linux/drivers/infiniband/core
Jason Gunthorpe 305d568b72 RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
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: e51060f08a ("IB: IP address based RDMA connection manager")
Reported-by: syzbot+dc3dfba010d7671e05f5@syzkaller.appspotmail.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
2021-09-23 17:03:09 -03:00
..
addr.c RDMA/addr: Be strict with gid size 2021-04-08 16:14:56 -03:00
agent.c
agent.h
cache.c IB/core: Shifting initialization of device->cache_lock 2021-07-16 10:57:28 -03:00
cgroup.c
cm_msgs.h RDMA/core: Add necessary spaces 2021-04-12 14:52:22 -03:00
cm_trace.c RDMA/cm: Replace pr_debug() call sites with tracepoints 2020-08-24 19:41:41 -03:00
cm_trace.h RDMA/cm: Add tracepoints to track MAD send operations 2020-08-24 19:41:41 -03:00
cm.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
cma_configfs.c RDMA: Support more than 255 rdma ports 2021-03-26 09:31:21 -03:00
cma_priv.h RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests 2021-09-23 17:03:09 -03:00
cma_trace.c
cma_trace.h RDMA/core: Move the rdma_show_ib_cm_event() macro 2020-08-24 16:01:47 -03:00
cma.c RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests 2021-09-23 17:03:09 -03:00
core_priv.h RDMA/core: Create clean QP creations interface for uverbs 2021-08-03 15:26:19 -03:00
counters.c RDMA: Split the alloc_hw_stats() ops to port and device variants 2021-06-16 20:58:29 -03:00
cq.c RDMA/core: Clean up cq pool mechanism 2020-12-10 15:05:17 -04:00
device.c RDMA: Globally allocate and release QP memory 2021-08-03 13:44:27 -03:00
ib_core_uverbs.c
iwcm.c RDMA/iwcm: Release resources if iw_cm module initialization fails 2021-07-30 10:01:40 -03:00
iwcm.h RDMA/core: Use refcount_t instead of atomic_t on refcount of iwcm_id_private 2021-06-08 14:35:44 -03:00
iwpm_msg.c RDMA/iwpm: Rely on the rdma_nl_[un]register() to ensure that requests are valid 2021-07-30 10:01:41 -03:00
iwpm_util.c RDMA/iwpm: Rely on the rdma_nl_[un]register() to ensure that requests are valid 2021-07-30 10:01:41 -03:00
iwpm_util.h RDMA/iwpm: Rely on the rdma_nl_[un]register() to ensure that requests are valid 2021-07-30 10:01:41 -03:00
lag.c RDMA/core: Consider flow label when building skb 2020-05-06 16:51:43 -03:00
mad_priv.h RDMA/core: Remove refcount from struct ib_mad_snoop_private 2021-06-08 14:43:28 -03:00
mad_rmpp.c RDMA/core: Remove redundant spaces 2021-04-12 14:56:48 -03:00
mad_rmpp.h
mad.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
Makefile RDMA/umem: Support importing dma-buf as user memory region 2021-01-20 16:07:52 -04:00
mr_pool.c
multicast.c RDMA/core: Use refcount_t instead of atomic_t on refcount of mcast_port 2021-06-08 14:45:07 -03:00
netlink.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
nldev.c RDMA/core: Replace the ib_port_data hw_stats pointers with a ib_port pointer 2021-06-16 20:58:29 -03:00
opa_smi.h RDMA: Support more than 255 rdma ports 2021-03-26 09:31:21 -03:00
packer.c
rdma_core.c RDMA/core: Correct misspellings of two words in comments 2021-03-26 11:58:26 -03:00
rdma_core.h IB/uverbs: Introduce create/destroy QP commands over ioctl 2020-05-21 20:39:36 -03:00
restrack.c RDMA: Globally allocate and release QP memory 2021-08-03 13:44:27 -03:00
restrack.h RDMA/restrack: Improve readability in task name management 2020-09-22 19:47:35 -03:00
roce_gid_mgmt.c RDMA: Fix kernel-doc warnings about wrong comment 2021-06-21 20:32:50 -03:00
rw.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
sa_query.c RDMA/core/sa_query: Retry SA queries 2021-08-25 13:42:47 -03:00
sa.h RDMA: Support more than 255 rdma ports 2021-03-26 09:31:21 -03:00
security.c IB/core: Removed port validity check from ib_get_cached_subnet_prefix 2021-06-21 20:49:32 -03:00
smi.c RDMA: Support more than 255 rdma ports 2021-03-26 09:31:21 -03:00
smi.h RDMA: Support more than 255 rdma ports 2021-03-26 09:31:21 -03:00
sysfs.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
trace.c RDMA/core: Clean up tracepoint headers 2020-07-06 14:54:46 -03:00
ucma.c RDMA/core: Use the DEVICE_ATTR_RO macro 2021-05-28 20:39:51 -03:00
ud_header.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
umem_dmabuf.c RDMA: Use the sg_table directly and remove the opencoded version from umem 2021-08-24 19:52:40 -03:00
umem_odp.c IB/core: Remove deprecated current_seq comments 2021-08-23 13:43:08 -03:00
umem.c RDMA: Use the sg_table directly and remove the opencoded version from umem 2021-08-24 19:52:40 -03:00
user_mad.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
uverbs_cmd.c RDMA/core: Create clean QP creations interface for uverbs 2021-08-03 15:26:19 -03:00
uverbs_ioctl.c IB/core: Split uverbs_get_const/default to consider target type 2021-03-11 20:20:36 -04:00
uverbs_main.c RDMA/core: Use refcount_t instead of atomic_t on refcount of ib_uverbs_device 2021-06-08 15:04:36 -03:00
uverbs_marshall.c
uverbs_std_types_async_fd.c RDMA/core: Make FD destroy callback void 2020-11-12 12:32:17 -04:00
uverbs_std_types_counters.c RDMA/core: Postpone uobject cleanup on failure till FD close 2020-11-12 12:32:17 -04:00
uverbs_std_types_cq.c RDMA/core: Postpone uobject cleanup on failure till FD close 2020-11-12 12:32:17 -04:00
uverbs_std_types_device.c RDMA/uverbs: Fix a NULL vs IS_ERR() bug 2021-05-19 15:32:07 -03:00
uverbs_std_types_dm.c RDMA/core: Postpone uobject cleanup on failure till FD close 2020-11-12 12:32:17 -04:00
uverbs_std_types_flow_action.c RDMA/core: Postpone uobject cleanup on failure till FD close 2020-11-12 12:32:17 -04:00
uverbs_std_types_mr.c RDMA/uverbs: Track dmabuf memory regions 2021-08-19 09:59:53 -03:00
uverbs_std_types_qp.c RDMA/core: Create clean QP creations interface for uverbs 2021-08-03 15:26:19 -03:00
uverbs_std_types_srq.c RDMA/core: Postpone uobject cleanup on failure till FD close 2020-11-12 12:32:17 -04:00
uverbs_std_types_wq.c RDMA/core: Postpone uobject cleanup on failure till FD close 2020-11-12 12:32:17 -04:00
uverbs_std_types.c RDMA/core: Make FD destroy callback void 2020-11-12 12:32:17 -04:00
uverbs_uapi.c RDMA/core: Fix incorrect print format specifier 2021-06-21 15:38:30 -03:00
uverbs.h RDMA/core: Use refcount_t instead of atomic_t on refcount of ib_uverbs_device 2021-06-08 15:04:36 -03:00
verbs.c RDMA/core: Create clean QP creations interface for uverbs 2021-08-03 15:26:19 -03:00