Skip to content

Commit

Permalink
Revert "libceph: use memalloc flags for net IO"
Browse files Browse the repository at this point in the history
This reverts commit 89baaa5.

Dirty page throttling should be sufficient for us in the general case
so there is no need to use __GFP_MEMALLOC - it would be needed only in
the swap-over-rbd case, which we currently don't support.  (It would
probably take approximately the commit that is being reverted to add
that support, but we would also need the "swap" option to distinguish
from the general case and make sure swap ceph_client-s aren't shared
with anything else.)  See ceph-devel threads [1] and [2] for the
details of why enabling pfmemalloc reserves for all cases is a bad
thing.

On top of potential system lockups related to drained emergency
reserves, this turned out to cause ceph lockups in case peers are on
the same host and communicating via loopback due to sk_filter()
dropping pfmemalloc skbs on the receiving side because the receiving
loopback socket is not tagged with SOCK_MEMALLOC.

[1] "SOCK_MEMALLOC vs loopback"
    http://www.spinics.net/lists/ceph-devel/msg22998.html
[2] "[PATCH] libceph: don't set memalloc flags in loopback case"
    http://www.spinics.net/lists/ceph-devel/msg23392.html

Conflicts:
	net/ceph/messenger.c [ context: tcp_nodelay option ]

Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Sage Weil <sage@redhat.com>
Cc: stable@vger.kernel.org # 3.18+, needs backporting
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Acked-by: Mike Christie <michaelc@cs.wisc.edu>
Acked-by: Mel Gorman <mgorman@suse.de>
  • Loading branch information
Ilya Dryomov committed Apr 7, 2015
1 parent f22e6e8 commit 6d7fdb0
Showing 1 changed file with 1 addition and 8 deletions.
9 changes: 1 addition & 8 deletions net/ceph/messenger.c
Original file line number Diff line number Diff line change
Expand Up @@ -484,7 +484,7 @@ static int ceph_tcp_connect(struct ceph_connection *con)
IPPROTO_TCP, &sock);
if (ret)
return ret;
sock->sk->sk_allocation = GFP_NOFS | __GFP_MEMALLOC;
sock->sk->sk_allocation = GFP_NOFS;

#ifdef CONFIG_LOCKDEP
lockdep_set_class(&sock->sk->sk_lock, &socket_class);
Expand Down Expand Up @@ -520,8 +520,6 @@ static int ceph_tcp_connect(struct ceph_connection *con)
ret);
}

sk_set_memalloc(sock->sk);

con->sock = sock;
return 0;
}
Expand Down Expand Up @@ -2808,11 +2806,8 @@ static void con_work(struct work_struct *work)
{
struct ceph_connection *con = container_of(work, struct ceph_connection,
work.work);
unsigned long pflags = current->flags;
bool fault;

current->flags |= PF_MEMALLOC;

mutex_lock(&con->mutex);
while (true) {
int ret;
Expand Down Expand Up @@ -2866,8 +2861,6 @@ static void con_work(struct work_struct *work)
con_fault_finish(con);

con->ops->put(con);

tsk_restore_flags(current, pflags, PF_MEMALLOC);
}

/*
Expand Down

0 comments on commit 6d7fdb0

Please sign in to comment.