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, 368 insertions, 0 deletions

diff --git a/io_uring/waitid.c b/io_uring/waitid.c
new file mode 100644
index 000000000..2b7560174
--- /dev/null
+++ b/io_uring/waitid.c
@@ -0,0 +1,368 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Support for async notification of waitid
+ */
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/fs.h>
+#include <linux/file.h>
+#include <linux/compat.h>
+#include <linux/io_uring.h>
+
+#include <uapi/linux/io_uring.h>
+
+#include "io_uring.h"
+#include "cancel.h"
+#include "waitid.h"
+#include "../kernel/exit.h"
+
+static void io_waitid_cb(struct io_tw_req tw_req, io_tw_token_t tw);
+
+#define IO_WAITID_CANCEL_FLAG	BIT(31)
+#define IO_WAITID_REF_MASK	GENMASK(30, 0)
+
+struct io_waitid {
+	struct file *file;
+	int which;
+	pid_t upid;
+	int options;
+	atomic_t refs;
+	struct wait_queue_head *head;
+	struct siginfo __user *infop;
+	struct waitid_info info;
+};
+
+static void io_waitid_free(struct io_kiocb *req)
+{
+	struct io_waitid_async *iwa = req->async_data;
+
+	put_pid(iwa->wo.wo_pid);
+	io_req_async_data_free(req);
+}
+
+static bool io_waitid_compat_copy_si(struct io_waitid *iw, int signo)
+{
+	struct compat_siginfo __user *infop;
+	bool ret;
+
+	infop = (struct compat_siginfo __user *) iw->infop;
+
+	if (!user_write_access_begin(infop, sizeof(*infop)))
+		return false;
+
+	unsafe_put_user(signo, &infop->si_signo, Efault);
+	unsafe_put_user(0, &infop->si_errno, Efault);
+	unsafe_put_user(iw->info.cause, &infop->si_code, Efault);
+	unsafe_put_user(iw->info.pid, &infop->si_pid, Efault);
+	unsafe_put_user(iw->info.uid, &infop->si_uid, Efault);
+	unsafe_put_user(iw->info.status, &infop->si_status, Efault);
+	ret = true;
+done:
+	user_write_access_end();
+	return ret;
+Efault:
+	ret = false;
+	goto done;
+}
+
+static bool io_waitid_copy_si(struct io_kiocb *req, int signo)
+{
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+	bool ret;
+
+	if (!iw->infop)
+		return true;
+
+	if (io_is_compat(req->ctx))
+		return io_waitid_compat_copy_si(iw, signo);
+
+	if (!user_write_access_begin(iw->infop, sizeof(*iw->infop)))
+		return false;
+
+	unsafe_put_user(signo, &iw->infop->si_signo, Efault);
+	unsafe_put_user(0, &iw->infop->si_errno, Efault);
+	unsafe_put_user(iw->info.cause, &iw->infop->si_code, Efault);
+	unsafe_put_user(iw->info.pid, &iw->infop->si_pid, Efault);
+	unsafe_put_user(iw->info.uid, &iw->infop->si_uid, Efault);
+	unsafe_put_user(iw->info.status, &iw->infop->si_status, Efault);
+	ret = true;
+done:
+	user_write_access_end();
+	return ret;
+Efault:
+	ret = false;
+	goto done;
+}
+
+static int io_waitid_finish(struct io_kiocb *req, int ret)
+{
+	int signo = 0;
+
+	if (ret > 0) {
+		signo = SIGCHLD;
+		ret = 0;
+	}
+
+	if (!io_waitid_copy_si(req, signo))
+		ret = -EFAULT;
+	io_waitid_free(req);
+	return ret;
+}
+
+static void io_waitid_remove_wq(struct io_kiocb *req)
+{
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+	struct wait_queue_head *head;
+
+	head = smp_load_acquire(&iw->head);
+	if (head) {
+		struct io_waitid_async *iwa = req->async_data;
+
+		smp_store_release(&iw->head, NULL);
+		spin_lock_irq(&head->lock);
+		list_del_init(&iwa->wo.child_wait.entry);
+		spin_unlock_irq(&head->lock);
+	}
+}
+
+static void io_waitid_complete(struct io_kiocb *req, int ret, bool copy_si)
+{
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+
+	/* anyone completing better be holding a reference */
+	WARN_ON_ONCE(!(atomic_read(&iw->refs) & IO_WAITID_REF_MASK));
+
+	lockdep_assert_held(&req->ctx->uring_lock);
+
+	hlist_del_init(&req->hash_node);
+	io_waitid_remove_wq(req);
+
+	if (copy_si)
+		ret = io_waitid_finish(req, ret);
+	else
+		io_waitid_free(req);
+	if (ret < 0)
+		req_set_fail(req);
+	io_req_set_res(req, ret, 0);
+}
+
+static bool __io_waitid_cancel(struct io_kiocb *req, bool copy_si)
+{
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+
+	lockdep_assert_held(&req->ctx->uring_lock);
+
+	/*
+	 * Mark us canceled regardless of ownership. This will prevent a
+	 * potential retry from a spurious wakeup.
+	 */
+	atomic_or(IO_WAITID_CANCEL_FLAG, &iw->refs);
+
+	/* claim ownership */
+	if (atomic_fetch_inc(&iw->refs) & IO_WAITID_REF_MASK)
+		return false;
+
+	io_waitid_complete(req, -ECANCELED, copy_si);
+	io_req_queue_tw_complete(req, -ECANCELED);
+	return true;
+}
+
+static bool io_waitid_cancel_cb(struct io_kiocb *req)
+{
+	return __io_waitid_cancel(req, true);
+}
+
+static bool io_waitid_cancel_nocopy_cb(struct io_kiocb *req)
+{
+	return __io_waitid_cancel(req, false);
+}
+
+int io_waitid_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd,
+		     unsigned int issue_flags)
+{
+	return io_cancel_remove(ctx, cd, issue_flags, &ctx->waitid_list, io_waitid_cancel_cb);
+}
+
+bool io_waitid_remove_all(struct io_ring_ctx *ctx, struct io_uring_task *tctx,
+			  bool cancel_all)
+{
+	return io_cancel_remove_all(ctx, tctx, &ctx->waitid_list, cancel_all,
+				       tctx ? io_waitid_cancel_cb : io_waitid_cancel_nocopy_cb);
+}
+
+static inline bool io_waitid_drop_issue_ref(struct io_kiocb *req)
+{
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+
+	if (!atomic_sub_return(1, &iw->refs))
+		return false;
+
+	io_waitid_remove_wq(req);
+
+	/*
+	 * Wakeup triggered, racing with us. It was prevented from
+	 * completing because of that, queue up the tw to do that.
+	 */
+	req->io_task_work.func = io_waitid_cb;
+	io_req_task_work_add(req);
+	return true;
+}
+
+static void io_waitid_cb(struct io_tw_req tw_req, io_tw_token_t tw)
+{
+	struct io_kiocb *req = tw_req.req;
+	struct io_waitid_async *iwa = req->async_data;
+	struct io_ring_ctx *ctx = req->ctx;
+	int ret;
+
+	io_tw_lock(ctx, tw);
+	if (unlikely(tw.cancel)) {
+		io_waitid_complete(req, -ECANCELED, false);
+		io_req_task_complete(tw_req, tw);
+		return;
+	}
+
+	ret = __do_wait(&iwa->wo);
+
+	/*
+	 * If we get -ERESTARTSYS here, we need to re-arm and check again
+	 * to ensure we get another callback. If the retry works, then we can
+	 * just remove ourselves from the waitqueue again and finish the
+	 * request.
+	 */
+	if (unlikely(ret == -ERESTARTSYS)) {
+		struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+
+		/* Don't retry if cancel found it meanwhile */
+		ret = -ECANCELED;
+		if (!(atomic_read(&iw->refs) & IO_WAITID_CANCEL_FLAG)) {
+			iw->head = &current->signal->wait_chldexit;
+			add_wait_queue(iw->head, &iwa->wo.child_wait);
+			ret = __do_wait(&iwa->wo);
+			if (ret == -ERESTARTSYS) {
+				/* retry armed, drop our ref */
+				io_waitid_drop_issue_ref(req);
+				return;
+			}
+			/* fall through to complete, will kill waitqueue */
+		}
+	}
+
+	io_waitid_complete(req, ret, true);
+	io_req_task_complete(tw_req, tw);
+}
+
+static int io_waitid_wait(struct wait_queue_entry *wait, unsigned mode,
+			  int sync, void *key)
+{
+	struct wait_opts *wo = container_of(wait, struct wait_opts, child_wait);
+	struct io_waitid_async *iwa = container_of(wo, struct io_waitid_async, wo);
+	struct io_kiocb *req = iwa->req;
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+	struct task_struct *p = key;
+
+	if (!pid_child_should_wake(wo, p))
+		return 0;
+
+	list_del_init(&wait->entry);
+	smp_store_release(&iw->head, NULL);
+
+	/* cancel is in progress */
+	if (atomic_fetch_inc(&iw->refs) & IO_WAITID_REF_MASK)
+		return 1;
+
+	req->io_task_work.func = io_waitid_cb;
+	__io_req_task_work_add(req, IOU_F_TWQ_IN_WAKE);
+	return 1;
+}
+
+int io_waitid_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+	struct io_waitid_async *iwa;
+
+	if (sqe->addr || sqe->buf_index || sqe->addr3 || sqe->waitid_flags)
+		return -EINVAL;
+
+	iwa = io_uring_alloc_async_data(NULL, req);
+	if (unlikely(!iwa))
+		return -ENOMEM;
+	iwa->req = req;
+
+	iw->which = READ_ONCE(sqe->len);
+	iw->upid = READ_ONCE(sqe->fd);
+	iw->options = READ_ONCE(sqe->file_index);
+	iw->head = NULL;
+	iw->infop = u64_to_user_ptr(READ_ONCE(sqe->addr2));
+	memset(&iw->info, 0, sizeof(iw->info));
+	return 0;
+}
+
+int io_waitid(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_waitid *iw = io_kiocb_to_cmd(req, struct io_waitid);
+	struct io_waitid_async *iwa = req->async_data;
+	struct io_ring_ctx *ctx = req->ctx;
+	int ret;
+
+	ret = kernel_waitid_prepare(&iwa->wo, iw->which, iw->upid, &iw->info,
+					iw->options, NULL);
+	if (ret)
+		goto done;
+
+	/*
+	 * Mark the request as busy upfront, in case we're racing with the
+	 * wakeup. If we are, then we'll notice when we drop this initial
+	 * reference again after arming.
+	 */
+	atomic_set(&iw->refs, 1);
+
+	/*
+	 * Cancel must hold the ctx lock, so there's no risk of cancelation
+	 * finding us until a) we remain on the list, and b) the lock is
+	 * dropped. We only need to worry about racing with the wakeup
+	 * callback.
+	 */
+	io_ring_submit_lock(ctx, issue_flags);
+
+	/*
+	 * iw->head is valid under the ring lock, and as long as the request
+	 * is on the waitid_list where cancelations may find it.
+	 */
+	iw->head = &current->signal->wait_chldexit;
+	hlist_add_head(&req->hash_node, &ctx->waitid_list);
+
+	init_waitqueue_func_entry(&iwa->wo.child_wait, io_waitid_wait);
+	iwa->wo.child_wait.private = req->tctx->task;
+	add_wait_queue(iw->head, &iwa->wo.child_wait);
+
+	ret = __do_wait(&iwa->wo);
+	if (ret == -ERESTARTSYS) {
+		/*
+		 * Nobody else grabbed a reference, it'll complete when we get
+		 * a waitqueue callback, or if someone cancels it.
+		 */
+		if (!io_waitid_drop_issue_ref(req)) {
+			io_ring_submit_unlock(ctx, issue_flags);
+			return IOU_ISSUE_SKIP_COMPLETE;
+		}
+
+		/*
+		 * Wakeup triggered, racing with us. It was prevented from
+		 * completing because of that, queue up the tw to do that.
+		 */
+		io_ring_submit_unlock(ctx, issue_flags);
+		return IOU_ISSUE_SKIP_COMPLETE;
+	}
+
+	hlist_del_init(&req->hash_node);
+	io_waitid_remove_wq(req);
+	ret = io_waitid_finish(req, ret);
+
+	io_ring_submit_unlock(ctx, issue_flags);
+done:
+	if (ret < 0)
+		req_set_fail(req);
+	io_req_set_res(req, ret, 0);
+	return IOU_COMPLETE;
+}