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, 407 insertions, 0 deletions

diff --git a/lib/buildid.c b/lib/buildid.c
new file mode 100644
index 000000000..c4b737640
--- /dev/null
+++ b/lib/buildid.c
@@ -0,0 +1,407 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/buildid.h>
+#include <linux/cache.h>
+#include <linux/elf.h>
+#include <linux/kernel.h>
+#include <linux/pagemap.h>
+#include <linux/fs.h>
+#include <linux/secretmem.h>
+
+#define BUILD_ID 3
+
+#define MAX_PHDR_CNT 256
+
+void freader_init_from_file(struct freader *r, void *buf, u32 buf_sz,
+			    struct file *file, bool may_fault)
+{
+	memset(r, 0, sizeof(*r));
+	r->buf = buf;
+	r->buf_sz = buf_sz;
+	r->file = file;
+	r->may_fault = may_fault;
+}
+
+void freader_init_from_mem(struct freader *r, const char *data, u64 data_sz)
+{
+	memset(r, 0, sizeof(*r));
+	r->data = data;
+	r->data_sz = data_sz;
+}
+
+static void freader_put_folio(struct freader *r)
+{
+	if (!r->folio)
+		return;
+	kunmap_local(r->addr);
+	folio_put(r->folio);
+	r->folio = NULL;
+}
+
+static int freader_get_folio(struct freader *r, loff_t file_off)
+{
+	/* check if we can just reuse current folio */
+	if (r->folio && file_off >= r->folio_off &&
+	    file_off < r->folio_off + folio_size(r->folio))
+		return 0;
+
+	freader_put_folio(r);
+
+	/* only use page cache lookup - fail if not already cached */
+	r->folio = filemap_get_folio(r->file->f_mapping, file_off >> PAGE_SHIFT);
+
+	if (IS_ERR(r->folio) || !folio_test_uptodate(r->folio)) {
+		if (!IS_ERR(r->folio))
+			folio_put(r->folio);
+		r->folio = NULL;
+		return -EFAULT;
+	}
+
+	r->folio_off = folio_pos(r->folio);
+	r->addr = kmap_local_folio(r->folio, 0);
+
+	return 0;
+}
+
+const void *freader_fetch(struct freader *r, loff_t file_off, size_t sz)
+{
+	size_t folio_sz;
+
+	/* provided internal temporary buffer should be sized correctly */
+	if (WARN_ON(r->buf && sz > r->buf_sz)) {
+		r->err = -E2BIG;
+		return NULL;
+	}
+
+	if (unlikely(file_off + sz < file_off)) {
+		r->err = -EOVERFLOW;
+		return NULL;
+	}
+
+	/* working with memory buffer is much more straightforward */
+	if (!r->buf) {
+		if (file_off + sz > r->data_sz) {
+			r->err = -ERANGE;
+			return NULL;
+		}
+		return r->data + file_off;
+	}
+
+	/* reject secretmem folios created with memfd_secret() */
+	if (secretmem_mapping(r->file->f_mapping)) {
+		r->err = -EFAULT;
+		return NULL;
+	}
+
+	/* use __kernel_read() for sleepable context */
+	if (r->may_fault) {
+		ssize_t ret;
+
+		ret = __kernel_read(r->file, r->buf, sz, &file_off);
+		if (ret != sz) {
+			r->err = (ret < 0) ? ret : -EIO;
+			return NULL;
+		}
+		return r->buf;
+	}
+
+	/* fetch or reuse folio for given file offset */
+	r->err = freader_get_folio(r, file_off);
+	if (r->err)
+		return NULL;
+
+	/* if requested data is crossing folio boundaries, we have to copy
+	 * everything into our local buffer to keep a simple linear memory
+	 * access interface
+	 */
+	folio_sz = folio_size(r->folio);
+	if (file_off + sz > r->folio_off + folio_sz) {
+		u64 part_sz = r->folio_off + folio_sz - file_off, off;
+
+		memcpy(r->buf, r->addr + file_off - r->folio_off, part_sz);
+		off = part_sz;
+
+		while (off < sz) {
+			/* fetch next folio */
+			r->err = freader_get_folio(r, r->folio_off + folio_sz);
+			if (r->err)
+				return NULL;
+			folio_sz = folio_size(r->folio);
+			part_sz = min_t(u64, sz - off, folio_sz);
+			memcpy(r->buf + off, r->addr, part_sz);
+			off += part_sz;
+		}
+
+		return r->buf;
+	}
+
+	/* if data fits in a single folio, just return direct pointer */
+	return r->addr + (file_off - r->folio_off);
+}
+
+void freader_cleanup(struct freader *r)
+{
+	if (!r->buf)
+		return; /* non-file-backed mode */
+
+	freader_put_folio(r);
+}
+
+/*
+ * Parse build id from the note segment. This logic can be shared between
+ * 32-bit and 64-bit system, because Elf32_Nhdr and Elf64_Nhdr are
+ * identical.
+ */
+static int parse_build_id(struct freader *r, unsigned char *build_id, __u32 *size,
+			  loff_t note_off, Elf32_Word note_size)
+{
+	const char note_name[] = "GNU";
+	const size_t note_name_sz = sizeof(note_name);
+	u32 build_id_off, new_off, note_end, name_sz, desc_sz;
+	const Elf32_Nhdr *nhdr;
+	const char *data;
+
+	if (check_add_overflow(note_off, note_size, &note_end))
+		return -EINVAL;
+
+	while (note_end - note_off > sizeof(Elf32_Nhdr) + note_name_sz) {
+		nhdr = freader_fetch(r, note_off, sizeof(Elf32_Nhdr) + note_name_sz);
+		if (!nhdr)
+			return r->err;
+
+		name_sz = READ_ONCE(nhdr->n_namesz);
+		desc_sz = READ_ONCE(nhdr->n_descsz);
+
+		new_off = note_off + sizeof(Elf32_Nhdr);
+		if (check_add_overflow(new_off, ALIGN(name_sz, 4), &new_off) ||
+		    check_add_overflow(new_off, ALIGN(desc_sz, 4), &new_off) ||
+		    new_off > note_end)
+			break;
+
+		if (nhdr->n_type == BUILD_ID &&
+		    name_sz == note_name_sz &&
+		    memcmp(nhdr + 1, note_name, note_name_sz) == 0 &&
+		    desc_sz > 0 && desc_sz <= BUILD_ID_SIZE_MAX) {
+			build_id_off = note_off + sizeof(Elf32_Nhdr) + ALIGN(note_name_sz, 4);
+
+			/* freader_fetch() will invalidate nhdr pointer */
+			data = freader_fetch(r, build_id_off, desc_sz);
+			if (!data)
+				return r->err;
+
+			memcpy(build_id, data, desc_sz);
+			memset(build_id + desc_sz, 0, BUILD_ID_SIZE_MAX - desc_sz);
+			if (size)
+				*size = desc_sz;
+			return 0;
+		}
+
+		note_off = new_off;
+	}
+
+	return -EINVAL;
+}
+
+/* Parse build ID from 32-bit ELF */
+static int get_build_id_32(struct freader *r, unsigned char *build_id, __u32 *size)
+{
+	const Elf32_Ehdr *ehdr;
+	const Elf32_Phdr *phdr;
+	__u32 phnum, phoff, i;
+
+	ehdr = freader_fetch(r, 0, sizeof(Elf32_Ehdr));
+	if (!ehdr)
+		return r->err;
+
+	/* subsequent freader_fetch() calls invalidate pointers, so remember locally */
+	phnum = READ_ONCE(ehdr->e_phnum);
+	phoff = READ_ONCE(ehdr->e_phoff);
+
+	/* set upper bound on amount of segments (phdrs) we iterate */
+	if (phnum > MAX_PHDR_CNT)
+		phnum = MAX_PHDR_CNT;
+
+	/* check that phoff is not large enough to cause an overflow */
+	if (phoff + phnum * sizeof(Elf32_Phdr) < phoff)
+		return -EINVAL;
+
+	for (i = 0; i < phnum; ++i) {
+		phdr = freader_fetch(r, phoff + i * sizeof(Elf32_Phdr), sizeof(Elf32_Phdr));
+		if (!phdr)
+			return r->err;
+
+		if (phdr->p_type == PT_NOTE &&
+		    !parse_build_id(r, build_id, size, READ_ONCE(phdr->p_offset),
+				    READ_ONCE(phdr->p_filesz)))
+			return 0;
+	}
+	return -EINVAL;
+}
+
+/* Parse build ID from 64-bit ELF */
+static int get_build_id_64(struct freader *r, unsigned char *build_id, __u32 *size)
+{
+	const Elf64_Ehdr *ehdr;
+	const Elf64_Phdr *phdr;
+	__u32 phnum, i;
+	__u64 phoff;
+
+	ehdr = freader_fetch(r, 0, sizeof(Elf64_Ehdr));
+	if (!ehdr)
+		return r->err;
+
+	/* subsequent freader_fetch() calls invalidate pointers, so remember locally */
+	phnum = READ_ONCE(ehdr->e_phnum);
+	phoff = READ_ONCE(ehdr->e_phoff);
+
+	/* set upper bound on amount of segments (phdrs) we iterate */
+	if (phnum > MAX_PHDR_CNT)
+		phnum = MAX_PHDR_CNT;
+
+	/* check that phoff is not large enough to cause an overflow */
+	if (phoff + phnum * sizeof(Elf64_Phdr) < phoff)
+		return -EINVAL;
+
+	for (i = 0; i < phnum; ++i) {
+		phdr = freader_fetch(r, phoff + i * sizeof(Elf64_Phdr), sizeof(Elf64_Phdr));
+		if (!phdr)
+			return r->err;
+
+		if (phdr->p_type == PT_NOTE &&
+		    !parse_build_id(r, build_id, size, READ_ONCE(phdr->p_offset),
+				    READ_ONCE(phdr->p_filesz)))
+			return 0;
+	}
+
+	return -EINVAL;
+}
+
+/* enough for Elf64_Ehdr, Elf64_Phdr, and all the smaller requests */
+#define MAX_FREADER_BUF_SZ 64
+
+static int __build_id_parse(struct file *file, unsigned char *build_id,
+			    __u32 *size, bool may_fault)
+{
+	const Elf32_Ehdr *ehdr;
+	struct freader r;
+	char buf[MAX_FREADER_BUF_SZ];
+	int ret;
+
+	freader_init_from_file(&r, buf, sizeof(buf), file, may_fault);
+
+	/* fetch first 18 bytes of ELF header for checks */
+	ehdr = freader_fetch(&r, 0, offsetofend(Elf32_Ehdr, e_type));
+	if (!ehdr) {
+		ret = r.err;
+		goto out;
+	}
+
+	ret = -EINVAL;
+
+	/* compare magic x7f "ELF" */
+	if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) != 0)
+		goto out;
+
+	/* only support executable file and shared object file */
+	if (ehdr->e_type != ET_EXEC && ehdr->e_type != ET_DYN)
+		goto out;
+
+	if (ehdr->e_ident[EI_CLASS] == ELFCLASS32)
+		ret = get_build_id_32(&r, build_id, size);
+	else if (ehdr->e_ident[EI_CLASS] == ELFCLASS64)
+		ret = get_build_id_64(&r, build_id, size);
+out:
+	freader_cleanup(&r);
+	return ret;
+}
+
+/**
+ * build_id_parse_nofault() - Parse build ID of ELF file mapped to vma
+ * @vma:      vma object
+ * @build_id: buffer to store build id, at least BUILD_ID_SIZE long
+ * @size:     returns actual build id size in case of success
+ *
+ * Assumes no page fault can be taken, so if relevant portions of ELF file are
+ * not already paged in, fetching of build ID fails.
+ *
+ * Return: 0 on success; negative error, otherwise
+ */
+int build_id_parse_nofault(struct vm_area_struct *vma, unsigned char *build_id, __u32 *size)
+{
+	if (!vma->vm_file)
+		return -EINVAL;
+
+	return __build_id_parse(vma->vm_file, build_id, size, false /* !may_fault */);
+}
+
+/**
+ * build_id_parse() - Parse build ID of ELF file mapped to VMA
+ * @vma:      vma object
+ * @build_id: buffer to store build id, at least BUILD_ID_SIZE long
+ * @size:     returns actual build id size in case of success
+ *
+ * Assumes faultable context and can cause page faults to bring in file data
+ * into page cache.
+ *
+ * Return: 0 on success; negative error, otherwise
+ */
+int build_id_parse(struct vm_area_struct *vma, unsigned char *build_id, __u32 *size)
+{
+	if (!vma->vm_file)
+		return -EINVAL;
+
+	return __build_id_parse(vma->vm_file, build_id, size, true /* may_fault */);
+}
+
+/**
+ * build_id_parse_file() - Parse build ID of ELF file
+ * @file:      file object
+ * @build_id: buffer to store build id, at least BUILD_ID_SIZE long
+ * @size:     returns actual build id size in case of success
+ *
+ * Assumes faultable context and can cause page faults to bring in file data
+ * into page cache.
+ *
+ * Return: 0 on success; negative error, otherwise
+ */
+int build_id_parse_file(struct file *file, unsigned char *build_id, __u32 *size)
+{
+	return __build_id_parse(file, build_id, size, true /* may_fault */);
+}
+
+/**
+ * build_id_parse_buf - Get build ID from a buffer
+ * @buf:      ELF note section(s) to parse
+ * @buf_size: Size of @buf in bytes
+ * @build_id: Build ID parsed from @buf, at least BUILD_ID_SIZE_MAX long
+ *
+ * Return: 0 on success, -EINVAL otherwise
+ */
+int build_id_parse_buf(const void *buf, unsigned char *build_id, u32 buf_size)
+{
+	struct freader r;
+	int err;
+
+	freader_init_from_mem(&r, buf, buf_size);
+
+	err = parse_build_id(&r, build_id, NULL, 0, buf_size);
+
+	freader_cleanup(&r);
+	return err;
+}
+
+#if IS_ENABLED(CONFIG_STACKTRACE_BUILD_ID) || IS_ENABLED(CONFIG_VMCORE_INFO)
+unsigned char vmlinux_build_id[BUILD_ID_SIZE_MAX] __ro_after_init;
+
+/**
+ * init_vmlinux_build_id - Compute and stash the running kernel's build ID
+ */
+void __init init_vmlinux_build_id(void)
+{
+	extern const void __start_notes;
+	extern const void __stop_notes;
+	unsigned int size = &__stop_notes - &__start_notes;
+
+	build_id_parse_buf(&__start_notes, vmlinux_build_id, size);
+}
+#endif