Commit 53c0441d authored Feb 07, 2024 by Jiri Pirko Committed by Jakub Kicinski Feb 08, 2024

dpll: fix possible deadlock during netlink dump operation



Recently, I've been hitting following deadlock warning during dpll pin
dump:

[52804.637962] ======================================================
[52804.638536] WARNING: possible circular locking dependency detected
[52804.639111] 6.8.0-rc2jiri+ #1 Not tainted
[52804.639529] ------------------------------------------------------
[52804.640104] python3/2984 is trying to acquire lock:
[52804.640581] ffff88810e642678 (nlk_cb_mutex-GENERIC){+.+.}-{3:3}, at: netlink_dump+0xb3/0x780
[52804.641417]
               but task is already holding lock:
[52804.642010] ffffffff83bde4c8 (dpll_lock){+.+.}-{3:3}, at: dpll_lock_dumpit+0x13/0x20
[52804.642747]
               which lock already depends on the new lock.

[52804.643551]
               the existing dependency chain (in reverse order) is:
[52804.644259]
               -> #1 (dpll_lock){+.+.}-{3:3}:
[52804.644836]        lock_acquire+0x174/0x3e0
[52804.645271]        __mutex_lock+0x119/0x1150
[52804.645723]        dpll_lock_dumpit+0x13/0x20
[52804.646169]        genl_start+0x266/0x320
[52804.646578]        __netlink_dump_start+0x321/0x450
[52804.647056]        genl_family_rcv_msg_dumpit+0x155/0x1e0
[52804.647575]        genl_rcv_msg+0x1ed/0x3b0
[52804.648001]        netlink_rcv_skb+0xdc/0x210
[52804.648440]        genl_rcv+0x24/0x40
[52804.648831]        netlink_unicast+0x2f1/0x490
[52804.649290]        netlink_sendmsg+0x36d/0x660
[52804.649742]        __sock_sendmsg+0x73/0xc0
[52804.650165]        __sys_sendto+0x184/0x210
[52804.650597]        __x64_sys_sendto+0x72/0x80
[52804.651045]        do_syscall_64+0x6f/0x140
[52804.651474]        entry_SYSCALL_64_after_hwframe+0x46/0x4e
[52804.652001]
               -> #0 (nlk_cb_mutex-GENERIC){+.+.}-{3:3}:
[52804.652650]        check_prev_add+0x1ae/0x1280
[52804.653107]        __lock_acquire+0x1ed3/0x29a0
[52804.653559]        lock_acquire+0x174/0x3e0
[52804.653984]        __mutex_lock+0x119/0x1150
[52804.654423]        netlink_dump+0xb3/0x780
[52804.654845]        __netlink_dump_start+0x389/0x450
[52804.655321]        genl_family_rcv_msg_dumpit+0x155/0x1e0
[52804.655842]        genl_rcv_msg+0x1ed/0x3b0
[52804.656272]        netlink_rcv_skb+0xdc/0x210
[52804.656721]        genl_rcv+0x24/0x40
[52804.657119]        netlink_unicast+0x2f1/0x490
[52804.657570]        netlink_sendmsg+0x36d/0x660
[52804.658022]        __sock_sendmsg+0x73/0xc0
[52804.658450]        __sys_sendto+0x184/0x210
[52804.658877]        __x64_sys_sendto+0x72/0x80
[52804.659322]        do_syscall_64+0x6f/0x140
[52804.659752]        entry_SYSCALL_64_after_hwframe+0x46/0x4e
[52804.660281]
               other info that might help us debug this:

[52804.661077]  Possible unsafe locking scenario:

[52804.661671]        CPU0                    CPU1
[52804.662129]        ----                    ----
[52804.662577]   lock(dpll_lock);
[52804.662924]                                lock(nlk_cb_mutex-GENERIC);
[52804.663538]                                lock(dpll_lock);
[52804.664073]   lock(nlk_cb_mutex-GENERIC);
[52804.664490]

The issue as follows: __netlink_dump_start() calls control->start(cb)
with nlk->cb_mutex held. In control->start(cb) the dpll_lock is taken.
Then nlk->cb_mutex is released and taken again in netlink_dump(), while
dpll_lock still being held. That leads to ABBA deadlock when another
CPU races with the same operation.

Fix this by moving dpll_lock taking into dumpit() callback which ensures
correct lock taking order.

Fixes: 9d71b54b ("dpll: netlink: Add DPLL framework base functions")
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Arkadiusz Kubalewski <arkadiusz.kubalewski@intel.com>
Link: https://lore.kernel.org/r/20240207115902.371649-1-jiri@resnulli.us


Signed-off-by: Jakub Kicinski <kuba@kernel.org>

parent 1f719a2f

Documentation/netlink/specs/dpll.yaml

+0 −4

Original line number	Diff line number	Diff line
		@@ -384,8 +384,6 @@ operations:
		- type

		dump:
		pre: dpll-lock-dumpit
		post: dpll-unlock-dumpit
		reply: *dev-attrs

		-
		@@ -473,8 +471,6 @@ operations:
		- fractional-frequency-offset

		dump:
		pre: dpll-lock-dumpit
		post: dpll-unlock-dumpit
		request:
		attributes:
		- id

drivers/dpll/dpll_netlink.c

+6 −14

Original line number	Diff line number	Diff line
		@@ -1199,6 +1199,7 @@ int dpll_nl_pin_get_dumpit(struct sk_buff skb, struct netlink_callback cb)
		unsigned long i;
		int ret = 0;

		mutex_lock(&dpll_lock);
		xa_for_each_marked_start(&dpll_pin_xa, i, pin, DPLL_REGISTERED,
		ctx->idx) {
		if (!dpll_pin_available(pin))
		@@ -1218,6 +1219,8 @@ int dpll_nl_pin_get_dumpit(struct sk_buff skb, struct netlink_callback cb)
		}
		genlmsg_end(skb, hdr);
		}
		mutex_unlock(&dpll_lock);

		if (ret == -EMSGSIZE) {
		ctx->idx = i;
		return skb->len;
		@@ -1373,6 +1376,7 @@ int dpll_nl_device_get_dumpit(struct sk_buff skb, struct netlink_callback cb)
		unsigned long i;
		int ret = 0;

		mutex_lock(&dpll_lock);
		xa_for_each_marked_start(&dpll_device_xa, i, dpll, DPLL_REGISTERED,
		ctx->idx) {
		hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
		@@ -1389,6 +1393,8 @@ int dpll_nl_device_get_dumpit(struct sk_buff skb, struct netlink_callback cb)
		}
		genlmsg_end(skb, hdr);
		}
		mutex_unlock(&dpll_lock);

		if (ret == -EMSGSIZE) {
		ctx->idx = i;
		return skb->len;
		@@ -1439,20 +1445,6 @@ dpll_unlock_doit(const struct genl_split_ops ops, struct sk_buff skb,
		mutex_unlock(&dpll_lock);
		}

		int dpll_lock_dumpit(struct netlink_callback *cb)
		{
		mutex_lock(&dpll_lock);

		return 0;
		}

		int dpll_unlock_dumpit(struct netlink_callback *cb)
		{
		mutex_unlock(&dpll_lock);

		return 0;
		}

		int dpll_pin_pre_doit(const struct genl_split_ops ops, struct sk_buff skb,
		struct genl_info *info)
		{

drivers/dpll/dpll_nl.c

+0 −4

Original line number	Diff line number	Diff line
		@@ -95,9 +95,7 @@ static const struct genl_split_ops dpll_nl_ops[] = {
		},
		{
		.cmd = DPLL_CMD_DEVICE_GET,
		.start = dpll_lock_dumpit,
		.dumpit = dpll_nl_device_get_dumpit,
		.done = dpll_unlock_dumpit,
		.flags = GENL_ADMIN_PERM \| GENL_CMD_CAP_DUMP,
		},
		{
		@@ -129,9 +127,7 @@ static const struct genl_split_ops dpll_nl_ops[] = {
		},
		{
		.cmd = DPLL_CMD_PIN_GET,
		.start = dpll_lock_dumpit,
		.dumpit = dpll_nl_pin_get_dumpit,
		.done = dpll_unlock_dumpit,
		.policy = dpll_pin_get_dump_nl_policy,
		.maxattr = DPLL_A_PIN_ID,
		.flags = GENL_ADMIN_PERM \| GENL_CMD_CAP_DUMP,

drivers/dpll/dpll_nl.h

+0 −2

Original line number	Diff line number	Diff line
		@@ -30,8 +30,6 @@ dpll_post_doit(const struct genl_split_ops ops, struct sk_buff skb,
		void
		dpll_pin_post_doit(const struct genl_split_ops ops, struct sk_buff skb,
		struct genl_info *info);
		int dpll_lock_dumpit(struct netlink_callback *cb);
		int dpll_unlock_dumpit(struct netlink_callback *cb);

		int dpll_nl_device_id_get_doit(struct sk_buff skb, struct genl_info info);
		int dpll_nl_device_get_doit(struct sk_buff skb, struct genl_info info);