In bpf_tcp_sendmsg() the sk_alloc_sg() may fail. In the case of
ENOMEM, it may also mean that we've partially filled the scatterlist
entries with pages. Later jumping to sk_stream_wait_memory()
we could further fail with an error for several reasons, however
we miss to call free_start_sg() if the local sk_msg_buff was used.

Fixes: 4f738adba30a ("bpf: create tcp_bpf_ulp allowing BPF to monitor socket TX/RX data")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

While working on bpf_tcp_sendmsg() code, I noticed that when a
sk->sk_err is set we error out with err = sk->sk_err. However
this is problematic since sk->sk_err is a positive error value
and therefore we will neither go into sk_stream_error() nor will
we report an error back to user space. I had this case with EPIPE
and user space was thinking sendmsg() succeeded since EPIPE is
a positive value, thinking we submitted 32 bytes. Fix it by
negating the sk->sk_err value.

Fixes: 4f738adba30a ("bpf: create tcp_bpf_ulp allowing BPF to monitor socket TX/RX data")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

In commit

  'bpf: bpf_compute_data uses incorrect cb structure' (8108a7751512)

we added the routine bpf_compute_data_end_sk_skb() to compute the
correct data_end values, but this has since been lost. In kernel
v4.14 this was correct and the above patch was applied in it
entirety. Then when v4.14 was merged into v4.15-rc1 net-next tree
we lost the piece that renamed bpf_compute_data_pointers to the
new function bpf_compute_data_end_sk_skb. This was done here,

e1ea2f9856b7 ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net")

When it conflicted with the following rename patch,

6aaae2b6c433 ("bpf: rename bpf_compute_data_end into bpf_compute_data_pointers")

Finally, after a refactor I thought even the function
bpf_compute_data_end_sk_skb() was no longer needed and it was
erroneously removed.

However, we never reverted the sk_skb_convert_ctx_access() usage of
tcp_skb_cb which had been committed and survived the merge conflict.
Here we fix this by adding back the helper and *_data_end_sk_skb()
usage. Using the bpf_skc_data_end mapping is not correct because it
expects a qdisc_skb_cb object but at the sock layer this is not the
case. Even though it happens to work here because we don't overwrite
any data in-use at the socket layer and the cb structure is cleared
later this has potential to create some subtle issues. But, even
more concretely the filter.c access check uses tcp_skb_cb.

And by some act of chance though,

struct bpf_skb_data_end {
        struct qdisc_skb_cb        qdisc_cb;             /*     0    28 */

        /* XXX 4 bytes hole, try to pack */

        void *                     data_meta;            /*    32     8 */
        void *                     data_end;             /*    40     8 */

        /* size: 48, cachelines: 1, members: 3 */
        /* sum members: 44, holes: 1, sum holes: 4 */
        /* last cacheline: 48 bytes */
};

and then tcp_skb_cb,

struct tcp_skb_cb {
	[...]
                struct {
                        __u32      flags;                /*    24     4 */
                        struct sock * sk_redir;          /*    32     8 */
                        void *     data_end;             /*    40     8 */
                } bpf;                                   /*          24 */
        };

So when we use offset_of() to track down the byte offset we get 40 in
either case and everything continues to work. Fix this mess and use
correct structures its unclear how long this might actually work for
until someone moves the structs around.

Reported-by: Martin KaFai Lau <kafai@fb.com>
Fixes: e1ea2f9856b7 ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net")
Fixes: 6aaae2b6c433 ("bpf: rename bpf_compute_data_end into bpf_compute_data_pointers")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Currently, when a sock is closed and the bpf_tcp_close() callback is
used we remove memory but do not free the skb. Call consume_skb() if
the skb is attached to the buffer.

Reported-by: syzbot+d464d2c20c717ef5a6a8@syzkaller.appspotmail.com
Fixes: 1aa12bdf1bfb ("bpf: sockmap, add sock close() hook to remove socks")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

After latest lock updates there is no longer anything preventing a
close and recvmsg call running in parallel. Additionally, we can
race update with close if we close a socket and simultaneously update
if via the BPF userspace API (note the cgroup ops are already run
with sock_lock held).

To resolve this take sock_lock in close and update paths.

Reported-by: syzbot+b680e42077a0d7c9a0c4@syzkaller.appspotmail.com
Fixes: e9db4ef6bf4c ("bpf: sockhash fix omitted bucket lock in sock_close")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

This removes locking from readers of RCU hash table. Its not
necessary.

Fixes: 81110384441a ("bpf: sockmap, add hash map support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

The current code, in the error path of sock_hash_ctx_update_elem,
checks if the sock has a psock in the user data and if so decrements
the reference count of the psock. However, if the error happens early
in the error path we may have never incremented the psock reference
count and if the psock exists because the sock is in another map then
we may inadvertently decrement the reference count.

Fix this by making the error path only call smap_release_sock if the
error happens after the increment.

Reported-by: syzbot+d464d2c20c717ef5a6a8@syzkaller.appspotmail.com
Fixes: 81110384441a ("bpf: sockmap, add hash map support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Add map_release_uref pointer to hashmap ops. This was dropped when
original sockhash code was ported into bpf-next before initial
commit.

Fixes: 81110384441a ("bpf: sockmap, add hash map support")
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

First the sk_callback_lock() was being used to protect both the
sock callback hooks and the psock->maps list. This got overly
convoluted after the addition of sockhash (in sockmap it made
some sense because masp and callbacks were tightly coupled) so
lets split out a specific lock for maps and only use the callback
lock for its intended purpose. This fixes a couple cases where
we missed using maps lock when it was in fact needed. Also this
makes it easier to follow the code because now we can put the
locking closer to the actual code its serializing.

Next, in sock_hash_delete_elem() the pattern was as follows,

  sock_hash_delete_elem()
     [...]
     spin_lock(bucket_lock)
     l = lookup_elem_raw()
     if (l)
        hlist_del_rcu()
        write_lock(sk_callback_lock)
         .... destroy psock ...
        write_unlock(sk_callback_lock)
     spin_unlock(bucket_lock)

The ordering is necessary because we only know the {p}sock after
dereferencing the hash table which we can't do unless we have the
bucket lock held. Once we have the bucket lock and the psock element
it is deleted from the hashmap to ensure any other path doing a lookup
will fail. Finally, the refcnt is decremented and if zero the psock
is destroyed.

In parallel with the above (or free'ing the map) a tcp close event
may trigger tcp_close(). Which at the moment omits the bucket lock
altogether (oops!) where the flow looks like this,

  bpf_tcp_close()
     [...]
     write_lock(sk_callback_lock)
     for each psock->maps // list of maps this sock is part of
         hlist_del_rcu(ref_hash_node);
         .... destroy psock ...
     write_unlock(sk_callback_lock)

Obviously, and demonstrated by syzbot, this is broken because
we can have multiple threads deleting entries via hlist_del_rcu().

To fix this we might be tempted to wrap the hlist operation in a
bucket lock but that would create a lock inversion problem. In
summary to follow locking rules the psocks maps list needs the
sk_callback_lock (after this patch maps_lock) but we need the bucket
lock to do the hlist_del_rcu.

To resolve the lock inversion problem pop the head of the maps list
repeatedly and remove the reference until no more are left. If a
delete happens in parallel from the BPF API that is OK as well because
it will do a similar action, lookup the lock in the map/hash, delete
it from the map/hash, and dec the refcnt. We check for this case
before doing a destroy on the psock to ensure we don't have two
threads tearing down a psock. The new logic is as follows,

  bpf_tcp_close()
  e = psock_map_pop(psock->maps) // done with map lock
  bucket_lock() // lock hash list bucket
  l = lookup_elem_raw(head, hash, key, key_size);
  if (l) {
     //only get here if elmnt was not already removed
     hlist_del_rcu()
     ... destroy psock...
  }
  bucket_unlock()

And finally for all the above to work add missing locking around  map
operations per above. Then add RCU annotations and use
rcu_dereference/rcu_assign_pointer to manage values relying on RCU so
that the object is not free'd from sock_hash_free() while it is being
referenced in bpf_tcp_close().

Reported-by: syzbot+0ce137753c78f7b6acc1@syzkaller.appspotmail.com
Fixes: 81110384441a ("bpf: sockmap, add hash map support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

If a hashmap is free'd with open socks it removes the reference to
the hash entry from the psock. If that is the last reference to the
psock then it will also be free'd by the reference counting logic.
However the current logic that removes the hash reference from the
list of references is broken. In smap_list_remove() we first check
if the sockmap entry matches and then check if the hashmap entry
matches. But, the sockmap entry sill always match because its NULL in
this case which causes the first entry to be removed from the list.
If this is always the "right" entry (because the user adds/removes
entries in order) then everything is OK but otherwise a subsequent
bpf_tcp_close() may reference a free'd object.

To fix this create two list handlers one for sockmap and one for
sockhash.

Reported-by: syzbot+0ce137753c78f7b6acc1@syzkaller.appspotmail.com
Fixes: 81110384441a ("bpf: sockmap, add hash map support")
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>