Btrfs: check prepare_uptodate_page() error code earlier
authorChris Mason <clm@fb.com>
Mon, 14 Dec 2015 23:40:44 +0000 (15:40 -0800)
committerChris Mason <clm@fb.com>
Tue, 15 Dec 2015 17:09:38 +0000 (09:09 -0800)
prepare_pages() may end up calling prepare_uptodate_page() twice if our
write only spans a single page.  But if the first call returns an error,
our page will be unlocked and its not safe to call it again.

This bug goes all the way back to 2011, and it's not something commonly
hit.

While we're here, add a more explicit check for the page being truncated
away.  The bare lock_page() alone is protected only by good thoughts and
i_mutex, which we're sure to regret eventually.

Reported-by: Dave Jones <dsj@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/file.c

index 8eb1f3c1b647da1f725906b0b0a19fceab4c4cbc..870150547f5c5968efb1b1532cca8f87310af99a 100644 (file)
@@ -1291,7 +1291,8 @@ out:
  * on error we return an unlocked page and the error value
  * on success we return a locked page and 0
  */
-static int prepare_uptodate_page(struct page *page, u64 pos,
+static int prepare_uptodate_page(struct inode *inode,
+                                struct page *page, u64 pos,
                                 bool force_uptodate)
 {
        int ret = 0;
@@ -1306,6 +1307,10 @@ static int prepare_uptodate_page(struct page *page, u64 pos,
                        unlock_page(page);
                        return -EIO;
                }
+               if (page->mapping != inode->i_mapping) {
+                       unlock_page(page);
+                       return -EAGAIN;
+               }
        }
        return 0;
 }
@@ -1324,6 +1329,7 @@ static noinline int prepare_pages(struct inode *inode, struct page **pages,
        int faili;
 
        for (i = 0; i < num_pages; i++) {
+again:
                pages[i] = find_or_create_page(inode->i_mapping, index + i,
                                               mask | __GFP_WRITE);
                if (!pages[i]) {
@@ -1333,13 +1339,17 @@ static noinline int prepare_pages(struct inode *inode, struct page **pages,
                }
 
                if (i == 0)
-                       err = prepare_uptodate_page(pages[i], pos,
+                       err = prepare_uptodate_page(inode, pages[i], pos,
                                                    force_uptodate);
-               if (i == num_pages - 1)
-                       err = prepare_uptodate_page(pages[i],
+               if (!err && i == num_pages - 1)
+                       err = prepare_uptodate_page(inode, pages[i],
                                                    pos + write_bytes, false);
                if (err) {
                        page_cache_release(pages[i]);
+                       if (err == -EAGAIN) {
+                               err = 0;
+                               goto again;
+                       }
                        faili = i - 1;
                        goto fail;
                }