Upgrade to Pro — share decks privately, control downloads, hide ads and more …

CRIUとSeccomp / criu-and-seccomp-and-me

CRIUとSeccomp / criu-and-seccomp-and-me

第14回 コンテナ技術の情報交換会@オンライン
https://ct-study.connpass.com/event/205571

Avatar for KONDO Uchio

KONDO Uchio

April 17, 2021
Tweet

More Decks by KONDO Uchio

Other Decks in Technology

Transcript

  1. ۙ౻Ӊஐ࿕ / GMO Pepabo, Inc. ୈ14ճ ίϯςφٕज़ͷ৘ใަ׵ձ@ΦϯϥΠϯ 2021/04/17 CRIUͱseccompͱ ֨ಆͨ͠࿩

    ʙશͯ͸ര଎ىಈͷͨΊͩͬͨʙ *NBHFIUUQTQJYBCBZDPNJNBHFTJECZ"OUSBOJBT
  2. γχΞɾϓϦϯγύϧΤϯδχΞ ۙ౻ Ӊஐ࿕ / @udzura https://blog.udzura.jp/ Uchio Kondo ٕज़෦ ٕज़ج൫νʔϜ

    @ GMOϖύϘ ΤϯδχΞΧϑΣʢ෱Ԭࢢ੺ẂנจԽձؗʣ αϙʔλʔ #Ruby #mruby #Rust #Containers #eBPF #CRIU #Seccomp #RubyKaigi #CloudNativeDays #Zumba #γϨϯ #FitBoxing2 --- ޷͖ͳγεςϜίʔϧʁ ΍ͬͺΓ unshare(2) Ͱ͢Ͷɻ
  3. ࠷ۙͷCRIU •3.13(Sep 11, 2019) ... libcriu.a ͕ϏϧυͰ͖ΔΑ͏ʹͳͬͨ by @udzura •3.14(π,

    Apr 29, 2020) ... clone3(2)ͱTime NS support, ଞ •3.15(Nov 04, 2020) ... MIPS support, cgroup v2 support, PID NS಺෦ ΁ͷϦετΞ, ... ଞ •Still developing... ຊൃදͷ$3*6ͷόʔδϣϯ͸Ͱ͢
  4. Mount namespace/root ͷ෼཭ •/proc ΛϚ΢ϯτ͠௚͢ͷͰMount NS΋unshared •ಠཱͨ͠ɺ͔ͭϗετͱେମಉ͡root filesystem͕ඞཁ •࡞Γํ: •ԾͷσΟϨΫτϦʹ

    / Λbind mountɺ/devͳͲ΋ݸผʹbind mount •ͦ͜ʹpivot_root͢Δ(chroot ͸μϝɺͦͷMount NSͰೝࣝ͞ΕΔ rootΛૠ͛ସ͑Δඞཁ͕͋ΔͨΊ)
  5. ࣮૷ͷΠϝʔδ /&8/4 /&81*%ͷϑϥάΛ༩͑ͯDMPOF  )PTUSPPU  ΛผͷՕॴʹCJOENPVOU TFUTJE QJWPU@SPPU GEΛEFWOVMM

    GE ΛݱࡏͷSPPU಺෦ͷϑΝΠϧʹ͠ 0@830/-:c0@"11&/%Ͱ։͘ ର৅ϓϩάϥϜʹFYFD
  6. miehistö ͰͷϦετΞ࣮૷ •miehistod ͷԼͰ criu restore ΛݺͿ •ͦͷࡍʹɺ --exec-cmd ͱ͍͏ΦϓγϣϯΛར༻͠ɺ

    ϦετΞޙʹcriuίϚϯυࣗ਎ΛrunmhϓϩάϥϜʹexec͢Δ • ਌ΛrunmhɺࢠΛϦετΞޙͷίϚϯυͱ͍͏ϓϩηεπϦʔΛ࡞੒͍ͨͨ͠Ί •·ͨɺbind mountͷѻ͍ʹ͍ͭͯcriu restoreʹ౉͢ඞཁ͕͋Δ •--external Φϓγϣϯʹ͍ͭͯ
  7. External bind mounts •root͔Βݟͯ֎෦σΟϨΫτϦΛBind mount͍ͯ͠Δ৘ใ͸ɺdump/ restore࣌ʹࣗಈݕ஌͞Εͳ͍ͷͰɺ໌ࣔతʹࢦఆ͢Δඞཁ͕͋Δɻ •dump࣌ʹˠ ֎෦Bind mountઌΛن໿ͰܾΊ͍ͯΔͷͰɺͦΕΒͷ ৘ใΛ

    --external ૬౰ͷΦϓγϣϯͱͯ͠criuαʔϏεʹ౉͢ •restore࣌ʹˠ ҎԼͷΑ͏ͳܗࣜͰίϚϯυʹ౉͢ --external mnt[__pids-d05739ddd0a12dd5040494c20461a197]:/sys/fs/cgroup/pids IUUQTDSJVPSH&YUFSOBM@CJOE@NPVOUT
  8. SCMP_ACT_TRACE ࢖͍ํ • fork͢Δ • [਌] ptrace(PTRACE_ATTACH); ptrace(PTRACE_O_TRACESECCOMP) ͢Δ •

    [਌] ptrace(PTRACE_CONT) ͢Δ • [਌] waitpid(-1, &status, WUNTRACED | WCONTINUED | __WALL) ͢Δ (-1 ͳͷ͸traceͯ͠Δ͞ΒʹࢠϓϩηεɾεϨουΛ௥͏ͨΊ) • [ࢠ] SCMP_ACT_TRACE ͳseccomp ctxΛϩʔυ͢Δ • [਌] ࢠͰ౰֘γεςϜίʔϧ͕ݺ͹ΕΔͱɺ౰֘ϓϩηεͷ৘ใ͕waitpidͷ໭Γ஋ɺ ptrace(PTRACE_GETEVENTMSG)ɺptrace(PTRACE_GETREGS)ͳͲͰऔಘͰ͖Δ
  9. ptrace ͷΦϓγϣϯ͕όοςΟϯά͢Δʁ •seccompͰͷτϨʔε͸ϓϩηεΛ ptrace(PTRACE_ATTACH) ͯ͠Ξ λονঢ়ଶʹ͠ɺఀࢭঢ়ଶΛݕ஌͢Δ •Ұํɺcriuͷ಺෦Ͱ͸ɺ ptrace(PTRACE_SEIZE) ͰΞλονঢ়ଶʹͯ͠ ptrace(PTRACE_INTERRUPT)

    Ͱ໌ࣔతʹࢭΊ͍ͯΔɻ •ผʑͷΦϓγϣϯͰࢭΊ͍ͯΔͷ͸໰୊ʹͳ͍ͬͯΔՄೳੑ͕͋Δ •͜ͷลΓͷύονΛແཧʹcriuʹ౰ͯͯ΋... • ݁ہseccomp ctx͕ϓϩηε͔Βൈ͚ͳ͍ͨΊɺ2ճ໨ͷ listen() ͕ptrace tracee͕ଘࡏ͠ͳ͍ѻ͍ʹͳΓENOSYS ʹͳΔ?
  10. ࢀߟهࣄͳͲ •அଓతʹॻ͍ͯΔseccompͱMiehistöͷϒϩά • ʮmrubyͱseccompͱptraceͰγεςϜίʔϧΛͱʹ͔͘௥͍͔͚Δʯ(2017/04) • ʮGrenadine: ʮී௨ͷΞϓϦέʔγϣϯʯ͕νΣοΫϙΠϯτ/ϦετΞͷԸܙΛڗड͢Δʯ (2019/03) • ʮWSAݚڀձ

    ୈ7ճ ͰCRIUͱMiehistöͷൃදΛ͠·ͨ͠ #WSAݚʯ(2020/11) • ʮ೚ҙͷϥΠϒϥϦίʔϧͰϓϩάϥϜΛఀࢭ͠ɺىಈ༻ͷCRIUΠϝʔδΛ࡞੒͢ΔΞϓ ϩʔνʹ͍ͭͯʯ(2020/12) IUUQTVE[VSBIBUFOBCMPHKQ