Re: [PATCH 3/3] zram: add mem_used_max via sysfs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Aug 14, 2014 at 11:32 AM, David Horner <ds2horner@xxxxxxxxx> wrote:
> On Thu, Aug 14, 2014 at 11:09 AM, Dan Streetman <ddstreet@xxxxxxxx> wrote:
>> On Wed, Aug 13, 2014 at 9:12 PM, Minchan Kim <minchan@xxxxxxxxxx> wrote:
>>> -       if (zram->limit_bytes &&
>>> -               zs_get_total_size_bytes(meta->mem_pool) > zram->limit_bytes) {
>>> +       total_bytes = zs_get_total_size_bytes(meta->mem_pool);
>>> +       if (zram->limit_bytes && total_bytes > zram->limit_bytes) {
>>
>> do you need to take the init_lock to read limit_bytes here?  It could
>> be getting changed between these checks...
>
> There is no real danger in freeing with an error.
> It is more timing than a race.
I probably should explain my reasoning.

any changes between getting the total value and the limit test are not
problematic (From race perspective).

1) If the actual total increases and the value returned under rates it, then
a) if this.total exceeds the limit - no problem it is rolled back as
it would if the actual total were used.
b) if this.total <= limit OK - as other process will be dinged (it
will see its own allocation)

2)  If the actual total decreases and the value returned overrates
rates it, then
a) if this.value <= limit then allocation great (actual has even more room)
b) if this.value > max it will be rolled back (as the other might be
as well) and process can compete again.

Is there a denial of service possible if 2.b repeats indefinitely.
Yes, but how to set it up reliably? And it is no different than a
single user exhausting the limit before any other users.
Yes, it is potentially a live false limit exhaustion, with one process
requesting an amount exceeding the limit but able to be allocated.
 But this is no worse than the rollback load we already have at the limit.

It would be better to check before the zs_malloc if the concern is
avoiding heavy processing in that function (as an optimization),  as
well as here.after allocation

But I see no real race or danger doing this unlocked.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]