On Mon, 11 Apr 2022, Linus Torvalds wrote:
> On Mon, Apr 11, 2022 at 4:43 AM Mikulas Patocka <mpatocka@xxxxxxxxxx> wrote:
> >
> > If you run a program compiled with OpenWatcom for Linux on a filesystem on
> > NVMe, all "stat" syscalls fail with -EOVERFLOW. The reason is that the
> > NVMe driver allocates a device with the major number 259 and it doesn't
> > pass the "old_valid_dev" test.
>
> OpenWatcom? Really?
Yes. I use OpenWatcom to verify that my programs are clean ANSI C without
any gccisms.
Other than that, it is not much useful - it has it's own libc, it's own
module format, and programs compiled with OpenWatcom cannot be linked with
existing *.a or *.so libraries.
> > This patch removes the tests - it's better to wrap around than to return
> > an error. (note that cp_old_stat also doesn't report an error and wraps
> > the number around)
>
> Hmm. We've used majors over 256 for a long time, but some of them are
> admittedly very rare (SCSI OSD?)
>
> Unfortunate. And in this case 259 aliases to 3, which is the old
> HD/IDE0 major number. That's not great - there would be other numbers
> that didn't have that problem (ie 4-6 are all currently only character
> device majors, I think).
Should we perhaps hash the number, take 16 bits of the hash and hope
than the collision won't happen?
> Anyway, I think that check is just bogus. The cp_new_stat() thing uses
> 'struct stat' and it has
>
> unsigned long st_dev; /* Device. */
> unsigned long st_rdev; /* Device number, if device. */
>
> so there's no reason to limit things to the old 8-bit behavior.
>
> Yes, it does that
>
> #define valid_dev(x) choose_32_64(old_valid_dev(x),true)
> #define encode_dev(x) choose_32_64(old_encode_dev,new_encode_dev)(x)
>
> static __always_inline u16 old_encode_dev(dev_t dev)
> {
> return (MAJOR(dev) << 8) | MINOR(dev);
> }
>
> which currently drops bits, but we should just *fix* that. We can put
> the high bits in the upper bits, not limit it to 16 bits when we have
> more space than that.
Yes - we can return values larger than 16-bit here. But there's a risk
that the userspace code will extract the values using macros like this and
lose the upper bits:
#define major(device) ((int)(((device) >> 8) & 0xFF))
#define minor(device) ((int)((device) & 0xff))
> Even the *really* old 'struct old_stat' doesn't really have a 16-bit
> st_dev/rdev.
>
> Linus
For me, the failure happens in cp_compat_stat (I have a 64-bit kernel). In
struct compat_stat in arch/x86/include/asm/compat.h, st_dev and st_rdev
are compat_dev_t which is 16-bit. But they are followed by 16-bit
paddings, so they could be extended.
If you have a native 32-bit kernel, it uses 'struct stat' defined at the
beginning of arch/x86/include/uapi/asm/stat.h that has 32-bit st_dev and
st_rdev. If you use a 64-bit kernel with 32-bit compat, it uses 'struct
compat_stat' defined in arch/x86/include/asm/compat.h that has 16-bit
st_dev and st_rdev. That's an inconsistency that should be resolved.
What did glibc do? Did it use 16-bit dev_t with following padding or
32-bit dev_t? (the current glibc just uses stat64 and 64-bit dev_t always)
Mikulas
