archive mirror
 help / color / mirror / Atom feed
From: Coly Li <>
To: Christoph Hellwig <>
	Alexander Ullrich <>,
	Diego Ercolani <>,
	Jan Szubiak <>,
	Marco Rebhan <>,
	Matthias Ferdinand <>,
	Victor Westerhuis <>,
	Vojtech Pavlik <>,
	Rolf Fokkens <>,
	Thorsten Knabe <>,,
	Kent Overstreet <>,
	Nix <>, Takashi Iwai <>
Subject: Re: [PATCH v5 2/2] bcache: avoid oversized read request in cache missing code path
Date: Mon, 7 Jun 2021 19:55:22 +0800	[thread overview]
Message-ID: <> (raw)
In-Reply-To: <>

On 6/7/21 7:06 PM, Christoph Hellwig wrote:
> On Mon, Jun 07, 2021 at 06:35:39PM +0800, Coly Li wrote:
>> +	/* Limitation for valid replace key size and cache_bio bvecs number */
>> +	size_limit = min_t(unsigned int, bio_max_segs(UINT_MAX) * PAGE_SECTORS,
>> +			   (1 << KEY_SIZE_BITS) - 1);
> bio_max_segs kaps the argument to BIO_MAX_VECS, so you might as well

It was suggested to not directly access BIO_MAX_VECS by you, maybe I
misunderstood you.

> directly write BIO_MAX_VECS.  Can you explain the PAGE_SECTORS here a bit
> more? Does this code path use discontiguous per-sector allocations?
> Preferably in a comment.

It is just because bch_bio_map() assume the maximum bio size is 1MB. It
was not true since the multiple pages bvecs
was merged in mainline kernel.
The PAGE_SECTORS part is legacy for 1MB maximum size bio (256*4KB), it
should be fixed/improved later to
use multiple pages for bio size > 1MB and replace bch_bio_map().

>> +	s->insert_bio_sectors = min3(size_limit, sectors, bio_sectors(bio));
> Also I don't really understand the units involved here.
> s->insert_bio_sectors, sectors, and bio_sectors is in unit of 512 byte
> sectors.  

Yes, they are in sectors, this is the maximum permitted bio size for 1MB
bio size. Now we can have bio > 1MB, and you modify
bio_alloc_bioset() parameter 'nr_iovecs from unsigned int to unsigned
short, so bio-size/page-size can be > 256 and overflow,
e.g. 258 overflows to 2, then the BUG in biovec_slab() is triggered.

I feel this is a long time existing issue in bcache code, and should be
fixed from bcache side, and your change helps to trigger
the problem explicitly.

>> -	miss = bio_next_split(bio, sectors, GFP_NOIO, &s->d->bio_split);
>> +	miss = bio_next_split(bio, s->insert_bio_sectors, GFP_NOIO, &s->d->bio_split);
> Overly long line.

Not any more. Now the line limit is 100 characters. Though I still
prefer 80 characters, place 86 characters in single line
makes the change more obvious.


Coly Li

  reply	other threads:[~2021-06-07 11:55 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-06-07 10:35 [PATCH 0/2] bcache fixes for Linux v5.13-rc6 Coly Li
2021-06-07 10:35 ` [PATCH v5 1/2] bcache: remove bcache device self-defined readahead Coly Li
2021-06-07 11:00   ` Christoph Hellwig
2021-06-07 10:35 ` [PATCH v5 2/2] bcache: avoid oversized read request in cache missing code path Coly Li
2021-06-07 11:06   ` Christoph Hellwig
2021-06-07 11:55     ` Coly Li [this message]
2021-06-07 12:08       ` Christoph Hellwig
2021-06-07 12:31         ` Coly Li
  -- strict thread matches above, loose matches on Subject: below --
2021-05-31 15:31 [PATCH v5 1/2] bcache: remove bcache device self-defined readahead Coly Li
2021-05-31 15:31 ` [PATCH v5 2/2] bcache: avoid oversized read request in cache missing code path Coly Li

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).