[PATCH v5 bpf-next 0/5] Enable writing xattr from BPF programs

[PATCH v5 bpf-next 0/5] Enable writing xattr from BPF programs

[Song Liu]

Add support to set and remove xattr from BPF program. Also add
security.bpf. xattr name prefix.

kfuncs are added to set and remove xattrs with security.bpf. name
prefix. Update kfuncs bpf_get_[file|dentry]_xattr to read xattrs
with security.bpf. name prefix. Note that BPF programs can read
user. xattrs, but not write and remove them.

Cover letter of v1 and v2:

Follow up discussion in LPC 2024 [1], that we need security.bpf xattr
prefix. This set adds "security.bpf." xattr name prefix, and allows
bpf kfuncs bpf_get_[file|dentry]_xattr() to read these xattrs.

[1] https://lpc.events/event/18/contributions/1940/

Changes v4 => v5
1. Let verifier pick proper kfunc (_locked or not _locked)  based on the
   calling context. (Alexei)
2. Remove the __failure test (6/6 of v4).

v4: https://lore.kernel.org/bpf/20241217063821.482857-1-song@kernel.org/

Changes v3 => v4
1. Do write permission check with inode locked. (Jan Kara)
2. Fix some source_inline warnings.

v3: https://lore.kernel.org/bpf/20241210220627.2800362-1-song@kernel.org/

Changes v2 => v3
1. Add kfuncs to set and remove xattr from BPF programs.

v2: https://lore.kernel.org/bpf/20241016070955.375923-1-song@kernel.org/

Changes v1 => v2
1. Update comment of bpf_get_[file|dentry]_xattr. (Jiri Olsa)
2. Fix comment for return value of bpf_get_[file|dentry]_xattr.

v1: https://lore.kernel.org/bpf/20241002214637.3625277-1-song@kernel.org/

Song Liu (5):
  fs/xattr: bpf: Introduce security.bpf. xattr name prefix
  selftests/bpf: Extend test fs_kfuncs to cover security.bpf. xattr
    names
  bpf: lsm: Add two more sleepable hooks
  bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs
  selftests/bpf: Test kfuncs that set and remove xattr from BPF programs

 fs/bpf_fs_kfuncs.c                            | 214 +++++++++++++++++-
 include/linux/bpf_lsm.h                       |   8 +
 include/uapi/linux/xattr.h                    |   4 +
 kernel/bpf/bpf_lsm.c                          |   2 +
 kernel/bpf/verifier.c                         |  55 ++++-
 tools/testing/selftests/bpf/bpf_kfuncs.h      |   5 +
 .../selftests/bpf/prog_tests/fs_kfuncs.c      | 162 ++++++++++++-
 .../selftests/bpf/progs/test_get_xattr.c      |  28 ++-
 .../bpf/progs/test_set_remove_xattr.c         | 133 +++++++++++
 9 files changed, 590 insertions(+), 21 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/progs/test_set_remove_xattr.c

--
2.43.5

[PATCH v5 bpf-next 1/5] fs/xattr: bpf: Introduce security.bpf. xattr name prefix

[Song Liu]

Introduct new xattr name prefix security.bpf., and enable reading these
xattrs from bpf kfuncs bpf_get_[file|dentry]_xattr().

As we are on it, correct the comments for return value of
bpf_get_[file|dentry]_xattr(), i.e. return length the xattr value on
success.

Signed-off-by: Song Liu <song@kernel.org>
Acked-by: Christian Brauner <brauner@kernel.org>
---
 fs/bpf_fs_kfuncs.c         | 19 ++++++++++++++-----
 include/uapi/linux/xattr.h |  4 ++++
 2 files changed, 18 insertions(+), 5 deletions(-)

diff --git a/fs/bpf_fs_kfuncs.c b/fs/bpf_fs_kfuncs.c
index 3fe9f59ef867..8a65184c8c2c 100644
--- a/fs/bpf_fs_kfuncs.c
+++ b/fs/bpf_fs_kfuncs.c
@@ -93,6 +93,11 @@ __bpf_kfunc int bpf_path_d_path(struct path *path, char *buf, size_t buf__sz)
 	return len;
 }
 
+static bool match_security_bpf_prefix(const char *name__str)
+{
+	return !strncmp(name__str, XATTR_NAME_BPF_LSM, XATTR_NAME_BPF_LSM_LEN);
+}
+
 /**
  * bpf_get_dentry_xattr - get xattr of a dentry
  * @dentry: dentry to get xattr from
@@ -101,9 +106,10 @@ __bpf_kfunc int bpf_path_d_path(struct path *path, char *buf, size_t buf__sz)
  *
  * Get xattr *name__str* of *dentry* and store the output in *value_ptr*.
  *
- * For security reasons, only *name__str* with prefix "user." is allowed.
+ * For security reasons, only *name__str* with prefix "user." or
+ * "security.bpf." is allowed.
  *
- * Return: 0 on success, a negative value on error.
+ * Return: length of the xattr value on success, a negative value on error.
  */
 __bpf_kfunc int bpf_get_dentry_xattr(struct dentry *dentry, const char *name__str,
 				     struct bpf_dynptr *value_p)
@@ -117,7 +123,9 @@ __bpf_kfunc int bpf_get_dentry_xattr(struct dentry *dentry, const char *name__st
 	if (WARN_ON(!inode))
 		return -EINVAL;
 
-	if (strncmp(name__str, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN))
+	/* Allow reading xattr with user. and security.bpf. prefix */
+	if (strncmp(name__str, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN) &&
+	    !match_security_bpf_prefix(name__str))
 		return -EPERM;
 
 	value_len = __bpf_dynptr_size(value_ptr);
@@ -139,9 +147,10 @@ __bpf_kfunc int bpf_get_dentry_xattr(struct dentry *dentry, const char *name__st
  *
  * Get xattr *name__str* of *file* and store the output in *value_ptr*.
  *
- * For security reasons, only *name__str* with prefix "user." is allowed.
+ * For security reasons, only *name__str* with prefix "user." or
+ * "security.bpf." is allowed.
  *
- * Return: 0 on success, a negative value on error.
+ * Return: length of the xattr value on success, a negative value on error.
  */
 __bpf_kfunc int bpf_get_file_xattr(struct file *file, const char *name__str,
 				   struct bpf_dynptr *value_p)
diff --git a/include/uapi/linux/xattr.h b/include/uapi/linux/xattr.h
index 9854f9cff3c6..c7c85bb504ba 100644
--- a/include/uapi/linux/xattr.h
+++ b/include/uapi/linux/xattr.h
@@ -83,6 +83,10 @@ struct xattr_args {
 #define XATTR_CAPS_SUFFIX "capability"
 #define XATTR_NAME_CAPS XATTR_SECURITY_PREFIX XATTR_CAPS_SUFFIX
 
+#define XATTR_BPF_LSM_SUFFIX "bpf."
+#define XATTR_NAME_BPF_LSM (XATTR_SECURITY_PREFIX XATTR_BPF_LSM_SUFFIX)
+#define XATTR_NAME_BPF_LSM_LEN (sizeof(XATTR_NAME_BPF_LSM) - 1)
+
 #define XATTR_POSIX_ACL_ACCESS  "posix_acl_access"
 #define XATTR_NAME_POSIX_ACL_ACCESS XATTR_SYSTEM_PREFIX XATTR_POSIX_ACL_ACCESS
 #define XATTR_POSIX_ACL_DEFAULT  "posix_acl_default"
-- 
2.43.5

[PATCH v5 bpf-next 2/5] selftests/bpf: Extend test fs_kfuncs to cover security.bpf. xattr names

[Song Liu]

Extend test_progs fs_kfuncs to cover different xattr names. Specifically:
xattr name "user.kfuncs" and "security.bpf.xxx" can be read from BPF
program with kfuncs bpf_get_[file|dentry]_xattr(); while "security.bpf"
and "security.selinux" cannot be read.

Signed-off-by: Song Liu <song@kernel.org>
---
 .../selftests/bpf/prog_tests/fs_kfuncs.c      | 37 ++++++++++++++-----
 .../selftests/bpf/progs/test_get_xattr.c      | 28 ++++++++++++--
 2 files changed, 51 insertions(+), 14 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c b/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
index 5a0b51157451..419f45b56472 100644
--- a/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
+++ b/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
@@ -12,7 +12,7 @@
 
 static const char testfile[] = "/tmp/test_progs_fs_kfuncs";
 
-static void test_xattr(void)
+static void test_get_xattr(const char *name, const char *value, bool allow_access)
 {
 	struct test_get_xattr *skel = NULL;
 	int fd = -1, err;
@@ -25,7 +25,7 @@ static void test_xattr(void)
 	close(fd);
 	fd = -1;
 
-	err = setxattr(testfile, "user.kfuncs", "hello", sizeof("hello"), 0);
+	err = setxattr(testfile, name, value, strlen(value) + 1, 0);
 	if (err && errno == EOPNOTSUPP) {
 		printf("%s:SKIP:local fs doesn't support xattr (%d)\n"
 		       "To run this test, make sure /tmp filesystem supports xattr.\n",
@@ -48,16 +48,23 @@ static void test_xattr(void)
 		goto out;
 
 	fd = open(testfile, O_RDONLY, 0644);
+
 	if (!ASSERT_GE(fd, 0, "open_file"))
 		goto out;
 
-	ASSERT_EQ(skel->bss->found_xattr_from_file, 1, "found_xattr_from_file");
-
 	/* Trigger security_inode_getxattr */
-	err = getxattr(testfile, "user.kfuncs", v, sizeof(v));
-	ASSERT_EQ(err, -1, "getxattr_return");
-	ASSERT_EQ(errno, EINVAL, "getxattr_errno");
-	ASSERT_EQ(skel->bss->found_xattr_from_dentry, 1, "found_xattr_from_dentry");
+	err = getxattr(testfile, name, v, sizeof(v));
+
+	if (allow_access) {
+		ASSERT_EQ(err, -1, "getxattr_return");
+		ASSERT_EQ(errno, EINVAL, "getxattr_errno");
+		ASSERT_EQ(skel->bss->found_xattr_from_file, 1, "found_xattr_from_file");
+		ASSERT_EQ(skel->bss->found_xattr_from_dentry, 1, "found_xattr_from_dentry");
+	} else {
+		ASSERT_EQ(err, strlen(value) + 1, "getxattr_return");
+		ASSERT_EQ(skel->bss->found_xattr_from_file, 0, "found_xattr_from_file");
+		ASSERT_EQ(skel->bss->found_xattr_from_dentry, 0, "found_xattr_from_dentry");
+	}
 
 out:
 	close(fd);
@@ -141,8 +148,18 @@ static void test_fsverity(void)
 
 void test_fs_kfuncs(void)
 {
-	if (test__start_subtest("xattr"))
-		test_xattr();
+	/* Matches xattr_names in progs/test_get_xattr.c */
+	if (test__start_subtest("user_xattr"))
+		test_get_xattr("user.kfuncs", "hello", true);
+
+	if (test__start_subtest("security_bpf_xattr"))
+		test_get_xattr("security.bpf.xxx", "hello", true);
+
+	if (test__start_subtest("security_bpf_xattr_error"))
+		test_get_xattr("security.bpf", "hello", false);
+
+	if (test__start_subtest("security_selinux_xattr_error"))
+		test_get_xattr("security.selinux", "hello", false);
 
 	if (test__start_subtest("fsverity"))
 		test_fsverity();
diff --git a/tools/testing/selftests/bpf/progs/test_get_xattr.c b/tools/testing/selftests/bpf/progs/test_get_xattr.c
index 66e737720f7c..358e3506e5b0 100644
--- a/tools/testing/selftests/bpf/progs/test_get_xattr.c
+++ b/tools/testing/selftests/bpf/progs/test_get_xattr.c
@@ -6,6 +6,7 @@
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
 #include "bpf_kfuncs.h"
+#include "bpf_misc.h"
 
 char _license[] SEC("license") = "GPL";
 
@@ -17,12 +18,23 @@ static const char expected_value[] = "hello";
 char value1[32];
 char value2[32];
 
+/* Matches caller of test_get_xattr() in prog_tests/fs_kfuncs.c */
+static const char * const xattr_names[] = {
+	/* The following work. */
+	"user.kfuncs",
+	"security.bpf.xxx",
+
+	/* The following do not work. */
+	"security.bpf",
+	"security.selinux"
+};
+
 SEC("lsm.s/file_open")
 int BPF_PROG(test_file_open, struct file *f)
 {
 	struct bpf_dynptr value_ptr;
 	__u32 pid;
-	int ret;
+	int ret, i;
 
 	pid = bpf_get_current_pid_tgid() >> 32;
 	if (pid != monitored_pid)
@@ -30,7 +42,11 @@ int BPF_PROG(test_file_open, struct file *f)
 
 	bpf_dynptr_from_mem(value1, sizeof(value1), 0, &value_ptr);
 
-	ret = bpf_get_file_xattr(f, "user.kfuncs", &value_ptr);
+	for (i = 0; i < ARRAY_SIZE(xattr_names); i++) {
+		ret = bpf_get_file_xattr(f, xattr_names[i], &value_ptr);
+		if (ret == sizeof(expected_value))
+			break;
+	}
 	if (ret != sizeof(expected_value))
 		return 0;
 	if (bpf_strncmp(value1, ret, expected_value))
@@ -44,7 +60,7 @@ int BPF_PROG(test_inode_getxattr, struct dentry *dentry, char *name)
 {
 	struct bpf_dynptr value_ptr;
 	__u32 pid;
-	int ret;
+	int ret, i;
 
 	pid = bpf_get_current_pid_tgid() >> 32;
 	if (pid != monitored_pid)
@@ -52,7 +68,11 @@ int BPF_PROG(test_inode_getxattr, struct dentry *dentry, char *name)
 
 	bpf_dynptr_from_mem(value2, sizeof(value2), 0, &value_ptr);
 
-	ret = bpf_get_dentry_xattr(dentry, "user.kfuncs", &value_ptr);
+	for (i = 0; i < ARRAY_SIZE(xattr_names); i++) {
+		ret = bpf_get_dentry_xattr(dentry, xattr_names[i], &value_ptr);
+		if (ret == sizeof(expected_value))
+			break;
+	}
 	if (ret != sizeof(expected_value))
 		return 0;
 	if (bpf_strncmp(value2, ret, expected_value))
-- 
2.43.5

[PATCH v5 bpf-next 3/5] bpf: lsm: Add two more sleepable hooks

[Song Liu]

Add bpf_lsm_inode_removexattr and bpf_lsm_inode_post_removexattr to list
sleepable_lsm_hooks. These two hooks are always called from sleepable
context.

Signed-off-by: Song Liu <song@kernel.org>
---
 kernel/bpf/bpf_lsm.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/kernel/bpf/bpf_lsm.c b/kernel/bpf/bpf_lsm.c
index 967492b65185..0a59df1c550a 100644
--- a/kernel/bpf/bpf_lsm.c
+++ b/kernel/bpf/bpf_lsm.c
@@ -316,7 +316,9 @@ BTF_ID(func, bpf_lsm_inode_getxattr)
 BTF_ID(func, bpf_lsm_inode_mknod)
 BTF_ID(func, bpf_lsm_inode_need_killpriv)
 BTF_ID(func, bpf_lsm_inode_post_setxattr)
+BTF_ID(func, bpf_lsm_inode_post_removexattr)
 BTF_ID(func, bpf_lsm_inode_readlink)
+BTF_ID(func, bpf_lsm_inode_removexattr)
 BTF_ID(func, bpf_lsm_inode_rename)
 BTF_ID(func, bpf_lsm_inode_rmdir)
 BTF_ID(func, bpf_lsm_inode_setattr)
-- 
2.43.5

[PATCH v5 bpf-next 4/5] bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs

[Song Liu]

Add the following kfuncs to set and remove xattrs from BPF programs:

  bpf_set_dentry_xattr
  bpf_remove_dentry_xattr
  bpf_set_dentry_xattr_locked
  bpf_remove_dentry_xattr_locked

The _locked version of these kfuncs are called from hooks where
dentry->d_inode is already locked. Instead of requiring the user
to know which version of the kfuncs to use, the verifier will pick
the proper kfunc based on the calling hook.

Signed-off-by: Song Liu <song@kernel.org>
---
 fs/bpf_fs_kfuncs.c      | 195 ++++++++++++++++++++++++++++++++++++++++
 include/linux/bpf_lsm.h |   8 ++
 kernel/bpf/verifier.c   |  55 +++++++++++-
 3 files changed, 256 insertions(+), 2 deletions(-)

diff --git a/fs/bpf_fs_kfuncs.c b/fs/bpf_fs_kfuncs.c
index 8a65184c8c2c..b3cc6330ab69 100644
--- a/fs/bpf_fs_kfuncs.c
+++ b/fs/bpf_fs_kfuncs.c
@@ -2,10 +2,12 @@
 /* Copyright (c) 2024 Google LLC. */
 
 #include <linux/bpf.h>
+#include <linux/bpf_lsm.h>
 #include <linux/btf.h>
 #include <linux/btf_ids.h>
 #include <linux/dcache.h>
 #include <linux/fs.h>
+#include <linux/fsnotify.h>
 #include <linux/file.h>
 #include <linux/mm.h>
 #include <linux/xattr.h>
@@ -161,6 +163,164 @@ __bpf_kfunc int bpf_get_file_xattr(struct file *file, const char *name__str,
 	return bpf_get_dentry_xattr(dentry, name__str, value_p);
 }
 
+static int bpf_xattr_write_permission(const char *name, struct inode *inode)
+{
+	if (WARN_ON(!inode))
+		return -EINVAL;
+
+	/* Only allow setting and removing security.bpf. xattrs */
+	if (!match_security_bpf_prefix(name))
+		return -EPERM;
+
+	return inode_permission(&nop_mnt_idmap, inode, MAY_WRITE);
+}
+
+static int __bpf_set_dentry_xattr(struct dentry *dentry, const char *name,
+				  const struct bpf_dynptr *value_p, int flags, bool lock_inode)
+{
+	struct bpf_dynptr_kern *value_ptr = (struct bpf_dynptr_kern *)value_p;
+	struct inode *inode = d_inode(dentry);
+	const void *value;
+	u32 value_len;
+	int ret;
+
+	value_len = __bpf_dynptr_size(value_ptr);
+	value = __bpf_dynptr_data(value_ptr, value_len);
+	if (!value)
+		return -EINVAL;
+
+	if (lock_inode)
+		inode_lock(inode);
+
+	ret = bpf_xattr_write_permission(name, inode);
+	if (ret)
+		goto out;
+
+	ret = __vfs_setxattr(&nop_mnt_idmap, dentry, inode, name,
+			     value, value_len, flags);
+	if (!ret) {
+		fsnotify_xattr(dentry);
+
+		/* This xattr is set by BPF LSM, so we do not call
+		 * security_inode_post_setxattr. This is the same as
+		 * security_inode_setsecurity().
+		 */
+	}
+out:
+	if (lock_inode)
+		inode_unlock(inode);
+	return ret;
+}
+
+/**
+ * bpf_set_dentry_xattr - set a xattr of a dentry
+ * @dentry: dentry to get xattr from
+ * @name__str: name of the xattr
+ * @value_p: xattr value
+ * @flags: flags to pass into filesystem operations
+ *
+ * Set xattr *name__str* of *dentry* to the value in *value_ptr*.
+ *
+ * For security reasons, only *name__str* with prefix "security.bpf."
+ * is allowed.
+ *
+ * The caller has not locked dentry->d_inode.
+ *
+ * Return: 0 on success, a negative value on error.
+ */
+__bpf_kfunc int bpf_set_dentry_xattr(struct dentry *dentry, const char *name__str,
+				     const struct bpf_dynptr *value_p, int flags)
+{
+	return __bpf_set_dentry_xattr(dentry, name__str, value_p, flags, true);
+}
+
+/**
+ * bpf_set_dentry_xattr_locked - set a xattr of a dentry
+ * @dentry: dentry to get xattr from
+ * @name__str: name of the xattr
+ * @value_p: xattr value
+ * @flags: flags to pass into filesystem operations
+ *
+ * Set xattr *name__str* of *dentry* to the value in *value_ptr*.
+ *
+ * For security reasons, only *name__str* with prefix "security.bpf."
+ * is allowed.
+ *
+ * The caller already locked dentry->d_inode.
+ *
+ * Return: 0 on success, a negative value on error.
+ */
+__bpf_kfunc int bpf_set_dentry_xattr_locked(struct dentry *dentry, const char *name__str,
+					    const struct bpf_dynptr *value_p, int flags)
+{
+	return __bpf_set_dentry_xattr(dentry, name__str, value_p, flags, false);
+}
+
+static int __bpf_remove_dentry_xattr(struct dentry *dentry, const char *name__str,
+				     bool lock_inode)
+{
+	struct inode *inode = d_inode(dentry);
+	int ret;
+
+	if (lock_inode)
+		inode_lock(inode);
+
+	ret = bpf_xattr_write_permission(name__str, inode);
+	if (ret)
+		goto out;
+
+	ret = __vfs_removexattr(&nop_mnt_idmap, dentry, name__str);
+	if (!ret) {
+		fsnotify_xattr(dentry);
+
+		/* This xattr is removed by BPF LSM, so we do not call
+		 * security_inode_post_removexattr.
+		 */
+	}
+out:
+	if (lock_inode)
+		inode_unlock(inode);
+	return ret;
+}
+
+/**
+ * bpf_remove_dentry_xattr - remove a xattr of a dentry
+ * @dentry: dentry to get xattr from
+ * @name__str: name of the xattr
+ *
+ * Rmove xattr *name__str* of *dentry*.
+ *
+ * For security reasons, only *name__str* with prefix "security.bpf."
+ * is allowed.
+ *
+ * The caller has not locked dentry->d_inode.
+ *
+ * Return: 0 on success, a negative value on error.
+ */
+__bpf_kfunc int bpf_remove_dentry_xattr(struct dentry *dentry, const char *name__str)
+{
+	return __bpf_remove_dentry_xattr(dentry, name__str, true);
+}
+
+/**
+ * bpf_remove_dentry_xattr_locked - remove a xattr of a dentry
+ * @dentry: dentry to get xattr from
+ * @name__str: name of the xattr
+ *
+ * Rmove xattr *name__str* of *dentry*.
+ *
+ * For security reasons, only *name__str* with prefix "security.bpf."
+ * is allowed.
+ *
+ * The caller already locked dentry->d_inode.
+ *
+ * Return: 0 on success, a negative value on error.
+ */
+__bpf_kfunc int bpf_remove_dentry_xattr_locked(struct dentry *dentry, const char *name__str)
+{
+	return __bpf_remove_dentry_xattr(dentry, name__str, false);
+}
+
 __bpf_kfunc_end_defs();
 
 BTF_KFUNCS_START(bpf_fs_kfunc_set_ids)
@@ -170,6 +330,10 @@ BTF_ID_FLAGS(func, bpf_put_file, KF_RELEASE)
 BTF_ID_FLAGS(func, bpf_path_d_path, KF_TRUSTED_ARGS)
 BTF_ID_FLAGS(func, bpf_get_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
 BTF_ID_FLAGS(func, bpf_get_file_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, bpf_set_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, bpf_remove_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, bpf_set_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, bpf_remove_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
 BTF_KFUNCS_END(bpf_fs_kfunc_set_ids)
 
 static int bpf_fs_kfuncs_filter(const struct bpf_prog *prog, u32 kfunc_id)
@@ -186,6 +350,37 @@ static const struct btf_kfunc_id_set bpf_fs_kfunc_set = {
 	.filter = bpf_fs_kfuncs_filter,
 };
 
+/* bpf_[set|remove]_dentry_xattr.* hooks have KF_TRUSTED_ARGS and
+ * KF_SLEEPABLE, so they are only available to sleepable hooks with
+ * dentry arguments.
+ *
+ * Setting and removing xattr requires exclusive lock on dentry->d_inode.
+ * Some hooks already locked d_inode, while some hooks have not locked
+ * d_inode. Therefore, we need different kfuncs for different hooks.
+ * Specifically, hooks in the following list (d_inode_locked_hooks)
+ * should call bpf_[set|remove]_dentry_xattr_locked; while other hooks
+ * should call bpf_[set|remove]_dentry_xattr.
+ */
+BTF_SET_START(d_inode_locked_hooks)
+BTF_ID(func, bpf_lsm_inode_post_removexattr)
+BTF_ID(func, bpf_lsm_inode_post_setattr)
+BTF_ID(func, bpf_lsm_inode_post_setxattr)
+BTF_ID(func, bpf_lsm_inode_removexattr)
+BTF_ID(func, bpf_lsm_inode_rmdir)
+BTF_ID(func, bpf_lsm_inode_setattr)
+BTF_ID(func, bpf_lsm_inode_setxattr)
+BTF_ID(func, bpf_lsm_inode_unlink)
+#ifdef CONFIG_SECURITY_PATH
+BTF_ID(func, bpf_lsm_path_unlink)
+BTF_ID(func, bpf_lsm_path_rmdir)
+#endif /* CONFIG_SECURITY_PATH */
+BTF_SET_END(d_inode_locked_hooks)
+
+bool bpf_lsm_has_d_inode_locked(const struct bpf_prog *prog)
+{
+	return btf_id_set_contains(&d_inode_locked_hooks, prog->aux->attach_btf_id);
+}
+
 static int __init bpf_fs_kfuncs_init(void)
 {
 	return register_btf_kfunc_id_set(BPF_PROG_TYPE_LSM, &bpf_fs_kfunc_set);
diff --git a/include/linux/bpf_lsm.h b/include/linux/bpf_lsm.h
index aefcd6564251..5147b10e16a2 100644
--- a/include/linux/bpf_lsm.h
+++ b/include/linux/bpf_lsm.h
@@ -48,6 +48,9 @@ void bpf_lsm_find_cgroup_shim(const struct bpf_prog *prog, bpf_func_t *bpf_func)
 
 int bpf_lsm_get_retval_range(const struct bpf_prog *prog,
 			     struct bpf_retval_range *range);
+
+bool bpf_lsm_has_d_inode_locked(const struct bpf_prog *prog);
+
 #else /* !CONFIG_BPF_LSM */
 
 static inline bool bpf_lsm_is_sleepable_hook(u32 btf_id)
@@ -86,6 +89,11 @@ static inline int bpf_lsm_get_retval_range(const struct bpf_prog *prog,
 {
 	return -EOPNOTSUPP;
 }
+
+static inline bool bpf_lsm_has_d_inode_locked(const struct bpf_prog *prog)
+{
+	return false;
+}
 #endif /* CONFIG_BPF_LSM */
 
 #endif /* _LINUX_BPF_LSM_H */
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index f27274e933e5..f0d240d46e54 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -206,6 +206,7 @@ static int ref_set_non_owning(struct bpf_verifier_env *env,
 static void specialize_kfunc(struct bpf_verifier_env *env,
 			     u32 func_id, u16 offset, unsigned long *addr);
 static bool is_trusted_reg(const struct bpf_reg_state *reg);
+static void remap_kfunc_locked_func_id(struct bpf_verifier_env *env, struct bpf_insn *insn);
 
 static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux)
 {
@@ -3224,10 +3225,12 @@ static int add_subprog_and_kfunc(struct bpf_verifier_env *env)
 			return -EPERM;
 		}
 
-		if (bpf_pseudo_func(insn) || bpf_pseudo_call(insn))
+		if (bpf_pseudo_func(insn) || bpf_pseudo_call(insn)) {
 			ret = add_subprog(env, i + insn->imm + 1);
-		else
+		} else {
+			remap_kfunc_locked_func_id(env, insn);
 			ret = add_kfunc_call(env, insn->imm, insn->off);
+		}
 
 		if (ret < 0)
 			return ret;
@@ -11690,6 +11693,10 @@ enum special_kfunc_type {
 	KF_bpf_get_kmem_cache,
 	KF_bpf_local_irq_save,
 	KF_bpf_local_irq_restore,
+	KF_bpf_set_dentry_xattr,
+	KF_bpf_remove_dentry_xattr,
+	KF_bpf_set_dentry_xattr_locked,
+	KF_bpf_remove_dentry_xattr_locked,
 };
 
 BTF_SET_START(special_kfunc_set)
@@ -11719,6 +11726,12 @@ BTF_ID(func, bpf_wq_set_callback_impl)
 #ifdef CONFIG_CGROUPS
 BTF_ID(func, bpf_iter_css_task_new)
 #endif
+#ifdef CONFIG_BPF_LSM
+BTF_ID(func, bpf_set_dentry_xattr)
+BTF_ID(func, bpf_remove_dentry_xattr)
+BTF_ID(func, bpf_set_dentry_xattr_locked)
+BTF_ID(func, bpf_remove_dentry_xattr_locked)
+#endif
 BTF_SET_END(special_kfunc_set)
 
 BTF_ID_LIST(special_kfunc_list)
@@ -11762,6 +11775,44 @@ BTF_ID_UNUSED
 BTF_ID(func, bpf_get_kmem_cache)
 BTF_ID(func, bpf_local_irq_save)
 BTF_ID(func, bpf_local_irq_restore)
+#ifdef CONFIG_BPF_LSM
+BTF_ID(func, bpf_set_dentry_xattr)
+BTF_ID(func, bpf_remove_dentry_xattr)
+BTF_ID(func, bpf_set_dentry_xattr_locked)
+BTF_ID(func, bpf_remove_dentry_xattr_locked)
+#else
+BTF_ID_UNUSED
+BTF_ID_UNUSED
+BTF_ID_UNUSED
+BTF_ID_UNUSED
+#endif
+
+/* Sometimes, we need slightly different verions of a kfunc for different
+ * contexts/hooks, for example, bpf_set_dentry_xattr vs.
+ * bpf_set_dentry_xattr_locked. The former kfunc need to lock the inode
+ * rwsem, while the latter is called with the inode rwsem held (by the
+ * caller).
+ *
+ * To avoid burden on the users, we allow either version of the kfunc in
+ * either context. Then the verifier will remap the kfunc to the proper
+ * version based the context.
+ */
+static void remap_kfunc_locked_func_id(struct bpf_verifier_env *env, struct bpf_insn *insn)
+{
+	u32 func_id = insn->imm;
+
+	if (bpf_lsm_has_d_inode_locked(env->prog)) {
+		if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr])
+			insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr_locked];
+		else if (func_id == special_kfunc_list[KF_bpf_remove_dentry_xattr])
+			insn->imm = special_kfunc_list[KF_bpf_remove_dentry_xattr_locked];
+	} else {
+		if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr_locked])
+			insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr];
+		else if (func_id == special_kfunc_list[KF_bpf_remove_dentry_xattr_locked])
+			insn->imm = special_kfunc_list[KF_bpf_remove_dentry_xattr];
+	}
+}
 
 static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta)
 {
-- 
2.43.5

[PATCH v5 bpf-next 5/5] selftests/bpf: Test kfuncs that set and remove xattr from BPF programs

[Song Liu]

Two sets of tests are added to exercise the not _locked and _locked
version of the kfuncs. For both tests, user space accesses xattr
security.bpf.foo on a testfile. The BPF program is triggered by user
space access (on LSM hook inode_[set|get]_xattr) and sets or removes
xattr security.bpf.bar. Then user space then validates that xattr
security.bpf.bar is set or removed as expected.

Note that, in both tests, the BPF programs use the not _locked kfuncs.
The verifier picks the proper kfuncs based on the calling context.

Signed-off-by: Song Liu <song@kernel.org>
---
 tools/testing/selftests/bpf/bpf_kfuncs.h      |   5 +
 .../selftests/bpf/prog_tests/fs_kfuncs.c      | 125 ++++++++++++++++
 .../bpf/progs/test_set_remove_xattr.c         | 133 ++++++++++++++++++
 3 files changed, 263 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/test_set_remove_xattr.c

diff --git a/tools/testing/selftests/bpf/bpf_kfuncs.h b/tools/testing/selftests/bpf/bpf_kfuncs.h
index 2eb3483f2fb0..8215c9b3115e 100644
--- a/tools/testing/selftests/bpf/bpf_kfuncs.h
+++ b/tools/testing/selftests/bpf/bpf_kfuncs.h
@@ -87,4 +87,9 @@ struct dentry;
  */
 extern int bpf_get_dentry_xattr(struct dentry *dentry, const char *name,
 			      struct bpf_dynptr *value_ptr) __ksym __weak;
+
+extern int bpf_set_dentry_xattr(struct dentry *dentry, const char *name__str,
+				const struct bpf_dynptr *value_p, int flags) __ksym __weak;
+extern int bpf_remove_dentry_xattr(struct dentry *dentry, const char *name__str) __ksym __weak;
+
 #endif
diff --git a/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c b/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
index 419f45b56472..43a26ec69a8e 100644
--- a/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
+++ b/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
@@ -8,6 +8,7 @@
 #include <unistd.h>
 #include <test_progs.h>
 #include "test_get_xattr.skel.h"
+#include "test_set_remove_xattr.skel.h"
 #include "test_fsverity.skel.h"
 
 static const char testfile[] = "/tmp/test_progs_fs_kfuncs";
@@ -72,6 +73,127 @@ static void test_get_xattr(const char *name, const char *value, bool allow_acces
 	remove(testfile);
 }
 
+/* xattr value we will set to security.bpf.foo */
+static const char value_foo[] = "hello";
+
+static void read_and_validate_foo(struct test_set_remove_xattr *skel)
+{
+	char value_out[32];
+	int err;
+
+	err = getxattr(testfile, skel->rodata->xattr_foo, value_out, sizeof(value_out));
+	ASSERT_EQ(err, sizeof(value_foo), "getxattr size foo");
+	ASSERT_EQ(strncmp(value_out, value_foo, sizeof(value_foo)), 0, "strncmp value_foo");
+}
+
+static void set_foo(struct test_set_remove_xattr *skel)
+{
+	ASSERT_OK(setxattr(testfile, skel->rodata->xattr_foo, value_foo, strlen(value_foo) + 1, 0),
+		  "setxattr foo");
+}
+
+static void validate_bar_match(struct test_set_remove_xattr *skel)
+{
+	char value_out[32];
+	int err;
+
+	err = getxattr(testfile, skel->rodata->xattr_bar, value_out, sizeof(value_out));
+	ASSERT_EQ(err, sizeof(skel->data->value_bar), "getxattr size bar");
+	ASSERT_EQ(strncmp(value_out, skel->data->value_bar, sizeof(skel->data->value_bar)), 0,
+		  "strncmp value_bar");
+}
+
+static void validate_bar_removed(struct test_set_remove_xattr *skel)
+{
+	char value_out[32];
+	int err;
+
+	err = getxattr(testfile, skel->rodata->xattr_bar, value_out, sizeof(value_out));
+	ASSERT_LT(err, 0, "getxattr size bar should fail");
+}
+
+static void test_set_remove_xattr(void)
+{
+	struct test_set_remove_xattr *skel = NULL;
+	int fd = -1, err;
+
+	fd = open(testfile, O_CREAT | O_RDONLY, 0644);
+	if (!ASSERT_GE(fd, 0, "create_file"))
+		return;
+
+	close(fd);
+	fd = -1;
+
+	skel = test_set_remove_xattr__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "test_set_remove_xattr__open_and_load"))
+		return;
+
+	/* Set security.bpf.foo to "hello" */
+	err = setxattr(testfile, skel->rodata->xattr_foo, value_foo, strlen(value_foo) + 1, 0);
+	if (err && errno == EOPNOTSUPP) {
+		printf("%s:SKIP:local fs doesn't support xattr (%d)\n"
+		       "To run this test, make sure /tmp filesystem supports xattr.\n",
+		       __func__, errno);
+		test__skip();
+		goto out;
+	}
+
+	if (!ASSERT_OK(err, "setxattr"))
+		goto out;
+
+	skel->bss->monitored_pid = getpid();
+	err = test_set_remove_xattr__attach(skel);
+	if (!ASSERT_OK(err, "test_set_remove_xattr__attach"))
+		goto out;
+
+	/* First, test not _locked version of the kfuncs with getxattr. */
+
+	/* Read security.bpf.foo and trigger test_inode_getxattr. This
+	 * bpf program will set security.bpf.bar to "world".
+	 */
+	read_and_validate_foo(skel);
+	validate_bar_match(skel);
+
+	/* Read security.bpf.foo and trigger test_inode_getxattr again.
+	 * This will remove xattr security.bpf.bar.
+	 */
+	read_and_validate_foo(skel);
+	validate_bar_removed(skel);
+
+	ASSERT_TRUE(skel->bss->set_security_bpf_bar_success, "set_security_bpf_bar_success");
+	ASSERT_TRUE(skel->bss->remove_security_bpf_bar_success, "remove_security_bpf_bar_success");
+	ASSERT_TRUE(skel->bss->set_security_selinux_fail, "set_security_selinux_fail");
+	ASSERT_TRUE(skel->bss->remove_security_selinux_fail, "remove_security_selinux_fail");
+
+	/* Second, test _locked version of the kfuncs, with setxattr */
+
+	/* Set security.bpf.foo and trigger test_inode_setxattr. This
+	 * bpf program will set security.bpf.bar to "world".
+	 */
+	set_foo(skel);
+	validate_bar_match(skel);
+
+	/* Set security.bpf.foo and trigger test_inode_setxattr again.
+	 * This will remove xattr security.bpf.bar.
+	 */
+	set_foo(skel);
+	validate_bar_removed(skel);
+
+	ASSERT_TRUE(skel->bss->locked_set_security_bpf_bar_success,
+		    "locked_set_security_bpf_bar_success");
+	ASSERT_TRUE(skel->bss->locked_remove_security_bpf_bar_success,
+		    "locked_remove_security_bpf_bar_success");
+	ASSERT_TRUE(skel->bss->locked_set_security_selinux_fail,
+		    "locked_set_security_selinux_fail");
+	ASSERT_TRUE(skel->bss->locked_remove_security_selinux_fail,
+		    "locked_remove_security_selinux_fail");
+
+out:
+	close(fd);
+	test_set_remove_xattr__destroy(skel);
+	remove(testfile);
+}
+
 #ifndef SHA256_DIGEST_SIZE
 #define SHA256_DIGEST_SIZE      32
 #endif
@@ -161,6 +283,9 @@ void test_fs_kfuncs(void)
 	if (test__start_subtest("security_selinux_xattr_error"))
 		test_get_xattr("security.selinux", "hello", false);
 
+	if (test__start_subtest("set_remove_xattr"))
+		test_set_remove_xattr();
+
 	if (test__start_subtest("fsverity"))
 		test_fsverity();
 }
diff --git a/tools/testing/selftests/bpf/progs/test_set_remove_xattr.c b/tools/testing/selftests/bpf/progs/test_set_remove_xattr.c
new file mode 100644
index 000000000000..e49be3cc4a33
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/test_set_remove_xattr.c
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2024 Meta Platforms, Inc. and affiliates. */
+
+#include "vmlinux.h"
+#include <errno.h>
+#include <bpf/bpf_tracing.h>
+#include "bpf_kfuncs.h"
+#include "bpf_misc.h"
+
+char _license[] SEC("license") = "GPL";
+
+__u32 monitored_pid;
+
+const char xattr_foo[] = "security.bpf.foo";
+const char xattr_bar[] = "security.bpf.bar";
+const char xattr_linux[] = "security.selinux";
+char value_bar[] = "world";
+char read_value[32];
+
+bool set_security_bpf_bar_success;
+bool remove_security_bpf_bar_success;
+bool set_security_selinux_fail;
+bool remove_security_selinux_fail;
+
+char name_buf[32];
+
+static inline bool name_match_foo(const char *name)
+{
+	bpf_probe_read_kernel(name_buf, sizeof(name_buf), name);
+
+	return !bpf_strncmp(name_buf, sizeof(xattr_foo), xattr_foo);
+}
+
+/* Test bpf_set_dentry_xattr and bpf_remove_dentry_xattr */
+SEC("lsm.s/inode_getxattr")
+int BPF_PROG(test_inode_getxattr, struct dentry *dentry, char *name)
+{
+	struct bpf_dynptr value_ptr;
+	__u32 pid;
+	int ret;
+
+	pid = bpf_get_current_pid_tgid() >> 32;
+	if (pid != monitored_pid)
+		return 0;
+
+	/* Only do the following for security.bpf.foo */
+	if (!name_match_foo(name))
+		return 0;
+
+	bpf_dynptr_from_mem(read_value, sizeof(read_value), 0, &value_ptr);
+
+	/* read security.bpf.bar */
+	ret = bpf_get_dentry_xattr(dentry, xattr_bar, &value_ptr);
+
+	if (ret < 0) {
+		/* If security.bpf.bar doesn't exist, set it */
+		bpf_dynptr_from_mem(value_bar, sizeof(value_bar), 0, &value_ptr);
+
+		ret = bpf_set_dentry_xattr(dentry, xattr_bar, &value_ptr, 0);
+		if (!ret)
+			set_security_bpf_bar_success = true;
+		ret = bpf_set_dentry_xattr(dentry, xattr_linux, &value_ptr, 0);
+		if (ret)
+			set_security_selinux_fail = true;
+	} else {
+		/* If security.bpf.bar exists, remove it */
+		ret = bpf_remove_dentry_xattr(dentry, xattr_bar);
+		if (!ret)
+			remove_security_bpf_bar_success = true;
+
+		ret = bpf_remove_dentry_xattr(dentry, xattr_linux);
+		if (ret)
+			remove_security_selinux_fail = true;
+	}
+
+	return 0;
+}
+
+bool locked_set_security_bpf_bar_success;
+bool locked_remove_security_bpf_bar_success;
+bool locked_set_security_selinux_fail;
+bool locked_remove_security_selinux_fail;
+
+/* Test bpf_set_dentry_xattr_locked and bpf_remove_dentry_xattr_locked.
+ * It not necessary to differentiate the _locked version and the
+ * not-_locked version in the BPF program. The verifier will fix them up
+ * properly.
+ */
+SEC("lsm.s/inode_setxattr")
+int BPF_PROG(test_inode_setxattr, struct mnt_idmap *idmap,
+	     struct dentry *dentry, const char *name,
+	     const void *value, size_t size, int flags)
+{
+	struct bpf_dynptr value_ptr;
+	__u32 pid;
+	int ret;
+
+	pid = bpf_get_current_pid_tgid() >> 32;
+	if (pid != monitored_pid)
+		return 0;
+
+	/* Only do the following for security.bpf.foo */
+	if (!name_match_foo(name))
+		return 0;
+
+	bpf_dynptr_from_mem(read_value, sizeof(read_value), 0, &value_ptr);
+
+	/* read security.bpf.bar */
+	ret = bpf_get_dentry_xattr(dentry, xattr_bar, &value_ptr);
+
+	if (ret < 0) {
+		/* If security.bpf.bar doesn't exist, set it */
+		bpf_dynptr_from_mem(value_bar, sizeof(value_bar), 0, &value_ptr);
+
+		ret = bpf_set_dentry_xattr(dentry, xattr_bar, &value_ptr, 0);
+		if (!ret)
+			locked_set_security_bpf_bar_success = true;
+		ret = bpf_set_dentry_xattr(dentry, xattr_linux, &value_ptr, 0);
+		if (ret)
+			locked_set_security_selinux_fail = true;
+	} else {
+		/* If security.bpf.bar exists, remove it */
+		ret = bpf_remove_dentry_xattr(dentry, xattr_bar);
+		if (!ret)
+			locked_remove_security_bpf_bar_success = true;
+
+		ret = bpf_remove_dentry_xattr(dentry, xattr_linux);
+		if (ret)
+			locked_remove_security_selinux_fail = true;
+	}
+
+	return 0;
+}
-- 
2.43.5

Re: [PATCH v5 bpf-next 4/5] bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs

[Alexei Starovoitov]

On Tue, Dec 17, 2024 at 8:48 PM Song Liu <song@kernel.org> wrote:
>
>
>  BTF_KFUNCS_START(bpf_fs_kfunc_set_ids)
> @@ -170,6 +330,10 @@ BTF_ID_FLAGS(func, bpf_put_file, KF_RELEASE)
>  BTF_ID_FLAGS(func, bpf_path_d_path, KF_TRUSTED_ARGS)
>  BTF_ID_FLAGS(func, bpf_get_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>  BTF_ID_FLAGS(func, bpf_get_file_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>  BTF_KFUNCS_END(bpf_fs_kfunc_set_ids)

The _locked() versions shouldn't be exposed to bpf prog.
Don't add them to the above set.

Also we need to somehow exclude them from being dumped into vmlinux.h

>  static int bpf_fs_kfuncs_filter(const struct bpf_prog *prog, u32 kfunc_id)
> @@ -186,6 +350,37 @@ static const struct btf_kfunc_id_set bpf_fs_kfunc_set = {
>         .filter = bpf_fs_kfuncs_filter,
>  };
>
> +/* bpf_[set|remove]_dentry_xattr.* hooks have KF_TRUSTED_ARGS and
> + * KF_SLEEPABLE, so they are only available to sleepable hooks with
> + * dentry arguments.
> + *
> + * Setting and removing xattr requires exclusive lock on dentry->d_inode.
> + * Some hooks already locked d_inode, while some hooks have not locked
> + * d_inode. Therefore, we need different kfuncs for different hooks.
> + * Specifically, hooks in the following list (d_inode_locked_hooks)
> + * should call bpf_[set|remove]_dentry_xattr_locked; while other hooks
> + * should call bpf_[set|remove]_dentry_xattr.
> + */
> +BTF_SET_START(d_inode_locked_hooks)
> +BTF_ID(func, bpf_lsm_inode_post_removexattr)
> +BTF_ID(func, bpf_lsm_inode_post_setattr)
> +BTF_ID(func, bpf_lsm_inode_post_setxattr)
> +BTF_ID(func, bpf_lsm_inode_removexattr)
> +BTF_ID(func, bpf_lsm_inode_rmdir)
> +BTF_ID(func, bpf_lsm_inode_setattr)
> +BTF_ID(func, bpf_lsm_inode_setxattr)
> +BTF_ID(func, bpf_lsm_inode_unlink)
> +#ifdef CONFIG_SECURITY_PATH
> +BTF_ID(func, bpf_lsm_path_unlink)
> +BTF_ID(func, bpf_lsm_path_rmdir)
> +#endif /* CONFIG_SECURITY_PATH */
> +BTF_SET_END(d_inode_locked_hooks)
> +
> +bool bpf_lsm_has_d_inode_locked(const struct bpf_prog *prog)
> +{
> +       return btf_id_set_contains(&d_inode_locked_hooks, prog->aux->attach_btf_id);
> +}
> +
>  static int __init bpf_fs_kfuncs_init(void)
>  {
>         return register_btf_kfunc_id_set(BPF_PROG_TYPE_LSM, &bpf_fs_kfunc_set);
> diff --git a/include/linux/bpf_lsm.h b/include/linux/bpf_lsm.h
> index aefcd6564251..5147b10e16a2 100644
> --- a/include/linux/bpf_lsm.h
> +++ b/include/linux/bpf_lsm.h
> @@ -48,6 +48,9 @@ void bpf_lsm_find_cgroup_shim(const struct bpf_prog *prog, bpf_func_t *bpf_func)
>
>  int bpf_lsm_get_retval_range(const struct bpf_prog *prog,
>                              struct bpf_retval_range *range);
> +
> +bool bpf_lsm_has_d_inode_locked(const struct bpf_prog *prog);
> +
>  #else /* !CONFIG_BPF_LSM */
>
>  static inline bool bpf_lsm_is_sleepable_hook(u32 btf_id)
> @@ -86,6 +89,11 @@ static inline int bpf_lsm_get_retval_range(const struct bpf_prog *prog,
>  {
>         return -EOPNOTSUPP;
>  }
> +
> +static inline bool bpf_lsm_has_d_inode_locked(const struct bpf_prog *prog)
> +{
> +       return false;
> +}
>  #endif /* CONFIG_BPF_LSM */
>
>  #endif /* _LINUX_BPF_LSM_H */
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index f27274e933e5..f0d240d46e54 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -206,6 +206,7 @@ static int ref_set_non_owning(struct bpf_verifier_env *env,
>  static void specialize_kfunc(struct bpf_verifier_env *env,
>                              u32 func_id, u16 offset, unsigned long *addr);
>  static bool is_trusted_reg(const struct bpf_reg_state *reg);
> +static void remap_kfunc_locked_func_id(struct bpf_verifier_env *env, struct bpf_insn *insn);
>
>  static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux)
>  {
> @@ -3224,10 +3225,12 @@ static int add_subprog_and_kfunc(struct bpf_verifier_env *env)
>                         return -EPERM;
>                 }
>
> -               if (bpf_pseudo_func(insn) || bpf_pseudo_call(insn))
> +               if (bpf_pseudo_func(insn) || bpf_pseudo_call(insn)) {
>                         ret = add_subprog(env, i + insn->imm + 1);
> -               else
> +               } else {
> +                       remap_kfunc_locked_func_id(env, insn);
>                         ret = add_kfunc_call(env, insn->imm, insn->off);
> +               }
>
>                 if (ret < 0)
>                         return ret;
> @@ -11690,6 +11693,10 @@ enum special_kfunc_type {
>         KF_bpf_get_kmem_cache,
>         KF_bpf_local_irq_save,
>         KF_bpf_local_irq_restore,
> +       KF_bpf_set_dentry_xattr,
> +       KF_bpf_remove_dentry_xattr,
> +       KF_bpf_set_dentry_xattr_locked,
> +       KF_bpf_remove_dentry_xattr_locked,
>  };
>
>  BTF_SET_START(special_kfunc_set)
> @@ -11719,6 +11726,12 @@ BTF_ID(func, bpf_wq_set_callback_impl)
>  #ifdef CONFIG_CGROUPS
>  BTF_ID(func, bpf_iter_css_task_new)
>  #endif
> +#ifdef CONFIG_BPF_LSM
> +BTF_ID(func, bpf_set_dentry_xattr)
> +BTF_ID(func, bpf_remove_dentry_xattr)
> +BTF_ID(func, bpf_set_dentry_xattr_locked)
> +BTF_ID(func, bpf_remove_dentry_xattr_locked)
> +#endif
>  BTF_SET_END(special_kfunc_set)

Do they need to be a part of special_kfunc_set ?
Where is the code that uses that?

>
>  BTF_ID_LIST(special_kfunc_list)
> @@ -11762,6 +11775,44 @@ BTF_ID_UNUSED
>  BTF_ID(func, bpf_get_kmem_cache)
>  BTF_ID(func, bpf_local_irq_save)
>  BTF_ID(func, bpf_local_irq_restore)
> +#ifdef CONFIG_BPF_LSM
> +BTF_ID(func, bpf_set_dentry_xattr)
> +BTF_ID(func, bpf_remove_dentry_xattr)
> +BTF_ID(func, bpf_set_dentry_xattr_locked)
> +BTF_ID(func, bpf_remove_dentry_xattr_locked)
> +#else
> +BTF_ID_UNUSED
> +BTF_ID_UNUSED
> +BTF_ID_UNUSED
> +BTF_ID_UNUSED
> +#endif
> +
> +/* Sometimes, we need slightly different verions of a kfunc for different

versions

> + * contexts/hooks, for example, bpf_set_dentry_xattr vs.
> + * bpf_set_dentry_xattr_locked. The former kfunc need to lock the inode
> + * rwsem, while the latter is called with the inode rwsem held (by the
> + * caller).
> + *
> + * To avoid burden on the users, we allow either version of the kfunc in
> + * either context. Then the verifier will remap the kfunc to the proper
> + * version based the context.
> + */
> +static void remap_kfunc_locked_func_id(struct bpf_verifier_env *env, struct bpf_insn *insn)
> +{
> +       u32 func_id = insn->imm;
> +
> +       if (bpf_lsm_has_d_inode_locked(env->prog)) {
> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr])
> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr_locked];
> +               else if (func_id == special_kfunc_list[KF_bpf_remove_dentry_xattr])
> +                       insn->imm = special_kfunc_list[KF_bpf_remove_dentry_xattr_locked];
> +       } else {
> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr_locked])
> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr];

This part is not necessary.
_locked() shouldn't be exposed and it should be an error
if bpf prog attempts to use invalid kfunc.

> +               else if (func_id == special_kfunc_list[KF_bpf_remove_dentry_xattr_locked])
> +                       insn->imm = special_kfunc_list[KF_bpf_remove_dentry_xattr];
> +       }
> +}
>
>  static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta)
>  {
> --
> 2.43.5
>

Re: [PATCH v5 bpf-next 4/5] bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs

[Song Liu]

Hi Alexei, 

Thanks for the review!

> On Dec 18, 2024, at 1:20 PM, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> 
> On Tue, Dec 17, 2024 at 8:48 PM Song Liu <song@kernel.org> wrote:
>> 
>> 
>> BTF_KFUNCS_START(bpf_fs_kfunc_set_ids)
>> @@ -170,6 +330,10 @@ BTF_ID_FLAGS(func, bpf_put_file, KF_RELEASE)
>> BTF_ID_FLAGS(func, bpf_path_d_path, KF_TRUSTED_ARGS)
>> BTF_ID_FLAGS(func, bpf_get_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>> BTF_ID_FLAGS(func, bpf_get_file_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>> BTF_KFUNCS_END(bpf_fs_kfunc_set_ids)
> 
> The _locked() versions shouldn't be exposed to bpf prog.
> Don't add them to the above set.
> 
> Also we need to somehow exclude them from being dumped into vmlinux.h
> 
>> static int bpf_fs_kfuncs_filter(const struct bpf_prog *prog, u32 kfunc_id)
>> @@ -186,6 +350,37 @@ static const struct btf_kfunc_id_set bpf_fs_kfunc_set = {
>>        .filter = bpf_fs_kfuncs_filter,
>> };

[...]

>> + */
>> +static void remap_kfunc_locked_func_id(struct bpf_verifier_env *env, struct bpf_insn *insn)
>> +{
>> +       u32 func_id = insn->imm;
>> +
>> +       if (bpf_lsm_has_d_inode_locked(env->prog)) {
>> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr])
>> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr_locked];
>> +               else if (func_id == special_kfunc_list[KF_bpf_remove_dentry_xattr])
>> +                       insn->imm = special_kfunc_list[KF_bpf_remove_dentry_xattr_locked];
>> +       } else {
>> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr_locked])
>> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr];
> 
> This part is not necessary.
> _locked() shouldn't be exposed and it should be an error
> if bpf prog attempts to use invalid kfunc.

I was implementing this in different way than the solution you and Kumar
suggested. Instead of updating this in add_kfunc_call, check_kfunc_call, 
and fixup_kfunc_call, remap_kfunc_locked_func_id happens before 
add_kfunc_call. Then, for the rest of the process, the verifier handles
_locked version and not _locked version as two different kfuncs. This is
why we need the _locked version in bpf_fs_kfunc_set_ids. I personally 
think this approach is a lot cleaner. 

I think the missing piece is to exclude the _locked version from 
vmlinux.h. Maybe we can achieve this by adding a different DECL_TAG 
to these kfuncs?

Does this make sense?

Thanks,
Song

Re: [PATCH v5 bpf-next 4/5] bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs

[Song Liu]

> On Dec 18, 2024, at 1:47 PM, Song Liu <songliubraving@meta.com> wrote:
> 
> Hi Alexei, 
> 
> Thanks for the review!
> 
>> On Dec 18, 2024, at 1:20 PM, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
>> 
>> On Tue, Dec 17, 2024 at 8:48 PM Song Liu <song@kernel.org> wrote:
>>> 
>>> 
>>> BTF_KFUNCS_START(bpf_fs_kfunc_set_ids)
>>> @@ -170,6 +330,10 @@ BTF_ID_FLAGS(func, bpf_put_file, KF_RELEASE)
>>> BTF_ID_FLAGS(func, bpf_path_d_path, KF_TRUSTED_ARGS)
>>> BTF_ID_FLAGS(func, bpf_get_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>>> BTF_ID_FLAGS(func, bpf_get_file_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>>> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>>> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>>> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>>> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
>>> BTF_KFUNCS_END(bpf_fs_kfunc_set_ids)
>> 
>> The _locked() versions shouldn't be exposed to bpf prog.
>> Don't add them to the above set.
>> 
>> Also we need to somehow exclude them from being dumped into vmlinux.h
>> 
>>> static int bpf_fs_kfuncs_filter(const struct bpf_prog *prog, u32 kfunc_id)
>>> @@ -186,6 +350,37 @@ static const struct btf_kfunc_id_set bpf_fs_kfunc_set = {
>>>       .filter = bpf_fs_kfuncs_filter,
>>> };
> 
> [...]
> 
>>> + */
>>> +static void remap_kfunc_locked_func_id(struct bpf_verifier_env *env, struct bpf_insn *insn)
>>> +{
>>> +       u32 func_id = insn->imm;
>>> +
>>> +       if (bpf_lsm_has_d_inode_locked(env->prog)) {
>>> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr])
>>> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr_locked];
>>> +               else if (func_id == special_kfunc_list[KF_bpf_remove_dentry_xattr])
>>> +                       insn->imm = special_kfunc_list[KF_bpf_remove_dentry_xattr_locked];
>>> +       } else {
>>> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr_locked])
>>> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr];
>> 
>> This part is not necessary.
>> _locked() shouldn't be exposed and it should be an error
>> if bpf prog attempts to use invalid kfunc.
> 
> I was implementing this in different way than the solution you and Kumar
> suggested. Instead of updating this in add_kfunc_call, check_kfunc_call, 
> and fixup_kfunc_call, remap_kfunc_locked_func_id happens before 
> add_kfunc_call. Then, for the rest of the process, the verifier handles
> _locked version and not _locked version as two different kfuncs. This is
> why we need the _locked version in bpf_fs_kfunc_set_ids. I personally 
> think this approach is a lot cleaner. 
> 
> I think the missing piece is to exclude the _locked version from 
> vmlinux.h. Maybe we can achieve this by adding a different DECL_TAG 
> to these kfuncs?

Looked into the code, I think it is doable:

1. Extend struct btf_kfunc_id_set with "struct btf_id_set8 *shadow_set",
   or a different name;
2. Add _locked kfuncs to shadow_set, and these kfuncs will not have 
   BTF_SET8_KFUNCS set. Then pahole will not generate DECL_TAG of 
   "bpf_kfunc" for these. 
3. __btf_kfunc_id_set_contains() will need to look up id from shadow_set.
   And the filter function needs to handle shadow_set. 

Does this sound sane? 

Thanks,
Song

Re: [PATCH v5 bpf-next 4/5] bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs

[Alexei Starovoitov]

On Wed, Dec 18, 2024 at 1:47 PM Song Liu <songliubraving@meta.com> wrote:
>
> Hi Alexei,
>
> Thanks for the review!
>
> > On Dec 18, 2024, at 1:20 PM, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> >
> > On Tue, Dec 17, 2024 at 8:48 PM Song Liu <song@kernel.org> wrote:
> >>
> >>
> >> BTF_KFUNCS_START(bpf_fs_kfunc_set_ids)
> >> @@ -170,6 +330,10 @@ BTF_ID_FLAGS(func, bpf_put_file, KF_RELEASE)
> >> BTF_ID_FLAGS(func, bpf_path_d_path, KF_TRUSTED_ARGS)
> >> BTF_ID_FLAGS(func, bpf_get_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> >> BTF_ID_FLAGS(func, bpf_get_file_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> >> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> >> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> >> +BTF_ID_FLAGS(func, bpf_set_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> >> +BTF_ID_FLAGS(func, bpf_remove_dentry_xattr_locked, KF_SLEEPABLE | KF_TRUSTED_ARGS)
> >> BTF_KFUNCS_END(bpf_fs_kfunc_set_ids)
> >
> > The _locked() versions shouldn't be exposed to bpf prog.
> > Don't add them to the above set.
> >
> > Also we need to somehow exclude them from being dumped into vmlinux.h
> >
> >> static int bpf_fs_kfuncs_filter(const struct bpf_prog *prog, u32 kfunc_id)
> >> @@ -186,6 +350,37 @@ static const struct btf_kfunc_id_set bpf_fs_kfunc_set = {
> >>        .filter = bpf_fs_kfuncs_filter,
> >> };
>
> [...]
>
> >> + */
> >> +static void remap_kfunc_locked_func_id(struct bpf_verifier_env *env, struct bpf_insn *insn)
> >> +{
> >> +       u32 func_id = insn->imm;
> >> +
> >> +       if (bpf_lsm_has_d_inode_locked(env->prog)) {
> >> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr])
> >> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr_locked];
> >> +               else if (func_id == special_kfunc_list[KF_bpf_remove_dentry_xattr])
> >> +                       insn->imm = special_kfunc_list[KF_bpf_remove_dentry_xattr_locked];
> >> +       } else {
> >> +               if (func_id == special_kfunc_list[KF_bpf_set_dentry_xattr_locked])
> >> +                       insn->imm =  special_kfunc_list[KF_bpf_set_dentry_xattr];
> >
> > This part is not necessary.
> > _locked() shouldn't be exposed and it should be an error
> > if bpf prog attempts to use invalid kfunc.
>
> I was implementing this in different way than the solution you and Kumar
> suggested. Instead of updating this in add_kfunc_call, check_kfunc_call,
> and fixup_kfunc_call, remap_kfunc_locked_func_id happens before
> add_kfunc_call. Then, for the rest of the process, the verifier handles
> _locked version and not _locked version as two different kfuncs. This is
> why we need the _locked version in bpf_fs_kfunc_set_ids. I personally
> think this approach is a lot cleaner.

I see. Blind rewrite in add_kfunc_call() looks simpler,
but allowing progs call _locked() version directly is not clean.

See specialize_kfunc() as an existing approach that does polymorphism.

_locked() doesn't need to be __bpf_kfunc annotated.
It can be just like bpf_dynptr_from_skb_rdonly.

There will be no issue with vmlinux.h as well.

Re: [PATCH v5 bpf-next 4/5] bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs

[Song Liu]

> On Dec 18, 2024, at 4:17 PM, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:

[...]

>>> This part is not necessary.
>>> _locked() shouldn't be exposed and it should be an error
>>> if bpf prog attempts to use invalid kfunc.
>> 
>> I was implementing this in different way than the solution you and Kumar
>> suggested. Instead of updating this in add_kfunc_call, check_kfunc_call,
>> and fixup_kfunc_call, remap_kfunc_locked_func_id happens before
>> add_kfunc_call. Then, for the rest of the process, the verifier handles
>> _locked version and not _locked version as two different kfuncs. This is
>> why we need the _locked version in bpf_fs_kfunc_set_ids. I personally
>> think this approach is a lot cleaner.
> 
> I see. Blind rewrite in add_kfunc_call() looks simpler,
> but allowing progs call _locked() version directly is not clean.

Agreed. 

> 
> See specialize_kfunc() as an existing approach that does polymorphism.
> 
> _locked() doesn't need to be __bpf_kfunc annotated.
> It can be just like bpf_dynptr_from_skb_rdonly.

I am thinking about a more modular approach. Instead of pushing the
polymorphism logic to verifer.c, we can have each btf_kfunc_id_set 
handle the remap of its kfuncs. Specifically, we can extend 
btf_kfunc_id_set as:

typedef u32 (*btf_kfunc_remap_t)(const struct bpf_prog *prog, u32 kfunc_id);

struct btf_kfunc_id_set {
        struct module *owner;
        struct btf_id_set8 *set;
        /* hidden_set contains kfuncs that are not marked as kfunc in
         * vmlinux.h. These kfuncs are usually a variation of a kfunc
         * in @set.
         */
        struct btf_id_set8 *hidden_set;
        btf_kfunc_filter_t filter;
        /* @remap method matches kfuncs in @set to proper version in
         * @hidden_set.
         */
        btf_kfunc_remap_t remap;
};

In this case, not_locked version of kfuncs will be added to @set;
while _locked kfuncs will be added to @hidden_set. @hidden_set 
will not be exposed in vmlinux.h. Then the new remap method is 
used to map not_locked kfuncs to _locked kfuncs for inode-locked 
context. 

We can also move bpf_dynptr_from_skb_rdonly to this model, and 
simplify specialize_kfunc(). 

I will send patch for this version for review. 

Thanks,
Song