Re: [Qemu-devel] [PATCH][resend] Add -f option to qemu-nbd

[Paolo Bonzini <pbonzini@redhat.com>]

On 01/18/2012 11:56 AM, Michael Tokarev wrote:
> On 18.01.2012 12:48, Chunyan Liu wrote:
>> Stefan, could you help commit it if it's OK? Thanks.
>> Same as in thread:
>> http://lists.gnu.org/archive/html/qemu-devel/2011-12/msg01083.html
>> but rebase it to latest code.
>
> There's a (trivial) fix sent against qemu-nbd which will
> make this patch to not apply again (the change of "fd"
> variable in main() into devfd and sockfd).  I dunno if
> it is applied to any branch or not.  Most active person
> in this area appears to be Paolo Bonzini (Cc'd).

I've not yet prepared a public NBD tree, but I will soon and I will 
apply that patch (strictly speaking I had a comment, but I can fix that 
up myself).

>> @@ -55,12 +55,13 @@ static void usage(const char *name)
>>   "  -o, --offset=OFFSET  offset into the image\n"
>>   "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
>>   "  -k, --socket=PATH    path to the unix socket\n"
>> -"                       (default '"SOCKET_PATH"')\n"
>> +"                       (default /var/lock/qemu-nbd-PID)\n"
>
> This is a semantic change.  Before, the socket was
> named after the nbd device name, which, lacking the
> -f option, was deterministic (given with -c option).
> Now, since pid is random for the user, we don't have
> a deterministic socket name anymore.  I don't think
> that using pid here makes any sense at all, especially
> having in mind the double-fork the daemon is doing
> at start.  I think that we should continue using the
> nbd device name here as before.

That is not possible.  The socket is needed by nbd_init, and you cannot 
know the name of the socket until you know the device name.

The socket name is really an internal detail when using -c.

> Note also that, while it is not currently supported
> anyway, you're making this more difficult to change
> the socket path by hardcoding it into two places.

Yes, that's true.  He could have two definitions (SOCKET_PATH with %d 
and SOCKET_PATH_HELP without) so that the occurrences would stay close 
in the source code.

>>   "  -r, --read-only      export read-only\n"
>>   "  -P, --partition=NUM  only expose partition NUM\n"
>>   "  -s, --snapshot       use snapshot file\n"
>>   "  -n, --nocache        disable host cache\n"
>>   "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
>> +"  -f, --find           find a free NBD device and connect FILE to it\n"
>>   "  -d, --disconnect     disconnect the specified device\n"
>>   "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
>>   "  -t, --persistent     don't exit on the last connection\n"
>> @@ -69,7 +70,7 @@ static void usage(const char *name)
>>   "  -V, --version        output version information and exit\n"
>>   "\n"
>>   "Report bugs to<anthony@codemonkey.ws>\n"
>> -    , name, NBD_DEFAULT_PORT, "DEVICE");
>> +    , name, NBD_DEFAULT_PORT);
>>   }
>>
>>   static void version(const char *name)
>> @@ -194,7 +195,8 @@ static void *show_parts(void *arg)
>>
>>   static void *nbd_client_thread(void *arg)
>>   {
>> -    int fd = *(int *)arg;
>> +    int fd = -1;
>> +    int find = *(int *)arg;
>
> You also can use !device condition here instead of extra
> variable: if device is set (non-NULL) use it, if it is not
> set, find.  This way there's no need to pass any args to
> this function at all.  But that's just my taste, nothing
> more.
>
>>       off_t size;
>>       size_t blocksize;
>>       uint32_t nbdflags;
>> @@ -213,9 +215,42 @@ static void *nbd_client_thread(void *arg)
>>           goto out;
>>       }
>>
>> -    ret = nbd_init(fd, sock, nbdflags, size, blocksize);
>> -    if (ret == -1) {
>> -        goto out;
>> +    if (!find) {
>> +        fd = open(device, O_RDWR);
>> +        if (fd == -1) {
>> +            fprintf(stderr, "Failed to open %s\n", device);
>> +            goto out;
>> +        }
>> +
>> +        ret = nbd_init(fd, sock, nbdflags, size, blocksize);
>> +        if (ret == -1) {
>> +            goto out;
>> +        }
>> +    } else {
>> +        int i = 0;
>> +        int max_nbd = 16;
>> +        device = g_malloc(64);
>> +
>> +        for (i = 0; i<  max_nbd; i++) {
>> +            snprintf(device, 64, "/dev/nbd%d", i);
>> +            fd = open(device, O_RDWR);
>> +            if (fd == -1) {
>> +                continue;
>> +            }
>> +
>> +            if (nbd_init(fd, sock, nbdflags, size, blocksize) == -1) {
>> +                close(fd);
>> +                continue;
>> +            }
>
> Here, I'm not sure it should ignore all possible errors.
> How about, say, EMFILE, or ENOENT, or lots of other possible
> error conditions which may be reported here instead of
> EBUSY?  Not that it is very important again...

Yes, though actually ENOENT is an important special case that is worth 
reporting to the user.  It happens when the nbd module is not loaded.

>> +
>> +            break;
>> +        }
>> +
>> +        if (i>= max_nbd) {
>> +            fprintf(stderr, "Fail to find a free nbd device\n");
>
> in all other places in qemu-nbd.c error reporting is done
> differently, namely, using err() or errx() routine (or
> warn()).  The difference is important: err(3) also reports
> strerror(errno) if errno is nonzero, so it will be clear
> _which_ error happened.  I understand that usage of err(3)
> here is a bit more fragile since it is not a main thread.
> Besides, it is "Failed" not "Fail".

Indeed, I suggested using errx in my review but err is better because of 
ENOENT.

Paolo