Add Unix domain socket support for listenTCP/listenHTTP for libevent on Posix #2073
Conversation
| { | ||
| bind_addr.family = AF_UNIX; | ||
| sockaddr_un* s = bind_addr.sockAddrUnix(); | ||
| enforce(s.sun_path.length > address.length, "Unix sockets cannot have that long a name."); |
There was a problem hiding this comment.
"that long a name" - it's generally good to be explicit in the error messages. The second part of enforce is lazy, so you can use things like format.
There was a problem hiding this comment.
Also this needs to be >= as \0 takes up space too.
There was a problem hiding this comment.
This is the same enforce and message that was used for the UDS implementation of requestHTTP.
There was a problem hiding this comment.
Phobos doesn't seem to specify the length either, but I could change it to enforce(s.sun_path.length > address.length, "Unix sockets cannot have names longer than %d characters.".format(s.sun_path.length - 1));?
There was a problem hiding this comment.
It shouldn't have to be >=, s.sun_path is a byte[].
There was a problem hiding this comment.
AFAICS, the path must always be \0 terminated regardless of the buffer length defined in sun_path, so using >= seems to be necessary indeed. The stdcpy below could otherwise also produce a buffer overflow when writing the zero byte.
There was a problem hiding this comment.
We're copying from address (string) to sun_path (byte[somelength]). Wouldn't > rather than >= make sure sun_path is large enough to hold address and \0?
| sockaddr_un* s = bind_addr.sockAddrUnix(); | ||
| enforce(s.sun_path.length > address.length, "Unix sockets cannot have that long a name."); | ||
| s.sun_family = AF_UNIX; | ||
| () @trusted { strcpy(cast(char*)s.sun_path.ptr,address.toStringz()); } (); |
There was a problem hiding this comment.
The toStringz seems to be unnecessary here. Why don't you simply copy it over (e.g. s[] = address[]) and then set the next char to \0?
There was a problem hiding this comment.
Also copied from the UDS implementation of requestHTTP.
There was a problem hiding this comment.
s.sun_path is a byte[]. It could be changed to () @trusted { s.sun_path[0 .. address.length] = (cast(byte[])address)[]; s.sun_path[address.length] = '\0'; } ();, which is more or less what Phobos does. Would that be better?
There was a problem hiding this comment.
Would be better in many cases because it always avoids a GC allocation, while toStringz needs to allocate if the string is not already zero terminated in memory for some reason (e.g. string literal or in a buffer that is padded with zeros).
| "dependencies": { | ||
| "vibe-d:http": {"path": "../../"}, | ||
| "vibe-d:web": {"path": "../../"}, | ||
| "vibe-d:core": {"path": "../../"} |
There was a problem hiding this comment.
FWIW you would only need vibe-d:web - it properly specified the dependencies on the other modules.
There was a problem hiding this comment.
Hmm, I think I copied this from another example's dub.json. Might be worth taking a look at others to see if they have unnecessary dependencies as well then.
There was a problem hiding this comment.
vibe-d:core is still required for the server example so the libevent configuration can be used for it, otherwise dub would say The sub configuration directive "vibe-d:core" -> "libevent" references a package that is not specified as a dependency and will have no effect.. Removing the vibe-d:http dependency from the server example and vibe-d:web from the client example though.
| import vibe.inet.url; | ||
| import vibe.http.server; | ||
| import vibe.http.router; | ||
| import std.stdio; |
There was a problem hiding this comment.
Use vibe.core.log for logging and in this case this looks like a left over.
There was a problem hiding this comment.
Yeah sorry, that's a leftover, will fix that.
| auto settings = new HTTPServerSettings; | ||
| settings.bindAddresses = ["/tmp/vibe.sock"]; | ||
|
|
||
| listenHTTP(settings, router); |
There was a problem hiding this comment.
Have a look at the other tests for examples on how you can verify the output of the server with requestHTTP
There was a problem hiding this comment.
Will check that out, thanks.
There was a problem hiding this comment.
Hmm, I'm not sure how much sense it would make adding tests for this PR considering the client side (i.e. requestHTTP, in master since 33027ee) for UDS doesn't have tests either.
There was a problem hiding this comment.
Would surely be good to add a test that tests both sides at once with a simple request. But since it depends on vibe-core fixes to make it work there, too, I'd defer that, though and treat this topic WIP in the meantime.
|
@wilzbach |
| import core.sys.posix.sys.un; | ||
| import core.stdc.string : strcpy; | ||
|
|
||
| if (address[0] == '/') |
There was a problem hiding this comment.
This disallows relative paths, is that necessary?
There was a problem hiding this comment.
No, but we couldn't simply check if it's a valid path because an IP address might also be a valid path.
@s-ludwig Any objections to using std.path.isValidPath(address) && std.socket.getAddressInfo(address, AddressInfoFlags.NUMERICHOST).empty here? Or is there a way to do this with vibe.core?
We could also just change it to address.canFind('/'), but then we wouldn't support Windows paths and paths below the working directory without ./.
|
I'm interested in UNIX socket implementations for vibe.d and landed on this page. Is this still, or already a thing? Simply to be informed. |
|
AFAICS, there are three open review comments that keep this from getting merged. @lesderid, if you don't have time to address them, I could probably take care of them in the near future, so that the remaining changes don't get lost here. |
|
@s-ludwig I should have time to get back to this PR this week or the next. Sorry for waiting so long with this. |
No description provided.