staging: lustre: ptlrpc: do not sleep if encpool reached max capacity

When encryption is enabled RPCs are encrypted just before being
sent. The encryption requires allocating memory in the encoding pool.
The current implementation in sptlrpc_enc_pool_get_pages() is
deadlock-prone. Indeed, if there is no more free pages in the pool,
all ptlrpcd threads can end up waiting in a queue, so there is no
thread available to process other requests. It means client is not
able to process replies from servers that yet contain last committed
transno useful to release memory allocated by previous requests,
including enc_pool pages.

To fix this, in sptlrpc_enc_pool_get_pages(), do not make ptlrpcd
threads wait in queue if encoding pool has already reached its maximum
capacity. Instead, return -ENOMEM. If functions calling ptl_send_rpc()
get -ENOMEM, then put back request in queue by moving it back to
RQ_PHASE_NEW phase.

As an optimization, do not call ptl_send_rpc() again for requests that
already failed to allocate in the enc_pool, as long as there is not
enough memory in the enc_pool to satisfy theirs needs.

In /sys/fs/lustre/sptlrpc/encrypt_page_pools, add a new 'out of mem'
stat to track how many requests fail to allocate memory in the
enc_pool.

Signed-off-by: Sebastien Buisson <sebastien.buisson@bull.net>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6356
Reviewed-on: http://review.whamcloud.com/15070


Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Dmitry Eremin <dmitry.eremin@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: James Simmons <jsimmons@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>