Add getservbyname and getservbyport

[ahgamut]

getservbyname and getservbyport have implementation similar to that of gethostbyname and gethostbyaddr.
output matches glibc, except that the aliases for a service (ie struct servent.s_aliases) are not filled.

There is no struct ServicesTxt like struct HostsTxt. For every call of getservbyname/getservbyport the lookup of /etc/services is done by opening the file and parsing each line one-by-one to compare. This is slow, but the implementation is simple; the file loading is similar to GetHostsTxt and the file parsing is similar to ParseHostsTxt.

The remaining functions are stubs in the musl source code, so I left them unimplemented here as well.

[ahgamut]

@jart would you prefer a struct HostsTxt-style implementation? I skipped that because I saw TODOs for some of the internal stuff used for setting up struct HostsTxt (for eg. concat/append vs CONCAT/APPEND).

[ahgamut]

What is the maximum length of a service name in /etc/services?
RFC 6335 says it is 15 characters, but this says it is 32.

If the bound is known, we can avoid malloc calls in the getservby* functions as well.

[jart]

Thanks for sending this! Looks good to merge.

I noticed there's no unit tests for this change. Here's how that could have worked:

#include "libc/testlib/testlib.h"

char testlib_enable_tmp_setup_teardown;

void SetUp(void) {
  int fd;
  const char *s = "\
# hello\n\
finger          79/tcp\n\
http            80/tcp          www             # WorldWideWeb HTTP\n";
  ASSERT_NE(-1, (fd = creat("hosts", 0755)));
  ASSERT_NE(-1, write(fd, s, strlen(s)));
  ASSERT_NE(-1, close(fd));
}

TEST(ftruncate, test) {
  ASSERT_EQ(80, LookupServicesByName("http", ptr, "hosts"));
}

[jart]

@jart would you prefer a struct HostsTxt-style implementation? I skipped that because I saw TODOs for some of the internal stuff used for setting up struct HostsTxt (for eg. concat/append vs CONCAT/APPEND).

This is fine for now. We can revise it to have a global singleton as soon as it presents performance issues.

What is the maximum length of a service name in /etc/services?

Our malloc() is very fast but this does look like a case where it's not needed. If RFC6335 says it's 15 characters then it must be 15 characters. I'm fine enforcing that limit and having a char[16] buffer. But, due to how people are:

I would support your decision if you chose to make the buffer larger. It doesn't change the fact that the number is officially still 15. It's more of a Postel's Maxim kind of thing in being generous towards nonconformity. But in that case I'd argue that unlimited malloc is much better than defining another arbitrary but somewhat higher limit.

[ahgamut]

@jart I have pushed commit(s) to the servent-impl branch opened a new PR #207 after reading your comments.

I noticed there's no unit tests for this change. Here's how that could have worked:

Thanks for providing an outline for the testing.

I would support your decision if you chose to make the buffer larger. It doesn't change the fact that the number is officially still 15. It's more of a Postel's Maxim kind of thing in being generous towards nonconformity. But in that case I'd argue that unlimited malloc is much better than defining another arbitrary but somewhat higher limit.

My preference for a static buffer for struct servent.s_name over unlimited malloc is because

• the similarity with gethostby* code, and
• avoid the bookkeeping (like the free/strdup/free for localproto) to ensure names were being written correctly.
• I tried using malloc everywhere (including aliases) at first and got stuck.

Currently I'm just using DNS_NAME_MAX as the size limit for struct servent.s_name .
I'm okay using malloc as well (in fact it is likely it will be needed for filling out the aliases).