Skip to contentSkip to conversation

rdma/cm: Add support for native Infiniband addressing

PATCH Superseded [email protected] 8 participants 28 messages Jan 22, 2013 – Mar 13, 2013 12 messages not in the archive

Subsystems (4)

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover rdma/cm: Add support for native Infiniband addressing

From: Sean Hefty <[email protected]>

This is the first group of patches in a series that adds the ability to handle native Infiniband addressing to the rdma_cm. I'm breaking the submission up to limit the number of patches that need to be reviewed at once, hoping that this helps with merging the code upstream. The full patch series is available from:

git://git.openfabrics.org/~shefty/rdma-dev.git for-next

Allowing the use of native IB addresses removes the requirement for IPoIB, which in turn allows us to offload name and/or address translation services to a user space daemon. The primary motivation is to support large scale fabrics, with address and name services either cached or bypassed completely. For example, IB GIDs are known or the information is exchanged out of band by an MPI process manager. However, another use case involves load balancing software. Currently the rdma cm cannot establish rdma connections through a load balancer, since the IP -> GID mapping is not well defined. An out of band mechanism could be used in such situations to determine the correct mapping, with the rdma cm still managing the connection.

The patch set introduces af_ib and sockaddr_ib. The kernel rdma_cm is updated accordingly, mainly to make its handling of addresses more generic. However, since sockaddr_ib is larger than sockaddr_in6, the rdma_ucm requires changes to its user to kernel interface. To provide backwards compatibility, the userspace ABI is extended to support the larger address size.

Changes from v3:
Updated to 3.8-rc4 kernel

Sean Hefty (9):
rdma/cm: define native IB address
rdma/cm: Include AF_IB in loopback and any address checks ib/addr: Add AF_IB support to ip_addr_size
rdma/cm: Update port reservation to support AF_IB rdma/cm: Allow user to specify AF_IB when binding rdma/cm: Do not modify sa_family when setting loopback address rdma/cm: Add helper functions to return id address information rdma/cm: Restrict AF_IB loopback to binding to IB devices only rdma/cm: Verify that source and dest sa_family are the same

 drivers/infiniband/core/addr.c |   20 ++-
 drivers/infiniband/core/cma.c  |  374 ++++++++++++++++++++++++++--------------
 include/linux/socket.h         |    2 +
 include/rdma/ib.h              |   89 ++++++++++
 include/rdma/ib_addr.h         |    6 +-
 include/rdma/rdma_cm.h         |    5 +
 6 files changed, 359 insertions(+), 137 deletions(-)
 create mode 100644 include/rdma/ib.h
Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1358891797-14625-1-git-send-email-sean.hefty-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org

#2 lore raw

PATCH v4 1/9 rdma/cm: define native IB address

From: Sean Hefty <[email protected]>

Define AF_IB and sockaddr_ib to allow the rdma_cm to use native IB addressing.

---
 include/linux/socket.h |    2 +
 include/rdma/ib.h      |   89 ++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 91 insertions(+), 0 deletions(-)
 create mode 100644 include/rdma/ib.h
include/linux/socket.h +2 -0
index 9a546ff..17a33f7 100644
@@ -167,6 +167,7 @@ struct ucred {
 #define AF_PPPOX	24	/* PPPoX sockets		*/
 #define AF_WANPIPE	25	/* Wanpipe API Sockets */
 #define AF_LLC		26	/* Linux LLC			*/
+#define AF_IB		27	/* Native InfiniBand address	*/
 #define AF_CAN		29	/* Controller Area Network      */
 #define AF_TIPC		30	/* TIPC sockets			*/
 #define AF_BLUETOOTH	31	/* Bluetooth sockets 		*/
@@ -210,6 +211,7 @@ struct ucred {
 #define PF_PPPOX	AF_PPPOX
 #define PF_WANPIPE	AF_WANPIPE
 #define PF_LLC		AF_LLC
+#define PF_IB		AF_IB
 #define PF_CAN		AF_CAN
 #define PF_TIPC		AF_TIPC
 #define PF_BLUETOOTH	AF_BLUETOOTH
include/rdma/ib.h new file +89 -0

89 lines not on this page · raw message

Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#3 lore raw

RE: [PATCH v4 1/9] rdma/cm: define native IB address

Quoted 19 lines

Define AF_IB and sockaddr_ib to allow the rdma_cm to use native IB addressing.

---
 include/linux/socket.h |    2 +
 include/rdma/ib.h      |   89 ++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 91 insertions(+), 0 deletions(-)
 create mode 100644 include/rdma/ib.h
include/linux/socket.h +1 -0
index 9a546ff..17a33f7 100644
@@ -167,6 +167,7 @@ struct ucred {
 #define AF_PPPOX	24	/* PPPoX sockets		*/
 #define AF_WANPIPE	25	/* Wanpipe API Sockets */
 #define AF_LLC		26	/* Linux LLC			*/
+#define AF_IB		27	/* Native InfiniBand address	*/

...

Quoted 1 line
include/rdma/ib.h +0 -0

...

include/rdma/ib.h line 87

Quoted 9 lines
include/rdma/ib.h +9 -0
+struct sockaddr_ib {
+	unsigned short int	sib_family;	/* AF_IB */
+	__be16			sib_pkey;
+	__be32			sib_flowinfo;
+	struct ib_addr		sib_addr;
+	__be64			sib_sid;
+	__be64			sib_sid_mask;
+	__u64			sib_scope_id;
+};

Dave/Roland/anyone, is there any feedback on this approach?

If there's hesitation to add new address families to socket.h, I could instead use definitions local to the rdma_cm. This has the potential to result in conflicts if the rdma_cm is expanded for other address families, though such conflicts seem unlikely.

- Sean

#4 lore raw

Re: [PATCH v4 1/9] rdma/cm: define native IB address

Quoted 34 lines · Sean
Quoted 18 lines

Define AF_IB and sockaddr_ib to allow the rdma_cm to use native IB addressing.

---
  include/linux/socket.h |    2 +
  include/rdma/ib.h      |   89 ++++++++++++++++++++++++++++++++++++++++++++++++
  2 files changed, 91 insertions(+), 0 deletions(-)

create mode 100644 include/rdma/ib.h

include/linux/socket.h +1 -0
index 9a546ff..17a33f7 100644
@@ -167,6 +167,7 @@ struct ucred {
  #define AF_PPPOX	24	/* PPPoX sockets		*/
  #define AF_WANPIPE	25	/* Wanpipe API Sockets */
  #define AF_LLC		26	/* Linux LLC			*/
+#define AF_IB		27	/* Native InfiniBand address	*/

...

Quoted 1 line
include/rdma/ib.h +0 -0

...

include/rdma/ib.h line 87

Quoted 9 lines
include/rdma/ib.h +9 -0
+struct sockaddr_ib {
+	unsigned short int	sib_family;	/* AF_IB */
+	__be16			sib_pkey;
+	__be32			sib_flowinfo;
+	struct ib_addr		sib_addr;
+	__be64			sib_sid;
+	__be64			sib_sid_mask;
+	__u64			sib_scope_id;
+};

Dave/Roland/anyone, is there any feedback on this approach?

Quoted 3 lines

If there's hesitation to add new address families to socket.h, I could instead use definitions local to the rdma_cm. This has the potential to result in conflicts if the rdma_cm is expanded for other address families, though such conflicts seem unlikely.

I don't see why not add new address family if it comes to serve a real world use case, which seems to be the case from the description you provided in the cover letter.

Or.

A message that is not in the archive (sent off-list or not yet received): 1828884A29C6694DAF28B7E6B8A8237368B99D59-P5GAC/[email protected]

#6 lore raw

RE: [PATCH v4 1/9] rdma/cm: define native IB address

Quoted 27 lines
Quoted 7 lines

Define AF_IB and sockaddr_ib to allow the rdma_cm to use native IB addressing.

---
 include/linux/socket.h |    2 +
 include/rdma/ib.h      |   89

++++++++++++++++++++++++++++++++++++++++++++++++

Quoted 12 lines
 2 files changed, 91 insertions(+), 0 deletions(-)
 create mode 100644 include/rdma/ib.h
include/linux/socket.h +1 -0
index 9a546ff..17a33f7 100644
@@ -167,6 +167,7 @@ struct ucred {
 #define AF_PPPOX	24	/* PPPoX sockets		*/
 #define AF_WANPIPE	25	/* Wanpipe API Sockets */
 #define AF_LLC		26	/* Linux LLC			*/
+#define AF_IB		27	/* Native InfiniBand address	*/

...

Quoted 1 line
include/rdma/ib.h +0 -0

...

Quoted 11 lines

include/rdma/ib.h line 87

Quoted 9 lines
include/rdma/ib.h +9 -0
+struct sockaddr_ib {
+	unsigned short int	sib_family;	/* AF_IB */
+	__be16			sib_pkey;
+	__be32			sib_flowinfo;
+	struct ib_addr		sib_addr;
+	__be64			sib_sid;
+	__be64			sib_sid_mask;
+	__u64			sib_scope_id;
+};

Dave/Roland/anyone, is there any feedback on this approach?

ping...

Seriously, there is a need to establish connections using native IB GIDs. It is preferable to add this functionality to the rdma_cm. Although the ib_cm can exchange IB CM messages based on GIDs, the rdma_cm is the only interface that provides applications with dynamic service IDs, path record queries, and usable multicast support.

The rdma_cm uses sockaddr for addressing information. The sockaddr_ib structure is defined within the RDMA tree, but I believe that the address family value belongs in socket.h. This helps to avoid any future conflicts. However, I can change the rdma_cm to use an internal enum for address values if there is an objection.

sockaddr_ib/AF_IB is usable with any IB network card. It allows an application to establish connections over IB without IBoIP being present, which is a real need. It also allows an application to use out of band mechanisms for discovering remote GIDs. Such mechanisms are needed to work through IP load balancing software and for MPI scalability.

I'm not trying to be impatient, but it's been 3 years...

- Sean
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1828884A29C6694DAF28B7E6B8A823736F35600F-P5GAC/[email protected]

#8 lore raw

Qperf and APM

Folks:
The qperf utility has -ap, -lap, -rap options in support of APM. Unfortunately they don't seem to work. Is there a way to make this work, or if you have any other ideas on testing failover using rdma please let me know. I have successfully used tcp_bw tests for this purpose already.

Thanks,
Suri

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 82EBBE0B04D4427381C29144BE078F2B-+IkoAhRkys/CbFgIbBqbbjGjJy/[email protected]

#10 lore raw

RE: Qperf and APM

Hi,

Can you be more specific? What are you running and how is it failing?

Thanks,
Dean

Quoted 15 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 9:40 AM
To: [email protected]
Cc: Hefty, Sean
Subject: Qperf and APM

Folks:
The qperf utility has -ap, -lap, -rap options in support of APM. Unfortunately they don't seem to work. Is there a way
to make this work, or if you have any other ideas on testing failover using rdma please let me know. I have successfully
used tcp_bw tests for this purpose already.

Quoted 7 lines

Thanks,
Suri

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 4AF12E8016D2BF46BCDFCE8FAA77A3580394D479-96pTJSsuoYQd3b2yrw5b5LfspsVTdybXVpNB7YpNyf8@public.gmane.org

#12 lore raw

RE: Qperf and APM

This is the error I see on the client(the same warning is seen with -lap and -rap options):

------------------------------
qperf <server-ip-addr> -ap 2 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw

    bw  =  3.39 GB/sec
Quoted 7 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 1:52 PM
To: Suresh Shelvapille; [email protected]
Cc: Hefty, Sean
Subject: RE: Qperf and APM

Quoted 29 lines

Hi,

Can you be more specific? What are you running and how is it failing?

Thanks,
Dean

Quoted 22 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 9:40 AM
To: [email protected]
Cc: Hefty, Sean
Subject: Qperf and APM

Folks:
The qperf utility has -ap, -lap, -rap options in support of APM. Unfortunately they don't seem to work. Is there a way
to make this work, or if you have any other ideas on testing failover using rdma please let me know. I have successfully
used tcp_bw tests for this purpose already.

Thanks,
Suri

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 58D79FF01489431AB6D87CD4977F2E45-+IkoAhRkys/CbFgIbBqbbjGjJy/[email protected]

#14 lore raw

RE: Qperf and APM

Have you tried ignoring the warning and checking of the alternate port works? The qperf code looks to be using the alternate port, but incorrectly fails to mark the argument as used.

Dean

Quoted 54 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 1:00 PM
To: Luick, Dean; [email protected]
Subject: RE: Qperf and APM

This is the error I see on the client(the same warning is seen with -lap and - rap options):

------------------------------
qperf <server-ip-addr> -ap 2 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw

    bw  =  3.39 GB/sec
Quoted 7 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 1:52 PM
To: Suresh Shelvapille; [email protected]
Cc: Hefty, Sean
Subject: RE: Qperf and APM

Quoted 17 lines

Hi,

Can you be more specific? What are you running and how is it failing?

Thanks,
Dean

Quoted 10 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 9:40 AM
To: [email protected]
Cc: Hefty, Sean
Subject: Qperf and APM

Folks:
The qperf utility has -ap, -lap, -rap options in support of APM.

Unfortunately

Quoted 5 lines
Quoted 5 lines

they don't seem to work. Is there a way
to make this work, or if you have any other ideas on testing failover using rdma please let me know. I have successfully
used tcp_bw tests for this purpose already.

Quoted 7 lines
Quoted 7 lines

Thanks,
Suri

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Quoted 4 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 4AF12E8016D2BF46BCDFCE8FAA77A3580394D4FB-96pTJSsuoYQd3b2yrw5b5LfspsVTdybXVpNB7YpNyf8@public.gmane.org

#16 lore raw

RE: Qperf and APM

I can see the packets going on the alternate path when the first link fails, but at the end of the test, the result is a failure, so I am not sure!
I tried this with both '-ap' and "-lap -rap" options. I am assuming they are equivalent.

------------------------------
[root]# qperf <server-ipaddr> -ap 2 -t 30 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw
failed to receive results: timed out

-------------------------
[root]# qperf <server-ipaddr> -lap 2 -rap 2 -t 30 rc_bw rc_bw:
warning: -lap set but not used in test rc_bw
warning: -rap set but not used in test rc_bw failed to receive results: timed out

thanks,
Suri

Quoted 6 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 3:57 PM
To: Suresh Shelvapille; [email protected]
Subject: RE: Qperf and APM

Quoted 63 lines

Have you tried ignoring the warning and checking of the alternate port works? The qperf code looks to be using the alternate port, but incorrectly fails to mark the argument as used.

Dean

Quoted 58 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 1:00 PM
To: Luick, Dean; [email protected]
Subject: RE: Qperf and APM

This is the error I see on the client(the same warning is seen with -lap and - rap options):

------------------------------
qperf <server-ip-addr> -ap 2 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw

    bw  =  3.39 GB/sec
Quoted 7 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 1:52 PM
To: Suresh Shelvapille; [email protected]
Cc: Hefty, Sean
Subject: RE: Qperf and APM

Quoted 17 lines

Hi,

Can you be more specific? What are you running and how is it failing?

Thanks,
Dean

Quoted 10 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 9:40 AM
To: [email protected]
Cc: Hefty, Sean
Subject: Qperf and APM

Folks:
The qperf utility has -ap, -lap, -rap options in support of APM.

Unfortunately

Quoted 5 lines
Quoted 5 lines

they don't seem to work. Is there a way
to make this work, or if you have any other ideas on testing failover using rdma please let me know. I have successfully
used tcp_bw tests for this purpose already.

Quoted 7 lines
Quoted 7 lines

Thanks,
Suri

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 8AF67C8E913E4EE49D65CABA83E3DEBF-+IkoAhRkys/CbFgIbBqbbjGjJy/[email protected]

#18 lore raw

RE: Qperf and APM

Using QLogic dual port HCAs, this works for me when I disable port 1 on the client, server, or both while the test is running:

# qperf <server ipaddr> -ap 2 -t 30 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw
    bw  =  272 MB/sec

The warning can be ignored. Using -ap is equivalent to using -rap and -lap with the same port.

The "failed to receive results" is qperf saying that once the test finished, the server did not contact the client.

Dean

Quoted 94 lines

-----Original Message-----
From: Suresh Shelvapille [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 4:24 PM
To: Luick, Dean; [email protected]
Subject: RE: Qperf and APM

I can see the packets going on the alternate path when the first link fails, but at the end of the test, the result is a
failure, so I am not sure!
I tried this with both '-ap' and "-lap -rap" options. I am assuming they are equivalent.

------------------------------
[root]# qperf <server-ipaddr> -ap 2 -t 30 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw
failed to receive results: timed out

-------------------------
[root]# qperf <server-ipaddr> -lap 2 -rap 2 -t 30 rc_bw rc_bw:
warning: -lap set but not used in test rc_bw
warning: -rap set but not used in test rc_bw failed to receive results: timed out

thanks,
Suri

Quoted 7 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 3:57 PM
To: Suresh Shelvapille; [email protected]
Subject: RE: Qperf and APM

Have you tried ignoring the warning and checking of the alternate port

works? The qperf code looks

Quoted 1 line

to be using the alternate port, but incorrectly fails to mark the argument as

used.

Quoted 11 lines

Dean

Quoted 8 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 1:00 PM
To: Luick, Dean; [email protected]
Subject: RE: Qperf and APM

This is the error I see on the client(the same warning is seen with -lap and

-

Quoted 30 lines
Quoted 30 lines

rap options):

------------------------------
qperf <server-ip-addr> -ap 2 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw

    bw  =  3.39 GB/sec
Quoted 22 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 1:52 PM
To: Suresh Shelvapille; [email protected]
Cc: Hefty, Sean
Subject: RE: Qperf and APM

Hi,

Can you be more specific? What are you running and how is it failing?

Thanks,
Dean

Quoted 8 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 9:40 AM
To: [email protected]
Cc: Hefty, Sean
Subject: Qperf and APM

Quoted 5 lines
Quoted 5 lines
Quoted 2 lines
Quoted 2 lines

Folks:
The qperf utility has -ap, -lap, -rap options in support of APM.

Unfortunately

Quoted 2 lines
Quoted 2 lines

they don't seem to work. Is there a way
to make this work, or if you have any other ideas on testing failover

using

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

rdma please let me know. I have successfully
used tcp_bw tests for this purpose already.

Thanks,
Suri

Quoted 11 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma"

in

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-

info.html

Quoted 5 lines
Quoted 5 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 4AF12E8016D2BF46BCDFCE8FAA77A3580394F46B-96pTJSsuoYQd3b2yrw5b5LfspsVTdybXVpNB7YpNyf8@public.gmane.org

#20 lore raw

RE: Qperf and APM

Dean:

The difference is, you got a BW number where as I don't! So, I don't know whether the test was successful or not. Do you have a Mellanox dual port card?

Mellanox folks: have you guys tried this on your dual port cards?

Thanks,
Suri

Quoted 14 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 13, 2013 6:25 PM
To: Suresh Shelvapille; [email protected]
Subject: RE: Qperf and APM

Using QLogic dual port HCAs, this works for me when I disable port 1 on the client, server, or both while the test is running:

# qperf <server ipaddr> -ap 2 -t 30 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw
    bw  =  272 MB/sec
Quoted 113 lines

The warning can be ignored. Using -ap is equivalent to using -rap and -lap with the same port.

The "failed to receive results" is qperf saying that once the test finished, the server did not contact the client.

Dean

Quoted 105 lines

-----Original Message-----
From: Suresh Shelvapille [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 4:24 PM
To: Luick, Dean; [email protected]
Subject: RE: Qperf and APM

I can see the packets going on the alternate path when the first link fails, but at the end of the test, the result is a
failure, so I am not sure!
I tried this with both '-ap' and "-lap -rap" options. I am assuming they are equivalent.

------------------------------
[root]# qperf <server-ipaddr> -ap 2 -t 30 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw
failed to receive results: timed out

-------------------------
[root]# qperf <server-ipaddr> -lap 2 -rap 2 -t 30 rc_bw rc_bw:
warning: -lap set but not used in test rc_bw
warning: -rap set but not used in test rc_bw failed to receive results: timed out

thanks,
Suri

Quoted 7 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 3:57 PM
To: Suresh Shelvapille; [email protected]
Subject: RE: Qperf and APM

Have you tried ignoring the warning and checking of the alternate port

works? The qperf code looks

Quoted 1 line

to be using the alternate port, but incorrectly fails to mark the argument as

used.

Quoted 11 lines

Dean

Quoted 8 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 1:00 PM
To: Luick, Dean; [email protected]
Subject: RE: Qperf and APM

This is the error I see on the client(the same warning is seen with -lap and

-

Quoted 30 lines
Quoted 30 lines

rap options):

------------------------------
qperf <server-ip-addr> -ap 2 rc_bw
rc_bw:
warning: -ap set but not used in test rc_bw

    bw  =  3.39 GB/sec
Quoted 22 lines

-----Original Message-----
From: Luick, Dean [mailto:[email protected]]
Sent: Wednesday, March 06, 2013 1:52 PM
To: Suresh Shelvapille; [email protected]
Cc: Hefty, Sean
Subject: RE: Qperf and APM

Hi,

Can you be more specific? What are you running and how is it failing?

Thanks,
Dean

Quoted 8 lines

-----Original Message-----
From: [email protected] [mailto:linux-rdma- [email protected]] On Behalf Of Suresh Shelvapille
Sent: Wednesday, March 06, 2013 9:40 AM
To: [email protected]
Cc: Hefty, Sean
Subject: Qperf and APM

Quoted 5 lines
Quoted 5 lines
Quoted 2 lines
Quoted 2 lines

Folks:
The qperf utility has -ap, -lap, -rap options in support of APM.

Unfortunately

Quoted 2 lines
Quoted 2 lines

they don't seem to work. Is there a way
to make this work, or if you have any other ideas on testing failover

using

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

rdma please let me know. I have successfully
used tcp_bw tests for this purpose already.

Thanks,
Suri

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma"

in

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-

info.html

Quoted 5 lines
Quoted 5 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#21 lore raw

Re: [PATCH v4 1/9] rdma/cm: define native IB address

In reply to [email protected]

include/rdma/ib.h line 87

Quoted 10 lines · On 22/01/2013 23:56, [email protected] wrote:
include/rdma/ib.h new file +10 -0
+
+struct sockaddr_ib {
+	unsigned short int	sib_family;	/* AF_IB */
+	__be16			sib_pkey;
+	__be32			sib_flowinfo;
+	struct ib_addr		sib_addr;
+	__be64			sib_sid;
+	__be64			sib_sid_mask;
+	__u64			sib_scope_id;
+};

just a nit, maybe reorder the fields to better cope with their IPv6 buddies (where there is such) from sockaddr_in6?

Also I see that both IPv6 header and IB GRH have a traffic class field which is skipped in both
cases for the related sockaddr_ structure, not sure why, is this something the kernel stack decides on and uses but not available for applications to read/modify?

struct sockaddr_in6 {

        unsigned short int      sin6_family;    /* AF_INET6 */
        __be16                  sin6_port;      /* Transport layer port 
# */
        __be32                  sin6_flowinfo;  /* IPv6 flow information */
        struct in6_addr         sin6_addr;      /* IPv6 address */
        __u32                   sin6_scope_id;  /* scope id (new in 

RFC2553) */
};

#22 lore raw

PATCH v4 2/9 rdma/cm: Include AF_IB in loopback and any address checks

In reply to a message not in the archive

From: Sean Hefty <[email protected]>

Enhance checks for loopback and any address to support AF_IB in addition to AF_INET and AF_INT6. This will allow future patches to use AF_IB when binding and resolving addresses.

---
 drivers/infiniband/core/cma.c |   40 ++++++++++++++++++++++++----------------
 1 files changed, 24 insertions(+), 16 deletions(-)
drivers/infiniband/core/cma.c +24 -16
index d789eea..9e3dc7f 100644
@@ -50,6 +50,7 @@
 #include <rdma/rdma_cm.h>
 #include <rdma/rdma_cm_ib.h>
 #include <rdma/rdma_netlink.h>
+#include <rdma/ib.h>
 #include <rdma/ib_cache.h>
 #include <rdma/ib_cm.h>
 #include <rdma/ib_sa.h>
@@ -679,26 +680,30 @@ EXPORT_SYMBOL(rdma_init_qp_attr);
 
 static inline int cma_zero_addr(struct sockaddr *addr)
 {
-	struct in6_addr *ip6;
-
-	if (addr->sa_family == AF_INET)
-		return ipv4_is_zeronet(
-			((struct sockaddr_in *)addr)->sin_addr.s_addr);
-	else {
-		ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
-		return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
-			ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
+	switch (addr->sa_family) {
+	case AF_INET:
+		return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
+	case AF_INET6:
+		return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
+	case AF_IB:
+		return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
+	default:
+		return 0;
 	}
 }
 
 static inline int cma_loopback_addr(struct sockaddr *addr)
 {
-	if (addr->sa_family == AF_INET)
-		return ipv4_is_loopback(
-			((struct sockaddr_in *) addr)->sin_addr.s_addr);
-	else
-		return ipv6_addr_loopback(
-			&((struct sockaddr_in6 *) addr)->sin6_addr);
+	switch (addr->sa_family) {
+	case AF_INET:
+		return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
+	case AF_INET6:
+		return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
+	case AF_IB:
+		return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
+	default:
+		return 0;
+	}
 }
 
 static inline int cma_any_addr(struct sockaddr *addr)
@@ -715,9 +720,12 @@ static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
 	case AF_INET:
 		return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
 		       ((struct sockaddr_in *) dst)->sin_addr.s_addr;
-	default:
+	case AF_INET6:
 		return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
 				     &((struct sockaddr_in6 *) dst)->sin6_addr);
+	default:
+		return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
+				   &((struct sockaddr_ib *) dst)->sib_addr);
 	}
 }
 
Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#23 lore raw

PATCH v4 4/9 rdma/cm: Update port reservation to support AF_IB

In reply to a message not in the archive

From: Sean Hefty <[email protected]>

The AF_IB uses a 64-bit service id (SID), which the user can control through the use of a mask. The rdma_cm will assign values to the unmasked portions of the SID based on the selected port space and port number.

Because the IB spec divides the SID range into several regions, a SID/mask combination may fall into one of the existing port space ranges as defined by the RDMA CM IP Annex. Map the AF_IB SID to the correct RDMA port space.

---
 drivers/infiniband/core/cma.c |  107 +++++++++++++++++++++++++++++++++--------
 include/rdma/rdma_cm.h        |    5 ++
 2 files changed, 91 insertions(+), 21 deletions(-)
drivers/infiniband/core/cma.c +86 -21
index e08f830..7eed4ee 100644
@@ -729,12 +729,22 @@ static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
 	}
 }
 
-static inline __be16 cma_port(struct sockaddr *addr)
+static __be16 cma_port(struct sockaddr *addr)
 {
-	if (addr->sa_family == AF_INET)
+	struct sockaddr_ib *sib;
+
+	switch (addr->sa_family) {
+	case AF_INET:
 		return ((struct sockaddr_in *) addr)->sin_port;
-	else
+	case AF_INET6:
 		return ((struct sockaddr_in6 *) addr)->sin6_port;
+	case AF_IB:
+		sib = (struct sockaddr_ib *) addr;
+		return htons((u16) (be64_to_cpu(sib->sib_sid) &
+				    be64_to_cpu(sib->sib_sid_mask)));
+	default:
+		return 0;
+	}
 }
 
 static inline int cma_any_port(struct sockaddr *addr)
@@ -2139,10 +2149,29 @@ EXPORT_SYMBOL(rdma_set_afonly);
 static void cma_bind_port(struct rdma_bind_list *bind_list,
 			  struct rdma_id_private *id_priv)
 {
-	struct sockaddr_in *sin;
+	struct sockaddr *addr;
+	struct sockaddr_ib *sib;
+	u64 sid, mask;
+	__be16 port;
 
-	sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
-	sin->sin_port = htons(bind_list->port);
+	addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
+	port = htons(bind_list->port);
+
+	switch (addr->sa_family) {
+	case AF_INET:
+		((struct sockaddr_in *) addr)->sin_port = port;
+		break;
+	case AF_INET6:
+		((struct sockaddr_in6 *) addr)->sin6_port = port;
+		break;
+	case AF_IB:
+		sib = (struct sockaddr_ib *) addr;
+		sid = be64_to_cpu(sib->sib_sid);
+		mask = be64_to_cpu(sib->sib_sid_mask);
+		sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
+		sib->sib_sid_mask = cpu_to_be64(~0ULL);
+		break;
+	}
 	id_priv->bind_list = bind_list;
 	hlist_add_head(&id_priv->node, &bind_list->owners);
 }
@@ -2280,31 +2309,67 @@ static int cma_bind_listen(struct rdma_id_private *id_priv)
 	return ret;
 }
 
-static int cma_get_port(struct rdma_id_private *id_priv)
+static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
 {
-	struct idr *ps;
-	int ret;
-
 	switch (id_priv->id.ps) {
 	case RDMA_PS_SDP:
-		ps = &sdp_ps;
-		break;
+		return &sdp_ps;
 	case RDMA_PS_TCP:
-		ps = &tcp_ps;
-		break;
+		return &tcp_ps;
 	case RDMA_PS_UDP:
-		ps = &udp_ps;
-		break;
+		return &udp_ps;
 	case RDMA_PS_IPOIB:
-		ps = &ipoib_ps;
-		break;
+		return &ipoib_ps;
 	case RDMA_PS_IB:
-		ps = &ib_ps;
-		break;
+		return &ib_ps;
 	default:
-		return -EPROTONOSUPPORT;
+		return NULL;
+	}
+}
+
+static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
+{
+	struct idr *ps = NULL;
+	struct sockaddr_ib *sib;
+	u64 sid_ps, mask, sid;
+
+	sib = (struct sockaddr_ib *) &id_priv->id.route.addr.src_addr;
+	mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
+	sid = be64_to_cpu(sib->sib_sid) & mask;
+
+	if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
+		sid_ps = RDMA_IB_IP_PS_IB;
+		ps = &ib_ps;
+	} else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
+		   (sid == (RDMA_IB_IP_PS_TCP & mask))) {
+		sid_ps = RDMA_IB_IP_PS_TCP;
+		ps = &tcp_ps;
+	} else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
+		   (sid == (RDMA_IB_IP_PS_UDP & mask))) {
+		sid_ps = RDMA_IB_IP_PS_UDP;
+		ps = &udp_ps;
 	}
 
+	if (ps) {
+		sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
+		sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
+						be64_to_cpu(sib->sib_sid_mask));
+	}
+	return ps;
+}
+
+static int cma_get_port(struct rdma_id_private *id_priv)
+{
+	struct idr *ps;
+	int ret;
+
+	if (id_priv->id.route.addr.src_addr.ss_family != AF_IB)
+		ps = cma_select_inet_ps(id_priv);
+	else
+		ps = cma_select_ib_ps(id_priv);
+	if (!ps)
+		return -EPROTONOSUPPORT;
+
 	mutex_lock(&lock);
 	if (cma_any_port((struct sockaddr *) &id_priv->id.route.addr.src_addr))
 		ret = cma_alloc_any_port(ps, id_priv);
include/rdma/rdma_cm.h +5 -0
index ad3a314..1e6c3c7 100644
@@ -70,6 +70,11 @@ enum rdma_port_space {
 	RDMA_PS_UDP   = 0x0111,
 };
 
+#define RDMA_IB_IP_PS_MASK   0xFFFFFFFFFFFF0000ULL
+#define RDMA_IB_IP_PS_TCP    0x0000000001060000ULL
+#define RDMA_IB_IP_PS_UDP    0x0000000001110000ULL
+#define RDMA_IB_IP_PS_IB     0x00000000013F0000ULL
+
 struct rdma_addr {
 	struct sockaddr_storage src_addr;
 	struct sockaddr_storage dst_addr;
Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1358891797-14625-5-git-send-email-sean.hefty-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org

#25 lore raw

Re: [PATCH v4 4/9] rdma/cm: Update port reservation to support AF_IB

Quoted 4 lines · On 22/01/2013 23:56, [email protected] wrote:

The AF_IB uses a 64-bit service id (SID), which the user can control through the use of a mask. The rdma_cm will assign values to the unmasked portions of the SID based on the selected port space and port number.

Something here I am not sure to follow -- if AF_IB consumers are allowed to provide 64-bit SID, how the IB port space is expressed in their SIDs?

Wouldn't it make sense to use a simpler approach that has a dedicated IB_PS in the port space portion of the RDMA-CM IP IBTA annex and let the consumer provide 16 bits for their part of the SID?

Or.

Quoted 9 lines

Because the IB spec divides the SID range into several regions, a SID/mask combination may fall into one of the existing port space ranges as defined by the RDMA CM IP Annex. Map the AF_IB SID to the correct RDMA port space.

[...]

include/rdma/rdma_cm.h enum rdma_port_space { · line 77

Quoted 8 lines
include/rdma/rdma_cm.h +4 -0
@@ -70,6 +70,11 @@ enum rdma_port_space {
  	RDMA_PS_UDP   = 0x0111,
  };
 
+#define RDMA_IB_IP_PS_MASK   0xFFFFFFFFFFFF0000ULL
+#define RDMA_IB_IP_PS_TCP    0x0000000001060000ULL
+#define RDMA_IB_IP_PS_UDP    0x0000000001110000ULL
+#define RDMA_IB_IP_PS_IB     0x00000000013F0000ULL
 

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 511B7474.3040201-VPRAkNaXOzVWk0Htik3J/[email protected]

#27 lore raw

RE: [PATCH v4 4/9] rdma/cm: Update port reservation to support AF_IB

Quoted 8 lines
Quoted 4 lines

The AF_IB uses a 64-bit service id (SID), which the user can control through the use of a mask. The rdma_cm will assign values to the unmasked portions of the SID based on the selected port space and port number.

Something here I am not sure to follow -- if AF_IB consumers are allowed to provide 64-bit SID, how
the IB port space is expressed in their SIDs?

The spec, notably the RDMA CM Annex, provides fixed values for the upper 48-bits of the service ID. Only the lower 16-bits are available to users. The upper bits end up identifying the various 'port spaces'.

(Note that there's no actual association between an IB port space and port spaces in the IP network stack.)

Quoted 4 lines

Wouldn't it make sense to use a simpler approach that has a dedicated IB_PS in the port space
portion of the RDMA-CM IP IBTA annex and let the consumer provide 16 bits for their part of the SID?

Ultimately, IB establishes communication using 64-bit service IDs, which is what sockaddr_ib exposes.

The SID mask was proposed by Jason as a way to simplify things. Conceptually, AF_IB has a single port space range, the 64-bit SID. The mask helps us break the SID into the various ranges, so we can clearly know which parts of the SID are fixed, and which can be variable. E.g. when connecting, the entire SID must be provided. However, when binding we want to support binding to a specific SID or only to a SID range (i.e. port space).

(I'm having trouble finding the original email threads where this was discussed.)

- Sean
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#28 lore raw

Re: [PATCH v4 4/9] rdma/cm: Update port reservation to support AF_IB

Quoted 9 lines · Sean

Ultimately, IB establishes communication using 64-bit service IDs, which is what sockaddr_ib exposes.

The SID mask was proposed by Jason as a way to simplify things. Conceptually, AF_IB has a single port space range, the 64-bit SID. The mask helps us break the SID into the various ranges, so we can clearly know which parts of the SID are fixed, and which can be variable. E.g. when connecting, the entire SID must be provided. However, when binding we want to support binding to a specific SID or only to a SID range (i.e. port space).

Right. AF_IB is about exposing the native IB CM, rather than a cooked version of it, so allowing user space to control the entire SID is desirable.

The intent of the masking was to provide a means for the kernel to atomically select a process-unique ID (such as the 16 bits the IP varient requires) for listening.

Other uses of IB CM would use a different prefix, but will still fundamentally require a machine unique portion.

This addresses a fundamental problem in ib_ucm - there is no way for multiple processes to co-ordinate on unique SIDs.

Sean, I've lost track of all this over time, but just to check in, with this new interface, would it be possible to obsolete ib_ucm? That seems desirable to me, as it is so flawed..

Jason

#29 lore raw

RE: [PATCH v4 4/9] rdma/cm: Update port reservation to support AF_IB

Quoted 3 lines

Sean, I've lost track of all this over time, but just to check in, with this new interface, would it be possible to obsolete ib_ucm? That seems desirable to me, as it is so flawed..

I'd love to obsolete ib_ucm. This interface should make it possible, though some additional implementation would be needed (e.g. for automatic path migration).

I have an immediate need to connect using native IB addresses. So without this, I'll be forced to use ib_ucm or roll a new user space CM.

A message that is not in the archive (sent off-list or not yet received): 1828884A29C6694DAF28B7E6B8A8237368B9A432-P5GAC/[email protected]

#31 lore raw

Re: [PATCH v4 4/9] rdma/cm: Update port reservation to support AF_IB

Quoted 7 lines · Sean
Quoted 3 lines

Sean, I've lost track of all this over time, but just to check in, with this new interface, would it be possible to obsolete ib_ucm? That seems desirable to me, as it is so flawed..

I'd love to obsolete ib_ucm. This interface should make it possible, though some additional implementation would be needed (e.g. for automatic path migration).

Okay - but there are some nice gaps in the API to cleanly place such things in future?

A long time ago we discussed having different paths for CM GMP, inbound data and outbound data (ie the full IB asymetric pathing model), did that get included in the API as well?

Quoted 3 lines

I have an immediate need to connect using native IB addresses. So without this, I'll be forced to use ib_ucm or roll a new user space CM.

Using ucm is alot of trouble. Many sites don't have it setup properly, don't have proper device node permissions and the issue with unique IDs ends up being critical..

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#32 lore raw

RE: [PATCH v4 4/9] rdma/cm: Update port reservation to support AF_IB

Quoted 3 lines

A long time ago we discussed having different paths for CM GMP, inbound data and outbound data (ie the full IB asymetric pathing model), did that get included in the API as well?

Yes - this support made it upstream.

#33 lore raw

PATCH v4 6/9 rdma/cm: Do not modify sa_family when setting loopback address

In reply to a message not in the archive

From: Sean Hefty <[email protected]>

cma_resolve_loopback is called after an rdma_cm_id has been bound to a specific sa_family and port. Once the source sa_family for the id has been set, do not modify it. Only the actual IP address portion of the source address needs to be set.

As part of this fix, we can simplify setting the source address by moving the loopback address assignment from cma_resolve_loopback to cma_bind_loopback. cma_bind_loopback is only invoked when the source address is the loopback address.

Finally, add loopback support for AF_IB as part of the change.

---
 drivers/infiniband/core/cma.c |   31 ++++++++++++++++++-------------
 1 files changed, 18 insertions(+), 13 deletions(-)
drivers/infiniband/core/cma.c +18 -13
index caea4ef3..bb22885 100644
@@ -1952,6 +1952,23 @@ err:
 }
 EXPORT_SYMBOL(rdma_resolve_route);
 
+static void cma_set_loopback(struct sockaddr *addr)
+{
+	switch (addr->sa_family) {
+	case AF_INET:
+		((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+		break;
+	case AF_INET6:
+		ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
+			      0, 0, 0, htonl(1));
+		break;
+	default:
+		ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
+			    0, 0, 0, htonl(1));
+		break;
+	}
+}
+
 static int cma_bind_loopback(struct rdma_id_private *id_priv)
 {
 	struct cma_device *cma_dev;
@@ -1992,6 +2009,7 @@ port_found:
 	ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
 	id_priv->id.port_num = p;
 	cma_attach_to_dev(id_priv, cma_dev);
+	cma_set_loopback((struct sockaddr *) &id_priv->id.route.addr.src_addr);
 out:
 	mutex_unlock(&lock);
 	return ret;
@@ -2039,7 +2057,6 @@ out:
 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
 {
 	struct cma_work *work;
-	struct sockaddr *src, *dst;
 	union ib_gid gid;
 	int ret;
 
@@ -2056,18 +2073,6 @@ static int cma_resolve_loopback(struct rdma_id_private *id_priv)
 	rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
 	rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
 
-	src = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
-	if (cma_zero_addr(src)) {
-		dst = (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
-		if ((src->sa_family = dst->sa_family) == AF_INET) {
-			((struct sockaddr_in *)src)->sin_addr =
-				((struct sockaddr_in *)dst)->sin_addr;
-		} else {
-			((struct sockaddr_in6 *)src)->sin6_addr =
-				((struct sockaddr_in6 *)dst)->sin6_addr;
-		}
-	}
-
 	work->id = id_priv;
 	INIT_WORK(&work->work, cma_work_handler);
 	work->old_state = RDMA_CM_ADDR_QUERY;
Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#34 lore raw

PATCH v4 7/9 rdma/cm: Add helper functions to return id address information

In reply to a message not in the archive

From: Sean Hefty <[email protected]>

Provide inline helpers to extract source and destination address data from the rdma_cm_id.

---
 drivers/infiniband/core/cma.c |  138 +++++++++++++++++++++--------------------
 1 files changed, 71 insertions(+), 67 deletions(-)
drivers/infiniband/core/cma.c +71 -67
index bb22885..0ebf123 100644
@@ -311,6 +311,21 @@ static void cma_release_dev(struct rdma_id_private *id_priv)
 	mutex_unlock(&lock);
 }
 
+static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
+{
+	return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
+}
+
+static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
+{
+	return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
+}
+
+static inline unsigned short cma_family(struct rdma_id_private *id_priv)
+{
+	return id_priv->id.route.addr.src_addr.ss_family;
+}
+
 static int cma_set_qkey(struct rdma_id_private *id_priv)
 {
 	struct ib_sa_mcmember_rec rec;
@@ -900,8 +915,7 @@ static void cma_cancel_operation(struct rdma_id_private *id_priv,
 		cma_cancel_route(id_priv);
 		break;
 	case RDMA_CM_LISTEN:
-		if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr)
-				&& !id_priv->cma_dev)
+		if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
 			cma_cancel_listens(id_priv);
 		break;
 	default:
@@ -1138,6 +1152,7 @@ static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
 	if (IS_ERR(id))
 		return NULL;
 
+	id_priv = container_of(id, struct rdma_id_private, id);
 	cma_save_net_info(&id->route.addr, &listen_id->route.addr,
 			  ip_ver, port, src, dst);
 
@@ -1152,19 +1167,17 @@ static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
 	if (rt->num_paths == 2)
 		rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
 
-	if (cma_any_addr((struct sockaddr *) &rt->addr.src_addr)) {
+	if (cma_any_addr(cma_src_addr(id_priv))) {
 		rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
 		rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
 		ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
 	} else {
-		ret = cma_translate_addr((struct sockaddr *) &rt->addr.src_addr,
-					 &rt->addr.dev_addr);
+		ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
 		if (ret)
 			goto err;
 	}
 	rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
 
-	id_priv = container_of(id, struct rdma_id_private, id);
 	id_priv->state = RDMA_CM_CONNECT;
 	return id_priv;
 
@@ -1188,7 +1201,7 @@ static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
 	if (IS_ERR(id))
 		return NULL;
 
-
+	id_priv = container_of(id, struct rdma_id_private, id);
 	if (cma_get_net_info(ib_event->private_data, listen_id->ps,
 			     &ip_ver, &port, &src, &dst))
 		goto err;
@@ -1197,13 +1210,11 @@ static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
 			  ip_ver, port, src, dst);
 
 	if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
-		ret = cma_translate_addr((struct sockaddr *) &id->route.addr.src_addr,
-					 &id->route.addr.dev_addr);
+		ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
 		if (ret)
 			goto err;
 	}
 
-	id_priv = container_of(id, struct rdma_id_private, id);
 	id_priv->state = RDMA_CM_CONNECT;
 	return id_priv;
 err:
@@ -1386,9 +1397,9 @@ static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
 		event.event = RDMA_CM_EVENT_DISCONNECTED;
 		break;
 	case IW_CM_EVENT_CONNECT_REPLY:
-		sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
+		sin = (struct sockaddr_in *) cma_src_addr(id_priv);
 		*sin = iw_event->local_addr;
-		sin = (struct sockaddr_in *) &id_priv->id.route.addr.dst_addr;
+		sin = (struct sockaddr_in *) cma_dst_addr(id_priv);
 		*sin = iw_event->remote_addr;
 		switch (iw_event->status) {
 		case 0:
@@ -1486,9 +1497,9 @@ static int iw_conn_req_handler(struct iw_cm_id *cm_id,
 	cm_id->context = conn_id;
 	cm_id->cm_handler = cma_iw_handler;
 
-	sin = (struct sockaddr_in *) &new_cm_id->route.addr.src_addr;
+	sin = (struct sockaddr_in *) cma_src_addr(conn_id);
 	*sin = iw_event->local_addr;
-	sin = (struct sockaddr_in *) &new_cm_id->route.addr.dst_addr;
+	sin = (struct sockaddr_in *) cma_dst_addr(conn_id);
 	*sin = iw_event->remote_addr;
 
 	ret = ib_query_device(conn_id->id.device, &attr);
@@ -1545,7 +1556,7 @@ static int cma_ib_listen(struct rdma_id_private *id_priv)
 
 	id_priv->cm_id.ib = id;
 
-	addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
+	addr = cma_src_addr(id_priv);
 	svc_id = cma_get_service_id(id_priv->id.ps, addr);
 	if (cma_any_addr(addr) && !id_priv->afonly)
 		ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
@@ -1576,7 +1587,7 @@ static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
 
 	id_priv->cm_id.iw = id;
 
-	sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
+	sin = (struct sockaddr_in *) cma_src_addr(id_priv);
 	id_priv->cm_id.iw->local_addr = *sin;
 
 	ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
@@ -1614,8 +1625,8 @@ static void cma_listen_on_dev(struct rdma_id_private *id_priv,
 	dev_id_priv = container_of(id, struct rdma_id_private, id);
 
 	dev_id_priv->state = RDMA_CM_ADDR_BOUND;
-	memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
-	       rdma_addr_size((struct sockaddr *) &id_priv->id.route.addr.src_addr));
+	memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
+	       rdma_addr_size(cma_src_addr(id_priv)));
 
 	cma_attach_to_dev(dev_id_priv, cma_dev);
 	list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
@@ -1673,29 +1684,28 @@ static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
 			      struct cma_work *work)
 {
-	struct rdma_addr *addr = &id_priv->id.route.addr;
+	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 	struct ib_sa_path_rec path_rec;
 	ib_sa_comp_mask comp_mask;
 	struct sockaddr_in6 *sin6;
 
 	memset(&path_rec, 0, sizeof path_rec);
-	rdma_addr_get_sgid(&addr->dev_addr, &path_rec.sgid);
-	rdma_addr_get_dgid(&addr->dev_addr, &path_rec.dgid);
-	path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr));
+	rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
+	rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
+	path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
 	path_rec.numb_path = 1;
 	path_rec.reversible = 1;
-	path_rec.service_id = cma_get_service_id(id_priv->id.ps,
-							(struct sockaddr *) &addr->dst_addr);
+	path_rec.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
 
 	comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
 		    IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
 		    IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
 
-	if (addr->src_addr.ss_family == AF_INET) {
+	if (cma_family(id_priv) == AF_INET) {
 		path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
 		comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
 	} else {
-		sin6 = (struct sockaddr_in6 *) &addr->src_addr;
+		sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
 		path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
 		comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
 	}
@@ -2009,7 +2019,7 @@ port_found:
 	ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
 	id_priv->id.port_num = p;
 	cma_attach_to_dev(id_priv, cma_dev);
-	cma_set_loopback((struct sockaddr *) &id_priv->id.route.addr.src_addr);
+	cma_set_loopback(cma_src_addr(id_priv));
 out:
 	mutex_unlock(&lock);
 	return ret;
@@ -2037,8 +2047,7 @@ static void addr_handler(int status, struct sockaddr *src_addr,
 		event.event = RDMA_CM_EVENT_ADDR_ERROR;
 		event.status = status;
 	} else {
-		memcpy(&id_priv->id.route.addr.src_addr, src_addr,
-		       rdma_addr_size(src_addr));
+		memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
 		event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
 	}
 
@@ -2115,11 +2124,11 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
 		return -EINVAL;
 
 	atomic_inc(&id_priv->refcount);
-	memcpy(&id->route.addr.dst_addr, dst_addr, rdma_addr_size(dst_addr));
+	memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
 	if (cma_any_addr(dst_addr))
 		ret = cma_resolve_loopback(id_priv);
 	else
-		ret = rdma_resolve_ip(&addr_client, (struct sockaddr *) &id->route.addr.src_addr,
+		ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
 				      dst_addr, &id->route.addr.dev_addr,
 				      timeout_ms, addr_handler, id_priv);
 	if (ret)
@@ -2180,7 +2189,7 @@ static void cma_bind_port(struct rdma_bind_list *bind_list,
 	u64 sid, mask;
 	__be16 port;
 
-	addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
+	addr = cma_src_addr(id_priv);
 	port = htons(bind_list->port);
 
 	switch (addr->sa_family) {
@@ -2279,7 +2288,7 @@ static int cma_check_port(struct rdma_bind_list *bind_list,
 	struct sockaddr *addr, *cur_addr;
 	struct hlist_node *node;
 
-	addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
+	addr = cma_src_addr(id_priv);
 	hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
 		if (id_priv == cur_id)
 			continue;
@@ -2288,7 +2297,7 @@ static int cma_check_port(struct rdma_bind_list *bind_list,
 		    cur_id->reuseaddr)
 			continue;
 
-		cur_addr = (struct sockaddr *) &cur_id->id.route.addr.src_addr;
+		cur_addr = cma_src_addr(cur_id);
 		if (id_priv->afonly && cur_id->afonly &&
 		    (addr->sa_family != cur_addr->sa_family))
 			continue;
@@ -2308,7 +2317,7 @@ static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
 	unsigned short snum;
 	int ret;
 
-	snum = ntohs(cma_port((struct sockaddr *) &id_priv->id.route.addr.src_addr));
+	snum = ntohs(cma_port(cma_src_addr(id_priv)));
 	if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
 		return -EACCES;
 
@@ -2359,7 +2368,7 @@ static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
 	struct sockaddr_ib *sib;
 	u64 sid_ps, mask, sid;
 
-	sib = (struct sockaddr_ib *) &id_priv->id.route.addr.src_addr;
+	sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
 	mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
 	sid = be64_to_cpu(sib->sib_sid) & mask;
 
@@ -2389,7 +2398,7 @@ static int cma_get_port(struct rdma_id_private *id_priv)
 	struct idr *ps;
 	int ret;
 
-	if (id_priv->id.route.addr.src_addr.ss_family != AF_IB)
+	if (cma_family(id_priv) != AF_IB)
 		ps = cma_select_inet_ps(id_priv);
 	else
 		ps = cma_select_ib_ps(id_priv);
@@ -2397,7 +2406,7 @@ static int cma_get_port(struct rdma_id_private *id_priv)
 		return -EPROTONOSUPPORT;
 
 	mutex_lock(&lock);
-	if (cma_any_port((struct sockaddr *) &id_priv->id.route.addr.src_addr))
+	if (cma_any_port(cma_src_addr(id_priv)))
 		ret = cma_alloc_any_port(ps, id_priv);
 	else
 		ret = cma_use_port(ps, id_priv);
@@ -2432,8 +2441,8 @@ int rdma_listen(struct rdma_cm_id *id, int backlog)
 
 	id_priv = container_of(id, struct rdma_id_private, id);
 	if (id_priv->state == RDMA_CM_IDLE) {
-		((struct sockaddr *) &id->route.addr.src_addr)->sa_family = AF_INET;
-		ret = rdma_bind_addr(id, (struct sockaddr *) &id->route.addr.src_addr);
+		id->route.addr.src_addr.ss_family = AF_INET;
+		ret = rdma_bind_addr(id, cma_src_addr(id_priv));
 		if (ret)
 			return ret;
 	}
@@ -2502,7 +2511,7 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
 			goto err1;
 	}
 
-	memcpy(&id->route.addr.src_addr, addr, rdma_addr_size(addr));
+	memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
 	if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
 		if (addr->sa_family == AF_INET)
 			id_priv->afonly = 1;
@@ -2525,19 +2534,18 @@ err1:
 }
 EXPORT_SYMBOL(rdma_bind_addr);
 
-static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
-			  struct rdma_route *route)
+static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
 {
 	struct cma_hdr *cma_hdr;
 	struct sdp_hh *sdp_hdr;
 
-	if (route->addr.src_addr.ss_family == AF_INET) {
+	if (cma_family(id_priv) == AF_INET) {
 		struct sockaddr_in *src4, *dst4;
 
-		src4 = (struct sockaddr_in *) &route->addr.src_addr;
-		dst4 = (struct sockaddr_in *) &route->addr.dst_addr;
+		src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
+		dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
 
-		switch (ps) {
+		switch (id_priv->id.ps) {
 		case RDMA_PS_SDP:
 			sdp_hdr = hdr;
 			if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
@@ -2559,10 +2567,10 @@ static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
 	} else {
 		struct sockaddr_in6 *src6, *dst6;
 
-		src6 = (struct sockaddr_in6 *) &route->addr.src_addr;
-		dst6 = (struct sockaddr_in6 *) &route->addr.dst_addr;
+		src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
+		dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
 
-		switch (ps) {
+		switch (id_priv->id.ps) {
 		case RDMA_PS_SDP:
 			sdp_hdr = hdr;
 			if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
@@ -2653,7 +2661,6 @@ static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
 			      struct rdma_conn_param *conn_param)
 {
 	struct ib_cm_sidr_req_param req;
-	struct rdma_route *route;
 	struct ib_cm_id	*id;
 	int ret;
 
@@ -2670,8 +2677,7 @@ static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
 		memcpy((void *) req.private_data + sizeof(struct cma_hdr),
 		       conn_param->private_data, conn_param->private_data_len);
 
-	route = &id_priv->id.route;
-	ret = cma_format_hdr((void *) req.private_data, id_priv->id.ps, route);
+	ret = cma_format_hdr((void *) req.private_data, id_priv);
 	if (ret)
 		goto out;
 
@@ -2683,9 +2689,8 @@ static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
 	}
 	id_priv->cm_id.ib = id;
 
-	req.path = route->path_rec;
-	req.service_id = cma_get_service_id(id_priv->id.ps,
-					    (struct sockaddr *) &route->addr.dst_addr);
+	req.path = id_priv->id.route.path_rec;
+	req.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
 	req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
 	req.max_cm_retries = CMA_MAX_CM_RETRIES;
 
@@ -2730,7 +2735,7 @@ static int cma_connect_ib(struct rdma_id_private *id_priv,
 	id_priv->cm_id.ib = id;
 
 	route = &id_priv->id.route;
-	ret = cma_format_hdr(private_data, id_priv->id.ps, route);
+	ret = cma_format_hdr(private_data, id_priv);
 	if (ret)
 		goto out;
 	req.private_data = private_data;
@@ -2739,8 +2744,7 @@ static int cma_connect_ib(struct rdma_id_private *id_priv,
 	if (route->num_paths == 2)
 		req.alternate_path = &route->path_rec[1];
 
-	req.service_id = cma_get_service_id(id_priv->id.ps,
-					    (struct sockaddr *) &route->addr.dst_addr);
+	req.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
 	req.qp_num = id_priv->qp_num;
 	req.qp_type = id_priv->id.qp_type;
 	req.starting_psn = id_priv->seq_num;
@@ -2779,10 +2783,10 @@ static int cma_connect_iw(struct rdma_id_private *id_priv,
 
 	id_priv->cm_id.iw = cm_id;
 
-	sin = (struct sockaddr_in*) &id_priv->id.route.addr.src_addr;
+	sin = (struct sockaddr_in *) cma_src_addr(id_priv);
 	cm_id->local_addr = *sin;
 
-	sin = (struct sockaddr_in*) &id_priv->id.route.addr.dst_addr;
+	sin = (struct sockaddr_in *) cma_dst_addr(id_priv);
 	cm_id->remote_addr = *sin;
 
 	ret = cma_modify_qp_rtr(id_priv, conn_param);
@@ -3547,29 +3551,29 @@ static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
 			id_stats->bound_dev_if =
 				id->route.addr.dev_addr.bound_dev_if;
 
-			if (id->route.addr.src_addr.ss_family == AF_INET) {
+			if (cma_family(id_priv) == AF_INET) {
 				if (ibnl_put_attr(skb, nlh,
 						  sizeof(struct sockaddr_in),
-						  &id->route.addr.src_addr,
+						  cma_src_addr(id_priv),
 						  RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
 					goto out;
 				}
 				if (ibnl_put_attr(skb, nlh,
 						  sizeof(struct sockaddr_in),
-						  &id->route.addr.dst_addr,
+						  cma_dst_addr(id_priv),
 						  RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
 					goto out;
 				}
-			} else if (id->route.addr.src_addr.ss_family == AF_INET6) {
+			} else if (cma_family(id_priv) == AF_INET6) {
 				if (ibnl_put_attr(skb, nlh,
 						  sizeof(struct sockaddr_in6),
-						  &id->route.addr.src_addr,
+						  cma_src_addr(id_priv),
 						  RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
 					goto out;
 				}
 				if (ibnl_put_attr(skb, nlh,
 						  sizeof(struct sockaddr_in6),
-						  &id->route.addr.dst_addr,
+						  cma_dst_addr(id_priv),
 						  RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
 					goto out;
 				}
Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#35 lore raw

Re: [PATCH v4 7/9] rdma/cm: Add helper functions to return id address information

Quoted 1 line · On 22/01/2013 23:56, [email protected] wrote:

Provide inline helpers to extract source and destination address data from the rdma_cm_id.

Sean,

FWIW - this patch and most of patch #6 can (and worth doing) be introduced as cleanup/maintenance patches to the rdma-cm without direct relation to the AF_IB effort.

Or.

#36 lore raw

PATCH v4 8/9 rdma/cm: Restrict AF_IB loopback to binding to IB devices only

In reply to a message not in the archive

From: Sean Hefty <[email protected]>

If a user specifies AF_IB as the source address for a loopback connection, limit the resolution to IB devices only.

---
 drivers/infiniband/core/cma.c |   28 ++++++++++++++++++++--------
 1 files changed, 20 insertions(+), 8 deletions(-)
drivers/infiniband/core/cma.c +20 -8
index 0ebf123..e7409b4 100644
@@ -1981,26 +1981,38 @@ static void cma_set_loopback(struct sockaddr *addr)
 
 static int cma_bind_loopback(struct rdma_id_private *id_priv)
 {
-	struct cma_device *cma_dev;
+	struct cma_device *cma_dev, *cur_dev;
 	struct ib_port_attr port_attr;
 	union ib_gid gid;
 	u16 pkey;
 	int ret;
 	u8 p;
 
+	cma_dev = NULL;
 	mutex_lock(&lock);
-	if (list_empty(&dev_list)) {
+	list_for_each_entry(cur_dev, &dev_list, list) {
+		if (cma_family(id_priv) == AF_IB &&
+		    rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
+			continue;
+
+		if (!cma_dev)
+			cma_dev = cur_dev;
+
+		for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
+			if (!ib_query_port(cur_dev->device, p, &port_attr) &&
+			    port_attr.state == IB_PORT_ACTIVE) {
+				cma_dev = cur_dev;
+				goto port_found;
+			}
+		}
+	}
+
+	if (!cma_dev) {
 		ret = -ENODEV;
 		goto out;
 	}
-	list_for_each_entry(cma_dev, &dev_list, list)
-		for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
-			if (!ib_query_port(cma_dev->device, p, &port_attr) &&
-			    port_attr.state == IB_PORT_ACTIVE)
-				goto port_found;
 
 	p = 1;
-	cma_dev = list_entry(dev_list.next, struct cma_device, list);
 
 port_found:
 	ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#37 lore raw

PATCH v4 9/9 rdma/cm: Verify that source and dest sa_family are the same

In reply to a message not in the archive

From: Sean Hefty <[email protected]>

---
 drivers/infiniband/core/cma.c |    8 +++-----
 1 files changed, 3 insertions(+), 5 deletions(-)
drivers/infiniband/core/cma.c +3 -5
index e7409b4..64799ea 100644
@@ -1849,14 +1849,9 @@ static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
 	struct rdma_addr *addr = &route->addr;
 	struct cma_work *work;
 	int ret;
-	struct sockaddr_in *src_addr = (struct sockaddr_in *)&route->addr.src_addr;
-	struct sockaddr_in *dst_addr = (struct sockaddr_in *)&route->addr.dst_addr;
 	struct net_device *ndev = NULL;
 	u16 vid;
 
-	if (src_addr->sin_family != dst_addr->sin_family)
-		return -EINVAL;
-
 	work = kzalloc(sizeof *work, GFP_KERNEL);
 	if (!work)
 		return -ENOMEM;
@@ -2132,6 +2127,9 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
 			return ret;
 	}
 
+	if (cma_family(id_priv) != dst_addr->sa_family)
+		return -EINVAL;
+
 	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
 		return -EINVAL;
 
Signature · 6 lines
1.7.3

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#38 lore raw

PATCH v4 3/9 ib/addr: Add AF_IB support to ip_addr_size

In reply to [email protected]

From: Sean Hefty <[email protected]>

Add support for AF_IB to ip_addr_size, and rename the function to account for the change. Give the compiler more control over whether the call should be inline or not by moving the definition into the .c file, removing the static inline, and exporting it.

---
 drivers/infiniband/core/addr.c |   20 ++++++++++++++++++--
 drivers/infiniband/core/cma.c  |   12 ++++++------
 include/rdma/ib_addr.h         |    6 +-----
 3 files changed, 25 insertions(+), 13 deletions(-)
drivers/infiniband/core/addr.c +18 -2
index eaec8d7..e90f2b2 100644
@@ -45,6 +45,7 @@
 #include <net/addrconf.h>
 #include <net/ip6_route.h>
 #include <rdma/ib_addr.h>
+#include <rdma/ib.h>
 
 MODULE_AUTHOR("Sean Hefty");
 MODULE_DESCRIPTION("IB Address Translation");
@@ -70,6 +71,21 @@ static LIST_HEAD(req_list);
 static DECLARE_DELAYED_WORK(work, process_req);
 static struct workqueue_struct *addr_wq;
 
+int rdma_addr_size(struct sockaddr *addr)
+{
+	switch (addr->sa_family) {
+	case AF_INET:
+		return sizeof(struct sockaddr_in);
+	case AF_INET6:
+		return sizeof(struct sockaddr_in6);
+	case AF_IB:
+		return sizeof(struct sockaddr_ib);
+	default:
+		return 0;
+	}
+}
+EXPORT_SYMBOL(rdma_addr_size);
+
 void rdma_addr_register_client(struct rdma_addr_client *client)
 {
 	atomic_set(&client->refcount, 1);
@@ -369,12 +385,12 @@ int rdma_resolve_ip(struct rdma_addr_client *client,
 			goto err;
 		}
 
-		memcpy(src_in, src_addr, ip_addr_size(src_addr));
+		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 	} else {
 		src_in->sa_family = dst_addr->sa_family;
 	}
 
-	memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
+	memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 	req->addr = addr;
 	req->callback = callback;
 	req->context = context;
drivers/infiniband/core/cma.c +6 -6
index 9e3dc7f..e08f830 100644
@@ -1584,7 +1584,7 @@ static void cma_listen_on_dev(struct rdma_id_private *id_priv,
 
 	dev_id_priv->state = RDMA_CM_ADDR_BOUND;
 	memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
-	       ip_addr_size((struct sockaddr *) &id_priv->id.route.addr.src_addr));
+	       rdma_addr_size((struct sockaddr *) &id_priv->id.route.addr.src_addr));
 
 	cma_attach_to_dev(dev_id_priv, cma_dev);
 	list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
@@ -1989,7 +1989,7 @@ static void addr_handler(int status, struct sockaddr *src_addr,
 		event.status = status;
 	} else {
 		memcpy(&id_priv->id.route.addr.src_addr, src_addr,
-		       ip_addr_size(src_addr));
+		       rdma_addr_size(src_addr));
 		event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
 	}
 
@@ -2079,7 +2079,7 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
 		return -EINVAL;
 
 	atomic_inc(&id_priv->refcount);
-	memcpy(&id->route.addr.dst_addr, dst_addr, ip_addr_size(dst_addr));
+	memcpy(&id->route.addr.dst_addr, dst_addr, rdma_addr_size(dst_addr));
 	if (cma_any_addr(dst_addr))
 		ret = cma_resolve_loopback(id_priv);
 	else
@@ -2410,7 +2410,7 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
 			goto err1;
 	}
 
-	memcpy(&id->route.addr.src_addr, addr, ip_addr_size(addr));
+	memcpy(&id->route.addr.src_addr, addr, rdma_addr_size(addr));
 	if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
 		if (addr->sa_family == AF_INET)
 			id_priv->afonly = 1;
@@ -3189,7 +3189,7 @@ int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
 	if (!mc)
 		return -ENOMEM;
 
-	memcpy(&mc->addr, addr, ip_addr_size(addr));
+	memcpy(&mc->addr, addr, rdma_addr_size(addr));
 	mc->context = context;
 	mc->id_priv = id_priv;
 
@@ -3234,7 +3234,7 @@ void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
 	id_priv = container_of(id, struct rdma_id_private, id);
 	spin_lock_irq(&id_priv->lock);
 	list_for_each_entry(mc, &id_priv->mc_list, list) {
-		if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
+		if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
 			list_del(&mc->list);
 			spin_unlock_irq(&id_priv->lock);
 
include/rdma/ib_addr.h +1 -5
index 9996539..f3ac0f2 100644
@@ -102,11 +102,7 @@ void rdma_addr_cancel(struct rdma_dev_addr *addr);
 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
 	      const unsigned char *dst_dev_addr);
 
-static inline int ip_addr_size(struct sockaddr *addr)
-{
-	return addr->sa_family == AF_INET6 ?
-	       sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in);
-}
+int rdma_addr_size(struct sockaddr *addr);
 
 static inline u16 ib_addr_get_pkey(struct rdma_dev_addr *dev_addr)
 {
Signature · 1 line
1.7.3
#39 lore raw

PATCH v4 5/9 rdma/cm: Allow user to specify AF_IB when binding

In reply to [email protected]

From: Sean Hefty <[email protected]>

Modify rdma_bind_addr to allow the user to specify AF_IB when binding to a device. AF_IB indicates that the user is not mapping an IP address to the native IB addressing. (The mapping may have already been done, or is not needed.)

---
 drivers/infiniband/core/cma.c |   34 ++++++++++++++++++++++++++++------
 1 files changed, 28 insertions(+), 6 deletions(-)
drivers/infiniband/core/cma.c +28 -6
index 7eed4ee..caea4ef3 100644
@@ -359,6 +359,27 @@ static int find_gid_port(struct ib_device *device, union ib_gid *gid, u8 port_nu
 	return -EADDRNOTAVAIL;
 }
 
+static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
+{
+	dev_addr->dev_type = ARPHRD_INFINIBAND;
+	rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
+	ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
+}
+
+static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
+{
+	int ret;
+
+	if (addr->sa_family != AF_IB) {
+		ret = rdma_translate_ip(addr, dev_addr);
+	} else {
+		cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
+		ret = 0;
+	}
+
+	return ret;
+}
+
 static int cma_acquire_dev(struct rdma_id_private *id_priv)
 {
 	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
@@ -1136,8 +1157,8 @@ static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
 		rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
 		ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
 	} else {
-		ret = rdma_translate_ip((struct sockaddr *) &rt->addr.src_addr,
-					&rt->addr.dev_addr);
+		ret = cma_translate_addr((struct sockaddr *) &rt->addr.src_addr,
+					 &rt->addr.dev_addr);
 		if (ret)
 			goto err;
 	}
@@ -1176,8 +1197,8 @@ static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
 			  ip_ver, port, src, dst);
 
 	if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
-		ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr,
-					&id->route.addr.dev_addr);
+		ret = cma_translate_addr((struct sockaddr *) &id->route.addr.src_addr,
+					 &id->route.addr.dev_addr);
 		if (ret)
 			goto err;
 	}
@@ -2454,7 +2475,8 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
 	struct rdma_id_private *id_priv;
 	int ret;
 
-	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
+	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
+	    addr->sa_family != AF_IB)
 		return -EAFNOSUPPORT;
 
 	id_priv = container_of(id, struct rdma_id_private, id);
@@ -2466,7 +2488,7 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
 		goto err1;
 
 	if (!cma_any_addr(addr)) {
-		ret = rdma_translate_ip(addr, &id->route.addr.dev_addr);
+		ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
 		if (ret)
 			goto err1;
 
Signature · 1 line
1.7.3