[Linux-diag-devel] Re: libsysfs & klibc

[H. P. A. <hp...@zy...>]

Christophe Varoqui wrote:
> 
> I finally got my nose into the details of this, which I should have done earlier.
> This is not dirent.h but mntent.h that libsysfs uses and is not present in klibc, my bad.
> 
> If mntent.h is not to be added to klibc, then the following patch ensures klibc compatibility at the expense of a little brain in sysfs_get_fs_mnt_path(). This approach was initialy adopted in udev.
> 
> Please let me know if the following patch is acceptable. I'd rather don't mess with the libsysfs building system myself, but I verified the lib builds fine with klcc with patch applied.

<mntent.h> I really don't think has any business being in klibc.

> 
> diff -ur sysfsutils-1.2.0/lib/sysfs_dir.c sysfsutils-1.2.0.klibcbuild/lib/sysfs_dir.c
> --- sysfsutils-1.2.0/lib/sysfs_dir.c    2004-09-13 15:45:15.000000000 +0200
> +++ sysfsutils-1.2.0.klibcbuild/lib/sysfs_dir.c 2005-04-04 10:57:31.000000000 +0200
> @@ -273,7 +273,11 @@
>                 errno = EACCES;
>                 return -1;
>         }
> +#ifdef __KLIBC__
> +       pgsize = PAGE_SIZE;
> +#else
>         pgsize = sysconf(_SC_PAGESIZE);
> +#endif
>         fbuf = (char *)calloc(1, pgsize+1);
>         if (fbuf == NULL) {
>                 dprintf("calloc failed\n");

This should be:

	pgsize = getpagesize();

... on both klibc and glibc.

> diff -ur sysfsutils-1.2.0/lib/sysfs_utils.c sysfsutils-1.2.0.klibcbuild/lib/sysfs_utils.c
> --- sysfsutils-1.2.0/lib/sysfs_utils.c  2004-09-13 15:45:15.000000000 +0200
> +++ sysfsutils-1.2.0.klibcbuild/lib/sysfs_utils.c       2005-04-04 10:40:52.000000000 +0200
> @@ -63,6 +63,10 @@
>  static int sysfs_get_fs_mnt_path(const char *fs_type,
>                                 char *mnt_path, size_t len)
>  {
> +#ifdef __KLIBC__
> +       safestrcpymax(mnt_path, "/sys", len);
> +       return 0;
> +#else
>         FILE *mnt;
>         struct mntent *mntent;
>         int ret = 0;
> @@ -99,6 +103,7 @@
>                 ret = -1;
> 
>         return ret;
> +#endif

Can't we just define that sysfs should be mounted on /sys, and be done 
with it?  It sounds like the kind of things that one shouldn't bother 
parameterizing; in fact, relying on /proc/mounts or /etc/mtab is 
actually likely to be *LESS* sane than just assuming /sys.

In other words, this is braindamage, fix it where it hurts.

	-hpa