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, 373 insertions, 0 deletions

diff --git a/fs/orangefs/waitqueue.c b/fs/orangefs/waitqueue.c
new file mode 100644
index 000000000..beafc33d5
--- /dev/null
+++ b/fs/orangefs/waitqueue.c
@@ -0,0 +1,373 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * (C) 2001 Clemson University and The University of Chicago
+ * (C) 2011 Omnibond Systems
+ *
+ * Changes by Acxiom Corporation to implement generic service_operation()
+ * function, Copyright Acxiom Corporation, 2005.
+ *
+ * See COPYING in top-level directory.
+ */
+
+/*
+ *  In-kernel waitqueue operations.
+ */
+
+#include "protocol.h"
+#include "orangefs-kernel.h"
+#include "orangefs-bufmap.h"
+
+static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
+		long timeout,
+		int flags)
+			__acquires(op->lock);
+static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
+	__releases(op->lock);
+
+/*
+ * What we do in this function is to walk the list of operations that are
+ * present in the request queue and mark them as purged.
+ * NOTE: This is called from the device close after client-core has
+ * guaranteed that no new operations could appear on the list since the
+ * client-core is anyway going to exit.
+ */
+void purge_waiting_ops(void)
+{
+	struct orangefs_kernel_op_s *op, *tmp;
+
+	spin_lock(&orangefs_request_list_lock);
+	list_for_each_entry_safe(op, tmp, &orangefs_request_list, list) {
+		gossip_debug(GOSSIP_WAIT_DEBUG,
+			     "pvfs2-client-core: purging op tag %llu %s\n",
+			     llu(op->tag),
+			     get_opname_string(op));
+		set_op_state_purged(op);
+		gossip_debug(GOSSIP_DEV_DEBUG,
+			     "%s: op:%s: op_state:%d: process:%s:\n",
+			     __func__,
+			     get_opname_string(op),
+			     op->op_state,
+			     current->comm);
+	}
+	spin_unlock(&orangefs_request_list_lock);
+}
+
+/*
+ * submits a ORANGEFS operation and waits for it to complete
+ *
+ * Note op->downcall.status will contain the status of the operation (in
+ * errno format), whether provided by pvfs2-client or a result of failure to
+ * service the operation.  If the caller wishes to distinguish, then
+ * op->state can be checked to see if it was serviced or not.
+ *
+ * Returns contents of op->downcall.status for convenience
+ */
+int service_operation(struct orangefs_kernel_op_s *op,
+		      const char *op_name,
+		      int flags)
+{
+	long timeout = MAX_SCHEDULE_TIMEOUT;
+	int ret = 0;
+
+	DEFINE_WAIT(wait_entry);
+
+	op->upcall.tgid = current->tgid;
+	op->upcall.pid = current->pid;
+
+retry_servicing:
+	op->downcall.status = 0;
+	gossip_debug(GOSSIP_WAIT_DEBUG,
+		     "%s: %s op:%p: process:%s: pid:%d:\n",
+		     __func__,
+		     op_name,
+		     op,
+		     current->comm,
+		     current->pid);
+
+	/*
+	 * If ORANGEFS_OP_NO_MUTEX was set in flags, we need to avoid
+	 * acquiring the request_mutex because we're servicing a
+	 * high priority remount operation and the request_mutex is
+	 * already taken.
+	 */
+	if (!(flags & ORANGEFS_OP_NO_MUTEX)) {
+		if (flags & ORANGEFS_OP_INTERRUPTIBLE)
+			ret = mutex_lock_interruptible(&orangefs_request_mutex);
+		else
+			ret = mutex_lock_killable(&orangefs_request_mutex);
+		/*
+		 * check to see if we were interrupted while waiting for
+		 * mutex
+		 */
+		if (ret < 0) {
+			op->downcall.status = ret;
+			gossip_debug(GOSSIP_WAIT_DEBUG,
+				     "%s: service_operation interrupted.\n",
+				     __func__);
+			return ret;
+		}
+	}
+
+	/* queue up the operation */
+	spin_lock(&orangefs_request_list_lock);
+	spin_lock(&op->lock);
+	set_op_state_waiting(op);
+	gossip_debug(GOSSIP_DEV_DEBUG,
+		     "%s: op:%s: op_state:%d: process:%s:\n",
+		     __func__,
+		     get_opname_string(op),
+		     op->op_state,
+		     current->comm);
+	/* add high priority remount op to the front of the line. */
+	if (flags & ORANGEFS_OP_PRIORITY)
+		list_add(&op->list, &orangefs_request_list);
+	else
+		list_add_tail(&op->list, &orangefs_request_list);
+	spin_unlock(&op->lock);
+	wake_up_interruptible(&orangefs_request_list_waitq);
+	if (!__is_daemon_in_service()) {
+		gossip_debug(GOSSIP_WAIT_DEBUG,
+			     "%s:client core is NOT in service.\n",
+			     __func__);
+		/*
+		 * Don't wait for the userspace component to return if
+		 * the filesystem is being umounted anyway.
+		 */
+		if (op->upcall.type == ORANGEFS_VFS_OP_FS_UMOUNT)
+			timeout = 0;
+		else
+			timeout = op_timeout_secs * HZ;
+	}
+	spin_unlock(&orangefs_request_list_lock);
+
+	if (!(flags & ORANGEFS_OP_NO_MUTEX))
+		mutex_unlock(&orangefs_request_mutex);
+
+	ret = wait_for_matching_downcall(op, timeout, flags);
+	gossip_debug(GOSSIP_WAIT_DEBUG,
+		     "%s: wait_for_matching_downcall returned %d for %p\n",
+		     __func__,
+		     ret,
+		     op);
+
+	/* got matching downcall; make sure status is in errno format */
+	if (!ret) {
+		spin_unlock(&op->lock);
+		op->downcall.status =
+		    orangefs_normalize_to_errno(op->downcall.status);
+		ret = op->downcall.status;
+		goto out;
+	}
+
+	/* failed to get matching downcall */
+	if (ret == -ETIMEDOUT) {
+		gossip_err("%s: %s -- wait timed out; aborting attempt.\n",
+			   __func__,
+			   op_name);
+	}
+
+	/*
+	 * remove a waiting op from the request list or
+	 * remove an in-progress op from the in-progress list.
+	 */
+	orangefs_clean_up_interrupted_operation(op);
+
+	op->downcall.status = ret;
+	/* retry if operation has not been serviced and if requested */
+	if (ret == -EAGAIN) {
+		op->attempts++;
+		timeout = op_timeout_secs * HZ;
+		gossip_debug(GOSSIP_WAIT_DEBUG,
+			     "orangefs: tag %llu (%s)"
+			     " -- operation to be retried (%d attempt)\n",
+			     llu(op->tag),
+			     op_name,
+			     op->attempts);
+
+		/*
+		 * io ops (ops that use the shared memory buffer) have
+		 * to be returned to their caller for a retry. Other ops
+		 * can just be recycled here.
+		 */
+		if (!op->uses_shared_memory)
+			goto retry_servicing;
+	}
+
+out:
+	gossip_debug(GOSSIP_WAIT_DEBUG,
+		     "%s: %s returning: %d for %p.\n",
+		     __func__,
+		     op_name,
+		     ret,
+		     op);
+	return ret;
+}
+
+/* This can get called on an I/O op if it had a bad service_operation. */
+bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
+{
+	u64 tag = op->tag;
+	if (!op_state_in_progress(op))
+		return false;
+
+	op->slot_to_free = op->upcall.req.io.buf_index;
+	memset(&op->upcall, 0, sizeof(op->upcall));
+	memset(&op->downcall, 0, sizeof(op->downcall));
+	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
+	op->upcall.req.cancel.op_tag = tag;
+	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
+	op->downcall.status = -1;
+	orangefs_new_tag(op);
+
+	spin_lock(&orangefs_request_list_lock);
+	/* orangefs_request_list_lock is enough of a barrier here */
+	if (!__is_daemon_in_service()) {
+		spin_unlock(&orangefs_request_list_lock);
+		return false;
+	}
+	spin_lock(&op->lock);
+	set_op_state_waiting(op);
+	gossip_debug(GOSSIP_DEV_DEBUG,
+		     "%s: op:%s: op_state:%d: process:%s:\n",
+		     __func__,
+		     get_opname_string(op),
+		     op->op_state,
+		     current->comm);
+	list_add(&op->list, &orangefs_request_list);
+	spin_unlock(&op->lock);
+	spin_unlock(&orangefs_request_list_lock);
+
+	gossip_debug(GOSSIP_WAIT_DEBUG,
+		     "Attempting ORANGEFS operation cancellation of tag %llu\n",
+		     llu(tag));
+	return true;
+}
+
+/*
+ * Change an op to the "given up" state and remove it from its list.
+ */
+static void
+	orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
+		__releases(op->lock)
+{
+	/*
+	 * handle interrupted cases depending on what state we were in when
+	 * the interruption is detected.
+	 *
+	 * Called with op->lock held.
+	 */
+
+	/*
+	 * List manipulation code elsewhere will ignore ops that
+	 * have been given up upon.
+	 */
+	op->op_state |= OP_VFS_STATE_GIVEN_UP;
+
+	if (list_empty(&op->list)) {
+		/* caught copying to/from daemon */
+		BUG_ON(op_state_serviced(op));
+		spin_unlock(&op->lock);
+		wait_for_completion(&op->waitq);
+	} else if (op_state_waiting(op)) {
+		/*
+		 * upcall hasn't been read; remove op from upcall request
+		 * list.
+		 */
+		spin_unlock(&op->lock);
+		spin_lock(&orangefs_request_list_lock);
+		list_del_init(&op->list);
+		spin_unlock(&orangefs_request_list_lock);
+		gossip_debug(GOSSIP_WAIT_DEBUG,
+			     "Interrupted: Removed op %p from request_list\n",
+			     op);
+	} else if (op_state_in_progress(op)) {
+		/* op must be removed from the in progress htable */
+		spin_unlock(&op->lock);
+		spin_lock(&orangefs_htable_ops_in_progress_lock);
+		list_del_init(&op->list);
+		spin_unlock(&orangefs_htable_ops_in_progress_lock);
+		gossip_debug(GOSSIP_WAIT_DEBUG,
+			     "Interrupted: Removed op %p"
+			     " from htable_ops_in_progress\n",
+			     op);
+	} else {
+		spin_unlock(&op->lock);
+		gossip_err("interrupted operation is in a weird state 0x%x\n",
+			   op->op_state);
+	}
+	reinit_completion(&op->waitq);
+}
+
+/*
+ * Sleeps on waitqueue waiting for matching downcall.
+ * If client-core finishes servicing, then we are good to go.
+ * else if client-core exits, we get woken up here, and retry with a timeout
+ *
+ * When this call returns to the caller, the specified op will no
+ * longer be in either the in_progress hash table or on the request list.
+ *
+ * Returns 0 on success and -errno on failure
+ * Errors are:
+ * EAGAIN in case we want the caller to requeue and try again..
+ * EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
+ * operation since client-core seems to be exiting too often
+ * or if we were interrupted.
+ *
+ * Returns with op->lock taken.
+ */
+static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
+		long timeout,
+		int flags)
+			__acquires(op->lock)
+{
+	long n;
+	int writeback = flags & ORANGEFS_OP_WRITEBACK,
+	    interruptible = flags & ORANGEFS_OP_INTERRUPTIBLE;
+
+	/*
+	 * There's a "schedule_timeout" inside of these wait
+	 * primitives, during which the op is out of the hands of the
+	 * user process that needs something done and is being
+	 * manipulated by the client-core process.
+	 */
+	if (writeback)
+		n = wait_for_completion_io_timeout(&op->waitq, timeout);
+	else if (!writeback && interruptible)
+		n = wait_for_completion_interruptible_timeout(&op->waitq,
+								      timeout);
+	else /* !writeback && !interruptible but compiler complains */
+		n = wait_for_completion_killable_timeout(&op->waitq, timeout);
+
+	spin_lock(&op->lock);
+
+	if (op_state_serviced(op))
+		return 0;
+
+	if (unlikely(n < 0)) {
+		gossip_debug(GOSSIP_WAIT_DEBUG,
+			     "%s: operation interrupted, tag %llu, %p\n",
+			     __func__,
+			     llu(op->tag),
+			     op);
+		return -EINTR;
+	}
+	if (op_state_purged(op)) {
+		gossip_debug(GOSSIP_WAIT_DEBUG,
+			     "%s: operation purged, tag %llu, %p, %d\n",
+			     __func__,
+			     llu(op->tag),
+			     op,
+			     op->attempts);
+		return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
+			 -EAGAIN :
+			 -EIO;
+	}
+	/* must have timed out, then... */
+	gossip_debug(GOSSIP_WAIT_DEBUG,
+		     "%s: operation timed out, tag %llu, %p, %d)\n",
+		     __func__,
+		     llu(op->tag),
+		     op,
+		     op->attempts);
+	return -ETIMEDOUT;
+}