Spinlock can be released twice in procsignal.c

From: "Maksim(dot)Melnikov" <m(dot)melnikov(at)postgrespro(dot)ru>
To: pgsql-hackers(at)lists(dot)postgresql(dot)org
Subject: Spinlock can be released twice in procsignal.c
Date: 2025-02-25 19:34:32
Message-ID: dca47527-2d8b-4e3b-b5a0-e2deb73371a4@postgrespro.ru
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi, it seems we can release spinlock twice in
/src/backend/storage/ipc/procsignal.c file, method ProcSignalInit.

void
ProcSignalInit(bool cancel_key_valid, int32 cancel_key)
{
    ProcSignalSlot *slot;
    uint64        barrier_generation;

..............................................................................

    slot = &ProcSignal->psh_slot[MyProcNumber];

    /* sanity check */
    SpinLockAcquire(&slot->pss_mutex);
    if (pg_atomic_read_u32(&slot->pss_pid) != 0)
    {
*SpinLockRelease(&slot->pss_mutex);*
        elog(LOG, "process %d taking over ProcSignal slot %d, but it's
not empty",
             MyProcPid, MyProcNumber);
    }

    /* Clear out any leftover signal reasons */
    MemSet(slot->pss_signalFlags, 0, NUM_PROCSIGNALS *
sizeof(sig_atomic_t));

......................

    slot->pss_cancel_key_valid = cancel_key_valid;
    slot->pss_cancel_key = cancel_key;
    pg_atomic_write_u32(&slot->pss_pid, MyProcPid);

*SpinLockRelease(&slot->pss_mutex);*

First in the if clause, second near the end of function. Such behavior
can lead to unpredictable concurrent issues.

In applied patch I removed spinlock release in if clause.

Attachment Content-Type Size
0001-Spinlock-can-be-released-twice-in-procsignal.c.patch text/x-patch 842 bytes

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Ekaterina Sokolova 2025-02-25 19:44:33 Proposal: Limitations of palloc inside checkpointer
Previous Message Masahiko Sawada 2025-02-25 18:58:39 Re: Restrict copying of invalidated replication slots