diff mbox series

[net] tcp: add sanity checks to rx zerocopy

Message ID 20240125103317.2334989-1-edumazet@google.com (mailing list archive)
State New, archived
Headers show
Series [net] tcp: add sanity checks to rx zerocopy | expand

Commit Message

Eric Dumazet Jan. 25, 2024, 10:33 a.m. UTC
TCP rx zerocopy intent is to map pages initially allocated
from NIC drivers, not pages owned by a fs.

This patch adds to can_map_frag() these additional checks:

- Page must not be a compound one.
- page->mapping must be NULL.

This fixes the panic reported by ZhangPeng.

syzbot was able to loopback packets built with sendfile(),
mapping pages owned by an ext4 file to TCP rx zerocopy.

r3 = socket$inet_tcp(0x2, 0x1, 0x0)
mmap(&(0x7f0000ff9000/0x4000)=nil, 0x4000, 0x0, 0x12, r3, 0x0)
r4 = socket$inet_tcp(0x2, 0x1, 0x0)
bind$inet(r4, &(0x7f0000000000)={0x2, 0x4e24, @multicast1}, 0x10)
connect$inet(r4, &(0x7f00000006c0)={0x2, 0x4e24, @empty}, 0x10)
r5 = openat$dir(0xffffffffffffff9c, &(0x7f00000000c0)='./file0\x00',
    0x181e42, 0x0)
fallocate(r5, 0x0, 0x0, 0x85b8)
sendfile(r4, r5, 0x0, 0x8ba0)
getsockopt$inet_tcp_TCP_ZEROCOPY_RECEIVE(r4, 0x6, 0x23,
    &(0x7f00000001c0)={&(0x7f0000ffb000/0x3000)=nil, 0x3000, 0x0, 0x0, 0x0,
    0x0, 0x0, 0x0, 0x0}, &(0x7f0000000440)=0x40)
r6 = openat$dir(0xffffffffffffff9c, &(0x7f00000000c0)='./file0\x00',
    0x181e42, 0x0)

Fixes: 93ab6cc69162 ("tcp: implement mmap() for zero copy receive")
Link: https://lore.kernel.org/netdev/5106a58e-04da-372a-b836-9d3d0bd2507b@huawei.com/T/
Reported-and-bisected-by: ZhangPeng <zhangpeng362@huawei.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Arjun Roy <arjunroy@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: linux-mm@vger.kernel.org
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: linux-fsdevel@vger.kernel.org
---
 net/ipv4/tcp.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

Comments

Matthew Wilcox Jan. 25, 2024, 4:07 p.m. UTC | #1
On Thu, Jan 25, 2024 at 10:33:17AM +0000, Eric Dumazet wrote:
> +++ b/net/ipv4/tcp.c
> @@ -1786,7 +1786,17 @@ static skb_frag_t *skb_advance_to_frag(struct sk_buff *skb, u32 offset_skb,
>  
>  static bool can_map_frag(const skb_frag_t *frag)
>  {
> -	return skb_frag_size(frag) == PAGE_SIZE && !skb_frag_off(frag);
> +	struct page *page;
> +
> +	if (skb_frag_size(frag) != PAGE_SIZE || skb_frag_off(frag))
> +		return false;
> +
> +	page = skb_frag_page(frag);
> +
> +	if (PageCompound(page) || page->mapping)
> +		return false;

I'm not entirely sure why you're testing PageCompound here.  If a driver
allocates a compound page, we'd still want to be able to insert it,
right?

I have a feeling that we want to fix this in the VM layer.  There are
some weird places calling vm_insert_page() and we should probably make
them all fail.

Something like this, perhaps?

diff --git a/mm/memory.c b/mm/memory.c
index 1a60faad2e49..ae0abab56d38 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1871,6 +1871,10 @@ static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte,
 
 	if (!pte_none(ptep_get(pte)))
 		return -EBUSY;
+	if (folio->mapping &&
+	    ((addr - vma->vm_start) / PAGE_SIZE + vma->vm_pgoff) !=
+	    (folio->index + folio_page_idx(folio, page)))
+		return -EINVAL;
 	/* Ok, finally just insert the thing.. */
 	folio_get(folio);
 	inc_mm_counter(vma->vm_mm, mm_counter_file(folio));
Matthew Wilcox Jan. 25, 2024, 4:09 p.m. UTC | #2
Fixing email address for linux-mm.

On Thu, Jan 25, 2024 at 04:07:50PM +0000, Matthew Wilcox wrote:
> On Thu, Jan 25, 2024 at 10:33:17AM +0000, Eric Dumazet wrote:
> > +++ b/net/ipv4/tcp.c
> > @@ -1786,7 +1786,17 @@ static skb_frag_t *skb_advance_to_frag(struct sk_buff *skb, u32 offset_skb,
> >  
> >  static bool can_map_frag(const skb_frag_t *frag)
> >  {
> > -	return skb_frag_size(frag) == PAGE_SIZE && !skb_frag_off(frag);
> > +	struct page *page;
> > +
> > +	if (skb_frag_size(frag) != PAGE_SIZE || skb_frag_off(frag))
> > +		return false;
> > +
> > +	page = skb_frag_page(frag);
> > +
> > +	if (PageCompound(page) || page->mapping)
> > +		return false;
> 
> I'm not entirely sure why you're testing PageCompound here.  If a driver
> allocates a compound page, we'd still want to be able to insert it,
> right?
> 
> I have a feeling that we want to fix this in the VM layer.  There are
> some weird places calling vm_insert_page() and we should probably make
> them all fail.
> 
> Something like this, perhaps?
> 
> diff --git a/mm/memory.c b/mm/memory.c
> index 1a60faad2e49..ae0abab56d38 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1871,6 +1871,10 @@ static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte,
>  
>  	if (!pte_none(ptep_get(pte)))
>  		return -EBUSY;
> +	if (folio->mapping &&
> +	    ((addr - vma->vm_start) / PAGE_SIZE + vma->vm_pgoff) !=
> +	    (folio->index + folio_page_idx(folio, page)))
> +		return -EINVAL;
>  	/* Ok, finally just insert the thing.. */
>  	folio_get(folio);
>  	inc_mm_counter(vma->vm_mm, mm_counter_file(folio));
>
Eric Dumazet Jan. 25, 2024, 4:54 p.m. UTC | #3
On Thu, Jan 25, 2024 at 5:09 PM Matthew Wilcox <willy@infradead.org> wrote:
>
>
> Fixing email address for linux-mm.
>
> On Thu, Jan 25, 2024 at 04:07:50PM +0000, Matthew Wilcox wrote:
> > On Thu, Jan 25, 2024 at 10:33:17AM +0000, Eric Dumazet wrote:
> > > +++ b/net/ipv4/tcp.c
> > > @@ -1786,7 +1786,17 @@ static skb_frag_t *skb_advance_to_frag(struct sk_buff *skb, u32 offset_skb,
> > >
> > >  static bool can_map_frag(const skb_frag_t *frag)
> > >  {
> > > -   return skb_frag_size(frag) == PAGE_SIZE && !skb_frag_off(frag);
> > > +   struct page *page;
> > > +
> > > +   if (skb_frag_size(frag) != PAGE_SIZE || skb_frag_off(frag))
> > > +           return false;
> > > +
> > > +   page = skb_frag_page(frag);
> > > +
> > > +   if (PageCompound(page) || page->mapping)
> > > +           return false;
> >
> > I'm not entirely sure why you're testing PageCompound here.  If a driver
> > allocates a compound page, we'd still want to be able to insert it,
> > right?

I tried to get something that would be free of merge conflicts, up to linux-4.18
I was not sure if I had to use compound_head(page) in order to test
for the mapping ?

page = compound_head(page);
if (page->mapping)
     return false;

I guess that we would have to adjust the page pointer based on
skb_frag_off(frag),
right now we bail if skb_frag_off(frag) is not zero.

I would leave this change for future kernels if there is interest.

> >
> > I have a feeling that we want to fix this in the VM layer.  There are
> > some weird places calling vm_insert_page() and we should probably make
> > them all fail.
> >
> > Something like this, perhaps?


Perhaps, but backports to stable versions (without folio) would be a
bit of a work ?

> >
> > diff --git a/mm/memory.c b/mm/memory.c
> > index 1a60faad2e49..ae0abab56d38 100644
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -1871,6 +1871,10 @@ static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte,
> >
> >       if (!pte_none(ptep_get(pte)))
> >               return -EBUSY;
> > +     if (folio->mapping &&
> > +         ((addr - vma->vm_start) / PAGE_SIZE + vma->vm_pgoff) !=
> > +         (folio->index + folio_page_idx(folio, page)))
> > +             return -EINVAL;
> >       /* Ok, finally just insert the thing.. */
> >       folio_get(folio);
> >       inc_mm_counter(vma->vm_mm, mm_counter_file(folio));
> >
Arjun Roy Jan. 25, 2024, 10:23 p.m. UTC | #4
On Thu, Jan 25, 2024 at 8:07 AM Matthew Wilcox <willy@infradead.org> wrote:
>
> On Thu, Jan 25, 2024 at 10:33:17AM +0000, Eric Dumazet wrote:
> > +++ b/net/ipv4/tcp.c
> > @@ -1786,7 +1786,17 @@ static skb_frag_t *skb_advance_to_frag(struct sk_buff *skb, u32 offset_skb,
> >
> >  static bool can_map_frag(const skb_frag_t *frag)
> >  {
> > -     return skb_frag_size(frag) == PAGE_SIZE && !skb_frag_off(frag);
> > +     struct page *page;
> > +
> > +     if (skb_frag_size(frag) != PAGE_SIZE || skb_frag_off(frag))
> > +             return false;
> > +
> > +     page = skb_frag_page(frag);
> > +
> > +     if (PageCompound(page) || page->mapping)
> > +             return false;
>
> I'm not entirely sure why you're testing PageCompound here.  If a driver
> allocates a compound page, we'd still want to be able to insert it,
> right?
>

Resend b/c I forgot I was in HTML mode email, oops.

Is there a common use case for a NIC driver to be doing this? I was
under the impression NIC drivers would get pages individually since it
would be harder to find physically contiguous groupings in general.

Anyways, a possible reason to not allow compound pages - if we have
some large memory range and different receiving users are interested
in different parts of the range in userspace - since the whole range
is pinned till everyone is done with it, it can lead to worse cases of
memory lingering when some user only wanted 10 bytes out of some N *
PAGE_SIZE range for a multi-minute long period.

-Arjun


> I have a feeling that we want to fix this in the VM layer.  There are
> some weird places calling vm_insert_page() and we should probably make
> them all fail.
>
> Something like this, perhaps?
>
> diff --git a/mm/memory.c b/mm/memory.c
> index 1a60faad2e49..ae0abab56d38 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1871,6 +1871,10 @@ static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte,
>
>         if (!pte_none(ptep_get(pte)))
>                 return -EBUSY;
> +       if (folio->mapping &&
> +           ((addr - vma->vm_start) / PAGE_SIZE + vma->vm_pgoff) !=
> +           (folio->index + folio_page_idx(folio, page)))
> +               return -EINVAL;
>         /* Ok, finally just insert the thing.. */
>         folio_get(folio);
>         inc_mm_counter(vma->vm_mm, mm_counter_file(folio));
patchwork-bot+netdevbpf@kernel.org Jan. 29, 2024, 12:10 p.m. UTC | #5
Hello:

This patch was applied to netdev/net.git (main)
by David S. Miller <davem@davemloft.net>:

On Thu, 25 Jan 2024 10:33:17 +0000 you wrote:
> TCP rx zerocopy intent is to map pages initially allocated
> from NIC drivers, not pages owned by a fs.
> 
> This patch adds to can_map_frag() these additional checks:
> 
> - Page must not be a compound one.
> - page->mapping must be NULL.
> 
> [...]

Here is the summary with links:
  - [net] tcp: add sanity checks to rx zerocopy
    https://git.kernel.org/netdev/net/c/577e4432f3ac

You are awesome, thank you!
diff mbox series

Patch

diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index a1c6de385ccef91fe3c3e072ac5d2a20f0394a2b..7e2481b9eae1b791e1ec65f39efa41837a9fcbd3 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -1786,7 +1786,17 @@  static skb_frag_t *skb_advance_to_frag(struct sk_buff *skb, u32 offset_skb,
 
 static bool can_map_frag(const skb_frag_t *frag)
 {
-	return skb_frag_size(frag) == PAGE_SIZE && !skb_frag_off(frag);
+	struct page *page;
+
+	if (skb_frag_size(frag) != PAGE_SIZE || skb_frag_off(frag))
+		return false;
+
+	page = skb_frag_page(frag);
+
+	if (PageCompound(page) || page->mapping)
+		return false;
+
+	return true;
 }
 
 static int find_next_mappable_frag(const skb_frag_t *frag,