[Toybox] cpio cleanups.

[Rob Landley]

I've done three cpio cleanup commits so far:

http://landley.net/hg/toybox/rev/1221
http://landley.net/hg/toybox/rev/1222
http://landley.net/hg/toybox/rev/1223

The first two are smallish and have commit comments saying what I did.
Here's my notes from doing the third:

  - Tweak arguments: add exclusion groups, make one of -iot required.
  - Redo help text: tabs instead of spaces after -x

loopfiles_stdin()
  - is only ever called from one place, with one function pointer
    - just inline it
  - why stdin? Why not arbitrary file so it could handle tar -T?
  - no error message if listed file not opened?
    - who said you have read access?
  - perror_msg() prepends command name, lower level verror_msg() does not

write_cpio_member()
  - you have a char array, and then you typecast it to a struct you never
    dereference the fields of. Just use the char array. (This way we don't
    have to worry about struct padding inserted by insane compiler.)
    - in fact, just use toybuf instead of hdr, sprintf returns correct size
  - Simplify if() to !IS_REG() && !IS_LNK()
  - snprintf... we control the string, every field length controlled.
    No need to specify length. Collate the format string.
    - Make the last 00000000 another %08 and feed in a 0 arg, for consistency
    - Add some (int) typecasting to shut up compiler warnings
  - replace write() with xwrite() (catches write errors, retries short writes).
  - use TT.outfd instead of hardwired 1 (for -F support).
  - if readlink() fails, perror_msg() about it.
  - we readlink() llen bytes and then write buf.st_size bytes, yet don't
    compare them for equality?
  - Remove strange lseek behavior in writing files, instead accumulate
    length of reads to figure out current position in file.
  - size_t out can't ever be negative so read error is lost, make it ssize_t
    instead.
  - what is this (nlen+2)&3 business? Padding? What's the 2 for?
    Ah, the 2 is the header size (110) and 3. That was non-obvious. So
    we're rounding the header size up to a multiple of 4 bytes.

main():
  - Don't need else "must use -iot" if argument parsing | handled it.

Query: how to deal with files > 2gb?
  - Which is worse: dying noisily or silently truncating them?

todo:
  accept kernel directory format (ala get_init_cpio.c)
  output kernel listing format (ala gen_initramfs_list.sh)
  code sharing with tar and zip...?

Test:
  archive file with permission 000 (can't open it, but for /dev node...)

 1394916875.0