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, 170 insertions, 0 deletions

diff --git a/drivers/nvme/target/trace.h b/drivers/nvme/target/trace.h
new file mode 100644
index 000000000..7f7ebf955
--- /dev/null
+++ b/drivers/nvme/target/trace.h
@@ -0,0 +1,170 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * NVM Express target device driver tracepoints
+ * Copyright (c) 2018 Johannes Thumshirn, SUSE Linux GmbH
+ *
+ * This is entirely based on drivers/nvme/host/trace.h
+ */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM nvmet
+
+#if !defined(_TRACE_NVMET_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_NVMET_H
+
+#include <linux/nvme.h>
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+#include "nvmet.h"
+
+const char *nvmet_trace_parse_admin_cmd(struct trace_seq *p, u8 opcode,
+		u8 *cdw10);
+const char *nvmet_trace_parse_nvm_cmd(struct trace_seq *p, u8 opcode,
+		u8 *cdw10);
+const char *nvmet_trace_parse_fabrics_cmd(struct trace_seq *p, u8 fctype,
+		u8 *spc);
+
+#define parse_nvme_cmd(qid, opcode, fctype, cdw10)			\
+	((opcode) == nvme_fabrics_command ?				\
+	 nvmet_trace_parse_fabrics_cmd(p, fctype, cdw10) :		\
+	(qid ?								\
+	 nvmet_trace_parse_nvm_cmd(p, opcode, cdw10) :			\
+	 nvmet_trace_parse_admin_cmd(p, opcode, cdw10)))
+
+const char *nvmet_trace_ctrl_id(struct trace_seq *p, u16 ctrl_id);
+#define __print_ctrl_id(ctrl_id)			\
+	nvmet_trace_ctrl_id(p, ctrl_id)
+
+const char *nvmet_trace_disk_name(struct trace_seq *p, char *name);
+#define __print_disk_name(name)				\
+	nvmet_trace_disk_name(p, name)
+
+#ifndef TRACE_HEADER_MULTI_READ
+static inline u16 nvmet_req_to_ctrl_id(struct nvmet_req *req)
+{
+	/*
+	 * The queue and controller pointers are not valid until an association
+	 * has been established.
+	 */
+	if (!req->sq || !req->sq->ctrl)
+		return 0;
+	return req->sq->ctrl->cntlid;
+}
+
+static inline void __assign_req_name(char *name, struct nvmet_req *req)
+{
+	if (!req->ns) {
+		memset(name, 0, DISK_NAME_LEN);
+		return;
+	}
+
+	strscpy_pad(name, req->ns->device_path, DISK_NAME_LEN);
+}
+#endif
+
+TRACE_EVENT(nvmet_req_init,
+	TP_PROTO(struct nvmet_req *req, struct nvme_command *cmd),
+	TP_ARGS(req, cmd),
+	TP_STRUCT__entry(
+		__field(struct nvme_command *, cmd)
+		__field(u16, ctrl_id)
+		__array(char, disk, DISK_NAME_LEN)
+		__field(int, qid)
+		__field(u16, cid)
+		__field(u8, opcode)
+		__field(u8, fctype)
+		__field(u8, flags)
+		__field(u32, nsid)
+		__field(u64, metadata)
+		__array(u8, cdw10, 24)
+	),
+	TP_fast_assign(
+		__entry->cmd = cmd;
+		__entry->ctrl_id = nvmet_req_to_ctrl_id(req);
+		__assign_req_name(__entry->disk, req);
+		__entry->qid = req->sq->qid;
+		__entry->cid = cmd->common.command_id;
+		__entry->opcode = cmd->common.opcode;
+		__entry->fctype = cmd->fabrics.fctype;
+		__entry->flags = cmd->common.flags;
+		__entry->nsid = le32_to_cpu(cmd->common.nsid);
+		__entry->metadata = le64_to_cpu(cmd->common.metadata);
+		memcpy(__entry->cdw10, &cmd->common.cdws,
+			sizeof(__entry->cdw10));
+	),
+	TP_printk("nvmet%s: %sqid=%d, cmdid=%u, nsid=%u, flags=%#x, "
+		  "meta=%#llx, cmd=(%s, %s)",
+		__print_ctrl_id(__entry->ctrl_id),
+		__print_disk_name(__entry->disk),
+		__entry->qid, __entry->cid, __entry->nsid,
+		__entry->flags, __entry->metadata,
+		show_opcode_name(__entry->qid, __entry->opcode,
+				__entry->fctype),
+		parse_nvme_cmd(__entry->qid, __entry->opcode,
+				__entry->fctype, __entry->cdw10))
+);
+
+TRACE_EVENT(nvmet_req_complete,
+	TP_PROTO(struct nvmet_req *req),
+	TP_ARGS(req),
+	TP_STRUCT__entry(
+		__field(u16, ctrl_id)
+		__array(char, disk, DISK_NAME_LEN)
+		__field(int, qid)
+		__field(int, cid)
+		__field(u64, result)
+		__field(u16, status)
+	),
+	TP_fast_assign(
+		__entry->ctrl_id = nvmet_req_to_ctrl_id(req);
+		__entry->qid = req->cq->qid;
+		__entry->cid = req->cqe->command_id;
+		__entry->result = le64_to_cpu(req->cqe->result.u64);
+		__entry->status = le16_to_cpu(req->cqe->status) >> 1;
+		__assign_req_name(__entry->disk, req);
+	),
+	TP_printk("nvmet%s: %sqid=%d, cmdid=%u, res=%#llx, status=%#x",
+		__print_ctrl_id(__entry->ctrl_id),
+		__print_disk_name(__entry->disk),
+		__entry->qid, __entry->cid, __entry->result, __entry->status)
+
+);
+
+#define aer_name(aer) { aer, #aer }
+
+TRACE_EVENT(nvmet_async_event,
+	TP_PROTO(struct nvmet_ctrl *ctrl, __le32 result),
+	TP_ARGS(ctrl, result),
+	TP_STRUCT__entry(
+		__field(int, ctrl_id)
+		__field(u32, result)
+	),
+	TP_fast_assign(
+		__entry->ctrl_id = ctrl->cntlid;
+		__entry->result = (le32_to_cpu(result) & 0xff00) >> 8;
+	),
+	TP_printk("nvmet%d: NVME_AEN=%#08x [%s]",
+		__entry->ctrl_id, __entry->result,
+		__print_symbolic(__entry->result,
+		aer_name(NVME_AER_NOTICE_NS_CHANGED),
+		aer_name(NVME_AER_NOTICE_ANA),
+		aer_name(NVME_AER_NOTICE_FW_ACT_STARTING),
+		aer_name(NVME_AER_NOTICE_DISC_CHANGED),
+		aer_name(NVME_AER_ERROR),
+		aer_name(NVME_AER_SMART),
+		aer_name(NVME_AER_CSS),
+		aer_name(NVME_AER_VS))
+	)
+);
+#undef aer_name
+
+#endif /* _TRACE_NVMET_H */
+
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_FILE trace
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>