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, 199 insertions, 0 deletions

diff --git a/lib/sg_split.c b/lib/sg_split.c
new file mode 100644
index 000000000..24e8f5e48
--- /dev/null
+++ b/lib/sg_split.c
@@ -0,0 +1,199 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2015 Robert Jarzmik <robert.jarzmik@free.fr>
+ *
+ * Scatterlist splitting helpers.
+ */
+
+#include <linux/scatterlist.h>
+#include <linux/slab.h>
+
+struct sg_splitter {
+	struct scatterlist *in_sg0;
+	int nents;
+	off_t skip_sg0;
+	unsigned int length_last_sg;
+
+	struct scatterlist *out_sg;
+};
+
+static int sg_calculate_split(struct scatterlist *in, int nents, int nb_splits,
+			      off_t skip, const size_t *sizes,
+			      struct sg_splitter *splitters, bool mapped)
+{
+	int i;
+	unsigned int sglen;
+	size_t size = sizes[0], len;
+	struct sg_splitter *curr = splitters;
+	struct scatterlist *sg;
+
+	for (i = 0; i < nb_splits; i++) {
+		splitters[i].in_sg0 = NULL;
+		splitters[i].nents = 0;
+	}
+
+	for_each_sg(in, sg, nents, i) {
+		sglen = mapped ? sg_dma_len(sg) : sg->length;
+		if (skip > sglen) {
+			skip -= sglen;
+			continue;
+		}
+
+		len = min_t(size_t, size, sglen - skip);
+		if (!curr->in_sg0) {
+			curr->in_sg0 = sg;
+			curr->skip_sg0 = skip;
+		}
+		size -= len;
+		curr->nents++;
+		curr->length_last_sg = len;
+
+		while (!size && (skip + len < sglen) && (--nb_splits > 0)) {
+			curr++;
+			size = *(++sizes);
+			skip += len;
+			len = min_t(size_t, size, sglen - skip);
+
+			curr->in_sg0 = sg;
+			curr->skip_sg0 = skip;
+			curr->nents = 1;
+			curr->length_last_sg = len;
+			size -= len;
+		}
+		skip = 0;
+
+		if (!size && --nb_splits > 0) {
+			curr++;
+			size = *(++sizes);
+		}
+
+		if (!nb_splits)
+			break;
+	}
+
+	return (size || !splitters[0].in_sg0) ? -EINVAL : 0;
+}
+
+static void sg_split_phys(struct sg_splitter *splitters, const int nb_splits)
+{
+	int i, j;
+	struct scatterlist *in_sg, *out_sg;
+	struct sg_splitter *split;
+
+	for (i = 0, split = splitters; i < nb_splits; i++, split++) {
+		in_sg = split->in_sg0;
+		out_sg = split->out_sg;
+		for (j = 0; j < split->nents; j++, out_sg++) {
+			*out_sg = *in_sg;
+			if (!j) {
+				out_sg->offset += split->skip_sg0;
+				out_sg->length -= split->skip_sg0;
+			}
+			sg_dma_address(out_sg) = 0;
+			sg_dma_len(out_sg) = 0;
+			in_sg = sg_next(in_sg);
+		}
+		out_sg[-1].length = split->length_last_sg;
+		sg_mark_end(out_sg - 1);
+	}
+}
+
+static void sg_split_mapped(struct sg_splitter *splitters, const int nb_splits)
+{
+	int i, j;
+	struct scatterlist *in_sg, *out_sg;
+	struct sg_splitter *split;
+
+	for (i = 0, split = splitters; i < nb_splits; i++, split++) {
+		in_sg = split->in_sg0;
+		out_sg = split->out_sg;
+		for (j = 0; j < split->nents; j++, out_sg++) {
+			sg_dma_address(out_sg) = sg_dma_address(in_sg);
+			sg_dma_len(out_sg) = sg_dma_len(in_sg);
+			if (!j) {
+				sg_dma_address(out_sg) += split->skip_sg0;
+				sg_dma_len(out_sg) -= split->skip_sg0;
+			}
+			in_sg = sg_next(in_sg);
+		}
+		sg_dma_len(--out_sg) = split->length_last_sg;
+	}
+}
+
+/**
+ * sg_split - split a scatterlist into several scatterlists
+ * @in: the input sg list
+ * @in_mapped_nents: the result of a dma_map_sg(in, ...), or 0 if not mapped.
+ * @skip: the number of bytes to skip in the input sg list
+ * @nb_splits: the number of desired sg outputs
+ * @split_sizes: the respective size of each output sg list in bytes
+ * @out: an array where to store the allocated output sg lists
+ * @out_mapped_nents: the resulting sg lists mapped number of sg entries. Might
+ *                    be NULL if sglist not already mapped (in_mapped_nents = 0)
+ * @gfp_mask: the allocation flag
+ *
+ * This function splits the input sg list into nb_splits sg lists, which are
+ * allocated and stored into out.
+ * The @in is split into :
+ *  - @out[0], which covers bytes [@skip .. @skip + @split_sizes[0] - 1] of @in
+ *  - @out[1], which covers bytes [@skip + split_sizes[0] ..
+ *                                 @skip + @split_sizes[0] + @split_sizes[1] -1]
+ * etc ...
+ * It will be the caller's duty to kfree() out array members.
+ *
+ * Returns 0 upon success, or error code
+ */
+int sg_split(struct scatterlist *in, const int in_mapped_nents,
+	     const off_t skip, const int nb_splits,
+	     const size_t *split_sizes,
+	     struct scatterlist **out, int *out_mapped_nents,
+	     gfp_t gfp_mask)
+{
+	int i, ret;
+	struct sg_splitter *splitters;
+
+	splitters = kzalloc_objs(*splitters, nb_splits, gfp_mask);
+	if (!splitters)
+		return -ENOMEM;
+
+	ret = sg_calculate_split(in, sg_nents(in), nb_splits, skip, split_sizes,
+			   splitters, false);
+	if (ret < 0)
+		goto err;
+
+	ret = -ENOMEM;
+	for (i = 0; i < nb_splits; i++) {
+		splitters[i].out_sg = kmalloc_objs(struct scatterlist,
+						   splitters[i].nents, gfp_mask);
+		if (!splitters[i].out_sg)
+			goto err;
+	}
+
+	/*
+	 * The order of these 3 calls is important and should be kept.
+	 */
+	sg_split_phys(splitters, nb_splits);
+	if (in_mapped_nents) {
+		ret = sg_calculate_split(in, in_mapped_nents, nb_splits, skip,
+					 split_sizes, splitters, true);
+		if (ret < 0)
+			goto err;
+		sg_split_mapped(splitters, nb_splits);
+	}
+
+	for (i = 0; i < nb_splits; i++) {
+		out[i] = splitters[i].out_sg;
+		if (out_mapped_nents)
+			out_mapped_nents[i] = splitters[i].nents;
+	}
+
+	kfree(splitters);
+	return 0;
+
+err:
+	for (i = 0; i < nb_splits; i++)
+		kfree(splitters[i].out_sg);
+	kfree(splitters);
+	return ret;
+}
+EXPORT_SYMBOL(sg_split);