[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <478CE41F.5010401@users.sourceforge.net>
Date: Tue, 15 Jan 2008 17:49:36 +0100 (MET)
From: Andrea Righi <righiandr@...rs.sourceforge.net>
To: Balbir Singh <balbir@...ux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@...llo.nl>, Valdis.Kletnieks@...edu,
LKML <linux-kernel@...r.kernel.org>
Subject: [RFC][PATCH] per-uid/gid I/O throttling (was Re: [RFC][PATCH] per-task
I/O throttling)
Allow to limit the I/O bandwidth for specific uid(s) or gid(s) imposing
additional delays on those processes that exceed the limits defined in a
configfs tree.
Examples:
Limit the I/O bandwidth for user www-data (UID 33) to 4MB/s:
root@...ux:/config/io-throttle# mkdir uid:33
root@...ux:/config/io-throttle# cd uid:33/
root@...ux:/config/io-throttle/uid:33# cat io-rate
io-rate: 0 KiB/sec
requested: 0 KiB
last_request: 0 jiffies
delta: 388202 jiffies
root@...ux:/config/io-throttle/uid:33# echo 4096 > io-rate
root@...ux:/config/io-throttle/uid:33# cat io-rate
io-rate: 4096 KiB/sec
requested: 0 KiB
last_request: 389271 jiffies
delta: 91 jiffies
Limit the I/O bandwidth of group backup (GID 34) to 512KB/s:
root@...ux:/config/io-throttle# mkdir gid:34
root@...ux:/config/io-throttle# cd gid:34/
root@...ux:/config/io-throttle/gid:34# cat io-rate
io-rate: 0 KiB/sec
requested: 0 KiB
last_request: 0 jiffies
delta: 403160 jiffies
root@...ux:/config/io-throttle/gid:34# echo 512 > io-rate
root@...ux:/config/io-throttle/gid:34# cat io-rate
io-rate: 512 KiB/sec
requested: 0 KiB
last_request: 403618 jiffies
delta: 80 jiffies
Remove the I/O limit for user www-data:
root@...ux:/config/io-throttle# echo 0 > uid:33/io-rate
root@...ux:/config/io-throttle# cat uid:33/io-rate
io-rate: 0 KiB/sec
requested: 0 KiB
last_request: 419009 jiffies
delta: 568 jiffies
or:
root@...ux:/config/io-throttle# rmdir uid:33
Future improvements:
* allow to limit also I/O operations per second (instead of KB/s only)
* extend grouping criteria (allow to define rules based on process containers,
process command, etc.)
Signed-off-by: Andrea Righi <a.righi@...eca.it>
---
diff -urpN linux-2.6.24-rc7/block/io-throttle.c linux-2.6.24-rc7-io-throttle/block/io-throttle.c
--- linux-2.6.24-rc7/block/io-throttle.c 1970-01-01 01:00:00.000000000 +0100
+++ linux-2.6.24-rc7-io-throttle/block/io-throttle.c 2008-01-15 17:25:06.000000000 +0100
@@ -0,0 +1,282 @@
+/*
+ * io-throttle.c
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ * Copyright (C) 2008 Andrea Righi <righiandr@...rs.sourceforge.net>
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/configfs.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
+#include <linux/jiffies.h>
+#include <linux/io-throttle.h>
+
+/* This should work for any reasonable size of uid_t */
+#define MAXUIDCHAR (sizeof(uid_t) * 5 / 2)
+
+/*
+ * Basic structure that identifies a single I/O throttling rule
+ */
+struct iothrottle {
+ struct config_item item;
+ unsigned long iorate;
+ unsigned long req;
+ unsigned long last_request;
+};
+
+/* Get an I/O-throttling item from a configfs item */
+static inline struct iothrottle *to_iothrottle(struct config_item *item)
+{
+ return item ? container_of(item, struct iothrottle, item) : NULL;
+}
+
+static ssize_t iothrottle_attr_show(struct config_item *item,
+ struct configfs_attribute *attr,
+ char *page);
+static ssize_t iothrottle_attr_store(struct config_item *item,
+ struct configfs_attribute *attr,
+ const char *page, size_t count);
+static struct config_item *iothrottle_make_item(struct config_group *group,
+ const char *name);
+static void iothrottle_release(struct config_item *item);
+
+/* I/O throttling item in configfs (identify a single I/O throttling rule) */
+static struct configfs_attribute iothrottle_attr_iorate = {
+ .ca_owner = THIS_MODULE,
+ .ca_name = "io-rate",
+ .ca_mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH,
+};
+
+/* I/O throttling element in configfs */
+static struct configfs_group_operations iothrottle_group_ops = {
+ .make_item = iothrottle_make_item,
+};
+
+static struct config_item_type iothrottle_group_type = {
+ .ct_group_ops = &iothrottle_group_ops,
+};
+
+/* Entire I/O throttling subsystem under configfs (/config/io-throttle) */
+struct configfs_subsystem iothrottle_subsys = {
+ .su_group = {
+ .cg_item = {
+ .ci_namebuf = "io-throttle",
+ .ci_type = &iothrottle_group_type,
+ },
+ },
+};
+
+/* List of configfs elements per group */
+static struct configfs_attribute *iothrottle_attrs[] = {
+ &iothrottle_attr_iorate,
+ NULL,
+};
+
+static struct configfs_item_operations iothrottle_item_ops = {
+ .release = iothrottle_release,
+ .show_attribute = iothrottle_attr_show,
+ .store_attribute = iothrottle_attr_store,
+};
+
+static struct config_item_type iothrottle_type = {
+ .ct_item_ops = &iothrottle_item_ops,
+ .ct_attrs = iothrottle_attrs,
+ .ct_owner = THIS_MODULE,
+};
+
+/* Print a I/O throttling rule details */
+static ssize_t iothrottle_attr_show(struct config_item *item,
+ struct configfs_attribute *attr,
+ char *page)
+{
+ ssize_t count;
+ unsigned long delta;
+ struct iothrottle *iot = to_iothrottle(item);
+
+ delta = (long)jiffies - (long)iot->last_request;
+ /* Print additional debugging stuff */
+ count = sprintf(page, " io-rate: %lu KiB/sec\n"
+ " requested: %lu KiB\n"
+ "last_request: %lu jiffies\n"
+ " delta: %lu jiffies\n",
+ iot->iorate, iot->req << 1, iot->last_request, delta);
+
+ return count;
+}
+
+/* Configure the attributes of an I/O throttling rule */
+static ssize_t iothrottle_attr_store(struct config_item *item,
+ struct configfs_attribute *attr,
+ const char *page, size_t count)
+{
+ struct iothrottle *iot = to_iothrottle(item);
+ unsigned long tmp;
+ char *p = (char *) page;
+
+ tmp = simple_strtoul(p, &p, 10);
+ if (!p || (*p && (*p != '\n')))
+ return -EINVAL;
+
+ iot->iorate = tmp;
+ iot->req = 0;
+ iot->last_request = jiffies;
+
+ return count;
+}
+
+/* Register a new I/O throttling rule */
+static struct config_item *iothrottle_make_item(struct config_group *group,
+ const char *name)
+{
+ struct iothrottle *iot;
+
+ iot = kzalloc(sizeof(*iot), GFP_KERNEL);
+ if (unlikely(!iot))
+ return NULL;
+
+ iot->last_request = jiffies;
+
+ config_item_init_type_name(&iot->item, name, &iothrottle_type);
+
+ return &iot->item;
+}
+
+/* Unregister an I/O throttling rule */
+static void iothrottle_release(struct config_item *item)
+{
+ kfree(to_iothrottle(item));
+}
+
+/* Get the opportune "struct iothrottle" item from configfs if present */
+static inline struct iothrottle *iothrottle_get_config(void)
+{
+ char idstr[MAXUIDCHAR + 1];
+ struct iothrottle *p = NULL;
+
+ memset(idstr, 0, sizeof(idstr));
+
+ /* UID-based rule */
+ snprintf(idstr, MAXUIDCHAR, "uid:%u", current->uid);
+ p = to_iothrottle(
+ config_group_find_item(&iothrottle_subsys.su_group, idstr));
+ if (p)
+ goto out_get_config;
+
+ /* GID-based rule */
+ snprintf(idstr, MAXUIDCHAR, "gid:%u", current->gid);
+ p = to_iothrottle(
+ config_group_find_item(&iothrottle_subsys.su_group, idstr));
+
+out_get_config:
+ return p;
+}
+
+/*
+ * Here is the main function of the I/O throttling mechanism.
+ *
+ * FIXME: potential race if the struct iothrottle *iot item is removed/modified
+ * in the middle of the io_throttle() execution.
+ */
+void io_throttle(int nr_sectors)
+{
+ struct iothrottle *iot;
+ unsigned long delta, n;
+ long sleep;
+
+ iot = iothrottle_get_config();
+ if (!iot || !iot->iorate)
+ return;
+
+ /*
+ * The concept is the following: evaluate the actual I/O rate of a
+ * process, looking at the sectors requested over the time elapsed from
+ * the last request. If the actual I/O rate is beyond the maximum
+ * allowed I/O rate then sleep the current task for the correct amount
+ * of time, in order to reduce the actual I/O rate under the allowed
+ * limit.
+ *
+ * The time to sleep is evaluated as:
+ *
+ * sleep = (sectors_requested / allowed_iorate) - time_elapsed
+ */
+ delta = (long)jiffies - (long)iot->last_request;
+ n = iot->req += nr_sectors;
+ do_div(n, iot->iorate);
+
+ /*
+ * Unable to evaluate delta (due to a too small interval of time
+ * between two requests) or n (due to a too small request).
+ *
+ * Account the requested sectors in iot->req and sum them to the
+ * sectors of the next request.
+ */
+ if (!delta || !n)
+ return;
+
+ /*
+ * Convert n in jiffies (remember that iot->iorate is in KB/s and we
+ * need to convert it in sectors/jiffies)
+ */
+ sleep = msecs_to_jiffies(n * 1000 / 2) - delta;
+ if (sleep > 0) {
+ printk(KERN_DEBUG
+ "io-throttle: process %i (%s) must sleep %lu jiffies\n",
+ task_pid_nr(current), current->comm, sleep);
+ schedule_timeout_uninterruptible(sleep);
+ }
+
+ /* OK, we stay under the limits. Reset statistics. */
+ iot->req = 0;
+ iot->last_request = jiffies;
+}
+EXPORT_SYMBOL(io_throttle);
+
+/* Register I/O throttling subsystem */
+static int __init iothrottle_init(void)
+{
+ int ret;
+
+ config_group_init(&iothrottle_subsys.su_group);
+ mutex_init(&iothrottle_subsys.su_mutex);
+ ret = configfs_register_subsystem(&iothrottle_subsys);
+ if (ret) {
+ printk(KERN_ERR "%s: error %d while registering subsystem\n",
+ iothrottle_subsys.su_group.cg_item.ci_namebuf,
+ ret);
+ goto out_unregister;
+ }
+
+ return 0;
+
+out_unregister:
+ configfs_unregister_subsystem(&iothrottle_subsys);
+
+ return ret;
+}
+
+/* Unregister I/O throttling subsystem */
+static void __exit iothrottle_exit(void)
+{
+ configfs_unregister_subsystem(&iothrottle_subsys);
+ mutex_destroy(&iothrottle_subsys.su_mutex);
+}
+
+module_init(iothrottle_init);
+module_exit(iothrottle_exit);
+MODULE_LICENSE("GPL");
diff -urpN linux-2.6.24-rc7/block/Kconfig linux-2.6.24-rc7-io-throttle/block/Kconfig
--- linux-2.6.24-rc7/block/Kconfig 2008-01-06 22:45:38.000000000 +0100
+++ linux-2.6.24-rc7-io-throttle/block/Kconfig 2008-01-15 15:30:21.000000000 +0100
@@ -40,6 +40,16 @@ config BLK_DEV_IO_TRACE
git://brick.kernel.dk/data/git/blktrace.git
+config IO_THROTTLE
+ tristate "Enable I/O throttling (EXPERIMENTAL)"
+ depends on EXPERIMENTAL
+ select CONFIGFS_FS
+ help
+ This allows to limit the maximum I/O rate for specific UID(s) or
+ GID(s).
+
+ Say N if unsure.
+
config LSF
bool "Support for Large Single Files"
depends on !64BIT
diff -urpN linux-2.6.24-rc7/block/ll_rw_blk.c linux-2.6.24-rc7-io-throttle/block/ll_rw_blk.c
--- linux-2.6.24-rc7/block/ll_rw_blk.c 2008-01-06 22:45:38.000000000 +0100
+++ linux-2.6.24-rc7-io-throttle/block/ll_rw_blk.c 2008-01-15 13:59:21.000000000 +0100
@@ -31,6 +31,7 @@
#include <linux/blktrace_api.h>
#include <linux/fault-inject.h>
#include <linux/scatterlist.h>
+#include <linux/io-throttle.h>
/*
* for max sense size
@@ -3221,6 +3222,8 @@ static inline void __generic_make_reques
if (bio_check_eod(bio, nr_sectors))
goto end_io;
+ io_throttle(nr_sectors);
+
/*
* Resolve the mapping until finished. (drivers are
* still free to implement/resolve their own stacking
diff -urpN linux-2.6.24-rc7/block/Makefile linux-2.6.24-rc7-io-throttle/block/Makefile
--- linux-2.6.24-rc7/block/Makefile 2008-01-06 22:45:38.000000000 +0100
+++ linux-2.6.24-rc7-io-throttle/block/Makefile 2008-01-15 13:59:21.000000000 +0100
@@ -12,3 +12,5 @@ obj-$(CONFIG_IOSCHED_CFQ) += cfq-iosched
obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o
obj-$(CONFIG_BLOCK_COMPAT) += compat_ioctl.o
+
+obj-$(CONFIG_IO_THROTTLE) += io-throttle.o
diff -urpN linux-2.6.24-rc7/include/linux/io-throttle.h linux-2.6.24-rc7-io-throttle/include/linux/io-throttle.h
--- linux-2.6.24-rc7/include/linux/io-throttle.h 1970-01-01 01:00:00.000000000 +0100
+++ linux-2.6.24-rc7-io-throttle/include/linux/io-throttle.h 2008-01-15 13:59:21.000000000 +0100
@@ -0,0 +1,10 @@
+#ifndef IO_THROTTLE_H
+#define IO_THROTTLE_H
+
+#ifdef CONFIG_IO_THROTTLE
+extern void io_throttle(int nr_sectors);
+#else
+static inline void io_throttle(int nr_sectors) { }
+#endif /* CONFIG_IO_THROTTLE */
+
+#endif
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists