The padding length calculation in calculate_padding() is wrong for the special case of 'p_mtu = 0'. It's calculating the padded length instead of the length to pad. So it should be: if (__predict_false(!pkt->p_mtu)) return ((last_unit + (WG_PKT_PADDING - 1)) & ~(WG_PKT_PADDING - 1)) - last_unit; However, it's impossible for 'p_mtu' to be zero. So I'd propose the following fix: @@ -1461,8 +1461,7 @@ calculate_padding(struct wg_packet *pkt) { unsigned int padded_size, last_unit = pkt->p_mbuf->m_pkthdr.len; - if (__predict_false(!pkt->p_mtu)) - return (last_unit + (WG_PKT_PADDING - 1)) & ~(WG_PKT_PADDING - 1); + KASSERT(pkt->p_mtu != 0, ("%s: packet %p has p_mtu = 0", __func__, m)); if (__predict_false(last_unit > pkt->p_mtu)) last_unit %= pkt->p_mtu;
Well, after testing, I found 'pkt->p_mtu' is actually 0 for keepalive packets; however, those packets also have a zero length, so calculate_padding() didn't cause any real problems. I suggest the following new patch: --- if_wg.c.orig 2023-10-12 09:06:16.983637264 +0800 +++ if_wg.c 2024-01-16 22:25:04.878629478 +0800 @@ -1461,8 +1461,11 @@ calculate_padding(struct wg_packet *pkt) { unsigned int padded_size, last_unit = pkt->p_mbuf->m_pkthdr.len; - if (__predict_false(!pkt->p_mtu)) - return (last_unit + (WG_PKT_PADDING - 1)) & ~(WG_PKT_PADDING - 1); + /* Keepalive packets don't set p_mtu, but also have a zreo length. */ + if (__predict_false(pkt->p_mtu == 0)) { + padded_size = (last_unit + (WG_PKT_PADDING - 1)) & ~(WG_PKT_PADDING - 1); + return padded_size - last_unit; + } if (__predict_false(last_unit > pkt->p_mtu)) last_unit %= pkt->p_mtu;
(In reply to Aaron LI from comment #1) I would perhaps reorganize it slightly at that point: static inline unsigned int calculate_padding(struct wg_packet *pkt) { unsigned int padded_size, last_unit = pkt->p_mbuf->m_pkthdr.len; padded_size = (last_unit + (WG_PKT_PADDING - 1)) & ~(WG_PKT_PADDING - 1); if (__predict_true(pkt->p_mtu != 0)) { if (__predict_false(last_unit > pkt->p_mtu)) last_unit %= pkt->p_mtu; if (pkt->p_mtu < padded_size) padded_size = pkt->p_mtu; } return (padded_size - last_unit); }
(In reply to Kyle Evans from comment #2) WHoops, sorry, of course we need to re-calculate padded_size.
A commit in branch main references this bug: URL: https://cgit.FreeBSD.org/src/commit/?id=b891f61ef538a4e9b4658b4b756635c8036a5788 commit b891f61ef538a4e9b4658b4b756635c8036a5788 Author: Aaron LI <aly@aaronly.me> AuthorDate: 2024-01-17 23:29:23 +0000 Commit: Kyle Evans <kevans@FreeBSD.org> CommitDate: 2024-01-17 23:29:52 +0000 if_wg: fix erroneous calculation in calculate_padding() for p_mtu == 0 In practice this is harmless; only keepalive packets may realistically have p_mtu == 0, and they'll also have no payload so the math works out the same either way. Still, let's prefer technical accuracy and calculate the amount of padding needed rather than the padded length... PR: 276363 sys/dev/wg/if_wg.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
A commit in branch stable/13 references this bug: URL: https://cgit.FreeBSD.org/src/commit/?id=bbda52e814e0c760b2beaeae40a2c76ff43d1975 commit bbda52e814e0c760b2beaeae40a2c76ff43d1975 Author: Aaron LI <aly@aaronly.me> AuthorDate: 2024-01-17 23:29:23 +0000 Commit: Kyle Evans <kevans@FreeBSD.org> CommitDate: 2024-01-30 05:37:46 +0000 if_wg: fix erroneous calculation in calculate_padding() for p_mtu == 0 In practice this is harmless; only keepalive packets may realistically have p_mtu == 0, and they'll also have no payload so the math works out the same either way. Still, let's prefer technical accuracy and calculate the amount of padding needed rather than the padded length... PR: 276363 (cherry picked from commit b891f61ef538a4e9b4658b4b756635c8036a5788) sys/dev/wg/if_wg.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
A commit in branch stable/14 references this bug: URL: https://cgit.FreeBSD.org/src/commit/?id=86986d381072171a5d6183701ae2f96c2d9a6406 commit 86986d381072171a5d6183701ae2f96c2d9a6406 Author: Aaron LI <aly@aaronly.me> AuthorDate: 2024-01-17 23:29:23 +0000 Commit: Kyle Evans <kevans@FreeBSD.org> CommitDate: 2024-01-30 05:37:29 +0000 if_wg: fix erroneous calculation in calculate_padding() for p_mtu == 0 In practice this is harmless; only keepalive packets may realistically have p_mtu == 0, and they'll also have no payload so the math works out the same either way. Still, let's prefer technical accuracy and calculate the amount of padding needed rather than the padded length... PR: 276363 (cherry picked from commit b891f61ef538a4e9b4658b4b756635c8036a5788) sys/dev/wg/if_wg.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
MFC'd to all supported branches; thanks!