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, 147 insertions, 0 deletions

diff --git a/scripts/check_extable.sh b/scripts/check_extable.sh
new file mode 100755
index 000000000..4b380564c
--- /dev/null
+++ b/scripts/check_extable.sh
@@ -0,0 +1,147 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# (c) 2015, Quentin Casasnovas <quentin.casasnovas@oracle.com>
+
+obj=$1
+
+file ${obj} | grep -q ELF || (echo "${obj} is not an ELF file." 1>&2 ; exit 0)
+
+# Bail out early if there isn't an __ex_table section in this object file.
+objdump -hj __ex_table ${obj} 2> /dev/null > /dev/null
+[ $? -ne 0 ] && exit 0
+
+white_list=.text,.fixup
+
+suspicious_relocs=$(objdump -rj __ex_table ${obj}  | tail -n +6 |
+			grep -v $(eval echo -e{${white_list}}) | awk '{print $3}')
+
+# No suspicious relocs in __ex_table, jobs a good'un
+[ -z "${suspicious_relocs}" ] && exit 0
+
+
+# After this point, something is seriously wrong since we just found out we
+# have some relocations in __ex_table which point to sections which aren't
+# white listed.  If you're adding a new section in the Linux kernel, and
+# you're expecting this section to contain code which can fault (i.e. the
+# __ex_table relocation to your new section is expected), simply add your
+# new section to the white_list variable above.  If not, you're probably
+# doing something wrong and the rest of this code is just trying to print
+# you more information about it.
+
+function find_section_offset_from_symbol()
+{
+    eval $(objdump -t ${obj} | grep ${1} | sed 's/\([0-9a-f]\+\) .\{7\} \([^ \t]\+\).*/section="\2"; section_offset="0x\1" /')
+
+    # addr2line takes addresses in hexadecimal...
+    section_offset=$(printf "0x%016x" $(( ${section_offset} + $2 )) )
+}
+
+function find_symbol_and_offset_from_reloc()
+{
+    # Extract symbol and offset from the objdump output
+    eval $(echo $reloc | sed 's/\([^+]\+\)+\?\(0x[0-9a-f]\+\)\?/symbol="\1"; symbol_offset="\2"/')
+
+    # When the relocation points to the begining of a symbol or section, it
+    # won't print the offset since it is zero.
+    if [ -z "${symbol_offset}" ]; then
+	symbol_offset=0x0
+    fi
+}
+
+function find_alt_replacement_target()
+{
+    # The target of the .altinstr_replacement is the relocation just before
+    # the .altinstr_replacement one.
+    eval $(objdump -rj .altinstructions ${obj} | grep -B1 "${section}+${section_offset}" | head -n1 | awk '{print $3}' |
+	   sed 's/\([^+]\+\)+\(0x[0-9a-f]\+\)/alt_target_section="\1"; alt_target_offset="\2"/')
+}
+
+function handle_alt_replacement_reloc()
+{
+    # This will define alt_target_section and alt_target_section_offset
+    find_alt_replacement_target ${section} ${section_offset}
+
+    echo "Error: found a reference to .altinstr_replacement in __ex_table:"
+    addr2line -fip -j ${alt_target_section} -e ${obj} ${alt_target_offset} | awk '{print "\t" $0}'
+
+    error=true
+}
+
+function is_executable_section()
+{
+    objdump -hwj ${section} ${obj} | grep -q CODE
+    return $?
+}
+
+function handle_suspicious_generic_reloc()
+{
+    if is_executable_section ${section}; then
+	# We've got a relocation to a non white listed _executable_
+	# section, print a warning so the developper adds the section to
+	# the white list or fix his code.  We try to pretty-print the file
+	# and line number where that relocation was added.
+	echo "Warning: found a reference to section \"${section}\" in __ex_table:"
+	addr2line -fip -j ${section} -e ${obj} ${section_offset} | awk '{print "\t" $0}'
+    else
+	# Something is definitively wrong here since we've got a relocation
+	# to a non-executable section, there's no way this would ever be
+	# running in the kernel.
+	echo "Error: found a reference to non-executable section \"${section}\" in __ex_table at offset ${section_offset}"
+	error=true
+    fi
+}
+
+function handle_suspicious_reloc()
+{
+    case "${section}" in
+	".altinstr_replacement")
+	    handle_alt_replacement_reloc ${section} ${section_offset}
+	    ;;
+	*)
+	    handle_suspicious_generic_reloc ${section} ${section_offset}
+	    ;;
+    esac
+}
+
+function diagnose()
+{
+
+    for reloc in ${suspicious_relocs}; do
+	# Let's find out where the target of the relocation in __ex_table
+	# is, this will define ${symbol} and ${symbol_offset}
+	find_symbol_and_offset_from_reloc ${reloc}
+
+	# When there's a global symbol at the place of the relocation,
+	# objdump will use it instead of giving us a section+offset, so
+	# let's find out which section is this symbol in and the total
+	# offset withing that section.
+	find_section_offset_from_symbol ${symbol} ${symbol_offset}
+
+	# In this case objdump was presenting us with a reloc to a symbol
+	# rather than a section. Now that we've got the actual section,
+	# we can skip it if it's in the white_list.
+	if [ -z "$( echo $section | grep -v $(eval echo -e{${white_list}}))" ]; then
+	    continue;
+	fi
+
+	# Will either print a warning if the relocation happens to be in a
+	# section we do not know but has executable bit set, or error out.
+	handle_suspicious_reloc
+    done
+}
+
+function check_debug_info() {
+    objdump -hj .debug_info ${obj} 2> /dev/null > /dev/null ||
+	echo -e "${obj} does not contain debug information, the addr2line output will be limited.\n" \
+	     "Recompile ${obj} with CONFIG_DEBUG_INFO to get a more useful output."
+}
+
+check_debug_info
+
+diagnose
+
+if [ "${error}" ]; then
+    exit 1
+fi
+
+exit 0