On Sun, Mar 31, 2019 at 10:53 AM Xin Long <lucien.xin@xxxxxxxxx> wrote: > > sctp memory accounting is added in this patchset by using > these kernel APIs on send side: > > - sk_mem_charge() > - sk_mem_uncharge() > - sk_wmem_schedule() > - sk_under_memory_pressure() > - sk_mem_reclaim() > > and these on receive side: > > - sk_mem_charge() > - sk_mem_uncharge() > - sk_rmem_schedule() > - sk_under_memory_pressure() > - sk_mem_reclaim() > > With sctp memory accounting, we can limit the memory allocation by > either sysctl: > > # sysctl -w net.sctp.sctp_mem="10 20 50" > > or cgroup: > > # echo $((8<<14)) > \ > /sys/fs/cgroup/memory/sctp_mem/memory.kmem.tcp.limit_in_bytes > > When the socket is under memory pressure, the send side will block > and wait, while the receive side will renege or drop. > I have tested this series with a tool which creates lot of SCTP sockets and writes into them, without never reading. Before the patch I was able to escape the cgroup limit and fill the system memory, and the OOM killer killed random processes: [ 317.348911] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),global_oom,task_memcg=/system.slice/ifup@eth0.service,task=dhclient,pid=188,uid=0 [ 317.349084] Out of memory: Killed process 188 (dhclient) total-vm:9484kB, anon-rss:1280kB, file-rss:1424kB, shmem-rss:0kB [ 317.743943] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),global_oom,task_memcg=/system.slice/systemd-journald.service,task=systemd-journal,pid=85,uid=0 [ 317.744093] Out of memory: Killed process 85 (systemd-journal) total-vm:24592kB, anon-rss:1024kB, file-rss:1112kB, shmem-rss:652kB [ 317.921049] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),global_oom,task_memcg=/system.slice/cron.service,task=cron,pid=222,uid=0 [ 317.921209] Out of memory: Killed process 222 (cron) total-vm:8692kB, anon-rss:276kB, file-rss:1540kB, shmem-rss:0kB Now the OOM killer behaves correctly and always kills the processes in the right cgroup: [ 512.100054] Tasks state (memory values in pages): [ 512.100122] [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name [ 512.100256] [ 838] 0 838 550 184 36864 0 0 sctprank [ 512.100452] [ 841] 0 841 550 18 36864 0 0 sctprank [ 512.100573] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),oom_memcg=/sctp,task_memcg=/sctp,task=sctprank,pid=838,uid=0 [ 512.100700] Memory cgroup out of memory: Killed process 838 (sctprank) total-vm:2200kB, anon-rss:64kB, file-rss:672kB, shmem-rss:0kB [ 512.100899] oom_reaper: reaped process 838 (sctprank), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB Series ACK. Tested-by: Matteo Croce <mcroce@xxxxxxxxxx> -- Matteo Croce per aspera ad upstream
