[Bug 203162] when close(fd) on a fifo fails with EINTR, the file descriptor is not really closed

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=203162

--- Comment #3 from Konstantin Belousov <kib at FreeBSD.org> ---
(In reply to Victor Stinner from comment #1)
I see the 'hang', where the parent is sleeping on open(O_WRONLY), and there is
no child.  The situation is not too mysterious, see the explanation below.

Test sets up two processes, parent does open(fifo, O_WRONLY), child does
open(fifo, O_RDONLY).  Either process can proceed further if other process
enters open(), or a signal is delivered interrupting the sleep waiting for the
residual open.  Assume that the parent sleeps in open waiting for child/reader,
child enters open and since there is writer waiting, the open in child
succeeded.  Child wakes up the parent to notify that open(O_WRONLY) can succeed
as well (there is reader after all).  And, the crusial point, assume that the
parent gets signal notification before it gets wakeup notice.

In this setup, open() in parent returns with EINTR.  On the other hand, child
successfully opened fifo for read, and after some more work, it exits.  But the
parent retries the open(O_WRONLY).  There is no child to open fifo, so open()
cannot succeed, it only gets interrupted by signal periodically.  And, parent
would not start another child until this open succeed.

Hopefully this can be followed.

FWIW, do not use truss, use kdump -i.

-- 
You are receiving this mail because:
You are the assignee for the bug.