cp: Fix perm race

[max-amb]

Fixes #10011

[github-actions]

GNU testsuite comparison:

GNU test failed: tests/cp/abuse. tests/cp/abuse is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/cp/acl. tests/cp/acl is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/cp/link-deref. tests/cp/link-deref is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/cp/link-preserve. tests/cp/link-preserve is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/cp/perm. tests/cp/perm is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/cp/preserve-mode. tests/cp/preserve-mode is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/cp/same-file. tests/cp/same-file is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/cp/thru-dangling. tests/cp/thru-dangling is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/misc/xattr. tests/misc/xattr is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/mv/dup-source. tests/mv/dup-source is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/mv/part-symlink. tests/mv/part-symlink is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/mv/trailing-slash. tests/mv/trailing-slash is passing on 'main'. Maybe you have to rebase?
GNU test failed: tests/mv/update. tests/mv/update is passing on 'main'. Maybe you have to rebase?

[sylvestre]

Please don't open pr with unfinished code

[codspeed-hq]

CodSpeed Performance Report

Merging #10030 will degrade performance by 46.76%

_{Comparing max-amb:fix_cp_perm_race (48c693f) with main (b852b1b)}

Summary

❌ 6 regressions
✅ 133 untouched
⏩ 37 skipped¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Efficiency
❌	cp_large_file[16]	363.2 µs	386.2 µs	-5.96%
❌	cp_preserve_metadata[(5, 4, 10)]	75 ms	134.2 ms	-44.11%
❌	cp_recursive_wide_tree[(6000, 800)]	190 ms	356.8 ms	-46.76%
❌	cp_recursive_deep_tree[(120, 4)]	13.7 ms	21.7 ms	-36.91%
❌	cp_archive_balanced_tree[(5, 4, 10)]	77.9 ms	137.2 ms	-43.24%
❌	cp_recursive_balanced_tree[(5, 4, 10)]	72.4 ms	129.4 ms	-44.05%

Footnotes

1. 37 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩