net: sched: refactor block offloads counter usage
authorVlad Buslov <vladbu@mellanox.com>
Mon, 26 Aug 2019 13:44:59 +0000 (16:44 +0300)
committerDavid S. Miller <davem@davemloft.net>
Mon, 26 Aug 2019 21:17:43 +0000 (14:17 -0700)
Without rtnl lock protection filters can no longer safely manage block
offloads counter themselves. Refactor cls API to protect block offloadcnt
with tcf_block->cb_lock that is already used to protect driver callback
list and nooffloaddevcnt counter. The counter can be modified by concurrent
tasks by new functions that execute block callbacks (which is safe with
previous patch that changed its type to atomic_t), however, block
bind/unbind code that checks the counter value takes cb_lock in write mode
to exclude any concurrent modifications. This approach prevents race
conditions between bind/unbind and callback execution code but allows for
concurrency for tc rule update path.

Move block offload counter, filter in hardware counter and filter flags
management from classifiers into cls hardware offloads API. Make functions
tcf_block_offload_{inc|dec}() and tc_cls_offload_cnt_update() to be cls API
private. Implement following new cls API to be used instead:

  tc_setup_cb_add() - non-destructive filter add. If filter that wasn't
  already in hardware is successfully offloaded, increment block offloads
  counter, set filter in hardware counter and flag. On failure, previously
  offloaded filter is considered to be intact and offloads counter is not
  decremented.

  tc_setup_cb_replace() - destructive filter replace. Release existing
  filter block offload counter and reset its in hardware counter and flag.
  Set new filter in hardware counter and flag. On failure, previously
  offloaded filter is considered to be destroyed and offload counter is
  decremented.

  tc_setup_cb_destroy() - filter destroy. Unconditionally decrement block
  offloads counter.

  tc_setup_cb_reoffload() - reoffload filter to single cb. Execute cb() and
  call tc_cls_offload_cnt_update() if cb() didn't return an error.

Refactor all offload-capable classifiers to atomically offload filters to
hardware, change block offload counter, and set filter in hardware counter
and flag by means of the new cls API functions.

Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/pkt_cls.h
include/net/sch_generic.h
net/sched/cls_api.c
net/sched/cls_bpf.c
net/sched/cls_flower.c
net/sched/cls_matchall.c
net/sched/cls_u32.c

index 64999ffcb486851e6058685d9e804e5ba5a6745b..612232492f6739289d94b215ef33914fe7a98bbe 100644 (file)
@@ -506,7 +506,22 @@ tcf_match_indev(struct sk_buff *skb, int ifindex)
 int tc_setup_flow_action(struct flow_action *flow_action,
                         const struct tcf_exts *exts);
 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
-                    void *type_data, bool err_stop);
+                    void *type_data, bool err_stop, bool rtnl_held);
+int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
+                   enum tc_setup_type type, void *type_data, bool err_stop,
+                   u32 *flags, unsigned int *in_hw_count, bool rtnl_held);
+int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
+                       enum tc_setup_type type, void *type_data, bool err_stop,
+                       u32 *old_flags, unsigned int *old_in_hw_count,
+                       u32 *new_flags, unsigned int *new_in_hw_count,
+                       bool rtnl_held);
+int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
+                       enum tc_setup_type type, void *type_data, bool err_stop,
+                       u32 *flags, unsigned int *in_hw_count, bool rtnl_held);
+int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
+                         bool add, flow_setup_cb_t *cb,
+                         enum tc_setup_type type, void *type_data,
+                         void *cb_priv, u32 *flags, unsigned int *in_hw_count);
 unsigned int tcf_exts_num_actions(struct tcf_exts *exts);
 
 struct tc_cls_u32_knode {
index d778c502decd3fd73d2342442bb2ebd090f9827f..f90e3b2a30654334367ea746d73a994b3812bedf 100644 (file)
@@ -439,37 +439,6 @@ static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
 #define tcf_proto_dereference(p, tp)                                   \
        rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
 
-static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
-{
-       if (*flags & TCA_CLS_FLAGS_IN_HW)
-               return;
-       *flags |= TCA_CLS_FLAGS_IN_HW;
-       atomic_inc(&block->offloadcnt);
-}
-
-static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
-{
-       if (!(*flags & TCA_CLS_FLAGS_IN_HW))
-               return;
-       *flags &= ~TCA_CLS_FLAGS_IN_HW;
-       atomic_dec(&block->offloadcnt);
-}
-
-static inline void
-tc_cls_offload_cnt_update(struct tcf_block *block, u32 *cnt,
-                         u32 *flags, bool add)
-{
-       if (add) {
-               if (!*cnt)
-                       tcf_block_offload_inc(block, flags);
-               (*cnt)++;
-       } else {
-               (*cnt)--;
-               if (!*cnt)
-                       tcf_block_offload_dec(block, flags);
-       }
-}
-
 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
 {
        struct qdisc_skb_cb *qcb;
index f2c2f8159e355a3b7b6a8e7c40bee30a560b03fb..6e612984e4a6e809cf834653201f4f8a6a7aa46c 100644 (file)
@@ -3000,37 +3000,185 @@ int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
 }
 EXPORT_SYMBOL(tcf_exts_dump_stats);
 
-int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
-                    void *type_data, bool err_stop)
+static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
+{
+       if (*flags & TCA_CLS_FLAGS_IN_HW)
+               return;
+       *flags |= TCA_CLS_FLAGS_IN_HW;
+       atomic_inc(&block->offloadcnt);
+}
+
+static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
+{
+       if (!(*flags & TCA_CLS_FLAGS_IN_HW))
+               return;
+       *flags &= ~TCA_CLS_FLAGS_IN_HW;
+       atomic_dec(&block->offloadcnt);
+}
+
+static void tc_cls_offload_cnt_update(struct tcf_block *block,
+                                     struct tcf_proto *tp, u32 *cnt,
+                                     u32 *flags, u32 diff, bool add)
+{
+       lockdep_assert_held(&block->cb_lock);
+
+       spin_lock(&tp->lock);
+       if (add) {
+               if (!*cnt)
+                       tcf_block_offload_inc(block, flags);
+               *cnt += diff;
+       } else {
+               *cnt -= diff;
+               if (!*cnt)
+                       tcf_block_offload_dec(block, flags);
+       }
+       spin_unlock(&tp->lock);
+}
+
+static void
+tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
+                        u32 *cnt, u32 *flags)
+{
+       lockdep_assert_held(&block->cb_lock);
+
+       spin_lock(&tp->lock);
+       tcf_block_offload_dec(block, flags);
+       *cnt = 0;
+       spin_unlock(&tp->lock);
+}
+
+static int
+__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
+                  void *type_data, bool err_stop)
 {
        struct flow_block_cb *block_cb;
        int ok_count = 0;
        int err;
 
-       down_read(&block->cb_lock);
-       /* Make sure all netdevs sharing this block are offload-capable. */
-       if (block->nooffloaddevcnt && err_stop) {
-               ok_count = -EOPNOTSUPP;
-               goto err_unlock;
-       }
-
        list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
                err = block_cb->cb(type, type_data, block_cb->cb_priv);
                if (err) {
-                       if (err_stop) {
-                               ok_count = err;
-                               goto err_unlock;
-                       }
+                       if (err_stop)
+                               return err;
                } else {
                        ok_count++;
                }
        }
-err_unlock:
+       return ok_count;
+}
+
+int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
+                    void *type_data, bool err_stop, bool rtnl_held)
+{
+       int ok_count;
+
+       down_read(&block->cb_lock);
+       ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
        up_read(&block->cb_lock);
        return ok_count;
 }
 EXPORT_SYMBOL(tc_setup_cb_call);
 
+/* Non-destructive filter add. If filter that wasn't already in hardware is
+ * successfully offloaded, increment block offloads counter. On failure,
+ * previously offloaded filter is considered to be intact and offloads counter
+ * is not decremented.
+ */
+
+int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
+                   enum tc_setup_type type, void *type_data, bool err_stop,
+                   u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
+{
+       int ok_count;
+
+       down_read(&block->cb_lock);
+       /* Make sure all netdevs sharing this block are offload-capable. */
+       if (block->nooffloaddevcnt && err_stop) {
+               ok_count = -EOPNOTSUPP;
+               goto err_unlock;
+       }
+
+       ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
+       if (ok_count > 0)
+               tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
+                                         ok_count, true);
+err_unlock:
+       up_read(&block->cb_lock);
+       return ok_count < 0 ? ok_count : 0;
+}
+EXPORT_SYMBOL(tc_setup_cb_add);
+
+/* Destructive filter replace. If filter that wasn't already in hardware is
+ * successfully offloaded, increment block offload counter. On failure,
+ * previously offloaded filter is considered to be destroyed and offload counter
+ * is decremented.
+ */
+
+int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
+                       enum tc_setup_type type, void *type_data, bool err_stop,
+                       u32 *old_flags, unsigned int *old_in_hw_count,
+                       u32 *new_flags, unsigned int *new_in_hw_count,
+                       bool rtnl_held)
+{
+       int ok_count;
+
+       down_read(&block->cb_lock);
+       /* Make sure all netdevs sharing this block are offload-capable. */
+       if (block->nooffloaddevcnt && err_stop) {
+               ok_count = -EOPNOTSUPP;
+               goto err_unlock;
+       }
+
+       tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
+
+       ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
+       if (ok_count > 0)
+               tc_cls_offload_cnt_update(block, tp, new_in_hw_count, new_flags,
+                                         ok_count, true);
+err_unlock:
+       up_read(&block->cb_lock);
+       return ok_count < 0 ? ok_count : 0;
+}
+EXPORT_SYMBOL(tc_setup_cb_replace);
+
+/* Destroy filter and decrement block offload counter, if filter was previously
+ * offloaded.
+ */
+
+int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
+                       enum tc_setup_type type, void *type_data, bool err_stop,
+                       u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
+{
+       int ok_count;
+
+       down_read(&block->cb_lock);
+       ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
+
+       tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
+       up_read(&block->cb_lock);
+       return ok_count < 0 ? ok_count : 0;
+}
+EXPORT_SYMBOL(tc_setup_cb_destroy);
+
+int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
+                         bool add, flow_setup_cb_t *cb,
+                         enum tc_setup_type type, void *type_data,
+                         void *cb_priv, u32 *flags, unsigned int *in_hw_count)
+{
+       int err = cb(type, type_data, cb_priv);
+
+       if (err) {
+               if (add && tc_skip_sw(*flags))
+                       return err;
+       } else {
+               tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
+                                         add);
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(tc_setup_cb_reoffload);
+
 int tc_setup_flow_action(struct flow_action *flow_action,
                         const struct tcf_exts *exts)
 {
index 3f7a9c02b70c570510fefae26a4b7fcedbabe432..bf10bdaf5012734012b34a30e0d31faebcbeed21 100644 (file)
@@ -163,17 +163,19 @@ static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
        cls_bpf.exts_integrated = obj->exts_integrated;
 
        if (oldprog)
-               tcf_block_offload_dec(block, &oldprog->gen_flags);
+               err = tc_setup_cb_replace(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
+                                         skip_sw, &oldprog->gen_flags,
+                                         &oldprog->in_hw_count,
+                                         &prog->gen_flags, &prog->in_hw_count,
+                                         true);
+       else
+               err = tc_setup_cb_add(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
+                                     skip_sw, &prog->gen_flags,
+                                     &prog->in_hw_count, true);
 
-       err = tc_setup_cb_call(block, TC_SETUP_CLSBPF, &cls_bpf, skip_sw);
-       if (prog) {
-               if (err < 0) {
-                       cls_bpf_offload_cmd(tp, oldprog, prog, extack);
-                       return err;
-               } else if (err > 0) {
-                       prog->in_hw_count = err;
-                       tcf_block_offload_inc(block, &prog->gen_flags);
-               }
+       if (prog && err) {
+               cls_bpf_offload_cmd(tp, oldprog, prog, extack);
+               return err;
        }
 
        if (prog && skip_sw && !(prog->gen_flags & TCA_CLS_FLAGS_IN_HW))
@@ -230,7 +232,7 @@ static void cls_bpf_offload_update_stats(struct tcf_proto *tp,
        cls_bpf.name = prog->bpf_name;
        cls_bpf.exts_integrated = prog->exts_integrated;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSBPF, &cls_bpf, false);
+       tc_setup_cb_call(block, TC_SETUP_CLSBPF, &cls_bpf, false, true);
 }
 
 static int cls_bpf_init(struct tcf_proto *tp)
@@ -673,15 +675,11 @@ static int cls_bpf_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb
                cls_bpf.name = prog->bpf_name;
                cls_bpf.exts_integrated = prog->exts_integrated;
 
-               err = cb(TC_SETUP_CLSBPF, &cls_bpf, cb_priv);
-               if (err) {
-                       if (add && tc_skip_sw(prog->gen_flags))
-                               return err;
-                       continue;
-               }
-
-               tc_cls_offload_cnt_update(block, &prog->in_hw_count,
-                                         &prog->gen_flags, add);
+               err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSBPF,
+                                           &cls_bpf, cb_priv, &prog->gen_flags,
+                                           &prog->in_hw_count);
+               if (err)
+                       return err;
        }
 
        return 0;
index 054123742e323baa402798708a8ae98d73f68e5a..cb816bbbd376a94c52efd409921b6a150b28731b 100644 (file)
@@ -419,10 +419,10 @@ static void fl_hw_destroy_filter(struct tcf_proto *tp, struct cls_fl_filter *f,
        cls_flower.command = FLOW_CLS_DESTROY;
        cls_flower.cookie = (unsigned long) f;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, false);
+       tc_setup_cb_destroy(block, tp, TC_SETUP_CLSFLOWER, &cls_flower, false,
+                           &f->flags, &f->in_hw_count, true);
        spin_lock(&tp->lock);
        list_del_init(&f->hw_list);
-       tcf_block_offload_dec(block, &f->flags);
        spin_unlock(&tp->lock);
 
        if (!rtnl_held)
@@ -466,18 +466,13 @@ static int fl_hw_replace_filter(struct tcf_proto *tp,
                goto errout;
        }
 
-       err = tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, skip_sw);
+       err = tc_setup_cb_add(block, tp, TC_SETUP_CLSFLOWER, &cls_flower,
+                             skip_sw, &f->flags, &f->in_hw_count, true);
        kfree(cls_flower.rule);
 
-       if (err < 0) {
+       if (err) {
                fl_hw_destroy_filter(tp, f, true, NULL);
                goto errout;
-       } else if (err > 0) {
-               f->in_hw_count = err;
-               err = 0;
-               spin_lock(&tp->lock);
-               tcf_block_offload_inc(block, &f->flags);
-               spin_unlock(&tp->lock);
        }
 
        if (skip_sw && !(f->flags & TCA_CLS_FLAGS_IN_HW)) {
@@ -509,7 +504,7 @@ static void fl_hw_update_stats(struct tcf_proto *tp, struct cls_fl_filter *f,
        cls_flower.cookie = (unsigned long) f;
        cls_flower.classid = f->res.classid;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, false);
+       tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, false, true);
 
        tcf_exts_stats_update(&f->exts, cls_flower.stats.bytes,
                              cls_flower.stats.pkts,
@@ -1844,21 +1839,16 @@ static int fl_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
 
                cls_flower.classid = f->res.classid;
 
-               err = cb(TC_SETUP_CLSFLOWER, &cls_flower, cb_priv);
+               err = tc_setup_cb_reoffload(block, tp, add, cb,
+                                           TC_SETUP_CLSFLOWER, &cls_flower,
+                                           cb_priv, &f->flags,
+                                           &f->in_hw_count);
                kfree(cls_flower.rule);
 
                if (err) {
-                       if (add && tc_skip_sw(f->flags)) {
-                               __fl_put(f);
-                               return err;
-                       }
-                       goto next_flow;
+                       __fl_put(f);
+                       return err;
                }
-
-               spin_lock(&tp->lock);
-               tc_cls_offload_cnt_update(block, &f->in_hw_count, &f->flags,
-                                         add);
-               spin_unlock(&tp->lock);
 next_flow:
                __fl_put(f);
        }
@@ -1886,7 +1876,7 @@ static int fl_hw_create_tmplt(struct tcf_chain *chain,
        /* We don't care if driver (any of them) fails to handle this
         * call. It serves just as a hint for it.
         */
-       tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, false);
+       tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, false, true);
        kfree(cls_flower.rule);
 
        return 0;
@@ -1902,7 +1892,7 @@ static void fl_hw_destroy_tmplt(struct tcf_chain *chain,
        cls_flower.command = FLOW_CLS_TMPLT_DESTROY;
        cls_flower.cookie = (unsigned long) tmplt;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, false);
+       tc_setup_cb_call(block, TC_SETUP_CLSFLOWER, &cls_flower, false, true);
 }
 
 static void *fl_tmplt_create(struct net *net, struct tcf_chain *chain,
index 455ea2793f9b37f9edb9a11754845c2b6d61dc39..911d1ea28bb2305a97bf9b18a43154c3d813c1bd 100644 (file)
@@ -75,8 +75,8 @@ static void mall_destroy_hw_filter(struct tcf_proto *tp,
        cls_mall.command = TC_CLSMATCHALL_DESTROY;
        cls_mall.cookie = cookie;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSMATCHALL, &cls_mall, false);
-       tcf_block_offload_dec(block, &head->flags);
+       tc_setup_cb_destroy(block, tp, TC_SETUP_CLSMATCHALL, &cls_mall, false,
+                           &head->flags, &head->in_hw_count, true);
 }
 
 static int mall_replace_hw_filter(struct tcf_proto *tp,
@@ -109,15 +109,13 @@ static int mall_replace_hw_filter(struct tcf_proto *tp,
                return err;
        }
 
-       err = tc_setup_cb_call(block, TC_SETUP_CLSMATCHALL, &cls_mall, skip_sw);
+       err = tc_setup_cb_add(block, tp, TC_SETUP_CLSMATCHALL, &cls_mall,
+                             skip_sw, &head->flags, &head->in_hw_count, true);
        kfree(cls_mall.rule);
 
-       if (err < 0) {
+       if (err) {
                mall_destroy_hw_filter(tp, head, cookie, NULL);
                return err;
-       } else if (err > 0) {
-               head->in_hw_count = err;
-               tcf_block_offload_inc(block, &head->flags);
        }
 
        if (skip_sw && !(head->flags & TCA_CLS_FLAGS_IN_HW))
@@ -312,16 +310,13 @@ static int mall_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
                return 0;
        }
 
-       err = cb(TC_SETUP_CLSMATCHALL, &cls_mall, cb_priv);
+       err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSMATCHALL,
+                                   &cls_mall, cb_priv, &head->flags,
+                                   &head->in_hw_count);
        kfree(cls_mall.rule);
 
-       if (err) {
-               if (add && tc_skip_sw(head->flags))
-                       return err;
-               return 0;
-       }
-
-       tc_cls_offload_cnt_update(block, &head->in_hw_count, &head->flags, add);
+       if (err)
+               return err;
 
        return 0;
 }
@@ -337,7 +332,7 @@ static void mall_stats_hw_filter(struct tcf_proto *tp,
        cls_mall.command = TC_CLSMATCHALL_STATS;
        cls_mall.cookie = cookie;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSMATCHALL, &cls_mall, false);
+       tc_setup_cb_call(block, TC_SETUP_CLSMATCHALL, &cls_mall, false, true);
 
        tcf_exts_stats_update(&head->exts, cls_mall.stats.bytes,
                              cls_mall.stats.pkts, cls_mall.stats.lastused);
index 8614088edd1be0cad1a9755ce11fac94304a65f0..a0e6fac613de215bd2ca588624e2f3edda7c19ff 100644 (file)
@@ -480,7 +480,7 @@ static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
        cls_u32.hnode.handle = h->handle;
        cls_u32.hnode.prio = h->prio;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
+       tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
 }
 
 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
@@ -498,7 +498,7 @@ static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
        cls_u32.hnode.handle = h->handle;
        cls_u32.hnode.prio = h->prio;
 
-       err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
+       err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
        if (err < 0) {
                u32_clear_hw_hnode(tp, h, NULL);
                return err;
@@ -522,8 +522,8 @@ static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
        cls_u32.command = TC_CLSU32_DELETE_KNODE;
        cls_u32.knode.handle = n->handle;
 
-       tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
-       tcf_block_offload_dec(block, &n->flags);
+       tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
+                           &n->flags, &n->in_hw_count, true);
 }
 
 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
@@ -552,13 +552,11 @@ static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
        if (n->ht_down)
                cls_u32.knode.link_handle = ht->handle;
 
-       err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
-       if (err < 0) {
+       err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
+                             &n->flags, &n->in_hw_count, true);
+       if (err) {
                u32_remove_hw_knode(tp, n, NULL);
                return err;
-       } else if (err > 0) {
-               n->in_hw_count = err;
-               tcf_block_offload_inc(block, &n->flags);
        }
 
        if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
@@ -1201,14 +1199,11 @@ static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
                        cls_u32.knode.link_handle = ht->handle;
        }
 
-       err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
-       if (err) {
-               if (add && tc_skip_sw(n->flags))
-                       return err;
-               return 0;
-       }
-
-       tc_cls_offload_cnt_update(block, &n->in_hw_count, &n->flags, add);
+       err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
+                                   &cls_u32, cb_priv, &n->flags,
+                                   &n->in_hw_count);
+       if (err)
+               return err;
 
        return 0;
 }