[Bug 2165140] Re: qrtr: ns: node limit of 64 breaks QRTR routing on large multi-node deployments

** Description changed: SRU Justification: - [ Impact ] + [Impact] -  * Commit 27d5e84e810b ("net: qrtr: ns: Limit the total number of nodes") -    introduced a hard cap of 64 on the number of QRTR nodes the namespace -    server (qrtr_ns) will track, including the host node itself. -    That limit was set as a DoS mitigation but is too low for legitimate -    large-scale deployments, such as Qualcomm AI200 systems, which register -    up to 384 QRTR nodes across their SoCs/cards. + * Commit 27d5e84e810b ("net: qrtr: ns: Limit the total number of nodes") + introduced a hard cap of 64 on the number of QRTR nodes the namespace + server (qrtr_ns) will track, including the host node itself. + That limit was set as a DoS mitigation but is too low for legitimate + large-scale deployments, such as Qualcomm AI200 systems, which register + up to 384 QRTR nodes across their SoCs/cards. -  * This patch fixes this issue by increasing the limit to 512. + * This patch fixes this issue by increasing the limit to 512. -  * This patch should be backported because it affects deployments with 64 -    or more Qualcomm AI100/AI200 targets. - - [ Test Plan ] - -  * Checking dmesg on an Ubuntu 24.04 system with at least 6.8.0-136-generic, -    having at least 64 remote QRTR nodes, it will show "QRTR clients exceed max -    node limit!" for every extra node. -  * Also, by running a QRTR client then sending a QRTR_TYPE_NEW_LOOKUP packet, -    only 64 nodes will show up (63 remote node + the local node). -  * After applying the fix, dmesg should no longer show the error message, and -    the lookup client will return a full list of nodes. + * This patch should be backported because it affects deployments with 64 + or more Qualcomm AI100/AI200 targets. [Fix] -  * Cherry-pick upstream commit: -    ff194cffd586 ("net: qrtr: ns: Raise node count limit to 512") + * Backport of upstream commit: + ff194cffd586 ("net: qrtr: ns: Raise node count limit to 512") - [ Where problems could occur ] + [Test Plan] -  * If a deployment exceeds 512 nodes, the original symptom returns identically -    (same log message, same rejection behavior). + * Checking dmesg on an Ubuntu 24.04 system with at least 6.8.0-136-generic, + having at least 64 remote QRTR nodes, it will show "QRTR clients exceed max + node limit!" for every extra node. + * Also, by running a QRTR client then sending a QRTR_TYPE_NEW_LOOKUP packet, + only 64 nodes will show up (63 remote node + the local node). + * After applying the fix, dmesg should no longer show the error message, and + the lookup client will return a full list of nodes. + + [Where problems could occur] + + * If a deployment exceeds 512 nodes, the original symptom returns identically + (same log message, same rejection behavior). -- You received this bug notification because you are subscribed to linux in Ubuntu. Matching subscriptions: Bgg, Bmail, Nb https://bugs.launchpad.net/bugs/2165140 Title: qrtr: ns: node limit of 64 breaks QRTR routing on large multi-node deployments Status in linux package in Ubuntu: New Bug description: SRU Justification: [Impact] * Commit 27d5e84e810b ("net: qrtr: ns: Limit the total number of nodes") introduced a hard cap of 64 on the number of QRTR nodes the namespace server (qrtr_ns) will track, including the host node itself. That limit was set as a DoS mitigation but is too low for legitimate large-scale deployments, such as Qualcomm AI200 systems, which register up to 384 QRTR nodes across their SoCs/cards. * This patch fixes this issue by increasing the limit to 512. * This patch should be backported because it affects deployments with 64 or more Qualcomm AI100/AI200 targets. [Fix] * Backport of upstream commit: ff194cffd586 ("net: qrtr: ns: Raise node count limit to 512") [Test Plan] * Checking dmesg on an Ubuntu 24.04 system with at least 6.8.0-136-generic, having at least 64 remote QRTR nodes, it will show "QRTR clients exceed max node limit!" for every extra node. * Also, by running a QRTR client then sending a QRTR_TYPE_NEW_LOOKUP packet, only 64 nodes will show up (63 remote node + the local node). * After applying the fix, dmesg should no longer show the error message, and the lookup client will return a full list of nodes. [Where problems could occur] * If a deployment exceeds 512 nodes, the original symptom returns identically (same log message, same rejection behavior). To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2165140/+subscriptions