Blame SOURCES/CVE-2021-4028.patch

3283a7
From 44d6b1169915913fae640dada5ed931dffff6772 Mon Sep 17 00:00:00 2001
3283a7
From: "C. Erastus Toe" <ctoe@redhat.com>
3283a7
Date: Fri, 18 Feb 2022 01:09:25 -0500
3283a7
Subject: [KPATCH CVE-2021-4028] kpatch fixes for CVE-2021-4028
3283a7
3283a7
Kernels:
3283a7
3.10.0-1160.21.1.el7
3283a7
3.10.0-1160.24.1.el7
3283a7
3.10.0-1160.25.1.el7
3283a7
3.10.0-1160.31.1.el7
3283a7
3.10.0-1160.36.2.el7
3283a7
3.10.0-1160.41.1.el7
3283a7
3.10.0-1160.42.2.el7
3283a7
3.10.0-1160.45.1.el7
3283a7
3.10.0-1160.49.1.el7
3283a7
3.10.0-1160.53.1.el7
3283a7
3283a7
Changes since last build:
3283a7
arches: x86_64 ppc64le
3283a7
cma.o: changed function: rdma_listen
3283a7
---------------------------
3283a7
3283a7
Kpatch-MR: https://gitlab.com/redhat/prdsc/rhel/src/kpatch/rhel-7/-/merge_requests/25
3283a7
Approved-by: Joe Lawrence (@joe.lawrence)
3283a7
Modifications: none
3283a7
3283a7
commit 00225da7d36440d257a98004ee2e7f288486245b
3283a7
Author: Kamal Heib <kheib@redhat.com>
3283a7
Date:   Sun Jan 16 11:38:26 2022 +0200
3283a7
3283a7
    RDMA/cma: Do not change route.addr.src_addr.ss_family
3283a7
3283a7
    Bugzilla: http://bugzilla.redhat.com/2032075
3283a7
    CVE: CVE-2021-4028
3283a7
    Conflicts:
3283a7
    The conditional statement to verify RDMA_CM_IDLE state is a
3283a7
    little bit different from the upstream to 7.9.z source. The commit
3283a7
    732d41c545bb (RDMA/cma: Make the locking for automatic state transition
3283a7
    more clear) is missing from 7.9.z. So, the context has been modified.
3283a7
3283a7
    commit bc0bdc5afaa740d782fbf936aaeebd65e5c2921d
3283a7
    Author: Jason Gunthorpe <jgg@nvidia.com>
3283a7
    Date:   Wed Sep 15 17:21:43 2021 -0300
3283a7
3283a7
        RDMA/cma: Do not change route.addr.src_addr.ss_family
3283a7
3283a7
        If the state is not idle then rdma_bind_addr() will immediately fail and
3283a7
        no change to global state should happen.
3283a7
3283a7
        For instance if the state is already RDMA_CM_LISTEN then this will corrupt
3283a7
        the src_addr and would cause the test in cma_cancel_operation():
3283a7
3283a7
                        if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
3283a7
3283a7
        To view a mangled src_addr, eg with a IPv6 loopback address but an IPv4
3283a7
        family, failing the test.
3283a7
3283a7
        This would manifest as this trace from syzkaller:
3283a7
3283a7
          BUG: KASAN: use-after-free in __list_add_valid+0x93/0xa0 lib/list_debug.c:26
3283a7
          Read of size 8 at addr ffff8881546491e0 by task syz-executor.1/32204
3283a7
3283a7
          CPU: 1 PID: 32204 Comm: syz-executor.1 Not tainted 5.12.0-rc8-syzkaller #0
3283a7
          Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
3283a7
          Call Trace:
3283a7
           __dump_stack lib/dump_stack.c:79 [inline]
3283a7
           dump_stack+0x141/0x1d7 lib/dump_stack.c:120
3283a7
           print_address_description.constprop.0.cold+0x5b/0x2f8 mm/kasan/report.c:232
3283a7
           __kasan_report mm/kasan/report.c:399 [inline]
3283a7
           kasan_report.cold+0x7c/0xd8 mm/kasan/report.c:416
3283a7
           __list_add_valid+0x93/0xa0 lib/list_debug.c:26
3283a7
           __list_add include/linux/list.h:67 [inline]
3283a7
           list_add_tail include/linux/list.h:100 [inline]
3283a7
           cma_listen_on_all drivers/infiniband/core/cma.c:2557 [inline]
3283a7
           rdma_listen+0x787/0xe00 drivers/infiniband/core/cma.c:3751
3283a7
           ucma_listen+0x16a/0x210 drivers/infiniband/core/ucma.c:1102
3283a7
           ucma_write+0x259/0x350 drivers/infiniband/core/ucma.c:1732
3283a7
           vfs_write+0x28e/0xa30 fs/read_write.c:603
3283a7
           ksys_write+0x1ee/0x250 fs/read_write.c:658
3283a7
           do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
3283a7
           entry_SYSCALL_64_after_hwframe+0x44/0xae
3283a7
3283a7
        Which is indicating that an rdma_id_private was destroyed without doing
3283a7
        cma_cancel_listens().
3283a7
3283a7
        Instead of trying to re-use the src_addr memory to indirectly create an
3283a7
        any address build one explicitly on the stack and bind to that as any
3283a7
        other normal flow would do.
3283a7
3283a7
        Link: https://lore.kernel.org/r/0-v1-9fbb33f5e201+2a-cma_listen_jgg@nvidia.com
3283a7
        Cc: stable@vger.kernel.org
3283a7
        Fixes: 732d41c545bb ("RDMA/cma: Make the locking for automatic state transition more clear")
3283a7
        Reported-by: syzbot+6bb0528b13611047209c@syzkaller.appspotmail.com
3283a7
        Tested-by: Hao Sun <sunhao.th@gmail.com>
3283a7
        Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
3283a7
        Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3283a7
3283a7
    Signed-off-by: Kamal Heib <kheib@redhat.com>
3283a7
3283a7
Signed-off-by: C. Erastus Toe <ctoe@redhat.com>
3283a7
---
3283a7
 drivers/infiniband/core/cma.c | 9 +++++++--
3283a7
 1 file changed, 7 insertions(+), 2 deletions(-)
3283a7
3283a7
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
3283a7
index ec473099e445..935f3979c8e2 100644
3283a7
--- a/drivers/infiniband/core/cma.c
3283a7
+++ b/drivers/infiniband/core/cma.c
3283a7
@@ -3467,8 +3467,13 @@ int rdma_listen(struct rdma_cm_id *id, int backlog)
3283a7
 
3283a7
 	id_priv = container_of(id, struct rdma_id_private, id);
3283a7
 	if (id_priv->state == RDMA_CM_IDLE) {
3283a7
-		id->route.addr.src_addr.ss_family = AF_INET;
3283a7
-		ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3283a7
+		struct sockaddr_in any_in = {
3283a7
+			.sin_family = AF_INET,
3283a7
+			.sin_addr.s_addr = htonl(INADDR_ANY),
3283a7
+		};
3283a7
+
3283a7
+		/* For a well behaved ULP state will be RDMA_CM_IDLE */
3283a7
+		ret = rdma_bind_addr(id, (struct sockaddr *)&any_in);
3283a7
 		if (ret)
3283a7
 			return ret;
3283a7
 	}
3283a7
-- 
3283a7
2.26.3
3283a7
3283a7