Samuel Thibault, le jeu. 17 avril 2025 23:31:50 +0200, a ecrit:
> Zhaoming Luo, le lun. 10 mars 2025 16:44:09 +0800, a ecrit:
> > I haven't tested this patch again some stress tests as I don't know why
> > the `dpkg-buildpackage -B` on my qemu got stuck at:
> >
> > ```
> > ...
> > ../scripts/evaluate-test.sh stdio-common/scanf14 $? false false >
> > /home/1speaker/apt-source/glibc-2.41/build-tree/hurd-i386-libc/stdio-common/scanf14.test-result
> > ```
>
> It seems it is related: without your patch the test passes.
> > - /* Install the ports and flags in the new descriptor slot. */
> > - __spin_lock (&d2->port.lock);
> > - if (flags & O_CLOEXEC)
> > - d2->flags = d_flags | FD_CLOEXEC;
> > - else
> > - /* dup clears FD_CLOEXEC. */
> > - d2->flags = d_flags & ~FD_CLOEXEC;
> > - _hurd_port_set (&d2->ctty, ctty);
> > - _hurd_port_locked_set (&d2->port, port); /* Unlocks D2. */
[...]
> > + /* Install the ports and flags in the new descriptor slot. */
> > + __spin_lock (&d2->port.lock);
> > + if (flags & O_CLOEXEC)
> > + d2->flags = d_flags | FD_CLOEXEC;
> > + else
> > + /* dup clears FD_CLOEXEC. */
> > + d2->flags = d_flags & ~FD_CLOEXEC;
> > + if (ctty != MACH_PORT_NULL)
The original code didn't have that if. Setting a MACH_PORT_NULL port does make
sense and fixes the scanf14 test.
The rest looks good, I'll fix this and commit, thanks!
Samuel
> > + _hurd_port_set (&d2->ctty, ctty);
> > + _hurd_port_locked_set (&d2->port, port); /* Unlocks D2. */
> > + }
> > }
> > }
> >
> > diff --git a/sysdeps/mach/hurd/fcntl.c b/sysdeps/mach/hurd/fcntl.c
> > index a65c190c..56d850e8 100644
> > --- a/sysdeps/mach/hurd/fcntl.c
> > +++ b/sysdeps/mach/hurd/fcntl.c
> > @@ -83,18 +83,46 @@ __libc_fcntl (int fd, int cmd, ...)
> > result = -1;
> > else
> > {
> > - /* Give the ports each a user ref for the new descriptor. */
> > - __mach_port_mod_refs (__mach_task_self (), port,
> > - MACH_PORT_RIGHT_SEND, 1);
> > - if (ctty != MACH_PORT_NULL)
> > - __mach_port_mod_refs (__mach_task_self (), ctty,
> > - MACH_PORT_RIGHT_SEND, 1);
> > -
> > - /* Install the ports and flags in the new descriptor. */
> > - if (ctty != MACH_PORT_NULL)
> > - _hurd_port_set (&new->ctty, ctty);
> > - new->flags = flags;
> > - _hurd_port_locked_set (&new->port, port); /* Unlocks NEW. */
> > + /* Give the io server port a user ref for the new descriptor. */
> > + err = __mach_port_mod_refs (__mach_task_self (), port,
> > + MACH_PORT_RIGHT_SEND, 1);
> > +
> > + if (err == KERN_UREFS_OVERFLOW)
> > + result = __hurd_fail (EMFILE);
> > + else if (err)
> > + result = __hurd_fail (EINVAL);
> > + else if (ctty != MACH_PORT_NULL)
> > + {
> > + /* We have confirmed the io server port has got a user ref
> > + count, now give ctty port a user ref for the new
> > + descriptor. */
> > + err = __mach_port_mod_refs (__mach_task_self (), ctty,
> > + MACH_PORT_RIGHT_SEND, 1);
> > +
> > + if (err)
> > + {
> > + /* In this case the io server port has got a ref count
> > + but the ctty port fails to get one, so we need to clean
> > + the ref count we just assigned. */
> > + __mach_port_mod_refs (__mach_task_self (), port,
> > + MACH_PORT_RIGHT_SEND, -1);
> > + if (err == KERN_UREFS_OVERFLOW)
> > + result = __hurd_fail (EMFILE);
> > + else
> > + result = __hurd_fail (EINVAL);
> > + }
> > + }
> > +
> > + if (!err)
> > + {
> > + /* The ref counts of the ports are incremented successfully. */
> > + /* Install the ports and flags in the new descriptor. */
> > + if (ctty != MACH_PORT_NULL)
> > + _hurd_port_set (&new->ctty, ctty);
> > + new->flags = flags;
> > + /* Unlocks NEW. */
> > + _hurd_port_locked_set (&new->port, port);
> > + }
> > }
> >
> > HURD_CRITICAL_END;
> > --
> > 2.47.2
> >
>
> --
> Samuel
> <N> bon comment on fait de l'investigation pour savoir qui est le vilain ?
> <s> on débranche le routeur et on regarde qui s'affole
> -+- #ens-mim administre -+-
--
Samuel
How do I type "for i in *.dvi do xdvi i done" in a GUI?
(Discussion in comp.os.linux.misc on the intuitiveness of interfaces.)
