projects / thirdparty / kernel / stable-queue.git / blame
| Commit | Line | Data |
|---|---|---|
| d60d0101 SL |
1 | From 7a0847107ec1c0477b91c8e67b50f5b0f982c6d8 Mon Sep 17 00:00:00 2001 |
| 2 | From: Alexei Starovoitov <ast@fb.com> | |
| 3 | Date: Thu, 9 May 2019 19:33:53 -0700 | |
| 4 | Subject: bpf: fix struct htab_elem layout | |
| 5 | ||
| 6 | commit 9f691549f76d488a0c74397b3e51e943865ea01f upstream. | |
| 7 | ||
| 8 | when htab_elem is removed from the bucket list the htab_elem.hash_node.next | |
| 9 | field should not be overridden too early otherwise we have a tiny race window | |
| 10 | between lookup and delete. | |
| 11 | The bug was discovered by manual code analysis and reproducible | |
| 12 | only with explicit udelay() in lookup_elem_raw(). | |
| 13 | ||
| 14 | Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") | |
| 15 | Reported-by: Jonathan Perry <jonperry@fb.com> | |
| 16 | Signed-off-by: Alexei Starovoitov <ast@kernel.org> | |
| 17 | Acked-by: Daniel Borkmann <daniel@iogearbox.net> | |
| 18 | Signed-off-by: David S. Miller <davem@davemloft.net> | |
| 19 | Signed-off-by: Chenbo Feng <fengc@google.com> | |
| 20 | Signed-off-by: Sasha Levin <sashal@kernel.org> | |
| 21 | --- | |
| 22 | kernel/bpf/hashtab.c | 28 ++++++++++++++++++++++------ | |
| 23 | 1 file changed, 22 insertions(+), 6 deletions(-) | |
| 24 | ||
| 25 | diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c | |
| 26 | index a36a532c056df..f9d53ac57f640 100644 | |
| 27 | --- a/kernel/bpf/hashtab.c | |
| 28 | +++ b/kernel/bpf/hashtab.c | |
| 29 | @@ -41,8 +41,13 @@ enum extra_elem_state { | |
| 30 | struct htab_elem { | |
| 31 | union { | |
| 32 | struct hlist_node hash_node; | |
| 33 | - struct bpf_htab *htab; | |
| 34 | - struct pcpu_freelist_node fnode; | |
| 35 | + struct { | |
| 36 | + void *padding; | |
| 37 | + union { | |
| 38 | + struct bpf_htab *htab; | |
| 39 | + struct pcpu_freelist_node fnode; | |
| 40 | + }; | |
| 41 | + }; | |
| 42 | }; | |
| 43 | union { | |
| 44 | struct rcu_head rcu; | |
| 45 | @@ -114,8 +119,10 @@ static int prealloc_elems_and_freelist(struct bpf_htab *htab) | |
| 46 | if (err) | |
| 47 | goto free_elems; | |
| 48 | ||
| 49 | - pcpu_freelist_populate(&htab->freelist, htab->elems, htab->elem_size, | |
| 50 | - htab->map.max_entries); | |
| 51 | + pcpu_freelist_populate(&htab->freelist, | |
| 52 | + htab->elems + offsetof(struct htab_elem, fnode), | |
| 53 | + htab->elem_size, htab->map.max_entries); | |
| 54 | + | |
| 55 | return 0; | |
| 56 | ||
| 57 | free_elems: | |
| 58 | @@ -148,6 +155,11 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr) | |
| 59 | int err, i; | |
| 60 | u64 cost; | |
| 61 | ||
| 62 | + BUILD_BUG_ON(offsetof(struct htab_elem, htab) != | |
| 63 | + offsetof(struct htab_elem, hash_node.pprev)); | |
| 64 | + BUILD_BUG_ON(offsetof(struct htab_elem, fnode.next) != | |
| 65 | + offsetof(struct htab_elem, hash_node.pprev)); | |
| 66 | + | |
| 67 | if (attr->map_flags & ~BPF_F_NO_PREALLOC) | |
| 68 | /* reserved bits should not be used */ | |
| 69 | return ERR_PTR(-EINVAL); | |
| 70 | @@ -429,9 +441,13 @@ static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key, | |
| 71 | int err = 0; | |
| 72 | ||
| 73 | if (prealloc) { | |
| 74 | - l_new = (struct htab_elem *)pcpu_freelist_pop(&htab->freelist); | |
| 75 | - if (!l_new) | |
| 76 | + struct pcpu_freelist_node *l; | |
| 77 | + | |
| 78 | + l = pcpu_freelist_pop(&htab->freelist); | |
| 79 | + if (!l) | |
| 80 | err = -E2BIG; | |
| 81 | + else | |
| 82 | + l_new = container_of(l, struct htab_elem, fnode); | |
| 83 | } else { | |
| 84 | if (atomic_inc_return(&htab->count) > htab->map.max_entries) { | |
| 85 | atomic_dec(&htab->count); | |
| 86 | -- | |
| 87 | 2.20.1 | |
| 88 |