| Summary: | openat2 syscall unimplemented on Linux | ||
|---|---|---|---|
| Product: | [Developer tools] valgrind | Reporter: | Drew DeVault <sir> |
| Component: | general | Assignee: | Julian Seward <jseward> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | pjfloyd, sam |
| Priority: | NOR | ||
| Version First Reported In: | unspecified | ||
| Target Milestone: | --- | ||
| Platform: | Other | ||
| OS: | Linux | ||
| Latest Commit: | Version Fixed In: | ||
| Sentry Crash Report: | |||
| Attachments: |
Program which demonstrates the use of the openat2 syscall
diff for opnenat2 patch for openat2 patch for openat2 |
||
|
Description
Drew DeVault
2021-03-02 20:03:21 UTC
Do you have a small testcase to reproduce this? Created attachment 136343 [details]
Program which demonstrates the use of the openat2 syscall
Sure thing. Here's a test program.
Created attachment 143181 [details]
diff for opnenat2
Here is the start of an implementation. Mostly it's just a copy/paste of sys_openat, with the args changed to use struct open_how*, and also to do more thorouhg checking of the args.
I'm not sure if the values of VKI_O_DIRECTORY / VKI___O_TMPFILE / VKI_O_TMPFILE are generic across Linux hardware platforms. If they are then they need to be moved to include/vki/vki-linux.h. If not they need to be moved to the platform versions of vki-linux.h
Created attachment 143183 [details]
patch for openat2
This time
- make sure masks are extended to unsigned long long
- use resolved field rather than mistakenly mode
I had a little difficulty compiling the example code, ended up copying/pasting a few ofd the octal flags from /usr/include/asm-generic/fcntl.h directly into the file.
I based the mode/flags and resolve checks on this comment in openat.h
* Arguments for how openat2(2) should open the target path. If only @flags and
* @mode are non-zero, then openat2(2) operates very similarly to openat(2).
*
* However, unlike openat(2), unknown or invalid bits in @flags result in
* -EINVAL rather than being silently ignored. @mode must be zero unless one of
* {O_CREAT, O_TMPFILE} are set.
Created attachment 143184 [details]
patch for openat2
Third time lucky with the flag checks.
Hmm. I should perhaps land this. Need some tests (scalar at least). Should be fixed with this commit commit 2c93baf53a652fd503c0c21bccb7e85363f43e58 (HEAD -> master, origin/master, origin/HEAD) Author: Paul Floyd <pjfloyd@wanadoo.fr> Date: Fri Jan 6 21:27:01 2023 +0100 Bug 433873 - openat2 syscall unimplemented on Linux This syscall is not exposed via glibc, so there is only an addition to the scalar test. |