epoll: introduce resource usage limits

It has been thought that the per-user file descriptors limit would also
limit the resources that a normal user can request via the epoll
interface. Vegard Nossum reported a very simple program (a modified
version attached) that can make a normal user to request a pretty large
amount of kernel memory, well within the its maximum number of fds. To
solve such problem, default limits are now imposed, and /proc based
configuration has been introduced. A new directory has been created,
named /proc/sys/fs/epoll/ and inside there, there are two configuration
points:

max_user_instances = Maximum number of devices - per user

max_user_watches = Maximum number of "watched" fds - per user

The current default for "max_user_watches" limits the memory used by epoll
to store "watches", to 1/32 of the amount of the low RAM. As example, a
256MB 32bit machine, will have "max_user_watches" set to roughly 90000.
That should be enough to not break existing heavy epoll users. The
default value for "max_user_instances" is set to 128, that should be
enough too.

This also changes the userspace, because a new error code can now come out
from EPOLL_CTL_ADD (-ENOSPC). The EMFILE from epoll_create() was already
listed, so that should be ok.

[[e-mail]: use get_current_user()]
Signed-off-by: Davide Libenzi <[e-mail]>
Cc: Michael Kerrisk <[e-mail]>
Cc: <[e-mail]>
Cc: Cyrill Gorcunov <[e-mail]>
Reported-by: Vegard Nossum <[e-mail]>
Signed-off-by: Andrew Morton <[e-mail]>
Signed-off-by: Linus Torvalds <[e-mail]>

Documentation/filesystems/proc.txt

···
   2.14	/proc/<pid>/io - Display the IO accounting fields
   2.15	/proc/<pid>/coredump_filter - Core dump filtering settings
   2.16	/proc/<pid>/mountinfo - Information about mounts
+  2.17	/proc/sys/fs/epoll - Configuration options for the epoll interface
 
 ------------------------------------------------------------------------------
 Preface
···
 
   Documentation/filesystems/sharedsubtree.txt
 
+2.17	/proc/sys/fs/epoll - Configuration options for the epoll interface
+--------------------------------------------------------
+
+This directory contains configuration options for the epoll(7) interface.
+
+max_user_instances
+------------------
+
+This is the maximum number of epoll file descriptors that a single user can
+have open at a given time. The default value is 128, and should be enough
+for normal users.
+
+max_user_watches
+----------------
+
+Every epoll file descriptor can store a number of files to be monitored
+for event readiness. Each one of these monitored files constitutes a "watch".
+This configuration option sets the maximum number of "watches" that are
+allowed for each user.
+Each "watch" costs roughly 90 bytes on a 32bit kernel, and roughly 160 bytes
+on a 64bit one.
+The current default value for  max_user_watches  is the 1/32 of the available
+low memory, divided for the "watch" cost in bytes.
+
+
 ------------------------------------------------------------------------------
+

fs/eventpoll.c

···
 
 #define EP_UNACTIVE_PTR ((void *) -1L)
 
+#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
+
 struct epoll_filefd {
 	struct file *file;
 	int fd;
···
 	 * holding ->lock.
 	 */
 	struct epitem *ovflist;
+
+	/* The user that created the eventpoll descriptor */
+	struct user_struct *user;
 };
 
 /* Wait structure used by the poll hooks */
···
 };
 
 /*
+ * Configuration options available inside /proc/sys/fs/epoll/
+ */
+/* Maximum number of epoll devices, per user */
+static int max_user_instances __read_mostly;
+/* Maximum number of epoll watched descriptors, per user */
+static int max_user_watches __read_mostly;
+
+/*
  * This mutex is used to serialize ep_free() and eventpoll_release_file().
  */
-static struct mutex epmutex;
+static DEFINE_MUTEX(epmutex);
 
 /* Safe wake up implementation */
 static struct poll_safewake psw;
···
 
 /* Slab cache used to allocate "struct eppoll_entry" */
 static struct kmem_cache *pwq_cache __read_mostly;
+
+#ifdef CONFIG_SYSCTL
+
+#include <linux/sysctl.h>
+
+static int zero;
+
+ctl_table epoll_table[] = {
+	{
+		.procname	= "max_user_instances",
+		.data		= &max_user_instances,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= &zero,
+	},
+	{
+		.procname	= "max_user_watches",
+		.data		= &max_user_watches,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= &zero,
+	},
+	{ .ctl_name = 0 }
+};
+#endif /* CONFIG_SYSCTL */
 
 
 /* Setup the structure that is used as key for the RB tree */
···
 	/* At this point it is safe to free the eventpoll item */
 	kmem_cache_free(epi_cache, epi);
 
+	atomic_dec(&ep->user->epoll_watches);
+
 	DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_remove(%p, %p)\n",
 		     current, ep, file));
 
···
 
 	mutex_unlock(&epmutex);
 	mutex_destroy(&ep->mtx);
+	atomic_dec(&ep->user->epoll_devs);
+	free_uid(ep->user);
 	kfree(ep);
 }
 
···
 
 static int ep_alloc(struct eventpoll **pep)
 {
-	struct eventpoll *ep = kzalloc(sizeof(*ep), GFP_KERNEL);
+	int error;
+	struct user_struct *user;
+	struct eventpoll *ep;
 
-	if (!ep)
-		return -ENOMEM;
+	user = get_current_user();
+	error = -EMFILE;
+	if (unlikely(atomic_read(&user->epoll_devs) >=
+			max_user_instances))
+		goto free_uid;
+	error = -ENOMEM;
+	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
+	if (unlikely(!ep))
+		goto free_uid;
 
 	spin_lock_init(&ep->lock);
 	mutex_init(&ep->mtx);
···
 	INIT_LIST_HEAD(&ep->rdllist);
 	ep->rbr = RB_ROOT;
 	ep->ovflist = EP_UNACTIVE_PTR;
+	ep->user = user;
 
 	*pep = ep;
 
 	DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_alloc() ep=%p\n",
 		     current, ep));
 	return 0;
+
+free_uid:
+	free_uid(user);
+	return error;
 }
 
 /*
···
 	struct epitem *epi;
 	struct ep_pqueue epq;
 
-	error = -ENOMEM;
+	if (unlikely(atomic_read(&ep->user->epoll_watches) >=
+		     max_user_watches))
+		return -ENOSPC;
 	if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL)))
-		goto error_return;
+		return -ENOMEM;
 
 	/* Item initialization follow here ... */
 	INIT_LIST_HEAD(&epi->rdllink);
···
 	 * install process. Namely an allocation for a wait queue failed due
 	 * high memory pressure.
 	 */
+	error = -ENOMEM;
 	if (epi->nwait < 0)
 		goto error_unregister;
 
···
 
 	spin_unlock_irqrestore(&ep->lock, flags);
 
+	atomic_inc(&ep->user->epoll_watches);
+
 	/* We have to call this outside the lock */
 	if (pwake)
 		ep_poll_safewake(&psw, &ep->poll_wait);
···
 	spin_unlock_irqrestore(&ep->lock, flags);
 
 	kmem_cache_free(epi_cache, epi);
-error_return:
+
 	return error;
 }
 
···
 			      flags & O_CLOEXEC);
 	if (fd < 0)
 		ep_free(ep);
+	atomic_inc(&ep->user->epoll_devs);
 
 error_return:
 	DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n",
···
 
 static int __init eventpoll_init(void)
 {
-	mutex_init(&epmutex);
+	struct sysinfo si;
+
+	si_meminfo(&si);
+	max_user_instances = 128;
+	max_user_watches = (((si.totalram - si.totalhigh) / 32) << PAGE_SHIFT) /
+		EP_ITEM_COST;
 
 	/* Initialize the structure used to perform safe poll wait head wake ups */
 	ep_poll_safewake_init(&psw);

include/linux/sched.h

···
 	atomic_t inotify_watches; /* How many inotify watches does this user have? */
 	atomic_t inotify_devs;	/* How many inotify devs does this user have opened? */
 #endif
+#ifdef CONFIG_EPOLL
+	atomic_t epoll_devs;	/* The number of epoll descriptors currently open */
+	atomic_t epoll_watches;	/* The number of file descriptors currently watched */
+#endif
 #ifdef CONFIG_POSIX_MQUEUE
 	/* protected by mq_lock	*/
 	unsigned long mq_bytes;	/* How many bytes can be allocated to mqueue? */

kernel/sysctl.c

···
 #ifdef CONFIG_INOTIFY_USER
 extern struct ctl_table inotify_table[];
 #endif
+#ifdef CONFIG_EPOLL
+extern struct ctl_table epoll_table[];
+#endif
 
 #ifdef HAVE_ARCH_PICK_MMAP_LAYOUT
 int sysctl_legacy_va_layout;
···
 		.child		= inotify_table,
 	},
 #endif	
+#ifdef CONFIG_EPOLL
+	{
+		.procname	= "epoll",
+		.mode		= 0555,
+		.child		= epoll_table,
+	},
+#endif
 #endif
 	{
 		.ctl_name	= KERN_SETUID_DUMPABLE,