Series comparison

With the in-kernel path-manager, it is possible to change the 'fullmesh'
flag. The code in mptcp_pm_nl_fullmesh() expects to change it only on
'subflow' endpoints, to recreate more or less subflows using the linked
address.

Unfortunately, the set_flags() hook was a bit more permissive, and
allowed 'implicit' endpoints to get the 'fullmesh' flag while it is not
allowed before.

That's what syzbot found, triggering the following warning:

WARNING: CPU: 0 PID: 6499 at net/mptcp/pm_netlink.c:1496 __mark_subflow_endp_available net/mptcp/pm_netlink.c:1496 [inline]
  WARNING: CPU: 0 PID: 6499 at net/mptcp/pm_netlink.c:1496 mptcp_pm_nl_fullmesh net/mptcp/pm_netlink.c:1980 [inline]
  WARNING: CPU: 0 PID: 6499 at net/mptcp/pm_netlink.c:1496 mptcp_nl_set_flags net/mptcp/pm_netlink.c:2003 [inline]
  WARNING: CPU: 0 PID: 6499 at net/mptcp/pm_netlink.c:1496 mptcp_pm_nl_set_flags+0x974/0xdc0 net/mptcp/pm_netlink.c:2064
  Modules linked in:
  CPU: 0 UID: 0 PID: 6499 Comm: syz.1.413 Not tainted 6.13.0-rc5-syzkaller-00172-gd1bf27c4e176 #0
  Hardware name: Google Compute Engine/Google Compute Engine, BIOS Google 09/13/2024
  RIP: 0010:__mark_subflow_endp_available net/mptcp/pm_netlink.c:1496 [inline]
  RIP: 0010:mptcp_pm_nl_fullmesh net/mptcp/pm_netlink.c:1980 [inline]
  RIP: 0010:mptcp_nl_set_flags net/mptcp/pm_netlink.c:2003 [inline]
  RIP: 0010:mptcp_pm_nl_set_flags+0x974/0xdc0 net/mptcp/pm_netlink.c:2064
  Code: 01 00 00 49 89 c5 e8 fb 45 e8 f5 e9 b8 fc ff ff e8 f1 45 e8 f5 4c 89 f7 be 03 00 00 00 e8 44 1d 0b f9 eb a0 e8 dd 45 e8 f5 90 <0f> 0b 90 e9 17 ff ff ff 89 d9 80 e1 07 38 c1 0f 8c c9 fc ff ff 48
  RSP: 0018:ffffc9000d307240 EFLAGS: 00010293
  RAX: ffffffff8bb72e03 RBX: 0000000000000000 RCX: ffff88807da88000
  RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
  RBP: ffffc9000d307430 R08: ffffffff8bb72cf0 R09: 1ffff1100b842a5e
  R10: dffffc0000000000 R11: ffffed100b842a5f R12: ffff88801e2e5ac0
  R13: ffff88805c214800 R14: ffff88805c2152e8 R15: 1ffff1100b842a5d
  FS:  00005555619f6500(0000) GS:ffff8880b8600000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 0000000020002840 CR3: 00000000247e6000 CR4: 00000000003526f0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
   <TASK>
   genl_family_rcv_msg_doit net/netlink/genetlink.c:1115 [inline]
   genl_family_rcv_msg net/netlink/genetlink.c:1195 [inline]
   genl_rcv_msg+0xb14/0xec0 net/netlink/genetlink.c:1210
   netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2542
   genl_rcv+0x28/0x40 net/netlink/genetlink.c:1219
   netlink_unicast_kernel net/netlink/af_netlink.c:1321 [inline]
   netlink_unicast+0x7f6/0x990 net/netlink/af_netlink.c:1347
   netlink_sendmsg+0x8e4/0xcb0 net/netlink/af_netlink.c:1891
   sock_sendmsg_nosec net/socket.c:711 [inline]
   __sock_sendmsg+0x221/0x270 net/socket.c:726
   ____sys_sendmsg+0x52a/0x7e0 net/socket.c:2583
   ___sys_sendmsg net/socket.c:2637 [inline]
   __sys_sendmsg+0x269/0x350 net/socket.c:2669
   do_syscall_x64 arch/x86/entry/common.c:52 [inline]
   do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83
   entry_SYSCALL_64_after_hwframe+0x77/0x7f
  RIP: 0033:0x7f5fe8785d29
  Code: ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 a8 ff ff ff f7 d8 64 89 01 48
  RSP: 002b:00007fff571f5558 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
  RAX: ffffffffffffffda RBX: 00007f5fe8975fa0 RCX: 00007f5fe8785d29
  RDX: 0000000000000000 RSI: 0000000020000480 RDI: 0000000000000007
  RBP: 00007f5fe8801b08 R08: 0000000000000000 R09: 0000000000000000
  R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
  R13: 00007f5fe8975fa0 R14: 00007f5fe8975fa0 R15: 00000000000011f4
   </TASK>

Here, syzbot managed to set the 'fullmesh' flag on an 'implicit' and
used -- according to 'id_avail_bitmap' -- endpoint, causing the PM to
try decrement the local_addr_used counter which is only incremented for
the 'subflow' endpoint.

Note that 'no type' endpoints -- not 'subflow', 'signal', 'implicit' --
are fine, because their ID will not be marked as used in the 'id_avail'
bitmap, and setting 'fullmesh' can help forcing the creation of subflow
when receiving an ADD_ADDR.

Fixes: 73c762c1f07d ("mptcp: set fullmesh flag in pm_netlink")
Reported-by: syzbot+cd16e79c1e45f3fe0377@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/6786ac51.050a0220.216c54.00a6.GAE@google.com
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/540
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
---
 net/mptcp/pm_netlink.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/mptcp/pm_netlink.c b/net/mptcp/pm_netlink.c
index XXXXXXX..XXXXXXX 100644
--- a/net/mptcp/pm_netlink.c
+++ b/net/mptcp/pm_netlink.c
@@ -XXX,XX +XXX,XX @@ int mptcp_pm_nl_set_flags(struct mptcp_pm_addr_entry *local,
 		return -EINVAL;
 	}
 	if ((local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
-	    (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
+	    ((entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) ||
+	     (entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT))) {
 		spin_unlock_bh(&pernet->lock);
 		NL_SET_ERR_MSG_ATTR(info->extack, attr, "invalid addr flags");
 		return -EINVAL;

---
base-commit: 9f10cc97addfb3a81b4aaf4b13f59b36b5bcc7a6
change-id: 20250116-mptcp-issue-540-49d1db3346a9

Best regards,
-- 
Matthieu Baerts (NGI0) <matttbe@kernel.org>

Here are a few patches linked to the 'set-flags' interface with the
in-kernel path-manager.

Patch 1: a fix for an issue reported by syzbot. For -net.

Patch 2: a small cleanup. For -next.

Patch 3: a small optimisation not to iterate over all subflows to do
         nothing. For -next.

Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
---
Changes in v2:
- patch 1: check the two flags in one instruction. (Paolo)
- patch 2-3: new patches, for next.
- Link to v1: https://lore.kernel.org/r/20250116-mptcp-issue-540-v1-1-b0ba2c126eb8@kernel.org

---
Matthieu Baerts (NGI0) (3):
      mptcp: pm: only set fullmesh for subflow endp
      mptcp: pm: remove unused ret value to set flags
      mptcp: pm: change to fullmesh only for 'subflow'

net/mptcp/pm_netlink.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
---
base-commit: c466642ac4d22d6f294d669d8a1a3535610cb45c
change-id: 20250116-mptcp-issue-540-49d1db3346a9

Best regards,
-- 
Matthieu Baerts (NGI0) <matttbe@kernel.org>

Unfortunately, the set_flags() hook was a bit more permissive, and
allowed 'implicit' endpoints to get the 'fullmesh' flag while it is not
allowed before.

That's what syzbot found, triggering the following warning:

Fixes: 73c762c1f07d ("mptcp: set fullmesh flag in pm_netlink")
Reported-by: syzbot+cd16e79c1e45f3fe0377@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/6786ac51.050a0220.216c54.00a6.GAE@google.com
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/540
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
---
Notes:
- v2: check the two flags in one instruction. (Paolo)
---
 net/mptcp/pm_netlink.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/mptcp/pm_netlink.c b/net/mptcp/pm_netlink.c
index XXXXXXX..XXXXXXX 100644
--- a/net/mptcp/pm_netlink.c
+++ b/net/mptcp/pm_netlink.c
@@ -XXX,XX +XXX,XX @@ int mptcp_pm_nl_set_flags(struct mptcp_pm_addr_entry *local,
 		return -EINVAL;
 	}
 	if ((local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
-	    (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
+	    (entry->flags & (MPTCP_PM_ADDR_FLAG_SIGNAL |
+			     MPTCP_PM_ADDR_FLAG_IMPLICIT))) {
 		spin_unlock_bh(&pernet->lock);
 		NL_SET_ERR_MSG_ATTR(info->extack, attr, "invalid addr flags");
 		return -EINVAL;

-- 
2.47.1

The returned value is not used, it can then be dropped.

Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
---
 net/mptcp/pm_netlink.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

-- 
2.47.1

If an entrypoint has no type -- so not 'subflow', 'signal', 'implicit' --
there are then no subflows to re-create from this local endpoint.

In this case, there is then no need to iterate over all connections to
do nothing. So stop early when this case is present.

Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
---
 net/mptcp/pm_netlink.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/net/mptcp/pm_netlink.c b/net/mptcp/pm_netlink.c
index XXXXXXX..XXXXXXX 100644
--- a/net/mptcp/pm_netlink.c
+++ b/net/mptcp/pm_netlink.c
@@ -XXX,XX +XXX,XX @@ static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
 }
 
 static void mptcp_nl_set_flags(struct net *net, struct mptcp_addr_info *addr,
-			       u8 bkup, u8 changed)
+			       u8 flags, u8 changed)
 {
+	u8 is_subflow = !!(flags & MPTCP_PM_ADDR_FLAG_SUBFLOW);
+	u8 bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
 	long s_slot = 0, s_num = 0;
 	struct mptcp_sock *msk;
 
+	if (changed == MPTCP_PM_ADDR_FLAG_FULLMESH && !is_subflow)
+		return;
+
 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
 		struct sock *sk = (struct sock *)msk;
 
@@ -XXX,XX +XXX,XX @@ static void mptcp_nl_set_flags(struct net *net, struct mptcp_addr_info *addr,
 		lock_sock(sk);
 		if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
 			mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
-		if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
+		if (is_subflow && (changed & MPTCP_PM_ADDR_FLAG_FULLMESH))
 			mptcp_pm_nl_fullmesh(msk, addr);
 		release_sock(sk);
 
@@ -XXX,XX +XXX,XX @@ int mptcp_pm_nl_set_flags(struct mptcp_pm_addr_entry *local,
 	struct mptcp_pm_addr_entry *entry;
 	struct pm_nl_pernet *pernet;
 	u8 lookup_by_id = 0;
-	u8 bkup = 0;
 
 	pernet = pm_nl_get_pernet(net);
 
@@ -XXX,XX +XXX,XX @@ int mptcp_pm_nl_set_flags(struct mptcp_pm_addr_entry *local,
 		}
 	}
 
-	if (local->flags & MPTCP_PM_ADDR_FLAG_BACKUP)
-		bkup = 1;
-
 	spin_lock_bh(&pernet->lock);
 	entry = lookup_by_id ? __lookup_addr_by_id(pernet, local->addr.id) :
 			       __lookup_addr(pernet, &local->addr);
@@ -XXX,XX +XXX,XX @@ int mptcp_pm_nl_set_flags(struct mptcp_pm_addr_entry *local,
 	*local = *entry;
 	spin_unlock_bh(&pernet->lock);
 
-	mptcp_nl_set_flags(net, &local->addr, bkup, changed);
+	mptcp_nl_set_flags(net, &local->addr, entry->flags, changed);
 	return 0;
 }

-- 
2.47.1