Re: [PATCH v2] zram: fix incorrectly stat with failed_reads

[Sergey Senozhatsky <sergey.senozhatsky@gmail.com>]

Hi,

On (08/14/14 08:02), Minchan Kim wrote:
> On Wed, Aug 13, 2014 at 11:43:04PM +0900, Sergey Senozhatsky wrote:
> > Hello,
> > 
> > On (08/13/14 20:32), Sergey Senozhatsky wrote:
> > > On (08/13/14 10:31), Jerome Marchand wrote:
> > > > Date: Wed, 13 Aug 2014 10:31:45 +0200
> > > > From: Jerome Marchand <jmarchan@redhat.com>
> > > > To: Chao Yu <chao2.yu@samsung.com>, minchan@kernel.org
> > > > CC: ngupta@vflare.org, linux-kernel@vger.kernel.org, 'Sergey Senozhatsky'
> > > >  <sergey.senozhatsky@gmail.com>, 'Andrew Morton'
> > > >  <akpm@linux-foundation.org>
> > > > Subject: Re: [PATCH v2] zram: fix incorrectly stat with failed_reads
> > > > User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101
> > > >  Thunderbird/24.6.0
> > > > 
> > > > On 08/13/2014 04:01 AM, Chao Yu wrote:
> > > > > Since we allocate a temporary buffer in zram_bvec_read to handle partial page
> > > > > operations in this commit 924bd88d703e53d30f393fac6117f8f1bc79aab6 (Staging:
> > > > > zram: allow partial page operations), our ->failed_reads value may be incorrect
> > > > > as we do not increase its value when failed to allocate the temporary buffer.
> > > > > 
> > > > > Let's fix this issue and correct the annotation of failed_reads.
> > > > > 
> > > > >  v2: clean codes of failed_{reads,writes} stat pointed out by Minchan Kim, and
> > > > > this cleanup also fix incorrectly stat when fail in zram_decompress_page.
> > > > > 
> > > > > Signed-off-by: Chao Yu <chao2.yu@samsung.com>
> > > > 
> > > > Acked-by: Jerome Marchand <jmarchan@redhat.com>
> > > 
> > > Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
> > > 
> > > 	-ss
> > > 
> > > > > ---
> > > > >  drivers/block/zram/zram_drv.c | 10 +++++++---
> > > > >  drivers/block/zram/zram_drv.h |  2 +-
> > > > >  2 files changed, 8 insertions(+), 4 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
> > > > > index dfa4024..d00831c 100644
> > > > > --- a/drivers/block/zram/zram_drv.c
> > > > > +++ b/drivers/block/zram/zram_drv.c
> > > > > @@ -378,7 +378,6 @@ static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
> > > > >  	/* Should NEVER happen. Return bio error if it does. */
> > > > >  	if (unlikely(ret)) {
> > > > >  		pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
> > > > > -		atomic64_inc(&zram->stats.failed_reads);
> > > > >  		return ret;
> > > > >  	}
> > > > >  
> > > > > @@ -547,8 +546,6 @@ out:
> > > > >  		zcomp_strm_release(zram->comp, zstrm);
> > > > >  	if (is_partial_io(bvec))
> > > > >  		kfree(uncmem);
> > > > > -	if (ret)
> > > > > -		atomic64_inc(&zram->stats.failed_writes);
> > > > >  	return ret;
> > > > >  }
> > > > >  
> > > > > @@ -566,6 +563,13 @@ static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
> > > > >  		ret = zram_bvec_write(zram, bvec, index, offset);
> > > > >  	}
> > > > >  
> > > > > +	if (unlikely(ret)) {
> > > > > +		if (rw == READ)
> > > > > +			atomic64_inc(&zram->stats.failed_reads);
> > > > > +		else
> > > > > +			atomic64_inc(&zram->stats.failed_writes);
> > > > > +	}
> > > > > +
> > > > >  	return ret;
> > > > >  }
> > > > >  
> > > > > diff --git a/drivers/block/zram/zram_drv.h b/drivers/block/zram/zram_drv.h
> > > > > index 5b0afde..e0f725c 100644
> > > > > --- a/drivers/block/zram/zram_drv.h
> > > > > +++ b/drivers/block/zram/zram_drv.h
> > > > > @@ -84,7 +84,7 @@ struct zram_stats {
> > > > >  	atomic64_t compr_data_size;	/* compressed size of pages stored */
> > > > >  	atomic64_t num_reads;	/* failed + successful */
> > > > >  	atomic64_t num_writes;	/* --do-- */
> > > > > -	atomic64_t failed_reads;	/* should NEVER! happen */
> > > > > +	atomic64_t failed_reads;	/* can happen when memory is too low */
> > 
> > 
> > this grabbed my attention.
> > 
> > 	"failed_writes; /* can happen when memory is too low */"
> > 
> > theoretically, we can have a misbehaving compression algorithm. so the
> > question is -- should we fail write() if compression has failed for any
> > other reason, rather than -ENOMEM? alternatively, we can store uncompressed
> > user mem, just the same way we do for `badly' compressed buffers, set a
> > FAILED_COMPRESSION flag (so we will not leak kernel memory on read() if we
> > failed to compress is_partial_io() bv_page), and on read() copy it out as is,
> > just as we do for `badly' compessed pages.
> > 
> > so, what do you think? /* I can take a look on it */
> 
> How often does it happen?
> I don't think it's valuable to maintain if it's too rare.
> Rather, I'd like to remove the comment on failed_writes. :)
> 

I don't think I've ever observed any errors. and, apparently, lzo
is always `return LZO_E_OK'. so let's just drop it.

	-ss