read() from a removed device

read() from a removed device

[Frank Schäfer]

Hi,

I just noticed that read() from a usb-serial device doesn't fail after 
the device got (physically) removed.
I would expect it to return -1 and set an appropriate errno, but it 
returns 0 (no error, no bytes read) instead.
Is this the correct/expected behavior or a bug ?

Regards,
Frank Schäfer

Re: read() from a removed device

[Florian Weimer]

* Frank Schäfer via Libc-help:

> I just noticed that read() from a usb-serial device doesn't fail after
> the device got (physically) removed.
> I would expect it to return -1 and set an appropriate errno, but it
> returns 0 (no error, no bytes read) instead.
> Is this the correct/expected behavior or a bug ?

You need toask on one of the kernel mailing lists.  glibc just calls
into the kernel and reports back whatever it receives from there.

Thanks,
Florian

Re: read() from a removed device

[Frank Schäfer]

Am 06.07.20 um 19:17 schrieb Florian Weimer:
> * Frank Schäfer via Libc-help:
>
>> I just noticed that read() from a usb-serial device doesn't fail after
>> the device got (physically) removed.
>> I would expect it to return -1 and set an appropriate errno, but it
>> returns 0 (no error, no bytes read) instead.
>> Is this the correct/expected behavior or a bug ?
> You need toask on one of the kernel mailing lists.  glibc just calls
> into the kernel and reports back whatever it receives from there.
Guess what the kernel devs say when I report this issue to them... ;)
Can I read your reply as "this is supposed to work as you expect from 
glibc side" ?
How does the "call into the kernel" look like ?

Thanks,
Frank


> Thanks,
> Florian
>

Re: read() from a removed device

[Florian Weimer]

* Frank Schäfer:

> Am 06.07.20 um 19:17 schrieb Florian Weimer:
>> * Frank Schäfer via Libc-help:
>>
>>> I just noticed that read() from a usb-serial device doesn't fail after
>>> the device got (physically) removed.
>>> I would expect it to return -1 and set an appropriate errno, but it
>>> returns 0 (no error, no bytes read) instead.
>>> Is this the correct/expected behavior or a bug ?
>> You need toask on one of the kernel mailing lists.  glibc just calls
>> into the kernel and reports back whatever it receives from there.

> Guess what the kernel devs say when I report this issue to them... ;)

Not sure what exactly you asked.  Maybe someone confused fread and read.

> How does the "call into the kernel" look like ?

Basically like this:

  xor %eax, %eax
  syscall 

Plus some code to put the error code into errno on failure.  And of
course zero return value is not a failure.  There is not much that can
go wrong there.

Thanks,
Florian