svn commit: r196558 - head/usr.bin/look
alc at cs.rice.edu
Sat Aug 29 20:56:04 UTC 2009
John Baldwin wrote:
> On Tuesday 25 August 2009 11:30:06 pm Colin Percival wrote:
>> Author: cperciva
>> Date: Wed Aug 26 03:30:06 2009
>> New Revision: 196558
>> URL: http://svn.freebsd.org/changeset/base/196558
>> Don't try to mmap the contents of empty files. This behaviour was harmless
>> prior to r195693, since historical behaviour of mmap(2) was to silently
>> ignore length-zero mmap requests; but mmap now returns EINVAL, which caused
>> look(1) to emit an error message and fail.
> FWIW, it did not silently ignore the request. Instead it rounded the size up
> to a page and mapped a page of data. However, if you then passed that pointer
> with a length of 0 to munmap() munmap() would fail.
I don't believe that your claim is correct; round_page(0) == 0. Thus,
the value of "size" that would be passed to vm_mmap() would be 0, which
would cause it to immediately return "0", indicating success. In this
case, I believe that the virtual address that would be returned by
mmap(2) would always be the "hint address" for where it should start
looking for free space, which would be the end of the heap/data segment,
unless the application specified its own hint.
In short, and the reason why I'm correcting you here, is to make clear
that we needn't worry about earlier versions of FreeBSD that don't
implement this change "leaking" or wasting memory from misuse of mmap(2)
in this way.
More information about the svn-src-all