Kernel crash on working with big memory chunk

[mihalicyn]

Consider the following code (write.lua):

local start = os.clock();

local f = io.open("/root/dev/write.lua", 'r')
local s = f:read('all')
f:close()

os.remove("/root/dev/out.txt");
local fout = io.open("/root/dev/out.txt", 'w+')
fout:write(s)
fout:close()

print("os.clock(): " .. (os.clock() - start) .. " seconds")

If we run that in user space - all is fine!
If we run that in kernel space we got copy of write.lua (content is code in listing).
If we try big file for example syssrc.tgz (~50MiB) we got freeze, try bigger (~150MiB) and got kernel crash (fatal page fault). Problem is in read_all function.

[mihalicyn]

After researching of problem internals I realized, that we have a problem with memory allocation (and we have no ENOMEM error and so on).

Consider the following code:

print("Strings test...")

string1 = "Hello, World!\n"
string = ""

count = 50000000

string = string1:rep(count)

print(string:len())
assert(string:len() == count * string1:len())

print("DONE")

NetBSD netbsd 7.1 NetBSD 7.1 (GENERIC.201703111743Z) i386

[mihalicyn]

PR/52225 created and fixed
http://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=52225

[mihalicyn]

PR/52225 solves problem for case when lua_alloc called in interruption handler context. In general case we can got kernel freeze:

		km_flag_t sleep = cpu_intr_p() || cpu_softintr_p() ?
			KM_NOSLEEP : KM_SLEEP;
//...

alloc_header_t *nhdr = kmem_intr_alloc(alloc_size, sleep);

For example. In my VirtualBox VM I have 1 GiB of RAM, and I trying to alloc ~300MiB and got freeze.