Santosh Shilimkar says:
====================
RDS: connection scalability and performance improvements
[v4]
Re-sending the same patches from v3 again since my repost of
patch 05/14 from v3 was whitespace damaged.
[v3]
Updated patch "[PATCH v2 05/14] RDS: defer the over_batch work to
send worker" as per David Miller's comment [4] to avoid the magic
value usage. Patch now makes use of already available but unused
send_batch_count module parameter. Rest of the patches are same as
earlier version v2 [3]
[v2]:
Dropped "[PATCH 05/15] RDS: increase size of hash-table to 8K" from
earlier version [1]. I plan to address the hash table scalability using
re-sizable hash tables as suggested by David Laight and David Miller [2]
This series addresses RDS connection bottlenecks on massive workloads and
improve the RDMA performance almost by 3X. RDS TCP also gets a small gain
of about 12%.
RDS is being used in massive systems with high scalability where several
hundred thousand end points and tens of thousands of local processes
are operating in tens of thousand sockets. Being RC(reliable connection),
socket bind and release happens very often and any inefficiencies in
bind hash look ups hurts the overall system performance. RDS bin hash-table
uses global spin-lock which is the biggest bottleneck. To make matter worst,
it uses rcu inside global lock for hash buckets.
This is being addressed by simply using per bucket rw lock which makes the
locking simple and very efficient. The hash table size is still an issue and
I plan to address it by using re-sizable hash tables as suggested on the list.
For RDS RDMA improvement, the completion handling is revamped so that we
can do batch completions. Both send and receive completion handlers are
split logically to achieve the same. RDS 8K messages being one of the
key usecase, mr pool is adapted to have the 8K mrs along with default 1M
mrs. And while doing this, few fixes and couple of bottlenecks seen with
rds_sendmsg() are addressed.
Series applies against 4.3-rc1 as well net-next. Its tested on Oracle
hardware with IB fabric for both bcopy as well as RDMA mode. RDS TCP is
tested with iXGB NIC. Like last time, iWARP transport is untested with
these changes. The patchset is also available at below git repo:
git://git.kernel.org/pub/scm/linux/kernel/git/ssantosh/linux.git net/rds/4.3-v3
As a side note, the IB HCA driver I used for testing misses at least 3
important patches in upstream to see the full blown IB performance and
am hoping to get that in mainline with help of them.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
8K message sizes are pretty important usecase for RDS current
workloads so we make provison to have 8K mrs available from the pool.
Based on number of SG's in the RDS message, we pick a pool to use.
Also to make sure that we don't under utlise mrs when say 8k messages
are dominating which could lead to 8k pull being exhausted, we fall-back
to 1m pool till 8k pool recovers for use.
This helps to at least push ~55 kB/s bidirectional data which
is a nice improvement.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
All HCA drivers seems to popullate max_mr caps and few of
them do both max_mr and max_fmr.
Hence update RDS code to make use of max_mr.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Fix below warning by marking rds_ib_fmr_wq static
net/rds/ib_rdma.c:87:25: warning: symbol 'rds_ib_fmr_wq' was not declared. Should it be static?
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
rds_ib_mr already keeps the pool handle which it associates
with. Lets use that instead of round about way of fetching
it from rds_ib_device.
No functional change.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
RDS IB mr pool has its own workqueue 'rds_ib_fmr_wq', so we need
to use queue_delayed_work() to kick the work. This was hurting
the performance since pool maintenance was less often triggered
from other path.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Just in case we are still handling the QP receive completion while the
rds_ibdev is released, drop the connection instead of crashing the kernel.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Similar to what we did with receive CQ completion handling, we split
the transmit completion handler so that it lets us implement batched
work completion handling.
We re-use the cq_poll routine and makes use of RDS_IB_SEND_OP to
identify the send vs receive completion event handler invocation.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
For better performance, we split the receive completion IRQ handler. That
lets us acknowledge several WCE events in one call. We also limit the WC
to max 32 to avoid latency. Acknowledging several completions in one call
instead of several calls each time will provide better performance since
less mutual exclusion locks are being performed.
In next patch, send completion is also split which re-uses the poll_cq()
and hence the code is moved to ib_cm.c
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
In Transport indepedent rds_sendmsg(), we shouldn't make decisions based
on RDS_LL_SEND_FULL which is used to manage the ring for RDMA based
transports. We can safely issue rds_send_xmit() and the using its
return value take decision on deferred work. This will also fix
the scenario where at times we are seeing connections stuck with
the LL_SEND_FULL bit getting set and never cleared.
We kick krdsd after any time we see -ENOMEM or -EAGAIN from the
ring allocation code.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Current process gives up if its send work over the batch limit.
The work queue will get kicked to finish off any other requests.
This fixes remainder condition from commit 443be0e5af ("RDS: make
sure not to loop forever inside rds_send_xmit").
The restart condition is only for the case where we reached to
over_batch code for some other reason so just retrying again
before giving up.
While at it, make sure we use already available 'send_batch_count'
parameter instead of magic value. The batch count threshold value
of 1024 came via commit 443be0e5af ("RDS: make sure not to loop
forever inside rds_send_xmit"). The idea is to process as big a
batch as we can but at the same time we don't hold other waiting
processes for send. Hence back-off after the send_batch_count
limit (1024) to avoid soft-lock ups.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
For the same reasons as commit 2f53384424 ("tcp: allow splice() to
build full TSO packets") and commit 35f9c09fe9 ("tcp: tcp_sendpages()
should call tcp_push() once"), rds_tcp_xmit may have multiple pages to
send, so use the MSG_MORE and MSG_SENDPAGE_NOTLAST as hints to
tcp_sendpage()
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Using the value of RDS_TCP_DEFAULT_BUFSIZE (128K)
clobbers efficient use of TSO because it inflates the size_goal
that is computed in tcp_sendmsg/tcp_sendpage and skews packet
latency, and the default values for these parameters actually
results in significantly better performance.
In request-response tests using rds-stress with a packet size of
100K with 16 threads (test parameters -q 100000 -a 256 -t16 -d16)
between a single pair of IP addresses achieves a throughput of
6-8 Gbps. Without this patch, throughput maxes at 2-3 Gbps under
equivalent conditions on these platforms.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit f711a6ae06 ("net/rds: RDS-TCP: Always create a new rds_sock
for an incoming connection.") modified rds-tcp so that an incoming SYN
would ignore an existing "client" TCP connection which had the local
port set to the transient port. The motivation for ignoring the existing
"client" connection in f711a6ae was to avoid race conditions and an
endless duel of reconnect attempts triggered by a restart/abort of one
of the nodes in the TCP connection.
However, having separate sockets for active and passive sides
is avoidable, and the simpler model of a single TCP socket for
both send and receives of all RDS connections associated with
that tcp socket makes for easier observability. We avoid the race
conditions from f711a6ae by attempting reconnects in rds_conn_shutdown
if, and only if, the (new) c_outgoing bit is set for RDS_TRANS_TCP.
The c_outgoing bit is initialized in __rds_conn_create().
A side-effect of re-using the client rds_connection for an incoming
SYN is the potential of encountering duelling SYNs, i.e., we
have an outgoing RDS_CONN_CONNECTING socket when we get the incoming
SYN. The logic to arbitrate this criss-crossing SYN exchange in
rds_tcp_accept_one() has been modified to emulate the BGP state
machine: the smaller IP address should back off from the connection attempt.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One global lock protecting hash-tables with 1024 buckets isn't
efficient and it shows up in a massive systems with truck
loads of RDS sockets serving multiple databases. The
perf data clearly highlights the contention on the rw
lock in these massive workloads.
When the contention gets worse, the code gets into a state where
it decides to back off on the lock. So while it has disabled interrupts,
it sits and backs off on this lock get. This causes the system to
become sluggish and eventually all sorts of bad things happen.
The simple fix is to move the lock into the hash bucket and
use per-bucket lock to improve the scalability.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
One need to take rds socket reference while using it and release it
once done with it. rds_add_bind() code path does not do that so
lets fix it.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
RDS bind and release locking scheme is very inefficient. It
uses RCU for maintaining the bind hash-table which is great but
it also needs to hold spinlock for [add/remove]_bound(). So
overall usecase, the hash-table concurrent speedup doesn't pay off.
In fact blocking nature of synchronize_rcu() makes the RDS
socket shutdown too slow which hurts RDS performance since
connection shutdown and re-connect happens quite often to
maintain the RC part of the protocol.
So we make the locking scheme simpler and more efficient by
replacing spin_locks with reader/writer locks and getting rid
off rcu for bind hash-table.
In subsequent patch, we also covert the global lock with per-bucket
lock to reduce the global lock contention.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
synchronize_rcu() slowing down un-necessarily the socket shutdown
path. It is used just kfree() the ip addresses in rds_ib_remove_ipaddr()
which is perfect usecase for kfree_rcu();
So lets use that to gain some speedup.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Pull networking fixes from David Miller:
1) Fix out-of-bounds array access in netfilter ipset, from Jozsef
Kadlecsik.
2) Use correct free operation on netfilter conntrack templates, from
Daniel Borkmann.
3) Fix route leak in SCTP, from Marcelo Ricardo Leitner.
4) Fix sizeof(pointer) in mac80211, from Thierry Reding.
5) Fix cache pointer comparison in ip6mr leading to missed unlock of
mrt_lock. From Richard Laing.
6) rds_conn_lookup() needs to consider network namespace in key
comparison, from Sowmini Varadhan.
7) Fix deadlock in TIPC code wrt broadcast link wakeups, from Kolmakov
Dmitriy.
8) Fix fd leaks in bpf syscall, from Daniel Borkmann.
9) Fix error recovery when installing ipv6 multipath routes, we would
delete the old route before we would know if we could fully commit
to the new set of nexthops. Fix from Roopa Prabhu.
10) Fix run-time suspend problems in r8152, from Hayes Wang.
11) In fec, don't program the MAC address into the chip when the clocks
are gated off. From Fugang Duan.
12) Fix poll behavior for netlink sockets when using rx ring mmap, from
Daniel Borkmann.
13) Don't allocate memory with GFP_KERNEL from get_stats64 in r8169
driver, from Corinna Vinschen.
14) In TCP Cubic congestion control, handle idle periods better where we
are application limited, in order to keep cwnd from growing out of
control. From Eric Dumzet.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (65 commits)
tcp_cubic: better follow cubic curve after idle period
tcp: generate CA_EVENT_TX_START on data frames
xen-netfront: respect user provided max_queues
xen-netback: respect user provided max_queues
r8169: Fix sleeping function called during get_stats64, v2
ether: add IEEE 1722 ethertype - TSN
netlink, mmap: fix edge-case leakages in nf queue zero-copy
netlink, mmap: don't walk rx ring on poll if receive queue non-empty
cxgb4: changes for new firmware 1.14.4.0
net: fec: add netif status check before set mac address
r8152: fix the runtime suspend issues
r8152: split DRIVER_VERSION
ipv6: fix ifnullfree.cocci warnings
add microchip LAN88xx phy driver
stmmac: fix check for phydev being open
net: qlcnic: delete redundant memsets
net: mv643xx_eth: use kzalloc
net: jme: use kzalloc() instead of kmalloc+memset
net: cavium: liquidio: use kzalloc in setup_glist()
net: ipv6: use common fib_default_rule_pref
...
- Create drivers/staging/rdma
- Move amso1100 driver to staging/rdma and schedule for deletion
- Move ipath driver to staging/rdma and schedule for deletion
- Add hfi1 driver to staging/rdma and set TODO for move to regular tree
- Initial support for namespaces to be used on RDMA devices
- Add RoCE GID table handling to the RDMA core caching code
- Infrastructure to support handling of devices with differing
read and write scatter gather capabilities
- Various iSER updates
- Kill off unsafe usage of global mr registrations
- Update SRP driver
- Misc. mlx4 driver updates
- Support for the mr_alloc verb
- Support for a netlink interface between kernel and user space cache
daemon to speed path record queries and route resolution
- Ininitial support for safe hot removal of verbs devices
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQIcBAABAgAGBQJV7v8wAAoJELgmozMOVy/d2dcP/3PXnGFPgFGJODKE6VCZtTvj
nooNXRKXjxv470UT5DiAX7SNcBxzzS7Zl/Lj+831H9iNXUyzuH31KtBOAZ3W03vZ
yXwCB2caOStSldTRSUUvPe2aIFPnyNmSpC4i6XcJLJMCFijKmxin5pAo8qE44BQU
yjhT+wC9P6LL5wZXsn/nFIMLjOFfu0WBFHNp3gs5j59paxlx5VeIAZk16aQZH135
m7YCyicwrS8iyWQl2bEXRMon2vlCHlX2RHmOJ4f/P5I0quNcGF2+d8Yxa+K1VyC5
zcb3OBezz+wZtvh16yhsDfSPqHWirljwID2VzOgRSzTJWvQjju8VkwHtkq6bYoBW
egIxGCHcGWsD0R5iBXLYr/tB+BmjbDObSm0AsR4+JvSShkeVA1IpeoO+19162ixE
n6CQnk2jCee8KXeIN4PoIKsjRSbIECM0JliWPLoIpuTuEhhpajftlSLgL5hf1dzp
HrSy6fXmmoRj7wlTa7DnYIC3X+ffwckB8/t1zMAm2sKnIFUTjtQXF7upNiiyWk4L
/T1QEzJ2bLQckQ9yY4v528SvBQwA4Dy1amIQB7SU8+2S//bYdUvhysWPkdKC4oOT
WlqS5PFDCI31MvNbbM3rUbMAD8eBAR8ACw9ZpGI/Rffm5FEX5W3LoxA8gfEBRuqt
30ZYFuW8evTL+YQcaV65
=EHLg
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull inifiniband/rdma updates from Doug Ledford:
"This is a fairly sizeable set of changes. I've put them through a
decent amount of testing prior to sending the pull request due to
that.
There are still a few fixups that I know are coming, but I wanted to
go ahead and get the big, sizable chunk into your hands sooner rather
than waiting for those last few fixups.
Of note is the fact that this creates what is intended to be a
temporary area in the drivers/staging tree specifically for some
cleanups and additions that are coming for the RDMA stack. We
deprecated two drivers (ipath and amso1100) and are waiting to hear
back if we can deprecate another one (ehca). We also put Intel's new
hfi1 driver into this area because it needs to be refactored and a
transfer library created out of the factored out code, and then it and
the qib driver and the soft-roce driver should all be modified to use
that library.
I expect drivers/staging/rdma to be around for three or four kernel
releases and then to go away as all of the work is completed and final
deletions of deprecated drivers are done.
Summary of changes for 4.3:
- Create drivers/staging/rdma
- Move amso1100 driver to staging/rdma and schedule for deletion
- Move ipath driver to staging/rdma and schedule for deletion
- Add hfi1 driver to staging/rdma and set TODO for move to regular
tree
- Initial support for namespaces to be used on RDMA devices
- Add RoCE GID table handling to the RDMA core caching code
- Infrastructure to support handling of devices with differing read
and write scatter gather capabilities
- Various iSER updates
- Kill off unsafe usage of global mr registrations
- Update SRP driver
- Misc mlx4 driver updates
- Support for the mr_alloc verb
- Support for a netlink interface between kernel and user space cache
daemon to speed path record queries and route resolution
- Ininitial support for safe hot removal of verbs devices"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (136 commits)
IB/ipoib: Suppress warning for send only join failures
IB/ipoib: Clean up send-only multicast joins
IB/srp: Fix possible protection fault
IB/core: Move SM class defines from ib_mad.h to ib_smi.h
IB/core: Remove unnecessary defines from ib_mad.h
IB/hfi1: Add PSM2 user space header to header_install
IB/hfi1: Add CSRs for CONFIG_SDMA_VERBOSITY
mlx5: Fix incorrect wc pkey_index assignment for GSI messages
IB/mlx5: avoid destroying a NULL mr in reg_user_mr error flow
IB/uverbs: reject invalid or unknown opcodes
IB/cxgb4: Fix if statement in pick_local_ip6adddrs
IB/sa: Fix rdma netlink message flags
IB/ucma: HW Device hot-removal support
IB/mlx4_ib: Disassociate support
IB/uverbs: Enable device removal when there are active user space applications
IB/uverbs: Explicitly pass ib_dev to uverbs commands
IB/uverbs: Fix race between ib_uverbs_open and remove_one
IB/uverbs: Fix reference counting usage of event files
IB/core: Make ib_dealloc_pd return void
IB/srp: Create an insecure all physical rkey only if needed
...
Only return a conn if the rds_conn_net(conn) matches the struct
net passed to rds_conn_lookup().
Fixes: 467fa15356 ("RDS-TCP: Support multiple RDS-TCP listen endpoints,
one per netns.")
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The majority of callers never check the return value, and even if they
did, they can't do anything about a failure.
All possible failure cases represent a bug in the caller, so just
WARN_ON inside the function instead.
This fixes a few random errors:
net/rd/iw.c infinite loops while it fails. (racing with EBUSY?)
This also lays the ground work to get rid of error return from the
drivers. Most drivers do not error, the few that do are broken since
it cannot be handled.
Since uverbs can legitimately make use of EBUSY, open code the
check.
Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The pd now has a local_dma_lkey member which completely replaces
ib_get_dma_mr, use it instead.
Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
An ib_client callback that is called with the lists_rwsem locked only for
read is protected from changes to the IB client lists, but not from
ib_unregister_device() freeing its client data. This is because
ib_unregister_device() will remove the device from the device list with
lists_rwsem locked for write, but perform the rest of the cleanup,
including the call to remove() without that lock.
Mark client data that is undergoing de-registration with a new going_down
flag in the client data context. Lock the client data list with lists_rwsem
for write in addition to using the spinlock, so that functions calling the
callback would be able to lock only lists_rwsem for read and let callbacks
sleep.
Since ib_unregister_client() now marks the client data context, no need for
remove() to search the context again, so pass the client data directly to
remove() callbacks.
Reviewed-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Memory allocated for 'ibmr' uses kzalloc_node() which already
initialises the memory to zero. There is no need to do
memset() 0 on that memory.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
FMR flush is an expensive and time consuming operation. Reduce the
frequency of FMR pool flush by 50% so that more FMR work gets accumulated
for more efficient flushing.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RDS FMR flush operation and also it races with connect/reconect
which happes a lot with RDS. FMR flush being on common rds_wq aggrevates
the problem. Lets push RDS FMR pool flush work to its own worker.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In rds_ib_flush_mr_pool(), dirty_count accounts the clean ones
which is wrong. This can lead to a negative dirty count value.
Lets fix it.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On rds_ib_frag_slab allocation failure, ensure rds_ib_incoming_slab
is not pointing to the detsroyed memory.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
>> net/rds/ib_recv.c:382:28: sparse: incorrect type in initializer (different base types)
net/rds/ib_recv.c:382:28: expected int [signed] can_wait
net/rds/ib_recv.c:382:28: got restricted gfp_t
net/rds/ib_recv.c:828:23: sparse: cast to restricted __le64
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Connection could have been dropped while the route is being resolved
so check for valid cm_id before initiating the connection.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rds_send_queue_rm() allows for the "current datagram" being queued
to exceed SO_SNDBUF thresholds by checking bytes queued without
counting in length of current datagram. (Since sk_sndbuf is set
to twice requested SO_SNDBUF value as a kernel heuristic this
is usually fine!)
If this "current datagram" squeezing past the threshold is itself
many times the size of the sk_sndbuf threshold itself then even
twice the SO_SNDBUF does not save us and it gets queued but
cannot be transmitted. Threads block and deadlock and device
becomes unusable. The check for this datagram not exceeding
SNDBUF thresholds (EMSGSIZE) is not done on this datagram as
that check is only done if queueing attempt fails.
(Datagrams that follow this datagram fail queueing attempts, go
through the check and eventually trip EMSGSIZE error but zero
length datagrams silently fail!)
This fix moves the check for datagrams exceeding SNDBUF limits
before any processing or queueing is attempted and returns EMSGSIZE
early in the rds_sndmsg() code. This change also ensures that all
datagrams get checked for exceeding SNDBUF/sk_sndbuf size limits
and the large datagrams that exceed those limits do not get to
rds_send_queue_rm() code for processing.
Signed-off-by: Mukesh Kacker <mukesh.kacker@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rds_send_drop_to() is used during socket tear down to find all the
messages on the socket and flush them . It can race with the
acking code unless it takes the m_rs_lock on each and every message.
This plugs a hole where we didn't take m_rs_lock on any message that
didn't have the RDS_MSG_ON_CONN set. Taking m_rs_lock avoids
double frees and other memory corruptions as the ack code trusts
the message m_rs pointer on a socket that had actually been freed.
We must take m_rs_lock to access m_rs. Because of lock nesting and
rs access, we also need to acquire rs_lock.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
During connection resets, we are destroying the rdma id too soon. We can't
destroy it when it is still in use. So lets move rdma_destroy_id() after
we clear the rings.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the asserion level since its not fatal and can be hit
in normal execution paths. There is no need to take the
system down.
We keep the WARN_ON() to detect the condition if we get
here with bad pages.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
WR(Work Requests )always generate a WC(Work Completion) with
signaled send. Default RDS ib code is setup for un-signaled
completion. Since RDS connction is persistent, we can end up
sending the data even after large-send when the remote end is
not active(for any reason).
By doing a signaled send at least once per large-send,
we can at least detect the problem in work completion
handler there by avoiding sending more data to
inactive remote.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rds_send_xmit() marks the rds message map flag after
xmit_[rdma/atomic]() which is clearly wrong. We need
to maintain the ownership between transport and rds.
Also take care of error path.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This helps to detect the accidental processes/apps trying to destroy
the RDS socket which they are sharing with other processes/apps.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ensure we don't keep sending the data if the link is congested.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If we get an ENOMEM during rds_ib_recv_refill, we might never come
back and refill again later. Patch makes sure to kick krdsd into
helping out.
To achieve this we add RDS_RECV_REFILL flag and update in the refill
path based on that so that at least some therad will keep posting
receive buffers.
Since krdsd and softirq both might race for refill, we decide to
schedule on work queue based on ring_low instead of ring_empty.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the ip address tables hasn't changed, there is no need to remove
them only to be added back again.
Lets fix it.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Destroy ib state early during shutdown. Otherwise we can get callbacks
after the QP isn't really able to handle them.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We were still seeing rare occurrences of the WARN_ON(recv->r_frag) which
indicates that the recv refill path was finding allocated frags in ring
entries that were marked free. These were usually followed by OOM crashes.
They only seem to be occurring in the presence of completion errors and
connection resets.
This patch ensures that we free the frag as we mark the ring entry free.
This should stop the refill path from finding allocated frags in ring
entries that were marked free.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In rds_cmsg_rdma_args() 'ret' is used by rds_pin_pages() which returns
number of pinned pages on success. And the same value is returned to the
caller of rds_cmsg_rdma_args() on success which is not intended.
Commit f4a3fc03c1 ("RDS: Clean up error handling in rds_cmsg_rdma_args")
removed the 'ret = 0' line which broke RDS RDMA mode.
Fix it by restoring the return value on rds_pin_pages() success
keeping the clean-up in place.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/cavium/Kconfig
The cavium conflict was overlapping dependency
changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Register pernet subsys init/stop functions that will set up
and tear down per-net RDS-TCP listen endpoints. Unregister
pernet subusys functions on 'modprobe -r' to clean up these
end points.
Enable keepalive on both accept and connect socket endpoints.
The keepalive timer expiration will ensure that client socket
endpoints will be removed as appropriate from the netns when
an interface is removed from a namespace.
Register a device notifier callback that will clean up all
sockets (and thus avoid the need to wait for keepalive timeout)
when the loopback device is unregistered from the netns indicating
that the netns is getting deleted.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Open the sockets calling sock_create_kern() with the correct struct net
pointer, and use that struct net pointer when verifying the
address passed to rds_bind().
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller
Santosh Shilimkar
Sowmini Varadhan
Linus Torvalds
Sasha Levin
Jason Gunthorpe
Sagi Grimberg
Haggai Eran
Wengang Wang
Mukesh Kacker
Santosh Shilimkar