2001-10-04 03:00:34

by John Levon

[permalink] [raw]
Subject: minor mmap bug ?


2.4.10

In do_mmap_pgoff() :

422 addr = get_unmapped_area(file, addr, len, pgoff, flags);
423 if (addr & ~PAGE_MASK)
424 return addr;

in get_unmapped_area() :

621 if (flags & MAP_FIXED) {
622 if (addr > TASK_SIZE - len)
623 return -EINVAL;
624 if (addr & ~PAGE_MASK)
625 return -EINVAL;
626 return addr;
627 }


So it seems that if we hit the first case (addr > TASK_SIZE - len),
we don't return with EINVAL immediately. This code is a little weird anyway.

Is this a bug ?

Also, what's the reason for :

405 if ((len = PAGE_ALIGN(len)) == 0)
406 return addr;


Also, shouldn't calc_vm_flags() mention MAP_LOCKED and the others specifically ?
The current "oh, it happens to be the same value" is distinctly non-obvious to me.


Also, why do several (all ?) of the do_mmap2()s do this :

51 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);

Is it something to do with mm->def_flags ?

thanks
john

--
" It is quite humbling to realize that the storage occupied by the longest line
from a typical Usenet posting is sufficient to provide a state space so vast
that all the computation power in the world can not conquer it."
- Dave Wallace