Skip to contentSkip to conversation

bpf: fix a potential use-after-free in bpf_link_free()

PATCH Applied Cong Wang 4 participants 5 messages Jun 2, 2024 – Jun 4, 2024

Subsystems (4)
lore
Share & tools
lore

Cong Wang

author
#0 lore raw

PATCH 1/1 bpf: fix a potential use-after-free in bpf_link_free()

From: Cong Wang <[email protected]>

After commit 1a80dbcb2dba, bpf_link can be freed by link->ops->dealloc_deferred, but the code still tests and uses link->ops->dealloc afterward, which leads to a use-after-free as reported by syzbot. Actually, one of them should be sufficient, so just call one of them instead of both. Also add a WARN_ON() in case of any problematic implementation.

Reported-by [email protected]
Fixes 1a80dbcb2dba ("bpf: support deferring bpf_link dealloc to after RCU grace period")
Cc Andrii Nakryiko <[email protected]>
---
 kernel/bpf/syscall.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/bpf/syscall.c +2 -2
index 2222c3ff88e7..d8f244069495 100644
@@ -2998,6 +2998,7 @@ static int bpf_obj_get(const union bpf_attr *attr)
 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
 		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
 {
+	WARN_ON(ops->dealloc && ops->dealloc_deferred);
 	atomic64_set(&link->refcnt, 1);
 	link->type = type;
 	link->id = 0;
@@ -3074,8 +3075,7 @@ static void bpf_link_free(struct bpf_link *link)
 			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
 		else
 			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
-	}
-	if (link->ops->dealloc)
+	} else if (link->ops->dealloc)
 		link->ops->dealloc(link);
 }
 
Signature · 1 line
2.34.1
Acked-by on v1 1/1
#1 lore raw
Quoted 30 lines · Cong Wang

From: Cong Wang <[email protected]>

After commit 1a80dbcb2dba, bpf_link can be freed by link->ops->dealloc_deferred, but the code still tests and uses link->ops->dealloc afterward, which leads to a use-after-free as reported by syzbot. Actually, one of them should be sufficient, so just call one of them instead of both. Also add a WARN_ON() in case of any problematic implementation.

Reported-by [email protected]
Fixes 1a80dbcb2dba ("bpf: support deferring bpf_link dealloc to after RCU grace period")
Cc Andrii Nakryiko <[email protected]>
---
 kernel/bpf/syscall.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/bpf/syscall.c +1 -0
index 2222c3ff88e7..d8f244069495 100644
@@ -2998,6 +2998,7 @@ static int bpf_obj_get(const union bpf_attr *attr)
 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
 		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
 {
+	WARN_ON(ops->dealloc && ops->dealloc_deferred);
 	atomic64_set(&link->refcnt, 1);
 	link->type = type;
 	link->id = 0;

kernel/bpf/syscall.c static void bpf_link_free(struct bpf_link *link) · line 3079

Quoted 7 lines
kernel/bpf/syscall.c +1 -2
@@ -3074,8 +3075,7 @@ static void bpf_link_free(struct bpf_link *link)
 			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
 		else
 			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
-	}
-	if (link->ops->dealloc)
+	} else if (link->ops->dealloc)
 		link->ops->dealloc(link);

nice catch

jirka

maintainer Acked-by Jiri Olsa
#2 lore raw
Quoted 32 lines · Jiri Olsa
Quoted 31 lines · Cong Wang

From: Cong Wang <[email protected]>

After commit 1a80dbcb2dba, bpf_link can be freed by link->ops->dealloc_deferred, but the code still tests and uses link->ops->dealloc afterward, which leads to a use-after-free as reported by syzbot. Actually, one of them should be sufficient, so just call one of them instead of both. Also add a WARN_ON() in case of any problematic implementation.

Reported-by [email protected]
Fixes 1a80dbcb2dba ("bpf: support deferring bpf_link dealloc to after RCU grace period")
Cc Andrii Nakryiko <[email protected]>
---
  kernel/bpf/syscall.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
kernel/bpf/syscall.c +1 -0
index 2222c3ff88e7..d8f244069495 100644
@@ -2998,6 +2998,7 @@ static int bpf_obj_get(const union bpf_attr *attr)
  void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
  		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
  {
+	WARN_ON(ops->dealloc && ops->dealloc_deferred);
  	atomic64_set(&link->refcnt, 1);
  	link->type = type;
  	link->id = 0;
@@ -3074,8 +3075,7 @@ static void bpf_link_free(struct bpf_link *link)
  			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
Quoted 8 lines

kernel/bpf/syscall.c static void bpf_link_free(struct bpf_link *link) · line 3079

Quoted 6 lines
kernel/bpf/syscall.c +1 -2
  		else
  			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
-	}
-	if (link->ops->dealloc)
+	} else if (link->ops->dealloc)
  		link->ops->dealloc(link);

nice catch

+1, thanks Cong !

Quoted 1 line

I think it would also be slightly nicer to just fetch the ops once, which wouldn't have caused the issue if it was done back then in the first place. Do you mind if I squash this in and then apply it to bpf tree? Looks as follows :

From 220fb40d88c7b5d1f3234a5f1b807f6067450832 Mon Sep 17 00:00:00 2001
Message-Id: <220fb40d88c7b5d1f3234a5f1b807f6067450832.1717431085.git.daniel@iogearbox.net>
From: Cong Wang <[email protected]>
Date: Sun, 2 Jun 2024 11:27:03 -0700
Subject: [PATCH bpf] bpf: Fix a potential use-after-free in bpf_link_free()

After commit 1a80dbcb2dba, bpf_link can be freed by link->ops->dealloc_deferred, but the code still tests and uses link->ops->dealloc afterward, which leads to a use-after-free as reported by syzbot. Actually, one of them should be sufficient, so just call one of them instead of both. Also add a WARN_ON() in case of any problematic implementation.

Fixes 1a80dbcb2dba ("bpf: support deferring bpf_link dealloc to after RCU grace period")
Reported-by [email protected]
---
 kernel/bpf/syscall.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
kernel/bpf/syscall.c +6 -5
index 5070fa20d05c..869265852d51 100644
@@ -2998,6 +2998,7 @@ static int bpf_obj_get(const union bpf_attr *attr)
 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
 		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
 {
+	WARN_ON(ops->dealloc && ops->dealloc_deferred);
 	atomic64_set(&link->refcnt, 1);
 	link->type = type;
 	link->id = 0;
@@ -3056,16 +3057,17 @@ static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
 /* bpf_link_free is guaranteed to be called from process context */
 static void bpf_link_free(struct bpf_link *link)
 {
+	const struct bpf_link_ops *ops = link->ops;
 	bool sleepable = false;
 
 	bpf_link_free_id(link->id);
 	if (link->prog) {
 		sleepable = link->prog->sleepable;
 		/* detach BPF program, clean up used resources */
-		link->ops->release(link);
+		ops->release(link);
 		bpf_prog_put(link->prog);
 	}
-	if (link->ops->dealloc_deferred) {
+	if (ops->dealloc_deferred) {
 		/* schedule BPF link deallocation; if underlying BPF program
 		 * is sleepable, we need to first wait for RCU tasks trace
 		 * sync, then go through "classic" RCU grace period
@@ -3074,9 +3076,8 @@ static void bpf_link_free(struct bpf_link *link)
 			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
 		else
 			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
-	}
-	if (link->ops->dealloc)
-		link->ops->dealloc(link);
+	} else if (ops->dealloc)
+		ops->dealloc(link);
 }
 
 static void bpf_link_put_deferred(struct work_struct *work)
Signature · 1 line
2.21.0
author
#3 lore raw
Quoted 42 lines · Daniel Borkmann
Quoted 32 lines · Jiri Olsa
Quoted 31 lines · Cong Wang

From: Cong Wang <[email protected]>

After commit 1a80dbcb2dba, bpf_link can be freed by link->ops->dealloc_deferred, but the code still tests and uses link->ops->dealloc afterward, which leads to a use-after-free as reported by syzbot. Actually, one of them should be sufficient, so just call one of them instead of both. Also add a WARN_ON() in case of any problematic implementation.

Reported-by [email protected]
Fixes 1a80dbcb2dba ("bpf: support deferring bpf_link dealloc to after RCU grace period")
Cc Andrii Nakryiko <[email protected]>
---
  kernel/bpf/syscall.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
kernel/bpf/syscall.c +1 -0
index 2222c3ff88e7..d8f244069495 100644
@@ -2998,6 +2998,7 @@ static int bpf_obj_get(const union bpf_attr *attr)
  void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
  		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
  {
+	WARN_ON(ops->dealloc && ops->dealloc_deferred);
  	atomic64_set(&link->refcnt, 1);
  	link->type = type;
  	link->id = 0;
@@ -3074,8 +3075,7 @@ static void bpf_link_free(struct bpf_link *link)
  			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
Quoted 8 lines

kernel/bpf/syscall.c static void bpf_link_free(struct bpf_link *link)

Quoted 6 lines
kernel/bpf/syscall.c +1 -2
  		else
  			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
-	}
-	if (link->ops->dealloc)
+	} else if (link->ops->dealloc)
  		link->ops->dealloc(link);

nice catch

Quoted 9 lines

+1, thanks Cong !

Quoted 1 line

I think it would also be slightly nicer to just fetch the ops once, which wouldn't have caused the issue if it was done back then in the first place. Do you mind if I squash this in and then apply it to bpf tree? Looks as follows :

Sounds good to me.

Thanks.

#4 lore raw
In reply to Cong Wang
This patch was applied to bpf/bpf.git (master) by Daniel Borkmann <[email protected]>:

Hello:

This patch was applied to bpf/bpf.git (master)
by Daniel Borkmann <[email protected]>:

Quoted 10 lines · On Sun, 2 Jun 2024 11:27:03 -0700 you wrote:

From: Cong Wang <[email protected]>

After commit 1a80dbcb2dba, bpf_link can be freed by link->ops->dealloc_deferred, but the code still tests and uses link->ops->dealloc afterward, which leads to a use-after-free as reported by syzbot. Actually, one of them should be sufficient, so just call one of them instead of both. Also add a WARN_ON() in case of any problematic implementation.

[...]

Here is the summary with links:
- [bpf] bpf: fix a potential use-after-free in bpf_link_free()

    https://git.kernel.org/bpf/bpf/c/2884dc7d08d9

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html