Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Ludvig Michaelsson via B4 Relay 
        <devnull+ludvig.michaelsson.yubico.com@kernel.org>
Date:   Wed, 21 Jun 2023 13:17:43 +0200
Subject: [PATCH] HID: hidraw: fix data race on device refcount
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
Message-Id: <20230621-hidraw-race-v1-1-a58e6ac69bab@yubico.com>
To:     Jiri Kosina <jikos@kernel.org>,
        Benjamin Tissoires <benjamin.tissoires@redhat.com>,
        =?utf-8?q?Andr=C3=A9_Almeida?= <andrealmeid@igalia.com>
Cc:     linux-input@vger.kernel.org, linux-kernel@vger.kernel.org,
        Ludvig Michaelsson <ludvig.michaelsson@yubico.com>
Reply-To: <ludvig.michaelsson@yubico.com>
Precedence: bulk

From: Ludvig Michaelsson <ludvig.michaelsson@yubico.com>

The hidraw_open() function increments the hidraw device reference
counter. The counter has no dedicated synchronization mechanism,
resulting in a potential data race when concurrently opening a device.

The race is a regression introduced by commit 8590222e4b02 ("HID:
hidraw: Replace hidraw device table mutex with a rwsem"). While
minors_rwsem is intended to protect the hidraw_table itself, by instead
acquiring the lock for writing, the reference counter is also protected.
This is symmetrical to hidraw_release().

Link: https://github.com/systemd/systemd/issues/27947
Fixes: 8590222e4b02 ("HID: hidraw: Replace hidraw device table mutex with a rwsem")
Signed-off-by: Ludvig Michaelsson <ludvig.michaelsson@yubico.com>
---
 drivers/hid/hidraw.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/hid/hidraw.c b/drivers/hid/hidraw.c
index 93e62b161501..e63c56a0d57f 100644
--- a/drivers/hid/hidraw.c
+++ b/drivers/hid/hidraw.c
@@ -272,7 +272,12 @@ static int hidraw_open(struct inode *inode, struct file *file)
 		goto out;
 	}
 
-	down_read(&minors_rwsem);
+	/*
+	 * Technically not writing to the hidraw_table but a write lock is
+	 * required to protect the device refcount. This is symmetrical to
+	 * hidraw_release().
+	 */
+	down_write(&minors_rwsem);
 	if (!hidraw_table[minor] || !hidraw_table[minor]->exist) {
 		err = -ENODEV;
 		goto out_unlock;
@@ -301,7 +306,7 @@ static int hidraw_open(struct inode *inode, struct file *file)
 	spin_unlock_irqrestore(&hidraw_table[minor]->list_lock, flags);
 	file->private_data = list;
 out_unlock:
-	up_read(&minors_rwsem);
+	up_write(&minors_rwsem);
 out:
 	if (err < 0)
 		kfree(list);

---
base-commit: 45a3e24f65e90a047bef86f927ebdc4c710edaa1
change-id: 20230621-hidraw-race-b51b11bf11ed

Best regards,
-- 
Ludvig Michaelsson <ludvig.michaelsson@yubico.com>