- 13 Apr, 2024 14 commits
-
-
Florian Westphal authored
Use socat, the different nc implementations have too much variance wrt. supported options. Avoid sleeping until listener is up, use busywait helper for this, this also greatly reduces test duration. Signed-off-by:
Florian Westphal <fw@strlen.de> Link: https://lore.kernel.org/r/20240411233624.8129-15-fw@strlen.deSigned-off-by:
Jakub Kicinski <kuba@kernel.org>
-
Florian Westphal authored
Also lower ping interval, wait times (helpers get called several times) and set nodad for ipv6 addresses: 20s down to 4s. Signed-off-by:
-
Florian Westphal authored
prefer socat over nc, nc has too many incompatible versions around. Signed-off-by:
-
Florian Westphal authored
While at it, use checktool helper. Signed-off-by:
-
Florian Westphal authored
The setup_ns helper makes the netns names random, so replace nsX with $nsX everywhere. Replace nc with socat, otherwise script fails on my system due to incompatible nc versions ("nc: cannot use -p and -l"). Signed-off-by: -
Florian Westphal authored
... so it doesn't have to be repeated everywhere. Signed-off-by:
-
Florian Westphal authored
Signed-off-by:
-
Florian Westphal authored
swap test for "ip" with "conntrack", former is already accounted for via setup_ns helper. Also switch to bash. Signed-off-by:
-
Florian Westphal authored
While at it, address warnings generated by shellcheck and fix following minor issues: - some distros place netem in 'extra' modules package, so add a skip check for netem-attach failure. - tc prints a warning for the 100mbit class: "Warning: sch_htb: quantum of class 10001 is big. Consider r2q change." Silence this by increasing the divisor. Signed-off-by:
-
Florian Westphal authored
Replace nc with socat. Too many different implementations of nc are around with incompatible options ("nc: cannot use -p and -l"). Signed-off-by: -
Florian Westphal authored
Only relevant change is that netns names have random suffix names, i.e. its safe to run this in parallel with other tests. Signed-off-by:
-
Florian Westphal authored
Also, fix two issues reported by Pablo Neira: 1. Must modprobe br_netfilter in case its not loaded, else sysctl cannot be set. 2. ping for netns4 fails if rp_filter is enabled in bridge netns, so set all and default to 0. Signed-off-by:
-
Florian Westphal authored
Doing so gets us dynamically generated netns names. Also: * do not assume rp_filter is disabled, if its on script failed * reduce timeout (-W) for "expected to fail" ping commands * don't print PASS line for basic sanity ping * shellcheck cleanups Signed-off-by:
-
Florian Westphal authored
.. so this can start re-using existing lib.sh infra in next patches. Several of these scripts will not work, e.g. because they assume rp_filter is disabled, or reliance on a particular version/flavor of "netcat" tool. Add config settings for them. nft_trans_stress.sh script is removed, it also exists in the nftables userspace selftests. I do not see a reason to keep two versions in different repositories/projects. The settings file is removed for now: It was used to increase the timeout to avoid slow scripts from getting zapped by the 45s timeout, but some of the slow scripts can be sped up. Re-add it later for scripts that cannot be sped up easily. Update MAINTAINERS to reflect that future updates to netfilter scripts should go through netfilter-devel@. Signed-off-by:
-
- 12 Apr, 2024 26 commits
-
-
David S. Miller authored
Louis Peens says: ==================== nfp: series of minor driver improvements This short series bundles now only includes a small update to add a board part number to devlink. Previously some dim patches also formed part of this series, these were dropped in v5. Patch1: Add new define for devlink string "board.part_number" Patch2: Make use of this field in the nfp driver Changes since V4: - Dropped the dim patches, as there is a more significant rework in progress to make it more flexible, as mentioned in the V4 review: https://lore.kernel.org/all/1712547870-112976-2-git-send-email-hengqi@linux.alibaba.com/ - Updated the devlink description of 'board.part_number' Changes since V3: - Fixed: Documentation/networking/devlink/devlink-info.rst:150: WARNING: Title underline too short. Changes since V2: - After some discussion on the previous series it was agreed that only the "board.part_number" field makes sense in the common code. The "board.model" field which was moved to devlink common code in V1 is now kept in the driver. The field is specific to the nfp driver, exposing the codename of the board. - In summary, add "board.part_number" to devlink, and populate it in the the nfp driver. Changes since V1: - Move nfp local defines to devlink common code as it is quite generic. - Add new 'dim' profile instead of using driver local overrides, as this allows use of the 'dim' helpers. - This expanded 2 patches to 4, as the common code changes are split into seperate patches. ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Fei Qin authored
Newer NIC will introduce a new part number, now add it into devlink device info. This patch also updates the information of "board.id" in nfp.rst to match the devlink-info.rst. Signed-off-by:
Fei Qin <fei.qin@corigine.com> Signed-off-by:
Louis Peens <louis.peens@corigine.com> Reviewed-by:
Jacob Keller <jacob.e.keller@intel.com> Signed-off-by:
-
Fei Qin authored
Add definition and documentation for the new generic info "board.part_number". The new one is for part number specific use, and board.id is modified to match the documentation in devlink-info. Signed-off-by:
-
Hechao Li authored
After commit dfa2f048 ("tcp: get rid of sysctl_tcp_adv_win_scale"), we noticed an application-level timeout due to reduced throughput. Before the commit, for a client that sets SO_RCVBUF to 65k, it takes around 22 seconds to transfer 10M data. After the commit, it takes 40 seconds. Because our application has a 30-second timeout, this regression broke the application. The reason that it takes longer to transfer data is that tp->scaling_ratio is initialized to a value that results in ~0.25 of rcvbuf. In our case, SO_RCVBUF is set to 65536 by the application, which translates to 2 * 65536 = 131,072 bytes in rcvbuf and hence a ~28k initial receive window. Later, even though the scaling_ratio is updated to a more accurate skb->len/skb->truesize, which is ~0.66 in our environment, the window stays at ~0.25 * rcvbuf. This is because tp->window_clamp does not change together with the tp->scaling_ratio update when autotuning is disabled due to SO_RCVBUF. As a result, the window size is capped at the initial window_clamp, which is also ~0.25 * rcvbuf, and never grows bigger. Most modern applications let the kernel do autotuning, and benefit from the increased scaling_ratio. But there are applications such as kafka that has a default setting of SO_RCVBUF=64k. This patch increases the initial scaling_ratio from ~25% to 50% in order to make it backward compatible with the original default sysctl_tcp_adv_win_scale for applications setting SO_RCVBUF. Fixes: dfa2f048 ("tcp: get rid of sysctl_tcp_adv_win_scale") Signed-off-by:
Hechao Li <hli@netflix.com> Reviewed-by:
Tycho Andersen <tycho@tycho.pizza> Reviewed-by:
Eric Dumazet <edumazet@google.com> Link: https://lore.kernel.org/netdev/20240402215405.432863-1-hli@netflix.com/Signed-off-by:
-
David S. Miller authored
Eric Woudstra says: ==================== rtl8226b/8221b add C45 instances and SerDes switching Based on the comments in [PATCH net-next] "Realtek RTL822x PHY rework to c45 and SerDes interface switching" Adds SerDes switching interface between 2500base-x and sgmii for rtl8221b and rtl8226b. Add get_rate_matching() for rtl8226b and rtl8221b, reading the serdes mode from phy. Driver instances are added for rtl8226b and rtl8221b for Clause 45 access only. The existing code is not touched, they use newly added functions. They also use the same rtl822xb_config_init() and rtl822xb_get_rate_matching() as these functions also can be used for direct Clause 45 access. Also Adds definition of MMC 31 registers, which cannot be used through C45-over-C22, only when phydev->is_c45 is set. Change rtlgen_get_speed() so the register value is passed as argument. Using Clause 45 access, this value is retrieved differently. Rename it to rtlgen_decode_speed() and add a call to it in rtl822x_c45_read_status(). Add rtl822x_c45_get_features() to set supported port for rtl8221b. Then 1 quirk is added for sfp modules known to have a rtl8221b behind RollBall, Clause 45 only, protocol. Changed in PATCH v4: * Changed switch to if statement in rtl822xb_get_rate_matching() * Removed setting ETHTOOL_LINK_MODE_MII_BIT in rtl822x_c45_get_features() Changed in PATCH v3: * Only apply to rtl8221b and rtl8226b phy's * Set phydev->rate_matching in .config_init() * Removed OEM SFP fixup for now, as there are modules with the same vendor name/PN, but with different PHY's. We found rtl8221b, but also the ty8821, which is not yet supported. Changed in PATCH v2: * Set author to Marek for the commit of the new C45 instances * Separate commit for setting supported ports * Renamed rtlgen_get_speed to rtlgen_decode_speed * Always fill in possible interfaces * Renamed sfp_fixup_oem_2_5g to sfp_fixup_oem_2_5gbaset * Only update phydev->interface when link is up Alexander Couzens (1): net: phy: realtek: configure SerDes mode for rtl822xb PHYs Eric Woudstra (3): net: phy: realtek: add get_rate_matching() for rtl822xb PHYs net: phy: realtek: Change rtlgen_get_speed() to rtlgen_decode_speed() net: phy: realtek: add rtl822x_c45_get_features() to set supported port ==================== Signed-off-by: -
Marek Behún authored
Add quirk for another RollBall copper transceiver: Turris RTSFP-2.5G, containing 2.5g capable RTL8221B PHY. Signed-off-by:
Marek Behún <kabel@kernel.org> Signed-off-by:
Eric Woudstra <ericwouds@gmail.com> Reviewed-by:
Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Signed-off-by:
-
Eric Woudstra authored
Sets ETHTOOL_LINK_MODE_TP_BIT in phydev->supported. Signed-off-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
Eric Woudstra authored
The value of the register to determine the speed, is retrieved differently when using Clause 45 only. To use the rtlgen_get_speed() function in this case, pass the value of the register as argument to rtlgen_get_speed(). The function would then always return 0, so change it to void. A better name for this function now is rtlgen_decode_speed(). Replace a call to genphy_read_status() followed by rtlgen_get_speed() with a call to rtlgen_read_status() in rtl822x_read_status(). Add reading speed to rtl822x_c45_read_status(). Signed-off-by:
-
Marek Behún authored
Collected from several commits in [PATCH net-next] "Realtek RTL822x PHY rework to c45 and SerDes interface switching" The instances are used by Clause 45 only accessible PHY's on several sfp modules, which are using RollBall protocol. Signed-off-by:
-
Eric Woudstra authored
Uses vendor register to determine if SerDes is setup in rate-matching mode. Rate-matching only supported when SerDes is set to 2500base-x. Signed-off-by:
-
Alexander Couzens authored
The rtl8221b and rtl8226b series support switching SerDes mode between 2500base-x and sgmii based on the negotiated copper speed. Configure this switching mode according to SerDes modes supported by host. There is an additional datasheet for RTL8226B/RTL8221B called "SERDES MODE SETTING FLOW APPLICATION NOTE" where a sequence is described to setup interface and rate adapter mode. However, there is no documentation about the meaning of registers and bits, it's literally just magic numbers and pseudo-code. Signed-off-by:
Alexander Couzens <lynxis@fe80.eu> [ refactored, dropped HiSGMII mode and changed commit message ] Signed-off-by:
-
Jakub Kicinski authored
Russell King says: ==================== net: dsa: allow phylink_mac_ops in DSA drivers This series showcases my idea of moving the phylink_mac_ops into DSA drivers, using mv88e6xxx as an example. Since I'm only changing one driver, providing the mac_ops has to be optional and the existing shims need to be kept for unconverted drivers. The first patch introduces a new helper that converts from the phylink_config structure that phylink uses to communicate with MAC drivers to the dsa_port structure. From this, DSA drivers can get the dsa_switch structure and thus their implementation specific data structure, and they can also retrieve the port index. The second patch adds the support to the core DSA layer to allow DSA drivers to provide phylink_mac_ops. The third patch converts mv88e6xxx to use this. I initially made this change after adding yet more phylink to DSA driver shims for my work with phylink-based EEE support, and decided that it was getting silly to keep implementing more and more shims. There are cases where shims don't work well - we had already tripped over a case a few years ago when the phylink mac_select_pcs operation was introduced. Phylink tested for the presence of this in the ops structure, but with DSA shims, this doesn't necessarily mean that the sub-driver supports this method. The only way to find that out is to call the method with dummy values and check the return code. The same thing was partly true when adding EEE support, and I ended up with this in phylink to determine whether the MAC supported EEE: +static bool phylink_mac_supports_eee(struct phylink *pl) +{ + return pl->mac_ops->mac_disable_tx_lpi && + pl->mac_ops->mac_enable_tx_lpi && + pl->config->lpi_capabilities; +} because merely testing for the presence of the operations is insufficient when shims are involved - and it wasn't possible to call these functions in the way that mac_select_pcs could be called. So, I think it's time to get away from this shimming model and instead have drivers directly interface to the various subsystems. This converts mv88e6xxx. I have similar patches for other DSA drivers that will be sent once this has been reviewed. ==================== Link: https://lore.kernel.org/r/ZhbrbM+d5UfgafGp@shell.armlinux.org.ukSigned-off-by: -
Russell King (Oracle) authored
Convert mv88e6xxx to provide its own phylink MAC operations, thus avoiding the shim layer in DSA's port.c Reviewed-by:
Florian Fainelli <florian.fainelli@broadcom.com> Signed-off-by:
Vladimir Oltean <olteanv@gmail.com> Reviewed-by:
-
Russell King (Oracle) authored
Rather than having a shim for each and every phylink MAC operation, allow DSA switch drivers to provide their own ops structure. When a DSA driver provides the phylink MAC operations, the shimmed ops must not be provided, so fail an attempt to register a switch with both the phylink_mac_ops in struct dsa_switch and the phylink_mac_* operations populated in dsa_switch_ops populated. Signed-off-by:
-
Russell King (Oracle) authored
We convert from a phylink_config struct to a dsa_port struct in many places, let's provide a helper for this. Reviewed-by:
-
Colin Ian King authored
The variable decrypted is being assigned a value that is never read, the control of flow after the assignment is via an return path and decrypted is not referenced in this path. The assignment is redundant and can be removed. Cleans up clang scan warning: net/tls/tls_sw.c:2150:4: warning: Value stored to 'decrypted' is never read [deadcode.DeadStores] Signed-off-by:
Colin Ian King <colin.i.king@gmail.com> Link: https://lore.kernel.org/r/20240410144136.289030-1-colin.i.king@gmail.comSigned-off-by:
-
Guillaume Nault authored
RTO_ONLINK was a flag used in ->flowi4_tos that allowed to alter the scope of an IPv4 route lookup. Setting this flag was equivalent to specifying RT_SCOPE_LINK in ->flowi4_scope. With commit ec20b283 ("ipv4: Set scope explicitly in ip_route_output()."), the last users of RTO_ONLINK have been removed. Therefore, we can now drop the code that checked this bit and stop modifying ->flowi4_scope in ip_route_output_key_hash(). Signed-off-by:
Guillaume Nault <gnault@redhat.com> Reviewed-by:
Przemek Kitszel <przemyslaw.kitszel@intel.com> Reviewed-by:
David Ahern <dsahern@kernel.org> Link: https://lore.kernel.org/r/57de760565cab55df7b129f523530ac6475865b2.1712754146.git.gnault@redhat.comSigned-off-by:
-
Jon Maloy authored
When reading received messages from a socket with MSG_PEEK, we may want to read the contents with an offset, like we can do with pread/preadv() when reading files. Currently, it is not possible to do that. In this commit, we add support for the SO_PEEK_OFF socket option for TCP, in a similar way it is done for Unix Domain sockets. In the iperf3 log examples shown below, we can observe a throughput improvement of 15-20 % in the direction host->namespace when using the protocol splicer 'pasta' (https://passt.top). This is a consistent result. pasta(1) and passt(1) implement user-mode networking for network namespaces (containers) and virtual machines by means of a translation layer between Layer-2 network interface and native Layer-4 sockets (TCP, UDP, ICMP/ICMPv6 echo). Received, pending TCP data to the container/guest is kept in kernel buffers until acknowledged, so the tool routinely needs to fetch new data from socket, skipping data that was already sent. At the moment this is implemented using a dummy buffer passed to recvmsg(). With this change, we don't need a dummy buffer and the related buffer copy (copy_to_user()) anymore. passt and pasta are supported in KubeVirt and libvirt/qemu. jmaloy@freyr:~/passt$ perf record -g ./pasta --config-net -f SO_PEEK_OFF not supported by kernel. jmaloy@freyr:~/passt# iperf3 -s ----------------------------------------------------------- Server listening on 5201 (test #1) ----------------------------------------------------------- Accepted connection from 192.168.122.1, port 44822 [ 5] local 192.168.122.180 port 5201 connected to 192.168.122.1 port 44832 [ ID] Interval Transfer Bitrate [ 5] 0.00-1.00 sec 1.02 GBytes 8.78 Gbits/sec [ 5] 1.00-2.00 sec 1.06 GBytes 9.08 Gbits/sec [ 5] 2.00-3.00 sec 1.07 GBytes 9.15 Gbits/sec [ 5] 3.00-4.00 sec 1.10 GBytes 9.46 Gbits/sec [ 5] 4.00-5.00 sec 1.03 GBytes 8.85 Gbits/sec [ 5] 5.00-6.00 sec 1.10 GBytes 9.44 Gbits/sec [ 5] 6.00-7.00 sec 1.11 GBytes 9.56 Gbits/sec [ 5] 7.00-8.00 sec 1.07 GBytes 9.20 Gbits/sec [ 5] 8.00-9.00 sec 667 MBytes 5.59 Gbits/sec [ 5] 9.00-10.00 sec 1.03 GBytes 8.83 Gbits/sec [ 5] 10.00-10.04 sec 30.1 MBytes 6.36 Gbits/sec - - - - - - - - - - - - - - - - - - - - - - - - - [ ID] Interval Transfer Bitrate [ 5] 0.00-10.04 sec 10.3 GBytes 8.78 Gbits/sec receiver ----------------------------------------------------------- Server listening on 5201 (test #2) ----------------------------------------------------------- ^Ciperf3: interrupt - the server has terminated jmaloy@freyr:~/passt# logout [ perf record: Woken up 23 times to write data ] [ perf record: Captured and wrote 5.696 MB perf.data (35580 samples) ] jmaloy@freyr:~/passt$ jmaloy@freyr:~/passt$ perf record -g ./pasta --config-net -f SO_PEEK_OFF supported by kernel. jmaloy@freyr:~/passt# iperf3 -s ----------------------------------------------------------- Server listening on 5201 (test #1) ----------------------------------------------------------- Accepted connection from 192.168.122.1, port 52084 [ 5] local 192.168.122.180 port 5201 connected to 192.168.122.1 port 52098 [ ID] Interval Transfer Bitrate [ 5] 0.00-1.00 sec 1.32 GBytes 11.3 Gbits/sec [ 5] 1.00-2.00 sec 1.19 GBytes 10.2 Gbits/sec [ 5] 2.00-3.00 sec 1.26 GBytes 10.8 Gbits/sec [ 5] 3.00-4.00 sec 1.36 GBytes 11.7 Gbits/sec [ 5] 4.00-5.00 sec 1.33 GBytes 11.4 Gbits/sec [ 5] 5.00-6.00 sec 1.21 GBytes 10.4 Gbits/sec [ 5] 6.00-7.00 sec 1.31 GBytes 11.2 Gbits/sec [ 5] 7.00-8.00 sec 1.25 GBytes 10.7 Gbits/sec [ 5] 8.00-9.00 sec 1.33 GBytes 11.5 Gbits/sec [ 5] 9.00-10.00 sec 1.24 GBytes 10.7 Gbits/sec [ 5] 10.00-10.04 sec 56.0 MBytes 12.1 Gbits/sec - - - - - - - - - - - - - - - - - - - - - - - - - [ ID] Interval Transfer Bitrate [ 5] 0.00-10.04 sec 12.9 GBytes 11.0 Gbits/sec receiver ----------------------------------------------------------- Server listening on 5201 (test #2) ----------------------------------------------------------- ^Ciperf3: interrupt - the server has terminated logout [ perf record: Woken up 20 times to write data ] [ perf record: Captured and wrote 5.040 MB perf.data (33411 samples) ] jmaloy@freyr:~/passt$ The perf record confirms this result. Below, we can observe that the CPU spends significantly less time in the function ____sys_recvmsg() when we have offset support. Without offset support: ---------------------- jmaloy@freyr:~/passt$ perf report -q --symbol-filter=do_syscall_64 \ -p ____sys_recvmsg -x --stdio -i perf.data | head -1 46.32% 0.00% passt.avx2 [kernel.vmlinux] [k] do_syscall_64 ____sys_recvmsg With offset support: ---------------------- jmaloy@freyr:~/passt$ perf report -q --symbol-filter=do_syscall_64 \ -p ____sys_recvmsg -x --stdio -i perf.data | head -1 28.12% 0.00% passt.avx2 [kernel.vmlinux] [k] do_syscall_64 ____sys_recvmsg Suggested-by:
Paolo Abeni <pabeni@redhat.com> Reviewed-by:
Stefano Brivio <sbrivio@redhat.com> Signed-off-by:
Jon Maloy <jmaloy@redhat.com> Reviewed-by:
-
Breno Leitao authored
Commit 3e2f544d ("net: get stats64 if device if driver is configured") moved the callback to dev_get_tstats64() to net core, so, unless the driver is doing some custom stats collection, it does not need to set .ndo_get_stats64. Since this driver is now relying in NETDEV_PCPU_STAT_TSTATS, then, it doesn't need to set the dev_get_tstats64() generic .ndo_get_stats64 function pointer. Signed-off-by:
Breno Leitao <leitao@debian.org> Link: https://lore.kernel.org/r/20240409133307.2058099-2-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 in 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 qmi_wwan driver and leverage the network core allocation instead. Signed-off-by:
-
Eric Dumazet authored
- Use for_each_netdev_dump() to no longer rely on net->dev_index_head hash table. - No longer care of net->dev_base_seq - Fix return value at the end of a dump, so that NLMSG_DONE can be appended to current skb, saving one recvmsg() system call. - No longer grab RTNL, RCU protection is enough, afer adding one READ_ONCE(mdev->input_enabled) in mpls_netconf_fill_devconf() Signed-off-by:
-
Asbjørn Sloth Tønnesen authored
include/net/flow_offload.h:351: warning: No description found for return value of 'flow_offload_has_one_action' Signed-off-by:
Asbjørn Sloth Tønnesen <ast@fiberby.net> Reviewed-by:
Pieter Jansen van Vuuren <pieter.jansen-van-vuuren@amd.com> Link: https://lore.kernel.org/r/20240410114718.15145-1-ast@fiberby.netSigned-off-by:
-
Carolina Jubran authored
Q counters are device-level counters that track specific events, among which are out_of_buffer events. These events occur when packets are dropped due to a lack of receive buffer in the RX queue. Expose the total number of out_of_buffer events on the VFs/SFs to their respective representor, using the "ip stats group link" under the name of "rx_missed". The "rx_missed" equals the sum of all Q counters out_of_buffer values allocated on the VFs/SFs. Signed-off-by:
Carolina Jubran <cjubran@nvidia.com> Reviewed-by:
Gal Pressman <gal@nvidia.com> Reviewed-by:
Aya Levin <ayal@nvidia.com> Reviewed-by:
Rahul Rameshbabu <rrameshbabu@nvidia.com> Signed-off-by:
Tariq Toukan <tariqt@nvidia.com> Reviewed-by:
-
Jakub Kicinski authored
Mina Almasry says: ==================== Minor cleanups to skb frag ref/unref This series is largely motivated by a recent discussion where there was some confusion on how to properly ref/unref pp pages vs non pp pages: https://lore.kernel.org/netdev/CAHS8izOoO-EovwMwAm9tLYetwikNPxC0FKyVGu1TPJWSz4bGoA@mail.gmail.com/T/#t There is some subtely there because pp uses page->pp_ref_count for refcounting, while non-pp uses get_page()/put_page() for ref counting. Getting the refcounting pairs wrong can lead to kernel crash. Additionally currently it may not be obvious to skb users unaware of page pool internals how to properly acquire a ref on a pp frag. It requires checking of skb->pp_recycle & is_pp_page() to make the correct calls and may require some handling at the call site aware of arguable pp internals. This series is a minor refactor with a couple of goals: 1. skb users should be able to ref/unref a frag using [__]skb_frag_[un]ref() functions without needing to understand pp concepts and pp_ref_count vs get/put_page() differences. 2. reference counting functions should have a mirror opposite. I.e. there should be a foo_unref() to every foo_ref() with a mirror opposite implementation (as much as possible). This is RFC to collect feedback if this change is desirable, but also so that I don't race with the fix for the issue Dragos is seeing for his crash. https://lore.kernel.org/lkml/CAHS8izN436pn3SndrzsCyhmqvJHLyxgCeDpWXA4r1ANt3RCDLQ@mail.gmail.com/T/ ==================== Link: https://lore.kernel.org/r/20240410190505.1225848-1-almasrymina@google.comSigned-off-by:
-
Mina Almasry authored
Refactor some of the skb frag ref/unref helpers for improved clarity. Implement napi_pp_get_page() to be the mirror counterpart of napi_pp_put_page(). Implement skb_page_ref() to be the mirror of skb_page_unref(). Improve __skb_frag_ref() to become a mirror counterpart of __skb_frag_unref(). Previously unref could handle pp & non-pp pages, while the ref could only handle non-pp pages. Now both the ref & unref helpers can correctly handle both pp & non-pp pages. Now that __skb_frag_ref() can handle both pp & non-pp pages, remove skb_pp_frag_ref(), and use __skb_frag_ref() instead. This lets us remove pp specific handling from skb_try_coalesce. Additionally, since __skb_frag_ref() can now handle both pp & non-pp pages, a latent issue in skb_shift() should now be fixed. Previously this function would do a non-pp ref & pp unref on potential pp frags (fragfrom). After this patch, skb_shift() should correctly do a pp ref/unref on pp frags. Signed-off-by:
Mina Almasry <almasrymina@google.com> Reviewed-by:
Dragos Tatulea <dtatulea@nvidia.com> Reviewed-by:
-
Mina Almasry authored
Add a new header, linux/skbuff_ref.h, which contains all the skb_*_ref() helpers. Many of the consumers of skbuff.h do not actually use any of the skb ref helpers, and we can speed up compilation a bit by minimizing this header file. Additionally in the later patch in the series we add page_pool support to skb_frag_ref(), which requires some page_pool dependencies. We can now add these dependencies to skbuff_ref.h instead of a very ubiquitous skbuff.h Signed-off-by:
-
