bcache: bugfix for race between moving_gc and bucket_invalidate

There is a possibility for a bucket to be invalidated by the allocator
while moving_gc was copying it's contents to another bucket, if the
bucket only held cached data. To prevent this moving checks for
a stale ptr (to an invalidated bucket), before and after reads.
It it finds one, it simply ignores moving that data. This only
affects bcache if the moving_gc was turned on, note that it's
off by default.

Signed-off-by: Nicholas Swenson <nks@daterainc.com>
Signed-off-by: Kent Overstreet <kmo@daterainc.com>

1 files changed, 10 insertions, 0 deletions

diff --git a/drivers/md/bcache/movinggc.c b/drivers/md/bcache/movinggc.c
index 30f347d4e609..f2f0998c4a91 100644
--- a/drivers/md/bcache/movinggc.c
+++ b/drivers/md/bcache/movinggc.c
@@ -64,11 +64,16 @@ static void write_moving_finish(struct closure *cl)
 
 static void read_moving_endio(struct bio *bio, int error)
 {
+        struct bbio *b = container_of(bio, struct bbio, bio);
         struct moving_io *io = container_of(bio->bi_private,
                                             struct moving_io, cl);
 
         if (error)
                 io->op.error = error;
+        else if (!KEY_DIRTY(&b->key) &&
+                 ptr_stale(io->op.c, &b->key, 0)) {
+                io->op.error = -EINTR;
+        }
 
         bch_bbio_endio(io->op.c, bio, error, "reading data to move");
 }
@@ -140,6 +145,11 @@ static void read_moving(struct cache_set *c)
                 if (!w)
                         break;
 
+                if (ptr_stale(c, &w->key, 0)) {
+                        bch_keybuf_del(&c->moving_gc_keys, w);
+                        continue;
+                }
+
                 io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec)
                              * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS),
                              GFP_KERNEL);