- 16 Feb, 2018 27 commits
-
-
David S. Miller authored
Sowmini Varadhan says: ==================== RDS: zerocopy support This is version 3 of the series, following up on review comments for http://patchwork.ozlabs.org/project/netdev/list/?series=28530 Review comments addressed Patch 4 - fix fragile use of skb->cb[], do not set ee_code incorrectly. Patch 5: - remove needless bzero of skb->cb[], consolidate err cleanup A brief overview of this feature follows. This patch series provides support for MSG_ZERCOCOPY on a PF_RDS socket based on the APIs and infrastructure added by Commit f214f915 ("tcp: enable MSG_ZEROCOPY") For single threaded rds-stress testing using rds-tcp with the ixgbe driver using 1M message sizes (-a 1M -q 1M) preliminary results show that there is a significant reduction in latency: about 90 usec with zerocopy, compared with 200 usec without zerocopy. This patchset modifies the above for zerocopy in the following manner. - if the MSG_ZEROCOPY flag is specified with rds_sendmsg(), and, - if the SO_ZEROCOPY socket option has been set on the PF_RDS socket, application pages sent down with rds_sendmsg are pinned. The pinning uses the accounting infrastructure added by a91dbff5 ("sock: ulimit on MSG_ZEROCOPY pages"). The message is unpinned when all references to the message go down to 0, and the message is freed by rds_message_purge. A multithreaded application using this infrastructure must send down a unique 32 bit cookie as ancillary data with each sendmsg invocation. The format of this ancillary data is described in Patch 5 of the series. The cookie is passed up to the application on the sk_error_queue when the message is unpinned, indicating to the application that it is now safe to free/reuse the message buffer. The details of the completion notification are provided in Patch 4 of this series. ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Sowmini Varadhan authored
Send a cookie with sendmsg() on PF_RDS sockets, and process the returned batched cookies in do_recv_completion() Signed-off-by:
Sowmini Varadhan <sowmini.varadhan@oracle.com> Acked-by:
Willem de Bruijn <willemb@google.com> Signed-off-by:
-
Sowmini Varadhan authored
Add support for basic PF_RDS client-server testing in msg_zerocopy Signed-off-by:
-
Sowmini Varadhan authored
If the MSG_ZEROCOPY flag is specified with rds_sendmsg(), and, if the SO_ZEROCOPY socket option has been set on the PF_RDS socket, application pages sent down with rds_sendmsg() are pinned. The pinning uses the accounting infrastructure added by Commit a91dbff5 ("sock: ulimit on MSG_ZEROCOPY pages") The payload bytes in the message may not be modified for the duration that the message has been pinned. A multi-threaded application using this infrastructure may thus need to be notified about send-completion so that it can free/reuse the buffers passed to rds_sendmsg(). Notification of send-completion will identify each message-buffer by a cookie that the application must specify as ancillary data to rds_sendmsg(). The ancillary data in this case has cmsg_level == SOL_RDS and cmsg_type == RDS_CMSG_ZCOPY_COOKIE. Signed-off-by:
Santosh Shilimkar <santosh.shilimkar@oracle.com> Acked-by:
-
Sowmini Varadhan authored
RDS removes a datagram (rds_message) from the retransmit queue when an ACK is received. The ACK indicates that the receiver has queued the RDS datagram, so that the sender can safely forget the datagram. When all references to the rds_message are quiesced, rds_message_purge is called to release resources used by the rds_message If the datagram to be removed had pinned pages set up, add an entry to the rs->rs_znotify_queue so that the notifcation will be sent up via rds_rm_zerocopy_callback() when the rds_message is eventually freed by rds_message_purge. rds_rm_zerocopy_callback() attempts to batch the number of cookies sent with each notification to a max of SO_EE_ORIGIN_MAX_ZCOOKIES. This is achieved by checking the tail skb in the sk_error_queue: if this has room for one more cookie, the cookie from the current notification is added; else a new skb is added to the sk_error_queue. Every invocation of rds_rm_zerocopy_callback() will trigger a ->sk_error_report to notify the application. Signed-off-by:
-
Sowmini Varadhan authored
allow the application to set SO_ZEROCOPY on the underlying sk of a PF_RDS socket Signed-off-by:
-
Sowmini Varadhan authored
The existing model holds a reference from the rds_sock to the rds_message, but the rds_message does not itself hold a sock_put() on the rds_sock. Instead the m_rs field in the rds_message is assigned when the message is queued on the sock, and nulled when the message is dequeued from the sock. We want to be able to notify userspace when the rds_message is actually freed (from rds_message_purge(), after the refcounts to the rds_message go to 0). At the time that rds_message_purge() is called, the message is no longer on the rds_sock retransmit queue. Thus the explicit reference for the m_rs is needed to send a notification that will signal to userspace that it is now safe to free/reuse any pages that may have been pinned down for zerocopy. This patch manages the m_rs assignment in the rds_message with the necessary refcount book-keeping. Signed-off-by:
-
Sowmini Varadhan authored
RDS would like to use the helper functions for managing pinned pages added by Commit a91dbff5 ("sock: ulimit on MSG_ZEROCOPY pages") Signed-off-by:
-
David S. Miller authored
It was mis-applied and the changes had rejects. Signed-off-by: -
David S. Miller authored
Alexander Aring says: ==================== net: sched: act: add extack support this patch series adds extack support for the TC action subsystem. As example I for the extack support in a TC action I choosed mirred action. - Alex Cc: David Ahern <dsahern@gmail.com> changes since v3: - adapt recommended changes from Davide Caratti, please check if I catch everything. Thanks. changes since v2: - remove newline in extack of generic walker handling Thanks to Davide Caratti - add kernel@mojatatu.com in cc ==================== Signed-off-by: -
Alexander Aring authored
This patch adds extack to tcf_action_init and tcf_action_init_1 functions. These are necessary to make individual extack handling in each act implementation. Based on work by David Ahern <dsahern@gmail.com> Cc: David Ahern <dsahern@gmail.com> Signed-off-by:
Alexander Aring <aring@mojatatu.com> Signed-off-by:
-
Alexander Aring authored
This patch is used by subsequent patches. It fixes code style issues caught by checkpatch. Signed-off-by:
-
Davide Caratti authored
When tca_action_flush() calls the action walk() and gets an error, a successful call to nla_nest_start() is not followed by a call to nla_nest_cancel(). It's harmless, as the skb is freed in the error path - but it's worth to fix this unbalance. Signed-off-by:
Davide Caratti <dcaratti@redhat.com> Signed-off-by:
-
David S. Miller authored
Andrew Lunn says: ==================== net: dsa: mv88e6xxx: Improve PTP access latency PTP needs to retrieve the hardware timestamps from the switch device in a low latency manor. However ethtool -S and bridge fdb show can hold the switch register access mutex for a long time. These patches changes the reading the statistics and the ATU so that the mutex is released and taken again between each statistic or ATU entry. The PTP code can then interleave its access to the hardware, keeping its latency low. ==================== Signed-off-by: -
Andrew Lunn authored
The PTP code needs low latency access to the PTP hardware timestamps. Reading all the ATU entries in one go adds a lot of latency to the PTP code. So take and release the reg_lock mutex for each individual MAC address in the ATU, allowing the PTP thread jump in between. Signed-off-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
Andrew Lunn authored
The PTP code needs low latency access to the PTP hardware timestamps. Reading all the statistics in one go adds a lot of latency to the PTP code. So take and release the reg_lock mutex for each individual statistics, allowing the PTP thread jump in between. Signed-off-by:
-
David S. Miller authored
Jon Maloy says: ==================== tipc: de-generealize topology server The topology server is partially based on a template that is much more generic than what we need. This results in a code that is unnecessarily hard to follow and keeping bug free. We now take the consequence of the fact that we only have one such server in TIPC, - with no prospects for introducing any more, and adapt the code to the specialized task is really is doing. ==================== Signed-off-by: -
Jon Maloy authored
We rename struct tipc_server to struct tipc_topsrv. This reflect its now specialized role as topology server. Accoringly, we change or add function prefixes to make it clearer which functionality those belong to. There are no functional changes in this commit. Acked-by:
Ying.Xue <ying.xue@windriver.com> Signed-off-by:
Jon Maloy <jon.maloy@ericsson.com> Signed-off-by:
-
Jon Maloy authored
We move the listener socket to struct tipc_server and give it its own work item. This makes it easier to follow the code, and entails some simplifications in the reception code in subscriber sockets. Acked-by:
-
Jon Maloy authored
In order to narrow the interface and dependencies between the topology server and the subscription/binding table functionality we move struct tipc_server inside the file server.c. This requires some code adaptations in other files, but those are mostly minor. The most important change is that we have to move the start/stop functions for the topology server to server.c, where they logically belong anyway. Acked-by:
-
Jon Maloy authored
Since we now have removed struct tipc_subscriber from the code, and only struct tipc_subscription remains, there is no longer need for long and awkward prefixes to distinguish between their pertaining functions. We now change all tipc_subscrp_* prefixes to tipc_sub_*. This is a purely cosmetic change. Acked-by:
-
Jon Maloy authored
After the previous changes it becomes logical to collapse the two-level creation of subscription instances into one. We do that here. We also rename the creation and deletion functions for more consistency. Acked-by:
-
Jon Maloy authored
Because of the requirement for total distribution transparency, users send subscriptions and receive topology events in their own host format. It is up to the topology server to determine this format and do the correct conversions to and from its own host format when needed. Until now, this has been handled in a rather non-transparent way inside the topology server and subscriber code, leading to unnecessary complexity when creating subscriptions and issuing events. We now improve this situation by adding two new macros, tipc_sub_read() and tipc_evt_write(). Both those functions calculate the need for conversion internally before performing their respective operations. Hence, all handling of such conversions become transparent to the rest of the code. Acked-by:
-
Jon Maloy authored
The message transmission and reception in the topology server is more generic than is currently necessary. By basing the funtionality on the fact that we only send items of type struct tipc_event and always receive items of struct tipc_subcr we can make several simplifications, and also get rid of some unnecessary dynamic memory allocations. Acked-by:
-
Jon Maloy authored
It is unnecessary to keep two structures, struct tipc_conn and struct tipc_subscriber, with a one-to-one relationship and still with different life cycles. The fact that the two often run in different contexts, and still may access each other via direct pointers constitutes an additional hazard, something we have experienced at several occasions, and still see happening. We have identified at least two remaining problems that are easier to fix if we simplify the topology server data structure somewhat. - When there is a race between a subscription up/down event and a timeout event, it is fully possible that the former might be delivered after the latter, leading to confusion for the receiver. - The function tipc_subcrp_timeout() is executing in interrupt context, while the following call chain is at least theoretically possible: tipc_subscrp_timeout() tipc_subscrp_send_event() tipc_conn_sendmsg() conn_put() tipc_conn_kref_release() sock_release(sock) I.e., we end up calling a function that might try to sleep in interrupt context. To eliminate this, we need to ensure that the tipc_conn structure and the socket, as well as the subscription instances, only are deleted in work queue context, i.e., after the timeout event really has been sent out. We now remove this unnecessary complexity, by merging data and functionality of the subscriber structure into struct tipc_conn and the associated file server.c. We thereafter add a spinlock and a new 'inactive' state to the subscription structure. Using those, both problems described above can be easily solved. Acked-by: -
Jon Maloy authored
Interaction between the functionality in server.c and subscr.c is done via function pointers installed in struct server. This makes the code harder to follow, and doesn't serve any obvious purpose. Here, we replace the function pointers with direct function calls. Acked-by:
-
Jon Maloy authored
The socket handling in the topology server is unnecessarily generic. It is prepared to handle both SOCK_RDM, SOCK_DGRAM and SOCK_STREAM type sockets, as well as the only socket type which is really used, SOCK_SEQPACKET. We now remove this redundant code to make the code more readable. Acked-by:
-
- 15 Feb, 2018 13 commits
-
-
David S. Miller authored
Merge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next Johan Hedberg says: ==================== pull request: bluetooth-next 2018-02-15 Here's the first bluetooth-next pull request targetting the 4.17 kernel release. - Fixes & cleanups to Atheros and Marvell drivers - Support for two new Realtek controllers - Support for new Intel Bluetooth controller - Fix for supporting multiple slave-role Bluetooth LE connections ==================== Signed-off-by:
-
Prashant Bhole authored
eBPF test fails due to verifier failure because log_buf is too small. Fixed by increasing log_buf size Signed-off-by:
Prashant Bhole <bhole_prashant_q7@lab.ntt.co.jp> Acked-by:
-
David Ahern authored
Remove rt_table_id from rtable. It was added for getroute to return the table id that was hit in the lookup. With the changes for fibmatch the table id can be extracted from the fib_info returned in the fib_result so it no longer needs to be in rtable directly. Signed-off-by:
David Ahern <dsahern@gmail.com> Signed-off-by:
-
David S. Miller authored
Brenda J. Butler says: ==================== tools: tc-testing: Plugin Architecture To make tdc.py more general, we are introducing a plugin architecture. This patch set first organizes the command line parameters, then introduces the plugin architecture and some example plugins. ==================== Signed-off-by: -
Brenda J. Butler authored
Signed-off-by:
Brenda J. Butler <bjb@mojatatu.com> Acked-by:
Lucas Bates <lucasb@mojatatu.com> Signed-off-by:
-
Brenda J. Butler authored
Run the command under test under valgrind. Produce an extra set of tap output for the memory check on each test. Signed-off-by:
-
Brenda J. Butler authored
Move the functionality of creating a namespace before the test suite and destroying it afterwards to a plugin. Signed-off-by:
-
Brenda J. Butler authored
Move the functionality that checks for root permissions into a plugin. Signed-off-by:
-
Brenda J. Butler authored
This should be a general test architecture, and yet allow specific tests to be done. Introduce a plugin architecture. An individual test has 4 stages, setup/execute/verify/teardown. Each plugin gets a chance to run a function at each stage, plus one call before all the tests are called ("pre" suite) and one after all the tests are called ("post" suite). In addition, just before each command is executed, the plugin gets a chance to modify the command using the "adjust_command" hook. This makes the test suite quite flexible. Future patches will take some functionality out of the tdc.py script and place it in plugins. To use the plugins, place the implementation in the plugins directory and run tdc.py. It will notice the plugins and use them. Signed-off-by: -
Brenda J. Butler authored
Split the test_runner function into the loop part (test_runner) and the contents (run_one_test) for maintainability. It makes it a little easier to catch exceptions in an individual test, and keep going (and flush a bunch of tap results for the skipped tests). Signed-off-by:
-
Brenda J. Butler authored
Separate the functionality of the command line parameters into "selection" parameters, "action" parameters and other parameters. "Selection" parameters are for choosing which tests on which to act. "Action" parameters are for choosing what to do with the selected tests. "Other" parameters are for global effect (like "help" or "verbose"). With this commit, we add the ability to name a directory as another selection mechanism. We can accumulate a number of tests by directory, file, category, or even by test id, instead of being constrained to run all tests in one collection or just one test. Signed-off-by:
-
David S. Miller authored
Kirill Tkhai says: ==================== net: Add ioctl() SIOCGSKNS cmd to allow obtaining net ns of tun device Currently, it's not possible to get or check net namespace, which was used to create tun socket. User may have two tun devices with the same names in different nets, and there is no way to differ them each other. The patchset adds support for ioctl() cmd SIOCGSKNS for tun devices. It will allow people to obtain net namespace file descriptor like we allow to do that for sockets in general. v2: Add new patch [2/3] to export open_related_ns(). ==================== Signed-off-by:
Kirill Tkhai <ktkhai@virtuozzo.com> Signed-off-by:
-
Kirill Tkhai authored
This patch adds possibility to get tun device's net namespace fd in the same way we allow to do that for sockets. Socket ioctl numbers do not intersect with tun-specific, and there is already SIOCSIFHWADDR used in tun code. So, SIOCGSKNS number is choosen instead of custom-made for this functionality. Note, that open_related_ns() uses plain get_net_ns() and it's safe (net can't be already dead at this moment): tun socket is allocated via sk_alloc() with zero last arg (kern = 0). So, each alive socket increments net::count, and the socket is definitely alive during ioctl syscall. Also, common variable net is introduced, so small cleanup in TUNSETIFF is made. Signed-off-by:
-
