- 05 Mar, 2024 8 commits
-
-
David Howells authored
Extract useful fields from a received ACK packet into the skb private data early on in the process of parsing incoming packets. This makes the ACK fields available even before we've matched the ACK up to a call and will allow us to deal with path MTU discovery probe responses even after the relevant call has been completed. Signed-off-by:David Howells <dhowells@redhat.com> cc: Marc Dionne <marc.dionne@auristor.com> cc: "David S. Miller" <davem@davemloft.net> cc: Eric Dumazet <edumazet@google.com> cc: Jakub Kicinski <kuba@kernel.org> cc: Paolo Abeni <pabeni@redhat.com> cc: linux-afs@lists.infradead.org cc: netdev@vger.kernel.org
-
David Howells authored
Clean up the DATA packet resending algorithm to retransmit packets as we come across them whilst walking the transmission buffer rather than queuing them for retransmission at the end. This can be done as ACK parsing - and thus the discarding of successful packets - is now done in the same thread rather than separately in softirq context and a locked section is no longer required. Signed-off-by: -
David Howells authored
Move the recording of a successfully transmitted DATA or ACK packet that will provide RTT probing to after the transmission. With the I/O thread model, this can be done because parsing of the responding ACK can no longer race with the post-transmission code. Move the various timeout-settings done after successfully transmitting a DATA packet into rxrpc_tstamp_data_packets() and eliminate a number of calls to get the current time. As a consequence we no longer need to cancel a proposed RTT probe on transmission failure. Signed-off-by: -
David Howells authored
Track the call timeouts as ktimes rather than jiffies as the latter's granularity is too high and only set the timer at the end of the event handling function. Signed-off-by: -
David Howells authored
There are three points that transmit PING ACKs and all of them use the same trace string. Change two of them to use different strings. Signed-off-by: -
David Howells authored
Once all the packets transmitted as part of a call have been acked, don't permit any resending. Signed-off-by: -
David Howells authored
Parse the received packets before going and processing timeouts as the timeouts may be reset by the reception of a packet. Signed-off-by: -
David Howells authored
Switch from keeping the transmission buffers in the rxrpc_txbuf struct and allocated from the slab, to allocating them using page fragment allocators (which uses raw pages), thereby allowing them to be passed to MSG_SPLICE_PAGES and avoid copying into the UDP buffers. Signed-off-by:
-
- 03 Mar, 2024 1 commit
-
-
David Howells authored
Use rxrpc_txbuf::kvec[0] instead of rxrpc_txbuf::wire to gain access to the Rx protocol header. In future, the wire header will be stored in a page frag, not in the rxrpc_txbuf struct making it possible to use MSG_SPLICE_PAGES when sending it. Similarly, access the ack header as being immediately after the wire header when filling out an ACK packet. Signed-off-by:
-
- 29 Feb, 2024 31 commits
-
-
David Howells authored
Move rxrpc_send_ACK() to output.c to so that it is with rxrpc_send_ack_packet() prior to merging the two. Signed-off-by: -
David Howells authored
Don't pick values out of the wire header in rxkad when setting up DATA packet security, but rather use other sources. This makes it easier to get rid of txb->wire. Signed-off-by: -
David Howells authored
Split (sub)packet preparation and timestamping out of the DATA packet transmission function to make it easier to glue multiple txbufs together into a jumbo DATA packet. This will require preparation and timestamping of all the subpackets in a txbuf, and these functions provide convenient points to place the required iteration. Signed-off-by: -
David Howells authored
Add a kvec[] to the rxrpc_txbuf struct to point to the contributory buffers for a packet. Start with just a single element for now, but this will be expanded later. Make the ACK sending function use it, which means that rxrpc_fill_out_ack() doesn't need to return the size of the sack table, padding and trailer. Make the data sending code use it, both in where sendmsg() packages code up into txbufs and where those txbufs are transmitted. Signed-off-by: -
David Howells authored
Merge together the DF and non-DF branches of the transmission function and always set the flag to the right thing before transmitting. If we see -EMSGSIZE from udp_sendmsg(), turn off DF and retry. Signed-off-by: -
David Howells authored
Don't reset the DF flag after transmission, but rather set it when needed since it should be a fast op now that we call IP directly. This includes turning it off for RESPONSE packets and, for the moment, ACK packets. In future, we will need to turn it on for ACK packets used to do path MTU discovery. Signed-off-by: -
David Howells authored
call->tx_transmitted and call->acks_prev_seq don't need to be managed with cmpxchg() and barriers as it's only used within the singular I/O thread. Signed-off-by: -
David Howells authored
Strip the atomic ops and barriering off of the call timer tracking as this is handled solely within the I/O thread, except for expect_term_by which is set by sendmsg(). Signed-off-by: -
David Howells authored
From AFS-3.3 a trailer containing extra info was added to the ACK packet format - but AF_RXRPC has the names of some of the fields mixed up compared to other AFS implementations. Rename the struct and the fields to make them match. Signed-off-by: -
David Howells authored
Add a field to rxrpc_txbuf in which to store the checksum to go in the header as this may get overwritten in the wire header struct when transmitting as part of a jumbo packet. Signed-off-by: -
David Howells authored
Convert the transmission buffer flags into a mask and use | and & rather than bitops functions (atomic ops are not required as only the I/O thread can manipulate them once submitted for transmission). The bottom byte can then correspond directly to the Rx protocol header flags. Signed-off-by: -
David Howells authored
Each Rx protocol packet contains a per-connection monotonically increasing serial number used to correlate outgoing messages with their replies - something that can be used for RTT calculation. Note this value in the rxrpc_txbuf struct in addition to the wire header and then log it in the rxrpc_retransmit trace for reference. Signed-off-by: -
Paolo Abeni authored
Breno Leitao says: ==================== net: collect tstats automatically The commit 34d21de9 ("net: Move {l,t,d}stats allocation to core and convert veth & vrf") added a field in struct_netdevice, which tells what type of statistics the driver supports. That field is used primarily to allocate stats structures automatically, but, it also could leveraged to simplify the drivers even further, such as, if the driver relies in the default stats collection, then it doesn't need to assign to .ndo_get_stats64. That means that drivers only assign functions to .ndo_get_stats64 if they are using something special. I started to move some of these drivers[1][2][3] to use the core allocation, and with this change in, I just need to touch the driver once, and be able to simplify the whole stats allocation and collection for generic case. There are 44 devices today that could benefit from this simplification. # grep -r .ndo_get_stats64 | grep dev_get_tstats64 | wc -l 44 As of today, netnext only has the `sit` driver fully ported to core stats allocation, hence the second patch. Links: [1] https://lore.kernel.org/all/20240227182338.2739884-1-leitao@debian.org/ [2] https://lore.kernel.org/all/20240222144117.1370101-1-leitao@debian.org/ [3] https://lore.kernel.org/all/20240223115839.3572852-1-leitao@debian.org/ ==================== Link: https://lore.kernel.org/r/20240228113125.3473685-1-leitao@debian.orgSigned-off-by:
Paolo Abeni <pabeni@redhat.com>
-
Breno Leitao authored
If the driver is using the network core allocation mechanism, by setting NETDEV_PCPU_STAT_TSTATS, as this driver is, then, it doesn't need to set the dev_get_tstats64() generic .ndo_get_stats64 function pointer. Since the network core calls it automatically, and .ndo_get_stats64 should only be set if the driver needs special treatment. This simplifies the driver, since all the generic statistics is now handled by core. Signed-off-by:
Breno Leitao <leitao@debian.org> Reviewed-by:
Simon Horman <horms@kernel.org> Signed-off-by:
-
Breno Leitao authored
If the network driver is relying in the net core to do stats allocation, then we want to dev_get_tstats64() instead of netdev_stats_to_stats64(), since there are per-cpu stats that needs to be taken in consideration. This will also simplify the drivers in regard to statistics. Once the driver sets NETDEV_PCPU_STAT_TSTATS, it doesn't not need to allocate the stacks, neither it needs to set `.ndo_get_stats64 = dev_get_tstats64` for the generic stats collection function anymore. Signed-off-by:
-
Yanteng Si authored
This is just a trivial fix for a typo in a comment, no functional changes. Reviewed-by:
Serge Semin <fancer.lancer@gmail.com> Signed-off-by:
Yanteng Si <siyanteng@loongson.cn> Link: https://lore.kernel.org/r/20240228112447.1490926-1-siyanteng@loongson.cnSigned-off-by:
-
Zhengchao Shao authored
The input parameter 'level' in do_raw_set/getsockopt() is not used. Therefore, remove it. Signed-off-by:
Zhengchao Shao <shaozhengchao@huawei.com> Reviewed-by:
Jiri Pirko <jiri@nvidia.com> Link: https://lore.kernel.org/r/20240228072505.640550-1-shaozhengchao@huawei.comSigned-off-by:
-
Paolo Abeni authored
Robert Marko says: ==================== net: dsa: mv88e6xxx: add Amethyst specific SMI GPIO function Amethyst family (MV88E6191X/6193X/6393X) has a simplified SMI GPIO setting via the Scratch and Misc register so it requires family specific function. In the v1 review, Andrew pointed out that it would make sense to rename the existing mv88e6xxx_g2_scratch_gpio_set_smi as it only works on the MV6390 family. Changes in v2: * Add rename of mv88e6xxx_g2_scratch_gpio_set_smi to mv88e6390_g2_scratch_gpio_set_smi ==================== Link: https://lore.kernel.org/r/20240227175457.2766628-1-robimarko@gmail.comSigned-off-by:
-
Robert Marko authored
The existing mv88e6390_g2_scratch_gpio_set_smi() cannot be used on the 88E6393X as it requires certain P0_MODE, it also checks the CPU mode as it impacts the bit setting value. This is all irrelevant for Amethyst (MV88E6191X/6193X/6393X) as only the default value of the SMI_PHY Config bit is set to CPU_MGD bootstrap pin value but it can be changed without restrictions so that GPIO pins 9 and 10 are used as SMI pins. So, introduce Amethyst specific function and call that if the Amethyst family wants to setup the external PHY. Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
Robert Marko <robimarko@gmail.com> Signed-off-by:
-
Robert Marko authored
The name mv88e6xxx_g2_scratch_gpio_set_smi is a bit ambiguous as it appears to only be applicable to the 6390 family, so lets rename it to mv88e6390_g2_scratch_gpio_set_smi to make it more obvious. Signed-off-by:
-
Eric Dumazet authored
I missed that inet6_dump_addr() is calling in6_dump_addrs() from two points. First one under RTNL protection, and second one under rcu_read_lock(). Since we want to remove RTNL use from inet6_dump_addr() very soon, no longer assume in6_dump_addrs() is protected by RTNL (even if this is still the case). Use rcu_read_lock() earlier to fix this lockdep splat: WARNING: suspicious RCU usage 6.8.0-rc5-syzkaller-01618-gf8cbf6bd #0 Not tainted net/ipv6/addrconf.c:5317 suspicious rcu_dereference_check() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 3 locks held by syz-executor.2/8834: #0: ffff88802f554678 (nlk_cb_mutex-ROUTE){+.+.}-{3:3}, at: __netlink_dump_start+0x119/0x780 net/netlink/af_netlink.c:2338 #1: ffffffff8f377a88 (rtnl_mutex){+.+.}-{3:3}, at: netlink_dump+0x676/0xda0 net/netlink/af_netlink.c:2265 #2: ffff88807e5f0580 (&ndev->lock){++--}-{2:2}, at: in6_dump_addrs+0xb8/0x1de0 net/ipv6/addrconf.c:5279 stack backtrace: CPU: 1 PID: 8834 Comm: syz-executor.2 Not tainted 6.8.0-rc5-syzkaller-01618-gf8cbf6bd #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/25/2024 Call Trace: <TASK> __dump_stack lib/dump_stack.c:88 [inline] dump_stack_lvl+0x1e7/0x2e0 lib/dump_stack.c:106 lockdep_rcu_suspicious+0x220/0x340 kernel/locking/lockdep.c:6712 in6_dump_addrs+0x1b47/0x1de0 net/ipv6/addrconf.c:5317 inet6_dump_addr+0x1597/0x1690 net/ipv6/addrconf.c:5428 netlink_dump+0x6a6/0xda0 net/netlink/af_netlink.c:2266 __netlink_dump_start+0x59d/0x780 net/netlink/af_netlink.c:2374 netlink_dump_start include/linux/netlink.h:340 [inline] rtnetlink_rcv_msg+0xcf7/0x10d0 net/core/rtnetlink.c:6555 netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2547 netlink_unicast_kernel net/netlink/af_netlink.c:1335 [inline] netlink_unicast+0x7ea/0x980 net/netlink/af_netlink.c:1361 netlink_sendmsg+0x8e0/0xcb0 net/netlink/af_netlink.c:1902 sock_sendmsg_nosec net/socket.c:730 [inline] __sock_sendmsg+0x221/0x270 net/socket.c:745 ____sys_sendmsg+0x525/0x7d0 net/socket.c:2584 ___sys_sendmsg net/socket.c:2638 [inline] __sys_sendmsg+0x2b0/0x3a0 net/socket.c:2667 Fixes: c3718936 ("ipv6: anycast: complete RCU handling of struct ifacaddr6") Reported-by:
syzbot <syzkaller@googlegroups.com> Signed-off-by:
Eric Dumazet <edumazet@google.com> Reviewed-by:
Jakub Kicinski <kuba@kernel.org>
-
Eric Dumazet authored
skb_defer_free_flush() is currently called from net_rx_action() and napi_threaded_poll(). We should also call it from __napi_busy_loop() otherwise there is the risk the percpu queue can grow until an IPI is forced from skb_attempt_defer_free() adding a latency spike. Signed-off-by:
Stanislav Fomichev <sdf@google.com> Reviewed-by:
-
Eric Dumazet authored
By moving some fields around, this patch shrinks holes size from 56 to 32, saving 24 bytes on 64bit arches. After the patch pahole gives the following for 'struct tcp_sock': /* size: 2304, cachelines: 36, members: 162 */ /* sum members: 2234, holes: 6, sum holes: 32 */ /* sum bitfield members: 34 bits, bit holes: 5, sum bit holes: 14 bits */ /* padding: 32 */ /* paddings: 3, sum paddings: 10 */ /* forced alignments: 1, forced holes: 1, sum forced holes: 12 */ Signed-off-by:
-
Breno Leitao authored
If br_multicast_init_stats() fails, there is no need to set lockdep classes. Just return from the error path. Signed-off-by:
Nikolay Aleksandrov <razor@blackwall.org> Link: https://lore.kernel.org/r/20240227182338.2739884-2-leitao@debian.orgSigned-off-by:
-
Breno Leitao authored
With commit 34d21de9 ("net: Move {l,t,d}stats allocation to core and convert veth & vrf"), stats allocation could be done on net core instead of this driver. With this new approach, the driver doesn't have to bother with error handling (allocation failure checking, making sure free happens in the right spot, etc). This is core responsibility now. Remove the allocation in the bridge driver and leverage the network core allocation. Signed-off-by:
-
Ido Schimmel authored
Rename some test cases to avoid overlapping test names which is problematic for the kernel test robot. No changes in the test's logic. Suggested-by:
Yujie Liu <yujie.liu@intel.com> Signed-off-by:
Ido Schimmel <idosch@nvidia.com> Link: https://lore.kernel.org/r/20240227170418.491442-1-idosch@nvidia.comSigned-off-by:
-
Zhengchao Shao authored
Since commit 446fda4f ("[NetLabel]: CIPSOv4 engine"), *bitmap_walk function only returns -1. Nearly 18 years have passed, -2 scenes never come up, so there's no need to consider it. Signed-off-by:
Paul Moore <paul@paul-moore.com> Link: https://lore.kernel.org/r/20240227093604.3574241-1-shaozhengchao@huawei.comSigned-off-by:
-
Jakub Kicinski authored
Eric Dumazet says: ==================== inet: implement lockless RTM_GETNETCONF ops This series removes RTNL use for RTM_GETNETCONF operations on AF_INET. - Annotate data-races to avoid possible KCSAN splats. - "ip -4 netconf show dev XXX" can be implemented without RTNL [1] - "ip -4 netconf" dumps can be implemented using RCU instead of RTNL [1] [1] This only refers to RTM_GETNETCONF operation, "ip" command also uses RTM_GETLINK dumps which are using RTNL at this moment. ==================== Link: https://lore.kernel.org/r/20240227092411.2315725-1-edumazet@google.comSigned-off-by: -
Eric Dumazet authored
1) inet_netconf_dump_devconf() can run under RCU protection instead of RTNL. 2) properly return 0 at the end of a dump, avoiding an an extra recvmsg() system call. 3) Do not use inet_base_seq() anymore, for_each_netdev_dump() has nice properties. Restarting a GETDEVCONF dump if a device has been added/removed or if net->ipv4.dev_addr_genid has changed is moot. Signed-off-by:
-
Eric Dumazet authored
"ip -4 netconf show dev XXXX" no longer acquires RTNL. Return -ENODEV instead of -EINVAL if no netdev or idev can be found. Signed-off-by:
-
Eric Dumazet authored
Add READ_ONCE() in ipv4_devconf_get() and corresponding WRITE_ONCE() in ipv4_devconf_set() Add IPV4_DEVCONF_RO() and IPV4_DEVCONF_ALL_RO() macros, and use them when reading devconf fields. Signed-off-by:
-
