From: Jeff Layton <jlayton@kernel.org>
Date: Fri, 5 Aug 2022 06:42:45 -0400
Subject: fscache: don't leak cookie access refs if invalidation is in progress
or failed
Git-commit: fb24771faf72a2fd62b3b6287af3c610c3ec9cf1
Patch-mainline: v6.0-rc1
References: jsc#SES-1880
It's possible for a request to invalidate a fscache_cookie will come in
while we're already processing an invalidation. If that happens we
currently take an extra access reference that will leak. Only call
__fscache_begin_cookie_access if the FSCACHE_COOKIE_DO_INVALIDATE bit
was previously clear.
Also, ensure that we attempt to clear the bit when the cookie is
"FAILED" and put the reference to avoid an access leak.
Fixes: 85e4ea1049c7 ("fscache: Fix invalidation/lookup race")
Suggested-by: David Howells <dhowells@redhat.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Luis Henriques <lhenriques@suse.com>
---
fs/fscache/cookie.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/fs/fscache/cookie.c b/fs/fscache/cookie.c
index 74920826d8f6..26a6d395737a 100644
--- a/fs/fscache/cookie.c
+++ b/fs/fscache/cookie.c
@@ -739,6 +739,9 @@ static void fscache_cookie_state_machine(struct fscache_cookie *cookie)
fallthrough;
case FSCACHE_COOKIE_STATE_FAILED:
+ if (test_and_clear_bit(FSCACHE_COOKIE_DO_INVALIDATE, &cookie->flags))
+ fscache_end_cookie_access(cookie, fscache_access_invalidate_cookie_end);
+
if (atomic_read(&cookie->n_accesses) != 0)
break;
if (test_bit(FSCACHE_COOKIE_DO_RELINQUISH, &cookie->flags)) {
@@ -1063,8 +1066,8 @@ void __fscache_invalidate(struct fscache_cookie *cookie,
return;
case FSCACHE_COOKIE_STATE_LOOKING_UP:
- __fscache_begin_cookie_access(cookie, fscache_access_invalidate_cookie);
- set_bit(FSCACHE_COOKIE_DO_INVALIDATE, &cookie->flags);
+ if (!test_and_set_bit(FSCACHE_COOKIE_DO_INVALIDATE, &cookie->flags))
+ __fscache_begin_cookie_access(cookie, fscache_access_invalidate_cookie);
fallthrough;
case FSCACHE_COOKIE_STATE_CREATING:
spin_unlock(&cookie->lock);