]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/commit
igb: avoid premature Rx buffer reuse
authorLi RongQing <lirongqing@baidu.com>
Thu, 21 Jan 2021 21:54:23 +0000 (13:54 -0800)
committerSeth Forshee <seth.forshee@canonical.com>
Thu, 8 Apr 2021 20:42:39 +0000 (15:42 -0500)
commiteb5e67d5f6bd4afdb095011e6ebd9ce2a6f06000
tree3b99245707b973e58e6e2aecd1c3cf028f6914f4
parente4a202654b1363df7723d0770a08dba75b212f66
igb: avoid premature Rx buffer reuse

BugLink: https://bugs.launchpad.net/bugs/1922601
[ Upstream commit 98dfb02aa22280bd8833836d1b00ab0488fa951f ]

Igb needs a similar fix as commit 75aab4e10ae6a ("i40e: avoid
premature Rx buffer reuse")

The page recycle code, incorrectly, relied on that a page fragment
could not be freed inside xdp_do_redirect(). This assumption leads to
that page fragments that are used by the stack/XDP redirect can be
reused and overwritten.

To avoid this, store the page count prior invoking xdp_do_redirect().

Longer explanation:

Intel NICs have a recycle mechanism. The main idea is that a page is
split into two parts. One part is owned by the driver, one part might
be owned by someone else, such as the stack.

t0: Page is allocated, and put on the Rx ring
              +---------------
used by NIC ->| upper buffer
(rx_buffer)   +---------------
              | lower buffer
              +---------------
  page count  == USHRT_MAX
  rx_buffer->pagecnt_bias == USHRT_MAX

t1: Buffer is received, and passed to the stack (e.g.)
              +---------------
              | upper buff (skb)
              +---------------
used by NIC ->| lower buffer
(rx_buffer)   +---------------
  page count  == USHRT_MAX
  rx_buffer->pagecnt_bias == USHRT_MAX - 1

t2: Buffer is received, and redirected
              +---------------
              | upper buff (skb)
              +---------------
used by NIC ->| lower buffer
(rx_buffer)   +---------------

Now, prior calling xdp_do_redirect():
  page count  == USHRT_MAX
  rx_buffer->pagecnt_bias == USHRT_MAX - 2

This means that buffer *cannot* be flipped/reused, because the skb is
still using it.

The problem arises when xdp_do_redirect() actually frees the
segment. Then we get:
  page count  == USHRT_MAX - 1
  rx_buffer->pagecnt_bias == USHRT_MAX - 2

From a recycle perspective, the buffer can be flipped and reused,
which means that the skb data area is passed to the Rx HW ring!

To work around this, the page count is stored prior calling
xdp_do_redirect().

Fixes: 9cbc948b5a20 ("igb: add XDP support")
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Tested-by: Vishakha Jambekar <vishakha.jambekar@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
drivers/net/ethernet/intel/igb/igb_main.c