x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 153 insertions, 0 deletions

diff --git a/fs/orangefs/acl.c b/fs/orangefs/acl.c
new file mode 100644
index 000000000..a01ef0c1b
--- /dev/null
+++ b/fs/orangefs/acl.c
@@ -0,0 +1,153 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * (C) 2001 Clemson University and The University of Chicago
+ *
+ * See COPYING in top-level directory.
+ */
+
+#include "protocol.h"
+#include "orangefs-kernel.h"
+#include "orangefs-bufmap.h"
+#include <linux/posix_acl_xattr.h>
+
+struct posix_acl *orangefs_get_acl(struct inode *inode, int type, bool rcu)
+{
+	struct posix_acl *acl;
+	int ret;
+	char *key = NULL, *value = NULL;
+
+	if (rcu)
+		return ERR_PTR(-ECHILD);
+
+	switch (type) {
+	case ACL_TYPE_ACCESS:
+		key = XATTR_NAME_POSIX_ACL_ACCESS;
+		break;
+	case ACL_TYPE_DEFAULT:
+		key = XATTR_NAME_POSIX_ACL_DEFAULT;
+		break;
+	default:
+		gossip_err("orangefs_get_acl: bogus value of type %d\n", type);
+		return ERR_PTR(-EINVAL);
+	}
+	/*
+	 * Rather than incurring a network call just to determine the exact
+	 * length of the attribute, I just allocate a max length to save on
+	 * the network call. Conceivably, we could pass NULL to
+	 * orangefs_inode_getxattr() to probe the length of the value, but
+	 * I don't do that for now.
+	 */
+	value = kmalloc(ORANGEFS_MAX_XATTR_VALUELEN, GFP_KERNEL);
+	if (!value)
+		return ERR_PTR(-ENOMEM);
+
+	gossip_debug(GOSSIP_ACL_DEBUG,
+		     "inode %pU, key %s, type %d\n",
+		     get_khandle_from_ino(inode),
+		     key,
+		     type);
+	ret = orangefs_inode_getxattr(inode, key, value,
+				      ORANGEFS_MAX_XATTR_VALUELEN);
+	/* if the key exists, convert it to an in-memory rep */
+	if (ret > 0) {
+		acl = posix_acl_from_xattr(&init_user_ns, value, ret);
+	} else if (ret == -ENODATA || ret == -ENOSYS) {
+		acl = NULL;
+	} else {
+		gossip_err("inode %pU retrieving acl's failed with error %d\n",
+			   get_khandle_from_ino(inode),
+			   ret);
+		acl = ERR_PTR(ret);
+	}
+	/* kfree(NULL) is safe, so don't worry if value ever got used */
+	kfree(value);
+	return acl;
+}
+
+int __orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
+{
+	int error = 0;
+	void *value = NULL;
+	size_t size = 0;
+	const char *name = NULL;
+
+	switch (type) {
+	case ACL_TYPE_ACCESS:
+		name = XATTR_NAME_POSIX_ACL_ACCESS;
+		break;
+	case ACL_TYPE_DEFAULT:
+		name = XATTR_NAME_POSIX_ACL_DEFAULT;
+		break;
+	default:
+		gossip_err("%s: invalid type %d!\n", __func__, type);
+		return -EINVAL;
+	}
+
+	gossip_debug(GOSSIP_ACL_DEBUG,
+		     "%s: inode %pU, key %s type %d\n",
+		     __func__, get_khandle_from_ino(inode),
+		     name,
+		     type);
+
+	if (acl) {
+		value = posix_acl_to_xattr(&init_user_ns, acl, &size, GFP_KERNEL);
+		if (!value)
+			return -ENOMEM;
+	}
+
+	gossip_debug(GOSSIP_ACL_DEBUG,
+		     "%s: name %s, value %p, size %zd, acl %p\n",
+		     __func__, name, value, size, acl);
+	/*
+	 * Go ahead and set the extended attribute now. NOTE: Suppose acl
+	 * was NULL, then value will be NULL and size will be 0 and that
+	 * will xlate to a removexattr. However, we don't want removexattr
+	 * complain if attributes does not exist.
+	 */
+	error = orangefs_inode_setxattr(inode, name, value, size, 0);
+
+	kfree(value);
+	if (!error)
+		set_cached_acl(inode, type, acl);
+	return error;
+}
+
+int orangefs_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
+		     struct posix_acl *acl, int type)
+{
+	int error;
+	struct iattr iattr;
+	int rc;
+	struct inode *inode = d_inode(dentry);
+
+	memset(&iattr, 0, sizeof iattr);
+
+	if (type == ACL_TYPE_ACCESS && acl) {
+		/*
+		 * posix_acl_update_mode checks to see if the permissions
+		 * described by the ACL can be encoded into the
+		 * object's mode. If so, it sets "acl" to NULL
+		 * and "mode" to the new desired value. It is up to
+		 * us to propagate the new mode back to the server...
+		 */
+		error = posix_acl_update_mode(&nop_mnt_idmap, inode,
+					      &iattr.ia_mode, &acl);
+		if (error) {
+			gossip_err("%s: posix_acl_update_mode err: %d\n",
+				   __func__,
+				   error);
+			return error;
+		}
+
+		if (inode->i_mode != iattr.ia_mode)
+			iattr.ia_valid = ATTR_MODE;
+
+	}
+
+	rc = __orangefs_set_acl(inode, acl, type);
+
+	if (!rc && (iattr.ia_valid == ATTR_MODE))
+		rc = __orangefs_setattr_mode(dentry, &iattr);
+
+	return rc;
+}