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-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20220915002235.v2.2.I8db2637ed3b9abc3961f3bc41ec25fa8b8ae6624@changeid>
Date:   Thu, 15 Sep 2022 00:22:54 +0000
From:   Sven van Ashbrook <svenva@...omium.org>
To:     LKML <linux-kernel@...r.kernel.org>,
        Herbert Xu <herbert@...dor.apana.org.au>,
        Dominik Brodowski <linux@...inikbrodowski.net>,
        Olivia Mackall <olivia@...enic.com>
Cc:     Alex Levin <levinale@...gle.com>,
        Andrey Pronin <apronin@...gle.com>,
        Stephen Boyd <swboyd@...gle.com>,
        Rajat Jain <rajatja@...gle.com>,
        Sven van Ashbrook <svenva@...omium.org>,
        Eric Biggers <ebiggers@...gle.com>,
        "Jason A. Donenfeld" <Jason@...c4.com>,
        linux-crypto@...r.kernel.org
Subject: [PATCH v2 2/2] hwrng: core: fix suspend/resume race condition

The hwrng fill function runs as a normal kthread. This thread
doesn't get frozen by the PM, i.e. it will keep running during,
or in, system suspend. It may call the client driver's
data_present()/data_read() functions during, or in, suspend;
which may generate errors or warnings. For example, if the
client driver uses an i2c bus, the following warning may be
intermittently generated:

  i2c: Transfer while suspended

Fix by converting the delay polled kthread into an ordered work
queue running a single, self-rearming delayed_work. Make the
workqueue WQ_FREEZABLE, so the PM will drain any work items
before going into suspend. This prevents client drivers from
being accessed during, or in, suspend.

Tested on a Chromebook containing an cr50 tpm over i2c. The test
consists of 31000 suspend/resume cycles. Occasional
"i2c: Transfer while suspended" warnings are seen. After applying
this patch, these warnings disappear.

This patch also does not appear to cause any regressions on the
ChromeOS test queues.

Signed-off-by: Sven van Ashbrook <svenva@...omium.org>
---

Changes in v2:
- refreshed patch set against latest tree

 drivers/char/hw_random/core.c | 95 +++++++++++++++++++----------------
 1 file changed, 51 insertions(+), 44 deletions(-)

diff --git a/drivers/char/hw_random/core.c b/drivers/char/hw_random/core.c
index 3675122c6cce..ee85ca97d215 100644
--- a/drivers/char/hw_random/core.c
+++ b/drivers/char/hw_random/core.c
@@ -17,7 +17,7 @@
 #include <linux/hw_random.h>
 #include <linux/random.h>
 #include <linux/kernel.h>
-#include <linux/kthread.h>
+#include <linux/workqueue.h>
 #include <linux/sched/signal.h>
 #include <linux/miscdevice.h>
 #include <linux/module.h>
@@ -28,14 +28,17 @@
 
 #define RNG_MODULE_NAME		"hw_random"
 
-static struct hwrng *current_rng;
 /* the current rng has been explicitly chosen by user via sysfs */
 static int cur_rng_set_by_user;
-static struct task_struct *hwrng_fill;
+static struct workqueue_struct *hwrng_wq;
+static struct delayed_work hwrng_fill_dwork;
+static size_t entropy_credit;
+/* Protects rng_list, current_rng, is_hwrng_wq_running */
+static DEFINE_MUTEX(rng_mutex);
 /* list of registered rngs */
 static LIST_HEAD(rng_list);
-/* Protects rng_list and current_rng */
-static DEFINE_MUTEX(rng_mutex);
+static struct hwrng *current_rng;
+static bool is_hwrng_wq_running;
 /* Protects rng read functions, data_avail, rng_buffer and rng_fillbuf */
 static DEFINE_MUTEX(reading_mutex);
 static int data_avail;
@@ -488,37 +491,29 @@ static int __init register_miscdev(void)
 	return misc_register(&rng_miscdev);
 }
 
-static int hwrng_fillfn(void *unused)
+static void hwrng_fillfn(struct work_struct *unused)
 {
-	size_t entropy, entropy_credit = 0; /* in 1/1024 of a bit */
+	unsigned short quality;
 	unsigned long delay;
+	struct hwrng *rng;
+	size_t entropy; /* in 1/1024 of a bit */
 	long rc;
 
-	while (!kthread_should_stop()) {
-		unsigned short quality;
-		struct hwrng *rng;
-
-		rng = get_current_rng();
-		if (IS_ERR(rng) || !rng)
-			break;
-		mutex_lock(&reading_mutex);
-		rc = rng_get_data(rng, rng_fillbuf,
-				  rng_buffer_size(), 1);
-		if (current_quality != rng->quality)
-			rng->quality = current_quality; /* obsolete */
-		quality = rng->quality;
-		mutex_unlock(&reading_mutex);
-		put_rng(rng);
-
-		if (!quality)
-			break;
+	rng = get_current_rng();
+	if (IS_ERR(rng) || !rng)
+		return;
+	mutex_lock(&reading_mutex);
+	rc = rng_get_data(rng, rng_fillbuf, rng_buffer_size(), 1);
+	if (current_quality != rng->quality)
+		rng->quality = current_quality; /* obsolete */
+	quality = rng->quality;
+	mutex_unlock(&reading_mutex);
+	put_rng(rng);
 
-		if (rc <= 0) {
-			pr_warn("hwrng: no data available\n");
-			msleep_interruptible(10000);
-			continue;
-		}
+	if (!quality)
+		return;
 
+	if (rc > 0) {
 		/* If we cannot credit at least one bit of entropy,
 		 * keep track of the remainder for the next iteration
 		 */
@@ -529,11 +524,11 @@ static int hwrng_fillfn(void *unused)
 		/* Outside lock, sure, but y'know: randomness. */
 		delay = add_hwgenerator_randomness((void *)rng_fillbuf, rc,
 						   entropy >> 10);
-		if (delay > 0)
-			schedule_timeout_interruptible(delay);
+	} else {
+		pr_warn("hwrng: no data available\n");
+		delay = 10 * HZ;
 	}
-	hwrng_fill = NULL;
-	return 0;
+	mod_delayed_work(hwrng_wq, &hwrng_fill_dwork, delay);
 }
 
 static void hwrng_manage_rngd(struct hwrng *rng)
@@ -541,14 +536,12 @@ static void hwrng_manage_rngd(struct hwrng *rng)
 	if (WARN_ON(!mutex_is_locked(&rng_mutex)))
 		return;
 
-	if (rng->quality == 0 && hwrng_fill)
-		kthread_stop(hwrng_fill);
-	if (rng->quality > 0 && !hwrng_fill) {
-		hwrng_fill = kthread_run(hwrng_fillfn, NULL, "hwrng");
-		if (IS_ERR(hwrng_fill)) {
-			pr_err("hwrng_fill thread creation failed\n");
-			hwrng_fill = NULL;
-		}
+	if (rng->quality == 0 && is_hwrng_wq_running) {
+		cancel_delayed_work(&hwrng_fill_dwork);
+		is_hwrng_wq_running = false;
+	} else if (rng->quality > 0 && !is_hwrng_wq_running) {
+		mod_delayed_work(hwrng_wq, &hwrng_fill_dwork, 0);
+		is_hwrng_wq_running = true;
 	}
 }
 
@@ -631,8 +624,7 @@ void hwrng_unregister(struct hwrng *rng)
 	new_rng = get_current_rng_nolock();
 	if (list_empty(&rng_list)) {
 		mutex_unlock(&rng_mutex);
-		if (hwrng_fill)
-			kthread_stop(hwrng_fill);
+		cancel_delayed_work_sync(&hwrng_fill_dwork);
 	} else
 		mutex_unlock(&rng_mutex);
 
@@ -703,17 +695,32 @@ static int __init hwrng_modinit(void)
 		return -ENOMEM;
 	}
 
+	/* ordered wq to mimic delay-polled kthread behaviour */
+	hwrng_wq = alloc_ordered_workqueue("hwrng",
+		WQ_FREEZABLE |	/* prevent work from running during suspend/resume */
+		WQ_MEM_RECLAIM	/* client drivers may need memory reclaim */
+	);
+	if (!hwrng_wq) {
+		kfree(rng_fillbuf);
+		kfree(rng_buffer);
+		return -ENOMEM;
+	}
+
 	ret = register_miscdev();
 	if (ret) {
+		destroy_workqueue(hwrng_wq);
 		kfree(rng_fillbuf);
 		kfree(rng_buffer);
 	}
 
+	INIT_DELAYED_WORK(&hwrng_fill_dwork, hwrng_fillfn);
+
 	return ret;
 }
 
 static void __exit hwrng_modexit(void)
 {
+	destroy_workqueue(hwrng_wq);
 	mutex_lock(&rng_mutex);
 	BUG_ON(current_rng);
 	kfree(rng_buffer);
-- 
2.37.2.789.g6183377224-goog

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ