Skip to contentSkip to conversation

driver core: Fix use-after-free and double free on glue directory

PATCH Superseded Muchun Song 2 participants 3 messages Jul 18, 2019 – Jul 24, 2019 Driver core

v5 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

PATCH v5 1/1 driver core: Fix use-after-free and double free on glue directory

There is a race condition between removing glue directory and adding a new device under the glue directory. It can be reproduced in following test:

path 1: Add the child device under glue dir
device_add()

    get_device_parent()
        mutex_lock(&gdp_mutex);
        ....
        /*find parent from glue_dirs.list*/
        list_for_each_entry(k, &dev->class->p->glue_dirs.list, entry)
            if (k->parent == parent_kobj) {
                kobj = kobject_get(k);
                break;
            }
        ....
        mutex_unlock(&gdp_mutex);
        ....
    ....
    kobject_add()
        kobject_add_internal()
            create_dir()
                sysfs_create_dir_ns()
                    if (kobj->parent)
                        parent = kobj->parent->sd;
                    ....
                    kernfs_create_dir_ns(parent)
                        kernfs_new_node()
                            kernfs_get(parent)
                        ....
                        /* link in */
                        rc = kernfs_add_one(kn);
                        if (!rc)
                            return kn;

                        kernfs_put(kn)
                            ....
                            repeat:
                            kmem_cache_free(kn)
                            kn = parent;

                            if (kn) {
                                if (atomic_dec_and_test(&kn->count))
                                    goto repeat;
                            }
                        ....

path2: Remove last child device under glue dir
device_del()

    cleanup_glue_dir()
        mutex_lock(&gdp_mutex);
        if (!kobject_has_children(glue_dir))
            kobject_del(glue_dir);
        kobject_put(glue_dir);
        mutex_unlock(&gdp_mutex);

Before path2 remove last child device under glue dir, If path1 add a new device under glue dir, the glue_dir kobject reference count will be increase to 2 via kobject_get(k) in get_device_parent(). And path1 has been called kernfs_new_node(), but not call kernfs_get(parent). Meanwhile, path2 call kobject_del(glue_dir) beacause 0 is returned by kobject_has_children(). This result in glue_dir->sd is freed and it's reference count will be 0. Then path1 call kernfs_get(parent) will trigger a warning in kernfs_get()(WARN_ON(!atomic_read(&kn->count))) and increase it's reference count to 1. Because glue_dir->sd is freed by path2, the next call kernfs_add_one() by path1 will fail(This is also use-after-free) and call atomic_dec_and_test() to decrease reference count. Because the reference count is decremented to 0, it will also call kmem_cache_free() to free glue_dir->sd again. This will result in double free.

In order to avoid this happening, we also should make sure that kernfs_node for glue_dir is released in path2 only when refcount for glue_dir kobj is 1 to fix this race.

The following calltrace is captured in kernel 4.14 with the following patch applied:

commit 726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

--------------------------------------------------------------------------

[    3.633703] WARNING: CPU: 4 PID: 513 at .../fs/kernfs/dir.c:494
                Here is WARN_ON(!atomic_read(&kn->count) in kernfs_get().
....
[    3.633986] Call trace:
[    3.633991]  kernfs_create_dir_ns+0xa8/0xb0
[    3.633994]  sysfs_create_dir_ns+0x54/0xe8
[    3.634001]  kobject_add_internal+0x22c/0x3f0
[    3.634005]  kobject_add+0xe4/0x118
[    3.634011]  device_add+0x200/0x870
[    3.634017]  _request_firmware+0x958/0xc38
[    3.634020]  request_firmware_into_buf+0x4c/0x70
....
[    3.634064] kernel BUG at .../mm/slub.c:294!
                Here is BUG_ON(object == fp) in set_freepointer().
....
[    3.634346] Call trace:
[    3.634351]  kmem_cache_free+0x504/0x6b8
[    3.634355]  kernfs_put+0x14c/0x1d8
[    3.634359]  kernfs_create_dir_ns+0x88/0xb0
[    3.634362]  sysfs_create_dir_ns+0x54/0xe8
[    3.634366]  kobject_add_internal+0x22c/0x3f0
[    3.634370]  kobject_add+0xe4/0x118
[    3.634374]  device_add+0x200/0x870
[    3.634378]  _request_firmware+0x958/0xc38
[    3.634381]  request_firmware_into_buf+0x4c/0x70

--------------------------------------------------------------------------

Fixes 726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

---

Change in v5:
1. Revert to the v1 fix.
2. Add some comment to explain why we need do this in cleanup_glue_dir().
Change in v4:
1. Add some kerneldoc comment.
2. Remove unlock_if_glue_dir().
3. Rename get_device_parent_locked_if_glue_dir() to get_device_parent_locked.
4. Update commit message.
Change in v3:
Add change log.
Change in v2:
Fix device_move() also.

 drivers/base/core.c | 54 ++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 53 insertions(+), 1 deletion(-)
drivers/base/core.c +53 -1
index 4aeaa0c92bda..1a67ee325584 100644
@@ -1825,7 +1825,59 @@ static void cleanup_glue_dir(struct device *dev, struct kobject *glue_dir)
 		return;
 
 	mutex_lock(&gdp_mutex);
-	if (!kobject_has_children(glue_dir))
+	/**
+	 * There is a race condition between removing glue directory
+	 * and adding a new device under the glue directory.
+	 *
+	 * path 1: Add the child device under glue dir
+	 * device_add()
+	 *	get_device_parent()
+	 *		mutex_lock(&gdp_mutex);
+	 *		....
+	 *		list_for_each_entry(k, &dev->class->p->glue_dirs.list,
+	 *				    entry)
+	 *		if (k->parent == parent_kobj) {
+	 *			kobj = kobject_get(k);
+	 *			break;
+	 *		}
+	 *		....
+	 *		mutex_unlock(&gdp_mutex);
+	 *		....
+	 *	....
+	 *	kobject_add()
+	 *		kobject_add_internal()
+	 *		create_dir()
+	 *			....
+	 *			if (kobj->parent)
+	 *				parent = kobj->parent->sd;
+	 *			....
+	 *			kernfs_create_dir_ns(parent)
+	 *				....
+	 *
+	 * path2: Remove last child device under glue dir
+	 * device_del()
+	 *	cleanup_glue_dir()
+	 *		....
+	 *		mutex_lock(&gdp_mutex);
+	 *		if (!kobject_has_children(glue_dir))
+	 *			kobject_del(glue_dir);
+	 *		....
+	 *		mutex_unlock(&gdp_mutex);
+	 *
+	 * Before path2 remove last child device under glue dir, if path1 add
+	 * a new device under glue dir, the glue_dir kobject reference count
+	 * will be increase to 2 via kobject_get(k) in get_device_parent().
+	 * And path1 has been called kernfs_create_dir_ns(). Meanwhile,
+	 * path2 call kobject_del(glue_dir) beacause 0 is returned by
+	 * kobject_has_children(). This result in glue_dir->sd is freed.
+	 * Then the path1 will see a stale "empty" but still potentially used
+	 * glue dir around.
+	 *
+	 * In order to avoid this happening, we also should make sure that
+	 * kernfs_node for glue_dir is released in path2 only when refcount
+	 * for glue_dir kobj is 1.
+	 */
+	if (!kobject_has_children(glue_dir) && kref_read(&glue_dir->kref) == 1)
 		kobject_del(glue_dir);
 	kobject_put(glue_dir);
 	mutex_unlock(&gdp_mutex);
Signature · 1 line
2.17.1
Reviewed-by on v5 1/1
#1 lore raw
Quoted 181 lines · Muchun Song

There is a race condition between removing glue directory and adding a new device under the glue directory. It can be reproduced in following test:

path 1: Add the child device under glue dir
device_add()

     get_device_parent()
         mutex_lock(&gdp_mutex);
         ....
         /*find parent from glue_dirs.list*/
         list_for_each_entry(k, &dev->class->p->glue_dirs.list, entry)
             if (k->parent == parent_kobj) {
                 kobj = kobject_get(k);
                 break;
             }
         ....
         mutex_unlock(&gdp_mutex);
         ....
     ....
     kobject_add()
         kobject_add_internal()
             create_dir()
                 sysfs_create_dir_ns()
                     if (kobj->parent)
                         parent = kobj->parent->sd;
                     ....
                     kernfs_create_dir_ns(parent)
                         kernfs_new_node()
                             kernfs_get(parent)
                         ....
                         /* link in */
                         rc = kernfs_add_one(kn);
                         if (!rc)
                             return kn;

                         kernfs_put(kn)
                             ....
                             repeat:
                             kmem_cache_free(kn)
                             kn = parent;

                             if (kn) {
                                 if (atomic_dec_and_test(&kn->count))
                                     goto repeat;
                             }
                         ....

path2: Remove last child device under glue dir
device_del()

     cleanup_glue_dir()
         mutex_lock(&gdp_mutex);
         if (!kobject_has_children(glue_dir))
             kobject_del(glue_dir);
         kobject_put(glue_dir);
         mutex_unlock(&gdp_mutex);

Before path2 remove last child device under glue dir, If path1 add a new device under glue dir, the glue_dir kobject reference count will be increase to 2 via kobject_get(k) in get_device_parent(). And path1 has been called kernfs_new_node(), but not call kernfs_get(parent). Meanwhile, path2 call kobject_del(glue_dir) beacause 0 is returned by kobject_has_children(). This result in glue_dir->sd is freed and it's reference count will be 0. Then path1 call kernfs_get(parent) will trigger a warning in kernfs_get()(WARN_ON(!atomic_read(&kn->count))) and increase it's reference count to 1. Because glue_dir->sd is freed by path2, the next call kernfs_add_one() by path1 will fail(This is also use-after-free) and call atomic_dec_and_test() to decrease reference count. Because the reference count is decremented to 0, it will also call kmem_cache_free() to free glue_dir->sd again. This will result in double free.

In order to avoid this happening, we also should make sure that kernfs_node for glue_dir is released in path2 only when refcount for glue_dir kobj is 1 to fix this race.

The following calltrace is captured in kernel 4.14 with the following patch applied:

commit 726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

--------------------------------------------------------------------------

[    3.633703] WARNING: CPU: 4 PID: 513 at .../fs/kernfs/dir.c:494
                 Here is WARN_ON(!atomic_read(&kn->count) in kernfs_get().
....
[    3.633986] Call trace:
[    3.633991]  kernfs_create_dir_ns+0xa8/0xb0
[    3.633994]  sysfs_create_dir_ns+0x54/0xe8
[    3.634001]  kobject_add_internal+0x22c/0x3f0
[    3.634005]  kobject_add+0xe4/0x118
[    3.634011]  device_add+0x200/0x870
[    3.634017]  _request_firmware+0x958/0xc38
[    3.634020]  request_firmware_into_buf+0x4c/0x70
....
[    3.634064] kernel BUG at .../mm/slub.c:294!
                 Here is BUG_ON(object == fp) in set_freepointer().
....
[    3.634346] Call trace:
[    3.634351]  kmem_cache_free+0x504/0x6b8
[    3.634355]  kernfs_put+0x14c/0x1d8
[    3.634359]  kernfs_create_dir_ns+0x88/0xb0
[    3.634362]  sysfs_create_dir_ns+0x54/0xe8
[    3.634366]  kobject_add_internal+0x22c/0x3f0
[    3.634370]  kobject_add+0xe4/0x118
[    3.634374]  device_add+0x200/0x870
[    3.634378]  _request_firmware+0x958/0xc38
[    3.634381]  request_firmware_into_buf+0x4c/0x70

--------------------------------------------------------------------------

Fixes 726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

---

Change in v5:
1. Revert to the v1 fix.
2. Add some comment to explain why we need do this in cleanup_glue_dir().
Change in v4:
1. Add some kerneldoc comment.
2. Remove unlock_if_glue_dir().
3. Rename get_device_parent_locked_if_glue_dir() to get_device_parent_locked.
4. Update commit message.
Change in v3:
Add change log.
Change in v2:
Fix device_move() also.

  drivers/base/core.c | 54 ++++++++++++++++++++++++++++++++++++++++++++-
  1 file changed, 53 insertions(+), 1 deletion(-)
drivers/base/core.c +43 -1
index 4aeaa0c92bda..1a67ee325584 100644
@@ -1825,7 +1825,59 @@ static void cleanup_glue_dir(struct device *dev, struct kobject *glue_dir)
  		return;
 
  	mutex_lock(&gdp_mutex);
-	if (!kobject_has_children(glue_dir))
+	/**
+	 * There is a race condition between removing glue directory
+	 * and adding a new device under the glue directory.
+	 *
+	 * path 1: Add the child device under glue dir
+	 * device_add()
+	 *	get_device_parent()
+	 *		mutex_lock(&gdp_mutex);
+	 *		....
+	 *		list_for_each_entry(k, &dev->class->p->glue_dirs.list,
+	 *				    entry)
+	 *		if (k->parent == parent_kobj) {
+	 *			kobj = kobject_get(k);
+	 *			break;
+	 *		}
+	 *		....
+	 *		mutex_unlock(&gdp_mutex);
+	 *		....
+	 *	....
+	 *	kobject_add()
+	 *		kobject_add_internal()
+	 *		create_dir()
+	 *			....
+	 *			if (kobj->parent)
+	 *				parent = kobj->parent->sd;
+	 *			....
+	 *			kernfs_create_dir_ns(parent)
+	 *				....
+	 *
+	 * path2: Remove last child device under glue dir
+	 * device_del()
+	 *	cleanup_glue_dir()
+	 *		....
+	 *		mutex_lock(&gdp_mutex);
+	 *		if (!kobject_has_children(glue_dir))
+	 *			kobject_del(glue_dir);
+	 *		....
+	 *		mutex_unlock(&gdp_mutex);
+	 *
+	 * Before path2 remove last child device under glue dir, if path1 add
+	 * a new device under glue dir, the glue_dir kobject reference count
+	 * will be increase to 2 via kobject_get(k) in get_device_parent().
+	 * And path1 has been called kernfs_create_dir_ns(). Meanwhile,

drivers/base/core.c static void cleanup_glue_dir(struct device *dev, struct kobject *glue_dir) · line 1880

Quoted 10 lines
drivers/base/core.c +10 -0
+	 * path2 call kobject_del(glue_dir) beacause 0 is returned by
+	 * kobject_has_children(). This result in glue_dir->sd is freed.
+	 * Then the path1 will see a stale "empty" but still potentially used
+	 * glue dir around.
+	 *
+	 * In order to avoid this happening, we also should make sure that
+	 * kernfs_node for glue_dir is released in path2 only when refcount
+	 * for glue_dir kobj is 1.
+	 */
+	if (!kobject_has_children(glue_dir) && kref_read(&glue_dir->kref) == 1)

Instead of hardcoding "1 " , can't we check against zero Like Prateek sood did in his patch.https://lkml.org/lkml/2019/5/1/3

+ref = kref_read(&glue_dir->kref);
+if  (!kobject_has_children(glue_dir) && --ref)

We have seen many pattern of this issue which is coming from different driver e.g uinput as well in 4.14 kernel.(https://lkml.org/lkml/2019/4/10/146) The reason why it  started coming after

726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

Looks good to me

-Mukesh

Quoted 3 lines
drivers/base/core.c +0 -0
  		kobject_del(glue_dir);
  	kobject_put(glue_dir);
  	mutex_unlock(&gdp_mutex);
#2 lore raw
Quoted 202 lines · Mukesh Ojha
Quoted 189 lines · Muchun Song

There is a race condition between removing glue directory and adding a new device under the glue directory. It can be reproduced in following test:

path 1: Add the child device under glue dir
device_add()
     get_device_parent()
         mutex_lock(&gdp_mutex);
         ....
         /*find parent from glue_dirs.list*/ list_for_each_entry(k, &dev->class->p->glue_dirs.list, entry) if (k->parent == parent_kobj) {
                 kobj = kobject_get(k);
                 break;
             }
         ....
         mutex_unlock(&gdp_mutex);
         ....
     ....
     kobject_add()
         kobject_add_internal()
             create_dir()
                 sysfs_create_dir_ns()
                     if (kobj->parent)
                         parent = kobj->parent->sd;
                     ....
                     kernfs_create_dir_ns(parent) kernfs_new_node() kernfs_get(parent)
                         ....
                         /* link in */
                         rc = kernfs_add_one(kn);
                         if (!rc)
                             return kn;

                         kernfs_put(kn)
                             ....
                             repeat:
                             kmem_cache_free(kn)
                             kn = parent;

                             if (kn) {
                                 if (atomic_dec_and_test(&kn->count)) goto repeat;
                             }
                         ....

path2: Remove last child device under glue dir
device_del()
     cleanup_glue_dir()
         mutex_lock(&gdp_mutex);
         if (!kobject_has_children(glue_dir))
             kobject_del(glue_dir);
         kobject_put(glue_dir);
         mutex_unlock(&gdp_mutex);

Before path2 remove last child device under glue dir, If path1 add a new device under glue dir, the glue_dir kobject reference count will be increase to 2 via kobject_get(k) in get_device_parent(). And path1 has been called kernfs_new_node(), but not call kernfs_get(parent). Meanwhile, path2 call kobject_del(glue_dir) beacause 0 is returned by kobject_has_children(). This result in glue_dir->sd is freed and it's reference count will be 0. Then path1 call kernfs_get(parent) will trigger a warning in kernfs_get()(WARN_ON(!atomic_read(&kn->count))) and increase it's reference count to 1. Because glue_dir->sd is freed by path2, the next call kernfs_add_one() by path1 will fail(This is also use-after-free) and call atomic_dec_and_test() to decrease reference count. Because the reference count is decremented to 0, it will also call kmem_cache_free() to free glue_dir->sd again. This will result in double free.

In order to avoid this happening, we also should make sure that kernfs_node for glue_dir is released in path2 only when refcount for glue_dir kobj is 1 to fix this race.

The following calltrace is captured in kernel 4.14 with the following patch applied:

commit 726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

--------------------------------------------------------------------------

[    3.633703] WARNING: CPU: 4 PID: 513 at .../fs/kernfs/dir.c:494 Here is WARN_ON(!atomic_read(&kn->count) in kernfs_get().
....
[    3.633986] Call trace:
[    3.633991]  kernfs_create_dir_ns+0xa8/0xb0
[    3.633994]  sysfs_create_dir_ns+0x54/0xe8
[    3.634001]  kobject_add_internal+0x22c/0x3f0
[    3.634005]  kobject_add+0xe4/0x118
[    3.634011]  device_add+0x200/0x870
[    3.634017]  _request_firmware+0x958/0xc38
[    3.634020]  request_firmware_into_buf+0x4c/0x70
....
[    3.634064] kernel BUG at .../mm/slub.c:294!
                 Here is BUG_ON(object == fp) in set_freepointer().
....
[    3.634346] Call trace:
[    3.634351]  kmem_cache_free+0x504/0x6b8
[    3.634355]  kernfs_put+0x14c/0x1d8
[    3.634359]  kernfs_create_dir_ns+0x88/0xb0
[    3.634362]  sysfs_create_dir_ns+0x54/0xe8
[    3.634366]  kobject_add_internal+0x22c/0x3f0
[    3.634370]  kobject_add+0xe4/0x118
[    3.634374]  device_add+0x200/0x870
[    3.634378]  _request_firmware+0x958/0xc38
[    3.634381]  request_firmware_into_buf+0x4c/0x70
--------------------------------------------------------------------------

Fixes 726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

---

Change in v5:
        1. Revert to the v1 fix.
        2. Add some comment to explain why we need do this in cleanup_glue_dir().
Change in v4:
        1. Add some kerneldoc comment.
        2. Remove unlock_if_glue_dir().
        3. Rename get_device_parent_locked_if_glue_dir() to get_device_parent_locked.
        4. Update commit message.
Change in v3:
        Add change log.
Change in v2:
        Fix device_move() also.

  drivers/base/core.c | 54 ++++++++++++++++++++++++++++++++++++++++++++-

  1 file changed, 53 insertions(+), 1 deletion(-)
drivers/base/core.c +40 -1
index 4aeaa0c92bda..1a67ee325584 100644
@@ -1825,7 +1825,59 @@ static void cleanup_glue_dir(struct device 
 *dev, struct kobject *glue_dir)
           return;
         mutex_lock(&gdp_mutex);
-    if (!kobject_has_children(glue_dir))
+    /**
+     * There is a race condition between removing glue directory
+     * and adding a new device under the glue directory.
+     *
+     * path 1: Add the child device under glue dir
+     * device_add()
+     *    get_device_parent()
+     *        mutex_lock(&gdp_mutex);
+     *        ....
+     *        list_for_each_entry(k, &dev->class->p->glue_dirs.list,
+     *                    entry)
+     *        if (k->parent == parent_kobj) {
+     *            kobj = kobject_get(k);
+     *            break;
+     *        }
+     *        ....
+     *        mutex_unlock(&gdp_mutex);
+     *        ....
+     *    ....
+     *    kobject_add()
+     *        kobject_add_internal()
+     *        create_dir()
+     *            ....
+     *            if (kobj->parent)
+     *                parent = kobj->parent->sd;
+     *            ....
+     *            kernfs_create_dir_ns(parent)
+     *                ....
+     *
+     * path2: Remove last child device under glue dir
+     * device_del()
+     *    cleanup_glue_dir()
+     *        ....
+     *        mutex_lock(&gdp_mutex);
+     *        if (!kobject_has_children(glue_dir))
+     *            kobject_del(glue_dir);
+     *        ....
+     *        mutex_unlock(&gdp_mutex);
+     *
+     * Before path2 remove last child device under glue dir, if 

path1 add
+     * a new device under glue dir, the glue_dir kobject reference count
+     * will be increase to 2 via kobject_get(k) in get_device_parent(). +     * And path1 has been called kernfs_create_dir_ns(). Meanwhile, +     * path2 call kobject_del(glue_dir) beacause 0 is returned by +     * kobject_has_children(). This result in glue_dir->sd is freed. +     * Then the path1 will see a stale "empty" but still potentially used

Quoted 8 lines

+     * glue dir around.
+     *
+     * In order to avoid this happening, we also should make sure that +     * kernfs_node for glue_dir is released in path2 only when refcount +     * for glue_dir kobj is 1.
+     */
+    if (!kobject_has_children(glue_dir) && kref_read(&glue_dir->kref) == 1)

Instead of hardcoding "1 " , can't we check against zero Like Prateek sood did in his patch.https://lkml.org/lkml/2019/5/1/3

Review

Quoted 2 lines
(quoted diff) +2 -0
+ref = kref_read(&glue_dir->kref);
+if  (!kobject_has_children(glue_dir) && --ref)

I meant if  (!kobject_has_children(glue_dir) && ! --ref)

Quoted 18 lines

We have seen many pattern of this issue which is coming from different driver e.g uinput as well in 4.14 kernel.(https://lkml.org/lkml/2019/4/10/146) The reason why it  started coming after

726e41097920 ("drivers: core: Remove glue dirs from sysfs earlier")

Looks good to me

-Mukesh

Quoted 3 lines

          kobject_del(glue_dir);
      kobject_put(glue_dir);
      mutex_unlock(&gdp_mutex);