Hi All!

I think I've spotted a bug in the initialisation procedure in the loop driver,
@line 1027-1060, when there is allocation failure:

----BEGIN CODE----
        loop_dev = kmalloc(max_loop * sizeof(struct loop_device), GFP_KERNEL);
        if (!loop_dev)
                return -ENOMEM;

        loop_sizes = kmalloc(max_loop * sizeof(int), GFP_KERNEL);
        if (!loop_sizes)
                goto out_sizes;

        loop_blksizes = kmalloc(max_loop * sizeof(int), GFP_KERNEL);
        if (!loop_blksizes)
                goto out_blksizes;

[...SNIP...]

out_sizes:
        kfree(loop_dev);
out_blksizes:
        kfree(loop_sizes);
----END CODE----

Two problems arise there:
1) Should allocation of loop_sizes fail, we'll try to free loop_sizes that
   was NOT allocated (no problem because kfree does not free a NULL pointer).
   But not very clean...

2) Should allocation of loop_blksizes fail, loop_dev will NOT be freed.
   That one looks more problematic.

Am I wrong? Have I missed something? Anyway it's been there for ages, I
suppose.

Patch is very straigthforward, and is attached : swap the last 2 pairs of
lines, so that out_blksizes is before out_sizes, and kfree(loop_sizes is
before kfree(loop_dev).

Please CC: answers and comments as I'm no longer subscribed.

Cheers,
Yann.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
|  0 662 376 056  | Software  Designer | \ / CAMPAIGN     |  ___               |
| --==< °_° >==-- °---.----------------:  X  AGAINST      |  \e/  There is no  |
| web: ymorin.free.fr | SETI@home  433 | / \ HTML MAIL    |   v   conspiracy.  |
°---------------------°----------------°------------------°--------------------°