Messages in this thread | | | Date | Mon, 5 Jan 2004 06:56:33 -0700 | From | Erik Andersen <> | Subject | Re: [PATCH] fix broken 2.4.x rt_sigprocmask error handling |
| |
On Mon Jan 05, 2004 at 11:23:02AM -0200, Marcelo Tosatti wrote: > Hi Erik, > > For how long the behaviour has been this? It is broken, but its brokenness > is harmless.
Not sure really. It is correct in 2.2.x, and 2.6.x, but is wrong in all 2.4.x kernels starting with 2.4.0, but I havn't bothered finding exactly which kernel broke things.
> Maybe some app rely on this behaviour? (I understand its wrong, but > still). > > Is there any major distribution which includes this fix?
I highly doubt anything relies on this. Glibc checks for this case before the syscall is made. I found the bug while running some the ltp testsuite on uClibc. I have since added a check to workaround the bug. but I would prefer to have it fixed properly in the kernel.
> Anyway, isnt it easier to move the "if (error)" up like this? > > --- signal.c.orig 2004-01-05 11:13:17.000000000 -0200 > +++ signal.c 2004-01-05 11:14:09.000000000 -0200 > @@ -888,11 +888,12 @@ > break; > } > > + if (error) > + goto out; > + > current->blocked = new_set; > recalc_sigpending(current); > spin_unlock_irq(¤t->sigmask_lock); > - if (error) > - goto out; > if (oset) > goto set_old; > } else if (oset) {
Only if you plan to ignore the locked spinlock. :-) The 2.2.x code does this:
-current->blocked = new_set; +if (!error) current->blocked = new_set;
I condiered that, but I thought it better to make the code more closely match 2.6, which looked a bit nicer.
-Erik
-- Erik B. Andersen http://codepoet-consulting.com/ --This message was written using 73% post-consumer electrons-- - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |