From: Ido Schimmel

Sent: 27 July 2020 13:15
On Thu, Jul 23, 2020 at 08:09:01AM +0200, Christoph Hellwig wrote:

quoted

Pass a sockptr_t to prepare for set_fs-less handling of the kernel
pointer from bpf-cgroup.

Note that the get case is pretty weird in that it actually copies data
back to userspace from setsockopt.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 include/net/ipv6.h       |  2 +-
 net/ipv6/ip6_flowlabel.c | 16 +++++++++-------
 net/ipv6/ipv6_sockglue.c |  2 +-
 3 files changed, 11 insertions(+), 9 deletions(-)

diff --git a/include/net/ipv6.h b/include/net/ipv6.h
index 262fc88dbd7e2f..4c9d89b5d73268 100644
--- a/include/net/ipv6.h
+++ b/include/net/ipv6.h

@@ -406,7 +406,7 @@ struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
 					 struct ip6_flowlabel *fl,
 					 struct ipv6_txoptions *fopt);
 void fl6_free_socklist(struct sock *sk);
-int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen);
+int ipv6_flowlabel_opt(struct sock *sk, sockptr_t optval, int optlen);
 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
 			   int flags);
 int ip6_flowlabel_init(void);

diff --git a/net/ipv6/ip6_flowlabel.c b/net/ipv6/ip6_flowlabel.c
index 27ee6de9beffc4..6b3c315f3d461a 100644
--- a/net/ipv6/ip6_flowlabel.c
+++ b/net/ipv6/ip6_flowlabel.c

@@ -371,7 +371,7 @@ static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned lo

 static struct ip6_flowlabel *
 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
-	  char __user *optval, int optlen, int *err_p)
+	  sockptr_t optval, int optlen, int *err_p)
 {
 	struct ip6_flowlabel *fl = NULL;
 	int olen;

@@ -401,7 +401,8 @@ fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
 		memset(fl->opt, 0, sizeof(*fl->opt));
 		fl->opt->tot_len = sizeof(*fl->opt) + olen;
 		err = -EFAULT;
-		if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
+		sockptr_advance(optval, CMSG_ALIGN(sizeof(*freq)));
+		if (copy_from_sockptr(fl->opt + 1, optval, olen))
 			goto done;

 		msg.msg_controllen = olen;

@@ -604,7 +605,7 @@ static int ipv6_flowlabel_renew(struct sock *sk, struct in6_flowlabel_req *freq)
 }

 static int ipv6_flowlabel_get(struct sock *sk, struct in6_flowlabel_req *freq,
-		void __user *optval, int optlen)
+		sockptr_t optval, int optlen)
 {
 	struct ipv6_fl_socklist *sfl, *sfl1 = NULL;
 	struct ip6_flowlabel *fl, *fl1 = NULL;

@@ -702,8 +703,9 @@ static int ipv6_flowlabel_get(struct sock *sk, struct in6_flowlabel_req *freq,
 		goto recheck;

 	if (!freq->flr_label) {
-		if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
-				 &fl->label, sizeof(fl->label))) {
+		sockptr_advance(optval,
+				offsetof(struct in6_flowlabel_req, flr_label));

Christoph,

I see a regression with IPv6 flowlabel that I bisected to this patch.
When passing '-F 0' to 'ping' the flow label should be random, yet it's
the same every time after this patch.

It seems that the pointer is never advanced after the call to
sockptr_advance() because it is passed by value and not by reference.
Even if you were to pass it by reference I think you would later need to
call sockptr_decrease() or something similar. Otherwise it is very
error-prone.

Depending on the other checks you may also be able to cross from
user addresses to kernel ones.
At the minimum sockptr_advance() has to fail if the boundary
would be crossed.

Maybe adding an offset to copy_to_sockptr() and copy_from_sockptr() is
better?

The 'is this a kernel or user copy' needs to use the base
address from the system call.
So you do need the offset passed in to copy_to/from_sockptr().

Clearly churn can be reduced by using a #define or static inline
for the common case.

The alternative is to pass a 'fat pointer' through than can
contain an offset as well as the user/kernel bases and
expected length.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)