fabrics: Simple URI parser

[tbzatek]

This is an initial implementation for a simple URI parser. Described in NVM Express Boot Specification (TP8012), used in struct nbft_info_discovery as a composite connection string, as opposed e.g. to separate struct fields in struct nbft_info_subsystem_ns.

Open questions/TODO:

• handle escaped ("percent-encoded") sequences
• any UTF-8 related tasks? (can't think of any right now)
• handle query
• handle fragment
• handle username + password

i.e. as defined in RFC 3986:

      URI         = scheme ":" hier-part [ "?" query ] [ "#" fragment ]

         foo://example.com:8042/over/there?name=ferret#nose
         \_/   \______________/\_________/ \_________/ \__/
          |           |            |            |        |
       scheme     authority       path        query   fragment

Applied to NVMe:

nvme+tcp://user:pass@host:port/subsys_nqn/nid?name=ferret#nose

Since the above tasks may involve struct nvme_fabrics_uri element changes, marking as WIP, seeking initial feedback. Any other possible elements to extract that I missed?

[tbzatek]

I think this is ready for review.

[tbzatek]

As a side note - while this work was initiated by the need from the NVM Express Boot Specification, I believe this libnvme URI parser could serve more general needs. There's simply so much potential in using URI as an alternative to the usual set of parameters. Perhaps more attractive to high-level/end-user-facing tools than low-level tools like nvme-cli. What about nvme-stas, any plans for using URIs, @martin-belanger ?

I haven't noticed any other mention elsewhere than in the Boot Specification. My primary concern is whether this API is universal enough before it's released and frozen. Alternatively, this might be moved into nvme-cli as a private API for nbft.c before it's mature enough.

[martin-belanger]

What about nvme-stas, any plans for using URIs, @martin-belanger ?

Hi @tbzatek - nvme-stas currently supports URIs read from NBFT records. We convert URIs into the familiar transport, traddr, and trsvcid parameters needed by libnvme. Since nvme-stas is written in Python, I was able to simply use the function urllib.parse.urlparse() without having to write my own URI parser.

Ref: https://github.com/linux-nvme/nvme-stas/blob/main/staslib/conf.py#L804

[tbzatek]

I was able to simply use the function urllib.parse.urlparse() without having to write my own URI parser.

I see, fair enough :-) I was originally considering using the uriparser C library but since it would be pulled in as dependency and we need to pack nvme-cli in initramfs, it was a no-go. Not to mention supporting old distros where we do regular rebases (RHEL 9).

[igaw]

My primary concern is whether this API is universal enough before it's released and frozen

I think we start with what you came up and see how it works out. In the worst case we introduce nvme_parse_uri2

[igaw]

Thanks!