rbd: use reference counting for the snap context
authorJosh Durgin <josh.durgin@dreamhost.com>
Mon, 5 Dec 2011 22:03:05 +0000 (14:03 -0800)
committerSage Weil <sage@inktank.com>
Tue, 31 Jul 2012 01:15:40 +0000 (18:15 -0700)
This prevents a race between requests with a given snap context and
header updates that free it. The osd client was already expecting the
snap context to be reference counted, since it get()s it in
ceph_osdc_build_request and put()s it when the request completes.

Also remove the second down_read()/up_read() on header_rwsem in
rbd_do_request, which wasn't actually preventing this race or
protecting any other data.

Signed-off-by: Josh Durgin <josh.durgin@dreamhost.com>
Reviewed-by: Alex Elder <elder@inktank.com>
drivers/block/rbd.c

index a6bbda2..988f944 100644 (file)
@@ -626,7 +626,7 @@ static void rbd_header_free(struct rbd_image_header *header)
        kfree(header->object_prefix);
        kfree(header->snap_sizes);
        kfree(header->snap_names);
-       kfree(header->snapc);
+       ceph_put_snap_context(header->snapc);
 }
 
 /*
@@ -902,13 +902,10 @@ static int rbd_do_request(struct request *rq,
        dout("rbd_do_request object_name=%s ofs=%lld len=%lld\n",
                object_name, len, ofs);
 
-       down_read(&rbd_dev->header_rwsem);
-
        osdc = &rbd_dev->rbd_client->client->osdc;
        req = ceph_osdc_alloc_request(osdc, flags, snapc, ops,
                                        false, GFP_NOIO, pages, bio);
        if (!req) {
-               up_read(&rbd_dev->header_rwsem);
                ret = -ENOMEM;
                goto done_pages;
        }
@@ -942,7 +939,6 @@ static int rbd_do_request(struct request *rq,
                                snapc,
                                &mtime,
                                req->r_oid, req->r_oid_len);
-       up_read(&rbd_dev->header_rwsem);
 
        if (linger_req) {
                ceph_osdc_set_request_linger(osdc, req);
@@ -1448,6 +1444,7 @@ static void rbd_rq_fn(struct request_queue *q)
                u64 ofs;
                int num_segs, cur_seg = 0;
                struct rbd_req_coll *coll;
+               struct ceph_snap_context *snapc;
 
                /* peek at request from block layer */
                if (!rq)
@@ -1474,21 +1471,20 @@ static void rbd_rq_fn(struct request_queue *q)
 
                spin_unlock_irq(q->queue_lock);
 
-               if (rbd_dev->snap_id != CEPH_NOSNAP) {
-                       bool snap_exists;
+               down_read(&rbd_dev->header_rwsem);
 
-                       down_read(&rbd_dev->header_rwsem);
-                       snap_exists = rbd_dev->snap_exists;
+               if (rbd_dev->snap_id != CEPH_NOSNAP && !rbd_dev->snap_exists) {
                        up_read(&rbd_dev->header_rwsem);
-
-                       if (!snap_exists) {
-                               dout("request for non-existent snapshot");
-                               spin_lock_irq(q->queue_lock);
-                               __blk_end_request_all(rq, -ENXIO);
-                               continue;
-                       }
+                       dout("request for non-existent snapshot");
+                       spin_lock_irq(q->queue_lock);
+                       __blk_end_request_all(rq, -ENXIO);
+                       continue;
                }
 
+               snapc = ceph_get_snap_context(rbd_dev->header.snapc);
+
+               up_read(&rbd_dev->header_rwsem);
+
                dout("%s 0x%x bytes at 0x%llx\n",
                     do_write ? "write" : "read",
                     size, blk_rq_pos(rq) * SECTOR_SIZE);
@@ -1498,6 +1494,7 @@ static void rbd_rq_fn(struct request_queue *q)
                if (!coll) {
                        spin_lock_irq(q->queue_lock);
                        __blk_end_request_all(rq, -ENOMEM);
+                       ceph_put_snap_context(snapc);
                        continue;
                }
 
@@ -1521,7 +1518,7 @@ static void rbd_rq_fn(struct request_queue *q)
                        /* init OSD command: write or read */
                        if (do_write)
                                rbd_req_write(rq, rbd_dev,
-                                             rbd_dev->header.snapc,
+                                             snapc,
                                              ofs,
                                              op_size, bio,
                                              coll, cur_seg);
@@ -1544,6 +1541,8 @@ next_seg:
                if (bp)
                        bio_pair_release(bp);
                spin_lock_irq(q->queue_lock);
+
+               ceph_put_snap_context(snapc);
        }
 }
 
@@ -1744,7 +1743,8 @@ static int __rbd_refresh_header(struct rbd_device *rbd_dev)
        /* rbd_dev->header.object_prefix shouldn't change */
        kfree(rbd_dev->header.snap_sizes);
        kfree(rbd_dev->header.snap_names);
-       kfree(rbd_dev->header.snapc);
+       /* osd requests may still refer to snapc */
+       ceph_put_snap_context(rbd_dev->header.snapc);
 
        rbd_dev->header.image_size = h.image_size;
        rbd_dev->header.total_snaps = h.total_snaps;