[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <16b39b14-a59f-d184-b4f5-4c184f227ed5@ursulin.net>
Date: Wed, 7 Mar 2018 17:06:55 +0000
From: Tvrtko Ursulin <tursulin@...ulin.net>
To: Bart Van Assche <Bart.VanAssche@....com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Cc: "tvrtko.ursulin@...el.com" <tvrtko.ursulin@...el.com>,
"hare@...e.com" <hare@...e.com>,
"jthumshirn@...e.de" <jthumshirn@...e.de>,
"axboe@...nel.dk" <axboe@...nel.dk>
Subject: Re: [PATCH 3/6] lib/scatterlist: Do not leak pages when high-order
allocation fails
On 07/03/18 16:16, Bart Van Assche wrote:
> On Wed, 2018-03-07 at 12:47 +0000, Tvrtko Ursulin wrote:
>> diff --git a/lib/scatterlist.c b/lib/scatterlist.c
>> index 9884be50a2c0..e13a759c5c49 100644
>> --- a/lib/scatterlist.c
>> +++ b/lib/scatterlist.c
>> @@ -493,7 +493,7 @@ struct scatterlist *sgl_alloc_order(unsigned long length, unsigned int order,
>> {
>> unsigned int chunk_len = PAGE_SIZE << order;
>> struct scatterlist *sgl, *sg;
>> - unsigned int nent;
>> + unsigned int nent, i;
>>
>> nent = round_up(length, chunk_len) >> (PAGE_SHIFT + order);
>>
>> @@ -517,11 +517,12 @@ struct scatterlist *sgl_alloc_order(unsigned long length, unsigned int order,
>>
>> sg_init_table(sgl, nent);
>> sg = sgl;
>> + i = 0;
>> while (length) {
>> struct page *page = alloc_pages(gfp, order);
>>
>> if (!page) {
>> - sgl_free(sgl);
>> + sgl_free_n_order(sgl, i, order);
>> return NULL;
>> }
>>
>> @@ -529,6 +530,7 @@ struct scatterlist *sgl_alloc_order(unsigned long length, unsigned int order,
>> sg_set_page(sg, page, chunk_len, 0);
>> length -= chunk_len;
>> sg = sg_next(sg);
>> + i++;
>> }
>
> Since the entire sg-list is zero-initialized before this loop starts, since
> the sg-list is not chained onto another sg-list before this function returns
> and since sgl_free_n_order() checks whether or not each page pointer is NULL
> before freeing it I think we don't need the new loop variable 'i' and that
> we can call sgl_free_order() instead of sgl_free_n_order().
Yes true, I've only realized that in a later patch. Can rebase to move
that change earlier in.
Regards,
Tvrtko
Powered by blists - more mailing lists