[PATCH 1/2] iscsi_tcp: Fix NULL pointer dereference in iscsi_sw_tcp_conn_get_param()

[Wenchao Hao]

kernel might crash in iscsi_sw_tcp_conn_get_param() because it dereference
an invalid address.

The initialization of iscsi_conn's dd_data is after device_register() of
struct iscsi_cls_conn, so iscsi_conn's dd_data might not initialized when
iscsi_sw_tcp_conn_get_param() is called.

Following stack would be reported and kernel would panic.

[449311.812887] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008
[449311.812893] Mem abort info:
[449311.812895] ESR = 0x96000004
[449311.812899] Exception class = DABT (current EL), IL = 32 bits
[449311.812901] SET = 0, FnV = 0
[449311.812903] EA = 0, S1PTW = 0
[449311.812905] Data abort info:
[449311.812907] ISV = 0, ISS = 0x00000004
[449311.812909] CM = 0, WnR = 0
[449311.812915] user pgtable: 4k pages, 48-bit VAs, pgdp = 00000000e26e7ace
[449311.812918] [0000000000000008] pgd=0000000000000000
[449311.812925] Internal error: Oops: 96000004 [#1] SMP
[449311.814974] Process iscsiadm (pid: 8286, stack limit = 0xffff800010f78000)
[449311.815570] CPU: 0 PID: 8286 Comm: iscsiadm Kdump: loaded Tainted: G B W 4.19.90-vhulk2201.1.0.h1021.kasan.eulerosv2r10.aarch64 #1
[449311.816584] sd 1:0:0:1: [sdg] Attached SCSI disk
[449311.816695] Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
[449311.817677] pstate: 40400005 (nZcv daif +PAN -UAO)
[449311.818121] pc : iscsi_sw_tcp_conn_get_param+0xec/0x300 [iscsi_tcp]
[449311.818688] lr : iscsi_sw_tcp_conn_get_param+0xe8/0x300 [iscsi_tcp]
[449311.819244] sp : ffff800010f7f890
[449311.819542] x29: ffff800010f7f890 x28: ffff8000cb1bea38
[449311.820025] x27: ffff800010911010 x26: ffff2000028887a4
[449311.820500] x25: ffff800009200d98 x24: ffff800010911000
[449311.820973] x23: 0000000000000000 x22: ffff8000cb1bea28
[449311.821458] x21: 0000000000000015 x20: ffff200081afa000
[449311.821934] x19: 1ffff000021eff20 x18: 0000000000000000
[449311.822414] x17: 0000000000000000 x16: ffff200080618220
[449311.822891] x15: 0000000000000000 x14: 0000000000000000
[449311.823413] x13: 0000000000000000 x12: 0000000000000000
[449311.823897] x11: 1ffff0001ab4f41f x10: ffff10001ab4f41f
[449311.824373] x9 : 0000000000000000 x8 : ffff8000d5a7a100
[449311.824847] x7 : 0000000000000000 x6 : dfff200000000000
[449311.825329] x5 : ffff1000021eff20 x4 : ffff8000cb1bea30
[449311.825806] x3 : ffff200002911178 x2 : ffff2000841ff000
[449311.826281] x1 : e0c234eab8420c00 x0 : ffff8000cb1bea38
[449311.826756] Call trace:
[449311.826987] iscsi_sw_tcp_conn_get_param+0xec/0x300 [iscsi_tcp]
[449311.827550] show_conn_ep_param_ISCSI_PARAM_CONN_ADDRESS+0xe4/0x100 [scsi_transport_iscsi]
[449311.828304] dev_attr_show+0x58/0xb0
[449311.828639] sysfs_kf_seq_show+0x124/0x210
[449311.829014] kernfs_seq_show+0x8c/0xa0
[449311.829362] seq_read+0x188/0x8a0
[449311.829667] kernfs_fop_read+0x250/0x398
[449311.830024] __vfs_read+0xe0/0x350
[449311.830339] vfs_read+0xbc/0x1c0
[449311.830635] ksys_read+0xdc/0x1b8
[449311.830941] __arm64_sys_read+0x50/0x60
[449311.831295] el0_svc_common+0xc8/0x320
[449311.831642] el0_svc_handler+0xf8/0x160
[449311.831998] el0_svc+0x10/0x218
[449311.832292] Code: f94006d7 910022e0 940007bb aa1c03e0 (f94006f9)

Signed-off-by: Wenchao Hao <haowe...@huawei.com>
Signed-off-by: Wu Bo <wub...@huawei.com>
---
drivers/scsi/iscsi_tcp.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/scsi/iscsi_tcp.c b/drivers/scsi/iscsi_tcp.c
index 1bc37593c88f..14db224486be 100644
--- a/drivers/scsi/iscsi_tcp.c
+++ b/drivers/scsi/iscsi_tcp.c
@@ -741,11 +741,16 @@ static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
{
struct iscsi_conn *conn = cls_conn->dd_data;
struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
- struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+ struct iscsi_sw_tcp_conn *tcp_sw_conn;
struct sockaddr_in6 addr;
struct socket *sock;
int rc;

+ if (!tcp_conn)
+ return -ENOTCONN;
+
+ tcp_sw_conn = tcp_conn->dd_data;
+
switch(param) {
case ISCSI_PARAM_CONN_PORT:
case ISCSI_PARAM_CONN_ADDRESS:
--
2.32.0

[Wenchao Hao]

iscsi_create_conn() would add newly alloced iscsi_cls_conn to connlist,
it means when userspace sends ISCSI_UEVENT_SET_PARAM, iscsi_conn_lookup()
would found this iscsi_cls_conn and call the set_param callback which is
iscsi_sw_tcp_conn_set_param(). While the iscsi_conn's dd_data might not
been initialized.

Signed-off-by: Wenchao Hao <haowe...@huawei.com>
Signed-off-by: Wu Bo <wub...@huawei.com>
---

drivers/scsi/iscsi_tcp.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/scsi/iscsi_tcp.c b/drivers/scsi/iscsi_tcp.c
index 14db224486be..a42449df6156 100644
--- a/drivers/scsi/iscsi_tcp.c
+++ b/drivers/scsi/iscsi_tcp.c
@@ -716,13 +716,17 @@ static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,

{
struct iscsi_conn *conn = cls_conn->dd_data;
struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
- struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+ struct iscsi_sw_tcp_conn *tcp_sw_conn;

switch(param) {
case ISCSI_PARAM_HDRDGST_EN:
iscsi_set_param(cls_conn, param, buf, buflen);
break;
case ISCSI_PARAM_DATADGST_EN:
+ if (!tcp_conn || !tcp_conn->dd_data)

+ return -ENOTCONN;
+
+ tcp_sw_conn = tcp_conn->dd_data;

iscsi_set_param(cls_conn, param, buf, buflen);
tcp_sw_conn->sendpage = conn->datadgst_en ?
sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
--
2.32.0

[Mike Christie]

Is this something you hit or from code review?

We have those state checks:

if ((conn->state == ISCSI_CONN_BOUND) ||
(conn->state == ISCSI_CONN_UP)) {
err = transport->set_param(conn, ev->u.set_param.param,

so we don't call set_param until after we have bound the
connection which will be after ISCSI_UEVENT_CREATE_CONN has returned.

Also for this specific bug iscsi_if_recv_msg is called with the
rx_queue_mutex, so set_param can only be called after the
ISCSI_UEVENT_CREATE_CONN cmd has returned.

[Mike Christie]

On 3/3/22 8:56 PM, Wenchao Hao wrote:

We are actually doing sysfs/device addition wrong.

We should be doing the 2 step setup where in step 1 we alloc/init.
When everything is allocated and initialized, then we should do
device_add which exposes us to sysfs. On the teardown side, we are
then supposed to do 2 steps where the remove function does device_del
which waits until sysfs accesses are completed. We can then tear
the structs down and free them and call device_put.

The exposure to NL would be similar where it goes into the wrapper
around device_add. However, see my comments on the other patch where
I don't think we can hit the bug you mention because every nl cmd
that calls into the drivers is done under the rx_queue_mutex.

I think we should separate the iscsi_create_conn function like we
do for sessions. This is going to be a little more involved because
you need to also convert iscsi_tcp_conn_setup and the drivers since
we can call into the drivers for the get_conn_param callout.

[Wenchao Hao]

It's from code review. I reviewed the code because the panic mentioned
in my first patch. The issue seems would not happen, so just ignore it.

> We have those state checks:
>
> if ((conn->state == ISCSI_CONN_BOUND) ||
> (conn->state == ISCSI_CONN_UP)) {
> err = transport->set_param(conn, ev->u.set_param.param,
>
> so we don't call set_param until after we have bound the
> connection which will be after ISCSI_UEVENT_CREATE_CONN has returned.
>
> Also for this specific bug iscsi_if_recv_msg is called with the
> rx_queue_mutex, so set_param can only be called after the
> ISCSI_UEVENT_CREATE_CONN cmd has returned.

> .
>

[Wenchao Hao]

I agree with this, and I would try to split device_add() from
iscsi_create_conn().

What's more I would do some check between sysfs files add/remove and
kernel object initialize/release to make a micro-refactoring

> The exposure to NL would be similar where it goes into the wrapper
> around device_add. However, see my comments on the other patch where
> I don't think we can hit the bug you mention because every nl cmd
> that calls into the drivers is done under the rx_queue_mutex.
>
> I think we should separate the iscsi_create_conn function like we
> do for sessions. This is going to be a little more involved because
> you need to also convert iscsi_tcp_conn_setup and the drivers since
> we can call into the drivers for the get_conn_param callout.

> .
>

[Wenchao Hao]

On 2022/3/3 23:03, Mike Christie wrote:

> On 3/3/22 8:56 PM, Wenchao Hao wrote:
>> kernel might crash in iscsi_sw_tcp_conn_get_param() because it dereference
>> an invalid address.
>>
>> The initialization of iscsi_conn's dd_data is after device_register() of
>> struct iscsi_cls_conn, so iscsi_conn's dd_data might not initialized when
>> iscsi_sw_tcp_conn_get_param() is called.
>>
>

> We are actually doing sysfs/device addition wrong.
>
> We should be doing the 2 step setup where in step 1 we alloc/init.
> When everything is allocated and initialized, then we should do
> device_add which exposes us to sysfs. On the teardown side, we are
> then supposed to do 2 steps where the remove function does device_del
> which waits until sysfs accesses are completed. We can then tear
> the structs down and free them and call device_put.
>

I reviewed the teardown flow of iscsi_cls_conn, it has already written
as what you saied.

> The exposure to NL would be similar where it goes into the wrapper
> around device_add. However, see my comments on the other patch where
> I don't think we can hit the bug you mention because every nl cmd
> that calls into the drivers is done under the rx_queue_mutex.
>
> I think we should separate the iscsi_create_conn function like we
> do for sessions. This is going to be a little more involved because
> you need to also convert iscsi_tcp_conn_setup and the drivers since
> we can call into the drivers for the get_conn_param callout.

> .
>

I hesitated about when should we call device_add(). I think there are
two places to call it.

The first one is in iscsi_conn_setup(), after some initialization of
conn, it keeps same with previous's implement and need not to change
drivers' code. What's more, the change can fix iscsi_tcp's NULL pointer
access. While this change can not make sure the LLDs related sources
are already initialized when iscsi_cls_conn is exposed to sysfs. It
means LLDs' callback are still responsible to check if the resources are
accessible.

Another one is in create_conn callback for each driver's
iscsi_transport. This need us to change each driver's code.

I send 2 patches which make changes in iscsi_conn_setup(), it's ok with
iscsi_tcp, would you help to review them?