On Wed, Dec 02, 2015 at 08:37:05AM -0500, Ted Unangst wrote: > This is a variant of the diffs sent by Daniel Micay, and then assembled by > Theo Buehler. I've looked it over and made a few tweaks. > > One: validate the junk in malloc hasn't been touched. I've tweaked this to > always be on if junk is on, but to only check the first 32 bytes. (Without the > atexit() handler, since I do not trust installing such by default.) I think > this is a decent compromise between checking and performance and complexity > and whatnot. > > Two: add chunk canaries at the end of allocations. I would like to do > something more interesting here (and by default, of course) but growing the > size of every allocation isn't free. Many userland applications already aim > for power of two sizing, so expanding by 4/8 bytes is far from free. However, > this is an ineresting feature, it's not that intrusive, and maybe we can build > from it. Include it as well. > > Hurray or nay?
FWIW, I would be very happy to see this go in. Your changes make sense to me, although I can't really judge the price of your point two on slower architectures. I believe it's slightly slower than Daniel's version although I don't have hard data to back this. I've been running with Daniel's patches and /etc/malloc.conf -> CJV on my two production laptops without any issues for more than a month now. Running with your diff and 'CJ' since Wednesday.