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, 275 insertions, 0 deletions

diff --git a/drivers/md/dm-vdo/dump.c b/drivers/md/dm-vdo/dump.c
new file mode 100644
index 000000000..00e575d7d
--- /dev/null
+++ b/drivers/md/dm-vdo/dump.c
@@ -0,0 +1,275 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright 2023 Red Hat
+ */
+
+#include "dump.h"
+
+#include <linux/module.h>
+
+#include "memory-alloc.h"
+#include "string-utils.h"
+
+#include "constants.h"
+#include "data-vio.h"
+#include "dedupe.h"
+#include "funnel-workqueue.h"
+#include "io-submitter.h"
+#include "logger.h"
+#include "types.h"
+#include "vdo.h"
+
+enum dump_options {
+	/* Work queues */
+	SHOW_QUEUES,
+	/* Memory pools */
+	SHOW_VIO_POOL,
+	/* Others */
+	SHOW_VDO_STATUS,
+	/* This one means an option overrides the "default" choices, instead of altering them. */
+	SKIP_DEFAULT
+};
+
+enum dump_option_flags {
+	/* Work queues */
+	FLAG_SHOW_QUEUES = (1 << SHOW_QUEUES),
+	/* Memory pools */
+	FLAG_SHOW_VIO_POOL = (1 << SHOW_VIO_POOL),
+	/* Others */
+	FLAG_SHOW_VDO_STATUS = (1 << SHOW_VDO_STATUS),
+	/* Special */
+	FLAG_SKIP_DEFAULT = (1 << SKIP_DEFAULT)
+};
+
+#define FLAGS_ALL_POOLS (FLAG_SHOW_VIO_POOL)
+#define DEFAULT_DUMP_FLAGS (FLAG_SHOW_QUEUES | FLAG_SHOW_VDO_STATUS)
+/* Another static buffer... log10(256) = 2.408+, round up: */
+#define DIGITS_PER_U64 (1 + sizeof(u64) * 2409 / 1000)
+
+static inline bool is_arg_string(const char *arg, const char *this_option)
+{
+	/* convention seems to be case-independent options */
+	return strncasecmp(arg, this_option, strlen(this_option)) == 0;
+}
+
+static void do_dump(struct vdo *vdo, unsigned int dump_options_requested,
+		    const char *why)
+{
+	u32 active, maximum;
+	s64 outstanding;
+
+	vdo_log_info("%s dump triggered via %s", VDO_LOGGING_MODULE_NAME, why);
+	active = get_data_vio_pool_active_requests(vdo->data_vio_pool);
+	maximum = get_data_vio_pool_maximum_requests(vdo->data_vio_pool);
+	outstanding = (atomic64_read(&vdo->stats.bios_submitted) -
+		       atomic64_read(&vdo->stats.bios_completed));
+	vdo_log_info("%u device requests outstanding (max %u), %lld bio requests outstanding, device '%s'",
+		     active, maximum, outstanding,
+		     vdo_get_device_name(vdo->device_config->owning_target));
+	if (((dump_options_requested & FLAG_SHOW_QUEUES) != 0) && (vdo->threads != NULL)) {
+		thread_id_t id;
+
+		for (id = 0; id < vdo->thread_config.thread_count; id++)
+			vdo_dump_work_queue(vdo->threads[id].queue);
+	}
+
+	vdo_dump_hash_zones(vdo->hash_zones);
+	dump_data_vio_pool(vdo->data_vio_pool,
+			   (dump_options_requested & FLAG_SHOW_VIO_POOL) != 0);
+	if ((dump_options_requested & FLAG_SHOW_VDO_STATUS) != 0)
+		vdo_dump_status(vdo);
+
+	vdo_report_memory_usage();
+	vdo_log_info("end of %s dump", VDO_LOGGING_MODULE_NAME);
+}
+
+static int parse_dump_options(unsigned int argc, char *const *argv,
+			      unsigned int *dump_options_requested_ptr)
+{
+	unsigned int dump_options_requested = 0;
+
+	static const struct {
+		const char *name;
+		unsigned int flags;
+	} option_names[] = {
+		{ "viopool", FLAG_SKIP_DEFAULT | FLAG_SHOW_VIO_POOL },
+		{ "vdo", FLAG_SKIP_DEFAULT | FLAG_SHOW_VDO_STATUS },
+		{ "pools", FLAG_SKIP_DEFAULT | FLAGS_ALL_POOLS },
+		{ "queues", FLAG_SKIP_DEFAULT | FLAG_SHOW_QUEUES },
+		{ "threads", FLAG_SKIP_DEFAULT | FLAG_SHOW_QUEUES },
+		{ "default", FLAG_SKIP_DEFAULT | DEFAULT_DUMP_FLAGS },
+		{ "all", ~0 },
+	};
+
+	bool options_okay = true;
+	unsigned int i;
+
+	for (i = 1; i < argc; i++) {
+		unsigned int j;
+
+		for (j = 0; j < ARRAY_SIZE(option_names); j++) {
+			if (is_arg_string(argv[i], option_names[j].name)) {
+				dump_options_requested |= option_names[j].flags;
+				break;
+			}
+		}
+		if (j == ARRAY_SIZE(option_names)) {
+			vdo_log_warning("dump option name '%s' unknown", argv[i]);
+			options_okay = false;
+		}
+	}
+	if (!options_okay)
+		return -EINVAL;
+	if ((dump_options_requested & FLAG_SKIP_DEFAULT) == 0)
+		dump_options_requested |= DEFAULT_DUMP_FLAGS;
+	*dump_options_requested_ptr = dump_options_requested;
+	return 0;
+}
+
+/* Dump as specified by zero or more string arguments. */
+int vdo_dump(struct vdo *vdo, unsigned int argc, char *const *argv, const char *why)
+{
+	unsigned int dump_options_requested = 0;
+	int result = parse_dump_options(argc, argv, &dump_options_requested);
+
+	if (result != 0)
+		return result;
+
+	do_dump(vdo, dump_options_requested, why);
+	return 0;
+}
+
+/* Dump everything we know how to dump */
+void vdo_dump_all(struct vdo *vdo, const char *why)
+{
+	do_dump(vdo, ~0, why);
+}
+
+/*
+ * Dump out the data_vio waiters on a waitq.
+ * wait_on should be the label to print for queue (e.g. logical or physical)
+ */
+static void dump_vio_waiters(struct vdo_wait_queue *waitq, char *wait_on)
+{
+	struct vdo_waiter *waiter, *first = vdo_waitq_get_first_waiter(waitq);
+	struct data_vio *data_vio;
+
+	if (first == NULL)
+		return;
+
+	data_vio = vdo_waiter_as_data_vio(first);
+
+	vdo_log_info("      %s is locked. Waited on by: vio %px pbn %llu lbn %llu d-pbn %llu lastOp %s",
+		     wait_on, data_vio, data_vio->allocation.pbn, data_vio->logical.lbn,
+		     data_vio->duplicate.pbn, get_data_vio_operation_name(data_vio));
+
+	for (waiter = first->next_waiter; waiter != first; waiter = waiter->next_waiter) {
+		data_vio = vdo_waiter_as_data_vio(waiter);
+		vdo_log_info("     ... and : vio %px pbn %llu lbn %llu d-pbn %llu lastOp %s",
+			     data_vio, data_vio->allocation.pbn, data_vio->logical.lbn,
+			     data_vio->duplicate.pbn,
+			     get_data_vio_operation_name(data_vio));
+	}
+}
+
+/*
+ * Encode various attributes of a data_vio as a string of one-character flags. This encoding is for
+ * logging brevity:
+ *
+ * R => vio completion result not VDO_SUCCESS
+ * W => vio is on a waitq
+ * D => vio is a duplicate
+ * p => vio is a partial block operation
+ * z => vio is a zero block
+ * d => vio is a discard
+ *
+ * The common case of no flags set will result in an empty, null-terminated buffer. If any flags
+ * are encoded, the first character in the string will be a space character.
+ */
+static void encode_vio_dump_flags(struct data_vio *data_vio, char buffer[8])
+{
+	char *p_flag = buffer;
+	*p_flag++ = ' ';
+	if (data_vio->vio.completion.result != VDO_SUCCESS)
+		*p_flag++ = 'R';
+	if (data_vio->waiter.next_waiter != NULL)
+		*p_flag++ = 'W';
+	if (data_vio->is_duplicate)
+		*p_flag++ = 'D';
+	if (data_vio->is_partial)
+		*p_flag++ = 'p';
+	if (data_vio->is_zero)
+		*p_flag++ = 'z';
+	if (data_vio->remaining_discard > 0)
+		*p_flag++ = 'd';
+	if (p_flag == &buffer[1]) {
+		/* No flags, so remove the blank space. */
+		p_flag = buffer;
+	}
+	*p_flag = '\0';
+}
+
+/* Implements buffer_dump_function. */
+void dump_data_vio(void *data)
+{
+	struct data_vio *data_vio = data;
+
+	/*
+	 * This just needs to be big enough to hold a queue (thread) name and a function name (plus
+	 * a separator character and NUL). The latter is limited only by taste.
+	 *
+	 * In making this static, we're assuming only one "dump" will run at a time. If more than
+	 * one does run, the log output will be garbled anyway.
+	 */
+	static char vio_completion_dump_buffer[100 + MAX_VDO_WORK_QUEUE_NAME_LEN];
+	static char vio_block_number_dump_buffer[sizeof("P L D") + 3 * DIGITS_PER_U64];
+	static char vio_flush_generation_buffer[sizeof(" FG") + DIGITS_PER_U64];
+	static char flags_dump_buffer[8];
+
+	/*
+	 * We're likely to be logging a couple thousand of these lines, and in some circumstances
+	 * syslogd may have trouble keeping up, so keep it BRIEF rather than user-friendly.
+	 */
+	vdo_dump_completion_to_buffer(&data_vio->vio.completion,
+				      vio_completion_dump_buffer,
+				      sizeof(vio_completion_dump_buffer));
+	if (data_vio->is_duplicate) {
+		snprintf(vio_block_number_dump_buffer,
+			 sizeof(vio_block_number_dump_buffer), "P%llu L%llu D%llu",
+			 data_vio->allocation.pbn, data_vio->logical.lbn,
+			 data_vio->duplicate.pbn);
+	} else if (data_vio_has_allocation(data_vio)) {
+		snprintf(vio_block_number_dump_buffer,
+			 sizeof(vio_block_number_dump_buffer), "P%llu L%llu",
+			 data_vio->allocation.pbn, data_vio->logical.lbn);
+	} else {
+		snprintf(vio_block_number_dump_buffer,
+			 sizeof(vio_block_number_dump_buffer), "L%llu",
+			 data_vio->logical.lbn);
+	}
+
+	if (data_vio->flush_generation != 0) {
+		snprintf(vio_flush_generation_buffer,
+			 sizeof(vio_flush_generation_buffer), " FG%llu",
+			 data_vio->flush_generation);
+	} else {
+		vio_flush_generation_buffer[0] = 0;
+	}
+
+	encode_vio_dump_flags(data_vio, flags_dump_buffer);
+
+	vdo_log_info("	vio %px %s%s %s %s%s", data_vio,
+		     vio_block_number_dump_buffer,
+		     vio_flush_generation_buffer,
+		     get_data_vio_operation_name(data_vio),
+		     vio_completion_dump_buffer,
+		     flags_dump_buffer);
+	/*
+	 * might want info on: wantUDSAnswer / operation / status
+	 * might want info on: bio / bios_merged
+	 */
+
+	dump_vio_waiters(&data_vio->logical.waiters, "lbn");
+
+	/* might want to dump more info from vio here */
+}