> [PATCH v2] net/unix: pass pidfd flags via SCM_PIDFD cmsg Please specify the target tree; net for fixes, net-next for others. https://www.kernel.org/doc/html/v6.11/process/maintainer-netdev.html [PATCH net-next v3] af_unix: pass ... From: Stas Sergeev <stsp2@xxxxxxxxx> Date: Thu, 14 Nov 2024 12:19:09 +0300 > Currently SCM_PIDFD cmsg cannot be sent via unix socket > (returns -EINVAL) and SO_PASSPIDFD doesn't support flags. > The created pidfd always has flags set to 0. > > This patch implements SCM_PIDFD cmsg in AF_UNIX socket, which > can be used to send flags to SO_PASSPIDFD-enabled recipient. > > Self-test is added for the propagation of PIDFD_NONBLOCK flag. > > This is mainly needed for the future extensions, like eg this one: > https://lore.kernel.org/lkml/8288a08e-448b-43c2-82dc-59f87d0d9072@xxxxxxxxx/T/#me1237e46deba8574b77834b7704e63559ffef9cb > where it was suggested to try solving the supplementary groups > problem with pidfd. > > Changes in v2: remove flags validation in scm_pidfd_recv(), as > suggested by Kuniyuki Iwashima <kuniyu@xxxxxxxxxx> You can put this changelog and the following CC: under '---' so that they will disappear during merge. > > Signed-off-by: Stas Sergeev <stsp2@xxxxxxxxx> > > CC: "David S. Miller" <davem@xxxxxxxxxxxxx> > CC: Eric Dumazet <edumazet@xxxxxxxxxx> > CC: Jakub Kicinski <kuba@xxxxxxxxxx> > CC: Paolo Abeni <pabeni@xxxxxxxxxx> > CC: Simon Horman <horms@xxxxxxxxxx> > CC: Shuah Khan <shuah@xxxxxxxxxx> > CC: Christian Brauner <brauner@xxxxxxxxxx> > CC: Jens Axboe <axboe@xxxxxxxxx> > CC: Willem de Bruijn <willemb@xxxxxxxxxx> > CC: Pavel Begunkov <asml.silence@xxxxxxxxx> > CC: Gabriel Krisman Bertazi <krisman@xxxxxxx> > CC: Mina Almasry <almasrymina@xxxxxxxxxx> > CC: Oleg Nesterov <oleg@xxxxxxxxxx> > CC: Tycho Andersen <tandersen@xxxxxxxxxxx> > CC: Al Viro <viro@xxxxxxxxxxxxxxxxxx> > CC: Kuniyuki Iwashima <kuniyu@xxxxxxxxxx> > CC: Gou Hao <gouhao@xxxxxxxxxxxxx> > CC: Abhishek Chauhan <quic_abchauha@xxxxxxxxxxx> > CC: Michal Luczaj <mhal@xxxxxxx> > CC: Kees Cook <kees@xxxxxxxxxx> > CC: Aleksa Sarai <cyphar@xxxxxxxxxx> > CC: linux-kernel@xxxxxxxxxxxxxxx > CC: netdev@xxxxxxxxxxxxxxx > CC: linux-kselftest@xxxxxxxxxxxxxxx > --- ^^^ Here > include/linux/pidfs.h | 9 +++ > include/linux/socket.h | 2 +- > include/net/af_unix.h | 1 + > include/net/scm.h | 3 +- > kernel/pid.c | 6 +- > net/core/scm.c | 14 ++++ > net/core/sock.c | 1 + > net/unix/af_unix.c | 3 + > .../testing/selftests/net/af_unix/scm_pidfd.c | 70 +++++++++++++++++-- > 9 files changed, 99 insertions(+), 10 deletions(-) > > diff --git a/include/linux/pidfs.h b/include/linux/pidfs.h > index 75bdf9807802..c4c5c1a0c2ad 100644 > --- a/include/linux/pidfs.h > +++ b/include/linux/pidfs.h > @@ -2,7 +2,16 @@ > #ifndef _LINUX_PID_FS_H > #define _LINUX_PID_FS_H > > +#include <uapi/linux/pidfd.h> > + > struct file *pidfs_alloc_file(struct pid *pid, unsigned int flags); > void __init pidfs_init(void); > > +static inline int pidfd_validate_flags(unsigned int flags) > +{ > + if (flags & ~(PIDFD_NONBLOCK | PIDFD_THREAD)) > + return -EINVAL; > + return 0; > +} > + > #endif /* _LINUX_PID_FS_H */ > diff --git a/include/linux/socket.h b/include/linux/socket.h > index d18cc47e89bd..ee27d391e5aa 100644 > --- a/include/linux/socket.h > +++ b/include/linux/socket.h > @@ -178,7 +178,7 @@ static inline size_t msg_data_left(struct msghdr *msg) > #define SCM_RIGHTS 0x01 /* rw: access rights (array of int) */ > #define SCM_CREDENTIALS 0x02 /* rw: struct ucred */ > #define SCM_SECURITY 0x03 /* rw: security label */ > -#define SCM_PIDFD 0x04 /* ro: pidfd (int) */ > +#define SCM_PIDFD 0x04 /* r: pidfd, w: pidfd_flags (int) */ > > struct ucred { > __u32 pid; > diff --git a/include/net/af_unix.h b/include/net/af_unix.h > index 63129c79b8cb..4bc197548c2f 100644 > --- a/include/net/af_unix.h > +++ b/include/net/af_unix.h > @@ -62,6 +62,7 @@ struct unix_skb_parms { > #ifdef CONFIG_SECURITY_NETWORK > u32 secid; /* Security ID */ > #endif > + u32 pidfd_flags; > u32 consumed; > } __randomize_layout; > > diff --git a/include/net/scm.h b/include/net/scm.h > index 0d35c7c77a74..1326edcacacb 100644 > --- a/include/net/scm.h > +++ b/include/net/scm.h > @@ -48,6 +48,7 @@ struct scm_cookie { > #ifdef CONFIG_SECURITY_NETWORK > u32 secid; /* Passed security ID */ > #endif > + u32 pidfd_flags; Now we consume 40 byes of cb[48]. If we need more storage in the future, we may want to save converted flags in __scm_send() and restore that in scm_pidfd_recv(). No need to do so now, just a note. > }; > > void scm_detach_fds(struct msghdr *msg, struct scm_cookie *scm); > @@ -154,7 +155,7 @@ static __inline__ void scm_pidfd_recv(struct msghdr *msg, struct scm_cookie *scm > if (!scm->pid) > return; > > - pidfd = pidfd_prepare(scm->pid, 0, &pidfd_file); > + pidfd = pidfd_prepare(scm->pid, scm->pidfd_flags, &pidfd_file); > > if (put_cmsg(msg, SOL_SOCKET, SCM_PIDFD, sizeof(int), &pidfd)) { > if (pidfd_file) { > diff --git a/kernel/pid.c b/kernel/pid.c > index 2715afb77eab..b1100ae8ea63 100644 > --- a/kernel/pid.c > +++ b/kernel/pid.c > @@ -629,10 +629,12 @@ static int pidfd_create(struct pid *pid, unsigned int flags) > SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags) > { > int fd; > + int err; > struct pid *p; > > - if (flags & ~(PIDFD_NONBLOCK | PIDFD_THREAD)) > - return -EINVAL; > + err = pidfd_validate_flags(flags); > + if (err) > + return err; > > if (pid <= 0) > return -EINVAL; > diff --git a/net/core/scm.c b/net/core/scm.c > index 4f6a14babe5a..3bcdecdacd7e 100644 > --- a/net/core/scm.c > +++ b/net/core/scm.c > @@ -23,6 +23,7 @@ > #include <linux/security.h> > #include <linux/pid_namespace.h> > #include <linux/pid.h> > +#include <linux/pidfs.h> > #include <linux/nsproxy.h> > #include <linux/slab.h> > #include <linux/errqueue.h> > @@ -210,6 +211,19 @@ int __scm_send(struct socket *sock, struct msghdr *msg, struct scm_cookie *p) > p->creds.gid = gid; > break; > } > + case SCM_PIDFD: > + { > + unsigned int flags; > + > + if (cmsg->cmsg_len != CMSG_LEN(sizeof(flags))) > + goto error; > + memcpy(&flags, CMSG_DATA(cmsg), sizeof(flags)); > + err = pidfd_validate_flags(flags); > + if (err) > + goto error; > + p->pidfd_flags = flags; > + break; > + } Now this allows sending pidfd without SO_PASSPIDFD, so you need to add a validation for "if (!msg->msg_control)" in __scm_recv_common(). > default: > goto error; > } > diff --git a/net/core/sock.c b/net/core/sock.c > index 039be95c40cf..d1fce437c035 100644 > --- a/net/core/sock.c > +++ b/net/core/sock.c > @@ -2930,6 +2930,7 @@ int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, > /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ > case SCM_RIGHTS: > case SCM_CREDENTIALS: > + case SCM_PIDFD: > break; > default: > return -EINVAL; > diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c > index 001ccc55ef0f..8b19dfec0221 100644 > --- a/net/unix/af_unix.c > +++ b/net/unix/af_unix.c > @@ -1892,6 +1892,7 @@ static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool sen > UNIXCB(skb).uid = scm->creds.uid; > UNIXCB(skb).gid = scm->creds.gid; > UNIXCB(skb).fp = NULL; > + UNIXCB(skb).pidfd_flags = scm->pidfd_flags; > unix_get_secdata(scm, skb); > if (scm->fp && send_fds) > err = unix_attach_fds(scm, skb); > @@ -2486,6 +2487,7 @@ int __unix_dgram_recvmsg(struct sock *sk, struct msghdr *msg, size_t size, > memset(&scm, 0, sizeof(scm)); > > scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid); > + scm.pidfd_flags = UNIXCB(skb).pidfd_flags; > unix_set_secdata(&scm, skb); > > if (!(flags & MSG_PEEK)) { > @@ -2873,6 +2875,7 @@ static int unix_stream_read_generic(struct unix_stream_read_state *state, > test_bit(SOCK_PASSPIDFD, &sock->flags)) { > /* Copy credentials */ > scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid); > + scm.pidfd_flags = UNIXCB(skb).pidfd_flags; > unix_set_secdata(&scm, skb); > check_creds = true; > }
