On Wed, May 02, 2012 at 01:01:35PM -0700, Ken Sumrall wrote: > I thought this patch enabled O_DIRECT functionality on the Mac: > > http://sourceforge.net/tracker/index.php?func=detail&aid=3140289&group_id=2406&atid=102406 Yeah, I was looking at this. It turns out that F_NOCACHE has the effect of O_DIRECT, but it does not require memory alignment. There are some hints that you get really bad performance if the read/write size is not a multiple of 4k, and/or is not well-aligned --- presumably in that case Mac OS uses a bounce buffer. In practice, Linux works the same way --- we'll fall back to using the page cache if the alignment requirements are not met. However, there have historically been bugs that cause this not to work correctly (and in fact caused data corruption): https://bugzilla.redhat.com/show_bug.cgi?id=471613 http://kerneltrap.org/mailarchive/linux-fsdevel/2008/11/14/4099714 ... and the formal documentation of O_DIRECT in the open(2) man page in Linux makes very little guarantees about what the requirements are for O_DIRECT to work correctly: The O_DIRECT flag may impose alignment restrictions on the length and address of userspace buffers and the file offset of I/Os. In Linux alignment restrictions vary by file system and kernel version and might be absent entirely. So there are two possibilities; one is that we change ext2fs_get_memalign() so it gets aligned memory on a best-efforts basis, but does not guarantee that the memory will be aligned. The other is we keep ext2fs_get_memalign() as an interface which returns an error if the request memory alignment can not be honored, and then add code so that we only try using ext2fs_get_memalign() if we are trying to use O_DIRECT (and make an exception for F_NOCACHE on Mac systems that don't have memalign or posix_memalign). The first is tempting, but it just seems a little dirty to me. On the other hand, we've never published a formal API specification for ext2fs_get_memalign(), and it's relatively unlikely there are users of the API outside of e2fsprogs. - Ted -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
