rsync - Apr 2008 - Bug#477931: rsync: Segfaults syncing the linux kernel archive.

Hi Wayne,
I just got this bug report about rsync 3.0.2 reproducibly crashing,
together with a backtrace and a patch; very helpful :-)

(Please preserve 477931-forwarded@bugs.debian.org in the CC so that you
response is archived in the Debian BTS, thanks.)

Paul Slootman


On Fri 25 Apr 2008, Kurt Roeckx wrote:
> Subject: Bug#477931: rsync: Segfaults syncing the linux kernel archive.
> From: Kurt Roeckx <kurt@roeckx.be>
> Package: rsync
> Version: 3.0.2-1
> Severity: serious
> Tags: patch
> 
> Hi,
> 
> I can consitently make rsync segfault when trying to sync the linux
> kernel archive using version 3.0.2.  It doesn't segfault using
> 2.6.9-2etch2.  This has worked for weeks but started happening yesterday.
> 
> This is what I got from gdb:
> Program received signal SIGSEGV, Segmentation fault.
> #0  0x0000000000434b84 in hashtable_find (tbl=0x0, key=-1,
>     allocate_if_missing=0) at hashtable.c:59
> 59              int key64 = tbl->key64;
> (gdb) bt
> #0  0x0000000000434b84 in hashtable_find (tbl=0x0, key=-1,
>     allocate_if_missing=0) at hashtable.c:59
> #1  0x0000000000430278 in check_prior (file=0x2ae3c6632e78, gnum=-1,
>     prev_ndx_p=0x7fffe8617564, flist_p=0x7fffe8617568) at hlink.c:269
> #2  0x0000000000430433 in hard_link_check (file=0x2ae3c6632e78, ndx=799511,
>     fname=0x7fffe861ba30
"scm/linux/kernel/git/torvalds/linux-2.6.git/objects/pack/pack-447fe96131da71732243925546a8d62bef99bea2.idx",
statret=0,
>     sxp=0x7fffe861b870, itemizing=1, code=FLOG) at hlink.c:298
> #3  0x00000000004112d6 in recv_generator (
>     fname=0x7fffe861ba30
"scm/linux/kernel/git/torvalds/linux-2.6.git/objects/pack/pack-447fe96131da71732243925546a8d62bef99bea2.idx",
file=0x2ae3c6632e78,
>     ndx=799511, itemizing=1, code=FLOG, f_out=3) at generator.c:1504
> #4  0x00000000004139c2 in generate_files (f_out=3, local_name=0x0)
>     at generator.c:2214
> #5  0x000000000041f651 in do_recv (f_in=3, f_out=3, local_name=0x0)
>     at main.c:839
> #6  0x000000000041fdbe in client_run (f_in=3, f_out=3, pid=-1, argc=1,
>     argv=0x682338) at main.c:1077
> #7  0x000000000043f035 in start_socket_client (
>     host=0x6827a8 "filehub.kernel.org", remote_argc=1,
remote_argv=0x682330,
>     argc=1, argv=0x682338) at clientserver.c:132
> #8  0x00000000004204c0 in start_client (argc=1, argv=0x682338) at
main.c:1243
> #9  0x0000000000420cf5 in main (argc=2, argv=0x682330) at main.c:1511
> (gdb) frame 1
> #1  0x0000000000430278 in check_prior (file=0x2ae3c6632e78, gnum=-1,
>     prev_ndx_p=0x7fffe8617564, flist_p=0x7fffe8617568) at hlink.c:269
> 269             if ((node = hashtable_find(prior_hlinks, gnum, 0)) != NULL)
{
> (gdb) p prior_hlinks
> $1 = (struct hashtable *) 0x0
> (gdb) p inc_recurse
> $2 = 0
> 
> prior_hlinks only seems to be used in case of !inc_recurse, so I
> have the feeling that I shouldn't be comming there in the first place.
> I'm guessing that the hashtable_find() should be inside an if
(inc_recurse)
> as all other hashtable_find()'s are also doing that.
> 
> I've attached a patch that does that.  It seems to be working.
> 
> PS: It's annoying that when a core file gets created that it's
written
> in destination path instead of the directory where I started rsync.
> 
> 
> Kurt
> 
--- hlink.c.old	2008-04-25 19:39:15.000000000 +0000
+++ hlink.c	2008-04-25 19:39:49.000000000 +0000
@@ -266,15 +266,18 @@
 		F_HL_PREV(file) = prev_ndx = F_HL_PREV(fp);
 	}
 
-	if ((node = hashtable_find(prior_hlinks, gnum, 0)) != NULL) {
-		assert(node->data != NULL);
-		if (CVAL(node->data, 0) != 0) {
-			*prev_ndx_p = -1;
-			*flist_p = NULL;
-			return node->data;
+	if (inc_recurse)
+	{
+		if ((node = hashtable_find(prior_hlinks, gnum, 0)) != NULL) {
+			assert(node->data != NULL);
+			if (CVAL(node->data, 0) != 0) {
+				*prev_ndx_p = -1;
+				*flist_p = NULL;
+				return node->data;
+			}
+			/* The prior file must have been skipped. */
+			F_HL_PREV(file) = -1;
 		}
-		/* The prior file must have been skipped. */
-		F_HL_PREV(file) = -1;
 	}
 
 	*prev_ndx_p = -1;

On Sat, Apr 26, 2008 at 10:33:42AM +0200, Paul Slootman
wrote:
> I just got this bug report about rsync 3.0.2 reproducibly crashing,
> together with a backtrace and a patch; very helpful :-)
Yes, quite helpful.  The fix is indeed the right one, though we can
avoid the re-indenting of all that code.  See the git repo for my
version:

    http://git.samba.org/rsync.git/?p=rsync.git

Thanks for the report and patch!

..wayne..