Memory leak when using small IPv6 subnets?

[xnpu]

Is it possible to use smaller IPv6 subnets?

When I configure my ipv6-pool like this:

XXXX:YYYY:ZZZZ:1391:0000:0005::/96,127

memory usage will go up very quickly until finally the Linux kernel decides to kill accel-ppp:

Aug 18 02:53:00 xxx-1 kernel: [842624.475618] Out of memory: Kill process 10909 (accel-pppd) score 234 or sacrifice child
Aug 18 02:53:00 xxxx-1 kernel: [842624.475687] Killed process 10909 (accel-pppd) total-vm:5907704kB, anon-rss:5840872kB, file-rss:4kB

If I attach an strace, I'll see an endless list of these until the process is killed:

[pid 12066] brk(0xaaa51000) = 0xaaa51000
[pid 12066] brk(0xaaa72000) = 0xaaa72000
[pid 12066] brk(0xaaa93000) = 0xaaa93000
[pid 12066] brk(0xaaab4000) = 0xaaab4000
[pid 12066] brk(0xaaad5000) = 0xaaad5000
[pid 12066] brk(0xaaaf6000) = 0xaaaf6000

[Dmitry]

96 by 127 is very large, do you really need so much addresses ?

[xnpu]

~200 or so is sufficient. I did not realise it would pre-allocate memory for every possible address.

I've now changed it to /96,112 which works fine. Will probably narrow it even further. Thanks!