Commit 40119211 authored by Vlad Buslov's avatar Vlad Buslov Committed by David S. Miller
Browse files

net: sched: refactor block offloads counter usage 



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: default avatarVlad Buslov <vladbu@mellanox.com>
Acked-by: default avatarJiri Pirko <jiri@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 97394bef

include/net/pkt_cls.h

+16 −1
Original line number Diff line number Diff line
@@ -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 {

include/net/sch_generic.h

+0 −31
Original line number Diff line number Diff line
@@ -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;

net/sched/cls_api.c

+162 −14
Original line number Diff line number Diff line
@@ -3000,13 +3000,97 @@ 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,

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;



	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)

				return err;

		} else {

			ok_count++;

		}

	}

	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) {
@@ -3014,22 +3098,86 @@ int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type, 
		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;

			}

		} else {

			ok_count++;

	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;

	return ok_count < 0 ? ok_count : 0;

}

EXPORT_SYMBOL(tc_setup_cb_call);

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)

net/sched/cls_bpf.c

+18 −20
Original line number Diff line number Diff line
@@ -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) {

	if (prog && err) {

		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 && 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))

		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;

			continue;

		}



		tc_cls_offload_cnt_update(block, &prog->in_hw_count,

					  &prog->gen_flags, add);

	}



	return 0;

net/sched/cls_flower.c

+14 −24
Original line number Diff line number Diff line
@@ -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;

		}



		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,

CRA Git | Maintained and supported by SUSTech CRA and CCSE