]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/commitdiff
4.14-stable patches
authorGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 6 Nov 2019 17:34:42 +0000 (18:34 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 6 Nov 2019 17:34:42 +0000 (18:34 +0100)
added patches:
net-flow_dissector-switch-to-siphash.patch

queue-4.14/net-flow_dissector-switch-to-siphash.patch [new file with mode: 0644]
queue-4.14/series

diff --git a/queue-4.14/net-flow_dissector-switch-to-siphash.patch b/queue-4.14/net-flow_dissector-switch-to-siphash.patch
new file mode 100644 (file)
index 0000000..43f0746
--- /dev/null
@@ -0,0 +1,389 @@
+From 55667441c84fa5e0911a0aac44fb059c15ba6da2 Mon Sep 17 00:00:00 2001
+From: Eric Dumazet <edumazet@google.com>
+Date: Tue, 22 Oct 2019 07:57:46 -0700
+Subject: net/flow_dissector: switch to siphash
+
+From: Eric Dumazet <edumazet@google.com>
+
+commit 55667441c84fa5e0911a0aac44fb059c15ba6da2 upstream.
+
+UDP IPv6 packets auto flowlabels are using a 32bit secret
+(static u32 hashrnd in net/core/flow_dissector.c) and
+apply jhash() over fields known by the receivers.
+
+Attackers can easily infer the 32bit secret and use this information
+to identify a device and/or user, since this 32bit secret is only
+set at boot time.
+
+Really, using jhash() to generate cookies sent on the wire
+is a serious security concern.
+
+Trying to change the rol32(hash, 16) in ip6_make_flowlabel() would be
+a dead end. Trying to periodically change the secret (like in sch_sfq.c)
+could change paths taken in the network for long lived flows.
+
+Let's switch to siphash, as we did in commit df453700e8d8
+("inet: switch IP ID generator to siphash")
+
+Using a cryptographically strong pseudo random function will solve this
+privacy issue and more generally remove other weak points in the stack.
+
+Packet schedulers using skb_get_hash_perturb() benefit from this change.
+
+Fixes: b56774163f99 ("ipv6: Enable auto flow labels by default")
+Fixes: 42240901f7c4 ("ipv6: Implement different admin modes for automatic flow labels")
+Fixes: 67800f9b1f4e ("ipv6: Call skb_get_hash_flowi6 to get skb->hash in ip6_make_flowlabel")
+Fixes: cb1ce2ef387b ("ipv6: Implement automatic flow label generation on transmit")
+Signed-off-by: Eric Dumazet <edumazet@google.com>
+Reported-by: Jonathan Berger <jonathann1@walla.com>
+Reported-by: Amit Klein <aksecurity@gmail.com>
+Reported-by: Benny Pinkas <benny@pinkas.net>
+Cc: Tom Herbert <tom@herbertland.com>
+Signed-off-by: David S. Miller <davem@davemloft.net>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+
+---
+ include/linux/skbuff.h       |    3 +-
+ include/net/flow_dissector.h |    3 +-
+ include/net/fq.h             |    2 -
+ include/net/fq_impl.h        |    4 +--
+ net/core/flow_dissector.c    |   48 +++++++++++++++++--------------------------
+ net/sched/sch_hhf.c          |    8 +++----
+ net/sched/sch_sfb.c          |   13 ++++++-----
+ net/sched/sch_sfq.c          |   14 +++++++-----
+ 8 files changed, 46 insertions(+), 49 deletions(-)
+
+--- a/include/linux/skbuff.h
++++ b/include/linux/skbuff.h
+@@ -1228,7 +1228,8 @@ static inline __u32 skb_get_hash_flowi6(
+       return skb->hash;
+ }
+-__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
++__u32 skb_get_hash_perturb(const struct sk_buff *skb,
++                         const siphash_key_t *perturb);
+ static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
+ {
+--- a/include/net/flow_dissector.h
++++ b/include/net/flow_dissector.h
+@@ -4,6 +4,7 @@
+ #include <linux/types.h>
+ #include <linux/in6.h>
++#include <linux/siphash.h>
+ #include <uapi/linux/if_ether.h>
+ /**
+@@ -229,7 +230,7 @@ struct flow_dissector {
+ struct flow_keys {
+       struct flow_dissector_key_control control;
+ #define FLOW_KEYS_HASH_START_FIELD basic
+-      struct flow_dissector_key_basic basic;
++      struct flow_dissector_key_basic basic __aligned(SIPHASH_ALIGNMENT);
+       struct flow_dissector_key_tags tags;
+       struct flow_dissector_key_vlan vlan;
+       struct flow_dissector_key_keyid keyid;
+--- a/include/net/fq.h
++++ b/include/net/fq.h
+@@ -70,7 +70,7 @@ struct fq {
+       struct list_head backlogs;
+       spinlock_t lock;
+       u32 flows_cnt;
+-      u32 perturbation;
++      siphash_key_t   perturbation;
+       u32 limit;
+       u32 memory_limit;
+       u32 memory_usage;
+--- a/include/net/fq_impl.h
++++ b/include/net/fq_impl.h
+@@ -105,7 +105,7 @@ static struct fq_flow *fq_flow_classify(
+       lockdep_assert_held(&fq->lock);
+-      hash = skb_get_hash_perturb(skb, fq->perturbation);
++      hash = skb_get_hash_perturb(skb, &fq->perturbation);
+       idx = reciprocal_scale(hash, fq->flows_cnt);
+       flow = &fq->flows[idx];
+@@ -255,7 +255,7 @@ static int fq_init(struct fq *fq, int fl
+       INIT_LIST_HEAD(&fq->backlogs);
+       spin_lock_init(&fq->lock);
+       fq->flows_cnt = max_t(u32, flows_cnt, 1);
+-      fq->perturbation = prandom_u32();
++      get_random_bytes(&fq->perturbation, sizeof(fq->perturbation));
+       fq->quantum = 300;
+       fq->limit = 8192;
+       fq->memory_limit = 16 << 20; /* 16 MBytes */
+--- a/net/core/flow_dissector.c
++++ b/net/core/flow_dissector.c
+@@ -889,45 +889,34 @@ out_bad:
+ }
+ EXPORT_SYMBOL(__skb_flow_dissect);
+-static u32 hashrnd __read_mostly;
++static siphash_key_t hashrnd __read_mostly;
+ static __always_inline void __flow_hash_secret_init(void)
+ {
+       net_get_random_once(&hashrnd, sizeof(hashrnd));
+ }
+-static __always_inline u32 __flow_hash_words(const u32 *words, u32 length,
+-                                           u32 keyval)
++static const void *flow_keys_hash_start(const struct flow_keys *flow)
+ {
+-      return jhash2(words, length, keyval);
+-}
+-
+-static inline const u32 *flow_keys_hash_start(const struct flow_keys *flow)
+-{
+-      const void *p = flow;
+-
+-      BUILD_BUG_ON(FLOW_KEYS_HASH_OFFSET % sizeof(u32));
+-      return (const u32 *)(p + FLOW_KEYS_HASH_OFFSET);
++      BUILD_BUG_ON(FLOW_KEYS_HASH_OFFSET % SIPHASH_ALIGNMENT);
++      return &flow->FLOW_KEYS_HASH_START_FIELD;
+ }
+ static inline size_t flow_keys_hash_length(const struct flow_keys *flow)
+ {
+-      size_t diff = FLOW_KEYS_HASH_OFFSET + sizeof(flow->addrs);
+-      BUILD_BUG_ON((sizeof(*flow) - FLOW_KEYS_HASH_OFFSET) % sizeof(u32));
+-      BUILD_BUG_ON(offsetof(typeof(*flow), addrs) !=
+-                   sizeof(*flow) - sizeof(flow->addrs));
++      size_t len = offsetof(typeof(*flow), addrs) - FLOW_KEYS_HASH_OFFSET;
+       switch (flow->control.addr_type) {
+       case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
+-              diff -= sizeof(flow->addrs.v4addrs);
++              len += sizeof(flow->addrs.v4addrs);
+               break;
+       case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
+-              diff -= sizeof(flow->addrs.v6addrs);
++              len += sizeof(flow->addrs.v6addrs);
+               break;
+       case FLOW_DISSECTOR_KEY_TIPC_ADDRS:
+-              diff -= sizeof(flow->addrs.tipcaddrs);
++              len += sizeof(flow->addrs.tipcaddrs);
+               break;
+       }
+-      return (sizeof(*flow) - diff) / sizeof(u32);
++      return len;
+ }
+ __be32 flow_get_u32_src(const struct flow_keys *flow)
+@@ -993,14 +982,15 @@ static inline void __flow_hash_consisten
+       }
+ }
+-static inline u32 __flow_hash_from_keys(struct flow_keys *keys, u32 keyval)
++static inline u32 __flow_hash_from_keys(struct flow_keys *keys,
++                                      const siphash_key_t *keyval)
+ {
+       u32 hash;
+       __flow_hash_consistentify(keys);
+-      hash = __flow_hash_words(flow_keys_hash_start(keys),
+-                               flow_keys_hash_length(keys), keyval);
++      hash = siphash(flow_keys_hash_start(keys),
++                     flow_keys_hash_length(keys), keyval);
+       if (!hash)
+               hash = 1;
+@@ -1010,12 +1000,13 @@ static inline u32 __flow_hash_from_keys(
+ u32 flow_hash_from_keys(struct flow_keys *keys)
+ {
+       __flow_hash_secret_init();
+-      return __flow_hash_from_keys(keys, hashrnd);
++      return __flow_hash_from_keys(keys, &hashrnd);
+ }
+ EXPORT_SYMBOL(flow_hash_from_keys);
+ static inline u32 ___skb_get_hash(const struct sk_buff *skb,
+-                                struct flow_keys *keys, u32 keyval)
++                                struct flow_keys *keys,
++                                const siphash_key_t *keyval)
+ {
+       skb_flow_dissect_flow_keys(skb, keys,
+                                  FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL);
+@@ -1063,7 +1054,7 @@ u32 __skb_get_hash_symmetric(const struc
+                          NULL, 0, 0, 0,
+                          FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL);
+-      return __flow_hash_from_keys(&keys, hashrnd);
++      return __flow_hash_from_keys(&keys, &hashrnd);
+ }
+ EXPORT_SYMBOL_GPL(__skb_get_hash_symmetric);
+@@ -1083,13 +1074,14 @@ void __skb_get_hash(struct sk_buff *skb)
+       __flow_hash_secret_init();
+-      hash = ___skb_get_hash(skb, &keys, hashrnd);
++      hash = ___skb_get_hash(skb, &keys, &hashrnd);
+       __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
+ }
+ EXPORT_SYMBOL(__skb_get_hash);
+-__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb)
++__u32 skb_get_hash_perturb(const struct sk_buff *skb,
++                         const siphash_key_t *perturb)
+ {
+       struct flow_keys keys;
+--- a/net/sched/sch_hhf.c
++++ b/net/sched/sch_hhf.c
+@@ -4,11 +4,11 @@
+  * Copyright (C) 2013 Nandita Dukkipati <nanditad@google.com>
+  */
+-#include <linux/jhash.h>
+ #include <linux/jiffies.h>
+ #include <linux/module.h>
+ #include <linux/skbuff.h>
+ #include <linux/vmalloc.h>
++#include <linux/siphash.h>
+ #include <net/pkt_sched.h>
+ #include <net/sock.h>
+@@ -125,7 +125,7 @@ struct wdrr_bucket {
+ struct hhf_sched_data {
+       struct wdrr_bucket buckets[WDRR_BUCKET_CNT];
+-      u32                perturbation;   /* hash perturbation */
++      siphash_key_t      perturbation;   /* hash perturbation */
+       u32                quantum;        /* psched_mtu(qdisc_dev(sch)); */
+       u32                drop_overlimit; /* number of times max qdisc packet
+                                           * limit was hit
+@@ -263,7 +263,7 @@ static enum wdrr_bucket_idx hhf_classify
+       }
+       /* Get hashed flow-id of the skb. */
+-      hash = skb_get_hash_perturb(skb, q->perturbation);
++      hash = skb_get_hash_perturb(skb, &q->perturbation);
+       /* Check if this packet belongs to an already established HH flow. */
+       flow_pos = hash & HHF_BIT_MASK;
+@@ -578,7 +578,7 @@ static int hhf_init(struct Qdisc *sch, s
+       sch->limit = 1000;
+       q->quantum = psched_mtu(qdisc_dev(sch));
+-      q->perturbation = prandom_u32();
++      get_random_bytes(&q->perturbation, sizeof(q->perturbation));
+       INIT_LIST_HEAD(&q->new_buckets);
+       INIT_LIST_HEAD(&q->old_buckets);
+--- a/net/sched/sch_sfb.c
++++ b/net/sched/sch_sfb.c
+@@ -22,7 +22,7 @@
+ #include <linux/errno.h>
+ #include <linux/skbuff.h>
+ #include <linux/random.h>
+-#include <linux/jhash.h>
++#include <linux/siphash.h>
+ #include <net/ip.h>
+ #include <net/pkt_sched.h>
+ #include <net/pkt_cls.h>
+@@ -49,7 +49,7 @@ struct sfb_bucket {
+  * (Section 4.4 of SFB reference : moving hash functions)
+  */
+ struct sfb_bins {
+-      u32               perturbation; /* jhash perturbation */
++      siphash_key_t     perturbation; /* siphash key */
+       struct sfb_bucket bins[SFB_LEVELS][SFB_NUMBUCKETS];
+ };
+@@ -221,7 +221,8 @@ static u32 sfb_compute_qlen(u32 *prob_r,
+ static void sfb_init_perturbation(u32 slot, struct sfb_sched_data *q)
+ {
+-      q->bins[slot].perturbation = prandom_u32();
++      get_random_bytes(&q->bins[slot].perturbation,
++                       sizeof(q->bins[slot].perturbation));
+ }
+ static void sfb_swap_slot(struct sfb_sched_data *q)
+@@ -317,9 +318,9 @@ static int sfb_enqueue(struct sk_buff *s
+               /* If using external classifiers, get result and record it. */
+               if (!sfb_classify(skb, fl, &ret, &salt))
+                       goto other_drop;
+-              sfbhash = jhash_1word(salt, q->bins[slot].perturbation);
++              sfbhash = siphash_1u32(salt, &q->bins[slot].perturbation);
+       } else {
+-              sfbhash = skb_get_hash_perturb(skb, q->bins[slot].perturbation);
++              sfbhash = skb_get_hash_perturb(skb, &q->bins[slot].perturbation);
+       }
+@@ -355,7 +356,7 @@ static int sfb_enqueue(struct sk_buff *s
+               /* Inelastic flow */
+               if (q->double_buffering) {
+                       sfbhash = skb_get_hash_perturb(skb,
+-                          q->bins[slot].perturbation);
++                          &q->bins[slot].perturbation);
+                       if (!sfbhash)
+                               sfbhash = 1;
+                       sfb_skb_cb(skb)->hashes[slot] = sfbhash;
+--- a/net/sched/sch_sfq.c
++++ b/net/sched/sch_sfq.c
+@@ -18,7 +18,7 @@
+ #include <linux/errno.h>
+ #include <linux/init.h>
+ #include <linux/skbuff.h>
+-#include <linux/jhash.h>
++#include <linux/siphash.h>
+ #include <linux/slab.h>
+ #include <linux/vmalloc.h>
+ #include <net/netlink.h>
+@@ -121,7 +121,7 @@ struct sfq_sched_data {
+       u8              headdrop;
+       u8              maxdepth;       /* limit of packets per flow */
+-      u32             perturbation;
++      siphash_key_t   perturbation;
+       u8              cur_depth;      /* depth of longest slot */
+       u8              flags;
+       unsigned short  scaled_quantum; /* SFQ_ALLOT_SIZE(quantum) */
+@@ -160,7 +160,7 @@ static inline struct sfq_head *sfq_dep_h
+ static unsigned int sfq_hash(const struct sfq_sched_data *q,
+                            const struct sk_buff *skb)
+ {
+-      return skb_get_hash_perturb(skb, q->perturbation) & (q->divisor - 1);
++      return skb_get_hash_perturb(skb, &q->perturbation) & (q->divisor - 1);
+ }
+ static unsigned int sfq_classify(struct sk_buff *skb, struct Qdisc *sch,
+@@ -609,9 +609,11 @@ static void sfq_perturbation(unsigned lo
+       struct Qdisc *sch = (struct Qdisc *)arg;
+       struct sfq_sched_data *q = qdisc_priv(sch);
+       spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
++      siphash_key_t nkey;
++      get_random_bytes(&nkey, sizeof(nkey));
+       spin_lock(root_lock);
+-      q->perturbation = prandom_u32();
++      q->perturbation = nkey;
+       if (!q->filter_list && q->tail)
+               sfq_rehash(sch);
+       spin_unlock(root_lock);
+@@ -690,7 +692,7 @@ static int sfq_change(struct Qdisc *sch,
+       del_timer(&q->perturb_timer);
+       if (q->perturb_period) {
+               mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
+-              q->perturbation = prandom_u32();
++              get_random_bytes(&q->perturbation, sizeof(q->perturbation));
+       }
+       sch_tree_unlock(sch);
+       kfree(p);
+@@ -746,7 +748,7 @@ static int sfq_init(struct Qdisc *sch, s
+       q->quantum = psched_mtu(qdisc_dev(sch));
+       q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum);
+       q->perturb_period = 0;
+-      q->perturbation = prandom_u32();
++      get_random_bytes(&q->perturbation, sizeof(q->perturbation));
+       if (opt) {
+               int err = sfq_change(sch, opt);
index 86d541f23972881416ba51ba5842d1cac6a220af..fac684cd87cbd849be0151b82cdcc9eab2f96e10 100644 (file)
@@ -46,5 +46,6 @@ net-use-skb_queue_empty_lockless-in-busy-poll-contexts.patch
 vxlan-check-tun_info-options_len-properly.patch
 erspan-fix-the-tun_info-options_len-check-for-erspan.patch
 inet-stop-leaking-jiffies-on-the-wire.patch
+net-flow_dissector-switch-to-siphash.patch
 kbuild-use-fmacro-prefix-map-to-make-__file__-a-rela.patch
 kbuild-add-fcf-protection-none-when-using-retpoline-.patch