commit f5c4834d93 upstream.
When reading the IPv6 addresses from the net-device, make sure to
avoid adding a duplicate entry to the GID table because of equality
between the default GID we generate and the default IPv6 link-local
address of the device.
Fixes: acc4fccf4e ("IB/mlx4: Make sure GID index 0 is always occupied")
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit e381835cf1 upstream.
When Ethernet netdev is not present for a port (e.g. when the link
layer type of the port is InfiniBand) it's possible to dereference a
null pointer when we do netdevice scanning.
To fix that, we move a section of code that needs to run only when
netdev is present to a proper if () statement.
Fixes: ad4885d279 ("IB/mlx4: Build the port IBoE GID table properly under bonding")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 85cbb7c728 upstream.
This particular reference count is not needed with the rcu protection,
and the current code leaks a reference count, causing a hang in
qib_qp_destroy().
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 7e6d3e5c70 upstream.
This patch addresses an issue where the legacy diagpacket is sent in
from the user, but the driver operates on only the extended
diagpkt. This patch specifically initializes the extended diagpkt
based on the legacy packet.
Reported-by: Rickard Strandqvist <rickard_strandqvist@spectrumdigital.se>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 911eccd284 upstream.
The code used a literal 1 in dispatching an IB_EVENT_PKEY_CHANGE.
As of the dual port qib QDR card, this is not necessarily correct.
Change to use the port as specified in the call.
Reported-by: Alex Estrin <alex.estrin@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 43bc889380 upstream.
The i386 ABI disagrees with most other ABIs regarding alignment of
data type larger than 4 bytes: on most ABIs a padding must be added at
end of the structures, while it is not required on i386.
So for most ABIs struct mlx5_ib_create_srq gets implicitly padded to be
aligned on a 8 bytes multiple, while for i386, such padding is not
added.
Tool pahole could be used to find such implicit padding:
$ pahole --anon_include \
--nested_anon_include \
--recursive \
--class_name mlx5_ib_create_srq \
drivers/infiniband/hw/mlx5/mlx5_ib.o
Then, structure layout can be compared between i386 and x86_64:
# +++ obj-i386/drivers/infiniband/hw/mlx5/mlx5_ib.o.pahole.txt 2014-03-28 11:43:07.386413682 +0100
# --- obj-x86_64/drivers/infiniband/hw/mlx5/mlx5_ib.o.pahole.txt 2014-03-27 13:06:17.788472721 +0100
# @@ -69,7 +68,6 @@ struct mlx5_ib_create_srq {
# __u64 db_addr; /* 8 8 */
# __u32 flags; /* 16 4 */
#
# - /* size: 20, cachelines: 1, members: 3 */
# - /* last cacheline: 20 bytes */
# + /* size: 24, cachelines: 1, members: 3 */
# + /* padding: 4 */
# + /* last cacheline: 24 bytes */
# };
ABI disagreement will make an x86_64 kernel try to read past
the buffer provided by an i386 binary.
When boundary check will be implemented, the x86_64 kernel will
refuse to read past the i386 userspace provided buffer and the
uverb will fail.
Anyway, if the structure lay in memory on a page boundary and
next page is not mapped, ib_copy_from_udata() will fail and the
uverb will fail.
This patch makes create_srq_user() takes care of the input
data size to handle the case where no padding was provided.
This way, x86_64 kernel will be able to handle struct mlx5_ib_create_srq
as sent by unpatched and patched i386 libmlx5.
Link: http://marc.info/?i=cover.1399309513.git.ydroneaud@opteya.com
Fixes: e126ba97db ("mlx5: Add driver for Mellanox Connect-IB adapter")
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit a8237b32a3 upstream.
The i386 ABI disagrees with most other ABIs regarding alignment of
data type larger than 4 bytes: on most ABIs a padding must be added at
end of the structures, while it is not required on i386.
So for most ABI struct mlx5_ib_create_cq get padded to be aligned on a
8 bytes multiple, while for i386, such padding is not added.
The tool pahole can be used to find such implicit padding:
$ pahole --anon_include \
--nested_anon_include \
--recursive \
--class_name mlx5_ib_create_cq \
drivers/infiniband/hw/mlx5/mlx5_ib.o
Then, structure layout can be compared between i386 and x86_64:
# +++ obj-i386/drivers/infiniband/hw/mlx5/mlx5_ib.o.pahole.txt 2014-03-28 11:43:07.386413682 +0100
# --- obj-x86_64/drivers/infiniband/hw/mlx5/mlx5_ib.o.pahole.txt 2014-03-27 13:06:17.788472721 +0100
# @@ -34,9 +34,8 @@ struct mlx5_ib_create_cq {
# __u64 db_addr; /* 8 8 */
# __u32 cqe_size; /* 16 4 */
#
# - /* size: 20, cachelines: 1, members: 3 */
# - /* last cacheline: 20 bytes */
# + /* size: 24, cachelines: 1, members: 3 */
# + /* padding: 4 */
# + /* last cacheline: 24 bytes */
# };
This ABI disagreement will make an x86_64 kernel try to read past the
buffer provided by an i386 binary.
When boundary check will be implemented, a x86_64 kernel will refuse
to read past the i386 userspace provided buffer and the uverb will
fail.
Anyway, if the structure lies in memory on a page boundary and next
page is not mapped, ib_copy_from_udata() will fail when trying to read
the 4 bytes of padding and the uverb will fail.
This patch makes create_cq_user() takes care of the input data size to
handle the case where no padding is provided.
This way, x86_64 kernel will be able to handle struct
mlx5_ib_create_cq as sent by unpatched and patched i386 libmlx5.
Link: http://marc.info/?i=cover.1399309513.git.ydroneaud@opteya.com
Fixes: e126ba97db ("mlx5: Add driver for Mellanox Connect-IB adapter")
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 37a967651c upstream.
Commit c804f07248 moved qib_assign_ctxt() to
do_qib_user_sdma_queue_create() but dropped the braces
around the statements.
This was spotted by coccicheck (coccinelle/spatch):
$ make C=2 CHECK=scripts/coccicheck drivers/infiniband/hw/qib/
CHECK drivers/infiniband/hw/qib/qib_file_ops.c
drivers/infiniband/hw/qib/qib_file_ops.c:1583:2-23: code aligned with following code on line 1587
This patch adds braces back.
Link: http://marc.info/?i=cover.1394485254.git.ydroneaud@opteya.com
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: infinipath@intel.com
Cc: Julia Lawall <julia.lawall@lip6.fr>
Cc: cocci@systeme.lip6.fr
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Tested-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Acked-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit f8b6c47a44 upstream.
The debugfs init code was incorrectly called before the idr mechanism
is used to get the unit number, so the dd->unit hasn't been
initialized. This caused the unit relative directory creation to fail
after the first.
This patch moves the init for the debugfs stuff until after all of the
failures and after the unit number has been determined.
A bug in unwind code in qib_alloc_devdata() is also fixed.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5bdb0f02ad upstream.
In case of error when writing to userspace, function ehca_create_cq()
does not set an error code before following its error path.
This patch sets the error code to -EFAULT when ib_copy_to_udata()
fails.
This was caught when using spatch (aka. coccinelle)
to rewrite call to ib_copy_{from,to}_udata().
Link: 75ebf2c103:ib_copy_udata.cocci
Link: http://marc.info/?i=cover.1394485254.git.ydroneaud@opteya.com
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 08e74c4b00 upstream.
In case of error when writing to userspace, the function mthca_create_cq()
does not set an error code before following its error path.
This patch sets the error code to -EFAULT when ib_copy_to_udata() fails.
This was caught when using spatch (aka. coccinelle)
to rewrite call to ib_copy_{from,to}_udata().
Link: 75ebf2c103:ib_copy_udata.cocci
Link: http://marc.info/?i=cover.1394485254.git.ydroneaud@opteya.com
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 9d194d1025 upstream.
In case of error while accessing to userspace memory, function
nes_create_qp() returns NULL instead of an error code wrapped through
ERR_PTR(). But NULL is not expected by ib_uverbs_create_qp(), as it
check for error with IS_ERR().
As page 0 is likely not mapped, it is going to trigger an Oops when
the kernel will try to dereference NULL pointer to access to struct
ib_qp's fields.
In some rare cases, page 0 could be mapped by userspace, which could
turn this bug to a vulnerability that could be exploited: the function
pointers in struct ib_device will be under userspace total control.
This was caught when using spatch (aka. coccinelle)
to rewrite calls to ib_copy_{from,to}_udata().
Link: https://www.gitorious.org/opteya/ib-hw-nes-create-qp-null
Link: 75ebf2c103:ib_copy_udata.cocci
Link: http://marc.info/?i=cover.1394485254.git.ydroneaud@opteya.com
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit a2cb0eb8a6 upstream.
Guard against a potential buffer overrun. The size to read from the
user is passed in, and due to the padding that needs to be taken into
account, as well as the place holder for the ICRC it is possible to
overflow the 32bit value which would cause more data to be copied from
user space than is allocated in the buffer.
Reported-by: Nico Golde <nico@ngolde.de>
Reported-by: Fabian Yamaguchi <fabs@goesec.de>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We should use rdma_vlan_dev_real_dev() instead of using vlan_dev_real_dev()
when building the GID table for a vlan interface.
Signed-off-by: Devesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Use correct value for obtaining traffic class from device
response for Query QP request.
Signed-off-by: Devesh Sharma <devesh.sharma@emulex.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Error reported at http://marc.info/?l=linux-rdma&m=138995755801039&w=2
Fix short to int cast for big endian systems.
Signed-off-by: Upinder Malhi <umalhi@cisco.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Remove Kconfig dependency of mlx5_ib/mlx5_core on X86, since there is
no such dependency in reality.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Research has shown that commit a77fcf8950 ("IB/qib: Use a single
txselect module parameter for serdes tuning") missed a key serdes init
sequence.
This patch add that sequence.
Cc: <stable@vger.kernel.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
For userspace RoCE UD QPs we need to know the GID format that the
kernel uses, e.g when working over older kernels. For that end, add a
new port capability IB_PORT_IP_BASED_GIDS and report it when query
port is issued.
Signed-off-by: Moni Shoua <monis@mellanox.co.il>
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
When scanning netdevices we need to check a few more conditions and
cases to build the IBoE GID table properly. For example, under
bonding we must make sure that when a port is down, the bond IP
address isn't programmed as a GID, since doing so will cause failure
with IB core flows that selects ports by GID.
Signed-off-by: Moni Shoua <monis@mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The IBoE code used to reset the GID table did it for all Ethernet
ports of the device. Since the whole architecture of generating GIDs
and responding to events is port-based, this is inefficient and can
lead to wrong content in the GID table. Change the reset flow to be
per-port.
Signed-off-by: Moni Shoua <monis@mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Updating the GID table under IBoE requires read/write from/to shared
data structures. These data structures are protected with the device
iboe lock. The flows that modify the GID table start from
1. Initializing the GID table
2. NETDEV events
3. INET or INET6 events
This patch makes sure that the flow of initializing the GID table is
consistent with the other two flows w.r.t on what step the lock is taken.
Signed-off-by: Moni Shoua <monis@mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
On the one hand, the invocation of netdev_master_upper_dev_get()
within mlx4_ib_scan_netdevs() must be done with rtnl lock held. On
the other hand, it's wrong to call rtnl_lock() from within this
function since it's also called by our netdev notifier callback.
Therefore move the locking to mlx4_ib_add() so that both cases are
covered.
Signed-off-by: Moni Shoua <monis@mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Make sure that for Ethernet ports, the port GID table index 0 is always
occupied with a default GID of the relevant IPv6 link-local adderss.
This provides better user experience for legacy applications that don't use
the RDMA CM and were working on index 0 prior to the IP addressing change.
Also, as GIDs are generated from IP addresses of the network devices that
are associated with the port, it's basically possible that the GID table
will be empty if no IP address was assigned. This doesn't comply with the
IB spec section 4.1.1 "GID usage and properties".
Signed-off-by: Moni Shoua <monis@mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
When the device has only Ethernet ports, don't try to allocate range
of steerable UD QPs since they aren't needed. This fixes an issue
where mlx4 VFs tried to allocate a range of UD steerable QPs, but
failed to do so.
Fixes: c1c9850112 ("IB/mlx4: Add support for steerable IB UD QPs")
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Set the return variable to an error code as done elsewhere in the function.
A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)
// <smpl>
(
if@p1 (\(ret < 0\|ret != 0\))
{ ... return ret; }
|
ret@p1 = 0
)
... when != ret = e1
when != &ret
*if(...)
{
... when != ret = e2
when forall
return ret;
}
// </smpl>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Set the return variable to an error code as done elsewhere in the function.
A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)
// <smpl>
(
if@p1 (\(ret < 0\|ret != 0\))
{ ... return ret; }
|
ret@p1 = 0
)
... when != ret = e1
when != &ret
*if(...)
{
... when != ret = e2
when forall
return ret;
}
// </smpl>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Currently Connect-IB does not support blocking multicast loopback, so
don't set IB_DEVICE_BLOCK_MULTICAST_LOOPBACK in the device caps.
Reported by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Commit c1be5232d2 ("Fix micro UAR allocator") broke binary compatibility
between libmlx5 and mlx5_ib since it defines a different value to the number
of micro UARs per page, leading to wrong calculation in libmlx5. This patch
defines struct mlx5_ib_alloc_ucontext_req_v2 as an extension to struct
mlx5_ib_alloc_ucontext_req. The extended size is determined in mlx5_ib_alloc_ucontext()
and in case of old library we use uuarn 0 which works fine -- this is
acheived due to create_user_qp() falling back from high to medium then to
low class where low class will return 0. For new libraries we use the
more sophisticated allocation algorithm.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Fix the RC QPs send queue overhead computation to take into account
two additional segments in the WQE which are needed for registration
operations.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Verify that reserved fields in struct mlx5_ib_resize_cq are cleared
before continuing execution of the verb. This is required to allow
making use of this area in future revisions.
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The current code attmepts to call ib_umem_get() even if the length is
zero, which causes a failure. Since the spec allows zero length work
queues, change the code so we don't call ib_umem_get() in those cases.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Implement resize CQ which is a mandatory verb in mlx5.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Modify CQ is used by ULPs like IPoIB to change moderation parameters. This
patch adds support in mlx5.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Put a wmb() to make sure the doorbell record is visible to the HCA before we
hit doorbell.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Use the possibly more efficient ether_addr_equal() instead of memcmp().
Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The GSI QP type is compatible with and should be allowed to send data
to/from any UD QP. This was found when testing ibacm on the same node
as an SA.
Cc: <stable@vger.kernel.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Building mem.o for 32 bits x86 triggers a GCC warning:
drivers/infiniband/hw/cxgb4/mem.c: In function '_c4iw_write_mem_dma_aligned':
drivers/infiniband/hw/cxgb4/mem.c:79:25: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
Silence that warning by casting "&wr_wait" to unsigned long before
casting it to __be64. That's what _c4iw_write_mem_inline() already does.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Remove including <linux/version.h> that don't need it.
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Roland Dreier <roland@purestorage.com>
There is a path in handle_vq() where kmem_cache_free() can be called
with pointer to a local variable. It can happen if vq_repbuf_alloc()
failed to allocate memory from cache and req is NULL.
The patch adds check if cache memory was allocated before freeing it.
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Svetlana Mavrina <another.karnil@gmail.com>
Reviewed-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This is called from qp_grp_and_vf_bind() and we are holding the
vf->lock so the allocation can't sleep.
Fixes: e3cf00d0a8 ('IB/usnic: Add Cisco VIC low-level hardware driver')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Now that ocrdma supports IP-based addressing, we need to depend on
INET, since ocrdma registers itself for net device events.
Signed-off-by: Roland Dreier <roland@purestorage.com>
Since mlx4_ib supports IP based addressing, a dependency on INET needs
to be added, since mlx4_ib registers itself for net device events.
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Moni Shoua
Mike Marciniszyn
Dennis Dalessandro
Yann Droneaud
Amir Vadai
Roland Dreier
Devesh Sharma
Upinder Malhi
Eli Cohen
Kumar Sanghvi
Moni Shoua
Matan Barak
Julia Lawall
Eli Cohen
Ding Tianhong
Ira Weiny
Paul Bolle
Wei Yongjun
Svetlana Mavrina
Dan Carpenter