Netdev List
 help / color / mirror / Atom feed
From: Eric Dumazet <edumazet@kernel.org>
To: "David S . Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
	Willem de Bruijn <willemb@google.com>
Cc: Kuniyuki Iwashima <kuniyu@google.com>,
	Simon Horman <horms@kernel.org>,
	netdev@vger.kernel.org, Eric Dumazet <edumazet@kernel.org>
Subject: [PATCH net-next 1/4] tap: detach queues from the NETDEV_UNREGISTER notifier
Date: Thu, 19 Nov 2026 08:22:55 +0100	[thread overview]
Message-ID: <20261119072258.3157902-2-edumazet@kernel.org> (raw)
In-Reply-To: <20261119072258.3157902-1-edumazet@kernel.org>

tap_del_queues() is currently called from the macvtap and ipvtap
dellink handlers. Two things in this area only work because
tap_open() runs under RTNL, which a following patch removes:

1) Not all unregistrations go through dellink. If register_netdevice()
   fails after the NETDEV_REGISTER notifier created the char device,
   or if macvlan_common_newlink() / ipvlan_link_new() fail after
   register_netdevice(), the device is unregistered without
   tap_del_queues() being called. Nothing can open the char device in
   this window today, because RTNL is held from the minor allocation
   to its release.

2) tap_del_queues() sets tap->numvtaps to MAX_TAP_QUEUES to "guarantee
   that any future tap_set_queue will fail", but since commit
   815f236d6227 ("macvtap: add TUNSETQUEUE ioctl") tap_set_queue()
   tests tap->numqueues, which is zero at this point. A later
   tap_set_queue() would succeed and write to tap->taps[MAX_TAP_QUEUES],
   one element past the array. Again, this can not happen today since
   the minor is released in the same RTNL section.

Call tap_del_queues() from the NETDEV_UNREGISTER notifier instead, so
that every unregistration path is covered. The notifier is also invoked
when the device moves to another netns, in which case the queues must
stay attached: only act when the device is really being unregistered.

Replace the numvtaps trick with an explicit tap->dead flag, tested in
tap_set_queue().

No functional change intended: none of the above is reachable as long
as tap_open() holds RTNL. These issues were found by code inspection,
with the help of an AI coding assistant, while looking at RTNL removal
in tap. They were not triggered.

Signed-off-by: Eric Dumazet <edumazet@kernel.org>
---
 drivers/net/ipvlan/ipvtap.c | 8 +++++---
 drivers/net/macvtap.c       | 8 +++++---
 drivers/net/tap.c           | 5 ++++-
 include/linux/if_tap.h      | 2 ++
 4 files changed, 16 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ipvlan/ipvtap.c b/drivers/net/ipvlan/ipvtap.c
index 66c949d9426103edea10b2580f096f399f139f4f..d17daa1040967a1d80cede5d740d957204fe1186 100644
--- a/drivers/net/ipvlan/ipvtap.c
+++ b/drivers/net/ipvlan/ipvtap.c
@@ -112,10 +112,7 @@ static int ipvtap_newlink(struct net_device *dev,
 static void __ipvtap_dellink(struct net *net, struct net_device *dev,
 			     struct list_head *head)
 {
-	struct ipvtap_dev *vlantap = netdev_priv(dev);
-
 	netdev_rx_handler_unregister(dev);
-	tap_del_queues(&vlantap->tap);
 	__ipvlan_link_delete(net, dev, head);
 }
 
@@ -185,6 +182,11 @@ static int ipvtap_device_event(struct notifier_block *unused,
 			return notifier_from_errno(err);
 		break;
 	case NETDEV_UNREGISTER:
+		/* NETDEV_UNREGISTER is also sent when the device moves to
+		 * another netns, queues must survive this.
+		 */
+		if (dev->reg_state == NETREG_UNREGISTERING)
+			tap_del_queues(&vlantap->tap);
 		/* vlan->minor == 0 if NETDEV_REGISTER above failed */
 		if (vlantap->tap.minor == 0)
 			break;
diff --git a/drivers/net/macvtap.c b/drivers/net/macvtap.c
index cc975dfb738051cd43d70e6b95d7499400e345aa..3e88ca35cff2f9c755f8aff694e1c4ca83566741 100644
--- a/drivers/net/macvtap.c
+++ b/drivers/net/macvtap.c
@@ -120,10 +120,7 @@ static int macvtap_newlink(struct net_device *dev,
 static void macvtap_dellink(struct net_device *dev,
 			    struct list_head *head)
 {
-	struct macvtap_dev *vlantap = netdev_priv(dev);
-
 	netdev_rx_handler_unregister(dev);
-	tap_del_queues(&vlantap->tap);
 	macvlan_dellink(dev, head);
 }
 
@@ -186,6 +183,11 @@ static int macvtap_device_event(struct notifier_block *unused,
 			return notifier_from_errno(err);
 		break;
 	case NETDEV_UNREGISTER:
+		/* NETDEV_UNREGISTER is also sent when the device moves to
+		 * another netns, queues must survive this.
+		 */
+		if (dev->reg_state == NETREG_UNREGISTERING)
+			tap_del_queues(&vlantap->tap);
 		/* vlan->minor == 0 if NETDEV_REGISTER above failed */
 		if (vlantap->tap.minor == 0)
 			break;
diff --git a/drivers/net/tap.c b/drivers/net/tap.c
index 0f3ad8fd6ab66003093d2aedd4d7e7a6c4f882e9..ded0be1dfdd1e3123acb08e75aa9a874c9e949b5 100644
--- a/drivers/net/tap.c
+++ b/drivers/net/tap.c
@@ -102,6 +102,9 @@ static int tap_enable_queue(struct tap_dev *tap, struct file *file,
 static int tap_set_queue(struct tap_dev *tap, struct file *file,
 			 struct tap_queue *q)
 {
+	if (tap->dead)
+		return -ENODEV;
+
 	if (tap->numqueues == MAX_TAP_QUEUES)
 		return -EBUSY;
 
@@ -248,7 +251,7 @@ void tap_del_queues(struct tap_dev *tap)
 	BUG_ON(tap->numvtaps);
 	BUG_ON(tap->numqueues);
 	/* guarantee that any future tap_set_queue will fail */
-	tap->numvtaps = MAX_TAP_QUEUES;
+	tap->dead = true;
 }
 EXPORT_SYMBOL_GPL(tap_del_queues);
 
diff --git a/include/linux/if_tap.h b/include/linux/if_tap.h
index 553552fa635c3e1e53d1a63c203d32e4c4fd5a4f..e9c4c9272585cdb14cf0ea7ea7d96acc3e8964b7 100644
--- a/include/linux/if_tap.h
+++ b/include/linux/if_tap.h
@@ -43,6 +43,8 @@ struct tap_dev {
 	int			numqueues;
 	netdev_features_t	tap_features;
 	int			minor;
+	/* Set once the net_device is unregistered, no queue can be added. */
+	bool			dead;
 
 	void (*update_features)(struct tap_dev *tap, netdev_features_t features);
 	void (*count_tx_dropped)(struct tap_dev *tap);
-- 
2.53.0


  reply	other threads:[~2026-11-19  7:23 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-19  7:22 [PATCH net-next 0/4] tap: do not hold RTNL in open(), release() and queue ioctls Eric Dumazet
2026-11-19  7:22 ` Eric Dumazet [this message]
2026-11-19  7:22 ` [PATCH net-next 2/4] tap: add a mutex to protect queue management Eric Dumazet
2026-11-19  7:22 ` [PATCH net-next 3/4] tap: do not hold RTNL in open(), release() and queue ioctls Eric Dumazet
2026-11-19  7:22 ` [PATCH net-next 4/4] tap: remove tap_get_tap_dev() and tap_put_tap_dev() Eric Dumazet

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261119072258.3157902-2-edumazet@kernel.org \
    --to=edumazet@kernel.org \
    --cc=davem@davemloft.net \
    --cc=horms@kernel.org \
    --cc=kuba@kernel.org \
    --cc=kuniyu@google.com \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=willemb@google.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox