Hi, Some time recently valgrind suppressions in the older backbranches stopped fully working. E.g.
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=skink&dt=2025-03-04%2022%3A31%3A25 failed on 14, with: 2025-03-04 22:33:13.359 UTC [3534679][postmaster][:0] LOG: database system is ready to accept connections ==3534881== VALGRINDERROR-BEGIN ==3534881== Syscall param socketcall.sendto(msg) points to uninitialised byte(s) ==3534881== at 0x4D975C7: __internal_syscall_cancel (cancellation.c:64) ==3534881== by 0x4D975EC: __syscall_cancel (cancellation.c:75) ==3534881== by 0x4E1A1F6: send (send.c:28) ==3534881== by 0x4F32DC: pgstat_send (pgstat.c:2970) ==3534881== by 0x4F33C5: pgstat_send_inquiry (pgstat.c:1889) ==3534881== by 0x4F5508: backend_read_statsfile (pgstat.c:4650) ==3534881== by 0x4F7812: pgstat_fetch_stat_dbentry (pgstat.c:2710) ==3534881== by 0x4EB3AB: rebuild_database_list (autovacuum.c:1055) ==3534881== by 0x4EBF2F: AutoVacLauncherMain (autovacuum.c:635) ==3534881== by 0x4EC2B1: StartAutoVacLauncher (autovacuum.c:422) ==3534881== by 0x4FC815: reaper (postmaster.c:3067) ==3534881== by 0x4D47DAF: ??? (in /usr/lib/x86_64-linux-gnu/libc.so.6) ==3534881== Address 0x1ffeffedac is on thread 1's stack ==3534881== in frame #4, created by pgstat_send_inquiry (pgstat.c:1882) ==3534881== Uninitialised value was created by a stack allocation ==3534881== at 0x4F3386: pgstat_send_inquiry (pgstat.c:1882) ==3534881== ==3534881== VALGRINDERROR-END { <insert_a_suppression_name_here> Memcheck:Param socketcall.sendto(msg) fun:__internal_syscall_cancel fun:__syscall_cancel fun:send fun:pgstat_send fun:pgstat_send_inquiry fun:backend_read_statsfile fun:pgstat_fetch_stat_dbentry fun:rebuild_database_list fun:AutoVacLauncherMain fun:StartAutoVacLauncher fun:reaper obj:/usr/lib/x86_64-linux-gnu/libc.so.6 } I was confused by this for a while, because we seem to have a suppression for it: { padding_pgstat_sendto Memcheck:Param socketcall.sendto(msg) fun:*send* fun:pgstat_send } An embarassing amount of staring later I realized this must be due to the cancellation related functions in the stack trace. I think all we need to do is to add a ... line to the two relevant suppressions, similar to what we do for other suppressions. See attached. I don't think we need it for other suppressions, they either already use ... or aren't below a syscall. Greetings, Andres Freund
>From 44200a64f3fd5843c5041d13d66ac89e0d44bb8a Mon Sep 17 00:00:00 2001 From: Andres Freund <and...@anarazel.de> Date: Wed, 5 Mar 2025 16:21:40 -0500 Subject: [PATCH v1] valgrind: Adjust suppressions to handle glibc changes In newer glibc versions two additional functions appear between send() and socketcall.send(msg): fun:__internal_syscall_cancel fun:__syscall_cancel Due to that our existing suppression do not work anymore. The problematic suppressions are only in < 15, as they aren't needed after 5891c7a8ed8. Author: Reviewed-by: Discussion: https://postgr.es/m/ Backpatch-through: 13 --- src/tools/valgrind.supp | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/tools/valgrind.supp b/src/tools/valgrind.supp index e3a179d210b..6caf8593211 100644 --- a/src/tools/valgrind.supp +++ b/src/tools/valgrind.supp @@ -19,6 +19,7 @@ Memcheck:Param socketcall.send(msg) + ... fun:*send* fun:pgstat_send } @@ -28,6 +29,7 @@ Memcheck:Param socketcall.sendto(msg) + ... fun:*send* fun:pgstat_send } -- 2.48.1.76.g4e746b1a31.dirty
