• Add test for PKT_ALIAS_RESET_ON_ADDR_CHANGE

• Test case for PKG_ALIAS_SAME_PORTS

• Fix case 4_udp and various issues

• Fix stupid errors (reversed length check, existing nat state)

• Factor out common utilities
• Zeroize packet buffer

I'm far from finished with this.
But thank you @kp for the intermediate, positive response.

• Simplify structure
• More simplification.
• Add UDP and reverse translation
• Fix packet header size calculation

• Add UDP and reverse translation

• More simplification.

• Simplify structure

dn_aqm_pie.c and dn_sched_fq_pie.c would considerably improve by replacing the t = DROP by goto drop.

• Fix spelling

• Fix build process

• Add to global build.

• Move location of test dir and update build procedure

In D30307#680794, @kp wrote:

I think you're also missing a makefile hookup somewhere. These tests may want to live in /usr/src/tests/sys/netinet/libalias instead, but I don't hold strong views on that.

• Renumber the test cases and rename the file according to scope

• Replace preprocessor constants by compiler generated values.

• Use OBJDIR
• Make random reproducable.
• Inspect kyua reports

Added @lev primary for testing of functionality.
There should be some performance improvements due to:

• some early breaks in the loops with many active flows
• separation of "complete" flows from configured "forwardings"
• use of NO_ADDR when possible

Added @lev for his offer to do performance tests.

• Make the tests work directly on the intermediate builds.

• Add a test for a known API violation.

• Test to create many instances and remove them in random order.

• Add test to recreate an instance multiple time.

• Fix meta data. => Now it works.

Ah, kyua report shows:

aliasdb_test:__test_cases_list__  ->  broken: Unknown test case metadata property 'desc'  [0.001s]

• Remove unnecessary static

• Simplify the test

Start as simple as possible and fail ...

[/usr/tests/sys/netinet/libalias]$ kyua list
aliasdb_test:__test_cases_list__

• Reclassify SCTP links when specifying missing parts.

• Further refactoring

In D30284#680337, @kp wrote:

In D30284#680334, @donner wrote:

k -> 6, i -> 1, to -> 2 => (p)fkkif(to)kif => 0xf661f261f

I think I prefer 0xFEEDC0DE, because it's more obviously human-generated and not a valid pointer.

• NO_ADDR for fragments
• Factor out common search pattens in in and out direction.

In D30284#680331, @kp wrote:

Although using a more-or-less unique combination would probably make it easier to figure out where it came from. On the third hand, pf_kkif_to_kif() is a pretty obvious place to look.

0xdeadc0de is used to signal a use after free.
Please consider a different code, which does not cause confusion in the future.

• Fix split point.

• NO_ADDR for fragments

In D30273#680148, @markj wrote:

In D30273#680142, @donner wrote:

In D30273#680113, @markj wrote:

checksum_ipv4() and checksum_ipv6() may fail without freeing the mbuf. In that case, with the patch the mbuf is leaked.

Can you point to a code line, where this can happen?
I didn't see one.

In checksum_ipv4() we call PULLUP_CHECK() on line 316. PULLUP_CHECK() looks like this:

293 #define PULLUP_CHECK(mbuf, length) do {                                 \                                                                                                 
294         pullup_len += length;                                           \                                                                                                 
295         if (((mbuf)->m_pkthdr.len < pullup_len) ||                      \                                                                                                 
296             (pullup_len > MHLEN)) {                                     \                                                                                                 
297                 return (EINVAL);                                        \                                                                                                 
298         }                                                               \                                                                                                 
299         if ((mbuf)->m_len < pullup_len &&                               \                                                                                                 
300             (((mbuf) = m_pullup((mbuf), pullup_len)) == NULL)) {        \                                                                                                 
301                 return (ENOBUFS);                                       \                                                                                                 
302         }                                                               \                                                                                                 
303 } while (0)

On line 297, we return an error without having freed the mbuf. On line 301 we return an error having freed the mbuf, and the caller of checksum_ipv4() will free the mbuf again. So in the first case, we are now not freeing the mbuf at all.

In D30273#680113, @markj wrote:

checksum_ipv4() and checksum_ipv6() may fail without freeing the mbuf. In that case, with the patch the mbuf is leaked.

In D30275#680105, @kp wrote:

struct libalias is exposed to userspace, right?

• Reestablish ABI without providing any functionality.
• Remove the last piece from the internal structure.

I do not take every hint from D30260.
Many of those compiler generated suggestions destroy the flow of reading.
Most of the others are a bit picky on the line length, but this is not reason to change existing code.

Thank you for the input.

• Incooperate style suggestions from D30260
• Canonify #ifdef spacing.

• Replace spaces with tabs

• Remove spaces before tabs.

• Fix balanced braces around "else".
• Fix some spacing.

Thank you for the cool idea! There are some interesting ideas which are worth to be applied.

Rebased to D30259

• Make it const.

• Silence warning about non-static declaration.