sandbox/linux/bpf_dsl/codegen.cc

   1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #include "sandbox/linux/bpf_dsl/codegen.h"
   6
   7 #include <linux/filter.h>
   8
   9 #include <limits>
  10 #include <utility>
  11
  12 #include "base/logging.h"
  13
  14 // This CodeGen implementation strives for simplicity while still
  15 // generating acceptable BPF programs under typical usage patterns
  16 // (e.g., by PolicyCompiler).
  17 //
  18 // The key to its simplicity is that BPF programs only support forward
  19 // jumps/branches, which allows constraining the DAG construction API
  20 // to make instruction nodes immutable. Immutable nodes admits a
  21 // simple greedy approach of emitting new instructions as needed and
  22 // then reusing existing ones that have already been emitted. This
  23 // cleanly avoids any need to compute basic blocks or apply
  24 // topological sorting because the API effectively sorts instructions
  25 // for us (e.g., before MakeInstruction() can be called to emit a
  26 // branch instruction, it must have already been called for each
  27 // branch path).
  28 //
  29 // This greedy algorithm is not without (theoretical) weakness though:
  30 //
  31 //   1. In the general case, we don't eliminate dead code.  If needed,
  32 //      we could trace back through the program in Compile() and elide
  33 //      any unneeded instructions, but in practice we only emit live
  34 //      instructions anyway.
  35 //
  36 //   2. By not dividing instructions into basic blocks and sorting, we
  37 //      lose an opportunity to move non-branch/non-return instructions
  38 //      adjacent to their successor instructions, which means we might
  39 //      need to emit additional jumps. But in practice, they'll
  40 //      already be nearby as long as callers don't go out of their way
  41 //      to interleave MakeInstruction() calls for unrelated code
  42 //      sequences.
  43
  44 namespace sandbox {
  45
  46 // kBranchRange is the maximum value that can be stored in
  47 // sock_filter's 8-bit jt and jf fields.
  48 const size_t kBranchRange = std::numeric_limits<uint8_t>::max();
  49
  50 const CodeGen::Node CodeGen::kNullNode;
  51
  52 CodeGen::CodeGen() : program_(), equivalent_(), memos_() {
  53 }
  54
  55 CodeGen::~CodeGen() {
  56 }
  57
  58 void CodeGen::Compile(CodeGen::Node head, Program* out) {
  59   DCHECK(out);
  60   out->assign(program_.rbegin() + Offset(head), program_.rend());
  61 }
  62
  63 CodeGen::Node CodeGen::MakeInstruction(uint16_t code,
  64                                        uint32_t k,
  65                                        Node jt,
  66                                        Node jf) {
  67   // To avoid generating redundant code sequences, we memoize the
  68   // results from AppendInstruction().
  69   auto res = memos_.insert(std::make_pair(MemoKey(code, k, jt, jf), kNullNode));
  70   CodeGen::Node* node = &res.first->second;
  71   if (res.second) {  // Newly inserted memo entry.
  72     *node = AppendInstruction(code, k, jt, jf);
  73   }
  74   return *node;
  75 }
  76
  77 CodeGen::Node CodeGen::AppendInstruction(uint16_t code,
  78                                          uint32_t k,
  79                                          Node jt,
  80                                          Node jf) {
  81   if (BPF_CLASS(code) == BPF_JMP) {
  82     CHECK_NE(BPF_JA, BPF_OP(code)) << "CodeGen inserts JAs as needed";
  83
  84     // Optimally adding jumps is rather tricky, so we use a quick
  85     // approximation: by artificially reducing |jt|'s range, |jt| will
  86     // stay within its true range even if we add a jump for |jf|.
  87     jt = WithinRange(jt, kBranchRange - 1);
  88     jf = WithinRange(jf, kBranchRange);
  89     return Append(code, k, Offset(jt), Offset(jf));
  90   }
  91
  92   CHECK_EQ(kNullNode, jf) << "Non-branch instructions shouldn't provide jf";
  93   if (BPF_CLASS(code) == BPF_RET) {
  94     CHECK_EQ(kNullNode, jt) << "Return instructions shouldn't provide jt";
  95   } else {
  96     // For non-branch/non-return instructions, execution always
  97     // proceeds to the next instruction; so we need to arrange for
  98     // that to be |jt|.
  99     jt = WithinRange(jt, 0);
 100     CHECK_EQ(0U, Offset(jt)) << "ICE: Failed to setup next instruction";
 101   }
 102   return Append(code, k, 0, 0);
 103 }
 104
 105 CodeGen::Node CodeGen::WithinRange(Node target, size_t range) {
 106   // Just use |target| if it's already within range.
 107   if (Offset(target) <= range) {
 108     return target;
 109   }
 110
 111   // Alternatively, look for an equivalent instruction within range.
 112   if (Offset(equivalent_.at(target)) <= range) {
 113     return equivalent_.at(target);
 114   }
 115
 116   // Otherwise, fall back to emitting a jump instruction.
 117   Node jump = Append(BPF_JMP | BPF_JA, Offset(target), 0, 0);
 118   equivalent_.at(target) = jump;
 119   return jump;
 120 }
 121
 122 CodeGen::Node CodeGen::Append(uint16_t code, uint32_t k, size_t jt, size_t jf) {
 123   if (BPF_CLASS(code) == BPF_JMP && BPF_OP(code) != BPF_JA) {
 124     CHECK_LE(jt, kBranchRange);
 125     CHECK_LE(jf, kBranchRange);
 126   } else {
 127     CHECK_EQ(0U, jt);
 128     CHECK_EQ(0U, jf);
 129   }
 130
 131   CHECK_LT(program_.size(), static_cast<size_t>(BPF_MAXINSNS));
 132   CHECK_EQ(program_.size(), equivalent_.size());
 133
 134   Node res = program_.size();
 135   program_.push_back(sock_filter{code, jt, jf, k});
 136   equivalent_.push_back(res);
 137   return res;
 138 }
 139
 140 size_t CodeGen::Offset(Node target) const {
 141   CHECK_LT(target, program_.size()) << "Bogus offset target node";
 142   return (program_.size() - 1) - target;
 143 }
 144
 145 // TODO(mdempsky): Move into a general base::Tuple helper library.
 146 bool CodeGen::MemoKeyLess::operator()(const MemoKey& lhs,
 147                                       const MemoKey& rhs) const {
 148   if (get<0>(lhs) != get<0>(rhs))
 149     return get<0>(lhs) < get<0>(rhs);
 150   if (get<1>(lhs) != get<1>(rhs))
 151     return get<1>(lhs) < get<1>(rhs);
 152   if (get<2>(lhs) != get<2>(rhs))
 153     return get<2>(lhs) < get<2>(rhs);
 154   if (get<3>(lhs) != get<3>(rhs))
 155     return get<3>(lhs) < get<3>(rhs);
 156   return false;
 157 }
 158
 159 }  // namespace sandbox