On 3/20/18 4:50 PM, Alexander Duyck wrote:
On Tue, Mar 20, 2018 at 4:21 PM, Yonghong Song <y...@fb.com> wrote:
One of our in-house projects, bpf-based NAT, hits a kernel BUG_ON at
function skb_segment(), line 3667. The bpf program attaches to
clsact ingress, calls bpf_skb_change_proto to change protocol
from ipv4 to ipv6 or from ipv6 to ipv4, and then calls bpf_redirect
to send the changed packet out.
3472 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3473 netdev_features_t features)
3474 {
3475 struct sk_buff *segs = NULL;
3476 struct sk_buff *tail = NULL;
...
3665 while (pos < offset + len) {
3666 if (i >= nfrags) {
3667 BUG_ON(skb_headlen(list_skb));
3668
3669 i = 0;
3670 nfrags = skb_shinfo(list_skb)->nr_frags;
3671 frag = skb_shinfo(list_skb)->frags;
3672 frag_skb = list_skb;
...
call stack:
...
#1 [ffff883ffef03558] __crash_kexec at ffffffff8110c525
#2 [ffff883ffef03620] crash_kexec at ffffffff8110d5cc
#3 [ffff883ffef03640] oops_end at ffffffff8101d7e7
#4 [ffff883ffef03668] die at ffffffff8101deb2
#5 [ffff883ffef03698] do_trap at ffffffff8101a700
#6 [ffff883ffef036e8] do_error_trap at ffffffff8101abfe
#7 [ffff883ffef037a0] do_invalid_op at ffffffff8101acd0
#8 [ffff883ffef037b0] invalid_op at ffffffff81a00bab
[exception RIP: skb_segment+3044]
RIP: ffffffff817e4dd4 RSP: ffff883ffef03860 RFLAGS: 00010216
RAX: 0000000000002bf6 RBX: ffff883feb7aaa00 RCX: 0000000000000011
RDX: ffff883fb87910c0 RSI: 0000000000000011 RDI: ffff883feb7ab500
RBP: ffff883ffef03928 R8: 0000000000002ce2 R9: 00000000000027da
R10: 000001ea00000000 R11: 0000000000002d82 R12: ffff883f90a1ee80
R13: ffff883fb8791120 R14: ffff883feb7abc00 R15: 0000000000002ce2
ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
#9 [ffff883ffef03930] tcp_gso_segment at ffffffff818713e7
--- <IRQ stack> ---
...
The triggering input skb has the following properties:
list_skb = skb->frag_list;
skb->nfrags != NULL && skb_headlen(list_skb) != 0
and skb_segment() is not able to handle a frag_list skb
if its headlen (list_skb->len - list_skb->data_len) is not 0.
This patch addressed the issue by handling skb_headlen(list_skb) != 0
case properly if list_skb->head_frag is true, which is expected in
most cases. The head frag is processed before list_skb->frags
are processed.
Reported-by: Diptanu Gon Choudhury <dipt...@fb.com>
Signed-off-by: Yonghong Song <y...@fb.com>
---
net/core/skbuff.c | 51 +++++++++++++++++++++++++++++++++++++--------------
1 file changed, 37 insertions(+), 14 deletions(-)
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 715c134..59bbc06 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3475,7 +3475,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
struct sk_buff *segs = NULL;
struct sk_buff *tail = NULL;
struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
- skb_frag_t *frag = skb_shinfo(head_skb)->frags;
+ skb_frag_t *frag = skb_shinfo(head_skb)->frags, *head_frag = NULL;
I think you misunderstood me. I wasn't saying you allocate head_frag.
I was saying you could move the declaration down.
Sorry for my misunderstanding. I did understand your intention of moving
the declaration down in order to save stack space. I thought that we
cannot really move declaration down (although it works in C, but
semantically it is not quite right, more later), so I moved on to
use runtime allocation. But indeed skb_frag_t is not big (16 bytes), it
could live on the stack.
unsigned int mss = skb_shinfo(head_skb)->gso_size;
unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
struct sk_buff *frag_skb = head_skb;
@@ -3664,19 +3664,39 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
while (pos < offset + len) {
So right here in the loop you could add a "skb_frag_t head_frag;" just
so we declare it here and save ourselves the stack space.
I actually tried to move "skb_frag_t head_frag". The stack size remains
the same, 0xc0. This is related to how C compiler allocates stack space.
The declaration place won't decide the stack size as long as the
declaration dictates the usage. The stack size is really determined by
liveness analysis.
Further, we have code like:
do {
....
while (pos < offset + len) {
if (i >= nfrags) {
...
head_frag = ...
}
... = head_frag; // head_frag access guaranteed after
// above definition, but it may not
// be in the same outer do-while loop.
}
...
} while (((offset += len) < head_skb->len);
So the use of head_frag maybe in different outer loop iterations.
So I feel the definition of head_frag should be outside the
outer do-while loop, which is the main function scope. I will add some
comments here.
if (i >= nfrags) {
- BUG_ON(skb_headlen(list_skb));
-
i = 0;
+ if (skb_headlen(list_skb)) {
+ struct page *page;
+
+ BUG_ON(!list_skb->head_frag);
+
+ page =
virt_to_head_page(list_skb->head);
+ if (!head_frag) {
+ head_frag =
kmalloc(sizeof(skb_frag_t),
+ GFP_KERNEL);
+ if (!head_frag)
+ goto err;
+ }
Please no memory allocation. I just meant you could allocate it on the
stack later.
+ head_frag->page.p = page;
+ head_frag->page_offset = list_skb->data
-
+ (unsigned char
*)page_address(page);
+ head_frag->size = skb_headlen(list_skb);
+ /* set i = -1 so we will pick head_frag
+ * instead of
skb_shinfo(list_skb)->frags
+ * when i == -1.
+ */
+ i = -1;
+ }
So it took me a bit to pick up on the fact that line below wasn't
removed. So we are basically trying to do this all in one pass now. Do
I have that right?
One thing you could look at doing to save yourself the extra "if"
later would be to pull frag pointer before you go through skb_headlen
check above. Then if you are going to use a head_frag you could just
do a "i--; frag--;" combination just to rewind and make the room for
the increment to come later. That way you don't have an invalid frag
pointer floating around. That way you only have to do this once
instead of having to do a conditional check per fragment.
Right. This indeed make code more cleaner.
nfrags = skb_shinfo(list_skb)->nr_frags;
- frag = skb_shinfo(list_skb)->frags;
This patch might be more readable if you were to just insert the
skb_headlen() bits down here and left the i=0 through frag = .. in one
piece.
Right. Will implement as suggested.
- frag_skb = list_skb;
-
- BUG_ON(!nfrags);
-
- if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
- skb_zerocopy_clone(nskb, frag_skb,
- GFP_ATOMIC))
- goto err;
+ if (nfrags) {
+ frag = skb_shinfo(list_skb)->frags;
+ frag_skb = list_skb;
+
+ if (skb_orphan_frags(frag_skb,
GFP_ATOMIC) ||
+ skb_zerocopy_clone(nskb, frag_skb,
+ GFP_ATOMIC))
+ goto err;
+ }
list_skb = list_skb->next;
}
@@ -3689,7 +3709,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
goto err;
}
- *nskb_frag = *frag;
+ *nskb_frag = (i == -1) ? *head_frag : *frag;
So this would be better as "*nskb_frag = (i < 0) ? head_frag : *frag;".
Good suggestion. Will implement as suggested.
__skb_frag_ref(nskb_frag);
size = skb_frag_size(nskb_frag);
@@ -3702,7 +3722,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
if (pos + size <= offset + len) {
i++;
- frag++;
+ if (i != 0)
+ frag++;
pos += size;
} else {
skb_frag_size_sub(nskb_frag, pos + size -
(offset + len));
@@ -3774,10 +3795,12 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
swap(tail->destructor, head_skb->destructor);
swap(tail->sk, head_skb->sk);
}
+ kfree(head_frag);
return segs;
err:
kfree_skb_list(segs);
+ kfree(head_frag);
return ERR_PTR(err);
}
EXPORT_SYMBOL_GPL(skb_segment);
--
2.9.5
