[RFC,v3] net/memif: allow for full key size in socket name
Checks
Commit Message
The key size for memif is 256 but the unix domain socket structure has
space for 100 bytes. Change it to use a larger buffer and not hard
code the keysize everywhere.
Not sure what purpose of socket is anyway since there is no code
which connects to it in the current tree anyway?
Still an RFC, have no way to test.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
---
v3 - fix checkpatch issues
drivers/net/memif/memif_socket.c | 29 ++++++++++++++++++-----------
drivers/net/memif/memif_socket.h | 4 +++-
2 files changed, 21 insertions(+), 12 deletions(-)
Comments
> -----Original Message-----
> From: Stephen Hemminger <stephen@networkplumber.org>
> Sent: Tuesday, July 16, 2019 7:21 PM
> To: dev@dpdk.org; Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at
> Cisco) <jgrajcia@cisco.com>
> Cc: Stephen Hemminger <stephen@networkplumber.org>
> Subject: [RFC v3] net/memif: allow for full key size in socket name
>
> The key size for memif is 256 but the unix domain socket structure has
> space for 100 bytes. Change it to use a larger buffer and not hard
> code the keysize everywhere.
>
> Not sure what purpose of socket is anyway since there is no code
> which connects to it in the current tree anyway?
See memif_connect_slave in memif_socket.c
>
> Still an RFC, have no way to test.
>
> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Tested-by: Jakub Grajciar <jgrajcia@cisco.com>
On 8/30/2019 8:17 AM, Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at
Cisco) wrote:
>
>
>> -----Original Message-----
>> From: Stephen Hemminger <stephen@networkplumber.org>
>> Sent: Tuesday, July 16, 2019 7:21 PM
>> To: dev@dpdk.org; Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at
>> Cisco) <jgrajcia@cisco.com>
>> Cc: Stephen Hemminger <stephen@networkplumber.org>
>> Subject: [RFC v3] net/memif: allow for full key size in socket name
>>
>> The key size for memif is 256 but the unix domain socket structure has
>> space for 100 bytes. Change it to use a larger buffer and not hard
>> code the keysize everywhere.
>>
>> Not sure what purpose of socket is anyway since there is no code
>> which connects to it in the current tree anyway?
>
> See memif_connect_slave in memif_socket.c
>
>>
>> Still an RFC, have no way to test.
>>
>> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
>
> Tested-by: Jakub Grajciar <jgrajcia@cisco.com>
>
Applied to dpdk-next-net/master, thanks.
On 7/16/2019 6:20 PM, Stephen Hemminger wrote:
> The key size for memif is 256 but the unix domain socket structure has
> space for 100 bytes. Change it to use a larger buffer and not hard
> code the keysize everywhere.
>
> Not sure what purpose of socket is anyway since there is no code
> which connects to it in the current tree anyway?
>
> Still an RFC, have no way to test.
>
> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> ---
> v3 - fix checkpatch issues
>
> drivers/net/memif/memif_socket.c | 29 ++++++++++++++++++-----------
> drivers/net/memif/memif_socket.h | 4 +++-
> 2 files changed, 21 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/net/memif/memif_socket.c b/drivers/net/memif/memif_socket.c
> index 01a935f87c9f..5eecbdc98040 100644
> --- a/drivers/net/memif/memif_socket.c
> +++ b/drivers/net/memif/memif_socket.c
> @@ -860,11 +860,16 @@ memif_listener_handler(void *arg)
> rte_free(cc);
> }
>
> +#define MEMIF_SOCKET_UN_SIZE \
> + (offsetof(struct sockaddr_un, sun_path) + MEMIF_SOCKET_KEY_LEN)
> +
> static struct memif_socket *
> -memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
> +memif_socket_create(struct pmd_internals *pmd,
> + const char *key, uint8_t listener)
> {
> struct memif_socket *sock;
> - struct sockaddr_un un;
> + struct sockaddr_un *un;
> + char un_buf[MEMIF_SOCKET_UN_SIZE];
> int sockfd;
> int ret;
> int on = 1;
> @@ -876,7 +881,7 @@ memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
> }
>
> sock->listener = listener;
> - rte_memcpy(sock->filename, key, 256);
> + strlcpy(sock->filename, key, MEMIF_SOCKET_KEY_LEN);
> TAILQ_INIT(&sock->dev_queue);
>
> if (listener != 0) {
> @@ -884,15 +889,16 @@ memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
> if (sockfd < 0)
> goto error;
>
> - un.sun_family = AF_UNIX;
> - memcpy(un.sun_path, sock->filename,
> - sizeof(un.sun_path) - 1);
> + memset(un_buf, 0, sizeof(un_buf));
> + un = (struct sockaddr_un *)un_buf;
> + un->sun_family = AF_UNIX;
> + strlcpy(un->sun_path, sock->filename, MEMIF_SOCKET_KEY_LEN);
>
> ret = setsockopt(sockfd, SOL_SOCKET, SO_PASSCRED, &on,
> sizeof(on));
> if (ret < 0)
> goto error;
> - ret = bind(sockfd, (struct sockaddr *)&un, sizeof(un));
> + ret = bind(sockfd, (struct sockaddr *)un, MEMIF_SOCKET_UN_SIZE);
Hi Jakub,
While testing your zero-copy patch [1], I stuck to a bind() error [2].
When provided a socket length bigger than "sizeof(struct sockaddr)", bind()
fails. I am testing this on a Fedora system.
I wonder if there is a check in glibc related to the length.
What was your test platform for the change?
[1]
https://patches.dpdk.org/patch/57817/
[2]
memif_socket_create(): NULL: Failed to setup socket /run/memif.sock: Invalid
argument
Hi Ferruh,
> Hi Jakub,
>
> While testing your zero-copy patch [1], I stuck to a bind() error [2].
> When provided a socket length bigger than "sizeof(struct sockaddr)", bind()
> fails. I am testing this on a Fedora system.
> I wonder if there is a check in glibc related to the length.
Zero-copy patch does not contain this fix, in fact, I can't find this patch in the commit log.
It was supposed to be applied on 2019-09-13.
http://git.dpdk.org/next/dpdk-next-net/log/?ofs=400
Was there any problem with apply? Maybe because the patch is RFC?
>
> What was your test platform for the change?
Ubuntu 18.04
>
>
>
> [1]
> https://patches.dpdk.org/patch/57817/
>
> [2]
> memif_socket_create(): NULL: Failed to setup socket /run/memif.sock: Invalid
> argument
On 10/7/2019 10:01 AM, Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at
Cisco) wrote:
>
> Hi Ferruh,
>
>> Hi Jakub,
>>
>> While testing your zero-copy patch [1], I stuck to a bind() error [2].
>> When provided a socket length bigger than "sizeof(struct sockaddr)", bind()
>> fails. I am testing this on a Fedora system.
>> I wonder if there is a check in glibc related to the length.
>
> Zero-copy patch does not contain this fix, in fact, I can't find this patch in the commit log.
> It was supposed to be applied on 2019-09-13.
> http://git.dpdk.org/next/dpdk-next-net/log/?ofs=400
> Was there any problem with apply? Maybe because the patch is RFC?
No zero-copy patch doesn't contain it. It is already in the repo. I hit the
issue while testing zero-copy, other than that there is no relation to zero-copy
feature J
The commit is in next-net tree, it is not released yet:
f018cfa797eb ("net/memif: allow for full key size in socket name")
Briefly, I am having issues with that commit, bind() fails, but it has your
Tested-by tag. Can you please confirm it again?
>
>>
>> What was your test platform for the change?
>
> Ubuntu 18.04
So you are on Linux as well, I wonder what is the difference.
>
>>
>>
>>
>> [1]
>> https://patches.dpdk.org/patch/57817/
>>
>> [2]
>> memif_socket_create(): NULL: Failed to setup socket /run/memif.sock: Invalid
>> argument
@@ -860,11 +860,16 @@ memif_listener_handler(void *arg)
rte_free(cc);
}
+#define MEMIF_SOCKET_UN_SIZE \
+ (offsetof(struct sockaddr_un, sun_path) + MEMIF_SOCKET_KEY_LEN)
+
static struct memif_socket *
-memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
+memif_socket_create(struct pmd_internals *pmd,
+ const char *key, uint8_t listener)
{
struct memif_socket *sock;
- struct sockaddr_un un;
+ struct sockaddr_un *un;
+ char un_buf[MEMIF_SOCKET_UN_SIZE];
int sockfd;
int ret;
int on = 1;
@@ -876,7 +881,7 @@ memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
}
sock->listener = listener;
- rte_memcpy(sock->filename, key, 256);
+ strlcpy(sock->filename, key, MEMIF_SOCKET_KEY_LEN);
TAILQ_INIT(&sock->dev_queue);
if (listener != 0) {
@@ -884,15 +889,16 @@ memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
if (sockfd < 0)
goto error;
- un.sun_family = AF_UNIX;
- memcpy(un.sun_path, sock->filename,
- sizeof(un.sun_path) - 1);
+ memset(un_buf, 0, sizeof(un_buf));
+ un = (struct sockaddr_un *)un_buf;
+ un->sun_family = AF_UNIX;
+ strlcpy(un->sun_path, sock->filename, MEMIF_SOCKET_KEY_LEN);
ret = setsockopt(sockfd, SOL_SOCKET, SO_PASSCRED, &on,
sizeof(on));
if (ret < 0)
goto error;
- ret = bind(sockfd, (struct sockaddr *)&un, sizeof(un));
+ ret = bind(sockfd, (struct sockaddr *)un, MEMIF_SOCKET_UN_SIZE);
if (ret < 0)
goto error;
ret = listen(sockfd, 1);
@@ -928,9 +934,10 @@ static struct rte_hash *
memif_create_socket_hash(void)
{
struct rte_hash_parameters params = { 0 };
+
params.name = MEMIF_SOCKET_HASH_NAME;
params.entries = 256;
- params.key_len = 256;
+ params.key_len = MEMIF_SOCKET_KEY_LEN;
params.hash_func = rte_jhash;
params.hash_func_init_val = 0;
return rte_hash_create(¶ms);
@@ -945,7 +952,7 @@ memif_socket_init(struct rte_eth_dev *dev, const char *socket_filename)
struct pmd_internals *tmp_pmd;
struct rte_hash *hash;
int ret;
- char key[256];
+ char key[MEMIF_SOCKET_KEY_LEN];
hash = rte_hash_find_existing(MEMIF_SOCKET_HASH_NAME);
if (hash == NULL) {
@@ -956,8 +963,8 @@ memif_socket_init(struct rte_eth_dev *dev, const char *socket_filename)
}
}
- memset(key, 0, 256);
- rte_memcpy(key, socket_filename, strlen(socket_filename));
+ memset(key, 0, MEMIF_SOCKET_KEY_LEN);
+ strlcpy(key, socket_filename, MEMIF_SOCKET_KEY_LEN);
ret = rte_hash_lookup_data(hash, key, (void **)&socket);
if (ret < 0) {
socket = memif_socket_create(pmd, key,
@@ -79,9 +79,11 @@ struct memif_socket_dev_list_elt {
};
#define MEMIF_SOCKET_HASH_NAME "memif-sh"
+#define MEMIF_SOCKET_KEY_LEN 256
+
struct memif_socket {
struct rte_intr_handle intr_handle; /**< interrupt handle */
- char filename[256]; /**< socket filename */
+ char filename[MEMIF_SOCKET_KEY_LEN]; /**< socket filename */
TAILQ_HEAD(, memif_socket_dev_list_elt) dev_queue;
/**< Queue of devices using this socket */