[lkml]   [2020]   [Jan]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] mm/page-writeback.c: avoid potential division by zero

On 2020/1/1 8:39 下午, Qian Cai wrote:
>> On Jan 1, 2020, at 4:32 AM, Wen Yang <> wrote:
>> The variables 'min', 'max' and 'bw' are unsigned long and
>> do_div truncates them to 32 bits, which means it can test
>> non-zero and be truncated to zero for division.
>> Fix this issue by using div64_ul() instead.
> How did you find out the issue? If it is caught by compilers, can you paste the original warnings? Also, can you figure out which commit introduced the issue in the first place, so it could be backported to stable if needed?

Thanks for your comments.
There are no compilation warnings here.

We found this issue by following these steps:
We were first inspired by commit b0ab99e7736a ("sched: Fix possible
divide by zero in avg_atom () calculation"), combined with our recently
analyzed mm code, we found this suspicious place.

And we also disassembled and confirmed it:

201 if (min) {
202 min *= this_bw;
203 do_div(min, tot_bw);
204 }

0xffffffff811c37da <__wb_calc_thresh+234>: xor %r10d,%r10d
0xffffffff811c37dd <__wb_calc_thresh+237>: test %rax,%rax
0xffffffff811c37e0 <__wb_calc_thresh+240>: je
0xffffffff811c3800 <__wb_calc_thresh+272>
0xffffffff811c37e2 <__wb_calc_thresh+242>: imul %r8,%rax
0xffffffff811c37e6 <__wb_calc_thresh+246>: mov %r9d,%r10d
---> truncates it to 32 bits here

0xffffffff811c37e9 <__wb_calc_thresh+249>: xor %edx,%edx
0xffffffff811c37eb <__wb_calc_thresh+251>: div %r10
0xffffffff811c37ee <__wb_calc_thresh+254>: imul %rbx,%rax
0xffffffff811c37f2 <__wb_calc_thresh+258>: shr $0x2,%rax
0xffffffff811c37f6 <__wb_calc_thresh+262>: mul %rcx
0xffffffff811c37f9 <__wb_calc_thresh+265>: shr $0x2,%rdx
0xffffffff811c37fd <__wb_calc_thresh+269>: mov %rdx,%r10

This issue was introduced by commit 693108a8a667 (“writeback: make
bdi->min/max_ratio handling cgroup writeback aware”).

Finally, we will summarize the above cases and plan to write a general
coccinelle rule to check for similar problems.

>> For the two variables 'numerator' and 'denominator',
>> though they are declared as long, they should actually be
>> unsigned long (according to the implementation of
>> the fprop_fraction_percpu() function).

 \ /
  Last update: 2020-01-02 04:58    [W:0.054 / U:27.836 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site