]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
net: sched: protect block offload-related fields with rw_semaphore
authorVlad Buslov <vladbu@mellanox.com>
Mon, 26 Aug 2019 13:44:57 +0000 (16:44 +0300)
committerDavid S. Miller <davem@davemloft.net>
Mon, 26 Aug 2019 21:17:43 +0000 (14:17 -0700)
In order to remove dependency on rtnl lock, extend tcf_block with 'cb_lock'
rwsem and use it to protect flow_block->cb_list and related counters from
concurrent modification. The lock is taken in read mode for read-only
traversal of cb_list in tc_setup_cb_call() and write mode in all other
cases. This approach ensures that:

- cb_list is not changed concurrently while filters is being offloaded on
  block.

- block->nooffloaddevcnt is checked while holding the lock in read mode,
  but is only changed by bind/unbind code when holding the cb_lock in write
  mode to prevent concurrent modification.

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/sch_generic.h
net/sched/cls_api.c

index d9f359af0b93e8463158d354e0a9ad3351efa3a1..a3eaf5f9d28ff498b57f05e6b5455cac68f5b8cc 100644 (file)
@@ -13,6 +13,7 @@
 #include <linux/refcount.h>
 #include <linux/workqueue.h>
 #include <linux/mutex.h>
+#include <linux/rwsem.h>
 #include <net/gen_stats.h>
 #include <net/rtnetlink.h>
 #include <net/flow_offload.h>
@@ -396,6 +397,7 @@ struct tcf_block {
        refcount_t refcnt;
        struct net *net;
        struct Qdisc *q;
+       struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
        struct flow_block flow_block;
        struct list_head owner_list;
        bool keep_dst;
index e0d8b456e9f56441ead1285083926783e35de5df..959b7ca1ca022ca4319a674060a85d5d55cc2250 100644 (file)
@@ -568,9 +568,11 @@ static void tc_indr_block_ing_cmd(struct net_device *dev,
 
        bo.block = &block->flow_block;
 
+       down_write(&block->cb_lock);
        cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
 
        tcf_block_setup(block, &bo);
+       up_write(&block->cb_lock);
 }
 
 static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
@@ -661,6 +663,7 @@ static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
        struct net_device *dev = q->dev_queue->dev;
        int err;
 
+       down_write(&block->cb_lock);
        if (!dev->netdev_ops->ndo_setup_tc)
                goto no_offload_dev_inc;
 
@@ -669,24 +672,31 @@ static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
         */
        if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
                NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
-               return -EOPNOTSUPP;
+               err = -EOPNOTSUPP;
+               goto err_unlock;
        }
 
        err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
        if (err == -EOPNOTSUPP)
                goto no_offload_dev_inc;
        if (err)
-               return err;
+               goto err_unlock;
 
        tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
+       up_write(&block->cb_lock);
        return 0;
 
 no_offload_dev_inc:
-       if (tcf_block_offload_in_use(block))
-               return -EOPNOTSUPP;
+       if (tcf_block_offload_in_use(block)) {
+               err = -EOPNOTSUPP;
+               goto err_unlock;
+       }
+       err = 0;
        block->nooffloaddevcnt++;
        tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
-       return 0;
+err_unlock:
+       up_write(&block->cb_lock);
+       return err;
 }
 
 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
@@ -695,6 +705,7 @@ static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
        struct net_device *dev = q->dev_queue->dev;
        int err;
 
+       down_write(&block->cb_lock);
        tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
 
        if (!dev->netdev_ops->ndo_setup_tc)
@@ -702,10 +713,12 @@ static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
        err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
        if (err == -EOPNOTSUPP)
                goto no_offload_dev_dec;
+       up_write(&block->cb_lock);
        return;
 
 no_offload_dev_dec:
        WARN_ON(block->nooffloaddevcnt-- == 0);
+       up_write(&block->cb_lock);
 }
 
 static int
@@ -820,6 +833,7 @@ static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
                return ERR_PTR(-ENOMEM);
        }
        mutex_init(&block->lock);
+       init_rwsem(&block->cb_lock);
        flow_block_init(&block->flow_block);
        INIT_LIST_HEAD(&block->chain_list);
        INIT_LIST_HEAD(&block->owner_list);
@@ -1355,6 +1369,8 @@ tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
        struct tcf_proto *tp, *tp_prev;
        int err;
 
+       lockdep_assert_held(&block->cb_lock);
+
        for (chain = __tcf_get_next_chain(block, NULL);
             chain;
             chain_prev = chain,
@@ -1393,6 +1409,8 @@ static int tcf_block_bind(struct tcf_block *block,
        struct flow_block_cb *block_cb, *next;
        int err, i = 0;
 
+       lockdep_assert_held(&block->cb_lock);
+
        list_for_each_entry(block_cb, &bo->cb_list, list) {
                err = tcf_block_playback_offloads(block, block_cb->cb,
                                                  block_cb->cb_priv, true,
@@ -1427,6 +1445,8 @@ static void tcf_block_unbind(struct tcf_block *block,
 {
        struct flow_block_cb *block_cb, *next;
 
+       lockdep_assert_held(&block->cb_lock);
+
        list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
                tcf_block_playback_offloads(block, block_cb->cb,
                                            block_cb->cb_priv, false,
@@ -2987,19 +3007,26 @@ int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
        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)
-               return -EOPNOTSUPP;
+       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)
-                               return err;
+                       if (err_stop) {
+                               ok_count = err;
+                               goto err_unlock;
+                       }
                } else {
                        ok_count++;
                }
        }
+err_unlock:
+       up_read(&block->cb_lock);
        return ok_count;
 }
 EXPORT_SYMBOL(tc_setup_cb_call);