- 21 1月, 2017 10 次提交
-
-
由 Jon Paul Maloy 提交于
As a preparation for the 'replicast' functionality we are going to introduce in the next commits, we need the broadcast base structure to store whether bearer broadcast is available at all from the currently used bearer or bearers. We do this by adding a new function tipc_bearer_bcast_support() to the bearer layer, and letting the bearer selection function in bcast.c use this to give a new boolean field, 'bcast_support' the appropriate value. Reviewed-by: NParthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com> Acked-by: NYing Xue <ying.xue@windriver.com> Signed-off-by: NJon Maloy <jon.maloy@ericsson.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Gianluca Borello 提交于
Provide a simple helper with the same semantics of strncpy_from_unsafe(): int bpf_probe_read_str(void *dst, int size, const void *unsafe_addr) This gives more flexibility to a bpf program. A typical use case is intercepting a file name during sys_open(). The current approach is: SEC("kprobe/sys_open") void bpf_sys_open(struct pt_regs *ctx) { char buf[PATHLEN]; // PATHLEN is defined to 256 bpf_probe_read(buf, sizeof(buf), ctx->di); /* consume buf */ } This is suboptimal because the size of the string needs to be estimated at compile time, causing more memory to be copied than often necessary, and can become more problematic if further processing on buf is done, for example by pushing it to userspace via bpf_perf_event_output(), since the real length of the string is unknown and the entire buffer must be copied (and defining an unrolled strnlen() inside the bpf program is a very inefficient and unfeasible approach). With the new helper, the code can easily operate on the actual string length rather than the buffer size: SEC("kprobe/sys_open") void bpf_sys_open(struct pt_regs *ctx) { char buf[PATHLEN]; // PATHLEN is defined to 256 int res = bpf_probe_read_str(buf, sizeof(buf), ctx->di); /* consume buf, for example push it to userspace via * bpf_perf_event_output(), but this time we can use * res (the string length) as event size, after checking * its boundaries. */ } Another useful use case is when parsing individual process arguments or individual environment variables navigating current->mm->arg_start and current->mm->env_start: using this helper and the return value, one can quickly iterate at the right offset of the memory area. The code changes simply leverage the already existent strncpy_from_unsafe() kernel function, which is safe to be called from a bpf program as it is used in bpf_trace_printk(). Signed-off-by: NGianluca Borello <g.borello@gmail.com> Acked-by: NAlexei Starovoitov <ast@kernel.org> Acked-by: NDaniel Borkmann <daniel@iogearbox.net> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 David S. Miller 提交于
Phil Sutter says: ==================== Retrieve number of VFs in a bus-agnostic way Previously, it was assumed that only PCI NICs would be capable of having virtual functions - with my proposed enhancement of dummy NIC driver implementing (fake) ones for testing purposes, this is no longer true. Discussion of said patch has led to the suggestion of implementing a bus-agnostic method for VF count retrieval so rtnetlink could work with both real VF-capable PCI NICs as well as my dummy modifications without introducing ugly hacks. The following series tries to achieve just that by introducing a bus type callback to retrieve a device's number of VFs, implementing this callback for PCI bus and finally adjusting rtnetlink to make use of the generalized infrastructure. ==================== Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Phil Sutter 提交于
Now that pci_bus_type has num_vf callback set, dev_num_vf can be implemented in a bus type independent way and the check for whether a PCI device is being handled in rtnetlink can be dropped. Signed-off-by: NPhil Sutter <phil@nwl.cc> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Phil Sutter 提交于
Signed-off-by: NPhil Sutter <phil@nwl.cc> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Phil Sutter 提交于
This allows for bus types to implement their own method of retrieving the number of virtual functions a NIC on that type of bus supports. Signed-off-by: NPhil Sutter <phil@nwl.cc> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Geliang Tang 提交于
Use hlist_entry_safe() instead of open-coding it. Signed-off-by: NGeliang Tang <geliangtang@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jakub Sitnicki 提交于
Commit b05229f4 ("gre6: Cleanup GREv6 transmit path, call common GRE functions") removed the ip6gre specific transmit function, but left the struct ipv6_tel_txoption definition. Clean it up. Signed-off-by: NJakub Sitnicki <jkbs@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Eric Dumazet 提交于
Shaohua Li made percpu_counter irq safe in commit 098faf58 ("percpu_counter: make APIs irq safe") We can safely remove BH disable/enable sections around various percpu_counter manipulations. Signed-off-by: NEric Dumazet <edumazet@google.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Arnd Bergmann 提交于
The two new variables are only used inside of an #ifdef and cause harmless warnings when that is disabled: drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c: In function 'init_one': drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c:4646:9: error: unused variable 'port_vec' [-Werror=unused-variable] drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c:4646:6: error: unused variable 'v' [-Werror=unused-variable] This adds another #ifdef around the declarations. Fixes: 96fe11f2 ("cxgb4: Implement ndo_get_phys_port_id for mgmt dev") Signed-off-by: NArnd Bergmann <arnd@arndb.de> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
- 20 1月, 2017 7 次提交
-
-
由 David Ahern 提交于
IPv6 deletes route entries associated with multipath routes on an admin down where IPv4 does not. For example: $ ip ro ls vrf red unreachable default metric 8192 1.1.1.0/24 metric 64 nexthop via 10.100.1.254 dev eth1 weight 1 nexthop via 10.100.2.254 dev eth2 weight 1 10.100.1.0/24 dev eth1 proto kernel scope link src 10.100.1.4 10.100.2.0/24 dev eth2 proto kernel scope link src 10.100.2.4 $ ip -6 ro ls vrf red 2001:db8:1::/120 dev eth1 proto kernel metric 256 pref medium 2001:db8:2:: dev red proto none metric 0 pref medium 2001:db8:2::/120 dev eth2 proto kernel metric 256 pref medium 2001:db8:11::/120 via 2001:db8:1::16 dev eth1 metric 1024 pref medium 2001:db8:11::/120 via 2001:db8:2::17 dev eth2 metric 1024 pref medium ... Set link down: $ ip li set eth1 down IPv4 retains the multihop route but flags eth1 route as dead: $ ip ro ls vrf red unreachable default metric 8192 1.1.1.0/24 nexthop via 10.100.1.16 dev eth1 weight 1 dead linkdown nexthop via 10.100.2.16 dev eth2 weight 1 10.100.2.0/24 dev eth2 proto kernel scope link src 10.100.2.4 and IPv6 deletes the route as part of flushing all routes for the device: $ ip -6 ro ls vrf red 2001:db8:2:: dev red proto none metric 0 pref medium 2001:db8:2::/120 dev eth2 proto kernel metric 256 pref medium 2001:db8:11::/120 via 2001:db8:2::17 dev eth2 metric 1024 pref medium ... Worse, on admin up of the device the multipath route has to be deleted to get this leg of the route re-added. This patch keeps routes that are part of a multipath route if ignore_routes_with_linkdown is set with the dead and linkdown flags enabling consistency between IPv4 and IPv6: $ ip -6 ro ls vrf red 2001:db8:2:: dev red proto none metric 0 pref medium 2001:db8:2::/120 dev eth2 proto kernel metric 256 pref medium 2001:db8:11::/120 via 2001:db8:1::16 dev eth1 metric 1024 dead linkdown pref medium 2001:db8:11::/120 via 2001:db8:2::17 dev eth2 metric 1024 pref medium ... Signed-off-by: NDavid Ahern <dsa@cumulusnetworks.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Eric Dumazet 提交于
Commit 04aeb56a ("net/mlx4_en: allocate non 0-order pages for RX ring with __GFP_NOMEMALLOC") added code that appears to be not needed at that time, since mlx4 never used __GFP_MEMALLOC allocations anyway. As using memory reserves is a must in some situations (swap over NFS or iSCSI), this patch adds this flag. Note that this driver does not reuse pages (yet) so we do not have to add anything else. Signed-off-by: NEric Dumazet <edumazet@google.com> Cc: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Cc: Tariq Toukan <tariqt@mellanox.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Timur Tabi 提交于
This reverts commit 3e884493. With commit 529ed127 ("net: phy: phy drivers should not set SUPPORTED_[Asym_]Pause"), phylib now handles automatically enabling pause frame support in the PHY, and the MAC driver should follow suit. Since the EMAC driver driver does this, we no longer need to force pause frames support. Signed-off-by: NTimur Tabi <timur@codeaurora.org> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Volodymyr Bendiuga 提交于
Some bus names are pretty long and do not fit into 17 chars. Increase therefore MII_BUS_ID_SIZE and phy_fixup.bus_id to larger number. Now mii_bus.id can host larger name. Signed-off-by: NVolodymyr Bendiuga <volodymyr.bendiuga@gmail.com> Signed-off-by: NMagnus Öberg <magnus.oberg@westermo.se> Reviewed-by: NAndrew Lunn <andrew@lunn.ch> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
This patch does the following: - MACB/GEM-PTP interface - registers and bitfields for TSU - capability flags to enable PTP per platform basis Signed-off-by: NAndrei Pistirica <andrei.pistirica@microchip.com> Acked-by: NNicolas Ferre <nicolas.ferre@atmel.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Tobias Klauser 提交于
The stats member of struct chnl_net is used nowhere in the code, so it might as well be removed. Signed-off-by: NTobias Klauser <tklauser@distanz.ch> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Arnd Bergmann 提交于
The new ARP support has pushed the stack size over the edge on ARM, as there are two large objects on the stack in this function (mask and tb) and both have now grown a bit more: net/sched/cls_flower.c: In function 'fl_change': net/sched/cls_flower.c:928:1: error: the frame size of 1072 bytes is larger than 1024 bytes [-Werror=frame-larger-than=] We can solve this by dynamically allocating one or both of them. I first tried to do it just for the mask, but that only saved 152 bytes on ARM, while this version just does it for the 'tb' array, bringing the stack size back down to 664 bytes. Fixes: 99d31326 ("net/sched: cls_flower: Support matching on ARP") Signed-off-by: NArnd Bergmann <arnd@arndb.de> Acked-by: NJiri Pirko <jiri@mellanox.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
- 19 1月, 2017 23 次提交
-
-
由 Tobias Klauser 提交于
The network stack no longer uses the last_rx member of struct net_device since the bonding driver switched to use its own private last_rx in commit 9f242738 ("bonding: use last_arp_rx in slave_last_rx()"). However, some drivers still (ab)use the field for their own purposes and some driver just update it without actually using it. Previously, there was an accompanying comment for the last_rx member added in commit 4dc89133 ("net: add a comment on netdev->last_rx") which asked drivers not to update is, unless really needed. However, this commend was removed in commit f8ff080d ("bonding: remove useless updating of slave->dev->last_rx"), so some drivers added later on still did update last_rx. Remove all usage of last_rx and switch three drivers (sky2, atp and smc91c92_cs) which actually read and write it to use their own private copy in netdev_priv. Compile-tested with allyesconfig and allmodconfig on x86 and arm. Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Jay Vosburgh <j.vosburgh@gmail.com> Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <andy@greyhouse.net> Cc: Mirko Lindner <mlindner@marvell.com> Cc: Stephen Hemminger <stephen@networkplumber.org> Signed-off-by: NTobias Klauser <tklauser@distanz.ch> Acked-by: NEric Dumazet <edumazet@google.com> Reviewed-by: NJay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Vivien Didelot 提交于
Now that the DSA Ethernet switches are true Linux devices, the CPU switch is not necessarily the first one. If its address is higher than the second switch on the same MDIO bus, its index will be 1, not 0. Avoid any confusion by using dst->cpu_switch instead of dst->ds[0]. Signed-off-by: NVivien Didelot <vivien.didelot@savoirfairelinux.com> Reviewed-by: NAndrew Lunn <andrew@lunn.ch> Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com> Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Vivien Didelot 提交于
Store a dsa_switch pointer to the CPU switch in the tree instead of only its index. This avoids the need to initialize it to -1. Signed-off-by: NVivien Didelot <vivien.didelot@savoirfairelinux.com> Reviewed-by: NAndrew Lunn <andrew@lunn.ch> Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Ivan Khoronzhuk 提交于
Check "ch" on NULL first, then get ctlr. Signed-off-by: NIvan Khoronzhuk <ivan.khoronzhuk@linaro.org> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 David S. Miller 提交于
Jason Wang says: ==================== vhost_net tx batching This series tries to implement tx batching support for vhost. This was done by using MSG_MORE as a hint for under layer socket. The backend (e.g tap) can then batch the packets temporarily in a list and submit it all once the number of bacthed exceeds a limitation. Tests shows obvious improvement on guest pktgen over over mlx4(noqueue) on host: Mpps -+% rx-frames = 0 0.91 +0% rx-frames = 4 1.00 +9.8% rx-frames = 8 1.00 +9.8% rx-frames = 16 1.01 +10.9% rx-frames = 32 1.07 +17.5% rx-frames = 48 1.07 +17.5% rx-frames = 64 1.08 +18.6% rx-frames = 64 (no MSG_MORE) 0.91 +0% Changes from V4: - stick to NAPI_POLL_WEIGHT for rx-frames is user specify a value greater than it. Changes from V3: - use ethtool instead of module parameter to control the maximum number of batched packets - avoid overhead when MSG_MORE were not set and no packet queued Changes from V2: - remove uselss queue limitation check (and we don't drop any packet now) Changes from V1: - drop NAPI handler since we don't use NAPI now - fix the issues that may exceeds max pending of zerocopy - more improvement on available buffer detection - move the limitation of batched pacekts from vhost to tuntap ==================== Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jason Wang 提交于
We can only process 1 packet at one time during sendmsg(). This often lead bad cache utilization under heavy load. So this patch tries to do some batching during rx before submitting them to host network stack. This is done through accepting MSG_MORE as a hint from sendmsg() caller, if it was set, batch the packet temporarily in a linked list and submit them all once MSG_MORE were cleared. Tests were done by pktgen (burst=128) in guest over mlx4(noqueue) on host: Mpps -+% rx-frames = 0 0.91 +0% rx-frames = 4 1.00 +9.8% rx-frames = 8 1.00 +9.8% rx-frames = 16 1.01 +10.9% rx-frames = 32 1.07 +17.5% rx-frames = 48 1.07 +17.5% rx-frames = 64 1.08 +18.6% rx-frames = 64 (no MSG_MORE) 0.91 +0% User were allowed to change per device batched packets through ethtool -C rx-frames. NAPI_POLL_WEIGHT were used as upper limitation to prevent bh from being disabled too long. Signed-off-by: NJason Wang <jasowang@redhat.com> Acked-by: NMichael S. Tsirkin <mst@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jason Wang 提交于
This patch tries to utilize tuntap rx batching by peeking the tx virtqueue during transmission, if there's more available buffers in the virtqueue, set MSG_MORE flag for a hint for backend (e.g tuntap) to batch the packets. Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com> Signed-off-by: NJason Wang <jasowang@redhat.com> Acked-by: NMichael S. Tsirkin <mst@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jason Wang 提交于
This patch tries to do several tweaks on vhost_vq_avail_empty() for a better performance: - check cached avail index first which could avoid userspace memory access. - using unlikely() for the failure of userspace access - check vq->last_avail_idx instead of cached avail index as the last step. This patch is need for batching supports which needs to peek whether or not there's still available buffers in the ring. Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com> Signed-off-by: NJason Wang <jasowang@redhat.com> Acked-by: NMichael S. Tsirkin <mst@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Mao Wenan 提交于
Relax ordering(RO) is one feature of 82599 NIC, to enable this feature can enhance the performance for some cpu architecure, such as SPARC and so on. Currently it only supports one special cpu architecture(SPARC) in 82599 driver to enable RO feature, this is not very common for other cpu architecture which really needs RO feature. This patch add one common config CONFIG_ARCH_WANT_RELAX_ORDER to set RO feature, and should define CONFIG_ARCH_WANT_RELAX_ORDER in sparc Kconfig firstly. Signed-off-by: NMao Wenan <maowenan@huawei.com> Reviewed-by: NAlexander Duyck <alexander.duyck@gmail.com> Reviewed-by: NAlexander Duyck <alexander.h.duyck@intel.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 David S. Miller 提交于
David Ahern says: ==================== net: ipv6: simplify rt6_fill_node Remove a couple of unnecessary input arguments to rt6_fill_node. ==================== Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 David Ahern 提交于
The prefix arg to rt6_fill_node is non-0 in only 1 path - rt6_dump_route where a user is requesting a prefix only dump. Simplify rt6_fill_node by removing the prefix arg and moving the prefix check to rt6_dump_route. Signed-off-by: NDavid Ahern <dsa@cumulusnetworks.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 David Ahern 提交于
All callers of rt6_fill_node pass 0 for nowait arg. Remove the arg and simplify rt6_fill_node accordingly. rt6_fill_node passes the nowait of 0 to ip6mr_get_route. Remove the nowait arg from it as well. Signed-off-by: NDavid Ahern <dsa@cumulusnetworks.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 David S. Miller 提交于
Xin Long says: ==================== sctp: add sender-side procedures for stream reconf ssn reset request chunk Patch 6/6 is to implement sender-side procedures for the Outgoing and Incoming SSN Reset Request Parameter described in rfc6525 section 5.1.2 and 5.1.3 Patches 1-5/6 are ahead of it to define some apis and asoc members for it. Note that with this patchset, asoc->reconf_enable has no chance yet to be set, until the patch "sctp: add get and set sockopt for reconf_enable" is applied in the future. As we can not just enable it when sctp is not capable of processing reconf chunk yet. v1->v2: - put these into a smaller group. - rename some temporary variables in the codes. - rename the titles of the commits and improve some changelogs. v2->v3: - re-split the patchset and make sure it has no dead codes for review. v3->v4: - move sctp_make_reconf() into patch 1/6 to avoid kbuild warning. - drop unused struct sctp_strreset_req. ==================== Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Xin Long 提交于
This patch is to implement sender-side procedures for the Outgoing and Incoming SSN Reset Request Parameter described in rfc6525 section 5.1.2 and 5.1.3. It is also add sockopt SCTP_RESET_STREAMS in rfc6525 section 6.3.2 for users. Note that the new asoc member strreset_outstanding is to make sure only one reconf request chunk on the fly as rfc6525 section 5.1.1 demands. Signed-off-by: NXin Long <lucien.xin@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Xin Long 提交于
This patch is to add sockopt SCTP_ENABLE_STREAM_RESET to get/set strreset_enable to indicate which reconf request type it supports, which is described in rfc6525 section 6.3.1. Signed-off-by: NXin Long <lucien.xin@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Xin Long 提交于
This patch is to add reconf_enable field in all of asoc ep and netns to indicate if they support stream reset. When initializing, asoc reconf_enable get the default value from ep reconf_enable which is from netns netns reconf_enable by default. It is also to add reconf_capable in asoc peer part to know if peer supports reconf_enable, the value is set if ext params have reconf chunk support when processing init chunk, just as rfc6525 section 5.1.1 demands. Signed-off-by: NXin Long <lucien.xin@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Xin Long 提交于
This patch is to add a primitive based on sctp primitive frame for sending stream reconf request. It works as the other primitives, and create a SCTP_CMD_REPLY command to send the request chunk out. sctp_primitive_RECONF would be the api to send a reconf request chunk. Signed-off-by: NXin Long <lucien.xin@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Xin Long 提交于
This patch is to add a per transport timer based on sctp timer frame for stream reconf chunk retransmission. It would start after sending a reconf request chunk, and stop after receiving the response chunk. If the timer expires, besides retransmitting the reconf request chunk, it would also do the same thing with data RTO timer. like to increase the appropriate error counts, and perform threshold management, possibly destroying the asoc if sctp retransmission thresholds are exceeded, just as section 5.1.1 describes. This patch is also to add asoc strreset_chunk, it is used to save the reconf request chunk, so that it can be retransmitted, and to check if the response is really for this request by comparing the information inside with the response chunk as well. Signed-off-by: NXin Long <lucien.xin@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Xin Long 提交于
This patch is to add asoc strreset_outseq and strreset_inseq for saving the reconf request sequence, initialize them when create assoc and process init, and also to define Incoming and Outgoing SSN Reset Request Parameter described in rfc6525 section 4.1 and 4.2, As they can be in one same chunk as section rfc6525 3.1-3 describes, it makes them in one function. Signed-off-by: NXin Long <lucien.xin@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 David S. Miller 提交于
Josef Bacik says: ==================== Rework inet_csk_get_port V3->V4: -Removed the random include of addrconf.h that is no longer needed. V2->V3: -Dropped the fastsock from the tb and instead just carry the saddrs, family, and ipv6 only flag. -Reworked the helper functions to deal with this change so I could still use them when checking the fast path. -Killed tb->num_owners as per Eric's request. -Attached a reproducer to the bottom of this email. V1->V2: -Added a new patch 'inet: collapse ipv4/v6 rcv_saddr_equal functions into one' at Hannes' suggestion. -Dropped ->bind_conflict and just use the new helper. -Fixed a compile bug from the original ->bind_conflict patch. The original description of the series follows: At some point recently the guys working on our load balancer added the ability to use SO_REUSEPORT. When they restarted their app with this option enabled they immediately hit a softlockup on what appeared to be the inet_bind_bucket->lock. Eventually what all of our debugging and discussion led us to was the fact that the application comes up without SO_REUSEPORT, shuts down which creates around 100k twsk's, and then comes up and tries to open a bunch of sockets using SO_REUSEPORT, which meant traversing the inet_bind_bucket owners list under the lock. Since this lock is needed for dealing with the twsk's and basically anything else related to connections we would softlockup, and sometimes not ever recover. To solve this problem I did what you see in Path 5/5. Once we have a SO_REUSEPORT socket on the tb->owners list we know that the socket has no conflicts with any of the other sockets on that list. So we can add a copy of the sock_common (really all we need is the recv_saddr but it seemed ugly to copy just the ipv6, ipv4, and flag to indicate if we were ipv6 only in there so I've copied the whole common) in order to check subsequent SO_REUSEPORT sockets. If they match the previous one then we can skip the expensive inet_csk_bind_conflict check. This is what eliminated the soft lockup that we were seeing. Patches 1-4 are cleanups and re-workings. For instance when we specify port == 0 we need to find an open port, but we would do two passes through inet_csk_bind_conflict every time we found a possible port. We would also keep track of the smallest_port value in order to try and use it if we found no port our first run through. This however made no sense as it would have had to fail the first pass through inet_csk_bind_conflict, so would not actually pass the second pass through either. Finally I split the function into two functions in order to make it easier to read and to distinguish between the two behaviors. I have tested this on one of our load balancing boxes during peak traffic and it hasn't fallen over. But this is not my area, so obviously feel free to point out where I'm being stupid and I'll get it fixed up and retested. Thanks, ==================== Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Josef Bacik 提交于
If we have non reuseport sockets on a tb we will set tb->fastreuseport to 0 and never set it again. Which means that in the future if we end up adding a bunch of reuseport sk's to that tb we'll have to do the expensive scan every time. Instead add the ipv4/ipv6 saddr fields to the bind bucket, as well as the family so we know what comparison to make, and the ipv6 only setting so we can make sure to compare with new sockets appropriately. Once one sk has made it onto the list we know that there are no potential bind conflicts on the owners list that match that sk's rcv_addr. So copy the sk's information into our bind bucket and set tb->fastruseport to FASTREUSESOCK_STRICT so we know we have to do an extra check for subsequent reuseport sockets and skip the expensive bind conflict check. Signed-off-by: NJosef Bacik <jbacik@fb.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Josef Bacik 提交于
inet_csk_get_port does two different things, it either scans for an open port, or it tries to see if the specified port is available for use. Since these two operations have different rules and are basically independent lets split them into two different functions to make them both more readable. Signed-off-by: NJosef Bacik <jbacik@fb.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Josef Bacik 提交于
This is just wasted time, we've already found a tb that doesn't have a bind conflict, and we don't drop the head lock so scanning again isn't going to give us a different answer. Instead move the tb->reuse setting logic outside of the found_tb path and put it in the success: path. Then make it so that we don't goto again if we find a bind conflict in the found_tb path as we won't reach this anymore when we are scanning for an ephemeral port. Signed-off-by: NJosef Bacik <jbacik@fb.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-