Resolve calls to functions that use go:linkname

[ldemailly]

git clone https://github.com/fortio/log
cd log
# btw... consider moving the main at the top for shorter install cmd/ is overused
go install github.com/google/capslock/cmd/capslock@latest 
# ( got capslock v0.2.4 )
capslock --output=verbose
[...]
CAPABILITY_NETWORK: 7 references (7 direct, 0 transitive)
Example callpath:
  fortio.org/log.LogAndCall$1$1
  http_logging.go:202:15:fortio.org/log.LogRequest
  http_logging.go:84:56:(net/http.Header).Get
[...]
CAPABILITY_ARBITRARY_EXECUTION: 30 references (0 direct, 30 transitive)
Example callpath:
  fortio.org/log.init#1
  logger.go:190:15:fortio.org/log.configFromEnv
  logger.go:201:8:fortio.org/log.Errf
  logger.go:520:11:fortio.org/log.logPrintf
  logger.go:414:19:fortio.org/log.logUnconditionalf
  logger.go:447:31:fortio.org/log.colorGID
  console_logging.go:144:55:fortio.org/log/goroutine.ID
  gid.go:20:22:fortio.org/log/goroutine.add

https://github.com/fortio/log

doing http.Header.Get shouldn't be "CAPABILITY_NETWORK" should it ?

and the hacky code in https://github.com/fortio/log/blob/main/goroutine/gid.go#L20
is using unsafe but... how does that turn into CAPABILITY_ARBITRARY_EXECUTION ?

[jcd2]

Thanks for the report!

doing http.Header.Get shouldn't be "CAPABILITY_NETWORK" should it ?

You're right, I'll make a change soon to add this (and some similar functions) to Capslock's list of functions under net/ that don't really have the network capability.

how does that turn into CAPABILITY_ARBITRARY_EXECUTION ?

It's the call on that line to a go:linkname function that gets it marked as arbitrary execution. Capslock isn't clever enough to follow go:linkname directives, but we'll look into adding that in the future.

[ldemailly]

ping?

[jcd2]

We've narrowed the set of functions in net that have capabilities, but I'm afraid we haven't done any work on resolving the effects of linkname yet, which would probably be implemented upstream of Capslock in golang.org/x/tools. I'll retitle this issue to cover that part.

[ldemailly]

Thanks for the reply,

We've narrowed the set of functions in net that have capabilities

I still get

CAPABILITY_NETWORK: 6 references (6 direct, 0 transitive)
Example callpath:
  fortio.org/log.LogAndCall$1
  http_logging.go:189:4:fortio.org/log.LogAndCall$1$1
  http_logging.go:197:17:net/http.Error

(maybe because of init/globals in net/http? but should callers of http.Header.Get get that label?)

ps: also tried with
$ go install github.com/google/capslock/cmd/capslock@main
go: downloading github.com/google/capslock v0.2.8-0.20250311052645-d72b1ed2fa57
same

[jcd2]

I still get

CAPABILITY_NETWORK: 6 references (6 direct, 0 transitive)
Example callpath:
  fortio.org/log.LogAndCall$1
  http_logging.go:189:4:fortio.org/log.LogAndCall$1$1
  http_logging.go:197:17:net/http.Error

Well, net/http.Error here writes an error response that gets sent over the network:

https://github.com/fortio/log/blob/main/http_logging.go#L197

so IMO it should have the network capability.

[ldemailly]

Indeed I hadn't noticed that the method flagged has changed, yet http.Error imo isn't network capabilites, it doesn't do anything if you don't have a prexisting http.Response - network capability would be imo either listening to a port or opening a new connection (eg either new server or new client sockets underlying); neither of which does that code do

[jcd2]

That's true, it's only affecting a network connection that already exists.

We've been thinking about adding finer-grained capabilities, and your case here would be a good one for us to improve on: distinguishing between opening or listening for new connections, and reading or writing data.

[ldemailly]

thanks, would io.Write be flagged as “network capabilities” though ? it shouldn’t imo

[jcd2]

thanks, would io.Write be flagged as “network capabilities” though ? it shouldn’t imo

If the Write call sends bytes over an existing network connection, I think it should. But ideally it would be a different, lesser capability than opening a new connection.