xen-blkback: fix shutdown race
authorRoger Pau Monne <roger.pau@citrix.com>
Tue, 4 Feb 2014 10:26:14 +0000 (11:26 +0100)
committerKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Fri, 7 Feb 2014 17:59:30 +0000 (12:59 -0500)
Introduce a new variable to keep track of the number of in-flight
requests. We need to make sure that when xen_blkif_put is called the
request has already been freed and we can safely free xen_blkif, which
was not the case before.

Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Tested-by: Matt Rushton <mrushton@amazon.com>
Reviewed-by: Matt Rushton <mrushton@amazon.com>
Cc: Matt Wilson <msw@amazon.com>
Cc: Ian Campbell <Ian.Campbell@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
drivers/block/xen-blkback/blkback.c
drivers/block/xen-blkback/common.h
drivers/block/xen-blkback/xenbus.c

index dcfe49fd3cb4a0b1ef6dd1206fac83993a248cb2..394fa2eabf873b23e29d9360f5883b00fa586c4a 100644 (file)
@@ -943,9 +943,7 @@ static void xen_blk_drain_io(struct xen_blkif *blkif)
 {
        atomic_set(&blkif->drain, 1);
        do {
-               /* The initial value is one, and one refcnt taken at the
-                * start of the xen_blkif_schedule thread. */
-               if (atomic_read(&blkif->refcnt) <= 2)
+               if (atomic_read(&blkif->inflight) == 0)
                        break;
                wait_for_completion_interruptible_timeout(
                                &blkif->drain_complete, HZ);
@@ -985,17 +983,30 @@ static void __end_block_io_op(struct pending_req *pending_req, int error)
         * the proper response on the ring.
         */
        if (atomic_dec_and_test(&pending_req->pendcnt)) {
-               xen_blkbk_unmap(pending_req->blkif,
+               struct xen_blkif *blkif = pending_req->blkif;
+
+               xen_blkbk_unmap(blkif,
                                pending_req->segments,
                                pending_req->nr_pages);
-               make_response(pending_req->blkif, pending_req->id,
+               make_response(blkif, pending_req->id,
                              pending_req->operation, pending_req->status);
-               xen_blkif_put(pending_req->blkif);
-               if (atomic_read(&pending_req->blkif->refcnt) <= 2) {
-                       if (atomic_read(&pending_req->blkif->drain))
-                               complete(&pending_req->blkif->drain_complete);
+               free_req(blkif, pending_req);
+               /*
+                * Make sure the request is freed before releasing blkif,
+                * or there could be a race between free_req and the
+                * cleanup done in xen_blkif_free during shutdown.
+                *
+                * NB: The fact that we might try to wake up pending_free_wq
+                * before drain_complete (in case there's a drain going on)
+                * it's not a problem with our current implementation
+                * because we can assure there's no thread waiting on
+                * pending_free_wq if there's a drain going on, but it has
+                * to be taken into account if the current model is changed.
+                */
+               if (atomic_dec_and_test(&blkif->inflight) && atomic_read(&blkif->drain)) {
+                       complete(&blkif->drain_complete);
                }
-               free_req(pending_req->blkif, pending_req);
+               xen_blkif_put(blkif);
        }
 }
 
@@ -1249,6 +1260,7 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
         * below (in "!bio") if we are handling a BLKIF_OP_DISCARD.
         */
        xen_blkif_get(blkif);
+       atomic_inc(&blkif->inflight);
 
        for (i = 0; i < nseg; i++) {
                while ((bio == NULL) ||
index f733d7627120e199e676d0892f5dedfd06affb63..e40326a7f7075fb1d985387dd4c62135aad26be3 100644 (file)
@@ -278,6 +278,7 @@ struct xen_blkif {
        /* for barrier (drain) requests */
        struct completion       drain_complete;
        atomic_t                drain;
+       atomic_t                inflight;
        /* One thread per one blkif. */
        struct task_struct      *xenblkd;
        unsigned int            waiting_reqs;
index 8afef67fecdd18a286d2365d121b39af8ba0322a..84973c6a856aba0c33c1ebf305d75fd7fb200fe0 100644 (file)
@@ -128,6 +128,7 @@ static struct xen_blkif *xen_blkif_alloc(domid_t domid)
        INIT_LIST_HEAD(&blkif->persistent_purge_list);
        blkif->free_pages_num = 0;
        atomic_set(&blkif->persistent_gnt_in_use, 0);
+       atomic_set(&blkif->inflight, 0);
 
        INIT_LIST_HEAD(&blkif->pending_free);