Skip to contentSkip to conversation

maximum of 256 loop devices - not enough for cdrom server

DISCUSSION roland 3 participants 3 messages May 26, 2005 – Jun 4, 2005

lore
Share & tools
lore

roland

author
#0 lore raw

Hello !
I want to build a cd-rom server and want to use the loop-driver (loop.c) for that.
That server will be HUGE and I will need to loopback mount more than 256 .iso images in the near future. Unfortunately, the loop driver doesn`t support >256 device nodes.

Is there a possible solution for this "problem" ?

regards
roland

#1 lore raw
Quoted 2 lines · roland

Is there a possible solution for this "problem" ?

Beware: not a kernel hacker, not even a proficient programmer.

It seems it would be easy, just changing the max current loop device number value (256) to something higher (i.e., 4096). Note, however, that loop devices are described by an array of structs statically allocated at module load time, so expect memory usage to grow linearly and search operations on the array become slower at least linearly.

There is one thing that maybe would need some attetion: minor number allocation for loopback block devices. Now they are assigned at module initialization, but I don't know the current status in the Linux kernel with respect to (major,minor) sizes and allocation.

Hope someone more knowledgeable jumps into this :-)

Greetings,

Signature · 2 lines
Jose Luis Domingo Lopez
Linux Registered User #189436     Debian Linux Sid (Linux 2.6.11.10)
#2 lore raw

On Thu, 26 May 2005 22:14:38 +0200 Jose Luis Domingo Lopez wrote:

| On Thursday, 26 May 2005, at 03:06:34 +0200,
| roland wrote:
| 
| > I want to build a cd-rom server and want to use the loop-driver (loop.c) for that.
| > That server will be HUGE and I will need to loopback mount more than 256 .iso images in the near future.
| > Unfortunately, the loop driver doesn`t support >256 device nodes.
| > 
| > Is there a possible solution for this "problem" ?

First, let's understand the kernel version. 2.4.x or 2.6.y? I'm only looking at 2.6.y.

Roland, have you tried increasing the max. value of max_loop yet?

| Beware: not a kernel hacker, not even a proficient programmer.
| 
| It seems it would be easy, just changing the max current loop device
| number value (256) to something higher (i.e., 4096). Note, however, that
| loop devices are described by an array of structs statically allocated at
| module load time, so expect memory usage to grow linearly and search operations 
| on the array become slower at least linearly.

The loop_device array is kmalloc-ed at load/init time.
On x86 (32-bit), struct loop_device is 336 bytes, and with kmalloc() having an upper limit of 128 KB, the max. number of loop_devices that can be allocated at one time is 128 * 1024 / 336 = 390, not a huge improvement over 256 loopback devices.

I don't know of any reason that vmalloc() couldn't be used here instead of kmalloc(), and once that hurdle is cleared (the 128 KB one), since minor number on 2.6.y is 20 bits, there can be plenty of loopback devices AFAIK.

I'm sure that Al or probably Andrew could answer this reliably.

| There is one thing that maybe would need some attetion: minor number
| allocation for loopback block devices. Now they are assigned at module
| initialization, but I don't know the current status in the Linux kernel
| with respect to (major,minor) sizes and allocation.
| 
| Hope someone more knowledgeable jumps into this :-)

---
~Randy
http://www.madrone.org/donate/RFD1.html