hci: fix double free in hci_req_sync

The approach taken to address the 'CVE-2024-35978' introduced another double-free vulnerability. commit 45d355a926ab ("Bluetooth: Fix memory leak in hci_req_sync_complete()")

'hdev->req_skb' double free scenario:

cpu1 cpu2
==== ====
sock_ioctl
sock_do_ioctl
hci_sock_ioctl
hci_dev_cmd
hci_req_sync

 __hci_req_sync              hci_rx_work
 kfree_skb(hdev->req_skb)    hci_event_packet
 (sleep)                     hci_req_sync_complete
  \__ Longer times,          kfree_skb(hdev->req_skb)
      reproduce well         hdev->req_skb = NULL

The longer cpu1 sleep in '__hci_req_sync', the more reproducible it is. We've tested it by inserting the 'msleep()' function, and it's frequently at 1000ms, and It has been consistently reproducible at 2000ms.

We confirmed the detection with various workloads that cause CPU1 to sleep. The call trace below is one of the KASAN has seen.

Bluetooth: hci0: unexpected cc 0x0c38 length: 249 > 2
==================================================================

BUG: KASAN: slab-use-after-free in skb_release_data+0x7d8/0x8a0 home/paran/linux/net/core/skbuff.c:1119
Write of size 1 at addr ffff00000947d3fe by task syz-executor.2/2010

CPU: 1 PID: 2010 Comm: syz-executor.2 Not tainted
6.10.0-rc4-00217-g35bb670d65fc-dirty #22 Hardware name: linux,dummy-virt (DT)
Call trace:
 dump_backtrace+0x318/0x348 home/paran/linux/arch/arm64/kernel/stacktrace.c:317
 show_stack+0x4c/0x80 home/paran/linux/arch/arm64/kernel/stacktrace.c:324
 __dump_stack home/paran/linux/lib/dump_stack.c:88 [inline]
 dump_stack_lvl+0x214/0x328 home/paran/linux/lib/dump_stack.c:114
 print_address_description home/paran/linux/mm/kasan/report.c:377 [inline]
 print_report+0x2ac/0x948 home/paran/linux/mm/kasan/report.c:488
 kasan_report+0xc8/0x148 home/paran/linux/mm/kasan/report.c:601
 __asan_report_store1_noabort+0x44/0x60 home/paran/linux/mm/kasan/report_generic.c:383
 skb_release_data+0x7d8/0x8a0 home/paran/linux/net/core/skbuff.c:1119
 skb_release_all+0x80/0xe0 home/paran/linux/net/core/skbuff.c:1173
 __kfree_skb home/paran/linux/net/core/skbuff.c:1187 [inline]
 kfree_skb_reason+0x138/0x3a8 home/paran/linux/net/core/skbuff.c:1223
 __hci_req_sync+0x404/0x948 [bluetooth]
 hci_req_sync+0xc0/0x138 [bluetooth]
 hci_dev_cmd+0x33c/0xc18 [bluetooth]
 hci_sock_ioctl+0x800/0xb68 [bluetooth]
 sock_do_ioctl+0xfc/0x2e0 home/paran/linux/net/socket.c:1222
 sock_ioctl+0x62c/0xab0 home/paran/linux/net/socket.c:1341
 vfs_ioctl+0x90/0x140 home/paran/linux/fs/ioctl.c:51
 __do_sys_ioctl home/paran/linux/fs/ioctl.c:907 [inline]
 __se_sys_ioctl home/paran/linux/fs/ioctl.c:893 [inline]
 __arm64_sys_ioctl+0x218/0x268 home/paran/linux/fs/ioctl.c:893
 __invoke_syscall home/paran/linux/arch/arm64/kernel/syscall.c:34 [inline]
 invoke_syscall+0xdc/0x460 home/paran/linux/arch/arm64/kernel/syscall.c:48
 el0_svc_common.constprop.0+0x2d4/0x3e8 home/paran/linux/arch/arm64/kernel/syscall.c:133
 do_el0_svc+0x60/0x98 home/paran/linux/arch/arm64/kernel/syscall.c:152
 el0_svc+0xc4/0x240 home/paran/linux/arch/arm64/kernel/entry-common.c:712
 el0t_64_sync_handler+0x120/0x130 home/paran/linux/arch/arm64/kernel/entry-common.c:730
 el0t_64_sync+0x190/0x198 home/paran/linux/arch/arm64/kernel/entry.S:598

Allocated by task 577:
 kasan_save_stack+0x48/0x90 home/paran/linux/mm/kasan/common.c:47
 kasan_save_track+0x38/0x60 home/paran/linux/mm/kasan/common.c:68
 kasan_save_alloc_info+0x64/0xc0 home/paran/linux/mm/kasan/generic.c:565
 unpoison_slab_object home/paran/linux/mm/kasan/common.c:312 [inline]
 __kasan_slab_alloc+0x100/0x110 home/paran/linux/mm/kasan/common.c:338
 kasan_slab_alloc home/paran/linux/./include/linux/kasan.h:201 [inline]
 slab_post_alloc_hook home/paran/linux/mm/slub.c:3941 [inline]
 slab_alloc_node home/paran/linux/mm/slub.c:4001 [inline]
 kmem_cache_alloc_noprof+0x2e8/0x630 home/paran/linux/mm/slub.c:4008
 skb_clone+0x1a4/0x4d0 home/paran/linux/net/core/skbuff.c:2052
 hci_cmd_work+0x78c/0x868 [bluetooth]
 process_one_work home/paran/linux/kernel/workqueue.c:3231 [inline]
 process_scheduled_works+0x9fc/0x1d98 home/paran/linux/kernel/workqueue.c:3312
 worker_thread+0x57c/0xf98 home/paran/linux/kernel/workqueue.c:3393
 kthread+0x3c8/0x478 home/paran/linux/kernel/kthread.c:389
 ret_from_fork+0x10/0x20 home/paran/linux/arch/arm64/kernel/entry.S:860

Freed by task 577:
 kasan_save_stack+0x48/0x90 home/paran/linux/mm/kasan/common.c:47
 kasan_save_track+0x38/0x60 home/paran/linux/mm/kasan/common.c:68
 kasan_save_free_info+0x64/0xf0 home/paran/linux/mm/kasan/generic.c:579
 poison_slab_object+0x168/0x270 home/paran/linux/mm/kasan/common.c:240
 __kasan_slab_free+0x34/0xa0 home/paran/linux/mm/kasan/common.c:256
 kasan_slab_free home/paran/linux/./include/linux/kasan.h:184 [inline]
 slab_free_hook home/paran/linux/mm/slub.c:2196 [inline]
 slab_free home/paran/linux/mm/slub.c:4437 [inline]
 kmem_cache_free+0x20c/0x670 home/paran/linux/mm/slub.c:4512
 kfree_skbmem+0x2b0/0x390 home/paran/linux/net/core/skbuff.c:1131
 __kfree_skb home/paran/linux/net/core/skbuff.c:1188 [inline]
 kfree_skb_reason+0x14c/0x3a8 home/paran/linux/net/core/skbuff.c:1223
 hci_req_sync_complete+0x114/0x308 [bluetooth]
 hci_event_packet+0xa10/0x12a0 [bluetooth]
 hci_rx_work+0x4d8/0xa80 [bluetooth]
 process_one_work home/paran/linux/kernel/workqueue.c:3231 [inline]
 process_scheduled_works+0x9fc/0x1d98 home/paran/linux/kernel/workqueue.c:3312
 worker_thread+0x57c/0xf98 home/paran/linux/kernel/workqueue.c:3393
 kthread+0x3c8/0x478 home/paran/linux/kernel/kthread.c:389
 ret_from_fork+0x10/0x20 home/paran/linux/arch/arm64/kernel/entry.S:860

The buggy address belongs to the object at ffff00000947d380
 which belongs to the cache skbuff_head_cache of size 232

The buggy address is located 126 bytes inside of
freed 232-byte region [ffff00000947d380, ffff00000947d468)

The buggy address belongs to the physical page:
page: refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x4947c

head: order:1 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0 memcg:ffff0000222a73b1
flags: 0x3fffe0000000040(head|node=0|zone=0|lastcpupid=0x1ffff)
page_type: 0xffffefff(slab)

raw: 03fffe0000000040 ffff0000133a41c0 fffffdffc0e3a090 fffffdffc1069290
raw: 0000000000000000 0000000000120012 00000001ffffefff ffff0000222a73b1

head: 03fffe0000000040 ffff0000133a41c0 fffffdffc0e3a090 fffffdffc1069290
head: 0000000000000000 0000000000120012 00000001ffffefff ffff0000222a73b1
head: 03fffe0000000001 fffffdffc0251f01 ffffffffffffffff 0000000000000000
head: 0000000000000002 0000000000000000 00000000ffffffff 0000000000000000

page dumped because: kasan: bad access detected

Memory state around the buggy address:
 ffff00000947d280: 00 00 00 00 00 fc fc fc fc fc fc fc fc fc fc fc
 ffff00000947d300: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>ffff00000947d380: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
                                                                ^
 ffff00000947d400: fb fb fb fb fb fb fb fb fb fb fb fb fb fc fc fc
 ffff00000947d480: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
==================================================================

We considered using the "hci_req_sync_lock" mutex, but concluded it wasn't a good solution due to the increased sleep intervals causing this issue. Instead, we introduced a spinlock member on 'struct hci_dev'.

Since applying our patch, we have repeatedly run the same tests in the syzkaller without encountering any issues.

---
 include/net/bluetooth/hci_core.h |  1 +
 net/bluetooth/hci_core.c         |  1 +
 net/bluetooth/hci_request.c      |  6 ++----
 net/bluetooth/hci_request.h      |  9 +++++++++
 net/bluetooth/hci_sync.c         | 13 +++----------
 5 files changed, 16 insertions(+), 14 deletions(-)

index c43716edf205..8b95061f063b 100644
@@ -519,6 +519,7 @@ struct hci_dev {
 	struct sk_buff		*recv_event;
 
 	struct mutex		req_lock;
+	spinlock_t		req_skb_lock;
 	wait_queue_head_t	req_wait_q;
 	__u32			req_status;
 	__u32			req_result;

index dd3b0f501018..138a6b19894d 100644
@@ -2572,6 +2572,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
 
 	mutex_init(&hdev->lock);
 	mutex_init(&hdev->req_lock);
+	spin_lock_init(&hdev->req_skb_lock);
 
 	ida_init(&hdev->unset_handle_ida);
 

index efea25eb56ce..6a109c1ad359 100644
@@ -106,8 +106,7 @@ void hci_req_sync_complete(struct hci_dev *hdev, u8 result, u16 opcode,
 		hdev->req_result = result;
 		hdev->req_status = HCI_REQ_DONE;
 		if (skb) {
-			kfree_skb(hdev->req_skb);
-			hdev->req_skb = skb_get(skb);
+			hci_req_skb_release_and_set(hdev, skb_get(skb));
 		}
 		wake_up_interruptible(&hdev->req_wait_q);
 	}

@@ -181,8 +180,7 @@ int __hci_req_sync(struct hci_dev *hdev, int (*func)(struct hci_request *req,
 		break;
 	}
 
-	kfree_skb(hdev->req_skb);
-	hdev->req_skb = NULL;
+	hci_req_skb_release_and_set(hdev, NULL);
 	hdev->req_status = hdev->req_result = 0;
 
 	bt_dev_dbg(hdev, "end: err %d", err);

index c91f2838f542..6526c78443bc 100644
@@ -28,6 +28,15 @@
 
 #define hci_req_sync_lock(hdev)   mutex_lock(&hdev->req_lock)
 #define hci_req_sync_unlock(hdev) mutex_unlock(&hdev->req_lock)
+#define hci_req_skb_release_and_set(hdev, val)		\
+({							\
+	if (hdev->req_skb) {				\
+		spin_lock(&hdev->req_skb_lock);		\
+		kfree_skb(hdev->req_skb);		\
+		hdev->req_skb = val;			\
+		spin_unlock(&hdev->req_skb_lock);	\
+	}						\
+})
 
 struct hci_request {
 	struct hci_dev		*hdev;

index a8a7d2b36870..25c8d858c82e 100644
@@ -33,8 +33,7 @@ static void hci_cmd_sync_complete(struct hci_dev *hdev, u8 result, u16 opcode,
 	hdev->req_status = HCI_REQ_DONE;
 
 	/* Free the request command so it is not used as response */
-	kfree_skb(hdev->req_skb);
-	hdev->req_skb = NULL;
+	hci_req_skb_release_and_set(hdev, NULL);
 
 	if (skb) {
 		struct sock *sk = hci_skb_sk(skb);

@@ -4935,10 +4934,7 @@ int hci_dev_open_sync(struct hci_dev *hdev)
 			hdev->sent_cmd = NULL;
 		}
 
-		if (hdev->req_skb) {
-			kfree_skb(hdev->req_skb);
-			hdev->req_skb = NULL;
-		}
+		hci_req_skb_release_and_set(hdev, NULL);
 
 		clear_bit(HCI_RUNNING, &hdev->flags);
 		hci_sock_dev_event(hdev, HCI_DEV_CLOSE);

@@ -5100,10 +5096,7 @@ int hci_dev_close_sync(struct hci_dev *hdev)
 	}
 
 	/* Drop last request */
-	if (hdev->req_skb) {
-		kfree_skb(hdev->req_skb);
-		hdev->req_skb = NULL;
-	}
+	hci_req_skb_release_and_set(hdev, NULL);
 
 	clear_bit(HCI_RUNNING, &hdev->flags);
 	hci_sock_dev_event(hdev, HCI_DEV_CLOSE);

2.45.2

Hi,

hci_req_sync is no longer called from hci_dev_cmd:

https://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next.git/commit/?id=6851d11d389ceb00c1220b267b9a04f54dbc4573

And I'm in process of removing hci_request.c completely.

Luiz Augusto von Dentz

Hi Luiz,

Thank you so much your hard work.

I saw the patch email at the link below.
Link:
https://lore.kernel.org/all/[email protected]/T/

Does the patch you mentioned apply to backports as well?

Warm regards,

Yunseong Kim

This is automated email and please do not reply to this email!

Dear submitter,

Thank you for submitting the patches to the linux bluetooth mailing list. This is a CI test results with your patch series:
PW Link:https://patchwork.kernel.org/project/bluetooth/list/?series=867274

---Test result---

Test Summary:

CheckPatch                    FAIL      1.82 seconds
GitLint                       FAIL      0.49 seconds
SubjectPrefix                 FAIL      0.31 seconds
BuildKernel                   PASS      30.54 seconds
CheckAllWarning               PASS      45.30 seconds
CheckSparse                   PASS      38.87 seconds
CheckSmatch                   PASS      107.07 seconds
BuildKernel32                 PASS      29.74 seconds
TestRunnerSetup               PASS      530.55 seconds
TestRunner_l2cap-tester       PASS      24.05 seconds
TestRunner_iso-tester         PASS      37.17 seconds
TestRunner_bnep-tester        PASS      4.93 seconds
TestRunner_mgmt-tester        FAIL      110.24 seconds
TestRunner_rfcomm-tester      PASS      7.56 seconds
TestRunner_sco-tester         PASS      15.26 seconds
TestRunner_ioctl-tester       PASS      8.16 seconds
TestRunner_mesh-tester        PASS      6.21 seconds
TestRunner_smp-tester         PASS      7.00 seconds
TestRunner_userchan-tester    PASS      10.04 seconds
IncrementalBuild              PASS      28.81 seconds

Details
##############################
Test: CheckPatch - FAIL
Desc: Run checkpatch.pl script
Output:
hci: fix double free in hci_req_sync
WARNING: Prefer a maximum 75 chars per line (possible unwrapped commit description?) #129:
6.10.0-rc4-00217-g35bb670d65fc-dirty #22 Hardware name: linux,dummy-virt (DT)

CHECK: Macro argument reuse 'hdev' - possible side-effects? #304: FILE: net/bluetooth/hci_request.h:31:
+#define hci_req_skb_release_and_set(hdev, val) \
+({ \
+ if (hdev->req_skb) { \
+ spin_lock(&hdev->req_skb_lock); \
+ kfree_skb(hdev->req_skb); \
+ hdev->req_skb = val; \
+ spin_unlock(&hdev->req_skb_lock); \
+ } \
+})

CHECK: Macro argument 'hdev' may be better as '(hdev)' to avoid precedence issues #304: FILE: net/bluetooth/hci_request.h:31:
+#define hci_req_skb_release_and_set(hdev, val) \
+({ \
+ if (hdev->req_skb) { \
+ spin_lock(&hdev->req_skb_lock); \
+ kfree_skb(hdev->req_skb); \
+ hdev->req_skb = val; \
+ spin_unlock(&hdev->req_skb_lock); \
+ } \
+})

total: 0 errors, 1 warnings, 2 checks, 78 lines checked

NOTE: For some of the reported defects, checkpatch may be able to
mechanically convert to the typical style using --fix or --fix-inplace.

/github/workspace/src/src/13718597.patch has style problems, please review.

NOTE: Ignored message types: UNKNOWN_COMMIT_ID

NOTE: If any of the errors are false positives, please report
them to the maintainer, see CHECKPATCH in MAINTAINERS.

##############################
Test: GitLint - FAIL
Desc: Run gitlint
Output:
hci: fix double free in hci_req_sync

WARNING: I3 - ignore-body-lines: gitlint will be switching from using Python regex 'match' (match beginning) to 'search' (match anywhere) semantics. Please review your ignore-body-lines.regex option accordingly. To remove this warning, set general.regex-style-search=True. More details: https://jorisroovers.github.io/gitlint/configuration/#regex-style-search
31: B1 Line exceeds max length (103>80): "BUG: KASAN: slab-use-after-free in skb_release_data+0x7d8/0x8a0 home/paran/linux/net/core/skbuff.c:1119"
44: B1 Line exceeds max length (86>80): " __asan_report_store1_noabort+0x44/0x60 home/paran/linux/mm/kasan/report_generic.c:383"
61: B1 Line exceeds max length (88>80): " el0_svc_common.constprop.0+0x2d4/0x3e8 home/paran/linux/arch/arm64/kernel/syscall.c:133"
64: B1 Line exceeds max length (87>80): " el0t_64_sync_handler+0x120/0x130 home/paran/linux/arch/arm64/kernel/entry-common.c:730"
139: B2 Line has trailing whitespace: "Since applying our patch, we have repeatedly run the same tests in "
142: B1 Line exceeds max length (122>80): "Link: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=89a32741f4217856066c198a4a7267bcdd1edd67"
##############################
Test: SubjectPrefix - FAIL
Desc: Check subject contains "Bluetooth" prefix
Output:
"Bluetooth: " prefix is not specified in the subject
##############################
Test: TestRunner_mgmt-tester - FAIL
Desc: Run mgmt-tester with test-runner
Output:
Total: 492, Passed: 489 (99.4%), Failed: 1, Not Run: 2

Failed Test Cases
LL Privacy - Add Device 4 (2 Devices to AL) Failed 0.163 seconds

---
Regards,
Linux Bluetooth

…