]> git.baikalelectronics.ru Git - kernel.git/commitdiff
rbd: support cloning across namespaces
authorIlya Dryomov <idryomov@gmail.com>
Wed, 22 Aug 2018 15:26:10 +0000 (17:26 +0200)
committerIlya Dryomov <idryomov@gmail.com>
Thu, 6 Sep 2018 14:18:04 +0000 (16:18 +0200)
If parent_get class method is not supported by the OSDs, fall back to
the legacy class method and assume that the parent is in the default
(i.e. "") namespace.  The "use the child's image namespace" workaround
is no longer needed because creating images within namespaces will
require parent_get aware OSDs.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
drivers/block/rbd.c

index bec5a50c98907eb054e919d3fe93a4f7d4030812..73ed5f3a862dfcde598227671d55c2491d33408b 100644 (file)
@@ -4207,11 +4207,13 @@ static ssize_t rbd_parent_show(struct device *dev,
 
                count += sprintf(&buf[count], "%s"
                            "pool_id %llu\npool_name %s\n"
+                           "pool_ns %s\n"
                            "image_id %s\nimage_name %s\n"
                            "snap_id %llu\nsnap_name %s\n"
                            "overlap %llu\n",
                            !count ? "" : "\n", /* first? */
                            spec->pool_id, spec->pool_name,
+                           spec->pool_ns ?: "",
                            spec->image_id, spec->image_name ?: "(unknown)",
                            spec->snap_id, spec->snap_name,
                            rbd_dev->parent_overlap);
@@ -4586,12 +4588,89 @@ static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
 
 struct parent_image_info {
        u64             pool_id;
+       const char      *pool_ns;
        const char      *image_id;
        u64             snap_id;
 
+       bool            has_overlap;
        u64             overlap;
 };
 
+/*
+ * The caller is responsible for @pii.
+ */
+static int decode_parent_image_spec(void **p, void *end,
+                                   struct parent_image_info *pii)
+{
+       u8 struct_v;
+       u32 struct_len;
+       int ret;
+
+       ret = ceph_start_decoding(p, end, 1, "ParentImageSpec",
+                                 &struct_v, &struct_len);
+       if (ret)
+               return ret;
+
+       ceph_decode_64_safe(p, end, pii->pool_id, e_inval);
+       pii->pool_ns = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
+       if (IS_ERR(pii->pool_ns)) {
+               ret = PTR_ERR(pii->pool_ns);
+               pii->pool_ns = NULL;
+               return ret;
+       }
+       pii->image_id = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
+       if (IS_ERR(pii->image_id)) {
+               ret = PTR_ERR(pii->image_id);
+               pii->image_id = NULL;
+               return ret;
+       }
+       ceph_decode_64_safe(p, end, pii->snap_id, e_inval);
+       return 0;
+
+e_inval:
+       return -EINVAL;
+}
+
+static int __get_parent_info(struct rbd_device *rbd_dev,
+                            struct page *req_page,
+                            struct page *reply_page,
+                            struct parent_image_info *pii)
+{
+       struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
+       size_t reply_len = PAGE_SIZE;
+       void *p, *end;
+       int ret;
+
+       ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
+                            "rbd", "parent_get", CEPH_OSD_FLAG_READ,
+                            req_page, sizeof(u64), reply_page, &reply_len);
+       if (ret)
+               return ret == -EOPNOTSUPP ? 1 : ret;
+
+       p = page_address(reply_page);
+       end = p + reply_len;
+       ret = decode_parent_image_spec(&p, end, pii);
+       if (ret)
+               return ret;
+
+       ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
+                            "rbd", "parent_overlap_get", CEPH_OSD_FLAG_READ,
+                            req_page, sizeof(u64), reply_page, &reply_len);
+       if (ret)
+               return ret;
+
+       p = page_address(reply_page);
+       end = p + reply_len;
+       ceph_decode_8_safe(&p, end, pii->has_overlap, e_inval);
+       if (pii->has_overlap)
+               ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
+
+       return 0;
+
+e_inval:
+       return -EINVAL;
+}
+
 /*
  * The caller is responsible for @pii.
  */
@@ -4621,6 +4700,7 @@ static int __get_parent_info_legacy(struct rbd_device *rbd_dev,
                return ret;
        }
        ceph_decode_64_safe(&p, end, pii->snap_id, e_inval);
+       pii->has_overlap = true;
        ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
 
        return 0;
@@ -4648,7 +4728,10 @@ static int get_parent_info(struct rbd_device *rbd_dev,
 
        p = page_address(req_page);
        ceph_encode_64(&p, rbd_dev->spec->snap_id);
-       ret = __get_parent_info_legacy(rbd_dev, req_page, reply_page, pii);
+       ret = __get_parent_info(rbd_dev, req_page, reply_page, pii);
+       if (ret > 0)
+               ret = __get_parent_info_legacy(rbd_dev, req_page, reply_page,
+                                              pii);
 
        __free_page(req_page);
        __free_page(reply_page);
@@ -4669,10 +4752,11 @@ static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
        if (ret)
                goto out_err;
 
-       dout("%s pool_id %llu image_id %s snap_id %llu overlap %llu\n",
-            __func__, pii.pool_id, pii.image_id, pii.snap_id, pii.overlap);
+       dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n",
+            __func__, pii.pool_id, pii.pool_ns, pii.image_id, pii.snap_id,
+            pii.has_overlap, pii.overlap);
 
-       if (pii.pool_id == CEPH_NOPOOL) {
+       if (pii.pool_id == CEPH_NOPOOL || !pii.has_overlap) {
                /*
                 * Either the parent never existed, or we have
                 * record of it but the image got flattened so it no
@@ -4681,6 +4765,10 @@ static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
                 * overlap to 0.  The effect of this is that all new
                 * requests will be treated as if the image had no
                 * parent.
+                *
+                * If !pii.has_overlap, the parent image spec is not
+                * applicable.  It's there to avoid duplication in each
+                * snapshot record.
                 */
                if (rbd_dev->parent_overlap) {
                        rbd_dev->parent_overlap = 0;
@@ -4708,20 +4796,14 @@ static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
         */
        if (!rbd_dev->parent_spec) {
                parent_spec->pool_id = pii.pool_id;
+               if (pii.pool_ns && *pii.pool_ns) {
+                       parent_spec->pool_ns = pii.pool_ns;
+                       pii.pool_ns = NULL;
+               }
                parent_spec->image_id = pii.image_id;
                pii.image_id = NULL;
                parent_spec->snap_id = pii.snap_id;
 
-               /* TODO: support cloning across namespaces */
-               if (rbd_dev->spec->pool_ns) {
-                       parent_spec->pool_ns = kstrdup(rbd_dev->spec->pool_ns,
-                                                      GFP_KERNEL);
-                       if (!parent_spec->pool_ns) {
-                               ret = -ENOMEM;
-                               goto out_err;
-                       }
-               }
-
                rbd_dev->parent_spec = parent_spec;
                parent_spec = NULL;     /* rbd_dev now owns this */
        }
@@ -4746,6 +4828,7 @@ static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
 out:
        ret = 0;
 out_err:
+       kfree(pii.pool_ns);
        kfree(pii.image_id);
        rbd_spec_put(parent_spec);
        return ret;