Commit 8edf695c authored May 09, 2023 by Jason Gunthorpe Committed by Jialin Zhang May 10, 2023

RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests

mainline inclusion
from mainline-v5.15-rc4
commit 305d568b
category: bugfix
bugzilla: https://gitee.com/src-openeuler/kernel/issues/I6X49E
CVE: CVE-2023-2176

Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=305d568b72f17f674155a2a8275f865f207b3808

---------------------------

The FSM can run in a circle allowing rdma_resolve_ip() to be called twice
on the same id_priv. While this cannot happen without going through the
work, it violates the invariant that the same address resolution
background request cannot be active twice.

       CPU 1                                  CPU 2

rdma_resolve_addr():
  RDMA_CM_IDLE -> RDMA_CM_ADDR_QUERY
  rdma_resolve_ip(addr_handler)  #1

			 process_one_req(): for #1
                          addr_handler():
                            RDMA_CM_ADDR_QUERY -> RDMA_CM_ADDR_BOUND
                            mutex_unlock(&id_priv->handler_mutex);
                            [.. handler still running ..]

rdma_resolve_addr():
  RDMA_CM_ADDR_BOUND -> RDMA_CM_ADDR_QUERY
  rdma_resolve_ip(addr_handler)
    !! two requests are now on the req_list

rdma_destroy_id():
 destroy_id_handler_unlock():
  _destroy_id():
   cma_cancel_operation():
    rdma_addr_cancel()

                          // process_one_req() self removes it
		          spin_lock_bh(&lock);
                           cancel_delayed_work(&req->work);
	                   if (!list_empty(&req->list)) == true

      ! rdma_addr_cancel() returns after process_on_req #1 is done

   kfree(id_priv)

			 process_one_req(): for #2
                          addr_handler():
	                    mutex_lock(&id_priv->handler_mutex);
                            !! Use after free on id_priv

rdma_addr_cancel() expects there to be one req on the list and only
cancels the first one. The self-removal behavior of the work only happens
after the handler has returned. This yields a situations where the
req_list can have two reqs for the same "handle" but rdma_addr_cancel()
only cancels the first one.

The second req remains active beyond rdma_destroy_id() and will
use-after-free id_priv once it inevitably triggers.

Fix this by remembering if the id_priv has called rdma_resolve_ip() and
always cancel before calling it again. This ensures the req_list never
gets more than one item in it and doesn't cost anything in the normal flow
that never uses this strange error path.

Link: https://lore.kernel.org/r/0-v1-3bc675b8006d+22-syz_cancel_uaf_jgg@nvidia.com


Cc: stable@vger.kernel.org
Fixes: e51060f0 ("IB: IP address based RDMA connection manager")
Reported-by:  <syzbot+dc3dfba010d7671e05f5@syzkaller.appspotmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
(cherry picked from commit 305d568b)
Signed-off-by: Liu Jian <liujian56@huawei.com>

Conflicts:
	drivers/infiniband/core/cma_priv.h
Reviewed-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Wang Weiyang <wangweiyang2@huawei.com>
Signed-off-by: Jialin Zhang <zhangjialin11@huawei.com>

parent 7afa2650

drivers/infiniband/core/cma.c

+23 −0

Original line number	Diff line number	Diff line
		@@ -1793,6 +1793,14 @@ static void cma_cancel_operation(struct rdma_id_private *id_priv,
		{
		switch (state) {
		case RDMA_CM_ADDR_QUERY:
		/*
		* We can avoid doing the rdma_addr_cancel() based on state,
		* only RDMA_CM_ADDR_QUERY has a work that could still execute.
		* Notice that the addr_handler work could still be exiting
		* outside this state, however due to the interaction with the
		* handler_mutex the work is guaranteed not to touch id_priv
		* during exit.
		*/
		rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
		break;
		case RDMA_CM_ROUTE_QUERY:
		@@ -3402,6 +3410,21 @@ int rdma_resolve_addr(struct rdma_cm_id id, struct sockaddr src_addr,
		if (dst_addr->sa_family == AF_IB) {
		ret = cma_resolve_ib_addr(id_priv);
		} else {
		/*
		* The FSM can return back to RDMA_CM_ADDR_BOUND after
		* rdma_resolve_ip() is called, eg through the error
		* path in addr_handler(). If this happens the existing
		* request must be canceled before issuing a new one.
		* Since canceling a request is a bit slow and this
		* oddball path is rare, keep track once a request has
		* been issued. The track turns out to be a permanent
		* state since this is the only cancel as it is
		* immediately before rdma_resolve_ip().
		*/
		if (id_priv->used_resolve_ip)
		rdma_addr_cancel(&id->route.addr.dev_addr);
		else
		id_priv->used_resolve_ip = 1;
		ret = rdma_resolve_ip(cma_src_addr(id_priv), dst_addr,
		&id->route.addr.dev_addr,
		timeout_ms, addr_handler,

drivers/infiniband/core/cma_priv.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -89,6 +89,7 @@ struct rdma_id_private {
		u8 reuseaddr;
		u8 afonly;
		u8 timeout;
		u8 used_resolve_ip;
		enum ib_gid_type gid_type;

		/*