Containers: Pagecache accounting and control subsystem (v1)

This patch adds pagecache accounting and control on top of Paul's container subsystem v7 posted at
http://lkml.org/lkml/2007/2/12/88

Comments, suggestions and criticisms are welcome.

Features:
--------
* New subsystem called 'pagecache_acct' is registered with containers
* Container pointer is added to struct address_space to keep track of associated container
* In filemap.c and swap_state.c, the corresponding container's pagecache_acct subsystem is charged and uncharged whenever a new page is added or removed from pagecache
* The accounting number include pages in swap cache and filesystem buffer pages apart from pagecache, basically everything under NR_FILE_PAGES is counted as pagecache. However this excluded mapped and anonymous pages
* Limits on pagecache can be set by echo 100000 > pagecache_limit on the /container file system. The unit is in kilobytes
* If the pagecache utilisation limit is exceeded, pagecache reclaim code is invoked to recover dirty and clean pagecache pages only.

Advantages:
-----------
* Does not add container pointers in struct page

Limitations:
-----------
* Code is not safe for container deletion/task migration
* Pagecache page reclaim needs performance improvements
* Global LRU is churned in search of pagecache pages

Usage:
-----

* Add patch on top of Paul container (v7) at kernel version 2.6.20
* Enable CONFIG_CONTAINER_PAGECACHE_ACCT in 'General Setup'
* Boot new kernel
* Mount container filesystem
mount -t container /container
cd /container
* Create new container
mkdir mybox
cd /container/mybox
* Add current shell to container
echo $$ > tasks
* There are two files pagecache_usage and pagecache_limit
* In order to set limit, echo value in kilobytes to pagecache_limit

	echo 100000 > pagecache_limit 
	#This would set 100MB limit on pagecache usage

* Trash the system from current shell using scp/cp/dd/tar etc
* Watch pagecache_usage and /proc/meminfo to verify behavior

* Only unmapped pagecache data will be accounted and controlled. These are memory used by cp, scp, tar etc. While file mmap will be controlled by Balbir's RSS controller.

ToDo:
----

* Merge with container RSS controller and eliminate redundant code
* Support and test task migration and container deletion
* Review reclaim performance
* Optimise page reclaim

---
 include/linux/fs.h             |    6 
 include/linux/pagecache_acct.h |   52 +++++
 init/Kconfig                   |    7 
 kernel/Makefile                |    1 
 kernel/pagecache_acct.c        |  376 +++++++++++++++++++++++++++++++++++++++++
 mm/filemap.c                   |    8 
 mm/vmscan.c                    |   76 ++++++++
 7 files changed, 524 insertions(+), 2 deletions(-)

@@ -447,6 +447,12 @@ struct address_space {
 	spinlock_t		private_lock;	/* for use by the address_space */
 	struct list_head	private_list;	/* ditto */
 	struct address_space	*assoc_mapping;	/* ditto */
+
+#ifdef CONFIG_CONTAINER_PAGECACHE_ACCT
+	struct container *container; 	/* Charge page to the right container
+					   using page->mapping */
+#endif
+
 } __attribute__((aligned(sizeof(long))));
 	/*
 	 * On most architectures that alignment is already the case; but

@@ -306,6 +306,13 @@ config CONTAINER_NS
           for instance virtual servers and checkpoint/restart
           jobs.
 
+config CONTAINER_PAGECACHE_ACCT
+	bool "Simple PageCache accounting container subsystem"
+	select CONTAINERS
+	help
+	  Provides a simple Resource Controller for monitoring the
+	  total pagecache consumed by the tasks in a container
+
 config RELAY
 	bool "Kernel->user space relay support (formerly relayfs)"
 	help

@@ -40,6 +40,7 @@ obj-$(CONFIG_CONTAINERS) += container.o
 obj-$(CONFIG_CPUSETS) += cpuset.o
 obj-$(CONFIG_CONTAINER_CPUACCT) += cpu_acct.o
 obj-$(CONFIG_CONTAINER_NS) += ns_container.o
+obj-$(CONFIG_CONTAINER_PAGECACHE_ACCT) += pagecache_acct.o
 obj-$(CONFIG_IKCONFIG) += configs.o
 obj-$(CONFIG_STOP_MACHINE) += stop_machine.o
 obj-$(CONFIG_AUDIT) += audit.o auditfilter.o

@@ -30,6 +30,7 @@
 #include <linux/security.h>
 #include <linux/syscalls.h>
 #include <linux/cpuset.h>
+#include <linux/pagecache_acct.h>
 #include "filemap.h"
 #include "internal.h"
 

@@ -117,6 +118,8 @@ void __remove_from_page_cache(struct pag
 	struct address_space *mapping = page->mapping;
 
 	radix_tree_delete(&mapping->page_tree, page->index);
+	/* Uncharge before the mapping is gone */
+	pagecache_acct_uncharge(page);
 	page->mapping = NULL;
 	mapping->nrpages--;
 	__dec_zone_page_state(page, NR_FILE_PAGES);

@@ -451,6 +454,11 @@ int add_to_page_cache(struct page *page,
 			__inc_zone_page_state(page, NR_FILE_PAGES);
 		}
 		write_unlock_irq(&mapping->tree_lock);
+		/* Unlock before charge, because we may reclaim this inline */
+		if (!error) {
+			pagecache_acct_init_page_ptr(page);
+			pagecache_acct_charge(page);
+		}
 		radix_tree_preload_end();
 	}
 	return error;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>