Skip to contentSkip to conversation

nbd/server: fix nbd_negotiate_handle_info

PATCH In review Vladimir Sementsov-Ogievskiy 2 participants 6 messages Nov 1, 2017 – Nov 3, 2017

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 nbd/server: fix nbd_negotiate_handle_info

namelen should be here, lenght is unrelated.

---
 nbd/server.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
nbd/server.c +1 -1
index 70b40ed27e..7fcec0af7e 100644
@@ -433,7 +433,7 @@ static int nbd_negotiate_handle_info(NBDClient *client, uint32_t length,
 
     /* Don't bother sending NBD_INFO_NAME unless client requested it */
     if (sendname) {
-        rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, length, name,
+        rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, namelen, name,
                                      errp);
         if (rc < 0) {
             return rc;
Signature · 1 line
2.11.1
#1 lore raw

PATCH 2/2 nbd/server: add assert to nbd_negotiate_handle_info

Add an assert here to make last length assignment meaningful and following return without tail dropping obvious.

---
 nbd/server.c | 1 +
 1 file changed, 1 insertion(+)
nbd/server.c +1 -0
index 7fcec0af7e..bcf0cdb47c 100644
@@ -423,6 +423,7 @@ static int nbd_negotiate_handle_info(NBDClient *client, uint32_t length,
             break;
         }
     }
+    assert(length == 0);
 
     exp = nbd_export_find(name);
     if (!exp) {
Signature · 1 line
2.11.1
Reviewed-by on v1 1/2
#4 lore raw
In reply to Vladimir Sementsov-Ogievskiy

On 11/01/2017 10:42 AM, Vladimir Sementsov-Ogievskiy wrote:

It's best to send a 0/2 cover letter for a series, even when both patches are small, as that helps automation tools.

Quoted 1 line

namelen should be here, lenght is unrelated.

s/lenght/length/

Broken in introduction in commit f37708f6; hence adding qemu-stable in cc.

Quoted 5 lines
---
 nbd/server.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

nbd/server.c static int nbd_negotiate_handle_info(NBDClient *client, uint32_t length, · line 436

Quoted 11 lines
nbd/server.c +1 -1
index 70b40ed27e..7fcec0af7e 100644
@@ -433,7 +433,7 @@ static int nbd_negotiate_handle_info(NBDClient *client, uint32_t length,
 
     /* Don't bother sending NBD_INFO_NAME unless client requested it */
     if (sendname) {
-        rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, length, name,
+        rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, namelen, name,

Interestingly enough, length == 0 at this point, so we would always report that the export name is '' (aka the default export), without actually being a protocol violation. Doesn't hurt qemu as a client, since we don't ask for NBD_INFO_NAME, but may break other NBD client implementations, if they then use NBD_OPT_GO on the '' name expecting it to resolve to the same non-empty name they just queried on NBD_OPT_INFO.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
Reviewed-by on v1 2/2
#2 lore raw

Re: [Qemu-devel] [PATCH 2/2] nbd/server: add assert to nbd_negotiate_handle_info

In reply to Vladimir Sementsov-Ogievskiy
Quoted 2 lines · Vladimir Sementsov-Ogievskiy

Add an assert here to make last length assignment meaningful and following return without tail dropping obvious.

Not quite sure I followed your explanation, which means it's difficult for me to propose an alternative wording. Maybe:

Add an assert here to make it obvious that the prior loop consumed the rest of the input, and that all further code in the function is focused on output.

On the other hand, if you are okay with it, I wouldn't mind squashing the first and second patches into one (as the first patch is then easier to read when it is obvious that we used the wrong length variable). But until I get your feedback (on either squashing the two patches or tweaking the wording), I'm just placing your patches as-is on my NBD queue for inclusion prior to rc0.

Quoted 5 lines
---
 nbd/server.c | 1 +
 1 file changed, 1 insertion(+)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#5 lore raw
In reply to Eric Blake
Quoted 4 lines · Eric Blake

On 11/01/2017 10:42 AM, Vladimir Sementsov-Ogievskiy wrote:

It's best to send a 0/2 cover letter for a series, even when both patches are small, as that helps automation tools.

ok.

Quoted 3 lines
Quoted 2 lines

namelen should be here, lenght is unrelated.

s/lenght/length/

my eternal problem...

Quoted 26 lines

Broken in introduction in commit f37708f6; hence adding qemu-stable in cc.

Quoted 4 lines
---
  nbd/server.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

nbd/server.c static int nbd_negotiate_handle_info(NBDClient *client, uint32_t length, · line 436

Quoted 10 lines
nbd/server.c +1 -1
index 70b40ed27e..7fcec0af7e 100644
@@ -433,7 +433,7 @@ static int nbd_negotiate_handle_info(NBDClient *client, uint32_t length,
 
      /* Don't bother sending NBD_INFO_NAME unless client requested it */
      if (sendname) {
-        rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, length, name,
+        rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, namelen, name,

Interestingly enough, length == 0 at this point, so we would always report that the export name is '' (aka the default export), without actually being a protocol violation. Doesn't hurt qemu as a client, since we don't ask for NBD_INFO_NAME, but may break other NBD client implementations, if they then use NBD_OPT_GO on the '' name expecting it to resolve to the same non-empty name they just queried on NBD_OPT_INFO.

Signature · 2 lines
Best regards,
Vladimir
#3 lore raw

Re: [Qemu-devel] [PATCH 2/2] nbd/server: add assert to nbd_negotiate_handle_info

In reply to Eric Blake
Quoted 10 lines · Eric Blake
Quoted 2 lines · Vladimir Sementsov-Ogievskiy

Add an assert here to make last length assignment meaningful and following return without tail dropping obvious.

Not quite sure I followed your explanation, which means it's difficult for me to propose an alternative wording. Maybe:

Add an assert here to make it obvious that the prior loop consumed the rest of the input, and that all further code in the function is focused on output.

Quoted 6 lines

On the other hand, if you are okay with it, I wouldn't mind squashing the first and second patches into one (as the first patch is then easier to read when it is obvious that we used the wrong length variable). But until I get your feedback (on either squashing the two patches or tweaking the wording), I'm just placing your patches as-is on my NBD queue for inclusion prior to rc0.

in commit message I mean two things:
1. assignment to length in the previous loop is useless without an assert
2. firstly I thought that the following return statement is a bug as it doesn't drop payload tail. And only then I noticed previous "if (requests != length / sizeof(request))". With an assert the correctness of the code is more evident.

However you may reword or squash it as you want, all is good.

Quoted 7 lines
Quoted 5 lines
---
  nbd/server.c | 1 +
  1 file changed, 1 insertion(+)
Signature · 2 lines
Best regards,
Vladimir