[PATCH net 0/3] skb frag: kmap_atomic fixes

[PATCH net 0/3] skb frag: kmap_atomic fixes

[Willem de Bruijn]

From: Willem de Bruijn <willemb@google.com>

skb frags may be backed by highmem and/or compound pages. Various
code calls kmap_atomic to safely access highmem pages. But this
needs additional care for compound pages. Fix a few issues:

patch 1 expect kmap mappings with CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP
patch 2 fixes kmap_atomic + compound page support in skb_seq_read
patch 3 fixes kmap_atomic + compound page support in esp

Willem de Bruijn (3):
  net: support kmap_local forced debugging in skb_frag_foreach
  net: compound page support in skb_seq_read
  esp: avoid unneeded kmap_atomic call

 include/linux/skbuff.h |  3 ++-
 net/core/skbuff.c      | 28 +++++++++++++++++++++++-----
 net/ipv4/esp4.c        |  7 +------
 net/ipv6/esp6.c        |  7 +------
 4 files changed, 27 insertions(+), 18 deletions(-)

-- 
2.30.0.284.gd98b1dd5eaa7-goog

[PATCH net 1/3] net: support kmap_local forced debugging in skb_frag_foreach

[Willem de Bruijn]

From: Willem de Bruijn <willemb@google.com>

Skb frags may be backed by highmem and/or compound pages. Highmem
pages need kmap_atomic mappings to access. But kmap_atomic maps a
single page, not the entire compound page.

skb_foreach_page iterates over an skb frag, in one step in the common
case, page by page only if kmap_atomic must be called for each page.
The decision logic is captured in skb_frag_must_loop.

CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP extends kmap from highmem to all
pages, to increase code coverage.

Extend skb_frag_must_loop to this new condition.

Link: https://lore.kernel.org/linux-mm/20210106180132.41dc249d@gandalf.local.home/
Fixes: 0e91a0c6984c ("mm/highmem: Provide CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP")
Reported-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Tested-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
---
 include/linux/skbuff.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 333bcdc39635..c858adfb5a82 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -366,7 +366,7 @@ static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
 static inline bool skb_frag_must_loop(struct page *p)
 {
 #if defined(CONFIG_HIGHMEM)
-	if (PageHighMem(p))
+	if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
 		return true;
 #endif
 	return false;
-- 
2.30.0.284.gd98b1dd5eaa7-goog

[PATCH net 2/3] net: compound page support in skb_seq_read

[Willem de Bruijn]

From: Willem de Bruijn <willemb@google.com>

skb_seq_read iterates over an skb, returning pointer and length of
the next data range with each call.

It relies on kmap_atomic to access highmem pages when needed.

An skb frag may be backed by a compound page, but kmap_atomic maps
only a single page. There are not enough kmap slots to always map all
pages concurrently.

Instead, if kmap_atomic is needed, iterate over each page.

As this increases the number of calls, avoid this unless needed.
The necessary condition is captured in skb_frag_must_loop.

I tried to make the change as obvious as possible. It should be easy
to verify that nothing changes if skb_frag_must_loop returns false.

Tested:
  On an x86 platform with
    CONFIG_HIGHMEM=y
    CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP=y
    CONFIG_NETFILTER_XT_MATCH_STRING=y

  Run
    ip link set dev lo mtu 1500
    iptables -A OUTPUT -m string --string 'badstring' -algo bm -j ACCEPT
    dd if=/dev/urandom of=in bs=1M count=20
    nc -l -p 8000 > /dev/null &
    nc -w 1 -q 0 localhost 8000 < in

Signed-off-by: Willem de Bruijn <willemb@google.com>
---
 include/linux/skbuff.h |  1 +
 net/core/skbuff.c      | 28 +++++++++++++++++++++++-----
 2 files changed, 24 insertions(+), 5 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index c858adfb5a82..68ffd3f115c1 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1203,6 +1203,7 @@ struct skb_seq_state {
 	struct sk_buff	*root_skb;
 	struct sk_buff	*cur_skb;
 	__u8		*frag_data;
+	__u16		frag_off;
 };
 
 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index f62cae3f75d8..4acf45154b17 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3442,6 +3442,7 @@ void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
 	st->root_skb = st->cur_skb = skb;
 	st->frag_idx = st->stepped_offset = 0;
 	st->frag_data = NULL;
+	st->frag_off = 0;
 }
 EXPORT_SYMBOL(skb_prepare_seq_read);
 
@@ -3496,14 +3497,27 @@ unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
 		st->stepped_offset += skb_headlen(st->cur_skb);
 
 	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
+		unsigned int pg_idx, pg_off, pg_sz;
+
 		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
-		block_limit = skb_frag_size(frag) + st->stepped_offset;
 
+		pg_idx = 0;
+		pg_off = skb_frag_off(frag);
+		pg_sz = skb_frag_size(frag);
+
+		if (skb_frag_must_loop(skb_frag_page(frag))) {
+			pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
+			pg_off = offset_in_page(pg_off + st->frag_off);
+			pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
+						    PAGE_SIZE - pg_off);
+		}
+
+		block_limit = pg_sz + st->stepped_offset;
 		if (abs_offset < block_limit) {
 			if (!st->frag_data)
-				st->frag_data = kmap_atomic(skb_frag_page(frag));
+				st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
 
-			*data = (u8 *) st->frag_data + skb_frag_off(frag) +
+			*data = (u8 *)st->frag_data + pg_off +
 				(abs_offset - st->stepped_offset);
 
 			return block_limit - abs_offset;
@@ -3514,8 +3528,12 @@ unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
 			st->frag_data = NULL;
 		}
 
-		st->frag_idx++;
-		st->stepped_offset += skb_frag_size(frag);
+		st->stepped_offset += pg_sz;
+		st->frag_off += pg_sz;
+		if (st->frag_off == skb_frag_size(frag)) {
+			st->frag_off = 0;
+			st->frag_idx++;
+		}
 	}
 
 	if (st->frag_data) {
-- 
2.30.0.284.gd98b1dd5eaa7-goog

Re: [PATCH net 2/3] net: compound page support in skb_seq_read

[Willem de Bruijn]

On Fri, Jan 8, 2021 at 12:11 PM Willem de Bruijn
<willemdebruijn.kernel@gmail.com> wrote:
>
> From: Willem de Bruijn <willemb@google.com>
>
> skb_seq_read iterates over an skb, returning pointer and length of
> the next data range with each call.
>
> It relies on kmap_atomic to access highmem pages when needed.
>
> An skb frag may be backed by a compound page, but kmap_atomic maps
> only a single page. There are not enough kmap slots to always map all
> pages concurrently.
>
> Instead, if kmap_atomic is needed, iterate over each page.
>
> As this increases the number of calls, avoid this unless needed.
> The necessary condition is captured in skb_frag_must_loop.
>
> I tried to make the change as obvious as possible. It should be easy
> to verify that nothing changes if skb_frag_must_loop returns false.
>
> Tested:
>   On an x86 platform with
>     CONFIG_HIGHMEM=y
>     CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP=y
>     CONFIG_NETFILTER_XT_MATCH_STRING=y
>
>   Run
>     ip link set dev lo mtu 1500
>     iptables -A OUTPUT -m string --string 'badstring' -algo bm -j ACCEPT
>     dd if=/dev/urandom of=in bs=1M count=20
>     nc -l -p 8000 > /dev/null &
>     nc -w 1 -q 0 localhost 8000 < in
>
> Signed-off-by: Willem de Bruijn <willemb@google.com>

I don't have a clear Fixes tag for this.

That was also true for commit c613c209c3f3 ("net: add
skb_frag_foreach_page and use with kmap_atomic"), which deals with the
same problem in a few other functions.

It goes back to when compound highmem pages may have appeared in the
skb frag. Possibly with vmsplice, around 2006. The skb_seq_read
interface itself was added in 2005.

Re: [PATCH net 2/3] net: compound page support in skb_seq_read

[Willem de Bruijn]

On Fri, Jan 8, 2021 at 12:11 PM Willem de Bruijn
<willemdebruijn.kernel@gmail.com> wrote:
>
> From: Willem de Bruijn <willemb@google.com>
>
> skb_seq_read iterates over an skb, returning pointer and length of
> the next data range with each call.
>
> It relies on kmap_atomic to access highmem pages when needed.
>
> An skb frag may be backed by a compound page, but kmap_atomic maps
> only a single page. There are not enough kmap slots to always map all
> pages concurrently.
>
> Instead, if kmap_atomic is needed, iterate over each page.
>
> As this increases the number of calls, avoid this unless needed.
> The necessary condition is captured in skb_frag_must_loop.
>
> I tried to make the change as obvious as possible. It should be easy
> to verify that nothing changes if skb_frag_must_loop returns false.
>
> Tested:
>   On an x86 platform with
>     CONFIG_HIGHMEM=y
>     CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP=y
>     CONFIG_NETFILTER_XT_MATCH_STRING=y
>
>   Run
>     ip link set dev lo mtu 1500
>     iptables -A OUTPUT -m string --string 'badstring' -algo bm -j ACCEPT
>     dd if=/dev/urandom of=in bs=1M count=20
>     nc -l -p 8000 > /dev/null &
>     nc -w 1 -q 0 localhost 8000 < in
>
> Signed-off-by: Willem de Bruijn <willemb@google.com>
> ---
>  include/linux/skbuff.h |  1 +
>  net/core/skbuff.c      | 28 +++++++++++++++++++++++-----
>  2 files changed, 24 insertions(+), 5 deletions(-)
>
> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> index c858adfb5a82..68ffd3f115c1 100644
> --- a/include/linux/skbuff.h
> +++ b/include/linux/skbuff.h
> @@ -1203,6 +1203,7 @@ struct skb_seq_state {
>         struct sk_buff  *root_skb;
>         struct sk_buff  *cur_skb;
>         __u8            *frag_data;
> +       __u16           frag_off;

frags can exceed 64k, so this needs to be __u32, like the other offsets.

I'll have to send a v2.

There's also something to be said for having a

  BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(skb->cb));

as it's getting close. But I won't do that in this stable fix.

[PATCH net 3/3] esp: avoid unneeded kmap_atomic call

[Willem de Bruijn]

From: Willem de Bruijn <willemb@google.com>

esp(6)_output_head uses skb_page_frag_refill to allocate a buffer for
the esp trailer.

It accesses the page with kmap_atomic to handle highmem. But
skb_page_frag_refill can return compound pages, of which
kmap_atomic only maps the first underlying page.

skb_page_frag_refill does not return highmem, because flag
__GFP_HIGHMEM is not set. ESP uses it in the same manner as TCP.
That also does not call kmap_atomic, but directly uses page_address,
in skb_copy_to_page_nocache. Do the same for ESP.

This issue has become easier to trigger with recent kmap local
debugging feature CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP.

Fixes: cac2661c53f3 ("esp4: Avoid skb_cow_data whenever possible")
Fixes: 03e2a30f6a27 ("esp6: Avoid skb_cow_data whenever possible")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
---
 net/ipv4/esp4.c | 7 +------
 net/ipv6/esp6.c | 7 +------
 2 files changed, 2 insertions(+), 12 deletions(-)

diff --git a/net/ipv4/esp4.c b/net/ipv4/esp4.c
index 8b07f3a4f2db..a3271ec3e162 100644
--- a/net/ipv4/esp4.c
+++ b/net/ipv4/esp4.c
@@ -443,7 +443,6 @@ static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
 {
 	u8 *tail;
-	u8 *vaddr;
 	int nfrags;
 	int esph_offset;
 	struct page *page;
@@ -485,14 +484,10 @@ int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *
 			page = pfrag->page;
 			get_page(page);
 
-			vaddr = kmap_atomic(page);
-
-			tail = vaddr + pfrag->offset;
+			tail = page_address(page) + pfrag->offset;
 
 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
 
-			kunmap_atomic(vaddr);
-
 			nfrags = skb_shinfo(skb)->nr_frags;
 
 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
diff --git a/net/ipv6/esp6.c b/net/ipv6/esp6.c
index 52c2f063529f..2b804fcebcc6 100644
--- a/net/ipv6/esp6.c
+++ b/net/ipv6/esp6.c
@@ -478,7 +478,6 @@ static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
 int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
 {
 	u8 *tail;
-	u8 *vaddr;
 	int nfrags;
 	int esph_offset;
 	struct page *page;
@@ -519,14 +518,10 @@ int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info
 			page = pfrag->page;
 			get_page(page);
 
-			vaddr = kmap_atomic(page);
-
-			tail = vaddr + pfrag->offset;
+			tail = page_address(page) + pfrag->offset;
 
 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
 
-			kunmap_atomic(vaddr);
-
 			nfrags = skb_shinfo(skb)->nr_frags;
 
 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
-- 
2.30.0.284.gd98b1dd5eaa7-goog