[Bug 3979] Use reflink for fast cross filesystem copy and rename

[bugzilla-daemon at mindrot.org]

https://bugzilla.mindrot.org/show_bug.cgi?id=3979

--- Comment #2 from Darren Tucker <dtucker at dtucker.net> ---
Comment on attachment 3972
  --> https://bugzilla.mindrot.org/attachment.cgi?id=3972
Proof of Concept patch

>+#ifdef reflink
>+		if (reflink(oldpath, newpath, 1) == -1) {
>+#else

Is "reflink" actually a macro?  It's not on the Solaris 11 VM I have
handy so won't this be a no-op?

>+static int
>+copy_data_via_reflink(uint32_t id)
>+{
>+	char *oldpath, *newpath;
>+	int r;
>+
>+	if ((r = sshbuf_get_cstring(iqueue, &oldpath, NULL)) != 0 ||
>+	    (r = sshbuf_get_cstring(iqueue, &newpath, NULL)) != 0)
>+		fatal_fr(r, "parse");
>+
>+	debug("request %u: copy-data via reflink from \"%s\"  "
>+	    "to \"%s\"", oldpath, newpath);
>+	r = reflink(oldpath, newpath, 1);
>+	free(oldpath);
>+	free(newpath);
>+
>+	return (r);
>+}
>+
> static void
> process_extended_copy_data(uint32_t id)
> {
>@@ -1997,6 +2020,10 @@
> 	int r, copy_until_eof, status = SSH2_FX_OP_UNSUPPORTED;
> 	size_t ret;
> 
>+	if (copy_data_via_reflink(id) == 0) {
>+		status = SSH2_FX_OK;
>+		goto out;
>+	}

How does this work?  copy_data_via_reflink() takes two pathnames, but
process_extended_copy_data() takes two file handles, two offsets and a
length.

A couple of other general questions:
1) it seems like reflink(2) is Solaris-specific?  it doesn't seem to be
present in the Illumos or FreeBSD versions I have handy.
2) is reflink(2) atomic like like(2) is?

> The sftp-server copy-data can be implemented using reflink() with a fallback to the existing read/write loop.

reflink() only seems to support a subset of the functionality of
extended-copy-data: both offsets and length all being zero so assuming
it could be made to work that'd have to be checked too.

-- 
You are receiving this mail because:
You are watching the assignee of the bug.