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, 316 insertions, 0 deletions

diff --git a/Documentation/sphinx/automarkup.py b/Documentation/sphinx/automarkup.py
new file mode 100644
index 000000000..c2227ab0a
--- /dev/null
+++ b/Documentation/sphinx/automarkup.py
@@ -0,0 +1,316 @@
+# SPDX-License-Identifier: GPL-2.0
+# Copyright 2019 Jonathan Corbet <corbet@lwn.net>
+#
+# Apply kernel-specific tweaks after the initial document processing
+# has been done.
+#
+from docutils import nodes
+import sphinx
+from sphinx import addnodes
+from sphinx.errors import NoUri
+import re
+from itertools import chain
+
+from kernel_abi import get_kernel_abi
+
+#
+# Regex nastiness.  Of course.
+# Try to identify "function()" that's not already marked up some
+# other way.  Sphinx doesn't like a lot of stuff right after a
+# :c:func: block (i.e. ":c:func:`mmap()`s" flakes out), so the last
+# bit tries to restrict matches to things that won't create trouble.
+#
+RE_function = re.compile(r'\b(([a-zA-Z_]\w+)\(\))', flags=re.ASCII)
+
+#
+# Sphinx 3 uses a different C role for each one of struct, union, enum and
+# typedef
+#
+RE_struct = re.compile(r'\b(struct)\s+([a-zA-Z_]\w+)', flags=re.ASCII)
+RE_union = re.compile(r'\b(union)\s+([a-zA-Z_]\w+)', flags=re.ASCII)
+RE_enum = re.compile(r'\b(enum)\s+([a-zA-Z_]\w+)', flags=re.ASCII)
+RE_typedef = re.compile(r'\b(typedef)\s+([a-zA-Z_]\w+)', flags=re.ASCII)
+
+#
+# Detects a reference to a documentation page of the form Documentation/... with
+# an optional extension
+#
+RE_doc = re.compile(r'(\bDocumentation/)?((\.\./)*[\w\-/]+)\.(rst|txt)')
+RE_abi_file = re.compile(r'(\bDocumentation/ABI/[\w\-/]+)')
+RE_abi_symbol = re.compile(r'(\b/(sys|config|proc)/[\w\-/]+)')
+
+RE_namespace = re.compile(r'^\s*..\s*c:namespace::\s*(\S+)\s*$')
+
+#
+# Reserved C words that we should skip when cross-referencing
+#
+Skipnames = [ 'for', 'if', 'register', 'sizeof', 'struct', 'unsigned' ]
+
+#
+# Common English words that should not be recognized as C identifiers
+# when following struct/union/enum/typedef keywords.
+# Example: "a simple struct that" in workqueue.rst should not be marked as code.
+#
+Skipidentifiers = [ 'that', 'which', 'where', 'whose' ]
+
+#
+# Many places in the docs refer to common system calls.  It is
+# pointless to try to cross-reference them and, as has been known
+# to happen, somebody defining a function by these names can lead
+# to the creation of incorrect and confusing cross references.  So
+# just don't even try with these names.
+#
+Skipfuncs = [ 'open', 'close', 'read', 'write', 'fcntl', 'mmap',
+              'select', 'poll', 'fork', 'execve', 'clone', 'ioctl',
+              'socket' ]
+
+c_namespace = ''
+
+#
+# Detect references to commits.
+#
+RE_git = re.compile(r'commit\s+(?P<rev>[0-9a-f]{12,40})(?:\s+\(".*?"\))?',
+    flags=re.IGNORECASE | re.DOTALL)
+
+def markup_refs(docname, app, node):
+    t = node.astext()
+    done = 0
+    repl = [ ]
+    #
+    # Associate each regex with the function that will markup its matches
+    #
+
+    markup_func = {RE_doc: markup_doc_ref,
+                           RE_abi_file: markup_abi_file_ref,
+                           RE_abi_symbol: markup_abi_ref,
+                           RE_function: markup_func_ref_sphinx3,
+                           RE_struct: markup_c_ref,
+                           RE_union: markup_c_ref,
+                           RE_enum: markup_c_ref,
+                           RE_typedef: markup_c_ref,
+                           RE_git: markup_git}
+
+    match_iterators = [regex.finditer(t) for regex in markup_func]
+    #
+    # Sort all references by the starting position in text
+    #
+    sorted_matches = sorted(chain(*match_iterators), key=lambda m: m.start())
+    for m in sorted_matches:
+        #
+        # Include any text prior to match as a normal text node.
+        #
+        if m.start() > done:
+            repl.append(nodes.Text(t[done:m.start()]))
+
+        #
+        # Call the function associated with the regex that matched this text and
+        # append its return to the text
+        #
+        repl.append(markup_func[m.re](docname, app, m))
+
+        done = m.end()
+    if done < len(t):
+        repl.append(nodes.Text(t[done:]))
+    return repl
+
+#
+# Keep track of cross-reference lookups that failed so we don't have to
+# do them again.
+#
+failed_lookups = { }
+def failure_seen(target):
+    return (target) in failed_lookups
+def note_failure(target):
+    failed_lookups[target] = True
+
+#
+# In sphinx3 we can cross-reference to C macro and function, each one with its
+# own C role, but both match the same regex, so we try both.
+#
+def markup_func_ref_sphinx3(docname, app, match):
+    base_target = match.group(2)
+    target_text = nodes.Text(match.group(0))
+    possible_targets = [base_target]
+    # Check if this document has a namespace, and if so, try
+    # cross-referencing inside it first.
+    if c_namespace:
+        possible_targets.insert(0, c_namespace + "." + base_target)
+
+    if base_target not in Skipnames:
+        for target in possible_targets:
+            if (target not in Skipfuncs) and not failure_seen(target):
+                lit_text = nodes.literal(classes=['xref', 'c', 'c-func'])
+                lit_text += target_text
+                xref = add_and_resolve_xref(app, docname, 'c', 'function',
+                                            target, contnode=lit_text)
+                if xref:
+                    return xref
+                note_failure(target)
+
+    return target_text
+
+def markup_c_ref(docname, app, match):
+    class_str = {RE_struct: 'c-struct',
+                 RE_union: 'c-union',
+                 RE_enum: 'c-enum',
+                 RE_typedef: 'c-type',
+                 }
+    reftype_str = {RE_struct: 'struct',
+                   RE_union: 'union',
+                   RE_enum: 'enum',
+                   RE_typedef: 'type',
+                   }
+
+    base_target = match.group(2)
+    target_text = nodes.Text(match.group(0))
+    possible_targets = [base_target]
+    # Check if this document has a namespace, and if so, try
+    # cross-referencing inside it first.
+    if c_namespace:
+        possible_targets.insert(0, c_namespace + "." + base_target)
+
+    # Skip common English words that match identifier pattern but are not C code.
+    if base_target in Skipidentifiers:
+        return target_text
+
+    if base_target not in Skipnames:
+        for target in possible_targets:
+            if not (match.re == RE_function and target in Skipfuncs):
+                lit_text = nodes.literal(classes=['xref', 'c', class_str[match.re]])
+                lit_text += target_text
+                xref = add_and_resolve_xref(app, docname, 'c',
+                                            reftype_str[match.re], target,
+                                            contnode=lit_text)
+                if xref:
+                    return xref
+
+    return target_text
+
+#
+# Try to replace a documentation reference of the form Documentation/... with a
+# cross reference to that page
+#
+def markup_doc_ref(docname, app, match):
+    absolute = match.group(1)
+    target = match.group(2)
+    if absolute:
+       target = "/" + target
+
+    xref = add_and_resolve_xref(app, docname, 'std', 'doc', target)
+    if xref:
+        return xref
+    else:
+        return nodes.Text(match.group(0))
+
+#
+# Try to replace a documentation reference for ABI symbols and files
+# with a cross reference to that page
+#
+def markup_abi_ref(docname, app, match, warning=False):
+    kernel_abi = get_kernel_abi()
+
+    fname = match.group(1)
+    target = kernel_abi.xref(fname)
+
+    # Kernel ABI doesn't describe such file or symbol
+    if not target:
+        if warning:
+            kernel_abi.log.warning("%s not found", fname)
+        return nodes.Text(match.group(0))
+
+    xref = add_and_resolve_xref(app, docname, 'std', 'ref', target)
+    if xref:
+        return xref
+    else:
+        return nodes.Text(match.group(0))
+
+def add_and_resolve_xref(app, docname, domain, reftype, target, contnode=None):
+    #
+    # Go through the dance of getting an xref out of the corresponding domain
+    #
+    dom_obj = app.env.domains[domain]
+    pxref = addnodes.pending_xref('', refdomain = domain, reftype = reftype,
+                                  reftarget = target, modname = None,
+                                  classname = None, refexplicit = False)
+
+    #
+    # XXX The Latex builder will throw NoUri exceptions here,
+    # work around that by ignoring them.
+    #
+    try:
+        xref = dom_obj.resolve_xref(app.env, docname, app.builder, reftype,
+                                    target, pxref, contnode)
+    except NoUri:
+        xref = None
+
+    if xref:
+        return xref
+    #
+    # We didn't find the xref; if a container node was supplied,
+    # mark it as a broken xref
+    #
+    if contnode:
+        contnode['classes'].append("broken_xref")
+    return contnode
+
+#
+# Variant of markup_abi_ref() that warns when a reference is not found
+#
+def markup_abi_file_ref(docname, app, match):
+    return markup_abi_ref(docname, app, match, warning=True)
+
+
+def get_c_namespace(app, docname):
+    source = app.env.doc2path(docname)
+    with open(source) as f:
+        for l in f:
+            match = RE_namespace.search(l)
+            if match:
+                return match.group(1)
+    return ''
+
+def markup_git(docname, app, match):
+    # While we could probably assume that we are running in a git
+    # repository, we can't know for sure, so let's just mechanically
+    # turn them into git.kernel.org links without checking their
+    # validity. (Maybe we can do something in the future to warn about
+    # these references if this is explicitly requested.)
+    text = match.group(0)
+    rev = match.group('rev')
+    return nodes.reference('', nodes.Text(text),
+        refuri=f'https://git.kernel.org/torvalds/c/{rev}')
+
+def auto_markup(app, doctree, name):
+    global c_namespace
+    c_namespace = get_c_namespace(app, name)
+    def text_but_not_a_reference(node):
+        # The nodes.literal test catches ``literal text``, its purpose is to
+        # avoid adding cross-references to functions that have been explicitly
+        # marked with cc:func:.
+        if not isinstance(node, nodes.Text) or isinstance(node.parent, nodes.literal):
+            return False
+
+        child_of_reference = False
+        parent = node.parent
+        while parent:
+            if isinstance(parent, nodes.Referential):
+                child_of_reference = True
+                break
+            parent = parent.parent
+        return not child_of_reference
+
+    #
+    # This loop could eventually be improved on.  Someday maybe we
+    # want a proper tree traversal with a lot of awareness of which
+    # kinds of nodes to prune.  But this works well for now.
+    #
+    for para in doctree.traverse(nodes.paragraph):
+        for node in para.traverse(condition=text_but_not_a_reference):
+            node.parent.replace(node, markup_refs(name, app, node))
+
+def setup(app):
+    app.connect('doctree-resolved', auto_markup)
+    return {
+        'parallel_read_safe': True,
+        'parallel_write_safe': True,
+        }