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, 290 insertions, 0 deletions

diff --git a/drivers/char/tpm/tpm-dev-common.c b/drivers/char/tpm/tpm-dev-common.c
new file mode 100644
index 000000000..f942c0c8e
--- /dev/null
+++ b/drivers/char/tpm/tpm-dev-common.c
@@ -0,0 +1,290 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2004 IBM Corporation
+ * Authors:
+ * Leendert van Doorn <leendert@watson.ibm.com>
+ * Dave Safford <safford@watson.ibm.com>
+ * Reiner Sailer <sailer@watson.ibm.com>
+ * Kylene Hall <kjhall@us.ibm.com>
+ *
+ * Copyright (C) 2013 Obsidian Research Corp
+ * Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
+ *
+ * Device file system interface to the TPM
+ */
+#include <linux/poll.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/workqueue.h>
+#include "tpm.h"
+#include "tpm-dev.h"
+
+static struct workqueue_struct *tpm_dev_wq;
+
+static ssize_t tpm_dev_transmit(struct tpm_chip *chip, struct tpm_space *space,
+				u8 *buf, size_t bufsiz)
+{
+	struct tpm_header *header = (void *)buf;
+	ssize_t ret, len;
+
+	if (chip->flags & TPM_CHIP_FLAG_TPM2)
+		tpm2_end_auth_session(chip);
+
+	ret = tpm2_prepare_space(chip, space, buf, bufsiz);
+	/* If the command is not implemented by the TPM, synthesize a
+	 * response with a TPM2_RC_COMMAND_CODE return for user-space.
+	 */
+	if (ret == -EOPNOTSUPP) {
+		header->length = cpu_to_be32(sizeof(*header));
+		header->tag = cpu_to_be16(TPM2_ST_NO_SESSIONS);
+		header->return_code = cpu_to_be32(TPM2_RC_COMMAND_CODE |
+						  TSS2_RESMGR_TPM_RC_LAYER);
+		ret = sizeof(*header);
+	}
+	if (ret)
+		goto out_rc;
+
+	len = tpm_transmit(chip, buf, bufsiz);
+	if (len < 0)
+		ret = len;
+
+	if (!ret)
+		ret = tpm2_commit_space(chip, space, buf, &len);
+	else
+		tpm2_flush_space(chip);
+
+out_rc:
+	return ret ? ret : len;
+}
+
+static void tpm_dev_async_work(struct work_struct *work)
+{
+	struct file_priv *priv =
+			container_of(work, struct file_priv, async_work);
+	ssize_t ret;
+
+	mutex_lock(&priv->buffer_mutex);
+	priv->command_enqueued = false;
+	ret = tpm_try_get_ops(priv->chip);
+	if (ret) {
+		priv->response_length = ret;
+		goto out;
+	}
+
+	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
+			       sizeof(priv->data_buffer));
+	tpm_put_ops(priv->chip);
+
+	/*
+	 * If ret is > 0 then tpm_dev_transmit returned the size of the
+	 * response. If ret is < 0 then tpm_dev_transmit failed and
+	 * returned an error code.
+	 */
+	if (ret != 0) {
+		priv->response_length = ret;
+		mod_timer(&priv->user_read_timer, jiffies + (120 * HZ));
+	}
+out:
+	mutex_unlock(&priv->buffer_mutex);
+	wake_up_interruptible(&priv->async_wait);
+}
+
+static void user_reader_timeout(struct timer_list *t)
+{
+	struct file_priv *priv = timer_container_of(priv, t, user_read_timer);
+
+	pr_warn("TPM user space timeout is deprecated (pid=%d)\n",
+		task_tgid_nr(current));
+
+	schedule_work(&priv->timeout_work);
+}
+
+static void tpm_timeout_work(struct work_struct *work)
+{
+	struct file_priv *priv = container_of(work, struct file_priv,
+					      timeout_work);
+
+	mutex_lock(&priv->buffer_mutex);
+	priv->response_read = true;
+	priv->response_length = 0;
+	memset(priv->data_buffer, 0, sizeof(priv->data_buffer));
+	mutex_unlock(&priv->buffer_mutex);
+	wake_up_interruptible(&priv->async_wait);
+}
+
+void tpm_common_open(struct file *file, struct tpm_chip *chip,
+		     struct file_priv *priv, struct tpm_space *space)
+{
+	priv->chip = chip;
+	priv->space = space;
+	priv->response_read = true;
+
+	mutex_init(&priv->buffer_mutex);
+	timer_setup(&priv->user_read_timer, user_reader_timeout, 0);
+	INIT_WORK(&priv->timeout_work, tpm_timeout_work);
+	INIT_WORK(&priv->async_work, tpm_dev_async_work);
+	init_waitqueue_head(&priv->async_wait);
+	file->private_data = priv;
+}
+
+ssize_t tpm_common_read(struct file *file, char __user *buf,
+			size_t size, loff_t *off)
+{
+	struct file_priv *priv = file->private_data;
+	ssize_t ret_size = 0;
+	int rc;
+
+	mutex_lock(&priv->buffer_mutex);
+
+	if (priv->response_length) {
+		priv->response_read = true;
+
+		ret_size = min_t(ssize_t, size, priv->response_length);
+		if (ret_size <= 0) {
+			priv->response_length = 0;
+			goto out;
+		}
+
+		rc = copy_to_user(buf, priv->data_buffer + *off, ret_size);
+		if (rc) {
+			memset(priv->data_buffer, 0, TPM_BUFSIZE);
+			priv->response_length = 0;
+			ret_size = -EFAULT;
+		} else {
+			memset(priv->data_buffer + *off, 0, ret_size);
+			priv->response_length -= ret_size;
+			*off += ret_size;
+		}
+	}
+
+out:
+	if (!priv->response_length) {
+		*off = 0;
+		timer_delete_sync(&priv->user_read_timer);
+		flush_work(&priv->timeout_work);
+	}
+	mutex_unlock(&priv->buffer_mutex);
+	return ret_size;
+}
+
+ssize_t tpm_common_write(struct file *file, const char __user *buf,
+			 size_t size, loff_t *off)
+{
+	struct file_priv *priv = file->private_data;
+	int ret = 0;
+
+	if (size > TPM_BUFSIZE)
+		return -E2BIG;
+
+	mutex_lock(&priv->buffer_mutex);
+
+	/* Cannot perform a write until the read has cleared either via
+	 * tpm_read or a user_read_timer timeout. This also prevents split
+	 * buffered writes from blocking here.
+	 */
+	if ((!priv->response_read && priv->response_length) ||
+	    priv->command_enqueued) {
+		ret = -EBUSY;
+		goto out;
+	}
+
+	if (copy_from_user(priv->data_buffer, buf, size)) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	if (size < 6 ||
+	    size < be32_to_cpu(*((__be32 *)(priv->data_buffer + 2)))) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	priv->response_length = 0;
+	priv->response_read = false;
+	*off = 0;
+
+	/*
+	 * If in nonblocking mode schedule an async job to send
+	 * the command return the size.
+	 * In case of error the err code will be returned in
+	 * the subsequent read call.
+	 */
+	if (file->f_flags & O_NONBLOCK) {
+		priv->command_enqueued = true;
+		queue_work(tpm_dev_wq, &priv->async_work);
+		mutex_unlock(&priv->buffer_mutex);
+		return size;
+	}
+
+	/* atomic tpm command send and result receive. We only hold the ops
+	 * lock during this period so that the tpm can be unregistered even if
+	 * the char dev is held open.
+	 */
+	if (tpm_try_get_ops(priv->chip)) {
+		ret = -EPIPE;
+		goto out;
+	}
+
+	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
+			       sizeof(priv->data_buffer));
+	tpm_put_ops(priv->chip);
+
+	if (ret > 0) {
+		priv->response_length = ret;
+		mod_timer(&priv->user_read_timer, jiffies + (120 * HZ));
+		ret = size;
+	}
+out:
+	mutex_unlock(&priv->buffer_mutex);
+	return ret;
+}
+
+__poll_t tpm_common_poll(struct file *file, poll_table *wait)
+{
+	struct file_priv *priv = file->private_data;
+	__poll_t mask = 0;
+
+	poll_wait(file, &priv->async_wait, wait);
+	mutex_lock(&priv->buffer_mutex);
+
+	/*
+	 * The response_length indicates if there is still response
+	 * (or part of it) to be consumed. Partial reads decrease it
+	 * by the number of bytes read, and write resets it the zero.
+	 */
+	if (priv->response_length)
+		mask = EPOLLIN | EPOLLRDNORM;
+	else
+		mask = EPOLLOUT | EPOLLWRNORM;
+
+	mutex_unlock(&priv->buffer_mutex);
+	return mask;
+}
+
+/*
+ * Called on file close
+ */
+void tpm_common_release(struct file *file, struct file_priv *priv)
+{
+	flush_work(&priv->async_work);
+	timer_delete_sync(&priv->user_read_timer);
+	flush_work(&priv->timeout_work);
+	file->private_data = NULL;
+	priv->response_length = 0;
+}
+
+int __init tpm_dev_common_init(void)
+{
+	tpm_dev_wq = alloc_workqueue("tpm_dev_wq", WQ_MEM_RECLAIM | WQ_PERCPU,
+				     0);
+
+	return !tpm_dev_wq ? -ENOMEM : 0;
+}
+
+void __exit tpm_dev_common_exit(void)
+{
+	if (tpm_dev_wq) {
+		destroy_workqueue(tpm_dev_wq);
+		tpm_dev_wq = NULL;
+	}
+}