linux-kernel - Re: [PATCH] block: Fix page refcounts for unaligned buffers in __bio_release

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <2dbe097c-a4b8-4f22-8c39-1bdecbee4581@cybernetics.com>
Date: Thu, 29 Feb 2024 15:40:09 -0500
From: Tony Battersby <tonyb@...ernetics.com>
To: Matthew Wilcox <willy@...radead.org>
Cc: Jens Axboe <axboe@...nel.dk>, Andrew Morton <akpm@...ux-foundation.org>,
 "Kirill A . Shutemov" <kirill.shutemov@...ux.intel.com>,
 Hugh Dickins <hughd@...gle.com>, Hannes Reinecke <hare@...e.de>,
 Keith Busch <kbusch@...nel.org>, linux-mm <linux-mm@...ck.org>,
 linux-block@...r.kernel.org, linux-fsdevel <linux-fsdevel@...r.kernel.org>,
 "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] block: Fix page refcounts for unaligned buffers in
 __bio_release_pages()

On 2/29/24 14:53, Matthew Wilcox wrote:
> On Thu, Feb 29, 2024 at 01:08:09PM -0500, Tony Battersby wrote:
>> Fix an incorrect number of pages being released for buffers that do not
>> start at the beginning of a page.
> Oh, I see what I did.  Wouldn't a simpler fix be to just set "done" to
> offset_in_page(fi.offset)?

Actually it would be:

ssize_t done = -offset_in_page(offset);

But then you have signed vs. unsigned comparison in the while(), or you
could rearrange the loop to avoid the negative, but then it gets clunky.

>
>> @@ -1152,7 +1152,7 @@ void __bio_release_pages(struct bio *bio, bool mark_dirty)
>>  
>>  	bio_for_each_folio_all(fi, bio) {
>>  		struct page *page;
>> -		size_t done = 0;
>> +		size_t nr_pages;
>>  
>>  		if (mark_dirty) {
>>  			folio_lock(fi.folio);
>> @@ -1160,10 +1160,11 @@ void __bio_release_pages(struct bio *bio, bool mark_dirty)
>>  			folio_unlock(fi.folio);
>>  		}
>>  		page = folio_page(fi.folio, fi.offset / PAGE_SIZE);
>> +		nr_pages = (fi.offset + fi.length - 1) / PAGE_SIZE -
>> +			   fi.offset / PAGE_SIZE + 1;
>>  		do {
>>  			bio_release_page(bio, page++);
>> -			done += PAGE_SIZE;
>> -		} while (done < fi.length);
>> +		} while (--nr_pages != 0);
>>  	}
>>  }
>>  EXPORT_SYMBOL_GPL(__bio_release_pages);
> The long-term path here, I think, is to replace this bio_release_page()
> with a bio_release_folio(folio, offset, length) which calls into
> a new unpin_user_folio(folio, nr) which calls gup_put_folio().

I developed the patch with the 6.1 stable series, which just has:

nr_pages = (fi.offset + fi.length - 1) / PAGE_SIZE -
	   fi.offset / PAGE_SIZE + 1;
folio_put_refs(fi.folio, nr_pages);

Which is another reason that I went for the direct nr_pages
calculation.  Would you still prefer the negative offset_in_page() approach?

Tony