[v2,3/3] scandir: fix wrong assumption about errno [BZ #17804]

  malloc and realloc may set errno to ENOMEM even if they are successful.
The scandir code wrongly assume that they do not change errno, this
causes scandir to fail with ENOMEM even if malloc succeed.

The code already handles that readdir might set errno by calling
__set_errno (0) to clear the error. Move that part at the end of the
loop to also take malloc and realloc into account.

Changelog:
	[BZ #17804]
	* dirent/scandir-tail.c (SCANDIR_TAIL): Move __set_errno (0) at the
	end of the loop. Improve comments.
---
 ChangeLog             |  6 ++++++
 dirent/scandir-tail.c | 13 ++++++++-----
 2 files changed, 14 insertions(+), 5 deletions(-)
  

On Fri, Dec 22, 2017 at 08:31:01PM +0100, Aurelien Jarno wrote:
> malloc and realloc may set errno to ENOMEM even if they are successful.
> The scandir code wrongly assume that they do not change errno, this
> causes scandir to fail with ENOMEM even if malloc succeed.
> 
> The code already handles that readdir might set errno by calling
> __set_errno (0) to clear the error. Move that part at the end of the
> loop to also take malloc and realloc into account.
> 
> Changelog:
> 	[BZ #17804]
> 	* dirent/scandir-tail.c (SCANDIR_TAIL): Move __set_errno (0) at the
> 	end of the loop. Improve comments.
> ---
>  ChangeLog             |  6 ++++++
>  dirent/scandir-tail.c | 13 ++++++++-----
>  2 files changed, 14 insertions(+), 5 deletions(-)
> 
> diff --git a/ChangeLog b/ChangeLog
> index 8368b3006d..713c5ade6d 100644
> --- a/ChangeLog
> +++ b/ChangeLog
> @@ -1,3 +1,9 @@
> +2017-12-22  Aurelien Jarno  <aurelien@aurel32.net>
> +
> +	[BZ #17804]
> +	* dirent/scandir-tail.c (SCANDIR_TAIL): Move __set_errno (0) at the
> +	end of the loop. Improve comments.
> +
>  2017-12-22  Zack Weinberg  <zackw@panix.com>
>  
>  	[BZ #22615]
> diff --git a/dirent/scandir-tail.c b/dirent/scandir-tail.c
> index 068c644c4e..927791fad0 100644
> --- a/dirent/scandir-tail.c
> +++ b/dirent/scandir-tail.c
> @@ -53,16 +53,14 @@ SCANDIR_TAIL (DIR *dp,
>          {
>            int selected = (*select) (d);
>  
> -	  /* The SELECT function might have changed errno.  It was
> -	     zero before and it need to be again to make the later
> -	     tests work.  */
> +	  /* The SELECT function might have set errno to non-zero on
> +	     success.  It was zero before and it need to be again to

s/it need to be/it needs to be/

> +	     make the later tests work.  */
>  	  __set_errno (0);
>  
>            if (!selected)
>              continue;
>          }
> -      else
> -        __set_errno (0);
>  
>        if (__glibc_unlikely (c.cnt == vsize))
>          {
> @@ -81,6 +79,11 @@ SCANDIR_TAIL (DIR *dp,
>        if (vnew == NULL)
>          break;
>        v[c.cnt++] = (DIRENT_TYPE *) memcpy (vnew, d, dsize);
> +
> +      /* Ignore errors from readdir, malloc or realloc.  These functions
> +	 might have set errno to non-zero on success.  It was zero before
> +	 and it need to be again to make the latter tests work.  */

Likewise.

> +      __set_errno (0);
>      }
>  
>    if (__glibc_likely (errno == 0))

Besides these two typos in comments, the patch is fine.