tail: reuse existing pipe when stdout is already a pipe

[dragutreis]

Summary

Fixes #12413

tail -c +1 /dev/null made an unnecessary pipe2 syscall because
splice_unbounded_broker unconditionally created a pipe before attempting
any data transfer.

Root cause

bounded_tail called print_target_section unconditionally even when the
file had no data to output. On Linux, print_target_section calls
splice_unbounded_broker, which always created a pipe regardless of whether
there was any data to transfer.

Fix

Add an is_file_exhausted helper and check it before calling
print_target_section in the unbounded case:

• Regular files: compare stream position against metadata length.
• Char/block devices (Unix): probe with a single-byte read; if data exists,
  write the byte directly to stdout since char devices don't support seeking.

[oech3]

It is not what I was expected at the linked PR. I meant | echo should allow reusing existing pipe |.

[oech3]

But it seems still useful for supressing other syscalls at some cases. Right?

[dragutreis]

Oh, thanks for clarifying, I'm not really sure what to do however, should I keep this pr and open a new one to fix the actual issue?

[oech3]

We should try splice_unbounded before _broker to omit unnecessary pipe broker (and myself tried to do that, but failed).

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/cut/bounded-memory (fails in this run but passes in the 'main' branch)
Skip an intermittent issue tests/date/resolution (fails in this run but passes in the 'main' branch)
Skip an intermittent issue tests/misc/tty-eof (fails in this run but passes in the 'main' branch)
Skip an intermittent issue tests/pr/bounded-memory (fails in this run but passes in the 'main' branch)
Note: The gnu test tests/printf/printf-surprise is now being skipped but was previously passing.
Note: The gnu test tests/rm/many-dir-entries-vs-OOM is now being skipped but was previously passing.
Congrats! The gnu test tests/tail/tail-n0f is now passing!

[oech3]

Actually, I was not able to fix it by splice_unbounded_auto which tries to avoid pipe2(). It might coming from different code path as we expected.

[dragutreis]

Hmmm, I'll look into it and try to fix it.

[oech3]

( to clarify, strace -c -e pipe,pipe2 tail -c +1 /dev/zero should pipe2, but strace -c -e pipe,pipe2 tail -c +1 /dev/zero|echo should not pipe2)

[dragutreis]

Used strace -k to track it down. The pipe2 call was coming from splice_unbounded_broker in pipes.rs, which unconditionally creates a broker pipe even when stdout is already a pipe. There's even a comment around line 109 noting it shouldn't be used when one side is already a pipe. Replacing it with splice_unbounded_auto fixes it, not sure why it didn't work when you tried.

$ strace -c -e pipe,pipe2 tail -c +1 /dev/zero
1 pipe2 (expected)
$ strace -c -e pipe,pipe2 sh -c 'tail -c +1 /dev/zero | echo'
1 pipe2 from bash only (expected)

[oech3]

Strange... But it it fixed the issue for you. Confirmed it. So this PR's diff is just 1 line.

[dragutreis]

Awesome, glad that fixed it.

[oech3]

Would you revert 1st change, make diff 1line, and edit description to note about _auto? Thanks.

[dragutreis]

Will do.