回复: Fix segfault while accessing half-initialized hash table in pgstat_shmem.c

Hi, Mikhail,

If pgstat_init_entry() errors on OOM, what would it returns to shheader, NULL?
That would bring trouble to dshash_delete_entry().

Thanks,
Steven

________________________________
发件人: Mikhail Kot <mikhail(dot)kot(at)databricks(dot)com>
已发送: 2025 年 9 月 04 日 星期四 05:39
收件人: pgsql-hackers(at)lists(dot)postgresql(dot)org <pgsql-hackers(at)lists(dot)postgresql(dot)org>
抄送: to(at)myrrc(dot)dev <to(at)myrrc(dot)dev>
主题: Re: Fix segfault while accessing half-initialized hash table in pgstat_shmem.c

Hi Michael, Steven, and Ranier,

> Anyway, couldn't we flip the order of the operations in
pgstat_init_entry() so as we do first an allocation and avoid any inconsistency
in the shared state?

The issue is not only in pgstat_init_entry(). Currently it errors on OOM but
this doesn't prevent us from calling pgstat_lock_entry() through
pgstat_get_entry_ref() which accesses a non-initialized lock.

Here's the second version of the patch. Now we remove inserted hash entry
on OOM which would prevent accessing the entry