task IO accounting: provide distinct tgid/tid I/O statistics

Report per-thread I/O statistics in /proc/pid/task/tid/io and aggregate parent I/O statistics in /proc/pid/io. This approach follows the same model used to account per-process and per-thread CPU times. As a practial application, this allows for example to quickly find the top I/O consumer when a process spawns many child threads that perform the actual I/O work, because the aggregated I/O statistics can always be found in /proc/pid/io. [ Oleg Nesterov points out that we should check that the task is still alive before we iterate over the threads, but also says that we can do that fixup on top of this later. - Linus ] Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by: Andrea Righi <righi.andrea@gmail.com> Cc: Matt Heaton <matt@hostmonster.com> Cc: Shailabh Nagar <nagar@watson.ibm.com> Acked-by-with-comments: Oleg Nesterov <oleg@tv-sign.ru> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

task IO accounting: provide distinct tgid/tid I/O statistics
Report per-thread I/O statistics in /proc/pid/task/tid/io and aggregate parent I/O statistics in /proc/pid/io. This approach follows the same model used to account per-process and per-thread CPU times. As a practial application, this allows for example to quickly find the top I/O consumer when a process spawns many child threads that perform the actual I/O work, because the aggregated I/O statistics can always be found in /proc/pid/io. [ Oleg Nesterov points out that we should check that the task is still alive before we iterate over the threads, but also says that we can do that fixup on top of this later. - Linus ] Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by: Andrea Righi <righi.andrea@gmail.com> Cc: Matt Heaton <matt@hostmonster.com> Cc: Shailabh Nagar <nagar@watson.ibm.com> Acked-by-with-comments: Oleg Nesterov <oleg@tv-sign.ru> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Andrea Righi · Linus Torvalds
1 parent 0c18d7a5df
Showing 4 changed files with 108 additions and 15 deletions Side-by-side Diff
fs/proc/base.c
include/linux/sched.h
kernel/exit.c
kernel/fork.c
@@ -2376,30 +2376,83 @@
 }
  
 #ifdef CONFIG_TASK_IO_ACCOUNTING
-static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
+static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
 {
+	u64 rchar, wchar, syscr, syscw;
+	struct task_io_accounting ioac;
+
+	if (!whole) {
+		rchar = task->rchar;
+		wchar = task->wchar;
+		syscr = task->syscr;
+		syscw = task->syscw;
+		memcpy(&ioac, &task->ioac, sizeof(ioac));
+	} else {
+		unsigned long flags;
+		struct task_struct *t = task;
+		rchar = wchar = syscr = syscw = 0;
+		memset(&ioac, 0, sizeof(ioac));
+
+		rcu_read_lock();
+		do {
+			rchar += t->rchar;
+			wchar += t->wchar;
+			syscr += t->syscr;
+			syscw += t->syscw;
+
+			ioac.read_bytes += t->ioac.read_bytes;
+			ioac.write_bytes += t->ioac.write_bytes;
+			ioac.cancelled_write_bytes +=
+					t->ioac.cancelled_write_bytes;
+			t = next_thread(t);
+		} while (t != task);
+		rcu_read_unlock();
+
+		if (lock_task_sighand(task, &flags)) {
+			struct signal_struct *sig = task->signal;
+
+			rchar += sig->rchar;
+			wchar += sig->wchar;
+			syscr += sig->syscr;
+			syscw += sig->syscw;
+
+			ioac.read_bytes += sig->ioac.read_bytes;
+			ioac.write_bytes += sig->ioac.write_bytes;
+			ioac.cancelled_write_bytes +=
+					sig->ioac.cancelled_write_bytes;
+
+			unlock_task_sighand(task, &flags);
+		}
+	}
+
 	return sprintf(buffer,
-#ifdef CONFIG_TASK_XACCT
 			"rchar: %llu\n"
 			"wchar: %llu\n"
 			"syscr: %llu\n"
 			"syscw: %llu\n"
-#endif
 			"read_bytes: %llu\n"
 			"write_bytes: %llu\n"
 			"cancelled_write_bytes: %llu\n",
-#ifdef CONFIG_TASK_XACCT
-			(unsigned long long)task->rchar,
-			(unsigned long long)task->wchar,
-			(unsigned long long)task->syscr,
-			(unsigned long long)task->syscw,
-#endif
-			(unsigned long long)task->ioac.read_bytes,
-			(unsigned long long)task->ioac.write_bytes,
-			(unsigned long long)task->ioac.cancelled_write_bytes);
+			(unsigned long long)rchar,
+			(unsigned long long)wchar,
+			(unsigned long long)syscr,
+			(unsigned long long)syscw,
+			(unsigned long long)ioac.read_bytes,
+			(unsigned long long)ioac.write_bytes,
+			(unsigned long long)ioac.cancelled_write_bytes);
 }
-#endif
  
+static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
+{
+	return do_io_accounting(task, buffer, 0);
+}
+
+static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
+{
+	return do_io_accounting(task, buffer, 1);
+}
+#endif /* CONFIG_TASK_IO_ACCOUNTING */
+
 /*
  * Thread groups
  */
@@ -2470,7 +2523,7 @@
 	REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
 #endif
 #ifdef CONFIG_TASK_IO_ACCOUNTING
-	INF("io",	S_IRUGO, pid_io_accounting),
+	INF("io",	S_IRUGO, tgid_io_accounting),
 #endif
 };
  
@@ -2796,6 +2849,9 @@
 #endif
 #ifdef CONFIG_FAULT_INJECTION
 	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
+#endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+	INF("io",	S_IRUGO, tid_io_accounting),
 #endif
 };
  
@@ -506,6 +506,10 @@
 	unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw;
 	unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
 	unsigned long inblock, oublock, cinblock, coublock;
+#ifdef CONFIG_TASK_XACCT
+	u64 rchar, wchar, syscr, syscw;
+#endif
+	struct task_io_accounting ioac;
  
 	/*
 	 * Cumulative ns of scheduled CPU time for dead threads in the
@@ -120,6 +120,18 @@
 		sig->nivcsw += tsk->nivcsw;
 		sig->inblock += task_io_get_inblock(tsk);
 		sig->oublock += task_io_get_oublock(tsk);
+#ifdef CONFIG_TASK_XACCT
+		sig->rchar += tsk->rchar;
+		sig->wchar += tsk->wchar;
+		sig->syscr += tsk->syscr;
+		sig->syscw += tsk->syscw;
+#endif /* CONFIG_TASK_XACCT */
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+		sig->ioac.read_bytes += tsk->ioac.read_bytes;
+		sig->ioac.write_bytes += tsk->ioac.write_bytes;
+		sig->ioac.cancelled_write_bytes +=
+					tsk->ioac.cancelled_write_bytes;
+#endif /* CONFIG_TASK_IO_ACCOUNTING */
 		sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 		sig = NULL; /* Marker for below. */
 	}
@@ -1366,6 +1378,21 @@
 		psig->coublock +=
 			task_io_get_oublock(p) +
 			sig->oublock + sig->coublock;
+#ifdef CONFIG_TASK_XACCT
+		psig->rchar += p->rchar + sig->rchar;
+		psig->wchar += p->wchar + sig->wchar;
+		psig->syscr += p->syscr + sig->syscr;
+		psig->syscw += p->syscw + sig->syscw;
+#endif /* CONFIG_TASK_XACCT */
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+		psig->ioac.read_bytes +=
+			p->ioac.read_bytes + sig->ioac.read_bytes;
+		psig->ioac.write_bytes +=
+			p->ioac.write_bytes + sig->ioac.write_bytes;
+		psig->ioac.cancelled_write_bytes +=
+				p->ioac.cancelled_write_bytes +
+				sig->ioac.cancelled_write_bytes;
+#endif /* CONFIG_TASK_IO_ACCOUNTING */
 		spin_unlock_irq(&p->parent->sighand->siglock);
 	}
  
@@ -812,6 +812,12 @@
 	sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 	sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 	sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
+#ifdef CONFIG_TASK_XACCT
+	sig->rchar = sig->wchar = sig->syscr = sig->syscw = 0;
+#endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+	memset(&sig->ioac, 0, sizeof(sig->ioac));
+#endif
 	sig->sum_sched_runtime = 0;
 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
...	...	@@ -2376,30 +2376,83 @@
2376	2376	}
2377	2377
2378	2378	#ifdef CONFIG_TASK_IO_ACCOUNTING
2379		-static int proc_pid_io_accounting(struct task_struct task, char buffer)
	2379	+static int do_io_accounting(struct task_struct task, char buffer, int whole)
2380	2380	{
	2381	+ u64 rchar, wchar, syscr, syscw;
	2382	+ struct task_io_accounting ioac;
	2383	+
	2384	+ if (!whole) {
	2385	+ rchar = task->rchar;
	2386	+ wchar = task->wchar;
	2387	+ syscr = task->syscr;
	2388	+ syscw = task->syscw;
	2389	+ memcpy(&ioac, &task->ioac, sizeof(ioac));
	2390	+ } else {
	2391	+ unsigned long flags;
	2392	+ struct task_struct *t = task;
	2393	+ rchar = wchar = syscr = syscw = 0;
	2394	+ memset(&ioac, 0, sizeof(ioac));
	2395	+
	2396	+ rcu_read_lock();
	2397	+ do {
	2398	+ rchar += t->rchar;
	2399	+ wchar += t->wchar;
	2400	+ syscr += t->syscr;
	2401	+ syscw += t->syscw;
	2402	+
	2403	+ ioac.read_bytes += t->ioac.read_bytes;
	2404	+ ioac.write_bytes += t->ioac.write_bytes;
	2405	+ ioac.cancelled_write_bytes +=
	2406	+ t->ioac.cancelled_write_bytes;
	2407	+ t = next_thread(t);
	2408	+ } while (t != task);
	2409	+ rcu_read_unlock();
	2410	+
	2411	+ if (lock_task_sighand(task, &flags)) {
	2412	+ struct signal_struct *sig = task->signal;
	2413	+
	2414	+ rchar += sig->rchar;
	2415	+ wchar += sig->wchar;
	2416	+ syscr += sig->syscr;
	2417	+ syscw += sig->syscw;
	2418	+
	2419	+ ioac.read_bytes += sig->ioac.read_bytes;
	2420	+ ioac.write_bytes += sig->ioac.write_bytes;
	2421	+ ioac.cancelled_write_bytes +=
	2422	+ sig->ioac.cancelled_write_bytes;
	2423	+
	2424	+ unlock_task_sighand(task, &flags);
	2425	+ }
	2426	+ }
	2427	+
2381	2428	return sprintf(buffer,
2382		-#ifdef CONFIG_TASK_XACCT
2383	2429	"rchar: %llu\n"
2384	2430	"wchar: %llu\n"
2385	2431	"syscr: %llu\n"
2386	2432	"syscw: %llu\n"
2387		-#endif
2388	2433	"read_bytes: %llu\n"
2389	2434	"write_bytes: %llu\n"
2390	2435	"cancelled_write_bytes: %llu\n",
2391		-#ifdef CONFIG_TASK_XACCT
2392		- (unsigned long long)task->rchar,
2393		- (unsigned long long)task->wchar,
2394		- (unsigned long long)task->syscr,
2395		- (unsigned long long)task->syscw,
2396		-#endif
2397		- (unsigned long long)task->ioac.read_bytes,
2398		- (unsigned long long)task->ioac.write_bytes,
2399		- (unsigned long long)task->ioac.cancelled_write_bytes);
	2436	+ (unsigned long long)rchar,
	2437	+ (unsigned long long)wchar,
	2438	+ (unsigned long long)syscr,
	2439	+ (unsigned long long)syscw,
	2440	+ (unsigned long long)ioac.read_bytes,
	2441	+ (unsigned long long)ioac.write_bytes,
	2442	+ (unsigned long long)ioac.cancelled_write_bytes);
2400	2443	}
2401		-#endif
2402	2444
	2445	+static int proc_tid_io_accounting(struct task_struct task, char buffer)
	2446	+{
	2447	+ return do_io_accounting(task, buffer, 0);
	2448	+}
	2449	+
	2450	+static int proc_tgid_io_accounting(struct task_struct task, char buffer)
	2451	+{
	2452	+ return do_io_accounting(task, buffer, 1);
	2453	+}
	2454	+#endif /* CONFIG_TASK_IO_ACCOUNTING */
	2455	+
2403	2456	/*
2404	2457	* Thread groups
2405	2458	*/
...	...	@@ -2470,7 +2523,7 @@
2470	2523	REG("coredump_filter", S_IRUGO\|S_IWUSR, coredump_filter),
2471	2524	#endif
2472	2525	#ifdef CONFIG_TASK_IO_ACCOUNTING
2473		- INF("io", S_IRUGO, pid_io_accounting),
	2526	+ INF("io", S_IRUGO, tgid_io_accounting),
2474	2527	#endif
2475	2528	};
2476	2529
...	...	@@ -2796,6 +2849,9 @@
2796	2849	#endif
2797	2850	#ifdef CONFIG_FAULT_INJECTION
2798	2851	REG("make-it-fail", S_IRUGO\|S_IWUSR, fault_inject),
	2852	+#endif
	2853	+#ifdef CONFIG_TASK_IO_ACCOUNTING
	2854	+ INF("io", S_IRUGO, tid_io_accounting),
2799	2855	#endif
2800	2856	};
2801	2857
...	...	@@ -506,6 +506,10 @@
506	506	unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw;
507	507	unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
508	508	unsigned long inblock, oublock, cinblock, coublock;
	509	+#ifdef CONFIG_TASK_XACCT
	510	+ u64 rchar, wchar, syscr, syscw;
	511	+#endif
	512	+ struct task_io_accounting ioac;
509	513
510	514	/*
511	515	* Cumulative ns of scheduled CPU time for dead threads in the
...	...	@@ -120,6 +120,18 @@
120	120	sig->nivcsw += tsk->nivcsw;
121	121	sig->inblock += task_io_get_inblock(tsk);
122	122	sig->oublock += task_io_get_oublock(tsk);
	123	+#ifdef CONFIG_TASK_XACCT
	124	+ sig->rchar += tsk->rchar;
	125	+ sig->wchar += tsk->wchar;
	126	+ sig->syscr += tsk->syscr;
	127	+ sig->syscw += tsk->syscw;
	128	+#endif /* CONFIG_TASK_XACCT */
	129	+#ifdef CONFIG_TASK_IO_ACCOUNTING
	130	+ sig->ioac.read_bytes += tsk->ioac.read_bytes;
	131	+ sig->ioac.write_bytes += tsk->ioac.write_bytes;
	132	+ sig->ioac.cancelled_write_bytes +=
	133	+ tsk->ioac.cancelled_write_bytes;
	134	+#endif /* CONFIG_TASK_IO_ACCOUNTING */
123	135	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
124	136	sig = NULL; /* Marker for below. */
125	137	}
...	...	@@ -1366,6 +1378,21 @@
1366	1378	psig->coublock +=
1367	1379	task_io_get_oublock(p) +
1368	1380	sig->oublock + sig->coublock;
	1381	+#ifdef CONFIG_TASK_XACCT
	1382	+ psig->rchar += p->rchar + sig->rchar;
	1383	+ psig->wchar += p->wchar + sig->wchar;
	1384	+ psig->syscr += p->syscr + sig->syscr;
	1385	+ psig->syscw += p->syscw + sig->syscw;
	1386	+#endif /* CONFIG_TASK_XACCT */
	1387	+#ifdef CONFIG_TASK_IO_ACCOUNTING
	1388	+ psig->ioac.read_bytes +=
	1389	+ p->ioac.read_bytes + sig->ioac.read_bytes;
	1390	+ psig->ioac.write_bytes +=
	1391	+ p->ioac.write_bytes + sig->ioac.write_bytes;
	1392	+ psig->ioac.cancelled_write_bytes +=
	1393	+ p->ioac.cancelled_write_bytes +
	1394	+ sig->ioac.cancelled_write_bytes;
	1395	+#endif /* CONFIG_TASK_IO_ACCOUNTING */
1369	1396	spin_unlock_irq(&p->parent->sighand->siglock);
1370	1397	}
1371	1398
...	...	@@ -812,6 +812,12 @@
812	812	sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
813	813	sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
814	814	sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
	815	+#ifdef CONFIG_TASK_XACCT
	816	+ sig->rchar = sig->wchar = sig->syscr = sig->syscw = 0;
	817	+#endif
	818	+#ifdef CONFIG_TASK_IO_ACCOUNTING
	819	+ memset(&sig->ioac, 0, sizeof(sig->ioac));
	820	+#endif
815	821	sig->sum_sched_runtime = 0;
816	822	INIT_LIST_HEAD(&sig->cpu_timers[0]);
817	823	INIT_LIST_HEAD(&sig->cpu_timers[1]);