I don't see how this can help. Instead of allocating in kernel a buffer large enough, you allocate a small one and then for almost every output line you allocate a larger one, copy over then contents and free the smaller one. At the end, you have to allocate the same large buffer as in the old solution. You just replace one allocation by n allocation, were n is in the order of the number of lines.

I don't think this helps in allocating, but it definitely hurst in performance, I think.

Unless I am mistaken, sbuf_new_for_sysctl does NOT set SBUF_AUTOEXTEND. It rather sets the drain function (SYSCTL_OUT), and sbuf_put_bytes will call that function rather than extending the buffer (even if SBUF_AUTOEXTEND would be set).

So, at least from the sbuf infrastructure, this seems to be the right call to make. If the SYSCTL_OUT is doing re-allocs and copy-overs (especially when the caller has not provided any buffer, but I believe in that instance, it will actually only count up the bytes needed, but never actually copy memory), i think thats needed to be fixed there. If the buffer provided to SYSCTL_OUT is large enough to hold the data, only small chunks should be copied over repeatedly, I think.

You are absolutely right. Using sbuf_new_for_sysctl() allows the operation to succeed with using only a small buffer. I sent a patch via e-mail, which I used for testing and which reduces the number of whitespace changes, add a simplified way of calculating the size. Up to you what you take from it.