- 24 Jun, 2019 26 commits
-
-
David S. Miller authored
Stefano Brivio says: ==================== Fix listing (IPv4, IPv6) and flushing (IPv6) of cached route exceptions For IPv6 cached routes, the commands 'ip -6 route list cache' and 'ip -6 route flush cache' don't work at all after route exceptions have been moved to a separate hash table in commit 2b760fcf ("ipv6: hook up exception table to store dst cache"). For IPv4 cached routes, the command 'ip route list cache' has also stopped working in kernel 3.5 after commit 4895c771 ("ipv4: Add FIB nexthop exceptions.") introduced storage for route exceptions as a separate entity. Fix this by allowing userspace to clearly request cached routes with the RTM_F_CLONED flag used as a filter (in conjuction with strict checking) and by retrieving and dumping cached routes if requested. If strict checking is not requested (iproute2 < 5.0.0), we don't have a way to consistently filter results on other selectors (e.g. on tables), so skip filtering entirely and dump both regular routes and exceptions. For IPv4, cache flushing uses a completely different mechanism, so it wasn't affected. Listing of exception routes (modified routes pre-3.5) was tested against these versions of kernel and iproute2: iproute2 kernel 4.14.0 4.15.0 4.19.0 5.0.0 5.1.0 3.5-rc4 + + + + + 4.4 4.9 4.14 4.15 4.19 5.0 5.1 fixed + + + + + For IPv6, a separate iproute2 patch is required. Versions of iproute2 and kernel tested: iproute2 kernel 4.14.0 4.15.0 4.19.0 5.0.0 5.1.0 5.1.0, patched 3.18 list + + + + + + flush + + + + + + 4.4 list + + + + + + flush + + + + + + 4.9 list + + + + + + flush + + + + + + 4.14 list + + + + + + flush + + + + + + 4.15 list flush 4.19 list flush 5.0 list flush 5.1 list flush with list + + + + + + fix flush + + + + v7: Make sure r->rtm_tos is initialised in 3/11, move loop over nexthop objects in 4/11, add comments about usage of "skip" counters in commit messages of 4/11 and 8/11 v6: Target for net-next, rebase and adapt to nexthop objects for IPv6 paths. Merge selftests into this series (as they were addressed for net-next). A number of minor changes detailed in logs of single patches. v5: Skip filtering altogether if no strict checking is requested: selecting routes or exceptions only would be inconsistent with the fact we can't filter on tables. Drop 1/8 (non-strict dump filter function no longer needed), replace 2/8 (don't use NLM_F_MATCH, decide to skip routes or exceptions in filter function), drop 6/8 (2/8 is enough for IPv6 too). Introduce dump_routes and dump_exceptions flags in filter, adapt other patches to that. v4: Fix the listing issue also for IPv4, making the behaviour consistent with IPv6. Honour NLM_F_MATCH as per RFC 3549 and allow usage of RTM_F_CLONED filter. Split patches into smaller logical changes. v3: Drop check on RTM_F_CLONED and rework logic of return values of rt6_dump_route() v2: Add count of routes handled in partial dumps, and skip them, in patch 1/2. ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Stefano Brivio authored
Instead of just listing and flushing two cached exceptions, create a relatively big number of them, and count how many are listed. Single netlink dump messages contain approximately 25 entries each, and this way we can make sure the partial dump tracking mechanism is working properly. While at it, also ensure that no cached routes can be listed after flush, and remove 'sleep 1' calls, they are not actually needed. v7: No changes v6: - Merge this patch into series including fix, as it's also targeted for net-next. No actual changes Signed-off-by:Stefano Brivio <sbrivio@redhat.com> Signed-off-by:
-
Stefano Brivio authored
This test checks that route exceptions can be successfully listed and flushed using ip -6 route {list,flush} cache. v7: No changes v6: - Merge this patch into series including fix, as it's also targeted for net-next - Drop left-over print of 'ip route list cache | wc -l' Signed-off-by: -
Stefano Brivio authored
When we perform an inexact match on FIB nodes via fib6_locate_1(), longer prefixes will be preferred to shorter ones. However, it might happen that a node, with higher fn_bit value than some other, has no valid routing information. In this case, we'll pick that node, but it will be discarded by the check on RTN_RTINFO in fib6_locate(), and we might miss nodes with valid routing information but with lower fn_bit value. This is apparent when a routing exception is created for a default route: # ip -6 route list fc00:1::/64 dev veth_A-R1 proto kernel metric 256 pref medium fc00:2::/64 dev veth_A-R2 proto kernel metric 256 pref medium fc00:4::1 via fc00:2::2 dev veth_A-R2 metric 1024 pref medium fe80::/64 dev veth_A-R1 proto kernel metric 256 pref medium fe80::/64 dev veth_A-R2 proto kernel metric 256 pref medium default via fc00:1::2 dev veth_A-R1 metric 1024 pref medium # ip -6 route list cache fc00:4::1 via fc00:2::2 dev veth_A-R2 metric 1024 expires 593sec mtu 1500 pref medium fc00:3::1 via fc00:1::2 dev veth_A-R1 metric 1024 expires 593sec mtu 1500 pref medium # ip -6 route flush cache # node for default route is discarded Failed to send flush request: No such process # ip -6 route list cache fc00:3::1 via fc00:1::2 dev veth_A-R1 metric 1024 expires 586sec mtu 1500 pref medium Check right away if the node has a RTN_RTINFO flag, before replacing the 'prev' pointer, that indicates the longest matching prefix found so far. Fixes: 38fbeeee ("ipv6: prepare fib6_locate() for exception table") Signed-off-by:
-
Stefano Brivio authored
Since commit 2b760fcf ("ipv6: hook up exception table to store dst cache"), route exceptions reside in a separate hash table, and won't be found by walking the FIB, so they won't be dumped to userspace on a RTM_GETROUTE message. This causes 'ip -6 route list cache' and 'ip -6 route flush cache' to have no function anymore: # ip -6 route get fc00:3::1 fc00:3::1 via fc00:1::2 dev veth_A-R1 src fc00:1::1 metric 1024 expires 539sec mtu 1400 pref medium # ip -6 route get fc00:4::1 fc00:4::1 via fc00:2::2 dev veth_A-R2 src fc00:2::1 metric 1024 expires 536sec mtu 1500 pref medium # ip -6 route list cache # ip -6 route flush cache # ip -6 route get fc00:3::1 fc00:3::1 via fc00:1::2 dev veth_A-R1 src fc00:1::1 metric 1024 expires 520sec mtu 1400 pref medium # ip -6 route get fc00:4::1 fc00:4::1 via fc00:2::2 dev veth_A-R2 src fc00:2::1 metric 1024 expires 519sec mtu 1500 pref medium because iproute2 lists cached routes using RTM_GETROUTE, and flushes them by listing all the routes, and deleting them with RTM_DELROUTE one by one. If cached routes are requested using the RTM_F_CLONED flag together with strict checking, or if no strict checking is requested (and hence we can't consistently apply filters), look up exceptions in the hash table associated with the current fib6_info in rt6_dump_route(), and, if present and not expired, add them to the dump. We might be unable to dump all the entries for a given node in a single message, so keep track of how many entries were handled for the current node in fib6_walker, and skip that amount in case we start from the same partially dumped node. When a partial dump restarts, as the starting node might change when 'sernum' changes, we have no guarantee that we need to skip the same amount of in-node entries. Therefore, we need two counters, and we need to zero the in-node counter if the node from which the dump is resumed differs. Note that, with the current version of iproute2, this only fixes the 'ip -6 route list cache': on a flush command, iproute2 doesn't pass RTM_F_CLONED and, due to this inconsistency, 'ip -6 route flush cache' is still unable to fetch the routes to be flushed. This will be addressed in a patch for iproute2. To flush cached routes, a procfs entry could be introduced instead: that's how it works for IPv4. We already have a rt6_flush_exception() function ready to be wired to it. However, this would not solve the issue for listing. Versions of iproute2 and kernel tested: iproute2 kernel 4.14.0 4.15.0 4.19.0 5.0.0 5.1.0 5.1.0, patched 3.18 list + + + + + + flush + + + + + + 4.4 list + + + + + + flush + + + + + + 4.9 list + + + + + + flush + + + + + + 4.14 list + + + + + + flush + + + + + + 4.15 list flush 4.19 list flush 5.0 list flush 5.1 list flush with list + + + + + + fix flush + + + + v7: - Explain usage of "skip" counters in commit message (suggested by David Ahern) v6: - Rebase onto net-next, use recently introduced nexthop walker - Make rt6_nh_dump_exceptions() a separate function (suggested by David Ahern) v5: - Use dump_routes and dump_exceptions from filter, ignore NLM_F_MATCH, update test results (flushing works with iproute2 < 5.0.0 now) v4: - Split NLM_F_MATCH and strict check handling in separate patches - Filter routes using RTM_F_CLONED: if it's not set, only return non-cached routes, and if it's set, only return cached routes: change requested by David Ahern and Martin Lau. This implies that iproute2 needs a separate patch to be able to flush IPv6 cached routes. This is not ideal because we can't fix the breakage caused by 2b760fcf entirely in kernel. However, two years have passed since then, and this makes it more tolerable v3: - More descriptive comment about expired exceptions in rt6_dump_route() - Swap return values of rt6_dump_route() (suggested by Martin Lau) - Don't zero skip_in_node in case we don't dump anything in a given pass (also suggested by Martin Lau) - Remove check on RTM_F_CLONED altogether: in the current UAPI semantic, it's just a flag to indicate the route was cloned, not to filter on routes v2: Add tracking of number of entries to be skipped in current node after a partial dump. As we restart from the same node, if not all the exceptions for a given node fit in a single message, the dump will not terminate, as suggested by Martin Lau. This is a concrete possibility, setting up a big number of exceptions for the same route actually causes the issue, suggested by David Ahern. Reported-by:
Jianlin Shi <jishi@redhat.com> Fixes: 2b760fcf ("ipv6: hook up exception table to store dst cache") Signed-off-by:
David Ahern <dsahern@gmail.com> Signed-off-by:
-
Stefano Brivio authored
In the next patch, we are going to add optional dump of exceptions to rt6_dump_route(). Change the return code of rt6_dump_route() to accomodate partial node dumps: we might dump multiple routes per node, and might be able to dump only a given number of them, so fib6_dump_node() will need to know how many routes have been dumped on partial dump, to restart the dump from the point where it was interrupted. Note that fib6_dump_node() is the only caller and already handles all non-negative return codes as success: those become -1 to signal that we're done with the node. If we fail, return 0, as we were unable to dump the single route in the node, but we're not done with it. Signed-off-by:
-
Stefano Brivio authored
If fc_nh_id isn't set, we shouldn't try to match against it. This actually matters just for the RTF_CACHE below (where this case is already handled): if iproute2 gets a route exception and tries to delete it, it won't reference it by fc_nh_id, even if a nexthop object might be associated to the originating route. Fixes: 5b98324e ("ipv6: Allow routes to use nexthop objects") Signed-off-by:
-
Stefano Brivio authored
This reverts commit 08e814c9: as we are preparing to fix listing and dumping of IPv6 cached routes, we need to allow RTM_F_CLONED as a flag to match routes against while dumping them. Signed-off-by:
-
Stefano Brivio authored
Since commit 4895c771 ("ipv4: Add FIB nexthop exceptions."), cached exception routes are stored as a separate entity, so they are not dumped on a FIB dump, even if the RTM_F_CLONED flag is passed. This implies that the command 'ip route list cache' doesn't return any result anymore. If the RTM_F_CLONED is passed, and strict checking requested, retrieve nexthop exception routes and dump them. If no strict checking is requested, filtering can't be performed consistently: dump everything in that case. With this, we need to add an argument to the netlink callback in order to track how many entries were already dumped for the last leaf included in a partial netlink dump. A single additional argument is sufficient, even if we traverse logically nested structures (nexthop objects, hash table buckets, bucket chains): it doesn't matter if we stop in the middle of any of those, because they are always traversed the same way. As an example, s_i values in [], s_fa values in (): node (fa) #1 [1] nexthop #1 bucket #1 -> #0 in chain (1) bucket #2 -> #0 in chain (2) -> #1 in chain (3) -> #2 in chain (4) bucket #3 -> #0 in chain (5) -> #1 in chain (6) nexthop #2 bucket #1 -> #0 in chain (7) -> #1 in chain (8) bucket #2 -> #0 in chain (9) -- node (fa) #2 [2] nexthop #1 bucket #1 -> #0 in chain (1) -> #1 in chain (2) bucket #2 -> #0 in chain (3) it doesn't matter if we stop at (3), (4), (7) for "node #1", or at (2) for "node #2": walking flattens all that. It would even be possible to drop the distinction between the in-tree (s_i) and in-node (s_fa) counter, but a further improvement might advise against this. This is only as accurate as the existing tracking mechanism for leaves: if a partial dump is restarted after exceptions are removed or expired, we might skip some non-dumped entries. To improve this, we could attach a 'sernum' attribute (similar to the one used for IPv6) to nexthop entities, and bump this counter whenever exceptions change: having a distinction between the two counters would make this more convenient. Listing of exception routes (modified routes pre-3.5) was tested against these versions of kernel and iproute2: iproute2 kernel 4.14.0 4.15.0 4.19.0 5.0.0 5.1.0 3.5-rc4 + + + + + 4.4 4.9 4.14 4.15 4.19 5.0 5.1 fixed + + + + + v7: - Move loop over nexthop objects to route.c, and pass struct fib_info and table ID to it, not a struct fib_alias (suggested by David Ahern) - While at it, note that the NULL check on fa->fa_info is redundant, and the check on RTNH_F_DEAD is also not consistent with what's done with regular route listing: just keep it for nhc_flags - Rename entry point function for dumping exceptions to fib_dump_info_fnhe(), and rearrange arguments for consistency with fib_dump_info() - Rename fnhe_dump_buckets() to fnhe_dump_bucket() and make it handle one bucket at a time - Expand commit message to describe why we can have a single "skip" counter for all exceptions stored in bucket chains in nexthop objects (suggested by David Ahern) v6: - Rebased onto net-next - Loop over nexthop paths too. Move loop over fnhe buckets to route.c, avoids need to export rt_fill_info() and to touch exceptions from fib_trie.c. Pass NULL as flow to rt_fill_info(), it now allows that (suggested by David Ahern) Fixes: 4895c771 ("ipv4: Add FIB nexthop exceptions.") Signed-off-by:
-
Stefano Brivio authored
In the next patch, we're going to use rt_fill_info() to dump exception routes upon RTM_GETROUTE with NLM_F_ROOT, meaning userspace is requesting a dump and not a specific route selection, which in turn implies the input interface is not relevant. Update rt_fill_info() to handle a NULL flowinfo. v7: If fl4 is NULL, explicitly set r->rtm_tos to 0: it's not initialised otherwise (spotted by David Ahern) v6: New patch Suggested-by: -
Stefano Brivio authored
This functionally reverts the check introduced by commit e8ba330a ("rtnetlink: Update fib dumps for strict data checking") as modified by commit e4e92fb1 ("net/ipv4: Bail early if user only wants prefix entries"). As we are preparing to fix listing of IPv4 cached routes, we need to give userspace a way to request them. Signed-off-by:
-
Stefano Brivio authored
The following patches add back the ability to dump IPv4 and IPv6 exception routes, and we need to allow selection of regular routes or exceptions. Use RTM_F_CLONED as filter to decide whether to dump routes or exceptions: iproute2 passes it in dump requests (except for IPv6 cache flush requests, this will be fixed in iproute2) and this used to work as long as exceptions were stored directly in the FIB, for both IPv4 and IPv6. Caveat: if strict checking is not requested (that is, if the dump request doesn't go through ip_valid_fib_dump_req()), we can't filter on protocol, tables or route types. In this case, filtering on RTM_F_CLONED would be inconsistent: we would fix 'ip route list cache' by returning exception routes and at the same time introduce another bug in case another selector is present, e.g. on 'ip route list cache table main' we would return all exception routes, without filtering on tables. Keep this consistent by applying no filters at all, and dumping both routes and exceptions, if strict checking is not requested. iproute2 currently filters results anyway, and no unwanted results will be presented to the user. The kernel will just dump more data than needed. v7: No changes v6: Rebase onto net-next, no changes v5: New patch: add dump_routes and dump_exceptions flags in filter and simply clear the unwanted one if strict checking is enabled, don't ignore NLM_F_MATCH and don't set filter_set if NLM_F_MATCH is set. Skip filtering altogether if no strict checking is requested: selecting routes or exceptions only would be inconsistent with the fact we can't filter on tables. Signed-off-by: -
Antoine Tenart authored
This patch updates the macb driver to use NAPI GRO helpers when receiving SKBs. This improves performances. Signed-off-by:
Antoine Tenart <antoine.tenart@bootlin.com> Signed-off-by:
-
Antoine Tenart authored
Use NAPI_POLL_WEIGHT, the default NAPI poll() weight instead of redefining our own value (which turns out to be 64 as well). Signed-off-by:
Nicolas Ferre <nicolas.ferre@microchip.com> Signed-off-by:
-
David S. Miller authored
Shijie Luo says: ==================== ipv4: fix bugs when enable route_localnet When enable route_localnet, route of the 127/8 address is enabled. But in some situations like arp_announce=2, ARP requests or reply work abnormally. This patchset fix some bugs when enable route_localnet. Change History: V2: - Change a single patch to a patchset. - Add bug fix for arp_ignore = 3. - Add a couple of test for enabling route_localnet in selftests. ==================== Signed-off-by: -
Shijie Luo authored
Add a simple scripts to exercise several situations when enable route_localnet. Signed-off-by:
Shijie Luo <luoshijie1@huawei.com> Signed-off-by:
Zhiqiang liu <liuzhiqiang26@huawei.com> Signed-off-by:
-
Shijie Luo authored
When arp_ignore=3, the NIC won't reply for scope host addresses, but if enable route_locanet, we need to reply ip address with head 127 and scope RT_SCOPE_HOST. Fixes: d0daebc3 ("ipv4: Add interface option to enable routing of 127.0.0.0/8") Signed-off-by:
-
Shijie Luo authored
Suppose we have two interfaces eth0 and eth1 in two hosts, follow the same steps in the two hosts: # sysctl -w net.ipv4.conf.eth1.route_localnet=1 # sysctl -w net.ipv4.conf.eth1.arp_announce=2 # ip route del 127.0.0.0/8 dev lo table local and then set ip to eth1 in host1 like: # ifconfig eth1 127.25.3.4/24 set ip to eth2 in host2 and ping host1: # ifconfig eth1 127.25.3.14/24 # ping -I eth1 127.25.3.4 Well, host2 cannot connect to host1. When set a ip address with head 127, the scope of the address defaults to RT_SCOPE_HOST. In this situation, host2 will use arp_solicit() to send a arp request for the mac address of host1 with ip address 127.25.3.14. When arp_announce=2, inet_select_addr() cannot select a correct saddr with condition ifa->ifa_scope > scope, because ifa_scope is RT_SCOPE_HOST and scope is RT_SCOPE_LINK. Then, inet_select_addr() will go to no_in_dev to lookup all interfaces to find a primary ip and finally get the primary ip of eth0. Here I add a localnet_scope defaults to RT_SCOPE_HOST, and when route_localnet is enabled, this value changes to RT_SCOPE_LINK to make inet_select_addr() find a correct primary ip as saddr of arp request. Fixes: d0daebc3 ("ipv4: Add interface option to enable routing of 127.0.0.0/8") Signed-off-by:
-
Xin Long authored
tipc_nl_compat_bearer_set() is only called by tipc_nl_compat_link_set() which already does the check for msg->req check, so remove it from tipc_nl_compat_bearer_set(), and do the same in tipc_nl_compat_media_set(). Signed-off-by:
Xin Long <lucien.xin@gmail.com> Signed-off-by:
-
David S. Miller authored
Ido Schimmel says: ==================== mlxsw: Thermal and hwmon extensions This patchset from Vadim includes various enhancements to thermal and hwmon code in mlxsw. Patch #1 adds a thermal zone for each inter-connect device (gearbox). These devices are present in SN3800 systems and code to expose their temperature via hwmon was added in commit 2e265a8b ("mlxsw: core: Extend hwmon interface with inter-connect temperature attributes"). Currently, there are multiple thermal zones in mlxsw and only a few cooling devices. Patch #2 detects the hottest thermal zone and the cooling devices are switched to follow its trends. RFC was sent last month [1]. Patch #3 allows to read and report negative temperature of the sensors mlxsw exposes via hwmon and thermal subsystems. v2 (Andrew Lunn): * In patch #3, replace '%u' with '%d' in mlxsw_hwmon_module_temp_show() [1] https://patchwork.ozlabs.org/patch/1107161/ ==================== Signed-off-by:
-
Vadim Pasternak authored
Extend macros MLXSW_REG_MTMP_TEMP_TO_MC() to allow support of negative temperature readout, since chip and others thermal components are capable of operating within the negative temperature. With no such support negative temperature will be consider as very high temperature and it will cause wrong readout and thermal shutdown. For negative values 2`s complement is used. Tested in chamber. Example of chip ambient temperature readout with chamber temperature: -10 Celsius: temp1: -6.0C (highest = -5.0C) -5 Celsius: temp1: -1.0C (highest = -1.0C) v2 (Andrew Lunn): * Replace '%u' with '%d' in mlxsw_hwmon_module_temp_show() Signed-off-by:
Vadim Pasternak <vadimp@mellanox.com> Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Acked-by:
Jiri Pirko <jiri@mellanox.com> Signed-off-by:
-
Vadim Pasternak authored
When multiple sensors are mapped to the same cooling device, the cooling device should be set according the worst sensor from the sensors associated with this cooling device. Provide the hottest thermal zone detection and enforce cooling device to follow the temperature trends of the hottest zone only. Prevent competition for the cooling device control from others zones, by "stable trend" indication. A cooling device will not perform any actions associated with a zone with a "stable trend". When other thermal zone is detected as a hottest, a cooling device is to be switched to following temperature trends of new hottest zone. Thermal zone score is represented by 32 bits unsigned integer and calculated according to the next formula: For T < TZ<t><i>, where t from {normal trip = 0, high trip = 1, hot trip = 2, critical = 3}: TZ<i> score = (T + (TZ<t><i> - T) / 2) / (TZ<t><i> - T) * 256 ** j; Highest thermal zone score s is set as MAX(TZ<i>score); Following this formula, if TZ<i> is in trip point higher than TZ<k>, the higher score is to be always assigned to TZ<i>. For two thermal zones located at the same kind of trip point, the higher score will be assigned to the zone which is closer to the next trip point. Thus, the highest score will always be assigned objectively to the hottest thermal zone. All the thermal zones initially are to be configured with mode "enabled" with the "step_wise" governor. Signed-off-by: -
Vadim Pasternak authored
Add a dedicated thermal zone for each inter-connect device. The current temperature is obtained from inter-connect temperature sensor and the default trip points are set to the same values as default ASIC trip points. These settings could be changed from the user space. A cooling device (fan) is bound to all inter-connect devices. Signed-off-by:
-
john.rutherford@dektech.com.au authored
Fix misalignment of policy statement in netlink.c due to automatic spatch code transformation. Fixes: 3b0f31f2 ("genetlink: make policy common to family") Acked-by:
Jon Maloy <jon.maloy@ericsson.com> Signed-off-by:
John Rutherford <john.rutherford@dektech.com.au> Signed-off-by:
-
Keerthy authored
Commit bfe59032 ("net: ethernet: ti: cpsw: use cpsw as drv data")changes the driver data to struct cpsw_common *cpsw. This is done only in probe/remove but the suspend/resume functions are still left with struct net_device *ndev. Hence fix both suspend & resume also to fetch the updated driver data. Fixes: bfe59032 ("net: ethernet: ti: cpsw: use cpsw as drv data") Signed-off-by:
Keerthy <j-keerthy@ti.com> Signed-off-by:
-
Xue Chaojing authored
This patch implement the statistical interface of ethtool, user can use ethtool -S to show hinic statistics. Signed-off-by:
Xue Chaojing <xuechaojing@huawei.com> Signed-off-by:
-
- 23 Jun, 2019 11 commits
-
-
David S. Miller authored
Wei Wang says: ==================== ipv6: avoid taking refcnt on dst during route lookup Ipv6 route lookup code always grabs refcnt on the dst for the caller. But for certain cases, grabbing refcnt is not always necessary if the call path is rcu protected and the caller does not cache the dst. Another issue in the route lookup logic is: When there are multiple custom rules, we have to do the lookup into each table associated to each rule individually. And when we can't find the route in one table, we grab and release refcnt on net->ipv6.ip6_null_entry before going to the next table. This operation is completely redundant, and causes false issue because net->ipv6.ip6_null_entry is a shared object. This patch set introduces a new flag RT6_LOOKUP_F_DST_NOREF for route lookup callers to set, to avoid any manipulation on the dst refcnt. And it converts the major input and output path to use it. The performance gain is noticable. I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues. Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port. Both hosts have multiple custom rules: - For incoming packets, only local table is traversed. - For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows: - Before the fix: 3.78Mpps - After the fix: 5.50Mpps v2->v3: - Handled fib6_rule_lookup() when CONFIG_IPV6_MULTIPLE_TABLES is not configured in patch 03 (suggested by David Ahern) - Removed the renaming of l3mdev_link_scope_lookup() in patch 05 (suggested by David Ahern) - Moved definition of ip6_route_output_flags() from an inline function in /net/ipv6/route.c to net/ipv6/route.c in order to address kbuild error in patch 05 v1->v2: - Added a helper ip6_rt_put_flags() in patch 3 suggested by David Miller ==================== Reviewed-by:
Eric Dumazet <edumazet@google.com> Signed-off-by:
-
Wei Wang authored
For tx path, in most cases, we still have to take refcnt on the dst cause the caller is caching the dst somewhere. But it still is beneficial to make use of RT6_LOOKUP_F_DST_NOREF flag while doing the route lookup. It is cause this flag prevents manipulating refcnt on net->ipv6.ip6_null_entry when doing fib6_rule_lookup() to traverse each routing table. The null_entry is a shared object and constant updates on it cause false sharing. We converted the current major lookup function ip6_route_output_flags() to make use of RT6_LOOKUP_F_DST_NOREF. Together with the change in the rx path, we see noticable performance boost: I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues. Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port. Both hosts have multiple custom rules: - For incoming packets, only local table is traversed. - For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows: - Before the fix: 3.78Mpps - After the fix: 5.50Mpps Signed-off-by:
Wei Wang <weiwan@google.com> Signed-off-by:
-
Wei Wang authored
ip6_route_input() is the key function to do the route lookup in the rx data path. All the callers to this function are already holding rcu lock. So it is fairly easy to convert it to not take refcnt on the dst: We pass in flag RT6_LOOKUP_F_DST_NOREF and do skb_dst_set_noref(). This saves a few atomic inc or dec operations and should boost performance overall. This also makes the logic more aligned with v4. Signed-off-by:
Mahesh Bandewar <maheshb@google.com> Signed-off-by:
-
Wei Wang authored
This patch specifically converts the rule lookup logic to honor this flag and not release refcnt when traversing each rule and calling lookup() on each routing table. Similar to previous patch, we also need some special handling of dst entries in uncached list because there is always 1 refcnt taken for them even if RT6_LOOKUP_F_DST_NOREF flag is set. Signed-off-by:
-
Wei Wang authored
Initialize rt6->rt6i_uncached on the following pre-allocated dsts: net->ipv6.ip6_null_entry net->ipv6.ip6_prohibit_entry net->ipv6.ip6_blk_hole_entry This is a preparation patch for later commits to be able to distinguish dst entries in uncached list by doing: !list_empty(rt6->rt6i_uncached) Signed-off-by:
-
Wei Wang authored
This new flag is to instruct the route lookup function to not take refcnt on the dst entry. The user which does route lookup with this flag must properly use rcu protection. ip6_pol_route() is the major route lookup function for both tx and rx path. In this function: Do not take refcnt on dst if RT6_LOOKUP_F_DST_NOREF flag is set, and directly return the route entry. The caller should be holding rcu lock when using this flag, and decide whether to take refcnt or not. One note on the dst cache in the uncached_list: As uncached_list does not consume refcnt, one refcnt is always returned back to the caller even if RT6_LOOKUP_F_DST_NOREF flag is set. Uncached dst is only possible in the output path. So in such call path, caller MUST check if the dst is in the uncached_list before assuming that there is no refcnt taken on the returned dst. Signed-off-by:
-
Russell King authored
There seems to be some confusion surrounding three PHY interface modes, specifically 1000BASE-X, 2500BASE-X and SGMII. Add some documentation to phylib detailing precisely what these interface modes refer to. Signed-off-by:
Russell King <rmk+kernel@armlinux.org.uk> Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
Qian Cai authored
The linux-next commit "inet: fix various use-after-free in defrags units" [1] introduced compilation warnings, ./include/net/inet_frag.h:117:1: warning: 'inline' is not at beginning of declaration [-Wold-style-declaration] static void inline fqdir_pre_exit(struct fqdir *fqdir) ^~~~~~ In file included from ./include/net/netns/ipv4.h:10, from ./include/net/net_namespace.h:20, from ./include/linux/netdevice.h:38, from ./include/linux/icmpv6.h:13, from ./include/linux/ipv6.h:86, from ./include/net/ipv6.h:12, from ./include/rdma/ib_verbs.h:51, from ./include/linux/mlx5/device.h:37, from ./include/linux/mlx5/driver.h:51, from drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c:37: [1] https://lore.kernel.org/netdev/20190618180900.88939-3-edumazet@google.com/Signed-off-by:Qian Cai <cai@lca.pw> Signed-off-by:
-
Rasmus Villemoes authored
This is a no-op that simply moves all locking and unlocking of ->reg_lock into trivial helpers. I did that to be able to easily add some ad hoc instrumentation to those helpers to get some information on contention and hold times of the mutex. Perhaps others want to do something similar at some point, so this frees them from doing the 'sed -i' yoga, and have a much smaller 'git diff' while fiddling. Signed-off-by:
Rasmus Villemoes <rasmus.villemoes@prevas.dk> Reviewed-by:
Vivien Didelot <vivien.didelot@gmail.com> Signed-off-by:
-
Sameeh Jubran authored
The current code of create_queues_with_size_backoff() allows the ring size to become as small as ENA_MIN_RING_SIZE/2. This is a bug since we don't want the queue ring to be smaller than ENA_MIN_RING_SIZE In this commit we change the loop's termination condition to look at the queue size of the next iteration instead of that of the current one, so that the minimal queue size again becomes ENA_MIN_RING_SIZE. Fixes: eece4d2a ("net: ena: add ethtool function for changing io queue sizes") Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
Sameeh Jubran <sameehj@amazon.com> Signed-off-by:
-
Colin Ian King authored
Currently pointer hwdev is dereferenced when assigning hwif before hwdev is null checked. Fix this by only derefencing hwdev after the null check. Addresses-Coverity: ("Dereference before null check") Fixes: 4fdc51bb ("hinic: add support for rss parameters with ethtool") Signed-off-by:Colin Ian King <colin.king@canonical.com> Signed-off-by:
-
- 22 Jun, 2019 3 commits
-
-
David S. Miller authored
René van Dorst says: ==================== net: mediatek: Add MT7621 TRGMII mode support Like many other mediatek SOCs, the MT7621 SOC and the internal MT7530 switch both supports TRGMII mode. MT7621 TRGMII speed is fix 1200MBit. v1->v2: - Fix breakage on non MT7621 SOC - Support 25MHz and 40MHz XTAL as MT7530 clocksource ==================== Tested-by:
"Frank Wunderlich" <frank-w@public-files.de> Acked-by:
Sean Wang <sean.wang@kernel.org> Signed-off-by:
-
René van Dorst authored
This patch add support TRGMII mode for MT7621 internal MT7530 switch. MT7621 TRGMII has only one fix speed mode of 1200MBit. Also adding support for mt7530 25MHz and 40MHz crystal clocksource. Values are based on Banana Pi R2 bsp [1]. Don't change MT7623 registers on a MT7621 device. [1] https://github.com/BPI-SINOVOIP/BPI-R2-bsp/blob/master/linux-mt/drivers/net/ethernet/mediatek/gsw_mt7623.c#L769Signed-off-by:
René van Dorst <opensource@vdorst.com> Tested-by:
-
René van Dorst authored
MT7621 SOC also supports TRGMII. TRGMII speed is 1200MBit. Signed-off-by:
-
