[PATCH v2] ubi: ubi_wl_put_peb: Fix infinite loop when wear-leveling work failed

[Zhihao Cheng]

Following process will trigger an infinite loop in ubi_wl_put_peb():

	ubifs_bgt		ubi_bgt
ubifs_leb_unmap
  ubi_leb_unmap
    ubi_eba_unmap_leb
      ubi_wl_put_peb	wear_leveling_worker
                          e1 = rb_entry(rb_first(&ubi->used)
			  e2 = get_peb_for_wl(ubi)
			  ubi_io_read_vid_hdr  // return err (flash fault)
			  out_error:
			    ubi->move_from = ubi->move_to = NULL
			    wl_entry_destroy(ubi, e1)
			      ubi->lookuptbl[e->pnum] = NULL
      retry:
        e = ubi->lookuptbl[pnum];	// return NULL
	if (e == ubi->move_from) {	// NULL == NULL gets true
	  goto retry;			// infinite loop !!!

$ top
  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     COMMAND
  7676 root     20   0       0      0      0 R 100.0  0.0  ubifs_bgt0_0

Fix it by:
 1) Letting ubi_wl_put_peb() returns directly if wearl leveling entry has
    been removed from 'ubi->lookuptbl'.
 2) Using 'ubi->wl_lock' protecting wl entry deletion to preventing an
    use-after-free problem for wl entry in ubi_wl_put_peb().

Fetch a reproducer in [Link].

Fixes: 43f9b25a9cdd7b1 ("UBI: bugfix: protect from volume removal")
Fixes: ee59ba8b064f692 ("UBI: Fix stale pointers in ubi->lookuptbl")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=216111
Signed-off-by: Zhihao Cheng <chengzhihao1 at huawei.com>
---
 v1->v2: Don't split wl_entry_destroy(), since kmem_cache_free() can be
         executed wrapped a spinlock, eg dtl_disable().
 drivers/mtd/ubi/wl.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/drivers/mtd/ubi/wl.c b/drivers/mtd/ubi/wl.c
index 55bae06cf408..ee0100740869 100644
--- a/drivers/mtd/ubi/wl.c
+++ b/drivers/mtd/ubi/wl.c
@@ -973,11 +973,11 @@ static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
 	spin_lock(&ubi->wl_lock);
 	ubi->move_from = ubi->move_to = NULL;
 	ubi->move_to_put = ubi->wl_scheduled = 0;
+	wl_entry_destroy(ubi, e1);
+	wl_entry_destroy(ubi, e2);
 	spin_unlock(&ubi->wl_lock);
 
 	ubi_free_vid_buf(vidb);
-	wl_entry_destroy(ubi, e1);
-	wl_entry_destroy(ubi, e2);
 
 out_ro:
 	ubi_ro_mode(ubi);
@@ -1253,6 +1253,18 @@ int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
 retry:
 	spin_lock(&ubi->wl_lock);
 	e = ubi->lookuptbl[pnum];
+	if (!e) {
+		/*
+		 * This wl entry has been removed for some errors by other
+		 * process (eg. wear leveling worker), corresponding process
+		 * (except __erase_worker, which cannot concurrent with
+		 * ubi_wl_put_peb) will set ubi ro_mode at the same time,
+		 * just ignore this wl entry.
+		 */
+		spin_unlock(&ubi->wl_lock);
+		up_read(&ubi->fm_protect);
+		return 0;
+	}
 	if (e == ubi->move_from) {
 		/*
 		 * User is putting the physical eraseblock which was selected to
-- 
2.31.1