Drop Malformed 3MF Paint Data Instead of Reading Past the Bitstream

Painted facets are decoded from a bitstream a nibble at a time with no bound
check, so a truncated or corrupt paint string in a 3MF (for example split
codes with no children behind them) read past the end and crashed on load and
slice. A one- or two-side split naming side 3 also indexed past the triangle's
vertices.

Every nibble read now goes through a bounds-checked reader. Loading validates
each triangle's tree and drops a malformed one with a warning, so the stored
data, used extruder states and later decoding all agree. deserialize() also
unwinds and clears any triangle whose tree is incomplete or malformed, and
has_facets() stops at a truncated triangle. Valid streams decode unchanged.

src/libslic3r/Format/bbs_3mf.cpp

@@ -1629,7 +1629,7 @@ void PlateData::parse_filament_info(GCodeProcessorResult *result)
         }
         while (it != m_plater_data.end())
         {
-            if (it->first > m_plater_data.size())
+            if (it->first <= 0 || static_cast<size_t>(it->first) > m_plater_data.size())
             {
                 add_error("invalid plate index");
                 return false;
@@ -2312,7 +2312,7 @@ void PlateData::parse_filament_info(GCodeProcessorResult *result)
         }
         while (it != m_plater_data.end())
         {
-            if (it->first > m_plater_data.size())
+            if (it->first <= 0 || static_cast<size_t>(it->first) > m_plater_data.size())
             {
                 add_error("invalid plate index");
                 return false;

src/libslic3r/Model.cpp

@@ -3702,7 +3702,11 @@ void FacetsAnnotation::set_triangle_from_string(int triangle_id, const std::stri
             m_data.bitstream.insert(m_data.bitstream.end(), bool(dec & (1 << i)));
     }
 
-    m_data.update_used_states(bitstream_start_idx);
+    if (!m_data.update_used_states(bitstream_start_idx)) {
+        BOOST_LOG_TRIVIAL(warning) << __FUNCTION__ << ": dropping malformed paint data of triangle " << triangle_id;
+        m_data.bitstream.resize(bitstream_start_idx);
+        m_data.triangles_to_split.pop_back();
+    }
 }
 
 bool FacetsAnnotation::equals(const FacetsAnnotation &other) const

src/libslic3r/TriangleSelector.cpp

@@ -1778,6 +1778,13 @@ TriangleSelector::TriangleSplittingData TriangleSelector::serialize() const {
     return out.data;
 }
 
+// A split code keeps the split side (one split) or the kept side (two splits) in its upper two
+// bits, where 3 is not a side. The value is ignored for a three-side split.
+static bool split_code_valid(int code)
+{
+    return (code & 0b11) == 3 || (code >> 2) != 3;
+}
+
 void TriangleSelector::deserialize(const TriangleSplittingData &data,
                                    bool                         needs_reset,
                                    EnforcerBlockerType          max_ebt,
@@ -1812,11 +1819,12 @@ void TriangleSelector::deserialize(const TriangleSplittingData &data,
 
     for (auto [triangle_id, ibit] : data.triangles_to_split) {
         assert(triangle_id < int(m_triangles.size()));
-        assert(ibit < int(data.bitstream.size()));
-        auto next_nibble = [&data, &ibit = ibit]() {
+        // Set when the bitstream runs out or holds an impossible split before this triangle's tree is complete.
+        bool corrupt = false;
+        auto next_nibble = [&data, &ibit = ibit, &corrupt]() {
             int n = 0;
-            for (int i = 0; i < 4; ++ i)
-                n |= data.bitstream[ibit ++] << i;
+            if (! data.read_nibble(ibit, n))
+                corrupt = true;
             return n;
         };
         // Decode a leaf state stored behind the "11" prefix: one nibble of (state-3) for states
@@ -1835,6 +1843,10 @@ void TriangleSelector::deserialize(const TriangleSplittingData &data,
             bool is_split = num_of_children != 0;
             // Only valid if not is_split.
             auto state = is_split ? EnforcerBlockerType::NONE : ((code & 0b1100) == 0b1100 ? decode_leaf_state() : EnforcerBlockerType(code >> 2));
+            if (is_split && ! split_code_valid(code))
+                corrupt = true;
+            if (corrupt)
+                break;
 
             // BBS
             if (state == to_delete_filament)
@@ -1849,7 +1861,7 @@ void TriangleSelector::deserialize(const TriangleSplittingData &data,
             }
 
             // Only valid if is_split.
-            int special_side = code >> 2;
+            int special_side = num_of_split_sides == 3 ? 0 : code >> 2;
 
             // Take care of the first iteration separately, so handling of the others is simpler.
             if (parents.empty()) {
@@ -1904,47 +1916,55 @@ void TriangleSelector::deserialize(const TriangleSplittingData &data,
             if (parents.empty())
                 break;
         }
+
+        if (corrupt) {
+            // Every split above allocated all of its children, so the partial tree unwinds cleanly.
+            BOOST_LOG_TRIVIAL(warning) << __FUNCTION__ << ": malformed paint data, dropping paint of triangle " << triangle_id;
+            undivide_triangle(triangle_id);
+            m_triangles[triangle_id].set_state(EnforcerBlockerType::NONE);
+        }
     }
 }
 
-void TriangleSelector::TriangleSplittingData::update_used_states(const size_t bitstream_start_idx) {
-    assert(bitstream_start_idx < this->bitstream.size());
-    assert(!this->bitstream.empty() && this->bitstream.size() != bitstream_start_idx);
-    assert((this->bitstream.size() - bitstream_start_idx) % 4 == 0);
+bool TriangleSelector::TriangleSplittingData::update_used_states(const size_t bitstream_start_idx) {
+    int      ibit   = static_cast<int>(bitstream_start_idx);
+    uint64_t states = 0;
+    do {
+        // Walk one triangle's tree depth-first, counting the nodes still to be read; a split node adds its children.
+        for (int pending_nodes = 1; pending_nodes > 0; --pending_nodes) {
+            int code;
+            if (!this->read_nibble(ibit, code))
+                return false;
 
-    if (this->bitstream.empty() || this->bitstream.size() == bitstream_start_idx)
-        return;
+            if (const int num_of_split_sides = code & 0b11; num_of_split_sides != 0) {
+                if (!split_code_valid(code))
+                    return false;
+                pending_nodes += num_of_split_sides + 1;
+                continue;
+            }
 
-    size_t nibble_idx = bitstream_start_idx;
-
-    auto read_next_nibble = [&data_bitstream = std::as_const(this->bitstream), &nibble_idx]() -> uint8_t {
-        assert(nibble_idx + 3 < data_bitstream.size());
-        uint8_t code = 0;
-        for (size_t bit_idx = 0; bit_idx < 4; ++bit_idx)
-            code |= data_bitstream[nibble_idx++] << bit_idx;
-        return code;
-    };
-
-    while (nibble_idx < this->bitstream.size()) {
-        const uint8_t code = read_next_nibble();
-
-        if (const bool is_split = (code & 0b11) != 0; is_split)
-            continue;
-
-        uint8_t facet_state;
-        if ((code & 0b1100) == 0b1100) {
-            // Leaf behind the "11" prefix: one nibble of (state-3), or 0b1111 + (state-18).
-            const uint8_t nibble = read_next_nibble();
-            facet_state = nibble == 0b1111 ? uint8_t(read_next_nibble() + 18) : uint8_t(nibble + 3);
-        } else {
-            facet_state = code >> 2;
+            int facet_state = code >> 2;
+            if (facet_state == 0b11) {
+                // Leaf behind the "11" prefix: one nibble of (state-3), or 0b1111 + (state-18).
+                int nibble;
+                if (!this->read_nibble(ibit, nibble))
+                    return false;
+                facet_state = nibble + 3;
+                if (nibble == 0b1111) {
+                    if (!this->read_nibble(ibit, nibble))
+                        return false;
+                    facet_state = nibble + 18;
+                }
+            }
+            states |= uint64_t(1) << facet_state;
         }
-        assert(facet_state < this->used_states.size());
-        if (facet_state >= this->used_states.size())
-            continue;
+    } while (static_cast<size_t>(ibit) < this->bitstream.size());
 
-        this->used_states[facet_state] = true;
-    }
+    // The leaf encoding tops out at state 33, so every state fits the 64-bit mask.
+    for (size_t state_idx = 0; state_idx < std::min<size_t>(this->used_states.size(), 64); ++state_idx)
+        if (states & (uint64_t(1) << state_idx))
+            this->used_states[state_idx] = true;
+    return true;
 }
 
 // Lightweight variant of deserialization, which only tests whether a face of test_state exists.
@@ -1956,11 +1976,12 @@ bool TriangleSelector::has_facets(const TriangleSplittingData &data, const Enfor
 
     for (const TriangleBitStreamMapping &triangle_id_and_ibit : data.triangles_to_split) {
         int ibit = triangle_id_and_ibit.bitstream_start_idx;
-        assert(ibit < int(data.bitstream.size()));
-        auto next_nibble = [&data, &ibit = ibit]() {
+        // Stop reading a triangle whose stream is truncated.
+        bool truncated = false;
+        auto next_nibble = [&data, &ibit = ibit, &truncated]() {
             int n = 0;
-            for (int i = 0; i < 4; ++ i)
-                n |= data.bitstream[ibit ++] << i;
+            if (! data.read_nibble(ibit, n))
+                truncated = true;
             return n;
         };
         // < 0 -> negative of a number of children
@@ -1978,6 +1999,8 @@ bool TriangleSelector::has_facets(const TriangleSplittingData &data, const Enfor
         };
 
         int state = num_children_or_state();
+        if (truncated)
+            continue;
         if (state < 0) {
             // Root is split.
             parents_children.clear();
@@ -1985,6 +2008,8 @@ bool TriangleSelector::has_facets(const TriangleSplittingData &data, const Enfor
             do {
                 if (-- parents_children.back() >= 0) {
                     int state = num_children_or_state();
+                    if (truncated)
+                        break;
                     if (state < 0)
                         // Child is split.
                         parents_children.emplace_back(- state);

src/libslic3r/TriangleSelector.hpp

@@ -297,8 +297,20 @@ public:
             std::fill(used_states.begin(), used_states.end(), false);
         }
 
-        // Update used states based on the bitstream. It just iterated over the bitstream from the bitstream_start_idx till the end.
-        void update_used_states(size_t bitstream_start_idx);
+        // Update used states from the triangle trees stored between bitstream_start_idx and the end of the bitstream.
+        // Returns false and leaves used states untouched if a tree is truncated or malformed.
+        bool update_used_states(size_t bitstream_start_idx);
+
+        // Read the 4-bit code at bit index ibit (LSB first) and advance ibit past it.
+        // Returns false without advancing when fewer than 4 bits remain.
+        bool read_nibble(int &ibit, int &nibble) const {
+            if (ibit < 0 || static_cast<size_t>(ibit) + 4 > bitstream.size())
+                return false;
+            nibble = 0;
+            for (int i = 0; i < 4; ++i)
+                nibble |= static_cast<int>(bitstream[ibit++]) << i;
+            return true;
+        }
 
     private:
         friend class cereal::access;

tests/libslic3r/test_3mf.cpp

@@ -19,6 +19,7 @@
 #include <boost/filesystem/operations.hpp>
 #include <boost/algorithm/string/predicate.hpp>
 #include <algorithm>
+#include <functional>
 
 #include <catch2/catch_tostring.hpp>
 #include <Eigen/Core>
@@ -184,16 +185,13 @@ static bool read_cad_recipe_entry(const std::string& path, std::string& out,
     return found;
 }
 
-// Rewrites the archive at `path` with the recipe entry back under the name it had before the
-// rename, which is what every project saved by an earlier build looks like on disk. Generated
-// rather than checked in because a whole project archive is not frozen evidence the way a bare
-// recipe blob is -- it has to be whatever today's exporter writes, with only the name aged.
-// miniz cannot rename in place and open_zip_writer truncates, so the entries are held across
-// the switch.
-static void rename_cad_recipe_entry_to_legacy(const std::string& path)
+// Rewrites the archive at `path`, letting `edit` change the name or data of each entry; returns
+// whether `edit` reported a change for any of them. miniz cannot edit in place and
+// open_zip_writer truncates, so the entries are held across the switch.
+static bool rewrite_3mf_entries(const std::string& path, const std::function<bool(std::string& name, std::string& data)>& edit)
 {
     std::vector<std::pair<std::string, std::string>> entries;
-    bool renamed = false;
+    bool changed = false;
     {
         mz_zip_archive zip;
         mz_zip_zero_struct(&zip);
@@ -208,22 +206,140 @@ static void rename_cad_recipe_entry_to_legacy(const std::string& path)
             std::string data((size_t) st.m_uncomp_size, '\0');
             if (st.m_uncomp_size > 0)
                 REQUIRE(mz_zip_reader_extract_to_mem(&zip, i, data.data(), data.size(), 0));
-            if (boost::algorithm::iequals(name, CAD_RECIPE_ENTRY)) {
-                name    = LEGACY_CAD_RECIPE_ENTRY;
-                renamed = true;
-            }
+            changed |= edit(name, data);
             entries.emplace_back(std::move(name), std::move(data));
         }
         close_zip_reader(&zip);
     }
-    // Without this the scenario would degrade silently into re-testing the new name if the
-    // exporter's constant ever moved again: every load below would still pass.
-    REQUIRE(renamed);
 
     Zipper out(path);
     for (const auto& e : entries)
         out.add_entry(e.first, e.second.data(), e.second.size());
     out.finalize();
+    return changed;
+}
+
+// Rewrites the archive at `path` with the recipe entry back under the name it had before the
+// rename, which is what every project saved by an earlier build looks like on disk. Generated
+// rather than checked in because a whole project archive is not frozen evidence the way a bare
+// recipe blob is -- it has to be whatever today's exporter writes, with only the name aged.
+static void rename_cad_recipe_entry_to_legacy(const std::string& path)
+{
+    const bool renamed = rewrite_3mf_entries(path, [](std::string& name, std::string&) {
+        if (!boost::algorithm::iequals(name, CAD_RECIPE_ENTRY))
+            return false;
+        name = LEGACY_CAD_RECIPE_ENTRY;
+        return true;
+    });
+    // Without this the scenario would degrade silently into re-testing the new name if the
+    // exporter's constant ever moved again: every load below would still pass.
+    REQUIRE(renamed);
+}
+
+// Replaces the first occurrence of `from` in any entry whose name ends with `suffix`.
+static bool replace_in_3mf_entry(const std::string& path, const std::string& suffix, const std::string& from, const std::string& to)
+{
+    bool replaced = false;
+    rewrite_3mf_entries(path, [&](std::string& name, std::string& data) {
+        if (replaced || !boost::algorithm::ends_with(name, suffix))
+            return false;
+        if (const size_t pos = data.find(from); pos != std::string::npos) {
+            data.replace(pos, from.size(), to);
+            replaced = true;
+        }
+        return replaced;
+    });
+    return replaced;
+}
+
+// Stores a one-plate project holding a cube whose first two facets are painted Extruder2 and
+// Extruder3, which the exporter writes as paint_color="8" and paint_color="0C".
+static void store_painted_cube(const std::string& path)
+{
+    Model        model;
+    ModelObject* object = model.add_object();
+    ModelVolume* volume = object->add_volume(make_cube(10., 10., 10.));
+    object->add_instance();
+    {
+        TriangleSelector selector(volume->mesh());
+        selector.set_facet(0, EnforcerBlockerType::Extruder2);
+        selector.set_facet(1, EnforcerBlockerType::Extruder3);
+        REQUIRE(volume->mmu_segmentation_facets.set(selector));
+    }
+    ScopedTemporaryDir backup_dir("orca_paint_src");
+    model.set_backup_path(backup_dir.string());
+
+    DynamicPrintConfig cfg;
+    PlateData          plate;
+    plate.plate_index = 0;
+    StoreParams sp;
+    sp.path     = path.c_str();
+    sp.model    = &model;
+    sp.config   = &cfg;
+    sp.strategy = SaveStrategy::Zip64 | SaveStrategy::Silence;
+    sp.plate_data_list.push_back(&plate);
+    REQUIRE(store_bbs_3mf(sp));
+}
+
+// Loads `path` through the BBS importer into `model`, releasing the plates it returns. The
+// importer stages metadata through `backup_dir`, which has to outlive the model.
+static bool load_project(const std::string& path, Model& model, const ScopedTemporaryDir& backup_dir)
+{
+    model.set_backup_path(backup_dir.string());
+    DynamicPrintConfig        config;
+    ConfigSubstitutionContext ctxt{ ForwardCompatibilitySubstitutionRule::Enable };
+    PlateDataPtrs             plates;
+    std::vector<Preset*>      project_presets;
+    bool   is_bbl_3mf = false, is_orca_3mf = false;
+    Semver file_version;
+    const bool loaded = load_bbs_3mf(path.c_str(), &config, &ctxt, &model, &plates, &project_presets, &is_bbl_3mf,
+                                     &is_orca_3mf, &file_version, nullptr, LoadStrategy::LoadModel | LoadStrategy::LoadConfig);
+    release_PlateData_list(plates);
+    return loaded;
+}
+
+TEST_CASE("A project with a plate id below 1 fails to load", "[3mf][Regression]")
+{
+    const int plate_id = GENERATE(0, -1);
+    INFO("plater_id " << plate_id);
+
+    ScopedTemporaryFile temp(".3mf");
+    store_painted_cube(temp.string());
+    {
+        ScopedTemporaryDir backup_dir("orca_plate_dst");
+        Model              model;
+        REQUIRE(load_project(temp.string(), model, backup_dir));
+    }
+
+    REQUIRE(replace_in_3mf_entry(temp.string(), "model_settings.config", "key=\"plater_id\" value=\"1\"",
+                                 "key=\"plater_id\" value=\"" + std::to_string(plate_id) + "\""));
+    ScopedTemporaryDir backup_dir("orca_plate_dst");
+    Model              model;
+    bool               loaded = true;
+    REQUIRE_NOTHROW(loaded = load_project(temp.string(), model, backup_dir));
+    REQUIRE_FALSE(loaded);
+}
+
+TEST_CASE("A project with malformed paint data loads without the damaged facet", "[3mf][Regression]")
+{
+    ScopedTemporaryFile temp(".3mf");
+    store_painted_cube(temp.string());
+    // Split codes with no children behind them: the stream runs out mid-tree.
+    REQUIRE(replace_in_3mf_entry(temp.string(), ".model", "paint_color=\"8\"", "paint_color=\"FFFFFFFFFFFFFFFF3\""));
+
+    ScopedTemporaryDir backup_dir("orca_paint_dst");
+    Model              model;
+    REQUIRE(load_project(temp.string(), model, backup_dir));
+    REQUIRE(model.objects.size() == 1);
+    const ModelVolume& volume = *model.objects.front()->volumes.front();
+    const auto&        data   = volume.mmu_segmentation_facets.get_data();
+    REQUIRE_FALSE(data.used_states[size_t(EnforcerBlockerType::Extruder2)]);
+    REQUIRE(data.used_states[size_t(EnforcerBlockerType::Extruder3)]);
+
+    TriangleSelector selector(volume.mesh());
+    REQUIRE_NOTHROW(selector.deserialize(data));
+    REQUIRE(selector.num_facets(EnforcerBlockerType::Extruder2) == 0);
+    REQUIRE(selector.num_facets(EnforcerBlockerType::Extruder3) == 1);
 }
 
 // The recipe lives only in the BBS-native backend, because that is the only one that runs:

tests/libslic3r/test_triangle_selector.cpp

@@ -3,6 +3,8 @@
 #include "libslic3r/TriangleSelector.hpp"
 #include "libslic3r/TriangleMesh.hpp"
 
+#include <algorithm>
+
 using namespace Slic3r;
 
 // A sphere gives well over ExtruderMax original facets, so every extruder state can be assigned
@@ -123,3 +125,77 @@ TEST_CASE("Extruder states match the CONST_FILAMENTS hex encoding", "[TriangleSe
     INFO("Hex " << c.hex << " -> extruder " << c.state);
     REQUIRE(TriangleSelector::has_facets(data, EnforcerBlockerType(c.state)));
 }
+
+// Pack 4-bit codes into a bitstream, least significant bit first, in the order the decoder reads them.
+static std::vector<bool> pack_nibbles(const std::vector<int> &nibbles)
+{
+    std::vector<bool> bitstream;
+    for (const int nibble : nibbles)
+        for (int bit = 0; bit < 4; ++bit)
+            bitstream.push_back((nibble >> bit) & 1);
+    return bitstream;
+}
+
+TEST_CASE("A valid paint stream with nested splits round-trips bit for bit", "[TriangleSelector]")
+{
+    const TriangleMesh mesh = test_mesh();
+
+    TriangleSelector::TriangleSplittingData data;
+    data.triangles_to_split.emplace_back(0, 0);
+    // A three-side split whose children, in stream order, are: a one-side split (side 2) into two
+    // leaves, a two-side split (side 1) into leaves of states 20, 0 and 8, then two plain leaves.
+    const std::vector<int> triangle_0 = {0b0011,
+                                         0b1001, 0b1000, 0b0100,
+                                         0b0110, 0b1100, 0b1111, 20 - 18, 0b0000, 0b1100, 8 - 3,
+                                         0b1000,
+                                         0b0100};
+    data.bitstream = pack_nibbles(triangle_0);
+    data.triangles_to_split.emplace_back(5, int(data.bitstream.size()));
+    const std::vector<bool> triangle_5 = pack_nibbles({0b1100, 3 - 3});
+    data.bitstream.insert(data.bitstream.end(), triangle_5.begin(), triangle_5.end());
+    data.reset_used_states();
+    REQUIRE(data.update_used_states(0));
+
+    TriangleSelector restored(mesh);
+    restored.deserialize(data);
+
+    REQUIRE(restored.num_facets(EnforcerBlockerType::Extruder20) == 1);
+    REQUIRE(restored.num_facets(EnforcerBlockerType::Extruder3) == 1);
+    REQUIRE(restored.serialize() == data);
+}
+
+TEST_CASE("A truncated or malformed paint stream drops only the damaged triangle", "[TriangleSelector][Regression]")
+{
+    struct Case { const char *name; std::vector<int> nibbles; };
+    const auto c = GENERATE(values<Case>({
+        {"three-side split missing two children", {0b0011, 0b1000, 0b1000}},
+        {"leaf missing its state nibble",         {0b1100}},
+        {"leaf missing its second state nibble",  {0b1100, 0b1111}},
+        {"splits nested past the end",            {0b0011, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF,
+                                                    0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF}},
+        {"one-side split of the nonexistent side 3", {0b1101, 0b1000, 0b1000}},
+    }));
+    INFO(c.name);
+
+    const TriangleMesh mesh = test_mesh();
+    TriangleSelector   intact(mesh);
+    intact.set_facet(0, EnforcerBlockerType::Extruder2);
+
+    // Triangle 0 stays intact, triangle 1 carries the damaged stream.
+    TriangleSelector::TriangleSplittingData data = intact.serialize();
+    data.triangles_to_split.emplace_back(1, int(data.bitstream.size()));
+    const std::vector<bool> damaged = pack_nibbles(c.nibbles);
+    data.bitstream.insert(data.bitstream.end(), damaged.begin(), damaged.end());
+
+    TriangleSelector restored(mesh);
+    REQUIRE_NOTHROW(restored.deserialize(data));
+    // Triangle 1 unwinds completely, so the selector holds exactly the intact paint.
+    REQUIRE(restored.serialize() == intact.serialize());
+
+    REQUIRE_NOTHROW(TriangleSelector::has_facets(data, EnforcerBlockerType::Extruder3));
+
+    TriangleSelector::TriangleSplittingData recomputed = data;
+    recomputed.reset_used_states();
+    REQUIRE_FALSE(recomputed.update_used_states(0));
+    REQUIRE(std::none_of(recomputed.used_states.begin(), recomputed.used_states.end(), [](bool used) { return used; }));
+}