From 7382d99a39582fd9227d63f03cbe1839ac76f86e Mon Sep 17 00:00:00 2001 From: Steffen Klassert Date: Wed, 23 Oct 2024 12:53:42 +0200 Subject: [PATCH] xfrm: Add support for per cpu xfrm state handling. BugLink: https://bugs.launchpad.net/bugs/2111953 [ Upstream commit 1ddf9916ac09313128e40d6581cef889c0b4ce84 ] Currently all flows for a certain SA must be processed by the same cpu to avoid packet reordering and lock contention of the xfrm state lock. To get rid of this limitation, the IETF standardized per cpu SAs in RFC 9611. This patch implements the xfrm part of it. We add the cpu as a lookup key for xfrm states and a config option to generate acquire messages for each cpu. With that, we can have on each cpu a SA with identical traffic selector so that flows can be processed in parallel on all cpus. Signed-off-by: Steffen Klassert Tested-by: Antony Antony Tested-by: Tobias Brunner Stable-dep-of: e952837f3ddb ("xfrm: state: fix out-of-bounds read during lookup") Signed-off-by: Sasha Levin Signed-off-by: Manuel Diewald Signed-off-by: Mehmet Basaran --- include/net/xfrm.h | 5 ++-- include/uapi/linux/xfrm.h | 2 ++ net/key/af_key.c | 7 +++-- net/xfrm/xfrm_compat.c | 6 ++-- net/xfrm/xfrm_state.c | 58 +++++++++++++++++++++++++++++++-------- net/xfrm/xfrm_user.c | 52 +++++++++++++++++++++++++++++++---- 6 files changed, 107 insertions(+), 23 deletions(-) diff --git a/include/net/xfrm.h b/include/net/xfrm.h index 203353857535..2e36ec842372 100644 --- a/include/net/xfrm.h +++ b/include/net/xfrm.h @@ -186,6 +186,7 @@ struct xfrm_state { refcount_t refcnt; spinlock_t lock; + u32 pcpu_num; struct xfrm_id id; struct xfrm_selector sel; struct xfrm_mark mark; @@ -1660,7 +1661,7 @@ struct xfrmk_spdinfo { u32 spdhmcnt; }; -struct xfrm_state *xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq); +struct xfrm_state *xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq, u32 pcpu_num); int xfrm_state_delete(struct xfrm_state *x); int xfrm_state_flush(struct net *net, u8 proto, bool task_valid, bool sync); int xfrm_dev_state_flush(struct net *net, struct net_device *dev, bool task_valid); @@ -1772,7 +1773,7 @@ int verify_spi_info(u8 proto, u32 min, u32 max, struct netlink_ext_ack *extack); int xfrm_alloc_spi(struct xfrm_state *x, u32 minspi, u32 maxspi, struct netlink_ext_ack *extack); struct xfrm_state *xfrm_find_acq(struct net *net, const struct xfrm_mark *mark, - u8 mode, u32 reqid, u32 if_id, u8 proto, + u8 mode, u32 reqid, u32 if_id, u32 pcpu_num, u8 proto, const xfrm_address_t *daddr, const xfrm_address_t *saddr, int create, unsigned short family); diff --git a/include/uapi/linux/xfrm.h b/include/uapi/linux/xfrm.h index 6a77328be114..1703b6fc868e 100644 --- a/include/uapi/linux/xfrm.h +++ b/include/uapi/linux/xfrm.h @@ -315,6 +315,7 @@ enum xfrm_attr_type_t { XFRMA_SET_MARK_MASK, /* __u32 */ XFRMA_IF_ID, /* __u32 */ XFRMA_MTIMER_THRESH, /* __u32 in seconds for input SA */ + XFRMA_SA_PCPU, /* __u32 */ __XFRMA_MAX #define XFRMA_OUTPUT_MARK XFRMA_SET_MARK /* Compatibility */ @@ -430,6 +431,7 @@ struct xfrm_userpolicy_info { #define XFRM_POLICY_LOCALOK 1 /* Allow user to override global policy */ /* Automatically expand selector to include matching ICMP payloads. */ #define XFRM_POLICY_ICMP 2 +#define XFRM_POLICY_CPU_ACQUIRE 4 __u8 share; }; diff --git a/net/key/af_key.c b/net/key/af_key.c index f79fb99271ed..c56bb4f451e6 100644 --- a/net/key/af_key.c +++ b/net/key/af_key.c @@ -1354,7 +1354,7 @@ static int pfkey_getspi(struct sock *sk, struct sk_buff *skb, const struct sadb_ } if (hdr->sadb_msg_seq) { - x = xfrm_find_acq_byseq(net, DUMMY_MARK, hdr->sadb_msg_seq); + x = xfrm_find_acq_byseq(net, DUMMY_MARK, hdr->sadb_msg_seq, UINT_MAX); if (x && !xfrm_addr_equal(&x->id.daddr, xdaddr, family)) { xfrm_state_put(x); x = NULL; @@ -1362,7 +1362,8 @@ static int pfkey_getspi(struct sock *sk, struct sk_buff *skb, const struct sadb_ } if (!x) - x = xfrm_find_acq(net, &dummy_mark, mode, reqid, 0, proto, xdaddr, xsaddr, 1, family); + x = xfrm_find_acq(net, &dummy_mark, mode, reqid, 0, UINT_MAX, + proto, xdaddr, xsaddr, 1, family); if (x == NULL) return -ENOENT; @@ -1417,7 +1418,7 @@ static int pfkey_acquire(struct sock *sk, struct sk_buff *skb, const struct sadb if (hdr->sadb_msg_seq == 0 || hdr->sadb_msg_errno == 0) return 0; - x = xfrm_find_acq_byseq(net, DUMMY_MARK, hdr->sadb_msg_seq); + x = xfrm_find_acq_byseq(net, DUMMY_MARK, hdr->sadb_msg_seq, UINT_MAX); if (x == NULL) return 0; diff --git a/net/xfrm/xfrm_compat.c b/net/xfrm/xfrm_compat.c index 655fe4ff8621..c6d93eed4e57 100644 --- a/net/xfrm/xfrm_compat.c +++ b/net/xfrm/xfrm_compat.c @@ -129,6 +129,7 @@ static const struct nla_policy compat_policy[XFRMA_MAX+1] = { [XFRMA_SET_MARK_MASK] = { .type = NLA_U32 }, [XFRMA_IF_ID] = { .type = NLA_U32 }, [XFRMA_MTIMER_THRESH] = { .type = NLA_U32 }, + [XFRMA_SA_PCPU] = { .type = NLA_U32 }, }; static struct nlmsghdr *xfrm_nlmsg_put_compat(struct sk_buff *skb, @@ -277,9 +278,10 @@ static int xfrm_xlate64_attr(struct sk_buff *dst, const struct nlattr *src) case XFRMA_SET_MARK_MASK: case XFRMA_IF_ID: case XFRMA_MTIMER_THRESH: + case XFRMA_SA_PCPU: return xfrm_nla_cpy(dst, src, nla_len(src)); default: - BUILD_BUG_ON(XFRMA_MAX != XFRMA_MTIMER_THRESH); + BUILD_BUG_ON(XFRMA_MAX != XFRMA_SA_PCPU); pr_warn_once("unsupported nla_type %d\n", src->nla_type); return -EOPNOTSUPP; } @@ -434,7 +436,7 @@ static int xfrm_xlate32_attr(void *dst, const struct nlattr *nla, int err; if (type > XFRMA_MAX) { - BUILD_BUG_ON(XFRMA_MAX != XFRMA_MTIMER_THRESH); + BUILD_BUG_ON(XFRMA_MAX != XFRMA_SA_PCPU); NL_SET_ERR_MSG(extack, "Bad attribute"); return -EOPNOTSUPP; } diff --git a/net/xfrm/xfrm_state.c b/net/xfrm/xfrm_state.c index 8a6e8656d014..f84ff3582b7c 100644 --- a/net/xfrm/xfrm_state.c +++ b/net/xfrm/xfrm_state.c @@ -679,6 +679,7 @@ struct xfrm_state *xfrm_state_alloc(struct net *net) x->lft.hard_packet_limit = XFRM_INF; x->replay_maxage = 0; x->replay_maxdiff = 0; + x->pcpu_num = UINT_MAX; spin_lock_init(&x->lock); } return x; @@ -1154,6 +1155,12 @@ static void xfrm_state_look_at(struct xfrm_policy *pol, struct xfrm_state *x, struct xfrm_state **best, int *acq_in_progress, int *error) { + /* We need the cpu id just as a lookup key, + * we don't require it to be stable. + */ + unsigned int pcpu_id = get_cpu(); + put_cpu(); + /* Resolution logic: * 1. There is a valid state with matching selector. Done. * 2. Valid state with inappropriate selector. Skip. @@ -1173,13 +1180,18 @@ static void xfrm_state_look_at(struct xfrm_policy *pol, struct xfrm_state *x, &fl->u.__fl_common)) return; + if (x->pcpu_num != UINT_MAX && x->pcpu_num != pcpu_id) + return; + if (!*best || + ((*best)->pcpu_num == UINT_MAX && x->pcpu_num == pcpu_id) || (*best)->km.dying > x->km.dying || ((*best)->km.dying == x->km.dying && (*best)->curlft.add_time < x->curlft.add_time)) *best = x; } else if (x->km.state == XFRM_STATE_ACQ) { - *acq_in_progress = 1; + if (!*best || x->pcpu_num == pcpu_id) + *acq_in_progress = 1; } else if (x->km.state == XFRM_STATE_ERROR || x->km.state == XFRM_STATE_EXPIRED) { if ((!x->sel.family || @@ -1208,6 +1220,13 @@ xfrm_state_find(const xfrm_address_t *daddr, const xfrm_address_t *saddr, unsigned short encap_family = tmpl->encap_family; unsigned int sequence; struct km_event c; + unsigned int pcpu_id; + + /* We need the cpu id just as a lookup key, + * we don't require it to be stable. + */ + pcpu_id = get_cpu(); + put_cpu(); to_put = NULL; @@ -1281,7 +1300,10 @@ xfrm_state_find(const xfrm_address_t *daddr, const xfrm_address_t *saddr, } found: - x = best; + if (!(pol->flags & XFRM_POLICY_CPU_ACQUIRE) || + (best && (best->pcpu_num == pcpu_id))) + x = best; + if (!x && !error && !acquire_in_progress) { if (tmpl->id.spi && (x0 = __xfrm_state_lookup_all(net, mark, daddr, @@ -1313,6 +1335,8 @@ found: xfrm_init_tempstate(x, fl, tmpl, daddr, saddr, family); memcpy(&x->mark, &pol->mark, sizeof(x->mark)); x->if_id = if_id; + if ((pol->flags & XFRM_POLICY_CPU_ACQUIRE) && best) + x->pcpu_num = pcpu_id; error = security_xfrm_state_alloc_acquire(x, pol->security, fl->flowi_secid); if (error) { @@ -1390,6 +1414,11 @@ found: x = NULL; error = -ESRCH; } + + /* Use the already installed 'fallback' while the CPU-specific + * SA acquire is handled*/ + if (best) + x = best; } out: if (x) { @@ -1521,12 +1550,14 @@ static void __xfrm_state_bump_genids(struct xfrm_state *xnew) unsigned int h; u32 mark = xnew->mark.v & xnew->mark.m; u32 if_id = xnew->if_id; + u32 cpu_id = xnew->pcpu_num; h = xfrm_dst_hash(net, &xnew->id.daddr, &xnew->props.saddr, reqid, family); hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) { if (x->props.family == family && x->props.reqid == reqid && x->if_id == if_id && + x->pcpu_num == cpu_id && (mark & x->mark.m) == x->mark.v && xfrm_addr_equal(&x->id.daddr, &xnew->id.daddr, family) && xfrm_addr_equal(&x->props.saddr, &xnew->props.saddr, family)) @@ -1549,7 +1580,7 @@ EXPORT_SYMBOL(xfrm_state_insert); static struct xfrm_state *__find_acq_core(struct net *net, const struct xfrm_mark *m, unsigned short family, u8 mode, - u32 reqid, u32 if_id, u8 proto, + u32 reqid, u32 if_id, u32 pcpu_num, u8 proto, const xfrm_address_t *daddr, const xfrm_address_t *saddr, int create) @@ -1566,6 +1597,7 @@ static struct xfrm_state *__find_acq_core(struct net *net, x->id.spi != 0 || x->id.proto != proto || (mark & x->mark.m) != x->mark.v || + x->pcpu_num != pcpu_num || !xfrm_addr_equal(&x->id.daddr, daddr, family) || !xfrm_addr_equal(&x->props.saddr, saddr, family)) continue; @@ -1599,6 +1631,7 @@ static struct xfrm_state *__find_acq_core(struct net *net, break; } + x->pcpu_num = pcpu_num; x->km.state = XFRM_STATE_ACQ; x->id.proto = proto; x->props.family = family; @@ -1627,7 +1660,7 @@ static struct xfrm_state *__find_acq_core(struct net *net, return x; } -static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq); +static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq, u32 pcpu_num); int xfrm_state_add(struct xfrm_state *x) { @@ -1653,7 +1686,7 @@ int xfrm_state_add(struct xfrm_state *x) } if (use_spi && x->km.seq) { - x1 = __xfrm_find_acq_byseq(net, mark, x->km.seq); + x1 = __xfrm_find_acq_byseq(net, mark, x->km.seq, x->pcpu_num); if (x1 && ((x1->id.proto != x->id.proto) || !xfrm_addr_equal(&x1->id.daddr, &x->id.daddr, family))) { to_put = x1; @@ -1663,7 +1696,7 @@ int xfrm_state_add(struct xfrm_state *x) if (use_spi && !x1) x1 = __find_acq_core(net, &x->mark, family, x->props.mode, - x->props.reqid, x->if_id, x->id.proto, + x->props.reqid, x->if_id, x->pcpu_num, x->id.proto, &x->id.daddr, &x->props.saddr, 0); __xfrm_state_bump_genids(x); @@ -1788,6 +1821,7 @@ static struct xfrm_state *xfrm_state_clone(struct xfrm_state *orig, x->props.flags = orig->props.flags; x->props.extra_flags = orig->props.extra_flags; + x->pcpu_num = orig->pcpu_num; x->if_id = orig->if_id; x->tfcpad = orig->tfcpad; x->replay_maxdiff = orig->replay_maxdiff; @@ -2043,13 +2077,14 @@ EXPORT_SYMBOL(xfrm_state_lookup_byaddr); struct xfrm_state * xfrm_find_acq(struct net *net, const struct xfrm_mark *mark, u8 mode, u32 reqid, - u32 if_id, u8 proto, const xfrm_address_t *daddr, + u32 if_id, u32 pcpu_num, u8 proto, const xfrm_address_t *daddr, const xfrm_address_t *saddr, int create, unsigned short family) { struct xfrm_state *x; spin_lock_bh(&net->xfrm.xfrm_state_lock); - x = __find_acq_core(net, mark, family, mode, reqid, if_id, proto, daddr, saddr, create); + x = __find_acq_core(net, mark, family, mode, reqid, if_id, pcpu_num, + proto, daddr, saddr, create); spin_unlock_bh(&net->xfrm.xfrm_state_lock); return x; @@ -2184,7 +2219,7 @@ xfrm_state_sort(struct xfrm_state **dst, struct xfrm_state **src, int n, /* Silly enough, but I'm lazy to build resolution list */ -static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq) +static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq, u32 pcpu_num) { unsigned int h = xfrm_seq_hash(net, seq); struct xfrm_state *x; @@ -2192,6 +2227,7 @@ static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 s hlist_for_each_entry_rcu(x, net->xfrm.state_byseq + h, byseq) { if (x->km.seq == seq && (mark & x->mark.m) == x->mark.v && + x->pcpu_num == pcpu_num && x->km.state == XFRM_STATE_ACQ) { xfrm_state_hold(x); return x; @@ -2201,12 +2237,12 @@ static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 s return NULL; } -struct xfrm_state *xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq) +struct xfrm_state *xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq, u32 pcpu_num) { struct xfrm_state *x; spin_lock_bh(&net->xfrm.xfrm_state_lock); - x = __xfrm_find_acq_byseq(net, mark, seq); + x = __xfrm_find_acq_byseq(net, mark, seq, pcpu_num); spin_unlock_bh(&net->xfrm.xfrm_state_lock); return x; } diff --git a/net/xfrm/xfrm_user.c b/net/xfrm/xfrm_user.c index 0f26500773d2..d5a67d7eed3b 100644 --- a/net/xfrm/xfrm_user.c +++ b/net/xfrm/xfrm_user.c @@ -738,6 +738,12 @@ static struct xfrm_state *xfrm_state_construct(struct net *net, if (attrs[XFRMA_IF_ID]) x->if_id = nla_get_u32(attrs[XFRMA_IF_ID]); + if (attrs[XFRMA_SA_PCPU]) { + x->pcpu_num = nla_get_u32(attrs[XFRMA_SA_PCPU]); + if (x->pcpu_num >= num_possible_cpus()) + goto error; + } + err = __xfrm_init_state(x, false, attrs[XFRMA_OFFLOAD_DEV], extack); if (err) goto error; @@ -1188,8 +1194,13 @@ static int copy_to_user_state_extra(struct xfrm_state *x, if (ret) goto out; } - if (x->mapping_maxage) + if (x->mapping_maxage) { ret = nla_put_u32(skb, XFRMA_MTIMER_THRESH, x->mapping_maxage); + if (ret) + goto out; + } + if (x->pcpu_num != UINT_MAX) + ret = nla_put_u32(skb, XFRMA_SA_PCPU, x->pcpu_num); out: return ret; } @@ -1585,6 +1596,7 @@ static int xfrm_alloc_userspi(struct sk_buff *skb, struct nlmsghdr *nlh, u32 mark; struct xfrm_mark m; u32 if_id = 0; + u32 pcpu_num = UINT_MAX; p = nlmsg_data(nlh); err = verify_spi_info(p->info.id.proto, p->min, p->max, extack); @@ -1601,8 +1613,16 @@ static int xfrm_alloc_userspi(struct sk_buff *skb, struct nlmsghdr *nlh, if (attrs[XFRMA_IF_ID]) if_id = nla_get_u32(attrs[XFRMA_IF_ID]); + if (attrs[XFRMA_SA_PCPU]) { + pcpu_num = nla_get_u32(attrs[XFRMA_SA_PCPU]); + if (pcpu_num >= num_possible_cpus()) { + err = -EINVAL; + goto out_noput; + } + } + if (p->info.seq) { - x = xfrm_find_acq_byseq(net, mark, p->info.seq); + x = xfrm_find_acq_byseq(net, mark, p->info.seq, pcpu_num); if (x && !xfrm_addr_equal(&x->id.daddr, daddr, family)) { xfrm_state_put(x); x = NULL; @@ -1611,7 +1631,7 @@ static int xfrm_alloc_userspi(struct sk_buff *skb, struct nlmsghdr *nlh, if (!x) x = xfrm_find_acq(net, &m, p->info.mode, p->info.reqid, - if_id, p->info.id.proto, daddr, + if_id, pcpu_num, p->info.id.proto, daddr, &p->info.saddr, 1, family); err = -ENOENT; @@ -2407,7 +2427,8 @@ static inline unsigned int xfrm_aevent_msgsize(struct xfrm_state *x) + nla_total_size_64bit(sizeof(struct xfrm_lifetime_cur)) + nla_total_size(sizeof(struct xfrm_mark)) + nla_total_size(4) /* XFRM_AE_RTHR */ - + nla_total_size(4); /* XFRM_AE_ETHR */ + + nla_total_size(4) /* XFRM_AE_ETHR */ + + nla_total_size(4); /* XFRMA_SA_PCPU */ } static int build_aevent(struct sk_buff *skb, struct xfrm_state *x, const struct km_event *c) @@ -2463,6 +2484,8 @@ static int build_aevent(struct sk_buff *skb, struct xfrm_state *x, const struct err = xfrm_if_id_put(skb, x->if_id); if (err) goto out_cancel; + if (x->pcpu_num != UINT_MAX) + err = nla_put_u32(skb, XFRMA_SA_PCPU, x->pcpu_num); nlmsg_end(skb, nlh); return 0; @@ -2727,6 +2750,13 @@ static int xfrm_add_acquire(struct sk_buff *skb, struct nlmsghdr *nlh, xfrm_mark_get(attrs, &mark); + if (attrs[XFRMA_SA_PCPU]) { + x->pcpu_num = nla_get_u32(attrs[XFRMA_SA_PCPU]); + err = -EINVAL; + if (x->pcpu_num >= num_possible_cpus()) + goto free_state; + } + err = verify_newpolicy_info(&ua->policy, extack); if (err) goto free_state; @@ -3054,6 +3084,7 @@ const struct nla_policy xfrma_policy[XFRMA_MAX+1] = { [XFRMA_SET_MARK_MASK] = { .type = NLA_U32 }, [XFRMA_IF_ID] = { .type = NLA_U32 }, [XFRMA_MTIMER_THRESH] = { .type = NLA_U32 }, + [XFRMA_SA_PCPU] = { .type = NLA_U32 }, }; EXPORT_SYMBOL_GPL(xfrma_policy); @@ -3195,7 +3226,8 @@ static void xfrm_netlink_rcv(struct sk_buff *skb) static inline unsigned int xfrm_expire_msgsize(void) { return NLMSG_ALIGN(sizeof(struct xfrm_user_expire)) - + nla_total_size(sizeof(struct xfrm_mark)); + + nla_total_size(sizeof(struct xfrm_mark)) + + nla_total_size(4); /* XFRMA_SA_PCPU */ } static int build_expire(struct sk_buff *skb, struct xfrm_state *x, const struct km_event *c) @@ -3221,6 +3253,11 @@ static int build_expire(struct sk_buff *skb, struct xfrm_state *x, const struct err = xfrm_if_id_put(skb, x->if_id); if (err) return err; + if (x->pcpu_num != UINT_MAX) { + err = nla_put_u32(skb, XFRMA_SA_PCPU, x->pcpu_num); + if (err) + return err; + } nlmsg_end(skb, nlh); return 0; @@ -3322,6 +3359,8 @@ static inline unsigned int xfrm_sa_len(struct xfrm_state *x) } if (x->if_id) l += nla_total_size(sizeof(x->if_id)); + if (x->pcpu_num) + l += nla_total_size(sizeof(x->pcpu_num)); /* Must count x->lastused as it may become non-zero behind our back. */ l += nla_total_size_64bit(sizeof(u64)); @@ -3422,6 +3461,7 @@ static inline unsigned int xfrm_acquire_msgsize(struct xfrm_state *x, + nla_total_size(sizeof(struct xfrm_user_tmpl) * xp->xfrm_nr) + nla_total_size(sizeof(struct xfrm_mark)) + nla_total_size(xfrm_user_sec_ctx_size(x->security)) + + nla_total_size(4) /* XFRMA_SA_PCPU */ + userpolicy_type_attrsize(); } @@ -3458,6 +3498,8 @@ static int build_acquire(struct sk_buff *skb, struct xfrm_state *x, err = xfrm_if_id_put(skb, xp->if_id); if (!err && xp->xdo.dev) err = copy_user_offload(&xp->xdo, skb); + if (!err && x->pcpu_num != UINT_MAX) + err = nla_put_u32(skb, XFRMA_SA_PCPU, x->pcpu_num); if (err) { nlmsg_cancel(skb, nlh); return err;