If kzalloc fails to allocate the bioc pointer, NULL is returned
directly.

Signed-off-by: Li zeming <zeming@nfschina.com>
---
 fs/btrfs/volumes.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 064ab2a79c80..f9cb815fe23d 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -5892,6 +5892,8 @@ static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_
 		 */
 		sizeof(u64) * (total_stripes),
 		GFP_NOFS|__GFP_NOFAIL);
+	if (!bioc)
+		return NULL;
 
 	atomic_set(&bioc->error, 0);
 	refcount_set(&bioc->refs, 1);
-- 
2.18.2

On 2022/10/25 16:28, Li zeming wrote:
> If kzalloc fails to allocate the bioc pointer, NULL is returned
> directly.

s/is returned/should be returned/

>
> Signed-off-by: Li zeming <zeming@nfschina.com>
> ---
>   fs/btrfs/volumes.c | 2 ++
>   1 file changed, 2 insertions(+)
>
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index 064ab2a79c80..f9cb815fe23d 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -5892,6 +5892,8 @@ static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_
>   		 */
>   		sizeof(u64) * (total_stripes),
>   		GFP_NOFS|__GFP_NOFAIL);

I think you can also remove the __GFP_NOFAIL flag.

Especially the only caller is properly handling the error.

With that __GFP_NOFAIL flag there, it should not fail, but we can not
just rely on NOFAIL flag to save our asses.

Otherwise looks good to me.

With above two points fixed, you can add my tag:

Reviewed-by: Qu Wenruo <wqu@suse.com>

Thanks,
Qu

> +	if (!bioc)
> +		return NULL;
>
>   	atomic_set(&bioc->error, 0);
>   	refcount_set(&bioc->refs, 1);

very thankful. I have fixed these two problems and am ready to release v2 patches.

On 25.10.22 г. 11:28 ч., Li zeming wrote:
> If kzalloc fails to allocate the bioc pointer, NULL is returned
> directly.
> 
> Signed-off-by: Li zeming <zeming@nfschina.com>

This patch clearly shows you haven't really understood the code. As is 
evident there is __GFP_NOFAIL flag so as per the guarantees for this 
flag we either loop infinitely trying to allocate a bioc or simply 
allocated it. So this check can never be triggered.

NAK
> ---
>   fs/btrfs/volumes.c | 2 ++
>   1 file changed, 2 insertions(+)
> 
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index 064ab2a79c80..f9cb815fe23d 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -5892,6 +5892,8 @@ static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_
>   		 */
>   		sizeof(u64) * (total_stripes),
>   		GFP_NOFS|__GFP_NOFAIL);
> +	if (!bioc)
> +		return NULL;
>   
>   	atomic_set(&bioc->error, 0);
>   	refcount_set(&bioc->refs, 1);

On 2022/10/25 17:29, Nikolay Borisov wrote:
>
>
> On 25.10.22 г. 11:28 ч., Li zeming wrote:
>> If kzalloc fails to allocate the bioc pointer, NULL is returned
>> directly.
>>
>> Signed-off-by: Li zeming <zeming@nfschina.com>
>
> This patch clearly shows you haven't really understood the code. As is
> evident there is __GFP_NOFAIL flag so as per the guarantees for this
> flag we either loop infinitely trying to allocate a bioc or simply
> allocated it. So this check can never be triggered.

I guess what he missed is just to also remove that NOFAIL flag.

NOFAIL will not 100% guarantee the allocation, and I don't see this
location to be so important, especially when the only caller is already
handing allocation failure.

Thanks,
Qu

>
> NAK
>> ---
>>   fs/btrfs/volumes.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
>> index 064ab2a79c80..f9cb815fe23d 100644
>> --- a/fs/btrfs/volumes.c
>> +++ b/fs/btrfs/volumes.c
>> @@ -5892,6 +5892,8 @@ static struct btrfs_io_context
>> *alloc_btrfs_io_context(struct btrfs_fs_info *fs_
>>            */
>>           sizeof(u64) * (total_stripes),
>>           GFP_NOFS|__GFP_NOFAIL);
>> +    if (!bioc)
>> +        return NULL;
>>       atomic_set(&bioc->error, 0);
>>       refcount_set(&bioc->refs, 1);