[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20231005162417.GA32420@redhat.com>
Date: Thu, 5 Oct 2023 18:24:17 +0200
From: Oleg Nesterov <oleg@...hat.com>
To: Li Nan <linan666@...weicloud.com>
Cc: Khazhy Kumykov <khazhy@...omium.org>, tj@...nel.org,
josef@...icpanda.com, axboe@...nel.dk, yukuai3@...wei.com,
cgroups@...r.kernel.org, linux-block@...r.kernel.org,
linux-kernel@...r.kernel.org, yi.zhang@...wei.com,
houtao1@...wei.com, yangerkun@...wei.com
Subject: Re: [PATCH] blk-throttle: Calculate allowed value only when the
throttle is enabled
Hi Li,
On 10/05, Li Nan wrote:
>
> >I don't think this change is sufficient to prevent kernel crash, as a
> >"clever" user could still set the bps_limit to U64_MAX - 1 (or another
> >large value), which probably would still result in the same crash. The
> >comment in mul_u64_u64_div_u64 suggests there's something we can do to
> >better handle the overflow case, but I'm not sure what it's referring
> >to. ("Will generate an #DE when the result doesn't fit u64, could fix
> >with an __ex_table[] entry when it becomes an issue.") Otherwise, we
>
> When (a * mul) overflows, a divide 0 error occurs in
> mul_u64_u64_div_u64(). Commit 3dc167ba5729 ("sched/cputime: Improve
> cputime_adjust()") changed func and said: "Will generate an #DE when the
> result doesn't fit u64, could fix with an __ex_table[] entry when it
> becomes an issue." But we are unsure of how to fix it. Could you please
> explain how to fix this issue.
Not sure I understand the question...
OK, we can change mul_u64_u64_div_u64() to trap the exception, say,
static inline u64 mul_u64_u64_div_u64(u64 a, u64 mul, u64 div)
{
u64 q;
asm ("mulq %2; 1: divq %3; 2:\n"
_ASM_EXTABLE_TYPE(1b, 2b, EX_TYPE_DEFAULT|EX_FLAG_CLEAR_AX)
: "=a" (q)
: "a" (a), "rm" (mul), "rm" (div)
: "rdx");
return q;
}
should (iiuc) return 0 if the result doesn't fit u64 or div == 0.
But even if we forget that this is x86-specific, how can this help?
What should calculate_bytes_allowed() do/return in this case?
> >probably need to remove the mul_u64_u64_div_u64 and check for
> >overflow/potential overflow ourselves?
probably yes...
Oleg.
Powered by blists - more mailing lists