os: apparent deadlock in TestFifoEOF on `dragonfly`

[gopherbot]

#!watchflakes
post <- goos == "dragonfly" && pkg == "os" && test == "TestFifoEOF" && `panic: test timed out`

Issue created automatically to collect these failures.

Example (log):

panic: test timed out after 3m0s
running tests:
	TestFifoEOF (3m0s)

syscall.Syscall(0x5, 0xc0000ba210, 0x20001, 0x0)
	/tmp/workdir/go/src/syscall/asm_unix_amd64.s:19 +0x9 fp=0xc0000afde0 sp=0xc0000afdd0 pc=0x47b169
syscall.Open({0xc0000ba1b0?, 0x28?}, 0x59e080?, 0x0)
	/tmp/workdir/go/src/syscall/zsyscall_dragonfly_amd64.go:848 +0x87 fp=0xc0000afe50 sp=0xc0000afde0 pc=0x479b07
os.open(...)
	/tmp/workdir/go/src/os/file_open_unix.go:15
os.openFileNolog({0xc0000ba1b0, 0x2f}, 0x1, 0x0)
	/tmp/workdir/go/src/os/file_unix.go:272 +0xe5 fp=0xc0000afea0 sp=0xc0000afe50 pc=0x49dec5
os.OpenFile({0xc0000ba1b0, 0x2f}, 0x1, 0x621dc4?)
	/tmp/workdir/go/src/os/file.go:334 +0x3e fp=0xc0000afed8 sp=0xc0000afea0 pc=0x49b8de
os_test.TestFifoEOF(0xc000146b60)
	/tmp/workdir/go/src/os/fifo_test.go:50 +0x168 fp=0xc0000aff70 sp=0xc0000afed8 pc=0x53b788
testing.tRunner(0xc000146b60, 0x5c59f0)

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "os" && test == "TestFifoEOF"

2023-09-18 20:29 dragonfly-amd64-622 go@5b123aea os.TestFifoEOF (log)

panic: test timed out after 3m0s
running tests:
	TestFifoEOF (3m0s)

syscall.Syscall(0x5, 0xc0000ba210, 0x20001, 0x0)
	/tmp/workdir/go/src/syscall/asm_unix_amd64.s:19 +0x9 fp=0xc0000afde0 sp=0xc0000afdd0 pc=0x47b169
syscall.Open({0xc0000ba1b0?, 0x28?}, 0x59e080?, 0x0)
	/tmp/workdir/go/src/syscall/zsyscall_dragonfly_amd64.go:848 +0x87 fp=0xc0000afe50 sp=0xc0000afde0 pc=0x479b07
os.open(...)
	/tmp/workdir/go/src/os/file_open_unix.go:15
os.openFileNolog({0xc0000ba1b0, 0x2f}, 0x1, 0x0)
	/tmp/workdir/go/src/os/file_unix.go:272 +0xe5 fp=0xc0000afea0 sp=0xc0000afe50 pc=0x49dec5
os.OpenFile({0xc0000ba1b0, 0x2f}, 0x1, 0x621dc4?)
	/tmp/workdir/go/src/os/file.go:334 +0x3e fp=0xc0000afed8 sp=0xc0000afea0 pc=0x49b8de
os_test.TestFifoEOF(0xc000146b60)
	/tmp/workdir/go/src/os/fifo_test.go:50 +0x168 fp=0xc0000aff70 sp=0xc0000afed8 pc=0x53b788
testing.tRunner(0xc000146b60, 0x5c59f0)

— watchflakes

[bcmills]

(attn @golang/dragonfly)

[bcmills]

See previously #25877, #36107, #55207.

[bcmills]

Note that in the goroutine dump, the os.Open goroutine is not present, suggesting that that call to syscall.Open succeeded (or perhaps failed spuriously?) and unblocked.

(https://cs.opensource.google/go/go/+/master:src/os/fifo_test.go;l=31-53;drc=f777726ff073f8066c017649b572bd8c40940a42)

[tuxillo]

I'm thinking the first step is probably updating that build environment from 6.2 to 6.4 .

I've tried the OpenBSD example code in #25877 but it finishes ok, at least in a current DragonFly version:

❯ ./t_fifo
reader: os.OpenFile
writer: os.OpenFile
reader: os.OpenFile done
❯ go version
go version go1.20.3 dragonfly/amd64
                       

[bcmills]

Note that the rate of these failures on the dragonfly builder is quite low. (I suspect it's a race condition somewhere, which would make it difficult to reproduce.)