lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230627172216.1359-1-demi@invisiblethingslab.com>
Date:   Tue, 27 Jun 2023 13:22:14 -0400
From:   Demi Marie Obenour <demi@...isiblethingslab.com>
To:     Juergen Gross <jgross@...e.com>,
        Stefano Stabellini <sstabellini@...nel.org>,
        Oleksandr Tyshchenko <oleksandr_tyshchenko@...m.com>,
        Jan Beulich <JBeulich@...e.com>,
        Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>
Cc:     Demi Marie Obenour <demi@...isiblethingslab.com>,
        Xen developer discussion <xen-devel@...ts.xenproject.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        stable@...r.kernel.org
Subject: [PATCH v3] xen: speed up grant-table reclaim

When a grant entry is still in use by the remote domain, Linux must put
it on a deferred list.  Normally, this list is very short, because
the PV network and block protocols expect the backend to unmap the grant
first.  However, Qubes OS's GUI protocol is subject to the constraints
of the X Window System, and as such winds up with the frontend unmapping
the window first.  As a result, the list can grow very large, resulting
in a massive memory leak and eventual VM freeze.

To partially solve this problem, make the number of entries that the VM
will attempt to free at each iteration tunable.  The default is still
10, but it can be overridden at compile-time (via Kconfig), boot-time
(via a kernel command-line option), or runtime (via sysfs).

This is Cc: stable because (when combined with appropriate userspace
changes) it fixes a severe performance and stability problem for Qubes
OS users.

Cc: stable@...r.kernel.org
Signed-off-by: Demi Marie Obenour <demi@...isiblethingslab.com>
---
 drivers/xen/grant-table.c | 40 ++++++++++++++++++++++++++++-----------
 1 file changed, 29 insertions(+), 11 deletions(-)

Changes since v2:
- use atomic_inc_return(x) and atomic_dec_return(x) instead of
  atomic_add_return(1, x) and atomic_sub_return(1, x) respectively.
- move module_param macro closer to the definition of
  free_per_iteration.
- add blank line between declarations and statements.

Changes since v1:
- drop setting default via Kconfig

diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c
index e1ec725c2819d4d5dede063eb00d86a6d52944c0..f13c3b76ad1eb7110e2a2981e9fa4e504174e431 100644
--- a/drivers/xen/grant-table.c
+++ b/drivers/xen/grant-table.c
@@ -498,14 +498,21 @@ static LIST_HEAD(deferred_list);
 static void gnttab_handle_deferred(struct timer_list *);
 static DEFINE_TIMER(deferred_timer, gnttab_handle_deferred);
 
+static atomic64_t deferred_count;
+static atomic64_t leaked_count;
+static unsigned int free_per_iteration = 10;
+module_param(free_per_iteration, uint, 0600);
+
 static void gnttab_handle_deferred(struct timer_list *unused)
 {
-	unsigned int nr = 10;
+	unsigned int nr = READ_ONCE(free_per_iteration);
+	const bool ignore_limit = nr == 0;
 	struct deferred_entry *first = NULL;
 	unsigned long flags;
+	size_t freed = 0;
 
 	spin_lock_irqsave(&gnttab_list_lock, flags);
-	while (nr--) {
+	while ((ignore_limit || nr--) && !list_empty(&deferred_list)) {
 		struct deferred_entry *entry
 			= list_first_entry(&deferred_list,
 					   struct deferred_entry, list);
@@ -515,10 +522,14 @@ static void gnttab_handle_deferred(struct timer_list *unused)
 		list_del(&entry->list);
 		spin_unlock_irqrestore(&gnttab_list_lock, flags);
 		if (_gnttab_end_foreign_access_ref(entry->ref)) {
+			uint64_t ret = atomic64_dec_return(&deferred_count);
+
 			put_free_entry(entry->ref);
-			pr_debug("freeing g.e. %#x (pfn %#lx)\n",
-				 entry->ref, page_to_pfn(entry->page));
+			pr_debug("freeing g.e. %#x (pfn %#lx), %llu remaining\n",
+				 entry->ref, page_to_pfn(entry->page),
+				 (unsigned long long)ret);
 			put_page(entry->page);
+			freed++;
 			kfree(entry);
 			entry = NULL;
 		} else {
@@ -530,21 +541,22 @@ static void gnttab_handle_deferred(struct timer_list *unused)
 		spin_lock_irqsave(&gnttab_list_lock, flags);
 		if (entry)
 			list_add_tail(&entry->list, &deferred_list);
-		else if (list_empty(&deferred_list))
-			break;
 	}
-	if (!list_empty(&deferred_list) && !timer_pending(&deferred_timer)) {
+	if (list_empty(&deferred_list))
+		WARN_ON(atomic64_read(&deferred_count));
+	else if (!timer_pending(&deferred_timer)) {
 		deferred_timer.expires = jiffies + HZ;
 		add_timer(&deferred_timer);
 	}
 	spin_unlock_irqrestore(&gnttab_list_lock, flags);
+	pr_debug("Freed %zu references", freed);
 }
 
 static void gnttab_add_deferred(grant_ref_t ref, struct page *page)
 {
 	struct deferred_entry *entry;
 	gfp_t gfp = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
-	const char *what = KERN_WARNING "leaking";
+	uint64_t leaked, deferred;
 
 	entry = kmalloc(sizeof(*entry), gfp);
 	if (!page) {
@@ -567,10 +579,16 @@ static void gnttab_add_deferred(grant_ref_t ref, struct page *page)
 			add_timer(&deferred_timer);
 		}
 		spin_unlock_irqrestore(&gnttab_list_lock, flags);
-		what = KERN_DEBUG "deferring";
+		deferred = atomic64_inc_return(&deferred_count);
+		leaked = atomic64_read(&leaked_count);
+		pr_debug("deferring g.e. %#x (pfn %#lx) (total deferred %llu, total leaked %llu)\n",
+			 ref, page ? page_to_pfn(page) : -1, deferred, leaked);
+	} else {
+		deferred = atomic64_read(&deferred_count);
+		leaked = atomic64_inc_return(&leaked_count);
+		pr_warn("leaking g.e. %#x (pfn %#lx) (total deferred %llu, total leaked %llu)\n",
+			ref, page ? page_to_pfn(page) : -1, deferred, leaked);
 	}
-	printk("%s g.e. %#x (pfn %#lx)\n",
-	       what, ref, page ? page_to_pfn(page) : -1);
 }
 
 int gnttab_try_end_foreign_access(grant_ref_t ref)
-- 
Sincerely,
Demi Marie Obenour (she/her/hers)
Invisible Things Lab

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ