Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH v2 1/2] f2fs: compress: introduce page array slab cache
To:     Jaegeuk Kim <jaegeuk@kernel.org>
CC:     <linux-f2fs-devel@lists.sourceforge.net>,
        <linux-kernel@vger.kernel.org>, <chao@kernel.org>
References: <20200914090514.50102-1-yuchao0@huawei.com>
 <20200929082306.GA1567825@google.com>
 <6e7639db-9120-d406-0a46-ec841845bb28@huawei.com>
 <20200929084739.GB1567825@google.com>
From:   Chao Yu <yuchao0@huawei.com>
Message-ID: <1b9774da-b2a8-2009-7796-9c576af1b4c4@huawei.com>
Date:   Tue, 29 Sep 2020 17:15:51 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:52.0) Gecko/20100101
 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: <20200929084739.GB1567825@google.com>
Content-Type: text/plain; charset="windows-1252"; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2020/9/29 16:47, Jaegeuk Kim wrote:
> On 09/29, Chao Yu wrote:
>> On 2020/9/29 16:23, Jaegeuk Kim wrote:
>>> I found a bug related to the number of page pointer allocation related to
>>> nr_cpages.
>>
>> Jaegeuk,
>>
>> If I didn't miss anything, you mean that nr_cpages could be larger
>> than nr_rpages, right? the problematic case here is lzo/lzo-rle:
>>
>> cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
>>
>> As we can't limited clen as we did for lz4/zstd:
>>
>> cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
> 
> Yes, I've seen some memory corruption in lzo test. Here is another patch to fix
> mem leak.
> 
> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> ---
>   fs/f2fs/compress.c | 67 ++++++++++++++++++++++++++++------------------
>   1 file changed, 41 insertions(+), 26 deletions(-)
> 
> diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
> index f086ac43ca825..ba2d4897744d8 100644
> --- a/fs/f2fs/compress.c
> +++ b/fs/f2fs/compress.c
> @@ -20,22 +20,20 @@
>   static struct kmem_cache *cic_entry_slab;
>   static struct kmem_cache *dic_entry_slab;
>   
> -static void *page_array_alloc(struct inode *inode)
> +static void *page_array_alloc(struct inode *inode, int nr)
>   {
>   	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> -	unsigned int size = sizeof(struct page *) <<
> -				F2FS_I(inode)->i_log_cluster_size;
> +	unsigned int size = sizeof(struct page *) * nr;
>   
>   	if (likely(size == sbi->page_array_slab_size))
>   		return kmem_cache_zalloc(sbi->page_array_slab, GFP_NOFS);
>   	return f2fs_kzalloc(sbi, size, GFP_NOFS);
>   }
>   
> -static void page_array_free(struct inode *inode, void *pages)
> +static void page_array_free(struct inode *inode, void *pages, int nr)
>   {
>   	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> -	unsigned int size = sizeof(struct page *) <<
> -				F2FS_I(inode)->i_log_cluster_size;
> +	unsigned int size = sizeof(struct page *) * nr;
>   
>   	if (!pages)
>   		return;
> @@ -162,13 +160,13 @@ int f2fs_init_compress_ctx(struct compress_ctx *cc)
>   	if (cc->rpages)
>   		return 0;
>   
> -	cc->rpages = page_array_alloc(cc->inode);
> +	cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
>   	return cc->rpages ? 0 : -ENOMEM;
>   }
>   
>   void f2fs_destroy_compress_ctx(struct compress_ctx *cc)
>   {
> -	page_array_free(cc->inode, cc->rpages);
> +	page_array_free(cc->inode, cc->rpages, cc->cluster_size);
>   	cc->rpages = NULL;
>   	cc->nr_rpages = 0;
>   	cc->nr_cpages = 0;
> @@ -602,7 +600,8 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
>   	struct f2fs_inode_info *fi = F2FS_I(cc->inode);
>   	const struct f2fs_compress_ops *cops =
>   				f2fs_cops[fi->i_compress_algorithm];
> -	unsigned int max_len, nr_cpages;
> +	unsigned int max_len, new_nr_cpages;
> +	struct page **new_cpages;
>   	int i, ret;
>   
>   	trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
> @@ -617,7 +616,7 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
>   	max_len = COMPRESS_HEADER_SIZE + cc->clen;
>   	cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
>   
> -	cc->cpages = page_array_alloc(cc->inode);
> +	cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);

Well, cc->nr_cpages will be set to cc->nr_rpages - 1 for zstd/lz4 cases, so
this will make cpages allocation fallback to kmalloc, which can cause more
memory use.

Thanks,

>   	if (!cc->cpages) {
>   		ret = -ENOMEM;
>   		goto destroy_compress_ctx;
> @@ -659,16 +658,28 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
>   	for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
>   		cc->cbuf->reserved[i] = cpu_to_le32(0);
>   
> -	nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
> +	new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
> +
> +	/* Now we're going to cut unnecessary tail pages */
> +	new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
> +	if (new_cpages) {
> +		ret = -ENOMEM;
> +		goto out_vunmap_cbuf;
> +	}
>   
>   	/* zero out any unused part of the last page */
>   	memset(&cc->cbuf->cdata[cc->clen], 0,
> -	       (nr_cpages * PAGE_SIZE) - (cc->clen + COMPRESS_HEADER_SIZE));
> +			(new_nr_cpages * PAGE_SIZE) -
> +			(cc->clen + COMPRESS_HEADER_SIZE));
>   
>   	vm_unmap_ram(cc->cbuf, cc->nr_cpages);
>   	vm_unmap_ram(cc->rbuf, cc->cluster_size);
>   
> -	for (i = nr_cpages; i < cc->nr_cpages; i++) {
> +	for (i = 0; i < cc->nr_cpages; i++) {
> +		if (i < new_nr_cpages) {
> +			new_cpages[i] = cc->cpages[i];
> +			continue;
> +		}
>   		f2fs_compress_free_page(cc->cpages[i]);
>   		cc->cpages[i] = NULL;
>   	}
> @@ -676,7 +687,9 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
>   	if (cops->destroy_compress_ctx)
>   		cops->destroy_compress_ctx(cc);
>   
> -	cc->nr_cpages = nr_cpages;
> +	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
> +	cc->cpages = new_cpages;
> +	cc->nr_cpages = new_nr_cpages;
>   
>   	trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
>   							cc->clen, ret);
> @@ -691,7 +704,7 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
>   		if (cc->cpages[i])
>   			f2fs_compress_free_page(cc->cpages[i]);
>   	}
> -	page_array_free(cc->inode, cc->cpages);
> +	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
>   	cc->cpages = NULL;
>   destroy_compress_ctx:
>   	if (cops->destroy_compress_ctx)
> @@ -730,7 +743,7 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
>   		goto out_free_dic;
>   	}
>   
> -	dic->tpages = page_array_alloc(dic->inode);
> +	dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
>   	if (!dic->tpages) {
>   		ret = -ENOMEM;
>   		goto out_free_dic;
> @@ -1203,7 +1216,7 @@ static int f2fs_write_compressed_pages(struct compress_ctx *cc,
>   	cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
>   	cic->inode = inode;
>   	atomic_set(&cic->pending_pages, cc->nr_cpages);
> -	cic->rpages = page_array_alloc(cc->inode);
> +	cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
>   	if (!cic->rpages)
>   		goto out_put_cic;
>   
> @@ -1297,11 +1310,13 @@ static int f2fs_write_compressed_pages(struct compress_ctx *cc,
>   	spin_unlock(&fi->i_size_lock);
>   
>   	f2fs_put_rpages(cc);
> +	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
> +	cc->cpages = NULL;
>   	f2fs_destroy_compress_ctx(cc);
>   	return 0;
>   
>   out_destroy_crypt:
> -	page_array_free(cc->inode, cic->rpages);
> +	page_array_free(cc->inode, cic->rpages, cc->cluster_size);
>   
>   	for (--i; i >= 0; i--)
>   		fscrypt_finalize_bounce_page(&cc->cpages[i]);
> @@ -1310,6 +1325,8 @@ static int f2fs_write_compressed_pages(struct compress_ctx *cc,
>   			continue;
>   		f2fs_put_page(cc->cpages[i], 1);
>   	}
> +	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
> +	cc->cpages = NULL;
>   out_put_cic:
>   	kmem_cache_free(cic_entry_slab, cic);
>   out_put_dnode:
> @@ -1345,7 +1362,7 @@ void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
>   		end_page_writeback(cic->rpages[i]);
>   	}
>   
> -	page_array_free(cic->inode, cic->rpages);
> +	page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
>   	kmem_cache_free(cic_entry_slab, cic);
>   }
>   
> @@ -1442,8 +1459,6 @@ int f2fs_write_multi_pages(struct compress_ctx *cc,
>   
>   		err = f2fs_write_compressed_pages(cc, submitted,
>   							wbc, io_type);
> -		page_array_free(cc->inode, cc->cpages);
> -		cc->cpages = NULL;
>   		if (!err)
>   			return 0;
>   		f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
> @@ -1468,7 +1483,7 @@ struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
>   	if (!dic)
>   		return ERR_PTR(-ENOMEM);
>   
> -	dic->rpages = page_array_alloc(cc->inode);
> +	dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
>   	if (!dic->rpages) {
>   		kmem_cache_free(dic_entry_slab, dic);
>   		return ERR_PTR(-ENOMEM);
> @@ -1487,7 +1502,7 @@ struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
>   		dic->rpages[i] = cc->rpages[i];
>   	dic->nr_rpages = cc->cluster_size;
>   
> -	dic->cpages = page_array_alloc(dic->inode);
> +	dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
>   	if (!dic->cpages)
>   		goto out_free;
>   
> @@ -1522,7 +1537,7 @@ void f2fs_free_dic(struct decompress_io_ctx *dic)
>   				continue;
>   			f2fs_compress_free_page(dic->tpages[i]);
>   		}
> -		page_array_free(dic->inode, dic->tpages);
> +		page_array_free(dic->inode, dic->tpages, dic->cluster_size);
>   	}
>   
>   	if (dic->cpages) {
> @@ -1531,10 +1546,10 @@ void f2fs_free_dic(struct decompress_io_ctx *dic)
>   				continue;
>   			f2fs_compress_free_page(dic->cpages[i]);
>   		}
> -		page_array_free(dic->inode, dic->cpages);
> +		page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
>   	}
>   
> -	page_array_free(dic->inode, dic->rpages);
> +	page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
>   	kmem_cache_free(dic_entry_slab, dic);
>   }
>   
>