BUG in fs/proc/generic.c:proc_file_read

BUG in fs/proc/generic.c:proc_file_read

[Nagendra Singh Tomar]

In short:
The hack used to be able to read proc files larger than 4k, breaks if the 
caller does lseek() after read()

Detailed:
I am providing a proc read interface to one of my proc files by using the 
given hack symantics in which every call to read_proc() writes the data 
starting at the begining of the page but sets "*start" artificially to 
indicate how many fields have been read. proc_file_read then correctly 
adjusts *ppos to signify the artificial position inside the proc file 
where the read pointer points. It is *artificial* beacuse it is not the 
byte offset but some other offset which only read_proc understands. Next 
time around when read_proc gets the *ppos to start reading from it knows 
how to calculate the exxat byte offset from the *artificial* *ppos 
provided.
This works fine with cat which simply calls read(). The "more" command 
though has the following call pattern

read(fd,buf,4096) = X
lseek(fd,X,SEEK_SET);

-- lseek modified file->f_pos to the byte offset value, which disturbed 
read_proc ---

read(fd,buf,4096) = 0;


the effect is that more never gives me data more than 4096 bytes worth.

My Question is. Is it a known problem ? 


Thanx 
tomar

Re: BUG in fs/proc/generic.c:proc_file_read

[Zwane Mwaikambo]

On Thu, 7 Aug 2003, Nagendra Singh Tomar wrote:

> In short:
> The hack used to be able to read proc files larger than 4k, breaks if the 
> caller does lseek() after read()
> 
> Detailed:
> I am providing a proc read interface to one of my proc files by using the 
> given hack symantics in which every call to read_proc() writes the data 
> starting at the begining of the page but sets "*start" artificially to 
> indicate how many fields have been read. proc_file_read then correctly 
> adjusts *ppos to signify the artificial position inside the proc file 
> where the read pointer points. It is *artificial* beacuse it is not the 
> byte offset but some other offset which only read_proc understands. Next 
> time around when read_proc gets the *ppos to start reading from it knows 
> how to calculate the exxat byte offset from the *artificial* *ppos 
> provided.
> This works fine with cat which simply calls read(). The "more" command 
> though has the following call pattern
> 
> read(fd,buf,4096) = X
> lseek(fd,X,SEEK_SET);
> 
> -- lseek modified file->f_pos to the byte offset value, which disturbed 
> read_proc ---
> 
> read(fd,buf,4096) = 0;
> 
> 
> the effect is that more never gives me data more than 4096 bytes worth.
> 
> My Question is. Is it a known problem ? 

Yep known issue, i need to get around to look and test the patch.

http://bugzilla.kernel.org/show_bug.cgi?id=952

-- 
function.linuxpower.ca

Re: BUG in fs/proc/generic.c:proc_file_read

[Zwane Mwaikambo]

On Fri, 8 Aug 2003, Zwane Mwaikambo wrote:

> On Thu, 7 Aug 2003, Nagendra Singh Tomar wrote:
> 
> > In short:
> > The hack used to be able to read proc files larger than 4k, breaks if the 
> > caller does lseek() after read()
> > 
> > My Question is. Is it a known problem ? 
> 
> Yep known issue, i need to get around to look and test the patch.
> 
> http://bugzilla.kernel.org/show_bug.cgi?id=952

Sorry ignore that, i should get some sleep.

-- 
function.linuxpower.ca

Re: BUG in fs/proc/generic.c:proc_file_read

[Rusty Russell]

In message <Pine.LNX.4.44.0308072346020.3811-100000@localhost.localdomain> you write:
> In short:
> The hack used to be able to read proc files larger than 4k, breaks if the 
> caller does lseek() after read()

Hmm, my more(1) does this too.  What a PITA.  less(1) does not.

I've never noticed it before, though.  I certainly didn't notice it
when I implemented the hack.

Of course, converting to seq_file is probably the nicest solution
these days.

Rusty.
--
  Anyone who quotes me in their sig is an idiot. -- Rusty Russell.