The motivation for this is to remove the SOL_SOCKET limitation from io_uring_cmd_getsockopt().
The reason for this limitation is that io_uring_cmd_getsockopt() passes a kernel pointer. The first idea would be to change the optval and optlen arguments to the protocol specific hooks also to sockptr_t, as that is already used for setsockopt() and also by do_sock_getsockopt() sk_getsockopt() and BPF_CGROUP_RUN_PROG_GETSOCKOPT(). But as Linus don't like 'sockptr_t' I used a different approach. Instead of passing the optlen as user or kernel pointer, we only ever pass a kernel pointer and do the translation from/to userspace in do_sock_getsockopt(). The simple solution would be to just remove the '__user' from the int *optlen argument, but it seems the compiler doesn't complain about '__user' vs. without it, so instead I used a helper struct in order to make sure everything compiles with a typesafe change. That together with get_optlen() and put_optlen() helper macros make it relatively easy to review and check the behaviour is most likely unchanged. In order to avoid uapi changes regarding different error code orders regarding -EFAULT, the real -EFAULT handling is deferred to get_optlen() and put_optlen(). This allows io_uring_cmd_getsockopt() to remove the SOL_SOCKET limitation. Removing 'sockptr_t optlen' from existing code is for patch for another day. Link: https://lore.kernel.org/io-uring/86b1dce5-4bb4-4a0b-9cff-e72f488bf...@samba.org/T/#t Cc: Jens Axboe <ax...@kernel.dk> Cc: Pavel Begunkov <asml.sile...@gmail.com> Cc: Breno Leitao <lei...@debian.org> Cc: Linus Torvalds <torva...@linux-foundation.org> Cc: Jakub Kicinski <k...@kernel.org> Cc: Christoph Hellwig <h...@lst.de> Cc: Karsten Keil <i...@linux-pingi.de> Cc: Ayush Sawal <ayush.sa...@chelsio.com> Cc: Andrew Lunn <andrew+net...@lunn.ch> Cc: "David S. Miller" <da...@davemloft.net> Cc: Eric Dumazet <eduma...@google.com> Cc: Paolo Abeni <pab...@redhat.com> Cc: Simon Horman <ho...@kernel.org> Cc: Kuniyuki Iwashima <kun...@amazon.com> Cc: Willem de Bruijn <will...@google.com> Cc: David Ahern <dsah...@kernel.org> Cc: Marcelo Ricardo Leitner <marcelo.leit...@gmail.com> Cc: Xin Long <lucien....@gmail.com> Cc: Neal Cardwell <ncardw...@google.com> Cc: Joerg Reuter <jreu...@yaina.de> Cc: Marcel Holtmann <mar...@holtmann.org> Cc: Johan Hedberg <johan.hedb...@gmail.com> Cc: Luiz Augusto von Dentz <luiz.de...@gmail.com> Cc: Oliver Hartkopp <socket...@hartkopp.net> Cc: Marc Kleine-Budde <m...@pengutronix.de> Cc: Robin van der Gracht <ro...@protonic.nl> Cc: Oleksij Rempel <o.rem...@pengutronix.de> Cc: ker...@pengutronix.de Cc: Alexander Aring <alex.ar...@gmail.com> Cc: Stefan Schmidt <ste...@datenfreihafen.org> Cc: Miquel Raynal <miquel.ray...@bootlin.com> Cc: Alexandra Winter <wint...@linux.ibm.com> Cc: Thorsten Winkler <twink...@linux.ibm.com> Cc: James Chapman <jchap...@katalix.com> Cc: Jeremy Kerr <j...@codeconstruct.com.au> Cc: Matt Johnston <m...@codeconstruct.com.au> Cc: Matthieu Baerts <matt...@kernel.org> Cc: Mat Martineau <martin...@kernel.org> Cc: Geliang Tang <geli...@kernel.org> Cc: Krzysztof Kozlowski <k...@kernel.org> Cc: Remi Denis-Courmont <courmi...@gmail.com> Cc: Allison Henderson <allison.hender...@oracle.com> Cc: David Howells <dhowe...@redhat.com> Cc: Marc Dionne <marc.dio...@auristor.com> Cc: Wenjia Zhang <wen...@linux.ibm.com> Cc: Jan Karcher <j...@linux.ibm.com> Cc: "D. Wythe" <alib...@linux.alibaba.com> Cc: Tony Lu <ton...@linux.alibaba.com> Cc: Wen Gu <gu...@linux.alibaba.com> Cc: Jon Maloy <jma...@redhat.com> Cc: Boris Pismenny <bor...@nvidia.com> Cc: John Fastabend <john.fastab...@gmail.com> Cc: Stefano Garzarella <sgarz...@redhat.com> Cc: Martin Schiller <m...@dev.tdt.de> Cc: "Björn Töpel" <bj...@kernel.org> Cc: Magnus Karlsson <magnus.karls...@intel.com> Cc: Maciej Fijalkowski <maciej.fijalkow...@intel.com> Cc: Jonathan Lemon <jonathan.le...@gmail.com> Cc: Alexei Starovoitov <a...@kernel.org> Cc: Daniel Borkmann <dan...@iogearbox.net> Cc: Jesper Dangaard Brouer <h...@kernel.org> CC: Stefan Metzmacher <me...@samba.org> Cc: net...@vger.kernel.org Cc: linux-ker...@vger.kernel.org Cc: linux-s...@vger.kernel.org Cc: linux-h...@vger.kernel.org Cc: linux-blueto...@vger.kernel.org Cc: linux-...@vger.kernel.org Cc: d...@vger.kernel.org Cc: linux-w...@vger.kernel.org Cc: linux-s...@vger.kernel.org Cc: mp...@lists.linux.dev Cc: linux-r...@vger.kernel.org Cc: rds-de...@oss.oracle.com Cc: linux-...@lists.infradead.org Cc: tipc-discussion@lists.sourceforge.net Cc: virtualizat...@lists.linux.dev Cc: linux-...@vger.kernel.org Cc: b...@vger.kernel.org Cc: isdn4li...@listserv.isdn4linux.de Cc: io-ur...@vger.kernel.org Signed-off-by: Stefan Metzmacher <me...@samba.org> --- include/linux/sockptr.h | 20 +++++++++++++++----- net/socket.c | 31 +++++++++++++++++++++++++++++-- 2 files changed, 44 insertions(+), 7 deletions(-) diff --git a/include/linux/sockptr.h b/include/linux/sockptr.h index 1baf66f26f4f..06ec7fd73028 100644 --- a/include/linux/sockptr.h +++ b/include/linux/sockptr.h @@ -170,20 +170,25 @@ static inline int check_zeroed_sockptr(sockptr_t src, size_t offset, } typedef struct { - int __user *up; + int *kp; } optlen_t; #define __check_optlen_t(__optlen) \ ({ \ optlen_t *__ptr __maybe_unused = &__optlen; \ - BUILD_BUG_ON(sizeof(*((__ptr)->up)) != sizeof(int)); \ + BUILD_BUG_ON(sizeof(*((__ptr)->kp)) != sizeof(int)); \ }) #define get_optlen(__val, __optlen) \ ({ \ long __err; \ __check_optlen_t(__optlen); \ - __err = get_user(__val, __optlen.up); \ + if ((__optlen).kp != NULL) { \ + (__val) = *((__optlen).kp); \ + __err = 0; \ + } else { \ + __err = -EFAULT; \ + } \ __err; \ }) @@ -191,13 +196,18 @@ typedef struct { ({ \ long __err; \ __check_optlen_t(__optlen); \ - __err = put_user(__val, __optlen.up); \ + if ((__optlen).kp != NULL) { \ + *((__optlen).kp) = (__val); \ + __err = 0; \ + } else { \ + __err = -EFAULT; \ + } \ __err; \ }) static inline sockptr_t OPTLEN_SOCKPTR(optlen_t optlen) { - return (sockptr_t) { .user = optlen.up, }; + return (sockptr_t) { .kernel = optlen.kp, .is_kernel = true }; } #endif /* _LINUX_SOCKPTR_H */ diff --git a/net/socket.c b/net/socket.c index fa2de12c10e6..81e5c9767bbc 100644 --- a/net/socket.c +++ b/net/socket.c @@ -2350,15 +2350,42 @@ int do_sock_getsockopt(struct socket *sock, bool compat, int level, } else if (unlikely(!ops->getsockopt)) { err = -EOPNOTSUPP; } else { - optlen_t _optlen = { .up = NULL, }; + optlen_t _optlen = { .kp = NULL, }; + int koptlen; if (WARN_ONCE(optval.is_kernel, "Invalid argument type")) return -EOPNOTSUPP; - _optlen.up = optlen.user; + if (optlen.is_kernel) { + _optlen.kp = optlen.kernel; + } else if (optlen.user != NULL) { + /* + * If optlen.user is NULL, + * we pass _optlen.kp = NULL + * in order to avoid breaking + * any uapi for getsockopt() + * implementations that ignore + * the optlen pointer completely + * or do any level and optname + * checking before hitting a + * potential -EFAULT condition. + * + * Also when optlen.user is not NULL, + * but copy_from_sockptr() causes -EFAULT, + * we'll pass optlen.kp = NULL in order + * to defer a possible -EFAULT return + * to the caller to get_optlen() and put_optlen(). + */ + if (copy_from_sockptr(&koptlen, optlen, sizeof(koptlen)) == 0) + _optlen.kp = &koptlen; + } err = ops->getsockopt(sock, level, optname, optval.user, _optlen); + if (err != -EFAULT && _optlen.kp == &koptlen) { + if (copy_to_sockptr(optlen, &koptlen, sizeof(koptlen))) + return -EFAULT; + } } if (!compat) -- 2.34.1
