loop: zero fill bio instead of return -EIO for partial read

commit 8268f5a7415d914fc855a86aa2284ac819dc6b2e trying to fix the loop device partial read information leak problem. But it changed the semantics of read behavior. When we read beyond the end of the device we should get 0 bytes, which is normal behavior, we should not just return -EIO

Instead of return -EIO, zero out the bio to avoid information leak in case of partail read.

Signed-off-by Dave Young <[email protected]>

---
 drivers/block/loop.c |   24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

drivers/block/loop.c +13 -11

@@ -357,14 +357,14 @@ lo_direct_splice_actor(struct pipe_inode
 	return __splice_from_pipe(pipe, sd, lo_splice_actor);
 }
 
-static int
+static ssize_t
 do_lo_receive(struct loop_device *lo,
 	      struct bio_vec *bvec, int bsize, loff_t pos)
 {
 	struct lo_read_data cookie;
 	struct splice_desc sd;
 	struct file *file;
-	long retval;
+	ssize_t retval;
 
 	cookie.lo = lo;
 	cookie.page = bvec->bv_page;

@@ -380,26 +380,28 @@ do_lo_receive(struct loop_device *lo,
 	file = lo->lo_backing_file;
 	retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
 
-	if (retval < 0)
-		return retval;
-	if (retval != bvec->bv_len)
-		return -EIO;
-	return 0;
+	return retval;
 }
 
 static int
 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
 {
 	struct bio_vec *bvec;
-	int i, ret = 0;
+	ssize_t s;
+	int i;
 
 	bio_for_each_segment(bvec, bio, i) {
-		ret = do_lo_receive(lo, bvec, bsize, pos);
-		if (ret < 0)
+		s = do_lo_receive(lo, bvec, bsize, pos);
+		if (s < 0)
+			return s;
+
+		if (s != bvec->bv_len) {
+			zero_fill_bio(bio);
 			break;
+		}
 		pos += bvec->bv_len;
 	}
-	return ret;
+	return 0;
 }
 
 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)

Quoted 5 lines · Dave Young

commit 8268f5a7415d914fc855a86aa2284ac819dc6b2e trying to fix the loop device partial read information leak problem. But it changed the semantics of read behavior. When we read beyond the end of the device we should get 0 bytes, which is normal behavior, we should not just return -EIO

Quoted 2 lines

Instead of return -EIO, zero out the bio to avoid information leak in case of partail read.

I tested the patch with a program which patterns the loop device, truncates the backing file, and then performs preads from various offsets within the loop device, validates the return values and inspects the contents. With this patch, everything works as expected.

By the way, truncating the backing file for a loop device is insane. Why would you do that? Also, if you really want all of the data gone, you'll have to flush the contents of the buffer cache for the loop device first. It's quite the head scratcher the first time you truncate a file, wait a few seconds, and then witness the file size grow. ;-)

Reviewed-by Jeff Moyer <[email protected]>

Quoted 63 lines

Signed-off-by Dave Young <[email protected]>

---
 drivers/block/loop.c |   24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

drivers/block/loop.c +13 -11

@@ -357,14 +357,14 @@ lo_direct_splice_actor(struct pipe_inode
 	return __splice_from_pipe(pipe, sd, lo_splice_actor);
 }
 
-static int
+static ssize_t
 do_lo_receive(struct loop_device *lo,
 	      struct bio_vec *bvec, int bsize, loff_t pos)
 {
 	struct lo_read_data cookie;
 	struct splice_desc sd;
 	struct file *file;
-	long retval;
+	ssize_t retval;
 
 	cookie.lo = lo;
 	cookie.page = bvec->bv_page;

@@ -380,26 +380,28 @@ do_lo_receive(struct loop_device *lo,
 	file = lo->lo_backing_file;
 	retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
 
-	if (retval < 0)
-		return retval;
-	if (retval != bvec->bv_len)
-		return -EIO;
-	return 0;
+	return retval;
 }
 
 static int
 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
 {
 	struct bio_vec *bvec;
-	int i, ret = 0;
+	ssize_t s;
+	int i;
 
 	bio_for_each_segment(bvec, bio, i) {
-		ret = do_lo_receive(lo, bvec, bsize, pos);
-		if (ret < 0)
+		s = do_lo_receive(lo, bvec, bsize, pos);
+		if (s < 0)
+			return s;
+
+		if (s != bvec->bv_len) {
+			zero_fill_bio(bio);
 			break;
+		}
 		pos += bvec->bv_len;
 	}
-	return ret;
+	return 0;
 }
 
 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)

Quoted 7 lines · Jeff Moyer

Quoted 5 lines · Dave Young

commit 8268f5a7415d914fc855a86aa2284ac819dc6b2e trying to fix the loop device partial read information leak problem. But it changed the semantics of read behavior. When we read beyond the end of the device we should get 0 bytes, which is normal behavior, we should not just return -EIO

Quoted 7 lines

Quoted 2 lines

Instead of return -EIO, zero out the bio to avoid information leak in case of partail read.

I tested the patch with a program which patterns the loop device, truncates the backing file, and then performs preads from various offsets within the loop device, validates the return values and inspects the contents. With this patch, everything works as expected.

Many thanks for review and test

Quoted 6 lines

By the way, truncating the backing file for a loop device is insane. Why would you do that? Also, if you really want all of the data gone, you'll have to flush the contents of the buffer cache for the loop device first. It's quite the head scratcher the first time you truncate a file, wait a few seconds, and then witness the file size grow. ;-)

I think nobody will intend to do that, but random operations could cause this happen..

Quoted 65 lines

Reviewed-by Jeff Moyer <[email protected]>

Quoted 62 lines

Signed-off-by Dave Young <[email protected]>

---
 drivers/block/loop.c |   24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

drivers/block/loop.c +13 -11

@@ -357,14 +357,14 @@ lo_direct_splice_actor(struct pipe_inode
 	return __splice_from_pipe(pipe, sd, lo_splice_actor);
 }
 
-static int
+static ssize_t
 do_lo_receive(struct loop_device *lo,
 	      struct bio_vec *bvec, int bsize, loff_t pos)
 {
 	struct lo_read_data cookie;
 	struct splice_desc sd;
 	struct file *file;
-	long retval;
+	ssize_t retval;
 
 	cookie.lo = lo;
 	cookie.page = bvec->bv_page;

@@ -380,26 +380,28 @@ do_lo_receive(struct loop_device *lo,
 	file = lo->lo_backing_file;
 	retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
 
-	if (retval < 0)
-		return retval;
-	if (retval != bvec->bv_len)
-		return -EIO;
-	return 0;
+	return retval;
 }
 
 static int
 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
 {
 	struct bio_vec *bvec;
-	int i, ret = 0;
+	ssize_t s;
+	int i;
 
 	bio_for_each_segment(bvec, bio, i) {
-		ret = do_lo_receive(lo, bvec, bsize, pos);
-		if (ret < 0)
+		s = do_lo_receive(lo, bvec, bsize, pos);
+		if (s < 0)
+			return s;
+
+		if (s != bvec->bv_len) {
+			zero_fill_bio(bio);
 			break;
+		}
 		pos += bvec->bv_len;
 	}
-	return ret;
+	return 0;
 }
 

Quoted 1 line

drivers/block/loop.c do_lo_receive(struct loop_device *lo, · line 407

Quoted 1 line

drivers/block/loop.c +0 -0

 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)