[PATCH] net/9p/usbg: Fix use-after-free on usb9pfs_clear_tx

[PATCH] net/9p/usbg: Fix use-after-free on usb9pfs_clear_tx

[Yizhou Zhao]

When p9_usbg_close() is called, it invokes usb9pfs_clear_tx() but does
not clear usb9pfs->client nor usb9pfs->in_req->context afterwards.
The 9p client core (p9_client_destroy) calls close() and then
kfree(clnt), so after p9_usbg_close returns the client struct is freed.

If the USB gadget is later disabled (e.g., host disconnect),
usb9pfs_disable calls usb9pfs_clear_tx again, which passes the dangling
usb9pfs->client pointer to p9_client_cb. Inside p9_client_cb,
p9_req_put(c, req) dereferences the freed client to access its reqs IDR,
causing a use-after-free.

Additionally, usb9pfs_clear_tx does not clear usb9pfs->in_req->context
after calling p9_client_cb, so a second invocation finds a stale request
pointer and attempts double-completion, further compounding the UAF risk.

Fix by:
1. Clearing usb9pfs->in_req->context to NULL in usb9pfs_clear_tx()
   before checking usb9pfs->client, so that even if the client is gone
   the stale request pointer is consumed and cannot cause
   double-completion.
2. Adding a NULL check on usb9pfs->client in usb9pfs_clear_tx() to skip
   p9_client_cb() when the client has already been destroyed.
3. Setting usb9pfs->client = NULL in p9_usbg_close() after calling
   usb9pfs_clear_tx(), to prevent any subsequent call from dereferencing
   the freed client pointer.

Fixes: a3be076dc174 ("net/9p/usbg: Add new usb gadget function transport")
Reported-by: Yizhou Zhao <zhaoyz24@mails.tsinghua.edu.cn>
Reported-by: Yuxiang Yang <yangyx22@mails.tsinghua.edu.cn>
Reported-by: Ao Wang <wangao@seu.edu.cn>
Reported-by: Xuewei Feng <fengxw06@126.com>
Reported-by: Qi Li <qli01@tsinghua.edu.cn>
Reported-by: Ke Xu <xuke@tsinghua.edu.cn>
Assisted-by: GLM:GLM-5.1
Signed-off-by: Yizhou Zhao <zhaoyz24@mails.tsinghua.edu.cn>
---
diff --git a/net/9p/trans_usbg.c b/net/9p/trans_usbg.c
index 1ce7033..c75de5c 100644
--- a/net/9p/trans_usbg.c
+++ b/net/9p/trans_usbg.c
@@ -435,6 +435,11 @@ static void usb9pfs_clear_tx(struct f_usb9pfs *usb9pfs)
 	if (!req)
 		return;
 
+	usb9pfs->in_req->context = NULL;
+
+	if (!usb9pfs->client)
+		return;
+
 	if (!req->t_err)
 		req->t_err = -ECONNRESET;
 
@@ -457,6 +462,7 @@ static void p9_usbg_close(struct p9_client *client)
 	client->status = Disconnected;
 
 	usb9pfs_clear_tx(usb9pfs);
+	usb9pfs->client = NULL;
 
 	opts = container_of(usb9pfs->function.fi,
 			    struct f_usb9pfs_opts, func_inst);

--
2.43.0

Re: [PATCH] net/9p/usbg: Fix use-after-free on usb9pfs_clear_tx

[Dominique Martinet]

Hi Michael,

I'm honestly not quite sure what to do with this and others UAF usb9pfs
patches -- you acked the first one I replied to, and the tag match is
obviously correct so I picked these two, but I'd like your opinion on
the rest for two reasons:
 - There seem to be plenty of rough edges around usb9pfs, and frankly I
don't consider this a trust boundary (the usecase being developping
embedded devices, you trust the server fully, so whatever happens
in corner cases doesn't matter)
 - You had a rework in progress (before leaving Pengutronix, which I can
 totally understand you don't have time or want to keep carrying going
 forward, but is someone else from Pengutronix taking your place?)

For all I know this is also correct, but I'm not sure slapping a bandaid
is the way to go, so I'd really appreciate if you could have a look at
this[1] and "net/9p/usbg: clear stale request context after abort"[2]
[1] https://lore.kernel.org/r/20260529081026.77732-1-zhaoyz24@mails.tsinghua.edu.cn
[2] https://lore.kernel.org/r/20260802152148.810390-1-marsy12010123@gmail.com

If you tell me to pick them up I will, and if I don't hear back from you
I'll take another look hopefully before the 7.4 deadline, but I honestly
don't have enough time for this so the most likely outcome will be
/dev/null

Thanks (and sorry Yizhou Zhao / Chengfeng Ye)
-- 
Dominique