]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
net: ethtool: take rss_lock for all rxfh changes
authorJakub Kicinski <kuba@kernel.org>
Thu, 26 Jun 2025 20:28:46 +0000 (13:28 -0700)
committerJakub Kicinski <kuba@kernel.org>
Mon, 30 Jun 2025 15:41:24 +0000 (08:41 -0700)
Always take the rss_lock in ethtool_set_rxfh(). We will want to
make a similar change in ethtool_set_rxfh_fields() and some
drivers lock that callback regardless of rss context ID being set.
Having some callbacks locked unconditionally and some only if
context ID is set would be very confusing.

ethtool handling is under rtnl_lock, so rss_lock is very unlikely
to ever be congested.

Link: https://patch.msgid.link/20250626202848.104457-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/ethtool/ioctl.c

index c34bac7bffd8cbf5fb7608ee58bc19d12a3c6543..ce7d720b3c791bc42f524760006fb59778c10300 100644 (file)
@@ -1334,9 +1334,11 @@ static noinline_for_stack int ethtool_set_rxfh_indir(struct net_device *dev,
        }
 
        rxfh_dev.hfunc = ETH_RSS_HASH_NO_CHANGE;
+
+       mutex_lock(&dev->ethtool->rss_lock);
        ret = ops->set_rxfh(dev, &rxfh_dev, extack);
        if (ret)
-               goto out;
+               goto out_unlock;
 
        /* indicate whether rxfh was set to default */
        if (user_size == 0)
@@ -1344,6 +1346,8 @@ static noinline_for_stack int ethtool_set_rxfh_indir(struct net_device *dev,
        else
                dev->priv_flags |= IFF_RXFH_CONFIGURED;
 
+out_unlock:
+       mutex_unlock(&dev->ethtool->rss_lock);
 out:
        kfree(rxfh_dev.indir);
        return ret;
@@ -1500,7 +1504,6 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
        struct netlink_ext_ack *extack = NULL;
        struct ethtool_rxnfc rx_rings;
        struct ethtool_rxfh rxfh;
-       bool locked = false; /* dev->ethtool->rss_lock taken */
        bool create = false;
        bool mod = false;
        u8 *rss_config;
@@ -1570,7 +1573,7 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
        rx_rings.cmd = ETHTOOL_GRXRINGS;
        ret = ops->get_rxnfc(dev, &rx_rings, NULL);
        if (ret)
-               goto out;
+               goto out_free;
 
        /* rxfh.indir_size == 0 means reset the indir table to default (master
         * context) or delete the context (other RSS contexts).
@@ -1586,7 +1589,7 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                                                  &rx_rings,
                                                  rxfh.indir_size);
                if (ret)
-                       goto out;
+                       goto out_free;
        } else if (rxfh.indir_size == 0) {
                if (rxfh.rss_context == 0) {
                        u32 *indir;
@@ -1608,30 +1611,27 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                                   useraddr + rss_cfg_offset + user_indir_len,
                                   rxfh.key_size)) {
                        ret = -EFAULT;
-                       goto out;
+                       goto out_free;
                }
        }
 
-       if (rxfh.rss_context) {
-               mutex_lock(&dev->ethtool->rss_lock);
-               locked = true;
-       }
+       mutex_lock(&dev->ethtool->rss_lock);
 
        if (rxfh.rss_context && rxfh_dev.rss_delete) {
                ret = ethtool_check_rss_ctx_busy(dev, rxfh.rss_context);
                if (ret)
-                       goto out;
+                       goto out_unlock;
        }
 
        if (create) {
                if (rxfh_dev.rss_delete) {
                        ret = -EINVAL;
-                       goto out;
+                       goto out_unlock;
                }
                ctx = ethtool_rxfh_ctx_alloc(ops, dev_indir_size, dev_key_size);
                if (!ctx) {
                        ret = -ENOMEM;
-                       goto out;
+                       goto out_unlock;
                }
 
                if (ops->create_rxfh_context) {
@@ -1644,7 +1644,7 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                                       GFP_KERNEL_ACCOUNT);
                        if (ret < 0) {
                                kfree(ctx);
-                               goto out;
+                               goto out_unlock;
                        }
                        WARN_ON(!ctx_id); /* can't happen */
                        rxfh.rss_context = ctx_id;
@@ -1653,7 +1653,7 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                ctx = xa_load(&dev->ethtool->rss_ctx, rxfh.rss_context);
                if (!ctx) {
                        ret = -ENOENT;
-                       goto out;
+                       goto out_unlock;
                }
        }
        rxfh_dev.hfunc = rxfh.hfunc;
@@ -1687,7 +1687,7 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                                xa_erase(&dev->ethtool->rss_ctx, rxfh.rss_context);
                        kfree(ctx);
                }
-               goto out;
+               goto out_unlock;
        }
        mod = !create && !rxfh_dev.rss_delete;
 
@@ -1708,13 +1708,13 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                if (WARN_ON(xa_load(&dev->ethtool->rss_ctx, rxfh_dev.rss_context))) {
                        /* context ID reused, our tracking is screwed */
                        kfree(ctx);
-                       goto out;
+                       goto out_unlock;
                }
                /* Allocate the exact ID the driver gave us */
                if (xa_is_err(xa_store(&dev->ethtool->rss_ctx, rxfh_dev.rss_context,
                                       ctx, GFP_KERNEL))) {
                        kfree(ctx);
-                       goto out;
+                       goto out_unlock;
                }
 
                /* Fetch the defaults for the old API, in the new API drivers
@@ -1730,7 +1730,7 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                if (WARN_ON(ret)) {
                        xa_erase(&dev->ethtool->rss_ctx, rxfh.rss_context);
                        kfree(ctx);
-                       goto out;
+                       goto out_unlock;
                }
        }
        if (rxfh_dev.rss_delete) {
@@ -1755,9 +1755,9 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev,
                        ctx->input_xfrm = rxfh_dev.input_xfrm;
        }
 
-out:
-       if (locked)
-               mutex_unlock(&dev->ethtool->rss_lock);
+out_unlock:
+       mutex_unlock(&dev->ethtool->rss_lock);
+out_free:
        kfree(rss_config);
        if (mod)
                ethtool_rss_notify(dev, rxfh.rss_context);