Merge branch 'master' into develop

CHANGELOG.md

@@ -2,6 +2,12 @@
 
 This is a list of notable changes to Hyperscan, in reverse chronological order.
 
+## [4.4.1] 2017-02-28
+- Bugfixes to fix issues where stale data was being referenced in scratch
+  memory. In particular this may have resulted in hs_close_stream()
+  referencing data from other previously scanned streams. This may result in
+  incorrect matches being been reported.
+
 ## [4.4.0] 2017-01-20
 - Introduce the "fat runtime" build. This will build several variants of the
   Hyperscan scanning engine specialised for different processor feature sets,
@@ -136,7 +142,9 @@ This is a list of notable changes to Hyperscan, in reverse chronological order.
   supplied with a NULL scratch pointer if no matches are required. This is in
   line with the behaviour of `hs_close_stream()`.
 - Disallow bounded repeats with a very large minimum repeat but no maximum,
-  i.e. {N,} for very large N.
+  i.e. {
+    N,
+} for very large N.
 - Reduce compile memory usage in literal set explansion for some large cases.
 
 ## [4.0.0] 2015-10-20

CMakeLists.txt

@@ -3,7 +3,7 @@ project (Hyperscan C CXX)
 
 set (HS_MAJOR_VERSION 4)
 set (HS_MINOR_VERSION 4)
-set (HS_PATCH_VERSION 0)
+set (HS_PATCH_VERSION 1)
 set (HS_VERSION ${HS_MAJOR_VERSION}.${HS_MINOR_VERSION}.${HS_PATCH_VERSION})
 
 set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)

src/nfa/lbr.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015-2016, Intel Corporation
+ * Copyright (c) 2015-2017, Intel Corporation
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
@@ -77,6 +77,7 @@ void lbrExpandState(const struct lbr_common *l, u64a offset,
 
     const struct RepeatInfo *info = getRepeatInfo(l);
     repeatUnpack(stream_state, info, offset, &lstate->ctrl);
+    lstate->lastEscape = 0;
 }
 
 static really_inline

src/nfa/limex_compile.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015-2016, Intel Corporation
+ * Copyright (c) 2015-2017, Intel Corporation
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
@@ -1803,6 +1803,12 @@ struct Factory {
             assert(cyclic != NO_STATE);
             maskSetBit(limex->repeatCyclicMask, cyclic);
         }
+        /* also include tugs in repeat cyclic mask */
+        for (NFAVertex v : args.tugs) {
+            u32 v_state = args.state_ids.at(v);
+            assert(v_state != NO_STATE);
+            maskSetBit(limex->repeatCyclicMask, v_state);
+        }
     }
 
     static

src/nfa/limex_internal.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015-2016, Intel Corporation
+ * Copyright (c) 2015-2017, Intel Corporation
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
@@ -151,7 +151,7 @@ struct LimExNFA##size {                                                     \
                                     *  followers */                         \
     u_##size compressMask; /**< switch off before compress */               \
     u_##size exceptionMask;                                                 \
-    u_##size repeatCyclicMask;                                              \
+    u_##size repeatCyclicMask; /**< also includes tug states */             \
     u_##size zombieMask; /**< zombie if in any of the set states */         \
     u_##size shift[MAX_SHIFT_COUNT];                                        \
     u32 shiftCount; /**< number of shift masks used */                      \

src/nfa/limex_runtime_impl.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015-2016, Intel Corporation
+ * Copyright (c) 2015-2017, Intel Corporation
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
@@ -393,19 +393,16 @@ void COMPRESS_REPEATS_FN(const IMPL_NFA_T *limex, void *dest, void *src,
         DEBUG_PRINTF("repeat %u\n", i);
         const struct NFARepeatInfo *info = GET_NFA_REPEAT_INFO_FN(limex, i);
 
-        if (!TESTBIT_STATE(s, info->cyclicState)) {
+        const ENG_STATE_T *tug_mask =
+            (const ENG_STATE_T *)((const char *)info + info->tugMaskOffset);
+        /* repeat may still be inspected if its tug state is on */
+        if (!TESTBIT_STATE(s, info->cyclicState)
+            && ISZERO_STATE(AND_STATE(s, LOAD_FROM_ENG(tug_mask)))) {
             DEBUG_PRINTF("is dead\n");
             continue;
         }
 
         const struct RepeatInfo *repeat = getRepeatInfo(info);
-        if (repeatHasMatch(repeat, &ctrl[i], state_base + info->stateOffset,
-                           offset) == REPEAT_STALE) {
-            DEBUG_PRINTF("is stale, clearing state\n");
-            CLEARBIT_STATE(&s, info->cyclicState);
-            continue;
-        }
-
         DEBUG_PRINTF("packing state (packedCtrlOffset=%u)\n",
                      info->packedCtrlOffset);
         repeatPack(state_base + info->packedCtrlOffset, repeat, &ctrl[i],
@@ -448,8 +445,11 @@ void EXPAND_REPEATS_FN(const IMPL_NFA_T *limex, void *dest, const void *src,
     for (u32 i = 0; i < limex->repeatCount; i++) {
         DEBUG_PRINTF("repeat %u\n", i);
         const struct NFARepeatInfo *info = GET_NFA_REPEAT_INFO_FN(limex, i);
+        const ENG_STATE_T *tug_mask =
+            (const ENG_STATE_T *)((const char *)info + info->tugMaskOffset);
 
-        if (!TESTBIT_STATE(cyclics, info->cyclicState)) {
+        if (!TESTBIT_STATE(cyclics, info->cyclicState)
+            && ISZERO_STATE(AND_STATE(cyclics, LOAD_FROM_ENG(tug_mask)))) {
             DEBUG_PRINTF("is dead\n");
             continue;
         }

src/nfa/repeat.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015-2016, Intel Corporation
+ * Copyright (c) 2015-2017, Intel Corporation
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
@@ -177,6 +177,10 @@ u64a repeatLastTopRange(const union RepeatControl *ctrl, const void *state) {
 
 u64a repeatLastTopBitmap(const union RepeatControl *ctrl) {
     const struct RepeatBitmapControl *xs = &ctrl->bitmap;
+    if (!xs->bitmap) {
+        /* last top was too long ago */
+        return 0;
+    }
     return xs->offset + 63 - clz64(xs->bitmap);
 }
 

src/rose/program_runtime.h

@@ -1209,6 +1209,8 @@ hwlmcb_rv_t roseEnginesEod(const struct RoseEngine *rose,
 
     const u8 *aa = getActiveLeafArray(rose, scratch->core_info.state);
     const u32 aaCount = rose->activeArrayCount;
+    const u32 qCount = rose->queueCount;
+    struct fatbit *aqa = scratch->aqa;
 
     const struct mmbit_sparse_iter *it = getByOffset(rose, iter_offset);
     assert(ISALIGNED(it));
@@ -1221,6 +1223,10 @@ hwlmcb_rv_t roseEnginesEod(const struct RoseEngine *rose,
          qi = mmbit_sparse_iter_next(aa, aaCount, qi, &idx, it, si_state)) {
         DEBUG_PRINTF("checking nfa %u\n", qi);
         struct mq *q = scratch->queues + qi;
+        if (!fatbit_set(aqa, qCount, qi)) {
+            initQueue(q, qi, rose, scratch);
+        }
+
         assert(q->nfa == getNfaByQueue(rose, qi));
         assert(nfaAcceptsEod(q->nfa));
 

src/runtime.c

@@ -186,6 +186,18 @@ void rawBlockExec(const struct RoseEngine *rose, struct hs_scratch *scratch) {
     roseBlockExec(rose, scratch);
 }
 
+static really_inline
+void pureLiteralInitScratch(struct hs_scratch *scratch, u64a offset) {
+    // Some init has already been done.
+    assert(offset == scratch->core_info.buf_offset);
+
+    scratch->tctxt.lit_offset_adjust = offset + 1;
+    scratch->tctxt.lastEndOffset = offset;
+    scratch->tctxt.delayLastEndOffset = offset;
+    scratch->tctxt.filledDelayedSlots = 0;
+    scratch->al_log_sum = 0;
+}
+
 static really_inline
 void pureLiteralBlockExec(const struct RoseEngine *rose,
                           struct hs_scratch *scratch) {
@@ -198,9 +210,8 @@ void pureLiteralBlockExec(const struct RoseEngine *rose,
     size_t length = scratch->core_info.len;
     DEBUG_PRINTF("rose engine %d\n", rose->runtimeImpl);
 
-    // RoseContext values that need to be set for use by roseCallback.
+    pureLiteralInitScratch(scratch, 0);
     scratch->tctxt.groups = rose->initialGroups;
-    scratch->tctxt.lit_offset_adjust = 1;
 
     hwlmExec(ftable, buffer, length, 0, roseCallback, scratch,
              rose->initialGroups);
@@ -743,9 +754,8 @@ void pureLiteralStreamExec(struct hs_stream *stream_state,
     DEBUG_PRINTF("::: streaming rose ::: offset = %llu len = %zu\n",
                  stream_state->offset, scratch->core_info.len);
 
-    // RoseContext values that need to be set for use by roseCallback.
+    pureLiteralInitScratch(scratch, stream_state->offset);
     scratch->tctxt.groups = loadGroups(rose, scratch->core_info.state);
-    scratch->tctxt.lit_offset_adjust = scratch->core_info.buf_offset + 1;
 
     // Pure literal cases don't have floatingMinDistance set, so we always
     // start the match region at zero.

unit/hyperscan/behaviour.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015-2016, Intel Corporation
+ * Copyright (c) 2015-2017, Intel Corporation
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
@@ -1172,6 +1172,106 @@ TEST(HyperscanTestBehaviour, Vectored7) {
     hs_free_database(db);
 }
 
+TEST(HyperscanTestBehaviour, MultiStream1) {
+    hs_error_t err;
+
+    // build a database
+    hs_database_t *db = buildDB("foo.*bar.*\\b", 0, 0, HS_MODE_STREAM);
+    ASSERT_TRUE(db != nullptr);
+
+    // alloc some scratch
+    hs_scratch_t *scratch = nullptr;
+    err = hs_alloc_scratch(db, &scratch);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_TRUE(scratch != nullptr);
+
+    hs_stream_t *stream = nullptr;
+    err = hs_open_stream(db, 0, &stream);
+    ASSERT_EQ(HS_SUCCESS, err);
+    ASSERT_TRUE(stream != nullptr);
+
+    hs_stream_t *stream2 = nullptr;
+    err = hs_open_stream(db, 0, &stream2);
+    ASSERT_EQ(HS_SUCCESS, err);
+    ASSERT_TRUE(stream2 != nullptr);
+
+    matchCount = 0;
+    const string data("foo        bara");
+    err = hs_scan_stream(stream, data.c_str(), data.size(), 0, scratch,
+                         countHandler, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(0U, matchCount); // hasn't matched until stream end
+
+    const string data2("foo        bar ");
+    err = hs_scan_stream(stream2, data2.c_str(), data2.size(), 0, scratch,
+                         nullptr, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(0U, matchCount);
+
+    err = hs_close_stream(stream, scratch, countHandler, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(1U, matchCount);
+
+    err = hs_close_stream(stream2, scratch, countHandler, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(1U, matchCount);
+
+    // teardown
+    err = hs_free_scratch(scratch);
+    ASSERT_EQ(HS_SUCCESS, err);
+    hs_free_database(db);
+}
+
+TEST(HyperscanTestBehaviour, MultiStream2) {
+    hs_error_t err;
+
+    // build a database
+    hs_database_t *db = buildDB("foo.*bar.*\\b", 0, 0, HS_MODE_STREAM);
+    ASSERT_TRUE(db != nullptr);
+
+    // alloc some scratch
+    hs_scratch_t *scratch = nullptr;
+    err = hs_alloc_scratch(db, &scratch);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_TRUE(scratch != nullptr);
+
+    hs_stream_t *stream = nullptr;
+    err = hs_open_stream(db, 0, &stream);
+    ASSERT_EQ(HS_SUCCESS, err);
+    ASSERT_TRUE(stream != nullptr);
+
+    hs_stream_t *stream2 = nullptr;
+    err = hs_open_stream(db, 0, &stream2);
+    ASSERT_EQ(HS_SUCCESS, err);
+    ASSERT_TRUE(stream2 != nullptr);
+
+    matchCount = 0;
+    const string data2("foo        bar ");
+    err = hs_scan_stream(stream2, data2.c_str(), data2.size(), 0, scratch,
+                         nullptr, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(0U, matchCount);
+
+    const string data("foo        bara");
+    err = hs_scan_stream(stream, data.c_str(), data.size(), 0, scratch,
+                         countHandler, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(0U, matchCount); // hasn't matched until stream end
+
+    err = hs_close_stream(stream2, scratch, countHandler, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(0U, matchCount);
+
+    err = hs_close_stream(stream, scratch, countHandler, nullptr);
+    ASSERT_EQ(HS_SUCCESS, err);
+    EXPECT_EQ(1U, matchCount);
+
+    // teardown
+    err = hs_free_scratch(scratch);
+    ASSERT_EQ(HS_SUCCESS, err);
+    hs_free_database(db);
+}
+
 TEST(regression, UE_1005) {
     hs_error_t err;
     vector<pattern> patterns;