sock: ulimit on MSG_ZEROCOPY pages
authorWillem de Bruijn <willemb@google.com>
Thu, 3 Aug 2017 20:29:43 +0000 (16:29 -0400)
committerDavid S. Miller <davem@davemloft.net>
Fri, 4 Aug 2017 04:37:30 +0000 (21:37 -0700)
Bound the number of pages that a user may pin.

Follow the lead of perf tools to maintain a per-user bound on memory
locked pages commit 789f90fcf6b0 ("perf_counter: per user mlock gift")

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/sched/user.h
include/linux/skbuff.h
net/core/skbuff.c

index 5d5415e129d43649f9af0840b3d3f5a2687605fa..3c07e4135127915b75f916fa12e99f077889a141 100644 (file)
@@ -36,7 +36,8 @@ struct user_struct {
        struct hlist_node uidhash_node;
        kuid_t uid;
 
-#if defined(CONFIG_PERF_EVENTS) || defined(CONFIG_BPF_SYSCALL)
+#if defined(CONFIG_PERF_EVENTS) || defined(CONFIG_BPF_SYSCALL) || \
+    defined(CONFIG_NET)
        atomic_long_t locked_vm;
 #endif
 };
index f5bdd93a87daab58297cb45d5b0c06b73f6aae87..8c0708d2e5e6655c648a5718000fac4103c7f321 100644 (file)
@@ -457,6 +457,11 @@ struct ubuf_info {
                };
        };
        atomic_t refcnt;
+
+       struct mmpin {
+               struct user_struct *user;
+               unsigned int num_pg;
+       } mmp;
 };
 
 #define skb_uarg(SKB)  ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
index dcee0f64f1faf0d4b5bf26bb77f1414caeb3cd91..42b62c716a33653267a7ad377c46cb1d251d7c69 100644 (file)
@@ -897,6 +897,44 @@ struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
 }
 EXPORT_SYMBOL_GPL(skb_morph);
 
+static int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
+{
+       unsigned long max_pg, num_pg, new_pg, old_pg;
+       struct user_struct *user;
+
+       if (capable(CAP_IPC_LOCK) || !size)
+               return 0;
+
+       num_pg = (size >> PAGE_SHIFT) + 2;      /* worst case */
+       max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+       user = mmp->user ? : current_user();
+
+       do {
+               old_pg = atomic_long_read(&user->locked_vm);
+               new_pg = old_pg + num_pg;
+               if (new_pg > max_pg)
+                       return -ENOBUFS;
+       } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
+                old_pg);
+
+       if (!mmp->user) {
+               mmp->user = get_uid(user);
+               mmp->num_pg = num_pg;
+       } else {
+               mmp->num_pg += num_pg;
+       }
+
+       return 0;
+}
+
+static void mm_unaccount_pinned_pages(struct mmpin *mmp)
+{
+       if (mmp->user) {
+               atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
+               free_uid(mmp->user);
+       }
+}
+
 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
 {
        struct ubuf_info *uarg;
@@ -913,6 +951,12 @@ struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
 
        BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
        uarg = (void *)skb->cb;
+       uarg->mmp.user = NULL;
+
+       if (mm_account_pinned_pages(&uarg->mmp, size)) {
+               kfree_skb(skb);
+               return NULL;
+       }
 
        uarg->callback = sock_zerocopy_callback;
        uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
@@ -956,6 +1000,8 @@ struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
 
                next = (u32)atomic_read(&sk->sk_zckey);
                if ((u32)(uarg->id + uarg->len) == next) {
+                       if (mm_account_pinned_pages(&uarg->mmp, size))
+                               return NULL;
                        uarg->len++;
                        uarg->bytelen = bytelen;
                        atomic_set(&sk->sk_zckey, ++next);
@@ -1038,6 +1084,8 @@ EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
 void sock_zerocopy_put(struct ubuf_info *uarg)
 {
        if (uarg && atomic_dec_and_test(&uarg->refcnt)) {
+               mm_unaccount_pinned_pages(&uarg->mmp);
+
                if (uarg->callback)
                        uarg->callback(uarg, uarg->zerocopy);
                else