[Python-bugs-list] [ python-Bugs-585792 ] Invalid mmap crashes Python interpreter

[noreply@sourceforge.net]

Bugs item #585792, was opened at 2002-07-24 01:52
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=585792&group_id=5470

Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: paul rubin (phr)
Assigned to: Neal Norwitz (nnorwitz)
Summary: Invalid mmap crashes Python interpreter

Initial Comment:
This is under Red Hat 7.2, Linux kernel 2.4.7.   "x" is
an empty file, i.e. zero bytes long.

import mmap
f = open(x,"r+")m = mmap.mmap(f.fileno(), 10)
print m[1]

Crashes Python with a bus error.  Python should catch
the bus error signal
and raise an appropriate exception that the script can
catch.


----------------------------------------------------------------------

Comment By: Nobody/Anonymous (nobody)
Date: 2002-08-12 00:45

Message:
Logged In: NO 

Please don't even think of adding a stat call to every
access.  The whole idea of mmap is to speed up access by
eliminating system calls when you touch the file contents. 
Checking the file size at map creation is reasonable.  Some
note should also be added to the mmap doc about possible
crashes if the file size changes.

>From a Python script writer's viewpoint, the most convenient
fix is for Python to catch the bus error signal and inspect
the stack to figure out that the trap occurred in the mmap
routine, then generate and return an appropriate Python
exception.  However, doing that is pretty
platform-dependent, and attempting it in all the Python
ports that support mmap doesn't sound practical.  

----------------------------------------------------------------------

Comment By: Martin v. Löwis (loewis)
Date: 2002-08-12 00:15

Message:
Logged In: YES 
user_id=21627

Ah, an old-style diff. I assume that goes into mmap_item?

I'd rather avoid the stat call on every access, and check
the size on map creation.

The Solaris man page says this about changing files:

     If the size of the mapped file changes  after  the 
call  to
     mmap()  as  a  result  of some other operation on the
mapped
     file, the effect of references to  portions  of  the 
mapped
     region  that  correspond to added or removed portions
of the
     file is unspecified.

So it appears that this is a case of "don"t do this, then",
and that it atleast won't crash - but we might want to
perform a few experiments.



----------------------------------------------------------------------

Comment By: Neal Norwitz (nnorwitz)
Date: 2002-08-11 08:05

Message:
Logged In: YES 
user_id=33168

Martin, could you please review the patch?

----------------------------------------------------------------------

Comment By: Neal Norwitz (nnorwitz)
Date: 2002-07-28 08:00

Message:
Logged In: YES 
user_id=33168

Attached is a patch which fixes this problem.  I'm not sure
this patch it correct nor appropriate.  But it stops the
crash.  I'm hoping some mmap expert can shed some light on this.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=585792&group_id=5470