]> git.openfabrics.org - ~shefty/rdma-win.git/log
~shefty/rdma-win.git
14 years agopop
Sean Hefty [Fri, 29 Jan 2010 05:07:46 +0000 (21:07 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 05:07:46 +0000 (21:07 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:43:30 +0000 (17:43 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:43:30 +0000 (17:43 -0800)]
pop

14 years agorefresh
Sean Hefty [Fri, 29 Jan 2010 01:42:31 +0000 (17:42 -0800)]
refresh

14 years agorefresh
Sean Hefty [Fri, 29 Jan 2010 01:42:31 +0000 (17:42 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Fri, 29 Jan 2010 01:42:21 +0000 (17:42 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Fri, 29 Jan 2010 01:42:21 +0000 (17:42 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-addr_ptr
Sean Hefty [Fri, 29 Jan 2010 01:42:19 +0000 (17:42 -0800)]
Refresh of dapl-addr_ptr

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:55 +0000 (17:41 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:55 +0000 (17:41 -0800)]
pop

14 years agorefresh
Sean Hefty [Fri, 29 Jan 2010 01:41:49 +0000 (17:41 -0800)]
refresh

14 years agorefresh
Sean Hefty [Fri, 29 Jan 2010 01:41:49 +0000 (17:41 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Fri, 29 Jan 2010 01:41:45 +0000 (17:41 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Fri, 29 Jan 2010 01:41:45 +0000 (17:41 -0800)]
refresh (create temporary patch)

14 years agoRefresh of librdmacm-pdata
Sean Hefty [Fri, 29 Jan 2010 01:41:43 +0000 (17:41 -0800)]
Refresh of librdmacm-pdata

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:37 +0000 (17:41 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:36 +0000 (17:41 -0800)]
pop

14 years agorefresh
Sean Hefty [Fri, 29 Jan 2010 01:41:30 +0000 (17:41 -0800)]
refresh

14 years agorefresh
Sean Hefty [Fri, 29 Jan 2010 01:41:30 +0000 (17:41 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Fri, 29 Jan 2010 01:41:18 +0000 (17:41 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Fri, 29 Jan 2010 01:41:17 +0000 (17:41 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-close
Sean Hefty [Fri, 29 Jan 2010 01:41:15 +0000 (17:41 -0800)]
Refresh of dapl-close

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:09 +0000 (17:41 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:09 +0000 (17:41 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:03 +0000 (17:41 -0800)]
pop

14 years agopop
Sean Hefty [Fri, 29 Jan 2010 01:41:02 +0000 (17:41 -0800)]
pop

14 years agorefresh
Sean Hefty [Thu, 28 Jan 2010 23:40:02 +0000 (15:40 -0800)]
refresh

14 years agorefresh
Sean Hefty [Thu, 28 Jan 2010 23:40:02 +0000 (15:40 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Thu, 28 Jan 2010 23:39:59 +0000 (15:39 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Thu, 28 Jan 2010 23:39:59 +0000 (15:39 -0800)]
refresh (create temporary patch)

14 years agoRefresh of libibverbs-delete-channel
Sean Hefty [Thu, 28 Jan 2010 23:39:57 +0000 (15:39 -0800)]
Refresh of libibverbs-delete-channel

14 years agopop
Sean Hefty [Thu, 28 Jan 2010 23:38:07 +0000 (15:38 -0800)]
pop

14 years agopop
Sean Hefty [Thu, 28 Jan 2010 23:38:07 +0000 (15:38 -0800)]
pop

14 years agolibibverbs/device: destroy completion channel when closing device
Sean Hefty [Fri, 22 Jan 2010 22:00:34 +0000 (14:00 -0800)]
libibverbs/device: destroy completion channel when closing device

ibv_close_device should destroy the completion channel
when closing the device and freeing the memory.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agolibibverbs/device: destroy completion channel when closing device
Sean Hefty [Fri, 22 Jan 2010 22:00:34 +0000 (14:00 -0800)]
libibverbs/device: destroy completion channel when closing device

ibv_close_device should destroy the completion channel
when closing the device and freeing the memory.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Thu, 28 Jan 2010 23:35:26 +0000 (15:35 -0800)]
pop

14 years agopop
Sean Hefty [Thu, 28 Jan 2010 23:35:26 +0000 (15:35 -0800)]
pop

14 years agodapl: use private_data_len for mem copies
Sean Hefty [Sat, 23 Jan 2010 00:22:36 +0000 (16:22 -0800)]
dapl: use private_data_len for mem copies

When copying private_data out of rdma_cm events, use the
reported private_data_len for the size, and not IB maximums.
This fixes a bug running over the librdmacm on windows, where
DAPL accessed invalid memory.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodapl/cma: fix referencing freed address
Sean Hefty [Sat, 23 Jan 2010 07:28:57 +0000 (23:28 -0800)]
dapl/cma: fix referencing freed address

DAPL uses a pointer to reference the local and remote addresses
of an endpoint.  It expects that those addresses are located
in memory that is always accessible.  Typically, for the local
address, the pointer references the address stored with the DAPL
HCA device.  However, for the cma provider, it changes this pointer
to reference the address stored with the rdma_cm_id.

This causes a problem when that endpoint is connected on the
passive side of a connection.  When connect requests are given
to DAPL, a new rdma_cm_id is associated with the request.  The
DAPL code replaces the current rdma_cm_id associated with a
user's endpoint with the new rdma_cm_id.  The old rdma_cm_id is
then deleted.  But the endpoint's local address pointer still
references the address stored with the old rdma_cm_id.  The
result is that any reference to the address will access freed
memory.

Fix this by keeping the local address pointer always pointing
to the address associated with the DAPL HCA device.  This is about
the best that can be done given the DAPL interface design.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agolibrdmacm: set private_data_len
Sean Hefty [Sat, 23 Jan 2010 00:16:46 +0000 (16:16 -0800)]
librdmacm: set private_data_len

Set the private_data_len for reported events.  This allows
DAPL to use the value when copying private data.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodapl: move close device after async thread is done using it
Sean Hefty [Fri, 22 Jan 2010 22:48:54 +0000 (14:48 -0800)]
dapl: move close device after async thread is done using it

Before calling ibv_close_device, wait for the asynchronous
processing thread to finish using the device.  This prevents
a use after free error.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodelete
Sean Hefty [Thu, 28 Jan 2010 23:34:29 +0000 (15:34 -0800)]
delete

14 years agodelete
Sean Hefty [Thu, 28 Jan 2010 23:34:29 +0000 (15:34 -0800)]
delete

14 years agonew
Sean Hefty [Thu, 28 Jan 2010 23:34:24 +0000 (15:34 -0800)]
new

14 years agonew
Sean Hefty [Thu, 28 Jan 2010 23:34:23 +0000 (15:34 -0800)]
new

14 years agosadfasf
Sean Hefty [Thu, 28 Jan 2010 23:34:20 +0000 (15:34 -0800)]
sadfasf

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:15:32 +0000 (00:15 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:15:32 +0000 (00:15 -0800)]
pop

14 years agodapl: use private_data_len for mem copies
Sean Hefty [Sat, 23 Jan 2010 00:22:36 +0000 (16:22 -0800)]
dapl: use private_data_len for mem copies

When copying private_data out of rdma_cm events, use the
reported private_data_len for the size, and not IB maximums.
This fixes a bug running over the librdmacm on windows, where
DAPL accessed invalid memory.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:15:18 +0000 (00:15 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:15:18 +0000 (00:15 -0800)]
pop

14 years agodapl/cma: fix referencing freed address
Sean Hefty [Sat, 23 Jan 2010 07:28:57 +0000 (23:28 -0800)]
dapl/cma: fix referencing freed address

DAPL uses a pointer to reference the local and remote addresses
of an endpoint.  It expects that those addresses are located
in memory that is always accessible.  Typically, for the local
address, the pointer references the address stored with the DAPL
HCA device.  However, for the cma provider, it changes this pointer
to reference the address stored with the rdma_cm_id.

This causes a problem when that endpoint is connected on the
passive side of a connection.  When connect requests are given
to DAPL, a new rdma_cm_id is associated with the request.  The
DAPL code replaces the current rdma_cm_id associated with a
user's endpoint with the new rdma_cm_id.  The old rdma_cm_id is
then deleted.  But the endpoint's local address pointer still
references the address stored with the old rdma_cm_id.  The
result is that any reference to the address will access freed
memory.

Fix this by keeping the local address pointer always pointing
to the address associated with the DAPL HCA device.  This is about
the best that can be done given the DAPL interface design.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:15:06 +0000 (00:15 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:15:05 +0000 (00:15 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:13:10 +0000 (00:13 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:13:10 +0000 (00:13 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:13:03 +0000 (00:13 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 08:13:03 +0000 (00:13 -0800)]
pop

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 08:04:31 +0000 (00:04 -0800)]
refresh

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 08:04:31 +0000 (00:04 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 08:04:30 +0000 (00:04 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 08:04:30 +0000 (00:04 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-addr_ptr
Sean Hefty [Sat, 23 Jan 2010 08:04:28 +0000 (00:04 -0800)]
Refresh of dapl-addr_ptr

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 07:29:42 +0000 (23:29 -0800)]
refresh

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 07:29:42 +0000 (23:29 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 07:29:40 +0000 (23:29 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 07:29:40 +0000 (23:29 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-addr_ptr
Sean Hefty [Sat, 23 Jan 2010 07:29:37 +0000 (23:29 -0800)]
Refresh of dapl-addr_ptr

14 years agonew
Sean Hefty [Sat, 23 Jan 2010 07:29:00 +0000 (23:29 -0800)]
new

14 years agonew
Sean Hefty [Sat, 23 Jan 2010 07:28:59 +0000 (23:28 -0800)]
new

14 years agodapl/cma: fix referencing freed address
Sean Hefty [Sat, 23 Jan 2010 07:28:57 +0000 (23:28 -0800)]
dapl/cma: fix referencing freed address

DAPL uses a pointer to reference the local and remote addresses
of an endpoint.  It expects that those addresses are located
in memory that is always accessible.  Typically, for the local
address, the pointer references the address stored with the DAPL
HCA device.  However, for the cma provider, it changes this pointer
to reference the address stored with the rdma_cm_id.

This causes a problem when that endpoint is connected on the
passive side of a connection.  When connect requests are given
to DAPL, a new rdma_cm_id is associated with the request.  The
DAPL code replaces the current rdma_cm_id associated with a
user's endpoint with the new rdma_cm_id.  The old rdma_cm_id is
then deleted.  But the endpoint's local address pointer still
references the address stored with the old rdma_cm_id.  The
result is that any reference to the address will access freed
memory.

Fix this by keeping the local address pointer always pointing
to the address associated with the DAPL HCA device.  This is about
the best that can be done given the DAPL interface design.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodapl/cma: fix referencing freed address
Sean Hefty [Sat, 23 Jan 2010 07:28:57 +0000 (23:28 -0800)]
dapl/cma: fix referencing freed address

DAPL uses a pointer to reference the local and remote addresses
of an endpoint.  It expects that those addresses are located
in memory that is always accessible.  Typically, for the local
address, the pointer references the address stored with the DAPL
HCA device.  However, for the cma provider, it changes this pointer
to reference the address stored with the rdma_cm_id.

This causes a problem when that endpoint is connected on the
passive side of a connection.  When connect requests are given
to DAPL, a new rdma_cm_id is associated with the request.  The
DAPL code replaces the current rdma_cm_id associated with a
user's endpoint with the new rdma_cm_id.  The old rdma_cm_id is
then deleted.  But the endpoint's local address pointer still
references the address stored with the old rdma_cm_id.  The
result is that any reference to the address will access freed
memory.

Fix this by keeping the local address pointer always pointing
to the address associated with the DAPL HCA device.  This is about
the best that can be done given the DAPL interface design.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 06:35:00 +0000 (22:35 -0800)]
refresh

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 06:35:00 +0000 (22:35 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 06:34:47 +0000 (22:34 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 06:34:47 +0000 (22:34 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-pdata2
Sean Hefty [Sat, 23 Jan 2010 06:34:44 +0000 (22:34 -0800)]
Refresh of dapl-pdata2

14 years agonew
Sean Hefty [Sat, 23 Jan 2010 06:29:41 +0000 (22:29 -0800)]
new

14 years agonew
Sean Hefty [Sat, 23 Jan 2010 06:29:41 +0000 (22:29 -0800)]
new

14 years agodapl: quick fix for wrong private data size
Sean Hefty [Sat, 23 Jan 2010 06:29:39 +0000 (22:29 -0800)]
dapl: quick fix for wrong private data size

DAPl expects the private data size to be up to 256 bytes, but
on windows the private data size is limited to 56 bytes.  As
a result, DAPL can access memory beyond the end of what's
allocated.

A more 'correct' fix is being submitted upstream to dapl.  This
is a simpler fix for the Windows 2.2 release only.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodapl: quick fix for wrong private data size
Sean Hefty [Sat, 23 Jan 2010 06:29:39 +0000 (22:29 -0800)]
dapl: quick fix for wrong private data size

DAPl expects the private data size to be up to 256 bytes, but
on windows the private data size is limited to 56 bytes.  As
a result, DAPL can access memory beyond the end of what's
allocated.

A more 'correct' fix is being submitted upstream to dapl.  This
is a simpler fix for the Windows 2.2 release only.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Sat, 23 Jan 2010 06:27:22 +0000 (22:27 -0800)]
pop

14 years agopop
Sean Hefty [Sat, 23 Jan 2010 06:27:22 +0000 (22:27 -0800)]
pop

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 01:16:36 +0000 (17:16 -0800)]
refresh

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 01:16:36 +0000 (17:16 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 01:16:34 +0000 (17:16 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 01:16:33 +0000 (17:16 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-pdata
Sean Hefty [Sat, 23 Jan 2010 01:16:31 +0000 (17:16 -0800)]
Refresh of dapl-pdata

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 01:13:38 +0000 (17:13 -0800)]
refresh

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 01:13:38 +0000 (17:13 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 01:13:35 +0000 (17:13 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 01:13:34 +0000 (17:13 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-pdata
Sean Hefty [Sat, 23 Jan 2010 01:13:31 +0000 (17:13 -0800)]
Refresh of dapl-pdata

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 01:05:50 +0000 (17:05 -0800)]
refresh

14 years agorefresh
Sean Hefty [Sat, 23 Jan 2010 01:05:50 +0000 (17:05 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 01:05:47 +0000 (17:05 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Sat, 23 Jan 2010 01:05:47 +0000 (17:05 -0800)]
refresh (create temporary patch)

14 years agoRefresh of dapl-pdata
Sean Hefty [Sat, 23 Jan 2010 01:05:44 +0000 (17:05 -0800)]
Refresh of dapl-pdata

14 years agonew
Sean Hefty [Sat, 23 Jan 2010 00:22:38 +0000 (16:22 -0800)]
new