Bug 82271
| Summary: | Can't bind UNIX socket through file descriptor opened with O_PATH | ||
|---|---|---|---|
| Product: | Networking | Reporter: | Steven Stewart-Gallus (sstewartgallus00) |
| Component: | Other | Assignee: | Stephen Hemminger (stephen) |
| Status: | RESOLVED WILL_NOT_FIX | ||
| Severity: | enhancement | CC: | alan, shemming |
| Priority: | P1 | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Kernel Version: | 3.13 | Subsystem: | |
| Regression: | No | Bisected commit-id: | |
| Attachments: | generic open program | ||
This is not possible given the UNIX domain socket API is permanently fixed by Posix standards. The only O_* constants defined by POSIX 2013 are: O_CLOEXEC, O_CREAT, O_DIRECTORY, O_EXCL, O_NOCTTY, O_NOFOLLOW, O_TRUNC, O_TTY_INIT, O_APPEND, O_DSYNC, O_NONBLOCK, O_RSYNC, O_SYNC, O_ACCMODE, O_EXEC, O_RDONLY, O_RDWR, O_SEARCH and O_WRONLY. Stephen Hemminger, behaviour relating to O_PATH has absolutely zero to do with UNIX sockets. If you have some other meaning you are unintelligible. s/with UNIX sockets/with POSIX/g > If the address family of the socket is AF_UNIX and the pathname in
> address names a symbolic link, bind() shall fail and set errno to
> [EADDRINUSE].
- POSIX 2013
Ah, I see what you mean now. That's really silly but I'll try to think
and see if there's an alternate way of getting what I want.
|
Created attachment 146391 [details] generic open program One might want to open files generically. However, one cannot open a UNIX socket file found in the filesystem. That fails with ENXIO. To workaround this one can use stat to check if the file is a socket first but that is racy. O_PATH offers a possible solution. As in the program below, one can open a file with O_PATH, check the file type with fstat and then open (and in the case of a socket bind or connect to) the file through the file descriptor. Unfortunately, while this works with connect, and open it fails with bind with EADDRINUSE.