From: Christian Brauner <brauner@kernel.org>
Date: Thu, 23 Apr 2026 11:56:09 +0200
Subject: [PATCH] eventpoll: fix ep_remove struct eventpoll / struct file UAF

CVE: CVE-2026-46242

Upstream commit a6dc643c69311677c574a0f17a3f4d66a5f3744b, adapted to the
pre-e9e5cd40d7c4 ("eventpoll: kill __ep_remove()") code shape still used
by this kernel, where the removal path is the monolithic __ep_remove(ep,
epi, force) rather than upstream's ep_remove()/ep_remove_epi()/
ep_remove_file() split.

ep_remove() used epi->ffd.file across the file->f_lock critical section
without holding a reference. A concurrent __fput() taking the
eventpoll_release() fastpath could free the file (and, for the
epoll-watches-epoll case, the watched struct eventpoll) mid-section,
leading to a use-after-free / attacker-controllable kmem_cache_free().

Pin the file with epi_fget() before entering the critical section and
release it afterwards. A failed pin means __fput() is already in flight
and will remove the epi via eventpoll_release_file() (which blocks on
ep->mtx), so it is safe to bail. The pin is taken only on the !force
path: the force path is reached from eventpoll_release_file() itself,
i.e. from within __fput() where f_count has already dropped to zero and
epi_fget() would necessarily fail; there @file is still valid and no
concurrent __fput() can race, so it keeps using epi->ffd.file directly.
A successful pin also makes the in-lock epi->dying recheck redundant; the
cheap lockless READ_ONCE(epi->dying) fast-path bailout is kept.

Fixes: 58c9b016e128 ("epoll: use refcount to reduce ep_mutex contention")
Reported-by: Jaeyoung Chung <jjy600901@snu.ac.kr>
Signed-off-by: Christian Brauner <brauner@kernel.org>
---
diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index f4efb5b..68e58d0 100644
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -821,6 +821,8 @@ static void ep_free(struct eventpoll *ep)
 	kfree(ep);
 }
 
+static struct file *epi_fget(const struct epitem *epi);
+
 /*
  * Removes a "struct epitem" from the eventpoll RB tree and deallocates
  * all the associated resources. Must be called with "mtx" held.
@@ -832,6 +834,7 @@ static void ep_free(struct eventpoll *ep)
 static bool __ep_remove(struct eventpoll *ep, struct epitem *epi, bool force)
 {
 	struct file *file = epi->ffd.file;
+	struct file *to_put = NULL;
 	struct epitems_head *to_free;
 	struct hlist_head *head;
 
@@ -842,13 +845,27 @@ static bool __ep_remove(struct eventpoll *ep, struct epitem *epi, bool force)
 	 */
 	ep_unregister_pollwait(ep, epi);
 
-	/* Remove the current item from the list of epoll hooks */
-	spin_lock(&file->f_lock);
-	if (epi->dying && !force) {
-		spin_unlock(&file->f_lock);
-		return false;
+	if (!force) {
+		/* cheap sync with eventpoll_release_file() */
+		if (unlikely(READ_ONCE(epi->dying)))
+			return false;
+		/*
+		 * Grabbing a reference proves we are not racing
+		 * eventpoll_release_file() and are not going to: a concurrent
+		 * __fput() cannot free @file (nor, for the epoll-on-epoll
+		 * case, the watched eventpoll) across the f_lock section
+		 * below. A failed pin means @file already reached refcount
+		 * zero and __fput() is in flight; it will remove this epi via
+		 * eventpoll_release_file(), so bail out here.
+		 */
+		file = epi_fget(epi);
+		if (!file)
+			return false;
+		to_put = file;
 	}
 
+	/* Remove the current item from the list of epoll hooks */
+	spin_lock(&file->f_lock);
 	to_free = NULL;
 	head = file->f_ep;
 	if (head->first == &epi->fllink && !epi->fllink.next) {
@@ -882,6 +899,8 @@ static bool __ep_remove(struct eventpoll *ep, struct epitem *epi, bool force)
 	kfree_rcu(epi, rcu);
 
 	percpu_counter_dec(&ep->user->epoll_watches);
+	if (to_put)
+		fput(to_put);
 	return true;
 }
 
-- 
2.43.0