Skip to contentSkip to conversation

sockptr_t fixes v2

PATCH In review Christoph Hellwig 3 participants 8 messages Jul 28, 2020 – Jul 29, 2020

Subsystems (10)
lore
Share & tools
lore
#0 lore raw

cover sockptr_t fixes v2

Hi Dave,

a bunch of fixes for the sockptr_t conversion

Changes since v1:
- fix a user pointer dereference braino in bpfilter

#1 lore raw

PATCH 1/4 netfilter: arp_tables: restore a SPDX identifier

This was accidentally removed in an unrelated commit.

Fixes c2f12630c60f ("netfilter: switch nf_setsockopt to sockptr_t")
---
 net/ipv4/netfilter/arp_tables.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/ipv4/netfilter/arp_tables.c +1 -1
index f5b26ef1782001..9a1567dbc022b6 100644
@@ -1,4 +1,4 @@
-
+// SPDX-License-Identifier: GPL-2.0-only
 /*
  * Packet matching code for ARP packets.
  *
Signature · 1 line
2.27.0
#2 lore raw

PATCH 2/4 net: make sockptr_is_null strict aliasing safe

In reply to Christoph Hellwig

While the kernel in general is not strict aliasing safe we can trivially do that in sockptr_is_null without affecting code generation, so always check the actually assigned union member.

---
 include/linux/sockptr.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
include/linux/sockptr.h +3 -1
index 7d5cdb2b30b5f0..b13ea1422f93a5 100644
@@ -64,7 +64,9 @@ static inline int __must_check init_user_sockptr(sockptr_t *sp, void __user *p)
 
 static inline bool sockptr_is_null(sockptr_t sockptr)
 {
-	return !sockptr.user && !sockptr.kernel;
+	if (sockptr_is_kernel(sockptr))
+		return !sockptr.kernel;
+	return !sockptr.user;
 }
 
 static inline int copy_from_sockptr(void *dst, sockptr_t src, size_t size)
Signature · 1 line
2.27.0
#3 lore raw

RE: [PATCH 2/4] net: make sockptr_is_null strict aliasing safe

From: Christoph Hellwig <[email protected]>

Quoted 5 lines

Sent: 28 July 2020 17:39

While the kernel in general is not strict aliasing safe we can trivially do that in sockptr_is_null without affecting code generation, so always check the actually assigned union member.

Even with 'strict aliasing' gcc (at least) guarantees that the members of a union alias each other.
It is about the only way so safely interpret a float as an int.

So when sockptr_t is a union testing either member is enough.

When it is a structure the changed form almost certainly adds code.

David

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

#4 lore raw

RE: [PATCH 2/4] net: make sockptr_is_null strict aliasing safe

Quoted 10 lines · David Laight

From: Christoph Hellwig <[email protected]>

Quoted 5 lines

Sent: 28 July 2020 17:39

While the kernel in general is not strict aliasing safe we can trivially do that in sockptr_is_null without affecting code generation, so always check the actually assigned union member.

Even with 'strict aliasing' gcc (at least) guarantees that the members of a union alias each other.
It is about the only way so safely interpret a float as an int.

The only?

  float given;
  int i;
  memcpy(&i, &given, sizeof(i));
  BUILD_BUG_ON(sizeof(i) > sizeof(given));
#5 lore raw

PATCH 3/4 net: remove sockptr_advance

In reply to Christoph Hellwig

sockptr_advance never properly worked. Replace it with _offset variants of copy_from_sockptr and copy_to_sockptr.

---
 drivers/crypto/chelsio/chtls/chtls_main.c | 12 +++++-----
 include/linux/sockptr.h                   | 27 +++++++++++------------
 net/dccp/proto.c                          |  5 ++---
 net/ipv4/netfilter/arp_tables.c           |  8 +++----
 net/ipv4/netfilter/ip_tables.c            |  8 +++----
 net/ipv4/tcp.c                            |  5 +++--
 net/ipv6/ip6_flowlabel.c                  | 11 ++++-----
 net/ipv6/netfilter/ip6_tables.c           |  8 +++----
 net/netfilter/x_tables.c                  |  7 +++---
 net/tls/tls_main.c                        |  6 ++---
 10 files changed, 49 insertions(+), 48 deletions(-)
drivers/crypto/chelsio/chtls/chtls_main.c +6 -6
index c3058dcdb33c5c..66d247efd5615b 100644
@@ -525,9 +525,9 @@ static int do_chtls_setsockopt(struct sock *sk, int optname,
 		/* Obtain version and type from previous copy */
 		crypto_info[0] = tmp_crypto_info;
 		/* Now copy the following data */
-		sockptr_advance(optval, sizeof(*crypto_info));
-		rc = copy_from_sockptr((char *)crypto_info + sizeof(*crypto_info),
-				optval,
+		rc = copy_from_sockptr_offset((char *)crypto_info +
+				sizeof(*crypto_info),
+				optval, sizeof(*crypto_info),
 				sizeof(struct tls12_crypto_info_aes_gcm_128)
 				- sizeof(*crypto_info));
 
@@ -542,9 +542,9 @@ static int do_chtls_setsockopt(struct sock *sk, int optname,
 	}
 	case TLS_CIPHER_AES_GCM_256: {
 		crypto_info[0] = tmp_crypto_info;
-		sockptr_advance(optval, sizeof(*crypto_info));
-		rc = copy_from_sockptr((char *)crypto_info + sizeof(*crypto_info),
-				    optval,
+		rc = copy_from_sockptr_offset((char *)crypto_info +
+				sizeof(*crypto_info),
+				optval, sizeof(*crypto_info),
 				sizeof(struct tls12_crypto_info_aes_gcm_256)
 				- sizeof(*crypto_info));
 
include/linux/sockptr.h +13 -14
index b13ea1422f93a5..9e6c81d474cba8 100644
@@ -69,19 +69,26 @@ static inline bool sockptr_is_null(sockptr_t sockptr)
 	return !sockptr.user;
 }
 
-static inline int copy_from_sockptr(void *dst, sockptr_t src, size_t size)
+static inline int copy_from_sockptr_offset(void *dst, sockptr_t src,
+		size_t offset, size_t size)
 {
 	if (!sockptr_is_kernel(src))
-		return copy_from_user(dst, src.user, size);
-	memcpy(dst, src.kernel, size);
+		return copy_from_user(dst, src.user + offset, size);
+	memcpy(dst, src.kernel + offset, size);
 	return 0;
 }
 
-static inline int copy_to_sockptr(sockptr_t dst, const void *src, size_t size)
+static inline int copy_from_sockptr(void *dst, sockptr_t src, size_t size)
+{
+	return copy_from_sockptr_offset(dst, src, 0, size);
+}
+
+static inline int copy_to_sockptr_offset(sockptr_t dst, size_t offset,
+		const void *src, size_t size)
 {
 	if (!sockptr_is_kernel(dst))
-		return copy_to_user(dst.user, src, size);
-	memcpy(dst.kernel, src, size);
+		return copy_to_user(dst.user + offset, src, size);
+	memcpy(dst.kernel + offset, src, size);
 	return 0;
 }
 
@@ -112,14 +119,6 @@ static inline void *memdup_sockptr_nul(sockptr_t src, size_t len)
 	return p;
 }
 
-static inline void sockptr_advance(sockptr_t sockptr, size_t len)
-{
-	if (sockptr_is_kernel(sockptr))
-		sockptr.kernel += len;
-	else
-		sockptr.user += len;
-}
-
 static inline long strncpy_from_sockptr(char *dst, sockptr_t src, size_t count)
 {
 	if (sockptr_is_kernel(src)) {
net/dccp/proto.c +2 -3
index 2e9e8449698fb4..d148ab1530e57b 100644
@@ -426,9 +426,8 @@ static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
 			return -ENOMEM;
 
 		sl->dccpsl_nr = optlen / sizeof(u32) - 1;
-		sockptr_advance(optval, sizeof(service));
-		if (copy_from_sockptr(sl->dccpsl_list, optval,
-				      optlen - sizeof(service)) ||
+		if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
+				sizeof(service), optlen - sizeof(service)) ||
 		    dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
 			kfree(sl);
 			return -EFAULT;
net/ipv4/netfilter/arp_tables.c +4 -4
index 9a1567dbc022b6..d1e04d2b5170ec 100644
@@ -971,8 +971,8 @@ static int do_replace(struct net *net, sockptr_t arg, unsigned int len)
 		return -ENOMEM;
 
 	loc_cpu_entry = newinfo->entries;
-	sockptr_advance(arg, sizeof(tmp));
-	if (copy_from_sockptr(loc_cpu_entry, arg, tmp.size) != 0) {
+	if (copy_from_sockptr_offset(loc_cpu_entry, arg, sizeof(tmp),
+			tmp.size) != 0) {
 		ret = -EFAULT;
 		goto free_newinfo;
 	}
@@ -1267,8 +1267,8 @@ static int compat_do_replace(struct net *net, sockptr_t arg, unsigned int len)
 		return -ENOMEM;
 
 	loc_cpu_entry = newinfo->entries;
-	sockptr_advance(arg, sizeof(tmp));
-	if (copy_from_sockptr(loc_cpu_entry, arg, tmp.size) != 0) {
+	if (copy_from_sockptr_offset(loc_cpu_entry, arg, sizeof(tmp),
+			tmp.size) != 0) {
 		ret = -EFAULT;
 		goto free_newinfo;
 	}
net/ipv4/netfilter/ip_tables.c +4 -4
index f2a9680303d8c0..f15bc21d730164 100644
@@ -1126,8 +1126,8 @@ do_replace(struct net *net, sockptr_t arg, unsigned int len)
 		return -ENOMEM;
 
 	loc_cpu_entry = newinfo->entries;
-	sockptr_advance(arg, sizeof(tmp));
-	if (copy_from_sockptr(loc_cpu_entry, arg, tmp.size) != 0) {
+	if (copy_from_sockptr_offset(loc_cpu_entry, arg, sizeof(tmp),
+			tmp.size) != 0) {
 		ret = -EFAULT;
 		goto free_newinfo;
 	}
@@ -1508,8 +1508,8 @@ compat_do_replace(struct net *net, sockptr_t arg, unsigned int len)
 		return -ENOMEM;
 
 	loc_cpu_entry = newinfo->entries;
-	sockptr_advance(arg, sizeof(tmp));
-	if (copy_from_sockptr(loc_cpu_entry, arg, tmp.size) != 0) {
+	if (copy_from_sockptr_offset(loc_cpu_entry, arg, sizeof(tmp),
+			tmp.size) != 0) {
 		ret = -EFAULT;
 		goto free_newinfo;
 	}
net/ipv4/tcp.c +3 -2
index 27de9380ed140e..4afec552f211b9 100644
@@ -2801,12 +2801,13 @@ static int tcp_repair_options_est(struct sock *sk, sockptr_t optbuf,
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct tcp_repair_opt opt;
+	size_t offset = 0;
 
 	while (len >= sizeof(opt)) {
-		if (copy_from_sockptr(&opt, optbuf, sizeof(opt)))
+		if (copy_from_sockptr_offset(&opt, optbuf, offset, sizeof(opt)))
 			return -EFAULT;
 
-		sockptr_advance(optbuf, sizeof(opt));
+		offset += sizeof(opt);
 		len -= sizeof(opt);
 
 		switch (opt.opt_code) {
net/ipv6/ip6_flowlabel.c +6 -5
index 215b6f5e733ec9..2d655260dedc75 100644
@@ -401,8 +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;
-		sockptr_advance(optval, CMSG_ALIGN(sizeof(*freq)));
-		if (copy_from_sockptr(fl->opt + 1, optval, olen))
+		if (copy_from_sockptr_offset(fl->opt + 1, optval,
+				CMSG_ALIGN(sizeof(*freq)), olen))
 			goto done;
 
 		msg.msg_controllen = olen;
@@ -703,9 +703,10 @@ static int ipv6_flowlabel_get(struct sock *sk, struct in6_flowlabel_req *freq,
 		goto recheck;
 
 	if (!freq->flr_label) {
-		sockptr_advance(optval,
-				offsetof(struct in6_flowlabel_req, flr_label));
-		if (copy_to_sockptr(optval, &fl->label, sizeof(fl->label))) {
+		size_t offset = offsetof(struct in6_flowlabel_req, flr_label);
+
+		if (copy_to_sockptr_offset(optval, offset, &fl->label,
+				sizeof(fl->label))) {
 			/* Intentionally ignore fault. */
 		}
 	}
net/ipv6/netfilter/ip6_tables.c +4 -4
index 1d52957a413f4a..2e2119bfcf1373 100644
@@ -1143,8 +1143,8 @@ do_replace(struct net *net, sockptr_t arg, unsigned int len)
 		return -ENOMEM;
 
 	loc_cpu_entry = newinfo->entries;
-	sockptr_advance(arg, sizeof(tmp));
-	if (copy_from_sockptr(loc_cpu_entry, arg, tmp.size) != 0) {
+	if (copy_from_sockptr_offset(loc_cpu_entry, arg, sizeof(tmp),
+			tmp.size) != 0) {
 		ret = -EFAULT;
 		goto free_newinfo;
 	}
@@ -1517,8 +1517,8 @@ compat_do_replace(struct net *net, sockptr_t arg, unsigned int len)
 		return -ENOMEM;
 
 	loc_cpu_entry = newinfo->entries;
-	sockptr_advance(arg, sizeof(tmp));
-	if (copy_from_sockptr(loc_cpu_entry, arg, tmp.size) != 0) {
+	if (copy_from_sockptr_offset(loc_cpu_entry, arg, sizeof(tmp),
+			tmp.size) != 0) {
 		ret = -EFAULT;
 		goto free_newinfo;
 	}
net/netfilter/x_tables.c +4 -3
index b97eb4b538fd4e..91bf6635ea9ee4 100644
@@ -1050,6 +1050,7 @@ EXPORT_SYMBOL_GPL(xt_check_target);
 void *xt_copy_counters(sockptr_t arg, unsigned int len,
 		       struct xt_counters_info *info)
 {
+	size_t offset;
 	void *mem;
 	u64 size;
 
@@ -1067,7 +1068,7 @@ void *xt_copy_counters(sockptr_t arg, unsigned int len,
 
 		memcpy(info->name, compat_tmp.name, sizeof(info->name) - 1);
 		info->num_counters = compat_tmp.num_counters;
-		sockptr_advance(arg, sizeof(compat_tmp));
+		offset = sizeof(compat_tmp);
 	} else
 #endif
 	{
@@ -1078,7 +1079,7 @@ void *xt_copy_counters(sockptr_t arg, unsigned int len,
 		if (copy_from_sockptr(info, arg, sizeof(*info)) != 0)
 			return ERR_PTR(-EFAULT);
 
-		sockptr_advance(arg, sizeof(*info));
+		offset = sizeof(*info);
 	}
 	info->name[sizeof(info->name) - 1] = '\0';
 
@@ -1092,7 +1093,7 @@ void *xt_copy_counters(sockptr_t arg, unsigned int len,
 	if (!mem)
 		return ERR_PTR(-ENOMEM);
 
-	if (copy_from_sockptr(mem, arg, len) == 0)
+	if (copy_from_sockptr_offset(mem, arg, offset, len) == 0)
 		return mem;
 
 	vfree(mem);
net/tls/tls_main.c +3 -3
index d77f7d821130db..bbc52b088d2968 100644
@@ -522,9 +522,9 @@ static int do_tls_setsockopt_conf(struct sock *sk, sockptr_t optval,
 		goto err_crypto_info;
 	}
 
-	sockptr_advance(optval, sizeof(*crypto_info));
-	rc = copy_from_sockptr(crypto_info + 1, optval,
-			       optlen - sizeof(*crypto_info));
+	rc = copy_from_sockptr_offset(crypto_info + 1, optval,
+				      sizeof(*crypto_info),
+				      optlen - sizeof(*crypto_info));
 	if (rc) {
 		rc = -EFAULT;
 		goto err_crypto_info;
Signature · 1 line
2.27.0
#6 lore raw

PATCH 4/4 net: improve the user pointer check in init_user_sockptr

In reply to Christoph Hellwig

Make sure not just the pointer itself but the whole range lies in the user address space. For that pass the length and then use the access_ok helper to do the check.

Fixes 6d04fe15f78a ("net: optimize the sockptr_t for unified kernel/user address spaces")
---
 include/linux/sockptr.h     | 18 ++++++------------
 net/ipv4/bpfilter/sockopt.c |  2 +-
 net/socket.c                |  2 +-
 3 files changed, 8 insertions(+), 14 deletions(-)
include/linux/sockptr.h +6 -12
index 9e6c81d474cba8..96840def9d69cc 100644
@@ -27,14 +27,6 @@ static inline sockptr_t KERNEL_SOCKPTR(void *p)
 {
 	return (sockptr_t) { .kernel = p };
 }
-
-static inline int __must_check init_user_sockptr(sockptr_t *sp, void __user *p)
-{
-	if ((unsigned long)p >= TASK_SIZE)
-		return -EFAULT;
-	sp->user = p;
-	return 0;
-}
 #else /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 typedef struct {
 	union {
@@ -53,14 +45,16 @@ static inline sockptr_t KERNEL_SOCKPTR(void *p)
 {
 	return (sockptr_t) { .kernel = p, .is_kernel = true };
 }
+#endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 
-static inline int __must_check init_user_sockptr(sockptr_t *sp, void __user *p)
+static inline int __must_check init_user_sockptr(sockptr_t *sp, void __user *p,
+		size_t size)
 {
-	sp->user = p;
-	sp->is_kernel = false;
+	if (!access_ok(p, size))
+		return -EFAULT;
+	*sp = (sockptr_t) { .user = p };
 	return 0;
 }
-#endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 
 static inline bool sockptr_is_null(sockptr_t sockptr)
 {
net/ipv4/bpfilter/sockopt.c +1 -1
index 94f18d2352d007..545b2640f0194d 100644
@@ -65,7 +65,7 @@ int bpfilter_ip_get_sockopt(struct sock *sk, int optname,
 
 	if (get_user(len, optlen))
 		return -EFAULT;
-	err = init_user_sockptr(&optval, user_optval);
+	err = init_user_sockptr(&optval, user_optval, len);
 	if (err)
 		return err;
 	return bpfilter_mbox_request(sk, optname, optval, len, false);
net/socket.c +1 -1
index 94ca4547cd7c53..aff52e81653ce3 100644
@@ -2105,7 +2105,7 @@ int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
 	if (optlen < 0)
 		return -EINVAL;
 
-	err = init_user_sockptr(&optval, user_optval);
+	err = init_user_sockptr(&optval, user_optval, optlen);
 	if (err)
 		return err;
 
Signature · 1 line
2.27.0
#7 lore raw

RE: [PATCH 4/4] net: improve the user pointer check in init_user_sockptr

From: Christoph Hellwig

Quoted 5 lines

Sent: 28 July 2020 17:39

Make sure not just the pointer itself but the whole range lies in the user address space. For that pass the length and then use the access_ok helper to do the check.

Now that the address is never changed it is enough to check the base address (although this would be slightly safer if sockaddr_t were 'const').
This is especially true given some code paths ignore the length (so the length must be checked later - which it will be).

Isn't TASK_SIZE the wrong 'constant'.
Looks pretty 'heavy' on x86-64.
You just need something between valid user and valid kernel addresses. So (1ull << 63) is fine on x86-64 (and probably all 64bit with non-overlapping user/kernel addresses).
For i386 you need (3ul << 30) - probably also common.

David

Quoted 77 lines
Fixes 6d04fe15f78a ("net: optimize the sockptr_t for unified kernel/user address spaces")
---
 include/linux/sockptr.h     | 18 ++++++------------
 net/ipv4/bpfilter/sockopt.c |  2 +-
 net/socket.c                |  2 +-
 3 files changed, 8 insertions(+), 14 deletions(-)
include/linux/sockptr.h +6 -12
index 9e6c81d474cba8..96840def9d69cc 100644
@@ -27,14 +27,6 @@ static inline sockptr_t KERNEL_SOCKPTR(void *p)
 {
 	return (sockptr_t) { .kernel = p };
 }
-
-static inline int __must_check init_user_sockptr(sockptr_t *sp, void __user *p)
-{
-	if ((unsigned long)p >= TASK_SIZE)
-		return -EFAULT;
-	sp->user = p;
-	return 0;
-}
 #else /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 typedef struct {
 	union {
@@ -53,14 +45,16 @@ static inline sockptr_t KERNEL_SOCKPTR(void *p)
 {
 	return (sockptr_t) { .kernel = p, .is_kernel = true };
 }
+#endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 
-static inline int __must_check init_user_sockptr(sockptr_t *sp, void __user *p)
+static inline int __must_check init_user_sockptr(sockptr_t *sp, void __user *p,
+		size_t size)
 {
-	sp->user = p;
-	sp->is_kernel = false;
+	if (!access_ok(p, size))
+		return -EFAULT;
+	*sp = (sockptr_t) { .user = p };
 	return 0;
 }
-#endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 
 static inline bool sockptr_is_null(sockptr_t sockptr)
 {
net/ipv4/bpfilter/sockopt.c +1 -1
index 94f18d2352d007..545b2640f0194d 100644
@@ -65,7 +65,7 @@ int bpfilter_ip_get_sockopt(struct sock *sk, int optname,
 
 	if (get_user(len, optlen))
 		return -EFAULT;
-	err = init_user_sockptr(&optval, user_optval);
+	err = init_user_sockptr(&optval, user_optval, len);
 	if (err)
 		return err;
 	return bpfilter_mbox_request(sk, optname, optval, len, false);
net/socket.c +1 -2
index 94ca4547cd7c53..aff52e81653ce3 100644
@@ -2105,7 +2105,7 @@ int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
 	if (optlen < 0)
 		return -EINVAL;
 
-	err = init_user_sockptr(&optval, user_optval);
+	err = init_user_sockptr(&optval, user_optval, optlen);
 	if (err)
 		return err;
 
--
Quoted 1 line

2.27.0

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