All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH v3 0/4] Update SELinuxfs out of tree and then swapover
@ 2020-08-19 19:59 Daniel Burgener
  2020-08-19 19:59 ` [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup Daniel Burgener
                   ` (3 more replies)
  0 siblings, 4 replies; 13+ messages in thread
From: Daniel Burgener @ 2020-08-19 19:59 UTC (permalink / raw)
  To: selinux; +Cc: stephen.smalley.work, omosnace, paul, linux-fsdevel, viro

v2: Clean up commit messages to accurately reflect current scope of
changes
v3: Remove all policy_capabilities directory manipulation.  Switch from
vfs_rename() to d_exchange() for directory exchanging.  Use appropriate
comment style.  Reuse inodes for temporary directories.

In the current implementation, on policy load /sys/fs/selinux is updated
by deleting the previous contents of
/sys/fs/selinux/{class,booleans} and then recreating them.  This means
that there is a period of time when the contents of these directories do
not exist which can cause race conditions as userspace relies on them for
information about the policy.  In addition, it means that error recovery
in the event of failure is challenging.

This patch series follows the design outlined by Al Viro in a previous
e-mail to the list[1].  This approach is to first create the new
directory structures out of tree, then to perform the swapover, and
finally to delete the old directories.  Not handled in this series is
error recovery in the event of failure.

Error recovery in the selinuxfs recreation is unhandled in the current
code, so this series will not cause any regression in this regard.
Handling directory recreation in this manner is a prerequisite to make
proper error handling possible.

In order to demonstrate the race condition that this series fixes, you
can use the following commands:

while true; do cat /sys/fs/selinux/class/service/perms/status
>/dev/null; done &
while true; do load_policy; done;

In the existing code, this will display errors fairly often as the class
lookup fails.  (In normal operation from systemd, this would result in a
permission check which would be allowed or denied based on policy settings
around unknown object classes.) After applying this patch series you
should expect to no longer see such error messages.

This series is relative to https://patchwork.kernel.org/patch/11705743/,
Stephen Smalley's series to split policy loading into a prep and commit.

[1] https://lore.kernel.org/selinux/20181002155810.GP32577@ZenIV.linux.org.uk/

Daniel Burgener (4):
  selinux: Create function for selinuxfs directory cleanup
  selinux: Refactor selinuxfs directory populating functions
  selinux: Standardize string literal usage for selinuxfs directory
    names
  selinux: Create new booleans and class dirs out of tree

 security/selinux/selinuxfs.c | 190 +++++++++++++++++++++++++++--------
 1 file changed, 148 insertions(+), 42 deletions(-)

-- 
2.25.4


^ permalink raw reply	[flat|nested] 13+ messages in thread

* [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup
  2020-08-19 19:59 [PATCH v3 0/4] Update SELinuxfs out of tree and then swapover Daniel Burgener
@ 2020-08-19 19:59 ` Daniel Burgener
  2020-08-20 14:53   ` Stephen Smalley
  2020-08-21 13:43   ` Paul Moore
  2020-08-19 19:59 ` [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions Daniel Burgener
                   ` (2 subsequent siblings)
  3 siblings, 2 replies; 13+ messages in thread
From: Daniel Burgener @ 2020-08-19 19:59 UTC (permalink / raw)
  To: selinux; +Cc: stephen.smalley.work, omosnace, paul, linux-fsdevel, viro

Separating the cleanup from the creation will simplify two things in
future patches in this series.  First, the creation can be made generic,
to create directories not tied to the selinux_fs_info structure.  Second,
we will ultimately want to reorder creation and deletion so that the
deletions aren't performed until the new directory structures have already
been moved into place.

Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
---
 security/selinux/selinuxfs.c | 39 +++++++++++++++++++++++-------------
 1 file changed, 25 insertions(+), 14 deletions(-)

diff --git a/security/selinux/selinuxfs.c b/security/selinux/selinuxfs.c
index 131816878e50..19670e9bcd72 100644
--- a/security/selinux/selinuxfs.c
+++ b/security/selinux/selinuxfs.c
@@ -355,6 +355,9 @@ static int sel_make_classes(struct selinux_fs_info *fsi,
 static struct dentry *sel_make_dir(struct dentry *dir, const char *name,
 			unsigned long *ino);
 
+/* declaration for sel_remove_old_policy_nodes */
+static void sel_remove_entries(struct dentry *de);
+
 static ssize_t sel_read_mls(struct file *filp, char __user *buf,
 				size_t count, loff_t *ppos)
 {
@@ -509,11 +512,33 @@ static const struct file_operations sel_policy_ops = {
 	.llseek		= generic_file_llseek,
 };
 
+static void sel_remove_old_policy_nodes(struct selinux_fs_info *fsi)
+{
+	u32 i;
+
+	/* bool_dir cleanup */
+	for (i = 0; i < fsi->bool_num; i++)
+		kfree(fsi->bool_pending_names[i]);
+	kfree(fsi->bool_pending_names);
+	kfree(fsi->bool_pending_values);
+	fsi->bool_num = 0;
+	fsi->bool_pending_names = NULL;
+	fsi->bool_pending_values = NULL;
+
+	sel_remove_entries(fsi->bool_dir);
+
+	/* class_dir cleanup */
+	sel_remove_entries(fsi->class_dir);
+
+}
+
 static int sel_make_policy_nodes(struct selinux_fs_info *fsi,
 				struct selinux_policy *newpolicy)
 {
 	int ret;
 
+	sel_remove_old_policy_nodes(fsi);
+
 	ret = sel_make_bools(fsi, newpolicy);
 	if (ret) {
 		pr_err("SELinux: failed to load policy booleans\n");
@@ -1348,17 +1373,6 @@ static int sel_make_bools(struct selinux_fs_info *fsi,
 	int *values = NULL;
 	u32 sid;
 
-	/* remove any existing files */
-	for (i = 0; i < fsi->bool_num; i++)
-		kfree(fsi->bool_pending_names[i]);
-	kfree(fsi->bool_pending_names);
-	kfree(fsi->bool_pending_values);
-	fsi->bool_num = 0;
-	fsi->bool_pending_names = NULL;
-	fsi->bool_pending_values = NULL;
-
-	sel_remove_entries(dir);
-
 	ret = -ENOMEM;
 	page = (char *)get_zeroed_page(GFP_KERNEL);
 	if (!page)
@@ -1873,9 +1887,6 @@ static int sel_make_classes(struct selinux_fs_info *fsi,
 	int rc, nclasses, i;
 	char **classes;
 
-	/* delete any existing entries */
-	sel_remove_entries(fsi->class_dir);
-
 	rc = security_get_classes(newpolicy, &classes, &nclasses);
 	if (rc)
 		return rc;
-- 
2.25.4


^ permalink raw reply related	[flat|nested] 13+ messages in thread

* [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions
  2020-08-19 19:59 [PATCH v3 0/4] Update SELinuxfs out of tree and then swapover Daniel Burgener
  2020-08-19 19:59 ` [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup Daniel Burgener
@ 2020-08-19 19:59 ` Daniel Burgener
  2020-08-20 14:55   ` Stephen Smalley
  2020-08-21 13:43   ` Paul Moore
  2020-08-19 19:59 ` [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names Daniel Burgener
  2020-08-19 19:59 ` [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree Daniel Burgener
  3 siblings, 2 replies; 13+ messages in thread
From: Daniel Burgener @ 2020-08-19 19:59 UTC (permalink / raw)
  To: selinux; +Cc: stephen.smalley.work, omosnace, paul, linux-fsdevel, viro

Make sel_make_bools and sel_make_classes take the specific elements of
selinux_fs_info that they need rather than the entire struct.

This will allow a future patch to pass temporary elements that are not in
the selinux_fs_info struct to these functions so that the original elements
can be preserved until we are ready to perform the switch over.

Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
---
 security/selinux/selinuxfs.c | 45 ++++++++++++++++++++----------------
 1 file changed, 25 insertions(+), 20 deletions(-)

diff --git a/security/selinux/selinuxfs.c b/security/selinux/selinuxfs.c
index 19670e9bcd72..cac585ce576b 100644
--- a/security/selinux/selinuxfs.c
+++ b/security/selinux/selinuxfs.c
@@ -346,10 +346,12 @@ static const struct file_operations sel_policyvers_ops = {
 };
 
 /* declaration for sel_write_load */
-static int sel_make_bools(struct selinux_fs_info *fsi,
-			struct selinux_policy *newpolicy);
-static int sel_make_classes(struct selinux_fs_info *fsi,
-			struct selinux_policy *newpolicy);
+static int sel_make_bools(struct selinux_policy *newpolicy, struct dentry *bool_dir,
+			  unsigned int *bool_num, char ***bool_pending_names,
+			  unsigned int **bool_pending_values);
+static int sel_make_classes(struct selinux_policy *newpolicy,
+			    struct dentry *class_dir,
+			    unsigned long *last_class_ino);
 
 /* declaration for sel_make_class_dirs */
 static struct dentry *sel_make_dir(struct dentry *dir, const char *name,
@@ -539,13 +541,15 @@ static int sel_make_policy_nodes(struct selinux_fs_info *fsi,
 
 	sel_remove_old_policy_nodes(fsi);
 
-	ret = sel_make_bools(fsi, newpolicy);
+	ret = sel_make_bools(newpolicy, fsi->bool_dir, &fsi->bool_num,
+			     &fsi->bool_pending_names, &fsi->bool_pending_values);
 	if (ret) {
 		pr_err("SELinux: failed to load policy booleans\n");
 		return ret;
 	}
 
-	ret = sel_make_classes(fsi, newpolicy);
+	ret = sel_make_classes(newpolicy, fsi->class_dir,
+			       &fsi->last_class_ino);
 	if (ret) {
 		pr_err("SELinux: failed to load policy classes\n");
 		return ret;
@@ -1359,13 +1363,13 @@ static void sel_remove_entries(struct dentry *de)
 
 #define BOOL_DIR_NAME "booleans"
 
-static int sel_make_bools(struct selinux_fs_info *fsi,
-			struct selinux_policy *newpolicy)
+static int sel_make_bools(struct selinux_policy *newpolicy, struct dentry *bool_dir,
+			  unsigned int *bool_num, char ***bool_pending_names,
+			  unsigned int **bool_pending_values)
 {
 	int ret;
 	ssize_t len;
 	struct dentry *dentry = NULL;
-	struct dentry *dir = fsi->bool_dir;
 	struct inode *inode = NULL;
 	struct inode_security_struct *isec;
 	char **names = NULL, *page;
@@ -1384,12 +1388,12 @@ static int sel_make_bools(struct selinux_fs_info *fsi,
 
 	for (i = 0; i < num; i++) {
 		ret = -ENOMEM;
-		dentry = d_alloc_name(dir, names[i]);
+		dentry = d_alloc_name(bool_dir, names[i]);
 		if (!dentry)
 			goto out;
 
 		ret = -ENOMEM;
-		inode = sel_make_inode(dir->d_sb, S_IFREG | S_IRUGO | S_IWUSR);
+		inode = sel_make_inode(bool_dir->d_sb, S_IFREG | S_IRUGO | S_IWUSR);
 		if (!inode) {
 			dput(dentry);
 			goto out;
@@ -1418,9 +1422,9 @@ static int sel_make_bools(struct selinux_fs_info *fsi,
 		inode->i_ino = i|SEL_BOOL_INO_OFFSET;
 		d_add(dentry, inode);
 	}
-	fsi->bool_num = num;
-	fsi->bool_pending_names = names;
-	fsi->bool_pending_values = values;
+	*bool_num = num;
+	*bool_pending_names = names;
+	*bool_pending_values = values;
 
 	free_page((unsigned long)page);
 	return 0;
@@ -1433,7 +1437,7 @@ static int sel_make_bools(struct selinux_fs_info *fsi,
 		kfree(names);
 	}
 	kfree(values);
-	sel_remove_entries(dir);
+	sel_remove_entries(bool_dir);
 
 	return ret;
 }
@@ -1880,8 +1884,9 @@ static int sel_make_class_dir_entries(struct selinux_policy *newpolicy,
 	return rc;
 }
 
-static int sel_make_classes(struct selinux_fs_info *fsi,
-			struct selinux_policy *newpolicy)
+static int sel_make_classes(struct selinux_policy *newpolicy,
+			    struct dentry *class_dir,
+			    unsigned long *last_class_ino)
 {
 
 	int rc, nclasses, i;
@@ -1892,13 +1897,13 @@ static int sel_make_classes(struct selinux_fs_info *fsi,
 		return rc;
 
 	/* +2 since classes are 1-indexed */
-	fsi->last_class_ino = sel_class_to_ino(nclasses + 2);
+	*last_class_ino = sel_class_to_ino(nclasses + 2);
 
 	for (i = 0; i < nclasses; i++) {
 		struct dentry *class_name_dir;
 
-		class_name_dir = sel_make_dir(fsi->class_dir, classes[i],
-					      &fsi->last_class_ino);
+		class_name_dir = sel_make_dir(class_dir, classes[i],
+					      last_class_ino);
 		if (IS_ERR(class_name_dir)) {
 			rc = PTR_ERR(class_name_dir);
 			goto out;
-- 
2.25.4


^ permalink raw reply related	[flat|nested] 13+ messages in thread

* [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names
  2020-08-19 19:59 [PATCH v3 0/4] Update SELinuxfs out of tree and then swapover Daniel Burgener
  2020-08-19 19:59 ` [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup Daniel Burgener
  2020-08-19 19:59 ` [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions Daniel Burgener
@ 2020-08-19 19:59 ` Daniel Burgener
  2020-08-20 14:56   ` Stephen Smalley
  2020-08-21 13:44   ` Paul Moore
  2020-08-19 19:59 ` [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree Daniel Burgener
  3 siblings, 2 replies; 13+ messages in thread
From: Daniel Burgener @ 2020-08-19 19:59 UTC (permalink / raw)
  To: selinux; +Cc: stephen.smalley.work, omosnace, paul, linux-fsdevel, viro

Switch class and policy_capabilities directory names to be referred to with
global constants, consistent with booleans directory name.  This will allow
for easy consistency of naming in future development.

Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
---
 security/selinux/selinuxfs.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/security/selinux/selinuxfs.c b/security/selinux/selinuxfs.c
index cac585ce576b..2a0e8b5f19d5 100644
--- a/security/selinux/selinuxfs.c
+++ b/security/selinux/selinuxfs.c
@@ -117,6 +117,10 @@ static void selinux_fs_info_free(struct super_block *sb)
 #define SEL_POLICYCAP_INO_OFFSET	0x08000000
 #define SEL_INO_MASK			0x00ffffff
 
+#define BOOL_DIR_NAME "booleans"
+#define CLASS_DIR_NAME "class"
+#define POLICYCAP_DIR_NAME "policy_capabilities"
+
 #define TMPBUFLEN	12
 static ssize_t sel_read_enforce(struct file *filp, char __user *buf,
 				size_t count, loff_t *ppos)
@@ -1361,8 +1365,6 @@ static void sel_remove_entries(struct dentry *de)
 	shrink_dcache_parent(de);
 }
 
-#define BOOL_DIR_NAME "booleans"
-
 static int sel_make_bools(struct selinux_policy *newpolicy, struct dentry *bool_dir,
 			  unsigned int *bool_num, char ***bool_pending_names,
 			  unsigned int **bool_pending_values)
@@ -2078,14 +2080,14 @@ static int sel_fill_super(struct super_block *sb, struct fs_context *fc)
 	if (ret)
 		goto err;
 
-	fsi->class_dir = sel_make_dir(sb->s_root, "class", &fsi->last_ino);
+	fsi->class_dir = sel_make_dir(sb->s_root, CLASS_DIR_NAME, &fsi->last_ino);
 	if (IS_ERR(fsi->class_dir)) {
 		ret = PTR_ERR(fsi->class_dir);
 		fsi->class_dir = NULL;
 		goto err;
 	}
 
-	fsi->policycap_dir = sel_make_dir(sb->s_root, "policy_capabilities",
+	fsi->policycap_dir = sel_make_dir(sb->s_root, POLICYCAP_DIR_NAME,
 					  &fsi->last_ino);
 	if (IS_ERR(fsi->policycap_dir)) {
 		ret = PTR_ERR(fsi->policycap_dir);
-- 
2.25.4


^ permalink raw reply related	[flat|nested] 13+ messages in thread

* [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree
  2020-08-19 19:59 [PATCH v3 0/4] Update SELinuxfs out of tree and then swapover Daniel Burgener
                   ` (2 preceding siblings ...)
  2020-08-19 19:59 ` [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names Daniel Burgener
@ 2020-08-19 19:59 ` Daniel Burgener
  2020-08-20 15:23   ` Stephen Smalley
  2020-08-21 13:44   ` Paul Moore
  3 siblings, 2 replies; 13+ messages in thread
From: Daniel Burgener @ 2020-08-19 19:59 UTC (permalink / raw)
  To: selinux; +Cc: stephen.smalley.work, omosnace, paul, linux-fsdevel, viro

In order to avoid concurrency issues around selinuxfs resource availability
during policy load, we first create new directories out of tree for
reloaded resources, then swap them in, and finally delete the old versions.

This fix focuses on concurrency in each of the two subtrees swapped, and
not concurrency between the trees.  This means that it is still possible
that subsequent reads to eg the booleans directory and the class directory
during a policy load could see the old state for one and the new for the other.
The problem of ensuring that policy loads are fully atomic from the perspective
of userspace is larger than what is dealt with here.  This commit focuses on
ensuring that the directories contents always match either the new or the old
policy state from the perspective of userspace.

In the previous implementation, on policy load /sys/fs/selinux is updated
by deleting the previous contents of
/sys/fs/selinux/{class,booleans} and then recreating them.  This means
that there is a period of time when the contents of these directories do not
exist which can cause race conditions as userspace relies on them for
information about the policy.  In addition, it means that error recovery in
the event of failure is challenging.

In order to demonstrate the race condition that this series fixes, you
can use the following commands:

while true; do cat /sys/fs/selinux/class/service/perms/status
>/dev/null; done &
while true; do load_policy; done;

In the existing code, this will display errors fairly often as the class
lookup fails.  (In normal operation from systemd, this would result in a
permission check which would be allowed or denied based on policy settings
around unknown object classes.) After applying this patch series you
should expect to no longer see such error messages.

Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
---
 security/selinux/selinuxfs.c | 113 ++++++++++++++++++++++++++++-------
 1 file changed, 90 insertions(+), 23 deletions(-)

diff --git a/security/selinux/selinuxfs.c b/security/selinux/selinuxfs.c
index 2a0e8b5f19d5..d1872adf0c47 100644
--- a/security/selinux/selinuxfs.c
+++ b/security/selinux/selinuxfs.c
@@ -20,6 +20,7 @@
 #include <linux/fs_context.h>
 #include <linux/mount.h>
 #include <linux/mutex.h>
+#include <linux/namei.h>
 #include <linux/init.h>
 #include <linux/string.h>
 #include <linux/security.h>
@@ -361,7 +362,11 @@ static int sel_make_classes(struct selinux_policy *newpolicy,
 static struct dentry *sel_make_dir(struct dentry *dir, const char *name,
 			unsigned long *ino);
 
-/* declaration for sel_remove_old_policy_nodes */
+/* declaration for sel_make_policy_nodes */
+static struct dentry *sel_make_disconnected_dir(struct super_block *sb,
+						unsigned long *ino);
+
+/* declaration for sel_make_policy_nodes */
 static void sel_remove_entries(struct dentry *de);
 
 static ssize_t sel_read_mls(struct file *filp, char __user *buf,
@@ -518,48 +523,94 @@ static const struct file_operations sel_policy_ops = {
 	.llseek		= generic_file_llseek,
 };
 
-static void sel_remove_old_policy_nodes(struct selinux_fs_info *fsi)
+static void sel_remove_old_bool_data(unsigned int bool_num, char **bool_names,
+				unsigned int *bool_values)
 {
 	u32 i;
 
 	/* bool_dir cleanup */
-	for (i = 0; i < fsi->bool_num; i++)
-		kfree(fsi->bool_pending_names[i]);
-	kfree(fsi->bool_pending_names);
-	kfree(fsi->bool_pending_values);
-	fsi->bool_num = 0;
-	fsi->bool_pending_names = NULL;
-	fsi->bool_pending_values = NULL;
-
-	sel_remove_entries(fsi->bool_dir);
-
-	/* class_dir cleanup */
-	sel_remove_entries(fsi->class_dir);
-
+	for (i = 0; i < bool_num; i++)
+		kfree(bool_names[i]);
+	kfree(bool_names);
+	kfree(bool_values);
 }
 
 static int sel_make_policy_nodes(struct selinux_fs_info *fsi,
 				struct selinux_policy *newpolicy)
 {
-	int ret;
+	int ret = 0;
+	struct dentry *tmp_parent, *tmp_bool_dir, *tmp_class_dir, *old_dentry;
+	unsigned int tmp_bool_num, old_bool_num;
+	char **tmp_bool_names, **old_bool_names;
+	unsigned int *tmp_bool_values, *old_bool_values;
+	unsigned long tmp_ino = fsi->last_ino; /* Don't increment last_ino in this function */
 
-	sel_remove_old_policy_nodes(fsi);
+	tmp_parent = sel_make_disconnected_dir(fsi->sb, &tmp_ino);
+	if (IS_ERR(tmp_parent))
+		return PTR_ERR(tmp_parent);
 
-	ret = sel_make_bools(newpolicy, fsi->bool_dir, &fsi->bool_num,
-			     &fsi->bool_pending_names, &fsi->bool_pending_values);
+	tmp_ino = fsi->bool_dir->d_inode->i_ino - 1; /* sel_make_dir will increment and set */
+	tmp_bool_dir = sel_make_dir(tmp_parent, BOOL_DIR_NAME, &tmp_ino);
+	if (IS_ERR(tmp_bool_dir)) {
+		ret = PTR_ERR(tmp_bool_dir);
+		goto out;
+	}
+
+	tmp_ino = fsi->class_dir->d_inode->i_ino - 1; /* sel_make_dir will increment and set */
+	tmp_class_dir = sel_make_dir(tmp_parent, CLASS_DIR_NAME, &tmp_ino);
+	if (IS_ERR(tmp_class_dir)) {
+		ret = PTR_ERR(tmp_class_dir);
+		goto out;
+	}
+
+	ret = sel_make_bools(newpolicy, tmp_bool_dir, &tmp_bool_num,
+			     &tmp_bool_names, &tmp_bool_values);
 	if (ret) {
 		pr_err("SELinux: failed to load policy booleans\n");
-		return ret;
+		goto out;
 	}
 
-	ret = sel_make_classes(newpolicy, fsi->class_dir,
+	ret = sel_make_classes(newpolicy, tmp_class_dir,
 			       &fsi->last_class_ino);
 	if (ret) {
 		pr_err("SELinux: failed to load policy classes\n");
-		return ret;
+		goto out;
 	}
 
-	return 0;
+	/* booleans */
+	old_dentry = fsi->bool_dir;
+	lock_rename(tmp_bool_dir, old_dentry);
+	d_exchange(tmp_bool_dir, fsi->bool_dir);
+
+	old_bool_num = fsi->bool_num;
+	old_bool_names = fsi->bool_pending_names;
+	old_bool_values = fsi->bool_pending_values;
+
+	fsi->bool_num = tmp_bool_num;
+	fsi->bool_pending_names = tmp_bool_names;
+	fsi->bool_pending_values = tmp_bool_values;
+
+	sel_remove_old_bool_data(old_bool_num, old_bool_names, old_bool_values);
+
+	fsi->bool_dir = tmp_bool_dir;
+	unlock_rename(tmp_bool_dir, old_dentry);
+
+	/* classes */
+	old_dentry = fsi->class_dir;
+	lock_rename(tmp_class_dir, old_dentry);
+	d_exchange(tmp_class_dir, fsi->class_dir);
+	fsi->class_dir = tmp_class_dir;
+	unlock_rename(tmp_class_dir, old_dentry);
+
+out:
+	/* Since the other temporary dirs are children of tmp_parent
+	 * this will handle all the cleanup in the case of a failure before
+	 * the swapover
+	 */
+	sel_remove_entries(tmp_parent);
+	dput(tmp_parent); /* d_genocide() only handles the children */
+
+	return ret;
 }
 
 static ssize_t sel_write_load(struct file *file, const char __user *buf,
@@ -1982,6 +2033,22 @@ static struct dentry *sel_make_dir(struct dentry *dir, const char *name,
 	return dentry;
 }
 
+static struct dentry *sel_make_disconnected_dir(struct super_block *sb,
+						unsigned long *ino)
+{
+	struct inode *inode = sel_make_inode(sb, S_IFDIR | S_IRUGO | S_IXUGO);
+
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+
+	inode->i_op = &simple_dir_inode_operations;
+	inode->i_fop = &simple_dir_operations;
+	inode->i_ino = ++(*ino);
+	/* directory inodes start off with i_nlink == 2 (for "." entry) */
+	inc_nlink(inode);
+	return d_obtain_alias(inode);
+}
+
 #define NULL_FILE_NAME "null"
 
 static int sel_fill_super(struct super_block *sb, struct fs_context *fc)
-- 
2.25.4


^ permalink raw reply related	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup
  2020-08-19 19:59 ` [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup Daniel Burgener
@ 2020-08-20 14:53   ` Stephen Smalley
  2020-08-21 13:43   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Stephen Smalley @ 2020-08-20 14:53 UTC (permalink / raw)
  To: Daniel Burgener, selinux; +Cc: omosnace, paul, linux-fsdevel, viro

On 8/19/20 3:59 PM, Daniel Burgener wrote:

> Separating the cleanup from the creation will simplify two things in
> future patches in this series.  First, the creation can be made generic,
> to create directories not tied to the selinux_fs_info structure.  Second,
> we will ultimately want to reorder creation and deletion so that the
> deletions aren't performed until the new directory structures have already
> been moved into place.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
Acked-by: Stephen Smalley <stephen.smalley.work@gmail.com>


^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions
  2020-08-19 19:59 ` [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions Daniel Burgener
@ 2020-08-20 14:55   ` Stephen Smalley
  2020-08-21 13:43   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Stephen Smalley @ 2020-08-20 14:55 UTC (permalink / raw)
  To: Daniel Burgener, selinux; +Cc: omosnace, paul, linux-fsdevel, viro

On 8/19/20 3:59 PM, Daniel Burgener wrote:

> Make sel_make_bools and sel_make_classes take the specific elements of
> selinux_fs_info that they need rather than the entire struct.
>
> This will allow a future patch to pass temporary elements that are not in
> the selinux_fs_info struct to these functions so that the original elements
> can be preserved until we are ready to perform the switch over.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
Acked-by: Stephen Smalley <stephen.smalley.work@gmail.com>

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names
  2020-08-19 19:59 ` [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names Daniel Burgener
@ 2020-08-20 14:56   ` Stephen Smalley
  2020-08-21 13:44   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Stephen Smalley @ 2020-08-20 14:56 UTC (permalink / raw)
  To: Daniel Burgener, selinux; +Cc: omosnace, paul, linux-fsdevel, viro

On 8/19/20 3:59 PM, Daniel Burgener wrote:

> Switch class and policy_capabilities directory names to be referred to with
> global constants, consistent with booleans directory name.  This will allow
> for easy consistency of naming in future development.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>

Not strictly needed for policy_capabilities since it isn't being removed 
anymore but no harm.

Acked-by: Stephen Smalley <stephen.smalley.work@gmail.com>


^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree
  2020-08-19 19:59 ` [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree Daniel Burgener
@ 2020-08-20 15:23   ` Stephen Smalley
  2020-08-21 13:44   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Stephen Smalley @ 2020-08-20 15:23 UTC (permalink / raw)
  To: Daniel Burgener, selinux; +Cc: omosnace, paul, linux-fsdevel, viro

On 8/19/20 3:59 PM, Daniel Burgener wrote:

> In order to avoid concurrency issues around selinuxfs resource availability
> during policy load, we first create new directories out of tree for
> reloaded resources, then swap them in, and finally delete the old versions.
>
> This fix focuses on concurrency in each of the two subtrees swapped, and
> not concurrency between the trees.  This means that it is still possible
> that subsequent reads to eg the booleans directory and the class directory
> during a policy load could see the old state for one and the new for the other.
> The problem of ensuring that policy loads are fully atomic from the perspective
> of userspace is larger than what is dealt with here.  This commit focuses on
> ensuring that the directories contents always match either the new or the old
> policy state from the perspective of userspace.
>
> In the previous implementation, on policy load /sys/fs/selinux is updated
> by deleting the previous contents of
> /sys/fs/selinux/{class,booleans} and then recreating them.  This means
> that there is a period of time when the contents of these directories do not
> exist which can cause race conditions as userspace relies on them for
> information about the policy.  In addition, it means that error recovery in
> the event of failure is challenging.
>
> In order to demonstrate the race condition that this series fixes, you
> can use the following commands:
>
> while true; do cat /sys/fs/selinux/class/service/perms/status
>> /dev/null; done &
> while true; do load_policy; done;
>
> In the existing code, this will display errors fairly often as the class
> lookup fails.  (In normal operation from systemd, this would result in a
> permission check which would be allowed or denied based on policy settings
> around unknown object classes.) After applying this patch series you
> should expect to no longer see such error messages.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>

Acked-by: Stephen Smalley <stephen.smalley.work@gmail.com>


^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup
  2020-08-19 19:59 ` [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup Daniel Burgener
  2020-08-20 14:53   ` Stephen Smalley
@ 2020-08-21 13:43   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Paul Moore @ 2020-08-21 13:43 UTC (permalink / raw)
  To: Daniel Burgener
  Cc: selinux, Stephen Smalley, Ondrej Mosnacek, linux-fsdevel, viro

On Wed, Aug 19, 2020 at 3:59 PM Daniel Burgener
<dburgener@linux.microsoft.com> wrote:
>
> Separating the cleanup from the creation will simplify two things in
> future patches in this series.  First, the creation can be made generic,
> to create directories not tied to the selinux_fs_info structure.  Second,
> we will ultimately want to reorder creation and deletion so that the
> deletions aren't performed until the new directory structures have already
> been moved into place.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
> ---
>  security/selinux/selinuxfs.c | 39 +++++++++++++++++++++++-------------
>  1 file changed, 25 insertions(+), 14 deletions(-)

Merged into selinux/next, thanks!

-- 
paul moore
www.paul-moore.com

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions
  2020-08-19 19:59 ` [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions Daniel Burgener
  2020-08-20 14:55   ` Stephen Smalley
@ 2020-08-21 13:43   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Paul Moore @ 2020-08-21 13:43 UTC (permalink / raw)
  To: Daniel Burgener
  Cc: selinux, Stephen Smalley, Ondrej Mosnacek, linux-fsdevel, viro

On Wed, Aug 19, 2020 at 3:59 PM Daniel Burgener
<dburgener@linux.microsoft.com> wrote:
>
> Make sel_make_bools and sel_make_classes take the specific elements of
> selinux_fs_info that they need rather than the entire struct.
>
> This will allow a future patch to pass temporary elements that are not in
> the selinux_fs_info struct to these functions so that the original elements
> can be preserved until we are ready to perform the switch over.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
> ---
>  security/selinux/selinuxfs.c | 45 ++++++++++++++++++++----------------
>  1 file changed, 25 insertions(+), 20 deletions(-)

Merged into selinux/next, thanks!

-- 
paul moore
www.paul-moore.com

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names
  2020-08-19 19:59 ` [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names Daniel Burgener
  2020-08-20 14:56   ` Stephen Smalley
@ 2020-08-21 13:44   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Paul Moore @ 2020-08-21 13:44 UTC (permalink / raw)
  To: Daniel Burgener
  Cc: selinux, Stephen Smalley, Ondrej Mosnacek, linux-fsdevel, viro

On Wed, Aug 19, 2020 at 3:59 PM Daniel Burgener
<dburgener@linux.microsoft.com> wrote:
>
> Switch class and policy_capabilities directory names to be referred to with
> global constants, consistent with booleans directory name.  This will allow
> for easy consistency of naming in future development.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
> ---
>  security/selinux/selinuxfs.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)

Merged into selinux/next, thanks!

-- 
paul moore
www.paul-moore.com

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree
  2020-08-19 19:59 ` [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree Daniel Burgener
  2020-08-20 15:23   ` Stephen Smalley
@ 2020-08-21 13:44   ` Paul Moore
  1 sibling, 0 replies; 13+ messages in thread
From: Paul Moore @ 2020-08-21 13:44 UTC (permalink / raw)
  To: Daniel Burgener
  Cc: selinux, Stephen Smalley, Ondrej Mosnacek, linux-fsdevel, viro

On Wed, Aug 19, 2020 at 3:59 PM Daniel Burgener
<dburgener@linux.microsoft.com> wrote:
>
> In order to avoid concurrency issues around selinuxfs resource availability
> during policy load, we first create new directories out of tree for
> reloaded resources, then swap them in, and finally delete the old versions.
>
> This fix focuses on concurrency in each of the two subtrees swapped, and
> not concurrency between the trees.  This means that it is still possible
> that subsequent reads to eg the booleans directory and the class directory
> during a policy load could see the old state for one and the new for the other.
> The problem of ensuring that policy loads are fully atomic from the perspective
> of userspace is larger than what is dealt with here.  This commit focuses on
> ensuring that the directories contents always match either the new or the old
> policy state from the perspective of userspace.
>
> In the previous implementation, on policy load /sys/fs/selinux is updated
> by deleting the previous contents of
> /sys/fs/selinux/{class,booleans} and then recreating them.  This means
> that there is a period of time when the contents of these directories do not
> exist which can cause race conditions as userspace relies on them for
> information about the policy.  In addition, it means that error recovery in
> the event of failure is challenging.
>
> In order to demonstrate the race condition that this series fixes, you
> can use the following commands:
>
> while true; do cat /sys/fs/selinux/class/service/perms/status
> >/dev/null; done &
> while true; do load_policy; done;
>
> In the existing code, this will display errors fairly often as the class
> lookup fails.  (In normal operation from systemd, this would result in a
> permission check which would be allowed or denied based on policy settings
> around unknown object classes.) After applying this patch series you
> should expect to no longer see such error messages.
>
> Signed-off-by: Daniel Burgener <dburgener@linux.microsoft.com>
> ---
>  security/selinux/selinuxfs.c | 113 ++++++++++++++++++++++++++++-------
>  1 file changed, 90 insertions(+), 23 deletions(-)

Merged into selinux/next, thanks!

-- 
paul moore
www.paul-moore.com

^ permalink raw reply	[flat|nested] 13+ messages in thread

end of thread, other threads:[~2020-08-21 13:46 UTC | newest]

Thread overview: 13+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-08-19 19:59 [PATCH v3 0/4] Update SELinuxfs out of tree and then swapover Daniel Burgener
2020-08-19 19:59 ` [PATCH v3 1/4] selinux: Create function for selinuxfs directory cleanup Daniel Burgener
2020-08-20 14:53   ` Stephen Smalley
2020-08-21 13:43   ` Paul Moore
2020-08-19 19:59 ` [PATCH v3 2/4] selinux: Refactor selinuxfs directory populating functions Daniel Burgener
2020-08-20 14:55   ` Stephen Smalley
2020-08-21 13:43   ` Paul Moore
2020-08-19 19:59 ` [PATCH v3 3/4] selinux: Standardize string literal usage for selinuxfs directory names Daniel Burgener
2020-08-20 14:56   ` Stephen Smalley
2020-08-21 13:44   ` Paul Moore
2020-08-19 19:59 ` [PATCH v3 4/4] selinux: Create new booleans and class dirs out of tree Daniel Burgener
2020-08-20 15:23   ` Stephen Smalley
2020-08-21 13:44   ` Paul Moore

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.