gh-153400: Add syscall fallbacks for copy_file_range/memfd_create

[daandemeyer]

glibc only grew copy_file_range() and memfd_create() in 2.27, and we compile the os functions out when the libc we build against doesn't have them. That loses them for good in a redistributable built against an older glibc, such as the python-build-standalone builds targeting glibc 2.17, even when the kernel it runs on implements the syscalls.

Keep calling the libc wrappers when they are available, so we don't lose their symbol versioning and _FORTIFY_SOURCE checks, and issue the syscall directly when they aren't. If the syscall number is missing as well, the functions are still left out. pidfd_open() and pidfd_getfd() already use raw syscalls, so nothing changes for them.

Include <sys/syscall.h> whenever it exists rather than only when the getrandom() syscall was detected, since _NR* is now needed for more than getrandom().

• Issue: Resolve newer Linux os functions at runtime instead of compiling them out against old glibc #153400

[read-the-docs-community]

Documentation build overview

📚 cpython-previews | 🛠️ Build #34049216 | 📁 Comparing 96a7a84 against main (219768f)

  🔍 Preview build  

25 files changed · ± 25 modified

± Modified

• library/ast.html
• library/concurrent.futures.html
• library/csv.html
• library/curses.html
• library/email.compat32-message.html
• library/email.contentmanager.html
• library/email.examples.html
• library/email.header.html
• library/email.headerregistry.html
• library/email.html
• and 15 more...

[vstinner]

The overall change LGTM.

[daandemeyer]

@vstinner Addressed comments, no other changes

[vstinner]

LGTM.

[jjhelmus]

This also closes #153463 and makes #153464 unnecessary

[vstinner]

I merged my PR gh-155518 which replace syscall() calls with glibc function calls. Oh, it created some conflicts in generated files. Would you mind to update your branch (git merge main) and resolve merge conflicts (make clinic)?

[daandemeyer]

@vstinner Done

[vstinner]

LGTM.

In general, I would prefer to only support an os function if it has a glibc (wrapper) function. But well, in practice, glibc takes time to add a new function. And getting the right {Python, glibc and kernel} combo can be complicated. So I'm fine with adding syscall() implementation for copy_file_range() and memfd_create().