/**
* \brief Run the program for the given literal ID, with the interpreter
* out of line.
*
* Assumes not in_anchored.
*/
static really_inline
hwlmcb_rv_t roseProcessMatch(const struct RoseEngine *t,
struct hs_scratch *scratch, u64a end,
size_t match_len, u32 id) {
DEBUG_PRINTF("id=%u\n", id);
const u32 *programs = getByOffset(t, t->litProgramOffset);
assert(id < t->literalCount);
const u64a som = 0;
const u8 flags = 0;
return roseRunProgram(t, scratch, programs[id], som, end, match_len, flags);
}
int roseAnchoredCallback(u64a start, u64a end, u32 id, void *ctx) {
struct hs_scratch *scratch = ctx;
assert(scratch && scratch->magic == SCRATCH_MAGIC);
struct RoseContext *tctxt = &scratch->tctxt;
struct core_info *ci = &scratch->core_info;
const struct RoseEngine *t = ci->rose;
u64a real_end = ci->buf_offset + end; // index after last byte
DEBUG_PRINTF("MATCH id=%u offsets=[???,%llu]\n", id, real_end);
DEBUG_PRINTF("STATE groups=0x%016llx\n", tctxt->groups);
if (can_stop_matching(scratch)) {
DEBUG_PRINTF("received a match when we're already dead!\n");
return MO_HALT_MATCHING;
}
const size_t match_len = 0;
/* delayed literals need to be delivered before real literals; however
* delayed literals only come from the floating table so if we are going
* to deliver a literal here it must be too early for a delayed literal */
/* no history checks from anchored region and we are before the flush
* boundary */
if (real_end <= t->floatingMinLiteralMatchOffset) {
roseFlushLastByteHistory(t, scratch, real_end);
tctxt->lastEndOffset = real_end;
}
const u32 *programs = getByOffset(t, t->litProgramOffset);
assert(id < t->literalCount);
const u8 flags = ROSE_PROG_FLAG_IN_ANCHORED;
if (roseRunProgram(t, scratch, programs[id], start, real_end, match_len,
flags) == HWLM_TERMINATE_MATCHING) {
assert(can_stop_matching(scratch));
DEBUG_PRINTF("caller requested termination\n");
return MO_HALT_MATCHING;
}
DEBUG_PRINTF("DONE groups=0x%016llx\n", tctxt->groups);
if (real_end > t->floatingMinLiteralMatchOffset) {
recordAnchoredLiteralMatch(t, scratch, id, real_end);
}
return MO_CONTINUE_MATCHING;
}
/**
* \brief Match callback adaptor used for matches from pure-literal cases.
*
* Literal match IDs in this path run limited Rose programs that do not use
* Rose state (which is not initialised in the pure-literal path). They can
* still, for example, check lookarounds or literal masks.
*/
hwlmcb_rv_t rosePureLiteralCallback(size_t start, size_t end, u32 id,
void *context) {
DEBUG_PRINTF("start=%zu, end=%zu, id=%u\n", start, end, id);
struct hs_scratch *scratch = context;
struct core_info *ci = &scratch->core_info;
const u64a real_end = (u64a)end + ci->buf_offset + 1;
const u64a som = 0;
const size_t match_len = end - start + 1;
const struct RoseEngine *rose = ci->rose;
const u32 *programs = getByOffset(rose, rose->litProgramOffset);
assert(id < rose->literalCount);
const u8 flags = 0;
return roseRunProgram(rose, scratch, programs[id], som, real_end, match_len,
flags);
}
static really_inline
void do_rebuild(const struct RoseEngine *t, struct hs_scratch *scratch) {
assert(t->drmatcherOffset);
assert(!can_stop_matching(scratch));
const struct HWLM *hwlm = getByOffset(t, t->drmatcherOffset);
size_t len = MIN(scratch->core_info.hlen, t->delayRebuildLength);
const u8 *buf = scratch->core_info.hbuf + scratch->core_info.hlen - len;
DEBUG_PRINTF("BEGIN FLOATING REBUILD over %zu bytes\n", len);
scratch->core_info.status &= ~STATUS_DELAY_DIRTY;
hwlmExec(hwlm, buf, len, 0, roseDelayRebuildCallback, scratch,
scratch->tctxt.groups);
assert(!can_stop_matching(scratch));
}
hwlmcb_rv_t roseDelayRebuildCallback(size_t start, size_t end, u32 id,
void *ctx) {
struct hs_scratch *scratch = ctx;
struct RoseContext *tctx = &scratch->tctxt;
struct core_info *ci = &scratch->core_info;
const struct RoseEngine *t = ci->rose;
size_t rb_len = MIN(ci->hlen, t->delayRebuildLength);
u64a real_end = ci->buf_offset - rb_len + end + 1; // index after last byte
#ifdef DEBUG
DEBUG_PRINTF("REBUILD MATCH id=%u offsets=[%llu,%llu]: ", id,
start + ci->buf_offset - rb_len, real_end);
printMatch(ci, start + ci->buf_offset - rb_len, real_end);
printf("\n");
#endif
DEBUG_PRINTF("STATE groups=0x%016llx\n", tctx->groups);
const u32 *delayRebuildPrograms =
getByOffset(t, t->litDelayRebuildProgramOffset);
assert(id < t->literalCount);
const u32 program = delayRebuildPrograms[id];
if (program) {
const u64a som = 0;
const size_t match_len = end - start + 1;
const u8 flags = 0;
UNUSED hwlmcb_rv_t rv = roseRunProgram(t, scratch, program, som,
real_end, match_len, flags);
assert(rv != HWLM_TERMINATE_MATCHING);
}
/* we are just repopulating the delay queue, groups should be
* already set from the original scan. */
return tctx->groups;
}
static rose_inline
void runEagerPrefixesStream(const struct RoseEngine *t,
struct hs_scratch *scratch) {
if (!t->eagerIterOffset
|| scratch->core_info.buf_offset >= EAGER_STOP_OFFSET) {
return;
}
char *state = scratch->core_info.state;
u8 *ara = getActiveLeftArray(t, state); /* indexed by offsets into
* left_table */
const u32 arCount = t->activeLeftCount;
const u32 qCount = t->queueCount;
const struct LeftNfaInfo *left_table = getLeftTable(t);
const struct mmbit_sparse_iter *it = getByOffset(t, t->eagerIterOffset);
struct mmbit_sparse_state si_state[MAX_SPARSE_ITER_STATES];
u32 idx = 0;
u32 ri = mmbit_sparse_iter_begin(ara, arCount, &idx, it, si_state);
for (; ri != MMB_INVALID;
ri = mmbit_sparse_iter_next(ara, arCount, ri, &idx, it, si_state)) {
const struct LeftNfaInfo *left = left_table + ri;
u32 qi = ri + t->leftfixBeginQueue;
DEBUG_PRINTF("leftfix %u of %u, maxLag=%u\n", ri, arCount, left->maxLag);
assert(!fatbit_isset(scratch->aqa, qCount, qi));
assert(left->eager);
assert(!left->infix);
struct mq *q = scratch->queues + qi;
const struct NFA *nfa = getNfaByQueue(t, qi);
s64a loc = MIN(scratch->core_info.len,
EAGER_STOP_OFFSET - scratch->core_info.buf_offset);
fatbit_set(scratch->aqa, qCount, qi);
initRoseQueue(t, qi, left, scratch);
if (scratch->core_info.buf_offset) {
s64a sp = left->transient ? -(s64a)scratch->core_info.hlen
: -(s64a)loadRoseDelay(t, state, left);
pushQueueAt(q, 0, MQE_START, sp);
if (scratch->core_info.buf_offset + sp > 0) {
loadStreamState(nfa, q, sp);
/* if the leftfix fix is currently in a match state, we cannot
* advance it. */
if (nfaInAnyAcceptState(nfa, q)) {
continue;
}
pushQueueAt(q, 1, MQE_END, loc);
} else {
pushQueueAt(q, 1, MQE_TOP, sp);
pushQueueAt(q, 2, MQE_END, loc);
nfaQueueInitState(q->nfa, q);
}
} else {
pushQueueAt(q, 0, MQE_START, 0);
pushQueueAt(q, 1, MQE_TOP, 0);
pushQueueAt(q, 2, MQE_END, loc);
nfaQueueInitState(nfa, q);
}
char alive = nfaQueueExecToMatch(q->nfa, q, loc);
if (!alive) {
DEBUG_PRINTF("queue %u dead, squashing\n", qi);
mmbit_unset(ara, arCount, ri);
fatbit_unset(scratch->aqa, qCount, qi);
scratch->tctxt.groups &= left->squash_mask;
} else if (q->cur == q->end) {
assert(alive != MO_MATCHES_PENDING);
/* unlike in block mode we cannot squash groups if there is no match
* in this block as we need the groups on for later stream writes */
/* TODO: investigate possibility of a method to suppress groups for
* a single stream block. */
DEBUG_PRINTF("queue %u finished, nfa lives\n", qi);
q->cur = q->end = 0;
pushQueueAt(q, 0, MQE_START, loc);
} else {
assert(alive == MO_MATCHES_PENDING);
DEBUG_PRINTF("queue %u unfinished, nfa lives\n", qi);
q->end--; /* remove end item */
}
}
}
static rose_inline
int roseEodRunIterator(const struct RoseEngine *t, u8 *state, u64a offset,
struct hs_scratch *scratch) {
if (!t->eodIterOffset) {
return MO_CONTINUE_MATCHING;
}
const struct RoseRole *roleTable = getRoleTable(t);
const struct RosePred *predTable = getPredTable(t);
const struct RoseIterMapping *iterMapBase
= getByOffset(t, t->eodIterMapOffset);
const struct mmbit_sparse_iter *it = getByOffset(t, t->eodIterOffset);
assert(ISALIGNED(iterMapBase));
assert(ISALIGNED(it));
// Sparse iterator state was allocated earlier
struct mmbit_sparse_state *s = scratch->sparse_iter_state;
struct fatbit *handled_roles = scratch->handled_roles;
const u32 numStates = t->rolesWithStateCount;
void *role_state = getRoleState(state);
u32 idx = 0;
u32 i = mmbit_sparse_iter_begin(role_state, numStates, &idx, it, s);
fatbit_clear(handled_roles);
for (; i != MMB_INVALID;
i = mmbit_sparse_iter_next(role_state, numStates, i, &idx, it, s)) {
DEBUG_PRINTF("pred state %u (iter idx=%u) is on\n", i, idx);
const struct RoseIterMapping *iterMap = iterMapBase + idx;
const struct RoseIterRole *roles = getByOffset(t, iterMap->offset);
assert(ISALIGNED(roles));
DEBUG_PRINTF("%u roles to consider\n", iterMap->count);
for (u32 j = 0; j != iterMap->count; j++) {
u32 role = roles[j].role;
assert(role < t->roleCount);
DEBUG_PRINTF("checking role %u, pred %u:\n", role, roles[j].pred);
const struct RoseRole *tr = roleTable + role;
if (fatbit_isset(handled_roles, t->roleCount, role)) {
DEBUG_PRINTF("role %u already handled by the walk, skip\n",
role);
continue;
}
// Special case: if this role is a trivial case (pred type simple)
// we don't need to check any history and we already know the pred
// role is on.
if (tr->flags & ROSE_ROLE_PRED_SIMPLE) {
DEBUG_PRINTF("pred type is simple, no need for checks\n");
} else {
assert(roles[j].pred < t->predCount);
const struct RosePred *tp = predTable + roles[j].pred;
if (!roseCheckPredHistory(tp, offset)) {
continue;
}
}
/* mark role as handled so we don't touch it again in this walk */
fatbit_set(handled_roles, t->roleCount, role);
DEBUG_PRINTF("fire report for role %u, report=%u\n", role,
tr->reportId);
int rv = scratch->tctxt.cb(offset, tr->reportId,
scratch->tctxt.userCtx);
if (rv == MO_HALT_MATCHING) {
return MO_HALT_MATCHING;
}
}
}
return MO_CONTINUE_MATCHING;
}
