Skip to contentSkip to conversation

problems in kernel 2.4.20 stat() seems to return always size zero for any fifo (named pipe)

DISCUSSION jds 2 participants 3 messages Mar 6, 2003 – Mar 7, 2003

lore
Share & tools
lore

jds

author
#0 lore raw

Hi:

I have a problem using stat() with kernel 2.4.20 The size of ANY fifo is always 0.
There is no backward compatibility.
Probably it is a kernel bug in this particular version. The problem shows up when using the ls, stat commands.

How reproducible:
Always

Steps to Reproduce:
1. mkfifo xx
2. cat > xx &
3. cat < xx
4. ^Z
5. echo "abcd" > xx

Actual Results: prw-rw-r-- 1 kostadin kostadin 0 Feb 21 13:54 xx

Expected Results: prw-rw-r-- 1 kostadin kostadin 5 Feb 21 13:54 xx

In Kernel 2.4.18-X the RedHat work perfect, e inclusive in 2.4.20-X the Redhat
beta 8.0.94

Helpme please.

#1 lore raw
Quoted 11 lines · jds

Hi:

I have a problem using stat() with kernel 2.4.20 The size of ANY fifo is always 0.
There is no backward compatibility.
Probably it is a kernel bug in this particular version. The problem shows up when using the ls, stat commands.

How reproducible:
Always

[SNIPPED...]

Have you ever tried this before? I'm not aware that stat or lstat is supposed to return the byte-count of a pipe of FIFO. It certainly doesn't on my Sun SunOS 5.5.1 or Linux 2.4.18. In fact, the Sun prevents ^Z when reading the FIFO interactively so your script won't test it.

The second echo "abcd" >xx blocks so the pipe never accepts anything anyway, there are never any "bytes- in-the-pipe" that should be visible anyway. You can't write to a pipe until a reader is reading, and if the reader is suspended, there is no requirement for writing to be buffered, i.e., a write to be accepted for such a pipe.

Pipes do not work like you propose and, if they ever did, that was the bug.

Quoted 6 lines

Expected Results: prw-rw-r-- 1 kostadin kostadin 5 Feb 21 13:54 xx

In Kernel 2.4.18-X the RedHat work perfect,
e inclusive in 2.4.20-X the Redhat
beta 8.0.94

Linux 2.4.18 does not accept writes to a pipe when a reader is not actively reading. As such, a byte-count cannot exist.

If you have a program that expects a byte-count from a FIFO, the program is broken. FIFOs must have readers before any writes are accepted.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.18 on an i686 machine (797.90 BogoMips). Why is the government concerned about the lunatic fringe? Think about it.

author
#2 lore raw

In fact, do you have any idea of what's wrong?
I tryed to fix this in diferent ways but it doesn't seem to work.

Do you have any sugestion?

---------- Original Message ----------
From: "Richard B. Johnson" <[email protected]>
To: jds <[email protected]>
Sent: Thu, 6 Mar 2003 14:21:49 -0500 (EST)
Subject: Re: problems in kernel 2.4.20 stat() seems to return always size zero for any fifo (named pipe)

Quoted 46 lines · jds
Quoted 4 lines

Hi:

Quoted 8 lines

I have a problem using stat() with kernel 2.4.20 The size of ANY fifo is always 0.
There is no backward compatibility.
Probably it is a kernel bug in this particular version. The problem shows up when using the ls, stat commands.

How reproducible:
Always

[SNIPPED...]

Have you ever tried this before? I'm not aware that stat or lstat is supposed to return the byte-count of a pipe of FIFO. It certainly doesn't on my Sun SunOS 5.5.1 or Linux 2.4.18. In fact, the Sun prevents ^Z when reading the FIFO interactively so your script won't test it.

The second echo "abcd" >xx blocks so the pipe never accepts anything anyway, there are never any "bytes- in-the-pipe" that should be visible anyway. You can't write to a pipe until a reader is reading, and if the reader is suspended, there is no requirement for writing to be buffered, i.e., a write to be accepted for such a pipe.

Pipes do not work like you propose and, if they ever did, that was the bug.

Quoted 6 lines

Expected Results: prw-rw-r-- 1 kostadin kostadin 5 Feb 21 13:54 xx

In Kernel 2.4.18-X the RedHat work perfect,
e inclusive in 2.4.20-X the Redhat
beta 8.0.94

Linux 2.4.18 does not accept writes to a pipe when a reader is not actively reading. As such, a byte-count cannot exist.

If you have a program that expects a byte-count from a FIFO, the program is broken. FIFOs must have readers before any writes are accepted.

Quoted 5 lines

Cheers,
Dick Johnson
Penguin : Linux version 2.4.18 on an i686 machine (797.90 BogoMips). Why is the government concerned about the lunatic fringe? Think about it.

------- End of Original Message -------