Apply the io-throttle controller to the opportune kernel functions. Both
accounting and throttling functionalities are performed by
cgroup_io_throttle().

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 block/blk-core.c      |    9 +++++++++
 fs/aio.c              |   31 ++++++++++++++++++++++++++++++-
 fs/buffer.c           |   20 +++++++++++++++++---
 fs/direct-io.c        |    4 ++++
 include/linux/sched.h |    3 +++
 kernel/fork.c         |    3 +++
 mm/filemap.c          |   18 +++++++++++++++++-
 mm/page-writeback.c   |   30 +++++++++++++++++++++++++++---
 mm/readahead.c        |    5 +++++
 9 files changed, 115 insertions(+), 8 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 4c222ba..431294f 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -26,6 +26,7 @@
 #include <linux/swap.h>
 #include <linux/writeback.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/interrupt.h>
 #include <linux/cpu.h>
 #include <linux/blktrace_api.h>
@@ -1482,7 +1483,15 @@ void submit_bio(int rw, struct bio *bio)
 		if (rw & WRITE) {
 			count_vm_events(PGPGOUT, count);
 		} else {
+			struct page *page = bio_iovec_idx(bio, 0)->bv_page;
+
 			task_io_account_read(bio->bi_size);
+			/*
+			 * Do not throttle page requests that need to be
+			 * urgently reclaimed.
+			 */
+			cgroup_io_throttle(bio->bi_bdev, bio->bi_size,
+				!(PageReclaim(page) || PageSwapCache(page)));
 			count_vm_events(PGPGIN, count);
 		}
 
diff --git a/fs/aio.c b/fs/aio.c
index 0051fd9..1f3abb3 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -22,6 +22,7 @@
 #include <linux/sched.h>
 #include <linux/fs.h>
 #include <linux/file.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/mm.h>
 #include <linux/mman.h>
 #include <linux/slab.h>
@@ -1558,6 +1559,8 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 {
 	struct kiocb *req;
 	struct file *file;
+	struct block_device *bdev;
+	struct inode *inode;
 	ssize_t ret;
 
 	/* enforce forwards compatibility on users */
@@ -1580,10 +1583,26 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 	if (unlikely(!file))
 		return -EBADF;
 
+	/*
+	 * Pre-account AIO activity: we over-account *all* the bytes here;
+	 * bytes read from the page cache and bytes written in already dirtied
+	 * pages (that do not generate real i/o on block devices) will be
+	 * subtracted later, following the path of aio_run_iocb().
+	 */
+	inode = file->f_mapping->host;
+	bdev = inode->i_sb->s_bdev;
+	ret = cgroup_io_throttle(bdev, iocb->aio_nbytes, 0);
+	if (unlikely(ret)) {
+		fput(file);
+		ret = -EAGAIN;
+		goto out_cgroup_io_throttle;
+	}
+
 	req = aio_get_req(ctx);		/* returns with 2 references to req */
 	if (unlikely(!req)) {
 		fput(file);
-		return -EAGAIN;
+		ret = -EAGAIN;
+		goto out_cgroup_io_throttle;
 	}
 	req->ki_filp = file;
 	if (iocb->aio_flags & IOCB_FLAG_RESFD) {
@@ -1622,12 +1641,14 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 		goto out_put_req;
 
 	spin_lock_irq(&ctx->ctx_lock);
+	set_in_aio();
 	aio_run_iocb(req);
 	if (!list_empty(&ctx->run_list)) {
 		/* drain the run list */
 		while (__aio_run_iocbs(ctx))
 			;
 	}
+	unset_in_aio();
 	spin_unlock_irq(&ctx->ctx_lock);
 	aio_put_req(req);	/* drop extra ref to req */
 	return 0;
@@ -1635,6 +1656,8 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 out_put_req:
 	aio_put_req(req);	/* drop extra ref to req */
 	aio_put_req(req);	/* drop i/o ref to req */
+out_cgroup_io_throttle:
+	cgroup_io_throttle(bdev, -iocb->aio_nbytes, 0);
 	return ret;
 }
 
@@ -1746,6 +1769,12 @@ asmlinkage long sys_io_cancel(aio_context_t ctx_id, struct iocb __user *iocb,
 	ret = -EAGAIN;
 	kiocb = lookup_kiocb(ctx, iocb, key);
 	if (kiocb && kiocb->ki_cancel) {
+		struct block_device *bdev;
+		struct inode *inode = kiocb->ki_filp->f_mapping->host;
+
+		bdev = inode->i_sb->s_bdev;
+		cgroup_io_throttle(bdev, -kiocb->ki_nbytes, 0);
+
 		cancel = kiocb->ki_cancel;
 		kiocb->ki_users ++;
 		kiocbSetCancelled(kiocb);
diff --git a/fs/buffer.c b/fs/buffer.c
index 4ffb5bb..6d4bf2c 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -35,6 +35,7 @@
 #include <linux/suspend.h>
 #include <linux/buffer_head.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/bio.h>
 #include <linux/notifier.h>
 #include <linux/cpu.h>
@@ -708,11 +709,14 @@ EXPORT_SYMBOL(mark_buffer_dirty_inode);
 static int __set_page_dirty(struct page *page,
 		struct address_space *mapping, int warn)
 {
+	ssize_t cgroup_io_acct = 0;
+	int ret = 0;
+
 	if (unlikely(!mapping))
 		return !TestSetPageDirty(page);
 
 	if (TestSetPageDirty(page))
-		return 0;
+		goto out;
 
 	spin_lock_irq(&mapping->tree_lock);
 	if (page->mapping) {	/* Race with truncate? */
@@ -723,14 +727,24 @@ static int __set_page_dirty(struct page *page,
 			__inc_bdi_stat(mapping->backing_dev_info,
 					BDI_RECLAIMABLE);
 			task_io_account_write(PAGE_CACHE_SIZE);
+			cgroup_io_acct = PAGE_CACHE_SIZE;
 		}
 		radix_tree_tag_set(&mapping->page_tree,
 				page_index(page), PAGECACHE_TAG_DIRTY);
 	}
 	spin_unlock_irq(&mapping->tree_lock);
 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
-
-	return 1;
+	ret = 1;
+out:
+	if (is_in_aio() && !cgroup_io_acct)
+		cgroup_io_acct = -PAGE_CACHE_SIZE;
+	if (cgroup_io_acct) {
+		struct block_device *bdev = (mapping->host &&
+				mapping->host->i_sb->s_bdev) ?
+				mapping->host->i_sb->s_bdev : NULL;
+		cgroup_io_throttle(bdev, cgroup_io_acct, 0);
+	}
+	return ret;
 }
 
 /*
diff --git a/fs/direct-io.c b/fs/direct-io.c
index 9606ee8..f5dcb91 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -35,6 +35,7 @@
 #include <linux/buffer_head.h>
 #include <linux/rwsem.h>
 #include <linux/uio.h>
+#include <linux/blk-io-throttle.h>
 #include <asm/atomic.h>
 
 /*
@@ -660,6 +661,9 @@ submit_page_section(struct dio *dio, struct page *page,
 		/*
 		 * Read accounting is performed in submit_bio()
 		 */
+		struct block_device *bdev = dio->bio ?
+					dio->bio->bi_bdev : NULL;
+		cgroup_io_throttle(bdev, len, 1);
 		task_io_account_write(len);
 	}
 
diff --git a/include/linux/sched.h b/include/linux/sched.h
index ba43675..9d4c755 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1250,6 +1250,9 @@ struct task_struct {
 	u64 rchar, wchar, syscr, syscw;
 #endif
 	struct task_io_accounting ioac;
+#ifdef CONFIG_CGROUP_IO_THROTTLE
+	atomic_t in_aio;
+#endif
 #if defined(CONFIG_TASK_XACCT)
 	u64 acct_rss_mem1;	/* accumulated rss usage */
 	u64 acct_vm_mem1;	/* accumulated virtual memory usage */
diff --git a/kernel/fork.c b/kernel/fork.c
index aed1ff7..f8cf5da 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1029,6 +1029,9 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	task_io_accounting_init(p);
 	acct_clear_integrals(p);
 
+#ifdef CONFIG_CGROUP_IO_THROTTLE
+	atomic_set(&p->in_aio, 0);
+#endif
 	p->it_virt_expires = cputime_zero;
 	p->it_prof_expires = cputime_zero;
 	p->it_sched_expires = 0;
diff --git a/mm/filemap.c b/mm/filemap.c
index 7567d86..bb80789 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -13,6 +13,7 @@
 #include <linux/slab.h>
 #include <linux/compiler.h>
 #include <linux/fs.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/uaccess.h>
 #include <linux/aio.h>
 #include <linux/capability.h>
@@ -1011,6 +1012,7 @@ static void do_generic_file_read(struct file *filp, loff_t *ppos,
 	pgoff_t prev_index;
 	unsigned long offset;      /* offset into pagecache page */
 	unsigned int prev_offset;
+	int was_page_ok = 0;
 	int error;
 
 	index = *ppos >> PAGE_CACHE_SHIFT;
@@ -1023,7 +1025,8 @@ static void do_generic_file_read(struct file *filp, loff_t *ppos,
 		struct page *page;
 		pgoff_t end_index;
 		loff_t isize;
-		unsigned long nr, ret;
+		ssize_t nr;
+		unsigned long ret;
 
 		cond_resched();
 find_page:
@@ -1051,6 +1054,8 @@ find_page:
 								desc, offset))
 				goto page_not_up_to_date_locked;
 			unlock_page(page);
+		} else {
+			was_page_ok = 1;
 		}
 page_ok:
 		/*
@@ -1080,6 +1085,17 @@ page_ok:
 		}
 		nr = nr - offset;
 
+		/*
+		 * De-account i/o in case of AIO read from the page cache.
+		 * AIO accounting was performed in io_submit_one().
+		 */
+		if (is_in_aio() && was_page_ok) {
+			struct block_device *bdev = (inode &&
+						inode->i_sb->s_bdev) ?
+						inode->i_sb->s_bdev : NULL;
+			cgroup_io_throttle(bdev, -nr, 0);
+		}
+
 		/* If users can be writing to this page using arbitrary
 		 * virtual addresses, take care about potential aliasing
 		 * before reading the page on the kernel side.
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 29b1d1e..c6207de 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -23,6 +23,7 @@
 #include <linux/init.h>
 #include <linux/backing-dev.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/blkdev.h>
 #include <linux/mpage.h>
 #include <linux/rmap.h>
@@ -430,6 +431,9 @@ static void balance_dirty_pages(struct address_space *mapping)
 	unsigned long write_chunk = sync_writeback_pages();
 
 	struct backing_dev_info *bdi = mapping->backing_dev_info;
+	struct block_device *bdev = (mapping->host &&
+					mapping->host->i_sb->s_bdev) ?
+					mapping->host->i_sb->s_bdev : NULL;
 
 	for (;;) {
 		struct writeback_control wbc = {
@@ -512,6 +516,14 @@ static void balance_dirty_pages(struct address_space *mapping)
 		return;		/* pdflush is already workin

This is the core io-throttle kernel infrastructure. It creates the basic
interfaces to cgroups and implements the I/O measurement and throttling
functions.

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 block/Makefile                  |    2 +
 block/blk-io-throttle.c         |  668 +++++++++++++++++++++++++++++++++++++++
 include/linux/blk-io-throttle.h |   41 +++
 include/linux/cgroup_subsys.h   |    6 +
 init/Kconfig                    |   10 +
 5 files changed, 727 insertions(+), 0 deletions(-)
 create mode 100644 block/blk-io-throttle.c
 create mode 100644 include/linux/blk-io-throttle.h

diff --git a/block/Makefile b/block/Makefile
index 208000b..b3afc86 100644
--- a/block/Makefile
+++ b/block/Makefile
@@ -13,6 +13,8 @@ obj-$(CONFIG_IOSCHED_AS)	+= as-iosched.o
 obj-$(CONFIG_IOSCHED_DEADLINE)	+= deadline-iosched.o
 obj-$(CONFIG_IOSCHED_CFQ)	+= cfq-iosched.o
 
+obj-$(CONFIG_CGROUP_IO_THROTTLE)	+= blk-io-throttle.o
+
 obj-$(CONFIG_BLK_DEV_IO_TRACE)	+= blktrace.o
 obj-$(CONFIG_BLOCK_COMPAT)	+= compat_ioctl.o
 obj-$(CONFIG_BLK_DEV_INTEGRITY)	+= blk-integrity.o
diff --git a/block/blk-io-throttle.c b/block/blk-io-throttle.c
new file mode 100644
index 0000000..6b0aa45
--- /dev/null
+++ b/block/blk-io-throttle.c
@@ -0,0 +1,668 @@
+/*
+ * blk-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 <righi.andrea@gmail.com>
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/cgroup.h>
+#include <linux/slab.h>
+#include <linux/gfp.h>
+#include <linux/err.h>
+#include <linux/sched.h>
+#include <linux/genhd.h>
+#include <linux/fs.h>
+#include <linux/jiffies.h>
+#include <linux/hardirq.h>
+#include <linux/list.h>
+#include <linux/seq_file.h>
+#include <linux/spinlock.h>
+#include <linux/uaccess.h>
+#include <linux/blk-io-throttle.h>
+
+#define IOTHROTTLE_BANDWIDTH	0
+#define IOTHROTTLE_IOPS		1
+
+/* The various types of throttling algorithms */
+enum iothrottle_strategy {
+	IOTHROTTLE_LEAKY_BUCKET = 0,
+	IOTHROTTLE_TOKEN_BUCKET = 1,
+};
+
+/**
+ * struct iothrottle_node - throttling rule of a single block device
+ * @node: list of per block device throttling rules
+ * @dev: block device number, used as key in the list
+ * @iorate: max i/o bandwidth (in bytes/s)
+ * @strategy: throttling strategy
+ * @timestamp: timestamp of the last i/o request for bandwidth limiting
+ *             (in jiffies)
+ * @stat: i/o activity counter (leaky bucket only)
+ * @bucket_size: bucket size in bytes (token bucket only)
+ * @token: token counter (token bucket only)
+ * @iops: max i/o operations per second
+ * @iops_stat: i/o operations counter (leaky bucket policy)
+ * @iops_timestamp: timestamp of the last i/o request for iops/sec limiting
+ *                  (in jiffies)
+ *
+ * Define a i/o throttling rule for a single block device.
+ *
+ * NOTE: limiting rules always refer to dev_t; if a block device is unplugged
+ * the limiting rules defined for that device persist and they are still valid
+ * if a new device is plugged and it uses the same dev_t number.
+ */
+struct iothrottle_node {
+	struct list_head node;
+	dev_t dev;
+
+	u64 iorate;
+	enum iothrottle_strategy strategy;
+	unsigned long timestamp;
+	atomic_long_t stat;
+	s64 bucket_size;
+	atomic_long_t token;
+
+	u64 iops;
+	atomic_long_t iops_stat;
+	unsigned long iops_timestamp;
+};
+
+/**
+ * struct iothrottle - throttling rules for a cgroup
+ * @css: pointer to the cgroup state
+ * @lock: spinlock used to protect write operations in the list
+ * @list: list of iothrottle_node elements
+ *
+ * Define multiple per-block device i/o throttling rules.
+ * Note: the list of the throttling rules is protected by RCU locking.
+ */
+struct iothrottle {
+	struct cgroup_subsys_state css;
+	spinlock_t lock;
+	struct list_head list;
+};
+
+static inline struct iothrottle *cgroup_to_iothrottle(struct cgroup *cgrp)
+{
+	return container_of(cgroup_subsys_state(cgrp, iothrottle_subsys_id),
+			    struct iothrottle, css);
+}
+
+static inline struct iothrottle *task_to_iothrottle(struct task_struct *task)
+{
+	return container_of(task_subsys_state(task, iothrottle_subsys_id),
+			    struct iothrottle, css);
+}
+
+/*
+ * Note: called with rcu_read_lock() or iot->lock held.
+ */
+static struct iothrottle_node *
+iothrottle_search_node(const struct iothrottle *iot, dev_t dev)
+{
+	struct iothrottle_node *n;
+
+	list_for_each_entry_rcu(n, &iot->list, node)
+		if (n->dev == dev)
+			return n;
+	return NULL;
+}
+
+/*
+ * Note: called with iot->lock held.
+ */
+static inline void iothrottle_insert_node(struct iothrottle *iot,
+						struct iothrottle_node *n)
+{
+	list_add_rcu(&n->node, &iot->list);
+}
+
+/*
+ * Note: called with iot->lock held.
+ */
+static inline void
+iothrottle_replace_node(struct iothrottle *iot, struct iothrottle_node *old,
+			struct iothrottle_node *new)
+{
+	list_replace_rcu(&old->node, &new->node);
+}
+
+/*
+ * Note: called with iot->lock held.
+ */
+static inline void
+iothrottle_delete_node(struct iothrottle *iot, struct iothrottle_node *n)
+{
+	list_del_rcu(&n->node);
+}
+
+/*
+ * Note: called from kernel/cgroup.c with cgroup_lock() held.
+ */
+static struct cgroup_subsys_state *
+iothrottle_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
+{
+	struct iothrottle *iot;
+
+	iot = kmalloc(sizeof(*iot), GFP_KERNEL);
+	if (unlikely(!iot))
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&iot->list);
+	spin_lock_init(&iot->lock);
+
+	return &iot->css;
+}
+
+/*
+ * Note: called from kernel/cgroup.c with cgroup_lock() held.
+ */
+static void iothrottle_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
+{
+	struct iothrottle_node *n, *p;
+	struct iothrottle *iot = cgroup_to_iothrottle(cgrp);
+
+	/*
+	 * don't worry about locking here, at this point there must be not any
+	 * reference to the list.
+	 */
+	list_for_each_entry_safe(n, p, &iot->list, node)
+		kfree(n);
+	kfree(iot);
+}
+
+static int iothrottle_read(struct cgroup *cgrp, struct cftype *cft,
+				struct seq_file *m)
+{
+	struct iothrottle *iot = cgroup_to_iothrottle(cgrp);
+	struct iothrottle_node *n;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(n, &iot->list, node) {
+		unsigned long delta;
+
+		BUG_ON(!n->dev);
+		switch (cft->private) {
+		case IOTHROTTLE_BANDWIDTH:
+			if (!n->iorate)
+				continue;
+			delta = jiffies_to_msecs((long)jiffies -
+					(long)n->timestamp);
+			seq_printf(m, "%u %u %llu %u %li %lli %li %lu\n",
+				MAJOR(n->dev), MINOR(n->dev),
+				n->iorate, n->strategy,
+				atomic_long_read(&n->stat),
+				n->bucket_size, atomic_long_read(&n->token),
+				delta);
+			break;
+		case IOTHROTTLE_IOPS:
+			if (!n->iops)
+				continue;
+			delta = jiffies_to_msecs((long)jiffies -
+					(long)n->iops_timestamp);
+			seq_printf(m, "%u %u %llu %li %lu\n",
+				MAJOR(n->dev), MINOR(n->dev),
+				n->iops, atomic_long_read(&n->iops_stat),
+				delta);
+			break;
+		}
+	}
+	rcu_read_unlock();
+	return 0;
+}
+
+static dev_t devname2dev_t(const char *buf)
+{
+	struct block_device *bdev;
+	dev_t dev = 0;
+	struct gendisk *disk;
+	int part;
+
+	/* use a lookup to validate the block device */
+	bdev = lookup_bdev(buf);
+	if (IS_ERR(bdev))
+		return 0;
+
+	/* only entire devices are allowed, not single partitions */
+	disk = get_gendisk(bdev->bd_dev, &part);
+	if (disk && !part) {
+		BUG_ON(!bdev->bd_inode);
+		dev = bdev->bd_inode->i_rdev;
+	}
+	bdput(bdev);
+
+	return dev;
+}
+
+/*
+ * The userspace input string must use one of the following syntaxes:
+ *
+ * blockio.bandwidth
+ * ~~~~~~~~~~~~~~~~~
+ * dev:0			<- delete a bandwidth limiting rule
+ * dev:bw-limit:0		<- set a leaky bucket throttling rule
+ * dev:bw-limit:1:bucket-size	<- set a token bucket throttling rule
+ *
+ * blockio.iops
+ * ~~~~~~~~~~~~
+ * dev:0			<- delete a iops/sec throttling rule
+ * dev:iops			<- set an iops/sec throttling rule
+ */
+static int iothrottle_parse_args(char *buf, size_t nbytes, int filetype,
+					dev_t *dev, u64 *iops, u64 *iorate,
+					enum iothrottle_strategy *strategy,
+					s64 *bucket_size)
+{
+	char *p;
+	int count = 0;
+	char *s[4];
+	unsigned long strategy_val;
+	int ret;
+
+	memset(s, 0, sizeof(s));
+	*dev = 0;
+	*iops = 0;
+	*iorate = 0;
+	*strategy = 0;
+	*bucket_size = 0;
+
+	/* split the colon-delimited input string into its elements */
+	while (count < ARRAY_SIZE(s)) {
+		p = strsep(&buf, ":");
+		if (!p)
+			break;
+		if (!*p)
+			continue;
+		s[count++] = p;
+	}
+
+	switch (filetype) {
+	case IOTHROTTLE_BANDWIDTH:
+		/* i/o bandwidth limit */
+		if (!s[1])
+			return -EINVAL;
+		ret = strict_strtoull(s[1], 10, iorate);
+		if (ret < 0)
+			return ret;
+		if (!*iorate)
+			goto out;
+		*iorate = ALIGN(*iorate, 1024);
+		/* throttling strategy */
+		if (!s[2])
+			return -EINVAL;
+		ret = strict_strtoul(s[2], 10, &strategy_val);
+		if (ret < 0)
+			return ret;
+		*strategy = (enum iothrottle_strategy)strategy_val;
+		switch (*strategy) {
+		case IOTHROTTLE_LEAKY_BUCKET:
+			goto out;
+		case IOTH

Documentation of the block device I/O bandwidth controller: description, usage,
advantages and design.

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 Documentation/controllers/io-throttle.txt |  328 +++++++++++++++++++++++++++++
 1 files changed, 328 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/controllers/io-throttle.txt

diff --git a/Documentation/controllers/io-throttle.txt b/Documentation/controllers/io-throttle.txt
new file mode 100644
index 0000000..f6b8bb9
--- /dev/null
+++ b/Documentation/controllers/io-throttle.txt
@@ -0,0 +1,328 @@
+
+               Block device I/O bandwidth controller
+
+----------------------------------------------------------------------
+1. DESCRIPTION
+
+This controller allows to limit the I/O bandwidth of specific block devices for
+specific process containers (cgroups) imposing additional delays on I/O
+requests for those processes that exceed the limits defined in the control
+group filesystem.
+
+Bandwidth limiting rules offer better control over QoS with respect to priority
+or weight-based solutions that only give information about applications'
+relative performance requirements. Nevertheless, priority based solutions are
+affected by performance bursts, when only low-priority requests are submitted
+to a general purpose resource dispatcher.
+
+The goal of the I/O bandwidth controller is to improve performance
+predictability and provide performance isolation of different control groups
+sharing the same block devices.
+
+NOTE #1: If you're looking for a way to improve the overall throughput of the
+system probably you should use a different solution.
+
+NOTE #2: The current implementation does not guarantee minimum bandwidth
+levels, the QoS is implemented only slowing down I/O "traffic" that exceeds the
+limits specified by the user; minimum I/O rate thresholds are supposed to be
+guaranteed if the user configures a proper I/O bandwidth partitioning of the
+block devices shared among the different cgroups (theoretically if the sum of
+all the single limits defined for a block device doesn't exceed the total I/O
+bandwidth of that device).
+
+----------------------------------------------------------------------
+2. USER INTERFACE
+
+A new I/O bandwidth limitation rule is described using the file
+blockio.bandwidth.
+
+The same file can be used to set multiple rules for different block devices
+relative to the same cgroup.
+
+2.1. Configure I/O bandwidth limiting rules
+
+The syntax to configure a limiting rule is the following:
+
+# /bin/echo DEV:BW:STRATEGY:BUCKET_SIZE > CGROUP/blockio.bandwidth
+
+- DEV is the name of the device the limiting rule is applied to.
+
+- BW is the maximum I/O bandwidth on DEV allowed by CGROUP; bandwidth must be
+  expressed in bytes/s. A generic I/O bandwidth limiting rule for a block
+  device DEV can be removed setting the BW value to 0.
+
+- STRATEGY is the throttling strategy used to throttle the applications' I/O
+  requests from/to device DEV. At the moment two different strategies can be
+  used:
+
+  0 = leaky bucket: the controller accepts at most B bytes (B = BW * time);
+		    further I/O requests are delayed scheduling a timeout for
+                    the tasks that made those requests.
+
+            Different I/O flow
+               | | |
+               | v |
+               |   v
+               v
+              .......
+              \     /
+               \   /  leaky-bucket
+                ---
+                |||
+                vvv
+             Smoothed I/O flow
+
+  1 = token bucket: BW tokens are added to the bucket every seconds; the bucket
+		    can hold at the most BUCKET_SIZE tokens; I/O requests are
+		    accepted if there are available tokens in the bucket; when
+		    a request of N bytes arrives N tokens are removed from the
+		    bucket; if fewer than N tokens are available the request is
+		    delayed until a sufficient amount of token is available in
+                    the bucket.
+
+            Tokens (I/O rate)
+                o
+                o
+                o
+              ....... <--.
+              \     /    | Bucket size (burst limit)
+               \ooo/     |
+                ---   <--'
+                 |ooo
+    Incoming --->|---> Conforming
+    I/O          |oo   I/O
+    requests  -->|-->  requests
+                 |
+            ---->|
+
+  Leaky bucket is more precise than token bucket to respect the bandwidth
+  limits, because bursty workloads are always smoothed. Token bucket, instead,
+  allows a small irregularity degree in the I/O flows (burst limit), and, for
+  this, it is better in terms of efficiency (bursty workloads are not smoothed
+  when there are sufficient tokens in the bucket).
+
+- BUCKET_SIZE is used only with token bucket (STRATEGY == 1) and defines the
+  size of the bucket in bytes.
+
+- CGROUP is the name of the limited process container.
+
+Also the following syntaxes are allowed:
+
+- remove an I/O bandwidth limiting rule
+# /bin/echo DEV:0 > CGROUP/blockio.bandwidth
+
+- configure a limiting rule using leaky bucket throttling (ignore bucket size):
+# /bin/echo DEV:BW:0 > CGROUP/blockio.bandwidth
+
+2.2. Show I/O bandwidth limiting rules
+
+All the defined rules and statistics for a specific cgroup can be shown reading
+the file blockio.bandwidth. The following syntax is used:
+
+$ cat CGROUP/blockio.bandwidth
+MAJOR MINOR BW STRATEGY LEAKY_STAT BUCKET_SIZE BUCKET_FILL TIME_DELTA
+
+- MAJOR is the major device number of DEV (defined above)
+
+- MINOR is the minor device number of DEV (defined above)
+
+- BW, STRATEGY and BUCKET_SIZE are the same parameters defined above
+
+- LEAKY_STAT is the amount of bytes currently allowed by the I/O bandwidth
+  controller (only used with leaky bucket strategy - STRATEGY == 0)
+
+- BUCKET_FILL represents the amount of tokens present in the bucket (only used
+  with token bucket strategy - STRATEGY == 1)
+
+- TIME_DELTA can be one of the following:
+  - the amount of jiffies elapsed from the last I/O request (token bucket)
+  - the amount of jiffies during which the bytes given by LEAKY_STAT have been
+    accumulated (leaky bucket)
+
+Multiple per-block device rules are reported in multiple rows
+(DEVi, i = 1 ..  n):
+
+$ cat CGROUP/blockio.bandwidth
+MAJOR1 MINOR1 BW1 STRATEGY1 LEAKY_STAT1 BUCKET_SIZE1 BUCKET_FILL1 TIME_DELTA1
+MAJOR1 MINOR1 BW2 STRATEGY2 LEAKY_STAT2 BUCKET_SIZE2 BUCKET_FILL2 TIME_DELTA2
+...
+MAJORn MINORn BWn STRATEGYn LEAKY_STATn BUCKET_SIZEn BUCKET_FILLn TIME_DELTAn
+
+2.3 Configure I/O operations/sec limiting rules
+
+The syntax to limit I/O operations/sec is the following:
+
+# /bin/echo DEV:IOPS > CGROUP/blockio.iops
+
+- DEV is the name of the device the limiting rule is applied to.
+
+- IOPS is the number of the maximum I/O operations per second that can be
+  issued on device DEV by the tasks belonging to cgroup CGROUP. The I/O
+  operations limit can be removed setting IOPS to 0.
+
+2.4 Show I/O operations/sec limiting rules
+
+The I/O operations limits and statistics for a specific cgroup can be shown
+reading the file blockio.ops. The following syntax is used:
+
+$ cat CGROUP/blockio.iops
+MAJOR MINOR IOPS IOPS_STAT TIME_DELTA
+
+- MAJOR is the major device number of DEV (defined above)
+
+- MINOR is the minor device number of DEV (defined above)
+
+- IOPS is the I/O operations/sec limit
+
+- IOPS_STAT is the number of I/O operations accumulated by the iops counter
+
+- TIME_DELTA can be one of the following:
+  - the amount of jiffies during which the number of I/O operations reported in
+    IOPS_STAT have been accumulated
+
+Multiple per-block device rules are reported in multiple rows
+(DEVi, i = 1 ..  n):
+
+$ cat CGROUP/blockio.bandwidth
+MAJOR1 MINOR1 IOPS1 IOPS_STAT1 TIME_DELTA1
+MAJOR2 MINOR2 IOPS2 IOPS_STAT2 TIME_DELTA2
+...
+MAJORn MINORn IOPSn IOPS_STATn TIME_DELTAn
+
+2.5. Examples
+
+* Mount the cgroup filesystem (blockio subsystem):
+  # mkdir /mnt/cgroup
+  # mount -t cgroup -oblockio blockio /mnt/cgroup
+
+* Instantiate the new cgroup "foo":
+  # mkdir /mnt/cgroup/foo
+  --> the cgroup foo has been created
+
+* Add the current shell process to the cgroup "foo":
+  # /bin/echo $$ > /mnt/cgroup/foo/tasks
+  --> the current shell has been added to the cgroup "foo"
+
+* Give maximum 1MiB/s of I/O bandwidth on /dev/sda for the cgroup "foo", using
+  leaky bucket throttling strategy:
+  # /bin/echo /dev/sda:$((1024 * 1024)):0:0 > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # sh
+  --> the subshell 'sh' is running in cgroup "foo" and it can use a maximum I/O
+      bandwidth of 1MiB/s on /dev/sda
+
+* Give maximum 8MiB/s of I/O bandwidth on /dev/sdb for the cgroup "foo", using
+  token bucket throttling strategy, bucket size = 8MiB:
+  # /bin/echo /dev/sdb:$((8 * 1024 * 1024)):1:$((8 * 1024 * 1024)) > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # sh
+  --> the subshell 'sh' is running in cgroup "foo" and it can use a maximum I/O
+      bandwidth of 1MiB/s on /dev/sda (controlled by leaky bucket throttling)
+      and 8MiB/s on /dev/sdb (controlled by token bucket throttling)
+
+* Run a benchmark doing I/O on /dev/sda and /dev/sdb; I/O limits and usage
+  defined for cgroup "foo" can be shown as following:
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 16 8388608 1 0 8388608 -522560 48
+  8 0 1048576 0 737280 0 0 216
+
+* Extend the maximum I/O bandwidth for the cgroup "foo" to 16MiB/s on /dev/sda:
+  # /bin/echo /dev/sda:$((16 * 1024 * 1024)):0:0 > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 16 8388608 1 0 8388608 -84432 206436
+  8 0 16777216 0 0 0 0 15212
+
+* Remove limiting rule on /dev/sdb for cgroup "foo":
+  # /bin/echo /dev/sdb:0:0:0 > /mnt/cgroup/foo/blockio.bandwidth
+  # cat /mnt/cgroup/foo/bl

The objective of the i/o bandwidth controller is to improve i/o performance
predictability of different cgroups sharing the same block devices.

Respect to other priority/weight-based solutions the approach used by this
controller is to explicitly choke applications' requests that directly (or
indirectly) generate i/o activity in the system.

The direct bandwidth limiting method has the advantage of improving the
performance predictability at the cost of reducing, in general, the overall
performance of the system (in terms of throughput).

Detailed informations about design, its goal and usage are described in the
documentation.

Tested against 2.6.26-rc8-mm1.

The all-in-one patch (and previous versions) can be found at:
http://download.systemimager.org/~arighi/linux/patches/io-throttle/

Changelog: (v6 -> v7)
  - added i/o operations per second throttling
  - fixed a build bug in x86 (undefined reference to `__udivdi3')
  - updated documentation

Following some results of a simple test I did to check the effectiveness of the
new iops throttling functionality (for Subrata: I'll post an update for the
io-throttle testcase in LTP ASAP).

testcase overview
=================
- cgroup #1: process P1 periodically reads a 5.5MB file and prints in stdout
             the time needed to read the file

- cgroup #2: a process P2 is started; P2 runs a lot of parallel md5sums of all
             the files under /usr (recursively)

We want to improve P1 responsiveness and better predict P1 performance,
regardless of the other i/o activities in the system, so we're going to measure
the times printed by P1 in stdout to evaluate the effectiveness of a each
tested solution for our particular requirement.

different configurations
========================
#1: no limiting at all
#2: plain CFQ priorities (P1 runs at real-time prio class 0, P2 runs at idle prio)
#3: iops throttling (P1 = unlimited, P2 = 50 iops/sec)
#4: bandwidth throttling (P1 = unlimited, P2 = 512KiB/s)
#5: bandwidth + iops throttling (P1 = unlimited, P2 = 512KiB/s and 50 iops/sec)
#6: aggressive bandwidth + iops throttling (P1 = unlimited, P2 = 128KiB/s and 10 iops/sec)

results (P1 response times)
===========================

   #1        #2        #3        #4        #5        #6
----------------------------------------------------------
 4.69724   4.68447   4.80822   4.37353   4.40609   4.37175
 4.71427   4.45847   4.40524   4.35441   4.37228   4.35842
 4.73120   4.46849   4.39400   4.36893   4.47388   4.36529
 4.83120   4.47956   4.37878   4.44221   4.36823   4.37942
 4.68060   4.49554   4.43058   4.40074   4.46004   4.37354
____________________ P2 starts here! _____________________
62.83110   7.06834   6.54557   7.10171   7.21964   5.35958
59.04400   6.92486  10.30330   5.38122   5.76458   4.89837
37.23380   7.11255   9.16971   8.32928   5.37017   5.51931
32.28180   7.26239   8.91513   6.27551   5.03347   4.79848
28.74150   7.19909   8.38274   5.00802   5.50771   4.72832

-Andrea
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

( This mail is sent to both OpenVZ and vserver ML)

Hello folks.

This is a new FS for virtual server.
Basically, Ths is "vhashify/vzcache" FS.
Just try it, please. I need your feed back.

URL:
http://www.digitalinfra.co.jp/20080720/hashfs.20080720.html


Best regards,

         --- Okajima, Jun. Tokyo, Japan.

To Mr. Herbert Poetzl :

See this URL:
http://www.mail-archive.com/vserver@list.linux-vserver.org/msg04247.html
>> BTW, I also am planning a new file system for Vserver.
>well, let's hear about it then ...

This is it.

From: Daniel Lezcano <dlezcano@fr.ibm.com>
Date: Sat, 19 Jul 2008 08:31:58 +0200

> [TCP] fix kernel panic with listening_get_next

Applied, thanks a lot Daniel.

_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

The mini-summit agenda has been updated at:

	http://wiki.openvz.org/Containers/Mini-summit_2008

Thanks.


















































Sauf indication contraire ci-dessus:
Compagnie IBM France
Siège Social : Tour Descartes, 2, avenue Gambetta, La Défense 5, 92400
Courbevoie
RCS Nanterre 552 118 465
Forme Sociale : S.A.S.
Capital Social : 542.737.118 ?
SIREN/SIRET : 552 118 465 02430
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

The containers mini-summit is happening at the Novotel Hotel.

Note that our breaks don't correspond to the break times given by
facilities.  I intend to ignore that - we can break for 5 mins to
collect coffee whenever - but if someone wants to update the times on
our agenda to better match these breaks that's fine.

I did however move the first two segments back by 30 minutes so that
we start at 8:30am.  Please note the time change!

-serge

----- Forwarded message from "C. Craig Ross" <ccr@linuxsymposium.org> -----

Date: Tue, 15 Jul 2008 17:01:19 -0400
From: "C. Craig Ross" <ccr@linuxsymposium.org>
To: "Serge E. Hallyn" <serue@us.ibm.com>,
	"Adams, Aland (OSLO R&D)" <aland.adams@hp.com>,
	"Brown, Len" <len.brown@intel.com>,
	"John W. Linville" <linville@tuxdriver.com>,
	Paul Moore <paul.moore@hp.com>
Subject: LS Mini Summit: Schedule Update & Info

Hi,

Here is the latest update.

Location
------------
Les Suites Hotel
1. Virtualization (Garden Suite Room) - Theatre Style Seating (35)
2. Linux Power Management (Rideau Suite) - U-Shape (20)
3. Linux Wireless LAN (Byward Suite) - U-Shape (25)

Novotel Hotel
1. Containers (Albion A) - U-Shape (35)
2. SE Linux (Albion B) - Theatre (45)

Schedule
-------------
08h30 - 16h30
With breaks at 10h00 - 10h30 and 15h30 - 16h00 and lunch at 12h30 - 13h30.
If the
breaks don't fit *exactly* with your schedule it should be ok since things
like coffee/snacks
shouldn't go back if you are 15 minutes early/late.

Registration
-----------------
Please send us your registration lists by Monday, July 21st (at the latest).
We are still trying to decide if we will be holding registration on Tuesday
or if we are just going to confirm your list against our registration list
so
that your attendees don't have to walk to the Congress Centre then walk
back.

We will provide specifics about room locations/access before the weekend.
Contact me directly if you have any questions but I think we're getting
there. :)

Cheers,

C.

On Tue, Jul 15, 2008 at 10:37 AM, C. Craig Ross <ccr@linuxsymposium.org>
wrote:

> Hi,
>
> Here are our attendance estimates, please confirm...
>
> Virtualization - 20
> Containers - 20
> SE Linux - 40
> Linux Power Management - 20
> Linux Wireless LAN - ... :-)
>
> We will be hosting the Mini Summits at Les Suites Hotel
> and most likely the Novotel Hotel (we will confirm soon).
> The rooms will be set up in theatre style seating
> unless you aren't doing formal presentations at which point
> a hollow square shape might be more beneficial.
>
> We also require confirmation for those of you who require
> projectors and screens or just flip charts and paper boards.
>
> Coffee breaks and lunches will be provided for all Mini Summits
> as they have generously been sponsored by HP. :-)
>
> Thank you.
>
> Cheers,
>
> C.
>
> P.S. You can just reply directly to me.
>

----- End forwarded message -----
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

Apply the io-throttle controller to the opportune kernel functions. Both
accounting and throttling functionalities are performed by
cgroup_io_throttle().

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 block/blk-core.c      |    9 +++++++++
 fs/aio.c              |   31 ++++++++++++++++++++++++++++++-
 fs/buffer.c           |   20 +++++++++++++++++---
 fs/direct-io.c        |    4 ++++
 include/linux/sched.h |    3 +++
 kernel/fork.c         |    3 +++
 mm/filemap.c          |   18 +++++++++++++++++-
 mm/page-writeback.c   |   30 +++++++++++++++++++++++++++---
 mm/readahead.c        |    5 +++++
 9 files changed, 115 insertions(+), 8 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 4c222ba..431294f 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -26,6 +26,7 @@
 #include <linux/swap.h>
 #include <linux/writeback.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/interrupt.h>
 #include <linux/cpu.h>
 #include <linux/blktrace_api.h>
@@ -1482,7 +1483,15 @@ void submit_bio(int rw, struct bio *bio)
 		if (rw & WRITE) {
 			count_vm_events(PGPGOUT, count);
 		} else {
+			struct page *page = bio_iovec_idx(bio, 0)->bv_page;
+
 			task_io_account_read(bio->bi_size);
+			/*
+			 * Do not throttle page requests that need to be
+			 * urgently reclaimed.
+			 */
+			cgroup_io_throttle(bio->bi_bdev, bio->bi_size,
+				!(PageReclaim(page) || PageSwapCache(page)));
 			count_vm_events(PGPGIN, count);
 		}
 
diff --git a/fs/aio.c b/fs/aio.c
index 0051fd9..1f3abb3 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -22,6 +22,7 @@
 #include <linux/sched.h>
 #include <linux/fs.h>
 #include <linux/file.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/mm.h>
 #include <linux/mman.h>
 #include <linux/slab.h>
@@ -1558,6 +1559,8 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 {
 	struct kiocb *req;
 	struct file *file;
+	struct block_device *bdev;
+	struct inode *inode;
 	ssize_t ret;
 
 	/* enforce forwards compatibility on users */
@@ -1580,10 +1583,26 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 	if (unlikely(!file))
 		return -EBADF;
 
+	/*
+	 * Pre-account AIO activity: we over-account *all* the bytes here;
+	 * bytes read from the page cache and bytes written in already dirtied
+	 * pages (that do not generate real i/o on block devices) will be
+	 * subtracted later, following the path of aio_run_iocb().
+	 */
+	inode = file->f_mapping->host;
+	bdev = inode->i_sb->s_bdev;
+	ret = cgroup_io_throttle(bdev, iocb->aio_nbytes, 0);
+	if (unlikely(ret)) {
+		fput(file);
+		ret = -EAGAIN;
+		goto out_cgroup_io_throttle;
+	}
+
 	req = aio_get_req(ctx);		/* returns with 2 references to req */
 	if (unlikely(!req)) {
 		fput(file);
-		return -EAGAIN;
+		ret = -EAGAIN;
+		goto out_cgroup_io_throttle;
 	}
 	req->ki_filp = file;
 	if (iocb->aio_flags & IOCB_FLAG_RESFD) {
@@ -1622,12 +1641,14 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 		goto out_put_req;
 
 	spin_lock_irq(&ctx->ctx_lock);
+	set_in_aio();
 	aio_run_iocb(req);
 	if (!list_empty(&ctx->run_list)) {
 		/* drain the run list */
 		while (__aio_run_iocbs(ctx))
 			;
 	}
+	unset_in_aio();
 	spin_unlock_irq(&ctx->ctx_lock);
 	aio_put_req(req);	/* drop extra ref to req */
 	return 0;
@@ -1635,6 +1656,8 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 out_put_req:
 	aio_put_req(req);	/* drop extra ref to req */
 	aio_put_req(req);	/* drop i/o ref to req */
+out_cgroup_io_throttle:
+	cgroup_io_throttle(bdev, -iocb->aio_nbytes, 0);
 	return ret;
 }
 
@@ -1746,6 +1769,12 @@ asmlinkage long sys_io_cancel(aio_context_t ctx_id, struct iocb __user *iocb,
 	ret = -EAGAIN;
 	kiocb = lookup_kiocb(ctx, iocb, key);
 	if (kiocb && kiocb->ki_cancel) {
+		struct block_device *bdev;
+		struct inode *inode = kiocb->ki_filp->f_mapping->host;
+
+		bdev = inode->i_sb->s_bdev;
+		cgroup_io_throttle(bdev, -kiocb->ki_nbytes, 0);
+
 		cancel = kiocb->ki_cancel;
 		kiocb->ki_users ++;
 		kiocbSetCancelled(kiocb);
diff --git a/fs/buffer.c b/fs/buffer.c
index 4ffb5bb..89808b1 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -35,6 +35,7 @@
 #include <linux/suspend.h>
 #include <linux/buffer_head.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/bio.h>
 #include <linux/notifier.h>
 #include <linux/cpu.h>
@@ -708,11 +709,14 @@ EXPORT_SYMBOL(mark_buffer_dirty_inode);
 static int __set_page_dirty(struct page *page,
 		struct address_space *mapping, int warn)
 {
+	ssize_t cgroup_io_acct = 0;
+	int ret = 0;
+
 	if (unlikely(!mapping))
 		return !TestSetPageDirty(page);
 
 	if (TestSetPageDirty(page))
-		return 0;
+		goto out;
 
 	spin_lock_irq(&mapping->tree_lock);
 	if (page->mapping) {	/* Race with truncate? */
@@ -723,14 +727,24 @@ static int __set_page_dirty(struct page *page,
 			__inc_bdi_stat(mapping->backing_dev_info,
 					BDI_RECLAIMABLE);
 			task_io_account_write(PAGE_CACHE_SIZE);
+			cgroup_io_acct = PAGE_CACHE_SIZE;
 		}
 		radix_tree_tag_set(&mapping->page_tree,
 				page_index(page), PAGECACHE_TAG_DIRTY);
 	}
 	spin_unlock_irq(&mapping->tree_lock);
 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
-
-	return 1;
+	ret = 1;
+out:
+	if (is_in_aio() && !cgroup_io_acct)
+		cgroup_io_acct = -PAGE_CACHE_SIZE;
+	if (cgroup_io_acct) {
+		struct block_device *bdev = (mapping->host &&
+		mapping->host->i_sb->s_bdev) ?
+		mapping->host->i_sb->s_bdev : NULL;
+		cgroup_io_throttle(bdev, cgroup_io_acct, 0);
+	}
+	return ret;
 }
 
 /*
diff --git a/fs/direct-io.c b/fs/direct-io.c
index 9606ee8..f5dcb91 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -35,6 +35,7 @@
 #include <linux/buffer_head.h>
 #include <linux/rwsem.h>
 #include <linux/uio.h>
+#include <linux/blk-io-throttle.h>
 #include <asm/atomic.h>
 
 /*
@@ -660,6 +661,9 @@ submit_page_section(struct dio *dio, struct page *page,
 		/*
 		 * Read accounting is performed in submit_bio()
 		 */
+		struct block_device *bdev = dio->bio ?
+					dio->bio->bi_bdev : NULL;
+		cgroup_io_throttle(bdev, len, 1);
 		task_io_account_write(len);
 	}
 
diff --git a/include/linux/sched.h b/include/linux/sched.h
index ba43675..9d4c755 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1250,6 +1250,9 @@ struct task_struct {
 	u64 rchar, wchar, syscr, syscw;
 #endif
 	struct task_io_accounting ioac;
+#ifdef CONFIG_CGROUP_IO_THROTTLE
+	atomic_t in_aio;
+#endif
 #if defined(CONFIG_TASK_XACCT)
 	u64 acct_rss_mem1;	/* accumulated rss usage */
 	u64 acct_vm_mem1;	/* accumulated virtual memory usage */
diff --git a/kernel/fork.c b/kernel/fork.c
index aed1ff7..f8cf5da 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1029,6 +1029,9 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	task_io_accounting_init(p);
 	acct_clear_integrals(p);
 
+#ifdef CONFIG_CGROUP_IO_THROTTLE
+	atomic_set(&p->in_aio, 0);
+#endif
 	p->it_virt_expires = cputime_zero;
 	p->it_prof_expires = cputime_zero;
 	p->it_sched_expires = 0;
diff --git a/mm/filemap.c b/mm/filemap.c
index 7567d86..bb80789 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -13,6 +13,7 @@
 #include <linux/slab.h>
 #include <linux/compiler.h>
 #include <linux/fs.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/uaccess.h>
 #include <linux/aio.h>
 #include <linux/capability.h>
@@ -1011,6 +1012,7 @@ static void do_generic_file_read(struct file *filp, loff_t *ppos,
 	pgoff_t prev_index;
 	unsigned long offset;      /* offset into pagecache page */
 	unsigned int prev_offset;
+	int was_page_ok = 0;
 	int error;
 
 	index = *ppos >> PAGE_CACHE_SHIFT;
@@ -1023,7 +1025,8 @@ static void do_generic_file_read(struct file *filp, loff_t *ppos,
 		struct page *page;
 		pgoff_t end_index;
 		loff_t isize;
-		unsigned long nr, ret;
+		ssize_t nr;
+		unsigned long ret;
 
 		cond_resched();
 find_page:
@@ -1051,6 +1054,8 @@ find_page:
 								desc, offset))
 				goto page_not_up_to_date_locked;
 			unlock_page(page);
+		} else {
+			was_page_ok = 1;
 		}
 page_ok:
 		/*
@@ -1080,6 +1085,17 @@ page_ok:
 		}
 		nr = nr - offset;
 
+		/*
+		 * De-account i/o in case of AIO read from the page cache.
+		 * AIO accounting was performed in io_submit_one().
+		 */
+		if (is_in_aio() && was_page_ok) {
+			struct block_device *bdev = (inode &&
+						inode->i_sb->s_bdev) ?
+						inode->i_sb->s_bdev : NULL;
+			cgroup_io_throttle(bdev, -nr, 0);
+		}
+
 		/* If users can be writing to this page using arbitrary
 		 * virtual addresses, take care about potential aliasing
 		 * before reading the page on the kernel side.
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 29b1d1e..c6207de 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -23,6 +23,7 @@
 #include <linux/init.h>
 #include <linux/backing-dev.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/blkdev.h>
 #include <linux/mpage.h>
 #include <linux/rmap.h>
@@ -430,6 +431,9 @@ static void balance_dirty_pages(struct address_space *mapping)
 	unsigned long write_chunk = sync_writeback_pages();
 
 	struct backing_dev_info *bdi = mapping->backing_dev_info;
+	struct block_device *bdev = (mapping->host &&
+					mapping->host->i_sb->s_bdev) ?
+					mapping->host->i_sb->s_bdev : NULL;
 
 	for (;;) {
 		struct writeback_control wbc = {
@@ -512,6 +516,14 @@ static void balance_dirty_pages(struct address_space *mapping)
 		return;		/* pdflush is already working th

This is the core io-throttle kernel infrastructure. It creates the basic
interfaces to cgroups and implements the I/O measurement and throttling
functions.

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 block/Makefile                  |    2 +
 block/blk-io-throttle.c         |  552 +++++++++++++++++++++++++++++++++++++++
 include/linux/blk-io-throttle.h |   41 +++
 include/linux/cgroup_subsys.h   |    6 +
 init/Kconfig                    |   10 +
 5 files changed, 611 insertions(+), 0 deletions(-)
 create mode 100644 block/blk-io-throttle.c
 create mode 100644 include/linux/blk-io-throttle.h

diff --git a/block/Makefile b/block/Makefile
index 208000b..b3afc86 100644
--- a/block/Makefile
+++ b/block/Makefile
@@ -13,6 +13,8 @@ obj-$(CONFIG_IOSCHED_AS)	+= as-iosched.o
 obj-$(CONFIG_IOSCHED_DEADLINE)	+= deadline-iosched.o
 obj-$(CONFIG_IOSCHED_CFQ)	+= cfq-iosched.o
 
+obj-$(CONFIG_CGROUP_IO_THROTTLE)	+= blk-io-throttle.o
+
 obj-$(CONFIG_BLK_DEV_IO_TRACE)	+= blktrace.o
 obj-$(CONFIG_BLOCK_COMPAT)	+= compat_ioctl.o
 obj-$(CONFIG_BLK_DEV_INTEGRITY)	+= blk-integrity.o
diff --git a/block/blk-io-throttle.c b/block/blk-io-throttle.c
new file mode 100644
index 0000000..f541e86
--- /dev/null
+++ b/block/blk-io-throttle.c
@@ -0,0 +1,552 @@
+/*
+ * blk-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 <righi.andrea@gmail.com>
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/cgroup.h>
+#include <linux/slab.h>
+#include <linux/gfp.h>
+#include <linux/err.h>
+#include <linux/sched.h>
+#include <linux/genhd.h>
+#include <linux/fs.h>
+#include <linux/jiffies.h>
+#include <linux/hardirq.h>
+#include <linux/list.h>
+#include <linux/seq_file.h>
+#include <linux/spinlock.h>
+#include <linux/uaccess.h>
+#include <linux/blk-io-throttle.h>
+
+/* The various types of throttling algorithms */
+enum iothrottle_strategy {
+	IOTHROTTLE_LEAKY_BUCKET = 0,
+	IOTHROTTLE_TOKEN_BUCKET = 1,
+};
+
+/**
+ * struct iothrottle_node - throttling rule of a single block device
+ * @node: list of per block device throttling rules
+ * @dev: block device number, used as key in the list
+ * @iorate: max i/o bandwidth (in bytes/s)
+ * @strategy: throttling strategy
+ * @timestamp: timestamp of the last I/O request (in jiffies)
+ * @stat: i/o activity counter (leaky bucket only)
+ * @bucket_size: bucket size in bytes (token bucket only)
+ * @token: token counter (token bucket only)
+ *
+ * Define a i/o throttling rule for a single block device.
+ *
+ * NOTE: limiting rules always refer to dev_t; if a block device is unplugged
+ * the limiting rules defined for that device persist and they are still valid
+ * if a new device is plugged and it uses the same dev_t number.
+ */
+struct iothrottle_node {
+	struct list_head node;
+	dev_t dev;
+	u64 iorate;
+	enum iothrottle_strategy strategy;
+	unsigned long timestamp;
+	atomic_long_t stat;
+	s64 bucket_size;
+	atomic_long_t token;
+};
+
+/**
+ * struct iothrottle - throttling rules for a cgroup
+ * @css: pointer to the cgroup state
+ * @lock: spinlock used to protect write operations in the list
+ * @list: list of iothrottle_node elements
+ *
+ * Define multiple per-block device i/o throttling rules.
+ * Note: the list of the throttling rules is protected by RCU locking.
+ */
+struct iothrottle {
+	struct cgroup_subsys_state css;
+	spinlock_t lock;
+	struct list_head list;
+};
+
+static inline struct iothrottle *cgroup_to_iothrottle(struct cgroup *cgrp)
+{
+	return container_of(cgroup_subsys_state(cgrp, iothrottle_subsys_id),
+			    struct iothrottle, css);
+}
+
+static inline struct iothrottle *task_to_iothrottle(struct task_struct *task)
+{
+	return container_of(task_subsys_state(task, iothrottle_subsys_id),
+			    struct iothrottle, css);
+}
+
+/*
+ * Note: called with rcu_read_lock() or iot->lock held.
+ */
+static struct iothrottle_node *
+iothrottle_search_node(const struct iothrottle *iot, dev_t dev)
+{
+	struct iothrottle_node *n;
+
+	list_for_each_entry_rcu(n, &iot->list, node)
+		if (n->dev == dev)
+			return n;
+	return NULL;
+}
+
+/*
+ * Note: called with iot->lock held.
+ */
+static inline void iothrottle_insert_node(struct iothrottle *iot,
+						struct iothrottle_node *n)
+{
+	list_add_rcu(&n->node, &iot->list);
+}
+
+/*
+ * Note: called with iot->lock held.
+ */
+static inline void
+iothrottle_replace_node(struct iothrottle *iot, struct iothrottle_node *old,
+			struct iothrottle_node *new)
+{
+	list_replace_rcu(&old->node, &new->node);
+}
+
+/*
+ * Note: called with iot->lock held.
+ */
+static struct iothrottle_node *
+iothrottle_delete_node(struct iothrottle *iot, dev_t dev)
+{
+	struct iothrottle_node *n;
+
+	list_for_each_entry(n, &iot->list, node)
+		if (n->dev == dev) {
+			list_del_rcu(&n->node);
+			return n;
+		}
+	return NULL;
+}
+
+/*
+ * Note: called from kernel/cgroup.c with cgroup_lock() held.
+ */
+static struct cgroup_subsys_state *
+iothrottle_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
+{
+	struct iothrottle *iot;
+
+	iot = kmalloc(sizeof(*iot), GFP_KERNEL);
+	if (unlikely(!iot))
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&iot->list);
+	spin_lock_init(&iot->lock);
+
+	return &iot->css;
+}
+
+/*
+ * Note: called from kernel/cgroup.c with cgroup_lock() held.
+ */
+static void iothrottle_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
+{
+	struct iothrottle_node *n, *p;
+	struct iothrottle *iot = cgroup_to_iothrottle(cgrp);
+
+	/*
+	 * don't worry about locking here, at this point there must be not any
+	 * reference to the list.
+	 */
+	list_for_each_entry_safe(n, p, &iot->list, node)
+		kfree(n);
+	kfree(iot);
+}
+
+static int iothrottle_read(struct cgroup *cgrp, struct cftype *cft,
+				struct seq_file *m)
+{
+	struct iothrottle *iot = cgroup_to_iothrottle(cgrp);
+	struct iothrottle_node *n;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(n, &iot->list, node) {
+		unsigned long delta;
+
+		BUG_ON(!n->dev);
+		delta = jiffies_to_msecs((long)jiffies - (long)n->timestamp);
+		seq_printf(m, "%u %u %llu %u %li %lli %li %lu\n",
+				MAJOR(n->dev), MINOR(n->dev), n->iorate,
+				n->strategy, atomic_long_read(&n->stat),
+				n->bucket_size, atomic_long_read(&n->token),
+				delta);
+	}
+	rcu_read_unlock();
+	return 0;
+}
+
+static dev_t devname2dev_t(const char *buf)
+{
+	struct block_device *bdev;
+	dev_t dev = 0;
+	struct gendisk *disk;
+	int part;
+
+	/* use a lookup to validate the block device */
+	bdev = lookup_bdev(buf);
+	if (IS_ERR(bdev))
+		return 0;
+
+	/* only entire devices are allowed, not single partitions */
+	disk = get_gendisk(bdev->bd_dev, &part);
+	if (disk && !part) {
+		BUG_ON(!bdev->bd_inode);
+		dev = bdev->bd_inode->i_rdev;
+	}
+	bdput(bdev);
+
+	return dev;
+}
+
+/*
+ * The userspace input string must use one of the following syntax:
+ *
+ * dev:0			<- delete a limiting rule
+ * dev:bw-limit:0		<- leaky bucket throttling rule
+ * dev:bw-limit:1:bucket-size	<- token bucket throttling rule
+ */
+static int iothrottle_parse_args(char *buf, size_t nbytes, dev_t *dev,
+					u64 *iorate,
+					enum iothrottle_strategy *strategy,
+					s64 *bucket_size)
+{
+	char *p;
+	int count = 0;
+	char *s[4];
+	unsigned long strategy_val;
+	int ret;
+
+	/* split the colon-delimited input string into its elements */
+	memset(s, 0, sizeof(s));
+	while (count < ARRAY_SIZE(s)) {
+		p = strsep(&buf, ":");
+		if (!p)
+			break;
+		if (!*p)
+			continue;
+		s[count++] = p;
+	}
+
+	/* i/o bandwidth limit */
+	if (!s[1])
+		return -EINVAL;
+	ret = strict_strtoull(s[1], 10, iorate);
+	if (ret < 0)
+		return ret;
+	if (!*iorate) {
+		/*
+		 * we're deleting a limiting rule, so just ignore the other
+		 * parameters
+		 */
+		*strategy = 0;
+		*bucket_size = 0;
+		goto out;
+	}
+	*iorate = ALIGN(*iorate, 1024);
+
+	/* throttling strategy */
+	if (!s[2])
+		return -EINVAL;
+	ret = strict_strtoul(s[2], 10, &strategy_val);
+	if (ret < 0)
+		return ret;
+	*strategy = (enum iothrottle_strategy)strategy_val;
+	switch (*strategy) {
+	case IOTHROTTLE_LEAKY_BUCKET:
+		/* leaky bucket ignores bucket size */
+		*bucket_size = 0;
+		goto out;
+	case IOTHROTTLE_TOKEN_BUCKET:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	/* bucket size */
+	if (!s[3])
+		return -EINVAL;
+	ret = strict_strtoll(s[3], 10, bucket_size);
+	if (ret < 0)
+		return ret;
+	if (*bucket_size < 0)
+		return -EINVAL;
+	*bucket_size = ALIGN(*bucket_size, 1024);
+out:
+	/* block device number */
+	*dev = devname2dev_t(s[0]);
+	return *dev ? 0 : -EINVAL;
+}
+
+static int iothrottle_write(struct cgroup *cgrp, struct cftype *cft,
+				const char *buffer)
+{
+	struct iothrottle *iot;
+	struct iothrottle_node *n, *newn = NULL;
+	dev_t dev;
+	u64 iorate;
+	enum iothrottle_strategy strategy;
+	s64 bucket_size;
+	char *buf;
+	size_t nbytes = strlen(buffer);
+	int ret = 0;
+
+	buf = kmalloc(nbytes + 1, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+	memcpy(buf, buffer, nbytes + 1);
+
+	ret = iothrottle_parse_args(buf, nbytes, &dev, &iorat

Documentation of the block device I/O bandwidth controller: description, usage,
advantages and design.

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 Documentation/controllers/io-throttle.txt |  282 +++++++++++++++++++++++++++++
 1 files changed, 282 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/controllers/io-throttle.txt

diff --git a/Documentation/controllers/io-throttle.txt b/Documentation/controllers/io-throttle.txt
new file mode 100644
index 0000000..9c53f6c
--- /dev/null
+++ b/Documentation/controllers/io-throttle.txt
@@ -0,0 +1,282 @@
+
+               Block device I/O bandwidth controller
+
+1. Description
+
+This controller allows to limit the I/O bandwidth of specific block devices for
+specific process containers (cgroups) imposing additional delays on I/O
+requests for those processes that exceed the limits defined in the control
+group filesystem.
+
+Bandwidth limiting rules offer better control over QoS with respect to priority
+or weight-based solutions that only give information about applications'
+relative performance requirements. Nevertheless, priority based solutions are
+affected by performance bursts, when only low-priority requests are submitted
+to a general purpose resource dispatcher.
+
+The goal of the I/O bandwidth controller is to improve performance
+predictability and provide performance isolation of different control groups
+sharing the same block devices.
+
+NOTE #1: If you're looking for a way to improve the overall throughput of the
+system probably you should use a different solution.
+
+NOTE #2: The current implementation does not guarantee minimum bandwidth
+levels, the QoS is implemented only slowing down I/O "traffic" that exceeds the
+limits specified by the user; minimum I/O rate thresholds are supposed to be
+guaranteed if the user configures a proper I/O bandwidth partitioning of the
+block devices shared among the different cgroups (theoretically if the sum of
+all the single limits defined for a block device doesn't exceed the total I/O
+bandwidth of that device).
+
+2. User Interface
+
+A new I/O bandwidth limitation rule is described using the file
+blockio.bandwidth.
+
+The same file can be used to set multiple rules for different block devices
+relative to the same cgroup.
+
+2.1. Configure I/O limiting rules
+
+The syntax to configure a limiting rule is the following:
+
+# /bin/echo DEV:BW:STRATEGY:BUCKET_SIZE > CGROUP/blockio.bandwidth
+
+- DEV is the name of the device the limiting rule is applied to.
+
+- BW is the maximum I/O bandwidth on DEVICE allowed by CGROUP; bandwidth must
+  be expressed in bytes/s. A generic I/O bandwidth limiting rule for a block
+  device DEV can be removed setting the BW value to 0.
+
+- STRATEGY is the throttling strategy used to throttle the applications' I/O
+  requests from/to device DEV. At the moment two different strategies can be
+  used:
+
+  0 = leaky bucket: the controller accepts at most B bytes (B = BW * time);
+		    further I/O requests are delayed scheduling a timeout for
+                    the tasks that made those requests.
+
+            Different I/O flow
+               | | |
+               | v |
+               |   v
+               v
+              .......
+              \     /
+               \   /  leaky-bucket
+                ---
+                |||
+                vvv
+             Smoothed I/O flow
+
+  1 = token bucket: BW tokens are added to the bucket every seconds; the bucket
+		    can hold at the most BUCKET_SIZE tokens; I/O requests are
+		    accepted if there are available tokens in the bucket; when
+		    a request of N bytes arrives N tokens are removed from the
+		    bucket; if fewer than N tokens are available the request is
+		    delayed until a sufficient amount of token is available in
+                    the bucket.
+
+            Tokens (I/O rate)
+                o
+                o
+                o
+              ....... <--.
+              \     /    | Bucket size (burst limit)
+               \ooo/     |
+                ---   <--'
+                 |ooo
+    Incoming --->|---> Conforming
+    I/O          |oo   I/O
+    requests  -->|-->  requests
+                 |
+            ---->|
+
+  Leaky bucket is more precise than token bucket to respect the bandwidth
+  limits, because bursty workloads are always smoothed. Token bucket, instead,
+  allows a small irregularity degree in the I/O flows (burst limit), and, for
+  this, it is better in terms of efficiency (bursty workloads are not smoothed
+  when there are sufficient tokens in the bucket).
+
+- BUCKET_SIZE is used only with token bucket (STRATEGY == 1) and defines the
+  size of the bucket in bytes.
+
+- CGROUP is the name of the limited process container.
+
+Also the following syntaxes are allowed:
+
+- remove an I/O bandwidth limiting rule
+# /bin/echo DEV:0 > CGROUP/blockio.bandwidth
+
+- configure a limiting rule using leaky bucket throttling (ignore bucket size):
+# /bin/echo DEV:BW:0 > CGROUP/blockio.bandwidth
+
+2.2. Show I/O limiting rules
+
+All the defined rules and statistics for a specific cgroup can be shown reading
+the file blockio.bandwidth. The following syntax is used:
+
+$ cat CGROUP/blockio.bandwidth
+MAJOR MINOR BW STRATEGY LEAKY_STAT BUCKET_SIZE BUCKET_FILL TIME_DELTA
+
+- MAJOR is the major device number of DEV (defined above)
+
+- MINOR is the minor device number of DEV (defined above)
+
+- BW, STRATEGY and BUCKET_SIZE are the same parameters defined above
+
+- LEAKY_STAT is the amount of bytes currently allowed by the I/O bandwidth
+  controller (only used with leaky bucket strategy - STRATEGY == 0)
+
+- BUCKET_FILL represents the amount of tokens present in the bucket (only used
+  with token bucket strategy - STRATEGY == 1)
+
+- TIME_DELTA can be one of the following:
+  - the amount of jiffies elapsed from the last I/O request (token bucket)
+  - the amount of jiffies during which the bytes given by LEAKY_STAT have been
+    accumulated (leaky bucket)
+
+Multiple per-block device rules are reported in multiple rows
+(DEVi, i = 1 ..  n):
+
+$ cat CGROUP/blockio.bandwidth
+MAJOR1 MINOR1 BW1 STRATEGY1 LEAKY_STAT1 BUCKET_SIZE1 BUCKET_FILL1 TIME_DELTA1
+MAJOR1 MINOR1 BW2 STRATEGY2 LEAKY_STAT2 BUCKET_SIZE2 BUCKET_FILL2 TIME_DELTA2
+...
+MAJORn MINORn BWn STRATEGYn LEAKY_STATn BUCKET_SIZEn BUCKET_FILLn TIME_DELTAn
+
+2.3. Examples
+
+* Mount the cgroup filesystem (blockio subsystem):
+  # mkdir /mnt/cgroup
+  # mount -t cgroup -oblockio blockio /mnt/cgroup
+
+* Instantiate the new cgroup "foo":
+  # mkdir /mnt/cgroup/foo
+  --> the cgroup foo has been created
+
+* Add the current shell process to the cgroup "foo":
+  # /bin/echo $$ > /mnt/cgroup/foo/tasks
+  --> the current shell has been added to the cgroup "foo"
+
+* Give maximum 1MiB/s of I/O bandwidth on /dev/sda for the cgroup "foo", using
+  leaky bucket throttling strategy:
+  # /bin/echo /dev/sda:$((1024 * 1024)):0:0 > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # sh
+  --> the subshell 'sh' is running in cgroup "foo" and it can use a maximum I/O
+      bandwidth of 1MiB/s on /dev/sda
+
+* Give maximum 8MiB/s of I/O bandwidth on /dev/sdb for the cgroup "foo", using
+  token bucket throttling strategy, bucket size = 8MiB:
+  # /bin/echo /dev/sdb:$((8 * 1024 * 1024)):1:$((8 * 1024 * 1024)) > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # sh
+  --> the subshell 'sh' is running in cgroup "foo" and it can use a maximum I/O
+      bandwidth of 1MiB/s on /dev/sda (controlled by leaky bucket throttling)
+      and 8MiB/s on /dev/sdb (controlled by token bucket throttling)
+
+* Run a benchmark doing I/O on /dev/sda and /dev/sdb; I/O limits and usage
+  defined for cgroup "foo" can be shown as following:
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 16 8388608 1 0 8388608 -522560 48
+  8 0 1048576 0 737280 0 0 216
+
+* Extend the maximum I/O bandwidth for the cgroup "foo" to 16MiB/s on /dev/sda:
+  # /bin/echo /dev/sda:$((16 * 1024 * 1024)):0:0 > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 16 8388608 1 0 8388608 -84432 206436
+  8 0 16777216 0 0 0 0 15212
+
+* Remove limiting rule on /dev/sdb for cgroup "foo":
+  # /bin/echo /dev/sdb:0:0:0 > /mnt/cgroup/foo/blockio.bandwidth
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 0 16777216 0 0 0 0 110388
+
+3. Advantages of providing this feature
+
+* Allow I/O traffic shaping for block device shared among different cgroups
+* Improve I/O performance predictability on block devices shared between
+  different cgroups
+* Limiting rules do not depend of the particular I/O scheduler (anticipatory,
+  deadline, CFQ, noop) and/or the type of the underlying block devices
+* The bandwidth limitations are guaranteed both for synchronous and
+  asynchronous operations, even the I/O passing through the page cache or
+  buffers and not only direct I/O (see below for details)
+* It is possible to implement a simple user-space application to dynamically
+  adjust the I/O workload of different process containers at run-time,
+  according to the particular users' requirements and applications' performance
+  constraints
+* It is even possible to implement event-based performance throttling
+  mechanisms; for example the same user-space application could actively
+  throttle the I/O bandwidth to reduce power consumption when the battery of a
+  mobile device is running low (power throttling) or when the temperature of a
+  hardware component is too high (thermal throttling)
+
+4. Design
+
+The I/O throttling is performed imposing an explicit timeout, via
+schedule_timeout_killable() on the processes that exceed the I/O bandwidth
+dedicated to the cgroup they belong to. I/O accounting happens per cgroup.
+
+It just works as expected for read operations: the

The objective of the i/o bandwidth controller is to improve i/o performance
predictability of different cgroups sharing the same block devices.

Respect to other priority/weight-based solutions the approach used by this
controller is to explicitly choke applications' requests that directly (or
indirectly) generate i/o activity in the system.

The direct bandwidth limiting method has the advantage of improving the
performance predictability at the cost of reducing, in general, the overall
performance of the system (in terms of throughput).

Detailed informations about design, its goal and usage are described in the
documentation.

Tested against 2.6.26-rc8-mm1.

The all-in-one patch (and previous versions) can be found at:
http://download.systemimager.org/~arighi/linux/patches/io-throttle/

Changelog: (v5 -> v6)
  - do not make kernel threads to sleep
  - do not throttle i/o for pages that need to be urgently reclaimed in
    submit_bio(READ, ...) (i.e. tasks such as pdflush and kswapd when
    performing writeout)
  - minor fixes and improvements (thanks to Li Zefan review)
  - fixed a small typo in the documentation (reported by Marco Innocenti)

TODO:
  - see documentation

-Andrea
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

Apply the io-throttle controller to the opportune kernel functions. Both
accounting and throttling functionalities are performed by
cgroup_io_throttle().

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 block/blk-core.c      |    2 ++
 fs/aio.c              |   31 ++++++++++++++++++++++++++++++-
 fs/buffer.c           |   20 +++++++++++++++++---
 fs/direct-io.c        |    4 ++++
 include/linux/sched.h |    3 +++
 kernel/fork.c         |    3 +++
 mm/filemap.c          |   18 +++++++++++++++++-
 mm/page-writeback.c   |   30 +++++++++++++++++++++++++++---
 mm/readahead.c        |    5 +++++
 9 files changed, 108 insertions(+), 8 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 4c222ba..bffce33 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -26,6 +26,7 @@
 #include <linux/swap.h>
 #include <linux/writeback.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/interrupt.h>
 #include <linux/cpu.h>
 #include <linux/blktrace_api.h>
@@ -1483,6 +1484,7 @@ void submit_bio(int rw, struct bio *bio)
 			count_vm_events(PGPGOUT, count);
 		} else {
 			task_io_account_read(bio->bi_size);
+			cgroup_io_throttle(bio->bi_bdev, bio->bi_size, 1);
 			count_vm_events(PGPGIN, count);
 		}
 
diff --git a/fs/aio.c b/fs/aio.c
index 0051fd9..1f3abb3 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -22,6 +22,7 @@
 #include <linux/sched.h>
 #include <linux/fs.h>
 #include <linux/file.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/mm.h>
 #include <linux/mman.h>
 #include <linux/slab.h>
@@ -1558,6 +1559,8 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 {
 	struct kiocb *req;
 	struct file *file;
+	struct block_device *bdev;
+	struct inode *inode;
 	ssize_t ret;
 
 	/* enforce forwards compatibility on users */
@@ -1580,10 +1583,26 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 	if (unlikely(!file))
 		return -EBADF;
 
+	/*
+	 * Pre-account AIO activity: we over-account *all* the bytes here;
+	 * bytes read from the page cache and bytes written in already dirtied
+	 * pages (that do not generate real i/o on block devices) will be
+	 * subtracted later, following the path of aio_run_iocb().
+	 */
+	inode = file->f_mapping->host;
+	bdev = inode->i_sb->s_bdev;
+	ret = cgroup_io_throttle(bdev, iocb->aio_nbytes, 0);
+	if (unlikely(ret)) {
+		fput(file);
+		ret = -EAGAIN;
+		goto out_cgroup_io_throttle;
+	}
+
 	req = aio_get_req(ctx);		/* returns with 2 references to req */
 	if (unlikely(!req)) {
 		fput(file);
-		return -EAGAIN;
+		ret = -EAGAIN;
+		goto out_cgroup_io_throttle;
 	}
 	req->ki_filp = file;
 	if (iocb->aio_flags & IOCB_FLAG_RESFD) {
@@ -1622,12 +1641,14 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 		goto out_put_req;
 
 	spin_lock_irq(&ctx->ctx_lock);
+	set_in_aio();
 	aio_run_iocb(req);
 	if (!list_empty(&ctx->run_list)) {
 		/* drain the run list */
 		while (__aio_run_iocbs(ctx))
 			;
 	}
+	unset_in_aio();
 	spin_unlock_irq(&ctx->ctx_lock);
 	aio_put_req(req);	/* drop extra ref to req */
 	return 0;
@@ -1635,6 +1656,8 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
 out_put_req:
 	aio_put_req(req);	/* drop extra ref to req */
 	aio_put_req(req);	/* drop i/o ref to req */
+out_cgroup_io_throttle:
+	cgroup_io_throttle(bdev, -iocb->aio_nbytes, 0);
 	return ret;
 }
 
@@ -1746,6 +1769,12 @@ asmlinkage long sys_io_cancel(aio_context_t ctx_id, struct iocb __user *iocb,
 	ret = -EAGAIN;
 	kiocb = lookup_kiocb(ctx, iocb, key);
 	if (kiocb && kiocb->ki_cancel) {
+		struct block_device *bdev;
+		struct inode *inode = kiocb->ki_filp->f_mapping->host;
+
+		bdev = inode->i_sb->s_bdev;
+		cgroup_io_throttle(bdev, -kiocb->ki_nbytes, 0);
+
 		cancel = kiocb->ki_cancel;
 		kiocb->ki_users ++;
 		kiocbSetCancelled(kiocb);
diff --git a/fs/buffer.c b/fs/buffer.c
index 4ffb5bb..89808b1 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -35,6 +35,7 @@
 #include <linux/suspend.h>
 #include <linux/buffer_head.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/bio.h>
 #include <linux/notifier.h>
 #include <linux/cpu.h>
@@ -708,11 +709,14 @@ EXPORT_SYMBOL(mark_buffer_dirty_inode);
 static int __set_page_dirty(struct page *page,
 		struct address_space *mapping, int warn)
 {
+	ssize_t cgroup_io_acct = 0;
+	int ret = 0;
+
 	if (unlikely(!mapping))
 		return !TestSetPageDirty(page);
 
 	if (TestSetPageDirty(page))
-		return 0;
+		goto out;
 
 	spin_lock_irq(&mapping->tree_lock);
 	if (page->mapping) {	/* Race with truncate? */
@@ -723,14 +727,24 @@ static int __set_page_dirty(struct page *page,
 			__inc_bdi_stat(mapping->backing_dev_info,
 					BDI_RECLAIMABLE);
 			task_io_account_write(PAGE_CACHE_SIZE);
+			cgroup_io_acct = PAGE_CACHE_SIZE;
 		}
 		radix_tree_tag_set(&mapping->page_tree,
 				page_index(page), PAGECACHE_TAG_DIRTY);
 	}
 	spin_unlock_irq(&mapping->tree_lock);
 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
-
-	return 1;
+	ret = 1;
+out:
+	if (is_in_aio() && !cgroup_io_acct)
+		cgroup_io_acct = -PAGE_CACHE_SIZE;
+	if (cgroup_io_acct) {
+		struct block_device *bdev = (mapping->host &&
+		mapping->host->i_sb->s_bdev) ?
+		mapping->host->i_sb->s_bdev : NULL;
+		cgroup_io_throttle(bdev, cgroup_io_acct, 0);
+	}
+	return ret;
 }
 
 /*
diff --git a/fs/direct-io.c b/fs/direct-io.c
index 9606ee8..f5dcb91 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -35,6 +35,7 @@
 #include <linux/buffer_head.h>
 #include <linux/rwsem.h>
 #include <linux/uio.h>
+#include <linux/blk-io-throttle.h>
 #include <asm/atomic.h>
 
 /*
@@ -660,6 +661,9 @@ submit_page_section(struct dio *dio, struct page *page,
 		/*
 		 * Read accounting is performed in submit_bio()
 		 */
+		struct block_device *bdev = dio->bio ?
+					dio->bio->bi_bdev : NULL;
+		cgroup_io_throttle(bdev, len, 1);
 		task_io_account_write(len);
 	}
 
diff --git a/include/linux/sched.h b/include/linux/sched.h
index ba43675..9d4c755 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1250,6 +1250,9 @@ struct task_struct {
 	u64 rchar, wchar, syscr, syscw;
 #endif
 	struct task_io_accounting ioac;
+#ifdef CONFIG_CGROUP_IO_THROTTLE
+	atomic_t in_aio;
+#endif
 #if defined(CONFIG_TASK_XACCT)
 	u64 acct_rss_mem1;	/* accumulated rss usage */
 	u64 acct_vm_mem1;	/* accumulated virtual memory usage */
diff --git a/kernel/fork.c b/kernel/fork.c
index aed1ff7..f8cf5da 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1029,6 +1029,9 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	task_io_accounting_init(p);
 	acct_clear_integrals(p);
 
+#ifdef CONFIG_CGROUP_IO_THROTTLE
+	atomic_set(&p->in_aio, 0);
+#endif
 	p->it_virt_expires = cputime_zero;
 	p->it_prof_expires = cputime_zero;
 	p->it_sched_expires = 0;
diff --git a/mm/filemap.c b/mm/filemap.c
index 7567d86..bb80789 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -13,6 +13,7 @@
 #include <linux/slab.h>
 #include <linux/compiler.h>
 #include <linux/fs.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/uaccess.h>
 #include <linux/aio.h>
 #include <linux/capability.h>
@@ -1011,6 +1012,7 @@ static void do_generic_file_read(struct file *filp, loff_t *ppos,
 	pgoff_t prev_index;
 	unsigned long offset;      /* offset into pagecache page */
 	unsigned int prev_offset;
+	int was_page_ok = 0;
 	int error;
 
 	index = *ppos >> PAGE_CACHE_SHIFT;
@@ -1023,7 +1025,8 @@ static void do_generic_file_read(struct file *filp, loff_t *ppos,
 		struct page *page;
 		pgoff_t end_index;
 		loff_t isize;
-		unsigned long nr, ret;
+		ssize_t nr;
+		unsigned long ret;
 
 		cond_resched();
 find_page:
@@ -1051,6 +1054,8 @@ find_page:
 								desc, offset))
 				goto page_not_up_to_date_locked;
 			unlock_page(page);
+		} else {
+			was_page_ok = 1;
 		}
 page_ok:
 		/*
@@ -1080,6 +1085,17 @@ page_ok:
 		}
 		nr = nr - offset;
 
+		/*
+		 * De-account i/o in case of AIO read from the page cache.
+		 * AIO accounting was performed in io_submit_one().
+		 */
+		if (is_in_aio() && was_page_ok) {
+			struct block_device *bdev = (inode &&
+						inode->i_sb->s_bdev) ?
+						inode->i_sb->s_bdev : NULL;
+			cgroup_io_throttle(bdev, -nr, 0);
+		}
+
 		/* If users can be writing to this page using arbitrary
 		 * virtual addresses, take care about potential aliasing
 		 * before reading the page on the kernel side.
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 29b1d1e..c6207de 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -23,6 +23,7 @@
 #include <linux/init.h>
 #include <linux/backing-dev.h>
 #include <linux/task_io_accounting_ops.h>
+#include <linux/blk-io-throttle.h>
 #include <linux/blkdev.h>
 #include <linux/mpage.h>
 #include <linux/rmap.h>
@@ -430,6 +431,9 @@ static void balance_dirty_pages(struct address_space *mapping)
 	unsigned long write_chunk = sync_writeback_pages();
 
 	struct backing_dev_info *bdi = mapping->backing_dev_info;
+	struct block_device *bdev = (mapping->host &&
+					mapping->host->i_sb->s_bdev) ?
+					mapping->host->i_sb->s_bdev : NULL;
 
 	for (;;) {
 		struct writeback_control wbc = {
@@ -512,6 +516,14 @@ static void balance_dirty_pages(struct address_space *mapping)
 		return;		/* pdflush is already working this queue */
 
 	/*
+	 * Apply the cgroup i/o throttling limitations. The accounting of write
+	 * activity in page cache is performed in __set_page_dirty(), but since
+	 * we cannot sleep there, 0 bytes are accounted here and the functi

Documentation of the block device I/O bandwidth controller: description, usage,
advantages and design.

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 Documentation/controllers/io-throttle.txt |  282 +++++++++++++++++++++++++++++
 1 files changed, 282 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/controllers/io-throttle.txt

diff --git a/Documentation/controllers/io-throttle.txt b/Documentation/controllers/io-throttle.txt
new file mode 100644
index 0000000..ab33633
--- /dev/null
+++ b/Documentation/controllers/io-throttle.txt
@@ -0,0 +1,282 @@
+
+               Block device I/O bandwidth controller
+
+1. Description
+
+This controller allows to limit the I/O bandwidth of specific block devices for
+specific process containers (cgroups) imposing additional delays on I/O
+requests for those processes that exceed the limits defined in the control
+group filesystem.
+
+Bandwidth limiting rules offer better control over QoS with respect to priority
+or weight-based solutions that only give information about applications'
+relative performance requirements. Nevertheless, priority based solutions are
+affected by performance bursts, when only low-priority requests are submitted
+to a general purpose resource dispatcher.
+
+The goal of the I/O bandwidth controller is to improve performance
+predictability and provide performance isolation of different control groups
+sharing the same block devices.
+
+NOTE #1: If you're looking for a way to improve the overall throughput of the
+system probably you should use a different solution.
+
+NOTE #2: The current implementation does not guarantee minimum bandwidth
+levels, the QoS is implemented only slowing down I/O "traffic" that exceeds the
+limits specified by the user; minimum I/O rate thresholds are supposed to be
+guaranteed if the user configures a proper I/O bandwidth partitioning of the
+block devices shared among the different cgroups (theoretically if the sum of
+all the single limits defined for a block device doesn't exceed the total I/O
+bandwidth of that device).
+
+2. User Interface
+
+A new I/O bandwidth limitation rule is described using the file
+blockio.bandwidth.
+
+The same file can be used to set multiple rules for different block devices
+relative to the same cgroup.
+
+2.1. Configure I/O limiting rules
+
+The syntax to configure a limiting rule is the following:
+
+# /bin/echo DEV:BW:STRATEGY:BUCKET_SIZE > CGROUP/blockio.bandwidth
+
+- DEV is the name of the device the limiting rule is applied to.
+
+- BW is the maximum I/O bandwidth on DEVICE allowed by CGROUP; bandwidth must
+  be expressed in bytes/s. A generic I/O bandwidth limiting rule for a block
+  device DEV can be removed setting the BW value to 0.
+
+- STRATEGY is the throttling strategy used to throttle the applications' I/O
+  requests from/to device DEV. At the moment two different strategies can be
+  used:
+
+  0 = leaky bucket: the controller accepts at most B bytes (B = BW * time);
+		    further I/O requests are delayed scheduling a timeout for
+                    the tasks that made those requests.
+
+            Different I/O flow
+               | | |
+               | v |
+               |   v
+               v
+              .......
+              \     /
+               \   /  leaky-bucket
+                ---
+                |||
+                vvv
+             Smoothed I/O flow
+
+  1 = token bucket: BW tokens are added to the bucket every seconds; the bucket
+		    can hold at the most BUCKET_SIZE tokens; I/O requests are
+		    accepted if there are available tokens in the bucket; when
+		    a request of N bytes arrives N tokens are removed from the
+		    bucket; if fewer than N tokens are available the request is
+		    delayed until a sufficient amount of token is available in
+                    the bucket.
+
+            Tokens (I/O rate)
+                o
+                o
+                o
+              ....... <--.
+              \     /    | Bucket size (burst limit)
+               \ooo/     |
+                ---   <--'
+                 |ooo
+    Incoming --->|---> Conforming
+    I/O          |oo   I/O
+    requests  -->|-->  requests
+                 |
+            ---->|
+
+  Leaky bucket is more precise than token bucket to respect the bandwidth
+  limits, because bursty workloads are always smoothed. Token bucket, instead,
+  allows a small irregularity degree in the I/O flows (burst limit), and, for
+  this, it is better in terms of efficiency (bursty workloads are not smoothed
+  when there are sufficient tokens in the bucket).
+
+- BUCKET_SIZE is used only with token bucket (STRATEGY == 1) and defines the
+  size of the bucket in bytes.
+
+- CGROUP is the name of the limited process container.
+
+Also the following syntaxes are allowed:
+
+- remove an I/O bandwidth limiting rule
+# /bin/echo DEV:0 > CGROUP/blockio.bandwidth
+
+- configure a limiting rule using leaky bucket throttling (ignore bucket size):
+# /bin/echo DEV:BW:0 > CGROUP/blockio.bandwidth
+
+2.2. Show I/O limiting rules
+
+All the defined rules and statistics for a specific cgroup can be shown reading
+the file blockio.bandwidth. The following syntax is used:
+
+$ cat CGROUP/blockio.bandwidth
+MAJOR MINOR BW STRATEGY LEAKY_STAT BUCKET_SIZE BUCKET_FILL TIME_DELTA
+
+- MAJOR is the major device number of DEV (defined above)
+
+- MINOR is the minor device number of DEV (defined above)
+
+- BW, STRATEGY and BUCKET_SIZE are the same parameters defined above
+
+- LEAKY_STAT is the amount of bytes currently allowed by the I/O bandwidth
+  controller (only used with leaky bucket strategy - STRATEGY == 0)
+
+- BUCKET_FILL represents the amount of tokens present in the bucket (only used
+  with token bucket strategy - STRATEGY == 1)
+
+- TIME_DELTA can be one of the following:
+  - the amount of jiffies elapsed from the last I/O request (token bucket)
+  - the amount of jiffies during which the bytes given by LEAKY_STAT have been
+    accumulated (leaky bucket)
+
+Multiple per-block device rules are reported in multiple rows
+(DEVi, i = 1 ..  n):
+
+$ cat CGROUP/blockio.bandwidth
+MAJOR1 MINOR1 BW1 STRATEGY1 LEAKY_STAT1 BUCKET_SIZE1 BUCKET_FILL1 TIME_DELTA1
+MAJOR1 MINOR1 BW2 STRATEGY2 LEAKY_STAT2 BUCKET_SIZE2 BUCKET_FILL2 TIME_DELTA2
+...
+MAJORn MINORn BWn STRATEGYn LEAKY_STATn BUCKET_SIZEn BUCKET_FILLn TIME_DELTAn
+
+2.3. Examples
+
+* Mount the cgroup filesystem (blockio subsystem):
+  # mkdir /mnt/cgroup
+  # mount -t cgroup -oblockio blockio /mnt/cgroup
+
+* Instantiate the new cgroup "foo":
+  # mkdir /mnt/cgroup/foo
+  --> the cgroup foo has been created
+
+* Add the current shell process to the cgroup "foo":
+  # /bin/echo $$ > /mnt/cgroup/foo/tasks
+  --> the current shell has been added to the cgroup "foo"
+
+* Give maximum 1MiB/s of I/O bandwidth on /dev/sda for the cgroup "foo", using
+  leaky bucket throttling strategy:
+  # /bin/echo /dev/sda:$((1024 * 1024)):0:0 > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # sh
+  --> the subshell 'sh' is running in cgroup "foo" and it can use a maximum I/O
+      bandwidth of 1MiB/s on /dev/sda
+
+* Give maximum 8MiB/s of I/O bandwidth on /dev/sdb for the cgroup "foo", using
+  token bucket throttling strategy, bucket size = 8MB:
+  # /bin/echo /dev/sdb:$((8 * 1024 * 1024)):1:$((8 * 1024 * 1024)) > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # sh
+  --> the subshell 'sh' is running in cgroup "foo" and it can use a maximum I/O
+      bandwidth of 1MiB/s on /dev/sda (controlled by leaky bucket throttling)
+      and 8MiB/s on /dev/sdb (controlled by token bucket throttling)
+
+* Run a benchmark doing I/O on /dev/sda and /dev/sdb; I/O limits and usage
+  defined for cgroup "foo" can be shown as following:
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 16 8388608 1 0 8388608 -522560 48
+  8 0 1048576 0 737280 0 0 216
+
+* Extend the maximum I/O bandwidth for the cgroup "foo" to 16MiB/s on /dev/sda:
+  # /bin/echo /dev/sda:$((16 * 1024 * 1024)):0:0 > \
+  > /mnt/cgroup/foo/blockio.bandwidth
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 16 8388608 1 0 8388608 -84432 206436
+  8 0 16777216 0 0 0 0 15212
+
+* Remove limiting rule on /dev/sdb for cgroup "foo":
+  # /bin/echo /dev/sdb:0:0:0 > /mnt/cgroup/foo/blockio.bandwidth
+  # cat /mnt/cgroup/foo/blockio.bandwidth
+  8 0 16777216 0 0 0 0 110388
+
+3. Advantages of providing this feature
+
+* Allow I/O traffic shaping for block device shared among different cgroups
+* Improve I/O performance predictability on block devices shared between
+  different cgroups
+* Limiting rules do not depend of the particular I/O scheduler (anticipatory,
+  deadline, CFQ, noop) and/or the type of the underlying block devices
+* The bandwidth limitations are guaranteed both for synchronous and
+  asynchronous operations, even the I/O passing through the page cache or
+  buffers and not only direct I/O (see below for details)
+* It is possible to implement a simple user-space application to dynamically
+  adjust the I/O workload of different process containers at run-time,
+  according to the particular users' requirements and applications' performance
+  constraints
+* It is even possible to implement event-based performance throttling
+  mechanisms; for example the same user-space application could actively
+  throttle the I/O bandwidth to reduce power consumption when the battery of a
+  mobile device is running low (power throttling) or when the temperature of a
+  hardware component is too high (thermal throttling)
+
+4. Design
+
+The I/O throttling is performed imposing an explicit timeout, via
+schedule_timeout_killable() on the processes that exceed the I/O bandwidth
+dedicated to the cgroup they belong to. I/O accounting happens per cgroup.
+
+It just works as expected for read operations: the 

The objective of the i/o bandwidth controller is to improve i/o performance
predictability of different cgroups sharing the same block devices.

Respect to other priority/weight-based solutions the approach used by this
controller is to explicitly choke applications' requests that directly (or
indirectly) generate i/o activity in the system.

The direct bandwidth limiting method has the advantage of improving the
performance predictability at the cost of reducing, in general, the overall
performance of the system (in terms of throughput).

Detailed informations about design, its goal and usage are described in the
documentation.

Tested against 2.6.26-rc8-mm1.

The all-in-one patch (and previous versions) can be found at:
http://download.systemimager.org/~arighi/linux/patches/io-throttle/

Thanks to Li Zefan for reviewing.

Changelog: (v4 -> v5)
  - rebase to 2.6.26-rc8-mm1
  - handle AIO opportunely: return -EAGAIN from sys_io_submit(), instead of
    making to sleep tasks doing AIO
  - userspace=>kernel interface now accepts the following syntaxes:
    * dev:0                       <- to delete a limiting rule
    * dev:bw-limit:0              <- define a leaky bucket throttling rule
    * dev:bw-limit:1:bucket-size  <- define a token bucket throttling rule
  - use .write_string and .read_seq_string to simplify iothrottle_read() and
    iothrottle_write() functions
  - use a enum structure to enumerate the various throttling algorithms

TODO:
  - see documentation

-Andrea
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers