Skip to contentSkip to conversation

v5.15-rc1 issue with the soft-iWARP driver

DISCUSSION Bart Van Assche 3 participants 3 messages Sep 15, 2021 – Sep 16, 2021 InfiniBand

lore
Share & tools
lore
#0 lore raw

Hi,

If I run test srp/015 from the blktests suite then the following appears in the kernel log:

==================================================================

BUG: KASAN: null-ptr-deref in __list_del_entry_valid+0x4d/0xe0
Read of size 8 at addr 0000000000000000 by task kworker/u16:3/161

CPU: 5 PID: 161 Comm: kworker/u16:3 Not tainted 5.15.0-rc1-dbg+ #2
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a-rebuilt.opensuse.org 04/01/2014
Workqueue: iw_cm_wq cm_work_handler [iw_cm]
Call Trace:
 show_stack+0x52/0x58
 dump_stack_lvl+0x5b/0x82
 kasan_report.cold+0x52/0x57
 __asan_load8+0x69/0x90
 __list_del_entry_valid+0x4d/0xe0
 _cma_cancel_listens+0x49/0x230 [rdma_cm]
 _destroy_id+0x4e/0x420 [rdma_cm]
 destroy_id_handler_unlock+0xc4/0x200 [rdma_cm]
 iw_conn_req_handler+0x335/0x370 [rdma_cm]
 cm_conn_req_handler+0x546/0x7d0 [iw_cm]
 cm_work_handler+0x419/0x480 [iw_cm]
 process_one_work+0x59d/0xb00
 worker_thread+0x8f/0x5c0
 kthread+0x1fc/0x230
 ret_from_fork+0x1f/0x30
==================================================================

I think this is a regression. This happened with commit a17a1faf5d3e ("RDMA/cma: Fix listener leak in rdma_cma_listen_on_all() failure").

Thanks,

Bart.

maintainer Reviewed-by Leon Romanovsky
#1 lore raw
Quoted 30 lines · Bart Van Assche

Hi,

If I run test srp/015 from the blktests suite then the following appears in the kernel log:

==================================================================

BUG: KASAN: null-ptr-deref in __list_del_entry_valid+0x4d/0xe0
Read of size 8 at addr 0000000000000000 by task kworker/u16:3/161

CPU: 5 PID: 161 Comm: kworker/u16:3 Not tainted 5.15.0-rc1-dbg+ #2
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a-rebuilt.opensuse.org 04/01/2014
Workqueue: iw_cm_wq cm_work_handler [iw_cm]
Call Trace:
 show_stack+0x52/0x58
 dump_stack_lvl+0x5b/0x82
 kasan_report.cold+0x52/0x57
 __asan_load8+0x69/0x90
 __list_del_entry_valid+0x4d/0xe0
 _cma_cancel_listens+0x49/0x230 [rdma_cm]
 _destroy_id+0x4e/0x420 [rdma_cm]
 destroy_id_handler_unlock+0xc4/0x200 [rdma_cm]
 iw_conn_req_handler+0x335/0x370 [rdma_cm]
 cm_conn_req_handler+0x546/0x7d0 [iw_cm]
 cm_work_handler+0x419/0x480 [iw_cm]
 process_one_work+0x59d/0xb00
 worker_thread+0x8f/0x5c0
 kthread+0x1fc/0x230
 ret_from_fork+0x1f/0x30
==================================================================
Quoted 2 lines

I think this is a regression. This happened with commit a17a1faf5d3e ("RDMA/cma: Fix listener leak in rdma_cma_listen_on_all() failure").

Gurk, this is a horrific mess, these list heads are multiplexing between 4 different roles depending on stuff. Thanks Bart

So let's go closer to the original patch:

From ca465e1f1f9b38fe916a36f7d80c5d25f2337c81 Mon Sep 17 00:00:00 2001
From: Tao Liu <[email protected]>
Date: Mon, 13 Sep 2021 17:33:44 +0800
Subject: [PATCH] RDMA/cma: Fix listener leak in rdma_cma_listen_on_all()
failure

If cma_listen_on_all() fails it leaves the per-device ID still on the listen_list but the state is not set to RDMA_CM_ADDR_BOUND.

When the cmid is eventually destroyed cma_cancel_listens() is not called due to the wrong state, however the per-device IDs are still holding the refcount preventing the ID from being destroyed, thus deadlocking:

task:rping state:D stack: 0 pid:19605 ppid: 47036 flags:0x00000084

 Call Trace:
  __schedule+0x29a/0x780
  ? free_unref_page_commit+0x9b/0x110
  schedule+0x3c/0xa0
  schedule_timeout+0x215/0x2b0
  ? __flush_work+0x19e/0x1e0
  wait_for_completion+0x8d/0xf0
  _destroy_id+0x144/0x210 [rdma_cm]
  ucma_close_id+0x2b/0x40 [rdma_ucm]
  __destroy_id+0x93/0x2c0 [rdma_ucm]
  ? __xa_erase+0x4a/0xa0
  ucma_destroy_id+0x9a/0x120 [rdma_ucm]
  ucma_write+0xb8/0x130 [rdma_ucm]
  vfs_write+0xb4/0x250
  ksys_write+0xb5/0xd0
  ? syscall_trace_enter.isra.19+0x123/0x190
  do_syscall_64+0x33/0x40
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Ensure that cma_listen_on_all() atomically unwinds its action under the lock during error.

Fixes c80a0c52d85c ("RDMA/cma: Add missing error handling of listen_id")
---
 drivers/infiniband/core/cma.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
drivers/infiniband/core/cma.c +10 -3
index 86ee3b01b3ee47..5aa58897965df4 100644
@@ -1746,15 +1746,16 @@ static void cma_cancel_route(struct rdma_id_private *id_priv)
 	}
 }
 
-static void cma_cancel_listens(struct rdma_id_private *id_priv)
+static void _cma_cancel_listens(struct rdma_id_private *id_priv)
 {
 	struct rdma_id_private *dev_id_priv;
 
+	lockdep_assert_held(&lock);
+
 	/*
 	 * Remove from listen_any_list to prevent added devices from spawning
 	 * additional listen requests.
 	 */
-	mutex_lock(&lock);
 	list_del(&id_priv->list);
 
 	while (!list_empty(&id_priv->listen_list)) {
@@ -1768,6 +1769,12 @@ static void cma_cancel_listens(struct rdma_id_private *id_priv)
 		rdma_destroy_id(&dev_id_priv->id);
 		mutex_lock(&lock);
 	}
+}
+
+static void cma_cancel_listens(struct rdma_id_private *id_priv)
+{
+	mutex_lock(&lock);
+	_cma_cancel_listens(id_priv);
 	mutex_unlock(&lock);
 }
 
@@ -2579,7 +2586,7 @@ static int cma_listen_on_all(struct rdma_id_private *id_priv)
 	return 0;
 
 err_listen:
-	list_del(&id_priv->list);
+	_cma_cancel_listens(id_priv);
 	mutex_unlock(&lock);
 	if (to_destroy)
 		rdma_destroy_id(&to_destroy->id);
Signature · 1 line
2.33.0
#2 lore raw
In reply to Bart Van Assche

-----"Bart Van Assche" <[email protected]> wrote: -----

Quoted 37 lines

To: [email protected], "Jason Gunthorpe" <[email protected]>
From: "Bart Van Assche" <[email protected]>
Date: 09/15/2021 06:54AM
Subject: [EXTERNAL] v5.15-rc1 issue with the soft-iWARP driver

Hi,

If I run test srp/015 from the blktests suite then the following appears
in the kernel log:

==================================================================

BUG: KASAN: null-ptr-deref in __list_del_entry_valid+0x4d/0xe0
Read of size 8 at addr 0000000000000000 by task kworker/u16:3/161

CPU: 5 PID: 161 Comm: kworker/u16:3 Not tainted 5.15.0-rc1-dbg+ #2
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
rel-1.14.0-0-g155821a-rebuilt.opensuse.org 04/01/2014
Workqueue: iw_cm_wq cm_work_handler [iw_cm]
Call Trace:
 show_stack+0x52/0x58
 dump_stack_lvl+0x5b/0x82
 kasan_report.cold+0x52/0x57
 __asan_load8+0x69/0x90
 __list_del_entry_valid+0x4d/0xe0
 _cma_cancel_listens+0x49/0x230 [rdma_cm]
 _destroy_id+0x4e/0x420 [rdma_cm]
 destroy_id_handler_unlock+0xc4/0x200 [rdma_cm]
 iw_conn_req_handler+0x335/0x370 [rdma_cm]
 cm_conn_req_handler+0x546/0x7d0 [iw_cm]
 cm_work_handler+0x419/0x480 [iw_cm]
 process_one_work+0x59d/0xb00
 worker_thread+0x8f/0x5c0
 kthread+0x1fc/0x230
 ret_from_fork+0x1f/0x30
==================================================================
Quoted 3 lines

I think this is a regression. This happened with commit a17a1faf5d3e ("RDMA/cma: Fix listener leak in rdma_cma_listen_on_all() failure").

So far I could not reproduce it with that patch (tried to get better understanding with siw debugging enabled). That doesn't mean much.

But I do see issues with rxe if running the complete srp set from blktest, which seem to be unrelated. I'll send an extra message.

Many thanks,
Bernard.