On 09/18/2011 09:59 PM, Paul Eggert wrote:
* lib/utimens.c (fdutimens): Don't invoke dup2; it's not needed
to work around the Linux kernel bug.
* modules/utimens (Depends-on): Remove dup2.
---
ChangeLog | 7 +++++++
lib/utimens.c | 13 ++++---------
modules/utimens | 1 -
3 files changed, 11 insertions(+), 10 deletions(-)
- /* Require that at least one of FD or FILE are valid. Works around
+ /* Require that at least one of FD or FILE are potentially valid, to avoid
a Linux bug where futimens (AT_FDCWD, NULL) changes "." rather
than failing. */
- if (!file)
+ if (fd< 0&& !file)
{
- if (fd< 0)
- {
- errno = EBADF;
- return -1;
- }
- if (dup2 (fd, fd) != fd)
- return -1;
+ errno = EBADF;
+ return -1;
}
I take it that the reason this still worked for a closed fd >= 0 is that
the kernel properly returns EBADF later on when we attempt the futimens;
and it was only negative fds where the kernel/glibc issue was leaking
through. But I don't see that test currently in the gnulib test suite.
So I'm pushing this additional followup, to fix that shortcoming in
the testsuite.
diff --git i/ChangeLog w/ChangeLog
index ddbbe0e..32f3723 100644
--- i/ChangeLog
+++ w/ChangeLog
@@ -1,5 +1,9 @@
2011-09-19 Eric Blake <ebl...@redhat.com>
+ futimens: enhance test
+ * tests/test-futimens.h (test_futimens): Also check for EBADF on
+ closed non-negative fd.
+
date: accept 'hence' as opposite of 'ago'
* lib/parse-datetime.y (relative_time_table): Add 'hence'.
* tests/test-parse-datetime.c (main): Enhance test.
diff --git i/tests/test-futimens.h w/tests/test-futimens.h
index 13e7d92..451edd9 100644
--- i/tests/test-futimens.h
+++ w/tests/test-futimens.h
@@ -81,6 +81,14 @@ test_futimens (int (*func) (int, struct timespec
const *),
ASSERT (func (-1, NULL) == -1);
ASSERT (errno == EBADF);
{
+ int fd = dup (0);
+ ASSERT (0 <= fd);
+ ASSERT (close (fd) == 0);
+ errno = 0;
+ ASSERT (func (fd, NULL) == -1);
+ ASSERT (errno == EBADF);
+ }
+ {
struct timespec ts[2] = { { Y2K, UTIME_BOGUS_POS }, { Y2K, 0 } };
errno = 0;
ASSERT (func (fd, ts) == -1);
--
Eric Blake ebl...@redhat.com +1-801-349-2682
Libvirt virtualization library http://libvirt.org
