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, 417 insertions, 0 deletions

diff --git a/lib/842/842_decompress.c b/lib/842/842_decompress.c
new file mode 100644
index 000000000..582085ef8
--- /dev/null
+++ b/lib/842/842_decompress.c
@@ -0,0 +1,417 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * 842 Software Decompression
+ *
+ * Copyright (C) 2015 Dan Streetman, IBM Corp
+ *
+ * See 842.h for details of the 842 compressed format.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+#define MODULE_NAME "842_decompress"
+
+#include "842.h"
+#include "842_debugfs.h"
+
+/* rolling fifo sizes */
+#define I2_FIFO_SIZE	(2 * (1 << I2_BITS))
+#define I4_FIFO_SIZE	(4 * (1 << I4_BITS))
+#define I8_FIFO_SIZE	(8 * (1 << I8_BITS))
+
+static u8 decomp_ops[OPS_MAX][4] = {
+	{ D8, N0, N0, N0 },
+	{ D4, D2, I2, N0 },
+	{ D4, I2, D2, N0 },
+	{ D4, I2, I2, N0 },
+	{ D4, I4, N0, N0 },
+	{ D2, I2, D4, N0 },
+	{ D2, I2, D2, I2 },
+	{ D2, I2, I2, D2 },
+	{ D2, I2, I2, I2 },
+	{ D2, I2, I4, N0 },
+	{ I2, D2, D4, N0 },
+	{ I2, D4, I2, N0 },
+	{ I2, D2, I2, D2 },
+	{ I2, D2, I2, I2 },
+	{ I2, D2, I4, N0 },
+	{ I2, I2, D4, N0 },
+	{ I2, I2, D2, I2 },
+	{ I2, I2, I2, D2 },
+	{ I2, I2, I2, I2 },
+	{ I2, I2, I4, N0 },
+	{ I4, D4, N0, N0 },
+	{ I4, D2, I2, N0 },
+	{ I4, I2, D2, N0 },
+	{ I4, I2, I2, N0 },
+	{ I4, I4, N0, N0 },
+	{ I8, N0, N0, N0 }
+};
+
+struct sw842_param {
+	u8 *in;
+	u8 bit;
+	u64 ilen;
+	u8 *out;
+	u8 *ostart;
+	u64 olen;
+};
+
+#define beN_to_cpu(d, s)					\
+	((s) == 2 ? be16_to_cpu(get_unaligned((__be16 *)d)) :	\
+	 (s) == 4 ? be32_to_cpu(get_unaligned((__be32 *)d)) :	\
+	 (s) == 8 ? be64_to_cpu(get_unaligned((__be64 *)d)) :	\
+	 0)
+
+static int next_bits(struct sw842_param *p, u64 *d, u8 n);
+
+static int __split_next_bits(struct sw842_param *p, u64 *d, u8 n, u8 s)
+{
+	u64 tmp = 0;
+	int ret;
+
+	if (n <= s) {
+		pr_debug("split_next_bits invalid n %u s %u\n", n, s);
+		return -EINVAL;
+	}
+
+	ret = next_bits(p, &tmp, n - s);
+	if (ret)
+		return ret;
+	ret = next_bits(p, d, s);
+	if (ret)
+		return ret;
+	*d |= tmp << s;
+	return 0;
+}
+
+static int next_bits(struct sw842_param *p, u64 *d, u8 n)
+{
+	u8 *in = p->in, b = p->bit, bits = b + n;
+
+	if (n > 64) {
+		pr_debug("next_bits invalid n %u\n", n);
+		return -EINVAL;
+	}
+
+	/* split this up if reading > 8 bytes, or if we're at the end of
+	 * the input buffer and would read past the end
+	 */
+	if (bits > 64)
+		return __split_next_bits(p, d, n, 32);
+	else if (p->ilen < 8 && bits > 32 && bits <= 56)
+		return __split_next_bits(p, d, n, 16);
+	else if (p->ilen < 4 && bits > 16 && bits <= 24)
+		return __split_next_bits(p, d, n, 8);
+
+	if (DIV_ROUND_UP(bits, 8) > p->ilen)
+		return -EOVERFLOW;
+
+	if (bits <= 8)
+		*d = *in >> (8 - bits);
+	else if (bits <= 16)
+		*d = be16_to_cpu(get_unaligned((__be16 *)in)) >> (16 - bits);
+	else if (bits <= 32)
+		*d = be32_to_cpu(get_unaligned((__be32 *)in)) >> (32 - bits);
+	else
+		*d = be64_to_cpu(get_unaligned((__be64 *)in)) >> (64 - bits);
+
+	*d &= GENMASK_ULL(n - 1, 0);
+
+	p->bit += n;
+
+	if (p->bit > 7) {
+		p->in += p->bit / 8;
+		p->ilen -= p->bit / 8;
+		p->bit %= 8;
+	}
+
+	return 0;
+}
+
+static int do_data(struct sw842_param *p, u8 n)
+{
+	u64 v;
+	int ret;
+
+	if (n > p->olen)
+		return -ENOSPC;
+
+	ret = next_bits(p, &v, n * 8);
+	if (ret)
+		return ret;
+
+	switch (n) {
+	case 2:
+		put_unaligned(cpu_to_be16((u16)v), (__be16 *)p->out);
+		break;
+	case 4:
+		put_unaligned(cpu_to_be32((u32)v), (__be32 *)p->out);
+		break;
+	case 8:
+		put_unaligned(cpu_to_be64((u64)v), (__be64 *)p->out);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	p->out += n;
+	p->olen -= n;
+
+	return 0;
+}
+
+static int __do_index(struct sw842_param *p, u8 size, u8 bits, u64 fsize)
+{
+	u64 index, offset, total = round_down(p->out - p->ostart, 8);
+	int ret;
+
+	ret = next_bits(p, &index, bits);
+	if (ret)
+		return ret;
+
+	offset = index * size;
+
+	/* a ring buffer of fsize is used; correct the offset */
+	if (total > fsize) {
+		/* this is where the current fifo is */
+		u64 section = round_down(total, fsize);
+		/* the current pos in the fifo */
+		u64 pos = total - section;
+
+		/* if the offset is past/at the pos, we need to
+		 * go back to the last fifo section
+		 */
+		if (offset >= pos)
+			section -= fsize;
+
+		offset += section;
+	}
+
+	if (offset + size > total) {
+		pr_debug("index%x %lx points past end %lx\n", size,
+			 (unsigned long)offset, (unsigned long)total);
+		return -EINVAL;
+	}
+
+	if (size != 2 && size != 4 && size != 8)
+		WARN(1, "__do_index invalid size %x\n", size);
+	else
+		pr_debug("index%x to %lx off %lx adjoff %lx tot %lx data %lx\n",
+			 size, (unsigned long)index,
+			 (unsigned long)(index * size), (unsigned long)offset,
+			 (unsigned long)total,
+			 (unsigned long)beN_to_cpu(&p->ostart[offset], size));
+
+	memcpy(p->out, &p->ostart[offset], size);
+	p->out += size;
+	p->olen -= size;
+
+	return 0;
+}
+
+static int do_index(struct sw842_param *p, u8 n)
+{
+	switch (n) {
+	case 2:
+		return __do_index(p, 2, I2_BITS, I2_FIFO_SIZE);
+	case 4:
+		return __do_index(p, 4, I4_BITS, I4_FIFO_SIZE);
+	case 8:
+		return __do_index(p, 8, I8_BITS, I8_FIFO_SIZE);
+	default:
+		return -EINVAL;
+	}
+}
+
+static int do_op(struct sw842_param *p, u8 o)
+{
+	int i, ret = 0;
+
+	if (o >= OPS_MAX)
+		return -EINVAL;
+
+	for (i = 0; i < 4; i++) {
+		u8 op = decomp_ops[o][i];
+
+		pr_debug("op is %x\n", op);
+
+		switch (op & OP_ACTION) {
+		case OP_ACTION_DATA:
+			ret = do_data(p, op & OP_AMOUNT);
+			break;
+		case OP_ACTION_INDEX:
+			ret = do_index(p, op & OP_AMOUNT);
+			break;
+		case OP_ACTION_NOOP:
+			break;
+		default:
+			pr_err("Internal error, invalid op %x\n", op);
+			return -EINVAL;
+		}
+
+		if (ret)
+			return ret;
+	}
+
+	if (sw842_template_counts)
+		atomic_inc(&template_count[o]);
+
+	return 0;
+}
+
+/**
+ * sw842_decompress
+ *
+ * Decompress the 842-compressed buffer of length @ilen at @in
+ * to the output buffer @out, using no more than @olen bytes.
+ *
+ * The compressed buffer must be only a single 842-compressed buffer,
+ * with the standard format described in the comments in 842.h
+ * Processing will stop when the 842 "END" template is detected,
+ * not the end of the buffer.
+ *
+ * Returns: 0 on success, error on failure.  The @olen parameter
+ * will contain the number of output bytes written on success, or
+ * 0 on error.
+ */
+int sw842_decompress(const u8 *in, unsigned int ilen,
+		     u8 *out, unsigned int *olen)
+{
+	struct sw842_param p;
+	int ret;
+	u64 op, rep, tmp, bytes, total;
+	u64 crc;
+
+	p.in = (u8 *)in;
+	p.bit = 0;
+	p.ilen = ilen;
+	p.out = out;
+	p.ostart = out;
+	p.olen = *olen;
+
+	total = p.olen;
+
+	*olen = 0;
+
+	do {
+		ret = next_bits(&p, &op, OP_BITS);
+		if (ret)
+			return ret;
+
+		pr_debug("template is %lx\n", (unsigned long)op);
+
+		switch (op) {
+		case OP_REPEAT:
+			ret = next_bits(&p, &rep, REPEAT_BITS);
+			if (ret)
+				return ret;
+
+			if (p.out == out) /* no previous bytes */
+				return -EINVAL;
+
+			/* copy rep + 1 */
+			rep++;
+
+			if (rep * 8 > p.olen)
+				return -ENOSPC;
+
+			while (rep-- > 0) {
+				memcpy(p.out, p.out - 8, 8);
+				p.out += 8;
+				p.olen -= 8;
+			}
+
+			if (sw842_template_counts)
+				atomic_inc(&template_repeat_count);
+
+			break;
+		case OP_ZEROS:
+			if (8 > p.olen)
+				return -ENOSPC;
+
+			memset(p.out, 0, 8);
+			p.out += 8;
+			p.olen -= 8;
+
+			if (sw842_template_counts)
+				atomic_inc(&template_zeros_count);
+
+			break;
+		case OP_SHORT_DATA:
+			ret = next_bits(&p, &bytes, SHORT_DATA_BITS);
+			if (ret)
+				return ret;
+
+			if (!bytes || bytes > SHORT_DATA_BITS_MAX)
+				return -EINVAL;
+
+			while (bytes-- > 0) {
+				ret = next_bits(&p, &tmp, 8);
+				if (ret)
+					return ret;
+				*p.out = (u8)tmp;
+				p.out++;
+				p.olen--;
+			}
+
+			if (sw842_template_counts)
+				atomic_inc(&template_short_data_count);
+
+			break;
+		case OP_END:
+			if (sw842_template_counts)
+				atomic_inc(&template_end_count);
+
+			break;
+		default: /* use template */
+			ret = do_op(&p, op);
+			if (ret)
+				return ret;
+			break;
+		}
+	} while (op != OP_END);
+
+	/*
+	 * crc(0:31) is saved in compressed data starting with the
+	 * next bit after End of stream template.
+	 */
+	ret = next_bits(&p, &crc, CRC_BITS);
+	if (ret)
+		return ret;
+
+	/*
+	 * Validate CRC saved in compressed data.
+	 */
+	if (crc != (u64)crc32_be(0, out, total - p.olen)) {
+		pr_debug("CRC mismatch for decompression\n");
+		return -EINVAL;
+	}
+
+	if (unlikely((total - p.olen) > UINT_MAX))
+		return -ENOSPC;
+
+	*olen = total - p.olen;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(sw842_decompress);
+
+static int __init sw842_init(void)
+{
+	if (sw842_template_counts)
+		sw842_debugfs_create();
+
+	return 0;
+}
+module_init(sw842_init);
+
+static void __exit sw842_exit(void)
+{
+	if (sw842_template_counts)
+		sw842_debugfs_remove();
+}
+module_exit(sw842_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Software 842 Decompressor");
+MODULE_AUTHOR("Dan Streetman <ddstreet@ieee.org>");