[PATCH 1/2] drivers: net: use xdp only inside tun_build_skb()

[PATCH 1/2] drivers: net: use xdp only inside tun_build_skb()

[Andrew Kanner]

Syzkaller reported the following issue:
=======================================
Too BIG xdp->frame_sz = 131072
WARNING: CPU: 0 PID: 5020 at net/core/filter.c:4121
  ____bpf_xdp_adjust_tail net/core/filter.c:4121 [inline]
WARNING: CPU: 0 PID: 5020 at net/core/filter.c:4121
  bpf_xdp_adjust_tail+0x466/0xa10 net/core/filter.c:4103
...
Call Trace:
 <TASK>
 bpf_prog_4add87e5301a4105+0x1a/0x1c
 __bpf_prog_run include/linux/filter.h:600 [inline]
 bpf_prog_run_xdp include/linux/filter.h:775 [inline]
 bpf_prog_run_generic_xdp+0x57e/0x11e0 net/core/dev.c:4721
 netif_receive_generic_xdp net/core/dev.c:4807 [inline]
 do_xdp_generic+0x35c/0x770 net/core/dev.c:4866
 tun_get_user+0x2340/0x3ca0 drivers/net/tun.c:1919
 tun_chr_write_iter+0xe8/0x210 drivers/net/tun.c:2043
 call_write_iter include/linux/fs.h:1871 [inline]
 new_sync_write fs/read_write.c:491 [inline]
 vfs_write+0x650/0xe40 fs/read_write.c:584
 ksys_write+0x12f/0x250 fs/read_write.c:637
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x38/0xb0 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x63/0xcd

frame_sz < PAGE_SIZE check was introduced in commit c8741e2bfe87
("xdp: Allow bpf_xdp_adjust_tail() to grow packet size"). But
tun_get_user() still provides an execution path to exceed XDP limits
for packet size. The way to mitigate this is to reduce xdp to be used
only in tun_build_skb(), without falling in tun_alloc_skb(), etc.

Reported-and-tested-by: syzbot+f817490f5bd20541b90a@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/000000000000774b9205f1d8a80d@google.com/T/
Link: https://syzkaller.appspot.com/bug?id=5335c7c62bfff89bbb1c8f14cdabebe91909060f
Signed-off-by: Andrew Kanner <andrew.kanner@gmail.com>
---
 drivers/net/tun.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index d75456adc62a..18ccbbe9830a 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1580,8 +1580,14 @@ static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
 }
 
 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
-			      int len, int noblock, bool zerocopy)
+			      int len, int noblock, bool zerocopy, int *skb_xdp)
 {
+	if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
+	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE) {
+		*skb_xdp = 0;
+		return false;
+	}
+
 	if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
 		return false;
 
@@ -1594,10 +1600,6 @@ static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
 	if (zerocopy)
 		return false;
 
-	if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
-	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
-		return false;
-
 	return true;
 }
 
@@ -1809,7 +1811,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 			zerocopy = true;
 	}
 
-	if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
+	if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy, &skb_xdp) && !frags) {
 		/* For the packet that is not easy to be processed
 		 * (e.g gso or jumbo packet), we will do it at after
 		 * skb was created with generic XDP routine.
-- 
2.39.3

[PATCH 2/2] drivers: net: prevent tun_can_build_skb() to exceed xdp size limits

[Andrew Kanner]

Tested with syzkaller repro with reduced packet size. It was
discovered that XDP_PACKET_HEADROOM is not checked in
tun_can_build_skb(), although pad may be incremented in
tun_build_skb().

Fixes: 7df13219d757 ("tun: reserve extra headroom only when XDP is set")
Link: https://syzkaller.appspot.com/text?tag=ReproC&x=12b2593ea80000
Signed-off-by: Andrew Kanner <andrew.kanner@gmail.com>
---
 drivers/net/tun.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 18ccbbe9830a..cdf2bd85b383 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1582,7 +1582,13 @@ static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
 			      int len, int noblock, bool zerocopy, int *skb_xdp)
 {
-	if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
+	int pad = TUN_RX_PAD;
+	struct bpf_prog *xdp_prog = rcu_dereference(tun->xdp_prog);
+
+	if (xdp_prog)
+		pad += XDP_PACKET_HEADROOM;
+
+	if (SKB_DATA_ALIGN(len + pad) +
 	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE) {
 		*skb_xdp = 0;
 		return false;
-- 
2.39.3

Re: [PATCH 1/2] drivers: net: use xdp only inside tun_build_skb()

[Jason Wang]

On Tue, Jul 25, 2023 at 6:14 AM Andrew Kanner <andrew.kanner@gmail.com> wrote:
>
> Syzkaller reported the following issue:
> =======================================
> Too BIG xdp->frame_sz = 131072
> WARNING: CPU: 0 PID: 5020 at net/core/filter.c:4121
>   ____bpf_xdp_adjust_tail net/core/filter.c:4121 [inline]
> WARNING: CPU: 0 PID: 5020 at net/core/filter.c:4121
>   bpf_xdp_adjust_tail+0x466/0xa10 net/core/filter.c:4103
> ...
> Call Trace:
>  <TASK>
>  bpf_prog_4add87e5301a4105+0x1a/0x1c
>  __bpf_prog_run include/linux/filter.h:600 [inline]
>  bpf_prog_run_xdp include/linux/filter.h:775 [inline]
>  bpf_prog_run_generic_xdp+0x57e/0x11e0 net/core/dev.c:4721
>  netif_receive_generic_xdp net/core/dev.c:4807 [inline]
>  do_xdp_generic+0x35c/0x770 net/core/dev.c:4866
>  tun_get_user+0x2340/0x3ca0 drivers/net/tun.c:1919
>  tun_chr_write_iter+0xe8/0x210 drivers/net/tun.c:2043
>  call_write_iter include/linux/fs.h:1871 [inline]
>  new_sync_write fs/read_write.c:491 [inline]
>  vfs_write+0x650/0xe40 fs/read_write.c:584
>  ksys_write+0x12f/0x250 fs/read_write.c:637
>  do_syscall_x64 arch/x86/entry/common.c:50 [inline]
>  do_syscall_64+0x38/0xb0 arch/x86/entry/common.c:80
>  entry_SYSCALL_64_after_hwframe+0x63/0xcd
>
> frame_sz < PAGE_SIZE check was introduced in commit c8741e2bfe87
> ("xdp: Allow bpf_xdp_adjust_tail() to grow packet size").

I wonder if this check makes sense for xdp generic when skb is linearized.

> But
> tun_get_user() still provides an execution path to exceed XDP limits
> for packet size.

I'd suggest being specific to the execution path, which is actually
the generic XDP.

> The way to mitigate this is to reduce xdp to be used
> only in tun_build_skb(), without falling in tun_alloc_skb(), etc.
>
> Reported-and-tested-by: syzbot+f817490f5bd20541b90a@syzkaller.appspotmail.com
> Closes: https://lore.kernel.org/all/000000000000774b9205f1d8a80d@google.com/T/
> Link: https://syzkaller.appspot.com/bug?id=5335c7c62bfff89bbb1c8f14cdabebe91909060f
> Signed-off-by: Andrew Kanner <andrew.kanner@gmail.com>
> ---
>  drivers/net/tun.c | 14 ++++++++------
>  1 file changed, 8 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/net/tun.c b/drivers/net/tun.c
> index d75456adc62a..18ccbbe9830a 100644
> --- a/drivers/net/tun.c
> +++ b/drivers/net/tun.c
> @@ -1580,8 +1580,14 @@ static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
>  }
>
>  static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
> -                             int len, int noblock, bool zerocopy)
> +                             int len, int noblock, bool zerocopy, int *skb_xdp)
>  {
> +       if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
> +           SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE) {
> +               *skb_xdp = 0;

So if XDP is enabled, the packet will bypass the bpf program which is
sub-optimal. I think we should at least drop the packet in this case.

Thanks

> +               return false;
> +       }
> +
>         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
>                 return false;
>
> @@ -1594,10 +1600,6 @@ static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
>         if (zerocopy)
>                 return false;
>
> -       if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
> -           SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
> -               return false;
> -
>         return true;
>  }
>
> @@ -1809,7 +1811,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
>                         zerocopy = true;
>         }
>
> -       if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
> +       if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy, &skb_xdp) && !frags) {
>                 /* For the packet that is not easy to be processed
>                  * (e.g gso or jumbo packet), we will do it at after
>                  * skb was created with generic XDP routine.
> --
> 2.39.3
>

Re: [PATCH 2/2] drivers: net: prevent tun_can_build_skb() to exceed xdp size limits

[Jason Wang]

On Tue, Jul 25, 2023 at 6:15 AM Andrew Kanner <andrew.kanner@gmail.com> wrote:
>
> Tested with syzkaller repro with reduced packet size. It was
> discovered that XDP_PACKET_HEADROOM is not checked in
> tun_can_build_skb(), although pad may be incremented in
> tun_build_skb().
>
> Fixes: 7df13219d757 ("tun: reserve extra headroom only when XDP is set")
> Link: https://syzkaller.appspot.com/text?tag=ReproC&x=12b2593ea80000
> Signed-off-by: Andrew Kanner <andrew.kanner@gmail.com>
> ---
>  drivers/net/tun.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/tun.c b/drivers/net/tun.c
> index 18ccbbe9830a..cdf2bd85b383 100644
> --- a/drivers/net/tun.c
> +++ b/drivers/net/tun.c
> @@ -1582,7 +1582,13 @@ static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
>  static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
>                               int len, int noblock, bool zerocopy, int *skb_xdp)
>  {
> -       if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
> +       int pad = TUN_RX_PAD;
> +       struct bpf_prog *xdp_prog = rcu_dereference(tun->xdp_prog);

This misses rcu read lock.

I wonder if things could be simpler if we move the limit check from
tun_can_build_skb() to tun_build_skb():

rcu_read_lock();
xdp_prog = rcu_dereference(tun->xdp_prog);
        if (xdp_prog)
                pad += XDP_PACKET_HEADROOM;
buflen += SKB_DATA_ALIGN(len + pad);
rcu_read_unlock();

Thanks

> +
> +       if (xdp_prog)
> +               pad += XDP_PACKET_HEADROOM;
> +
> +       if (SKB_DATA_ALIGN(len + pad) +
>             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE) {
>                 *skb_xdp = 0;
>                 return false;
> --
> 2.39.3
>

Re: [PATCH 2/2] drivers: net: prevent tun_can_build_skb() to exceed xdp size limits

[Andrew Kanner]

On Tue, Jul 25, 2023 at 11:39:46AM +0800, Jason Wang wrote:
> On Tue, Jul 25, 2023 at 6:15 AM Andrew Kanner <andrew.kanner@gmail.com> wrote:
> >
> > Tested with syzkaller repro with reduced packet size. It was
> > discovered that XDP_PACKET_HEADROOM is not checked in
> > tun_can_build_skb(), although pad may be incremented in
> > tun_build_skb().
> >
> > Fixes: 7df13219d757 ("tun: reserve extra headroom only when XDP is set")
> > Link: https://syzkaller.appspot.com/text?tag=ReproC&x=12b2593ea80000
> > Signed-off-by: Andrew Kanner <andrew.kanner@gmail.com>
> > ---
> >  drivers/net/tun.c | 8 +++++++-
> >  1 file changed, 7 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/net/tun.c b/drivers/net/tun.c
> > index 18ccbbe9830a..cdf2bd85b383 100644
> > --- a/drivers/net/tun.c
> > +++ b/drivers/net/tun.c
> > @@ -1582,7 +1582,13 @@ static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
> >  static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
> >                               int len, int noblock, bool zerocopy, int *skb_xdp)
> >  {
> > -       if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
> > +       int pad = TUN_RX_PAD;
> > +       struct bpf_prog *xdp_prog = rcu_dereference(tun->xdp_prog);
> 
> This misses rcu read lock.
> 
> I wonder if things could be simpler if we move the limit check from
> tun_can_build_skb() to tun_build_skb():
> 
> rcu_read_lock();
> xdp_prog = rcu_dereference(tun->xdp_prog);
>         if (xdp_prog)
>                 pad += XDP_PACKET_HEADROOM;
> buflen += SKB_DATA_ALIGN(len + pad);
> rcu_read_unlock();
> 
> Thanks
> 

Thanks, I missed the part with rcu read lock for some reason.

It's a good idea to move / reduce duplication. Let me think and try to
fix according to your comments, I will resend it as v2.

-- 
Andrew Kanner