possible use-after-free in drivers/mtd/ubi/wl.c: erase_worker

Hi,

I think I found a potential problem in drivers/mtd/ubi/wl.c in erase_worker():

1050         ubi_err("failed to erase PEB %d, error %d", pnum, err);
1051         kfree(wl_wrk);
1052         kmem_cache_free(ubi_wl_entry_slab, e);

1053
1054 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1055 err == -EBUSY) {
1056 int err1;
1057
1058 /* Re-schedule the LEB for erasure */
1059 err1 = schedule_erase(ubi, e, 0);

The pointer e is freed at line 1052 (kmem_cache_free), but later it is passed to schedule_erase which will eventually call erase_worker where it will be dereferenced and/or freed again.

It seems to have been introduced in commit
784c145444e7dd58ae740d406155b72ac658f151

Emese

Yes. That does seem to be a problem. Care to create a patch that solves this by deferring the frees until after the err variable is checked?

josh

Well-spotted, thanks. I've just pushed this patch to the UBIFS git tree:

From: Artem Bityutskiy <[email protected]>
Subject: [PATCH] UBI: fix use-after-free on error path

When we fail to erase a PEB, we free the corresponding erase entry object, but then re-schedule this object if the error code was something like -EAGAIN. Obviously, it is a bug to use the object after we have freed it.

---
 drivers/mtd/ubi/wl.c |    7 ++++---
 1 files changed, 4 insertions(+), 3 deletions(-)

index 42c684c..036d213 100644
@@ -1049,7 +1049,6 @@ static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
 
 	ubi_err("failed to erase PEB %d, error %d", pnum, err);
 	kfree(wl_wrk);
-	kmem_cache_free(ubi_wl_entry_slab, e);
 
 	if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
 	    err == -EBUSY) {

@@ -1062,14 +1061,16 @@ static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
 			goto out_ro;
 		}
 		return err;
-	} else if (err != -EIO) {
+	}
+
+	kmem_cache_free(ubi_wl_entry_slab, e);
+	if (err != -EIO)
 		/*
 		 * If this is not %-EIO, we have no idea what to do. Scheduling
 		 * this physical eraseblock for erasure again would cause
 		 * errors again and again. Well, lets switch to R/O mode.
 		 */
 		goto out_ro;
-	}
 
 	/* It is %-EIO, the PEB went bad */
 

1.7.7.3

-- 
Best Regards,
Artem Bityutskiy