blktrace: unblock tracers in the case of unsuccess

Hello.
Please consider/apply
Thanks

Edward O. Shishkin
Principal Software Engineer
Red Hat Czech

Fixup for bz 595356
Problem. Steps to Reproduce:
---

# blktrace -d /dev/sda -w 30
# rm -f sda.blktrace.0
# mkdir sda.blktrace.0
# blktrace -d /dev/sda -w 30
Open output file ./sda.blktrace.0 failed: 21/Is a directory
FAILED to start thread on CPU 0: 1/Operation not permitted

blktrace does not end after 30 seconds, it will be running until you kill it. Moreover, next attempts to run blktrace failed with error.

The bug:
---
If open_ios() fails, then tracer_wait_unblock() in the thread_main() waits for event, wich will never occur:
in the case of unsuccess tracers are not unblocked.

Fixup:
---
Make sure unblock_tracers() is also called in the case on unsuccess (when nthreads_running != ncpus).
---

---

@@ -1848,7 +1848,7 @@
 
 	for (cpu = 0; cpu < ncpus; cpu++)
 		if (start_tracer(cpu))
-			break;
+			return;
 
 	wait_tracers_ready(cpu);
 

@@ -2589,8 +2589,8 @@
 	}
 
 	start_tracers();
-	if (nthreads_running == ncpus) {
-		unblock_tracers();
+	unblock_tracers();
+ 	if (nthreads_running == ncpus) {
 		start_buts();
 		if (net_mode == Net_client)
 			printf("blktrace: connected!\n");