_page);
+unlock:
+		vfs_unlock_two_pages(src_page, dest_page);
 		put_page(dest_page);
 		put_page(src_page);