[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-block] [PATCH v2 12/22] nbd/client: Improve error handling in

From: Eric Blake
Subject: Re: [Qemu-block] [PATCH v2 12/22] nbd/client: Improve error handling in nbd_negotiate_simple_meta_context()
Date: Mon, 17 Dec 2018 09:26:41 -0600
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.3.1

On 12/15/18 9:19 AM, Richard W.M. Jones wrote:
On Sat, Dec 15, 2018 at 07:53:14AM -0600, Eric Blake wrote:
Always allocate space for the reply returned by the server and
hoist the trace earlier, as it is more interesting to trace the
server's reply (even if it is unexpected) than parroting our
request only on success.  After all, skipping the allocation
for a wrong size was merely a micro-optimization that only
benefitted a broken server, rather than the common case of a
compliant server that meets our expectations.

Then turn the reply handling into a loop (even though we still
never iterate more than once), to make this code easier to use
when later patches do support multiple server replies.  This
changes the error message for a server with two replies (a
corner case we are unlikely to hit in practice) from:

Unexpected reply type 4 (meta context), expected 0 (ack)


Server replied with more than one context

Signed-off-by: Eric Blake <address@hidden>

v2: split patch into easier-to-review pieces [Rich, Vladimir]
  nbd/client.c | 16 ++++++++++++----
  1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/nbd/client.c b/nbd/client.c
index bcccd5f555e..b6a85fc3ef8 100644
--- a/nbd/client.c
+++ b/nbd/client.c
@@ -684,10 +684,11 @@ static int nbd_negotiate_simple_meta_context(QIOChannel 
          return ret;

-    if (reply.type == NBD_REP_META_CONTEXT) {
+    while (reply.type == NBD_REP_META_CONTEXT) {

I'm not sure I understand why this change is safe.

As far as I can see reply.type is only updated in the loop by
nbd_receive_option_reply, and that reads from the server, and so the
server might keep sending NBD_REP_META_CONTEXT packets (instead of the
expected NBD_REP_ACK), so it could now loop forever against a
malicious server?  (This is not taking into account any later patches)

The loop can execute at most twice:


          char *name;

-        if (reply.length != sizeof(info->context_id) + context_len) {
+        if (reply.length <= sizeof(info->context_id) ||
+            reply.length > NBD_MAX_BUFFER_SIZE) {
              error_setg(errp, "Failed to negotiate meta context '%s', server "
                         "answered with unexpected length %" PRIu32, context,
@@ -708,6 +709,15 @@ static int nbd_negotiate_simple_meta_context(QIOChannel 
              return -1;
          name[reply.length] = '\0';
+        trace_nbd_opt_meta_reply(name, info->context_id);
+        if (received) {
+            error_setg(errp, "Server replied with more than one context");
+            g_free(name);
+            nbd_send_opt_abort(ioc);
+            return -1;
+        }

If the server replies with a second context, we break the loop by complaining.

The old code accepted at most one context, by complaining if the server's second reply was not ACK; the new code accepts at most one context, by complaining if the server sent more than one context, so the net effect of killing the connection for a misbehaving server response to SET is unchanged.

However, your point about a misbehaving server providing an infinite stream of responses to NBD_OPT_LIST or NBD_OPT_LIST_META_CONTEXT is an interesting question, and may be worth asking upstream to see if the NBD protocol should be tweaked to document any boundaries at how many listings a server might send before a client should worry about the server being malicious. (Does not affect this patch, but pre-existing when we call nbd_receive_list() for servers that lack NBD_OPT_GO, and does impact the later patches in this series that call NBD_OPT_LIST_META_CONTEXT for 'qemu-nbd --list').

Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

reply via email to

[Prev in Thread] Current Thread [Next in Thread]