core/surface/buffer: Buffer lock/release fixes (#7110)

src/desktop/LayerSurface.cpp

@@ -242,7 +242,7 @@ void CLayerSurface::onCommit() {
 
     if (!mapped) {
         // we're re-mapping if this is the case
-        if (layerSurface->surface && !layerSurface->surface->current.buffer) {
+        if (layerSurface->surface && !layerSurface->surface->current.texture) {
             fadingOut = false;
             geometry  = {};
             g_pHyprRenderer->arrangeLayersForMonitor(monitorID);

src/desktop/WLSurface.cpp

@@ -57,12 +57,12 @@ bool CWLSurface::small() const {
     if (!validMapped(m_pWindowOwner) || !exists())
         return false;
 
-    if (!m_pResource->current.buffer)
+    if (!m_pResource->current.texture)
         return false;
 
     const auto O = m_pWindowOwner.lock();
 
-    return O->m_vReportedSize.x > m_pResource->current.buffer->size.x + 1 || O->m_vReportedSize.y > m_pResource->current.buffer->size.y + 1;
+    return O->m_vReportedSize.x > m_pResource->current.bufferSize.x + 1 || O->m_vReportedSize.y > m_pResource->current.bufferSize.y + 1;
 }
 
 Vector2D CWLSurface::correctSmallVec() const {
@@ -76,37 +76,36 @@ Vector2D CWLSurface::correctSmallVec() const {
 }
 
 Vector2D CWLSurface::correctSmallVecBuf() const {
-    if (!exists() || !small() || m_bFillIgnoreSmall || !m_pResource->current.buffer)
+    if (!exists() || !small() || m_bFillIgnoreSmall || !m_pResource->current.texture)
         return {};
 
     const auto SIZE = getViewporterCorrectedSize();
-    const auto BS   = m_pResource->current.buffer->size;
+    const auto BS   = m_pResource->current.bufferSize;
 
     return Vector2D{(BS.x - SIZE.x) / 2, (BS.y - SIZE.y) / 2}.clamp({}, {INFINITY, INFINITY});
 }
 
 Vector2D CWLSurface::getViewporterCorrectedSize() const {
-    if (!exists() || !m_pResource->current.buffer)
+    if (!exists() || !m_pResource->current.texture)
         return {};
 
-    return m_pResource->current.viewport.hasDestination ? m_pResource->current.viewport.destination : m_pResource->current.buffer->size;
+    return m_pResource->current.viewport.hasDestination ? m_pResource->current.viewport.destination : m_pResource->current.bufferSize;
 }
 
 CRegion CWLSurface::computeDamage() const {
-    if (!m_pResource->current.buffer)
+    if (!m_pResource->current.texture)
         return {};
 
     CRegion damage = m_pResource->accumulateCurrentBufferDamage();
-    damage.transform(wlTransformToHyprutils(m_pResource->current.transform), m_pResource->current.buffer->size.x, m_pResource->current.buffer->size.y);
+    damage.transform(wlTransformToHyprutils(m_pResource->current.transform), m_pResource->current.bufferSize.x, m_pResource->current.bufferSize.y);
 
-    const auto BUFSIZE    = m_pResource->current.buffer->size;
+    const auto BUFSIZE    = m_pResource->current.bufferSize;
     const auto CORRECTVEC = correctSmallVecBuf();
 
     if (m_pResource->current.viewport.hasSource)
         damage.intersect(m_pResource->current.viewport.source);
 
-    const auto SCALEDSRCSIZE =
+    const auto SCALEDSRCSIZE = m_pResource->current.viewport.hasSource ? m_pResource->current.viewport.source.size() * m_pResource->current.scale : m_pResource->current.bufferSize;
-        m_pResource->current.viewport.hasSource ? m_pResource->current.viewport.source.size() * m_pResource->current.scale : m_pResource->current.buffer->size;
 
     damage.scale({BUFSIZE.x / SCALEDSRCSIZE.x, BUFSIZE.y / SCALEDSRCSIZE.y});
     damage.translate(CORRECTVEC);
@@ -114,7 +113,7 @@ CRegion CWLSurface::computeDamage() const {
     // go from buffer coords in the damage to hl logical
 
     const auto     BOX   = getSurfaceBoxGlobal();
-    const Vector2D SCALE = BOX.has_value() ? BOX->size() / m_pResource->current.buffer->size :
+    const Vector2D SCALE = BOX.has_value() ? BOX->size() / m_pResource->current.bufferSize :
                                              Vector2D{1.0 / m_pResource->current.scale, 1.0 / m_pResource->current.scale /* Wrong... but we can't really do better */};
 
     damage.scale(SCALE);

src/desktop/Window.cpp

@@ -1097,18 +1097,18 @@ bool CWindow::opaque() {
     if (PWORKSPACE->m_fAlpha.value() != 1.f)
         return false;
 
-    if (m_bIsX11 && m_pXWaylandSurface && m_pXWaylandSurface->surface && m_pXWaylandSurface->surface->current.buffer)
+    if (m_bIsX11 && m_pXWaylandSurface && m_pXWaylandSurface->surface && m_pXWaylandSurface->surface->current.texture)
-        return m_pXWaylandSurface->surface->current.buffer->opaque;
+        return m_pXWaylandSurface->surface->current.texture->m_bOpaque;
 
-    if (!m_pWLSurface->resource() || !m_pWLSurface->resource()->current.buffer)
+    if (!m_pWLSurface->resource() || !m_pWLSurface->resource()->current.texture)
         return false;
 
     // TODO: this is wrong
     const auto EXTENTS = m_pXDGSurface->surface->current.opaque.getExtents();
-    if (EXTENTS.w >= m_pXDGSurface->surface->current.buffer->size.x && EXTENTS.h >= m_pXDGSurface->surface->current.buffer->size.y)
+    if (EXTENTS.w >= m_pXDGSurface->surface->current.bufferSize.x && EXTENTS.h >= m_pXDGSurface->surface->current.bufferSize.y)
         return true;
 
-    return m_pWLSurface->resource()->current.buffer->opaque;
+    return m_pWLSurface->resource()->current.texture->m_bOpaque;
 }
 
 float CWindow::rounding() {

src/events/Windows.cpp

@@ -752,7 +752,7 @@ void Events::listener_commitWindow(void* owner, void* data) {
 
     // tearing: if solitary, redraw it. This still might be a single surface window
     const auto PMONITOR = g_pCompositor->getMonitorFromID(PWINDOW->m_iMonitorID);
-    if (PMONITOR && PMONITOR->solitaryClient.lock() == PWINDOW && PWINDOW->canBeTorn() && PMONITOR->tearingState.canTear && PWINDOW->m_pWLSurface->resource()->current.buffer) {
+    if (PMONITOR && PMONITOR->solitaryClient.lock() == PWINDOW && PWINDOW->canBeTorn() && PMONITOR->tearingState.canTear && PWINDOW->m_pWLSurface->resource()->current.texture) {
         CRegion damageBox{PWINDOW->m_pWLSurface->resource()->accumulateCurrentBufferDamage()};
 
         if (!damageBox.empty()) {

src/helpers/Monitor.cpp

@@ -801,17 +801,16 @@ bool CMonitor::attemptDirectScanout() {
 
     const auto PSURFACE = g_pXWaylandManager->getWindowSurface(PCANDIDATE);
 
-    if (!PSURFACE || !PSURFACE->current.buffer || PSURFACE->current.buffer->size != vecPixelSize || PSURFACE->current.transform != transform)
+    if (!PSURFACE || !PSURFACE->current.buffer || PSURFACE->current.bufferSize != vecPixelSize || PSURFACE->current.transform != transform)
         return false;
 
     // we can't scanout shm buffers.
-    if (!PSURFACE->current.buffer->dmabuf().success)
+    if (!PSURFACE->current.buffer || !PSURFACE->current.texture || !PSURFACE->current.texture->m_pEglImage /* dmabuf */)
         return false;
 
     // FIXME: make sure the buffer actually follows the available scanout dmabuf formats
     // and comes from the appropriate device. This may implode on multi-gpu!!
-
+    output->state->setBuffer(PSURFACE->current.buffer->buffer.lock());
-    output->state->setBuffer(PSURFACE->current.buffer);
     output->state->setPresentationMode(tearingState.activelyTearing ? Aquamarine::eOutputPresentationMode::AQ_OUTPUT_PRESENTATION_IMMEDIATE :
                                                                       Aquamarine::eOutputPresentationMode::AQ_OUTPUT_PRESENTATION_VSYNC);
 

src/managers/PointerManager.cpp

@@ -4,6 +4,7 @@
 #include "../protocols/PointerGestures.hpp"
 #include "../protocols/FractionalScale.hpp"
 #include "../protocols/core/Compositor.hpp"
+#include "../protocols/core/Seat.hpp"
 #include "eventLoop/EventLoopManager.hpp"
 #include "SeatManager.hpp"
 #include <cstring>
@@ -156,15 +157,15 @@ void CPointerManager::setCursorSurface(SP<CWLSurface> surf, const Vector2D& hots
         currentCursorImage.destroySurface = surf->events.destroy.registerListener([this](std::any data) { resetCursorImage(); });
         currentCursorImage.commitSurface  = surf->resource()->events.commit.registerListener([this](std::any data) {
             damageIfSoftware();
-            currentCursorImage.size  = currentCursorImage.surface->resource()->current.buffer ? currentCursorImage.surface->resource()->current.buffer->size : Vector2D{};
+            currentCursorImage.size  = currentCursorImage.surface->resource()->current.texture ? currentCursorImage.surface->resource()->current.bufferSize : Vector2D{};
             currentCursorImage.scale = currentCursorImage.surface ? currentCursorImage.surface->resource()->current.scale : 1.F;
             recheckEnteredOutputs();
             updateCursorBackend();
             damageIfSoftware();
         });
 
-        if (surf->resource()->current.buffer) {
+        if (surf->resource()->current.texture) {
-            currentCursorImage.size = surf->resource()->current.buffer->size;
+            currentCursorImage.size = surf->resource()->current.bufferSize;
             timespec now;
             clock_gettime(CLOCK_MONOTONIC, &now);
             surf->resource()->frame(&now);
@@ -430,16 +431,39 @@ SP<Aquamarine::IBuffer> CPointerManager::renderHWCursorBuffer(SP<CPointerManager
         // clear buffer
         memset(bufPtr, 0, std::get<2>(bufData));
 
-        auto texBuffer = currentCursorImage.pBuffer ? currentCursorImage.pBuffer : currentCursorImage.surface->resource()->current.buffer;
+        if (currentCursorImage.pBuffer) {
+            auto texAttrs = currentCursorImage.pBuffer->shm();
 
-        if (texBuffer) {
+            if (!texAttrs.success) {
-            auto textAttrs = texBuffer->shm();
+                Debug::log(TRACE, "Cannot use dumb copy on dmabuf cursor buffers");
-            auto texData   = texBuffer->beginDataPtr(GBM_BO_TRANSFER_WRITE);
+                return nullptr;
+            }
+
+            auto texData = currentCursorImage.pBuffer->beginDataPtr(GBM_BO_TRANSFER_WRITE);
             auto texPtr  = std::get<0>(texData);
-            Debug::log(TRACE, "cursor texture {}x{} {} {} {}", textAttrs.size.x, textAttrs.size.y, (void*)texPtr, textAttrs.format, textAttrs.stride);
+            Debug::log(TRACE, "cursor texture {}x{} {} {} {}", texAttrs.size.x, texAttrs.size.y, (void*)texPtr, texAttrs.format, texAttrs.stride);
             // copy cursor texture
-            for (int i = 0; i < texBuffer->shm().size.y; i++)
+            for (int i = 0; i < texAttrs.size.y; i++)
-                memcpy(bufPtr + i * buf->dmabuf().strides[0], texPtr + i * textAttrs.stride, textAttrs.stride);
+                memcpy(bufPtr + i * buf->dmabuf().strides[0], texPtr + i * texAttrs.stride, texAttrs.stride);
+        } else if (currentCursorImage.surface && currentCursorImage.surface->resource()->role->role() == SURFACE_ROLE_CURSOR) {
+            const auto SURFACE   = currentCursorImage.surface->resource();
+            auto&      shmBuffer = CCursorSurfaceRole::cursorPixelData(SURFACE);
+            Debug::log(TRACE, "cursor texture pixel data length: {}B", shmBuffer.size());
+
+            if (shmBuffer.data()) {
+                // copy cursor texture
+                // assume format is 32bpp
+                size_t STRIDE = 4 * SURFACE->current.bufferSize.x;
+                for (int i = 0; i < SURFACE->current.bufferSize.y; i++)
+                    memcpy(bufPtr + i * buf->dmabuf().strides[0], shmBuffer.data() + i * STRIDE, STRIDE);
+            } else {
+                // if there is no data, hide the cursor
+                memset(bufPtr, '\0', buf->size.x * buf->size.y * 4 /* assume 32bpp */);
+            }
+
+        } else {
+            Debug::log(TRACE, "Unsupported cursor buffer/surface, falling back to sw (can't dumb copy)");
+            return nullptr;
         }
 
         buf->endDataPtr();
@@ -740,7 +764,7 @@ void CPointerManager::onMonitorLayoutChange() {
 }
 
 SP<CTexture> CPointerManager::getCurrentCursorTexture() {
-    if (!currentCursorImage.pBuffer && (!currentCursorImage.surface || !currentCursorImage.surface->resource()->current.buffer))
+    if (!currentCursorImage.pBuffer && (!currentCursorImage.surface || !currentCursorImage.surface->resource()->current.texture))
         return nullptr;
 
     if (currentCursorImage.pBuffer) {
@@ -749,7 +773,7 @@ SP<CTexture> CPointerManager::getCurrentCursorTexture() {
         return currentCursorImage.bufferTex;
     }
 
-    return currentCursorImage.surface->resource()->current.buffer->texture;
+    return currentCursorImage.surface->resource()->current.texture;
 }
 
 void CPointerManager::attachPointer(SP<IPointer> pointer) {

src/protocols/DRMSyncobj.cpp

@@ -47,7 +47,7 @@ CDRMSyncobjSurfaceResource::CDRMSyncobjSurfaceResource(SP<CWpLinuxDrmSyncobjSurf
             return;
         }
 
-        if ((acquireTimeline || releaseTimeline) && !surface->pending.buffer) {
+        if ((acquireTimeline || releaseTimeline) && !surface->pending.texture) {
             resource->error(WP_LINUX_DRM_SYNCOBJ_SURFACE_V1_ERROR_NO_BUFFER, "Missing buffer");
             surface->pending.rejected = true;
             return;

src/protocols/InputMethodV2.cpp

@@ -107,14 +107,14 @@ CInputMethodPopupV2::CInputMethodPopupV2(SP<CZwpInputPopupSurfaceV2> resource_,
     });
 
     listeners.commitSurface = surface->events.commit.registerListener([this](std::any d) {
-        if (pSurface->current.buffer && !mapped) {
+        if (pSurface->current.texture && !mapped) {
             mapped = true;
             pSurface->map();
             events.map.emit();
             return;
         }
 
-        if (!pSurface->current.buffer && mapped) {
+        if (!pSurface->current.texture && mapped) {
             mapped = false;
             pSurface->unmap();
             events.unmap.emit();

src/protocols/LayerShell.cpp

@@ -44,7 +44,7 @@ CLayerShellResource::CLayerShellResource(SP<CZwlrLayerSurfaceV1> resource_, SP<C
         current           = pending;
         pending.committed = 0;
 
-        bool attachedBuffer = surface->current.buffer;
+        bool attachedBuffer = surface->current.texture;
 
         if (attachedBuffer && !configured) {
             surface->error(-1, "layerSurface was not configured, but a buffer was attached");

src/protocols/SessionLock.cpp

@@ -24,7 +24,7 @@ CSessionLockSurface::CSessionLockSurface(SP<CExtSessionLockSurfaceV1> resource_,
     resource->setAckConfigure([this](CExtSessionLockSurfaceV1* r, uint32_t serial) { ackdConfigure = true; });
 
     listeners.surfaceCommit = pSurface->events.commit.registerListener([this](std::any d) {
-        if (!pSurface->current.buffer) {
+        if (!pSurface->current.texture) {
             LOGM(ERR, "SessionLock attached a null buffer");
             resource->error(EXT_SESSION_LOCK_SURFACE_V1_ERROR_NULL_BUFFER, "Null buffer attached");
             return;

src/protocols/Viewporter.cpp

@@ -54,13 +54,13 @@ CViewportResource::CViewportResource(SP<CWpViewport> resource_, SP<CWLSurfaceRes
     });
 
     listeners.surfacePrecommit = surface->events.precommit.registerListener([this](std::any d) {
-        if (!surface || !surface->pending.buffer)
+        if (!surface || !surface->pending.texture)
             return;
 
         if (surface->pending.viewport.hasSource) {
             auto& src = surface->pending.viewport.source;
 
-            if (src.w + src.x > surface->pending.buffer->size.x || src.h + src.y > surface->pending.buffer->size.y) {
+            if (src.w + src.x > surface->pending.bufferSize.x || src.h + src.y > surface->pending.bufferSize.y) {
                 resource->error(WP_VIEWPORT_ERROR_BAD_VALUE, "Box doesn't fit");
                 surface->pending.rejected = true;
                 return;

src/protocols/XDGShell.cpp

@@ -347,12 +347,12 @@ CXDGSurfaceResource::CXDGSurfaceResource(SP<CXdgSurface> resource_, SP<CXDGWMBas
         if (toplevel)
             toplevel->current = toplevel->pending;
 
-        if (initialCommit && surface->pending.buffer) {
+        if (initialCommit && surface->pending.texture) {
             resource->error(-1, "Buffer attached before initial commit");
             return;
         }
 
-        if (surface->current.buffer && !mapped) {
+        if (surface->current.texture && !mapped) {
             // this forces apps to not draw CSD.
             if (toplevel)
                 toplevel->setMaximized(true);
@@ -363,7 +363,7 @@ CXDGSurfaceResource::CXDGSurfaceResource(SP<CXdgSurface> resource_, SP<CXDGWMBas
             return;
         }
 
-        if (!surface->current.buffer && mapped) {
+        if (!surface->current.texture && mapped) {
             mapped = false;
             events.unmap.emit();
             surface->unmap();

src/protocols/core/Compositor.cpp

@@ -1,5 +1,6 @@
 #include "Compositor.hpp"
 #include "Output.hpp"
+#include "Seat.hpp"
 #include "../types/WLBuffer.hpp"
 #include <algorithm>
 #include <ranges>
@@ -9,6 +10,7 @@
 #include "../PresentationTime.hpp"
 #include "../DRMSyncobj.hpp"
 #include "../../render/Renderer.hpp"
+#include <cstring>
 
 #define LOGM PROTO::compositor->protoLog
 
@@ -19,8 +21,6 @@ class CDefaultSurfaceRole : public ISurfaceRole {
     }
 };
 
-SP<CDefaultSurfaceRole> defaultRole = makeShared<CDefaultSurfaceRole>();
-
 CWLCallbackResource::CWLCallbackResource(SP<CWlCallback> resource_) : resource(resource_) {
     ;
 }
@@ -63,7 +63,7 @@ CWLSurfaceResource::CWLSurfaceResource(SP<CWlSurface> resource_) : resource(reso
 
     resource->setData(this);
 
-    role = defaultRole;
+    role = makeShared<CDefaultSurfaceRole>();
 
     resource->setDestroy([this](CWlSurface* r) { destroy(); });
     resource->setOnDestroy([this](CWlSurface* r) { destroy(); });
@@ -76,40 +76,41 @@ CWLSurfaceResource::CWLSurfaceResource(SP<CWlSurface> resource_) : resource(reso
             pending.texture.reset();
         } else {
             auto res           = CWLBufferResource::fromResource(buffer);
-            pending.buffer  = res && res->buffer ? res->buffer.lock() : nullptr;
+            pending.buffer     = res && res->buffer ? makeShared<CHLBufferReference>(res->buffer.lock(), self.lock()) : nullptr;
             pending.size       = res && res->buffer ? res->buffer->size : Vector2D{};
             pending.texture    = res && res->buffer ? res->buffer->texture : nullptr;
-            if (res)
+            pending.bufferSize = res && res->buffer ? res->buffer->size : Vector2D{};
-                res->released = false;
         }
 
-        Vector2D oldBufSize = current.buffer ? current.buffer->size : Vector2D{};
+        Vector2D oldBufSize = current.buffer ? current.bufferSize : Vector2D{};
-        Vector2D newBufSize = pending.buffer ? pending.buffer->size : Vector2D{};
+        Vector2D newBufSize = pending.buffer ? pending.bufferSize : Vector2D{};
 
         if (oldBufSize != newBufSize || current.buffer != pending.buffer)
             pending.bufferDamage = CBox{{}, {INT32_MAX, INT32_MAX}};
     });
 
     resource->setCommit([this](CWlSurface* r) {
-        if (pending.buffer)
+        if (pending.texture)
-            pending.bufferDamage.intersect(CBox{{}, pending.buffer->size});
+            pending.bufferDamage.intersect(CBox{{}, pending.bufferSize});
 
-        if (!pending.buffer)
+        if (!pending.texture)
             pending.size = {};
         else if (pending.viewport.hasDestination)
             pending.size = pending.viewport.destination;
         else if (pending.viewport.hasSource)
             pending.size = pending.viewport.source.size();
         else {
-            Vector2D tfs = pending.transform % 2 == 1 ? Vector2D{pending.buffer->size.y, pending.buffer->size.x} : pending.buffer->size;
+            Vector2D tfs = pending.transform % 2 == 1 ? Vector2D{pending.bufferSize.y, pending.bufferSize.x} : pending.bufferSize;
             pending.size = tfs / pending.scale;
         }
 
         pending.damage.intersect(CBox{{}, pending.size});
 
         events.precommit.emit();
-        if (pending.rejected)
+        if (pending.rejected) {
+            dropPendingBuffer();
             return;
+        }
 
         if (stateLocks <= 0)
             commitPendingState();
@@ -160,6 +161,14 @@ void CWLSurfaceResource::destroy() {
     PROTO::compositor->destroyResource(this);
 }
 
+void CWLSurfaceResource::dropPendingBuffer() {
+    pending.buffer.reset();
+}
+
+void CWLSurfaceResource::dropCurrentBuffer() {
+    current.buffer.reset();
+}
+
 SP<CWLSurfaceResource> CWLSurfaceResource::fromResource(wl_resource* res) {
     auto data = (CWLSurfaceResource*)(((CWlSurface*)wl_resource_get_user_data(res))->data());
     return data ? data->self.lock() : nullptr;
@@ -240,7 +249,7 @@ void CWLSurfaceResource::frame(timespec* now) {
 }
 
 void CWLSurfaceResource::resetRole() {
-    role = defaultRole;
+    role = makeShared<CDefaultSurfaceRole>();
 }
 
 void CWLSurfaceResource::bfHelper(std::vector<SP<CWLSurfaceResource>> nodes, std::function<void(SP<CWLSurfaceResource>, const Vector2D&, void*)> fn, void* data) {
@@ -254,6 +263,8 @@ void CWLSurfaceResource::bfHelper(std::vector<SP<CWLSurfaceResource>> nodes, std
         for (auto& c : n->subsurfaces) {
             if (c->zIndex >= 0)
                 break;
+            if (c->surface.expired())
+                continue;
             nodes2.push_back(c->surface.lock());
         }
     }
@@ -277,6 +288,8 @@ void CWLSurfaceResource::bfHelper(std::vector<SP<CWLSurfaceResource>> nodes, std
         for (auto& c : n->subsurfaces) {
             if (c->zIndex < 0)
                 continue;
+            if (c->surface.expired())
+                continue;
             nodes2.push_back(c->surface.lock());
         }
     }
@@ -343,14 +356,9 @@ void CWLSurfaceResource::unmap() {
 }
 
 void CWLSurfaceResource::releaseBuffers(bool onlyCurrent) {
-    if (current.buffer && !current.buffer->resource->released)
-        current.buffer->sendRelease();
-    if (pending.buffer && !pending.buffer->resource->released && !onlyCurrent)
-        pending.buffer->sendRelease();
-
-    pending.buffer.reset();
     if (!onlyCurrent)
-        current.buffer.reset();
+        dropPendingBuffer();
+    dropCurrentBuffer();
 }
 
 void CWLSurfaceResource::error(int code, const std::string& str) {
@@ -375,18 +383,18 @@ CBox CWLSurfaceResource::extends() {
 }
 
 Vector2D CWLSurfaceResource::sourceSize() {
-    if (!current.buffer)
+    if (!current.texture)
         return {};
 
     if (current.viewport.hasSource)
         return current.viewport.source.size();
 
-    Vector2D trc = current.transform % 2 == 1 ? Vector2D{current.buffer->size.y, current.buffer->size.x} : current.buffer->size;
+    Vector2D trc = current.transform % 2 == 1 ? Vector2D{current.bufferSize.y, current.bufferSize.x} : current.bufferSize;
     return trc / current.scale;
 }
 
 CRegion CWLSurfaceResource::accumulateCurrentBufferDamage() {
-    if (!current.buffer)
+    if (!current.texture)
         return {};
 
     CRegion surfaceDamage = current.damage;
@@ -398,7 +406,7 @@ CRegion CWLSurfaceResource::accumulateCurrentBufferDamage() {
     if (current.viewport.hasSource)
         surfaceDamage.translate(current.viewport.source.pos());
 
-    Vector2D trc = current.transform % 2 == 1 ? Vector2D{current.buffer->size.y, current.buffer->size.x} : current.buffer->size;
+    Vector2D trc = current.transform % 2 == 1 ? Vector2D{current.bufferSize.y, current.bufferSize.x} : current.bufferSize;
 
     return surfaceDamage.scale(current.scale).transform(wlTransformToHyprutils(invertTransform(current.transform)), trc.x, trc.y).add(current.bufferDamage);
 }
@@ -421,16 +429,21 @@ void CWLSurfaceResource::commitPendingState() {
     pending.damage.clear();
     pending.bufferDamage.clear();
 
-    if (current.buffer && current.buffer->texture)
+    if (current.texture)
-        current.buffer->texture->m_eTransform = wlTransformToHyprutils(current.transform);
+        current.texture->m_eTransform = wlTransformToHyprutils(current.transform);
 
-    if (current.buffer && !current.buffer->resource->released) {
+    if (current.buffer && current.buffer->buffer) {
-        current.buffer->update(accumulateCurrentBufferDamage());
+        current.buffer->buffer->update(accumulateCurrentBufferDamage());
+
+        // if the surface is a cursor, update the shm buffer
+        // TODO: don't update the entire texture
+        if (role->role() == SURFACE_ROLE_CURSOR)
+            updateCursorShm();
 
         // release the buffer if it's synchronous as update() has done everything thats needed
         // so we can let the app know we're done.
-        if (current.buffer->isSynchronous())
+        if (current.buffer->buffer->isSynchronous())
-            current.buffer->sendReleaseWithSurface(self.lock());
+            dropCurrentBuffer();
     }
 
     // TODO: we should _accumulate_ and not replace above if sync
@@ -455,22 +468,39 @@ void CWLSurfaceResource::commitPendingState() {
     }
 
     // for async buffers, we can only release the buffer once we are unrefing it from current.
-    if (previousBuffer && !previousBuffer->isSynchronous() && !previousBuffer->resource->released) {
+    // if the backend took it, ref it with the lambda. Otherwise, the end of this scope will release it.
-        if (previousBuffer->lockedByBackend) {
+    if (previousBuffer && previousBuffer->buffer && !previousBuffer->buffer->isSynchronous()) {
-            previousBuffer->hlEvents.backendRelease = previousBuffer->events.backendRelease.registerListener([this, previousBuffer](std::any data) {
+        if (previousBuffer->buffer->lockedByBackend && !previousBuffer->buffer->hlEvents.backendRelease) {
-                if (!self.expired()) // could be dead in the dtor
+            previousBuffer->buffer->lock();
-                    previousBuffer->sendReleaseWithSurface(self.lock());
+            previousBuffer->buffer->unlockOnBufferRelease(self);
-                else
-                    previousBuffer->sendRelease();
-                previousBuffer->hlEvents.backendRelease.reset();
-            });
-        } else
-            previousBuffer->sendReleaseWithSurface(self.lock());
-
-        previousBuffer->resource->released = true; // set it here regardless so we dont set more listeners for backendRelease
         }
     }
 
+    lastBuffer = current.buffer ? current.buffer->buffer : WP<IHLBuffer>{};
+}
+
+void CWLSurfaceResource::updateCursorShm() {
+    auto buf = current.buffer ? current.buffer : lastBuffer;
+
+    if (!buf)
+        return;
+
+    // TODO: actually use damage
+    auto& shmData  = CCursorSurfaceRole::cursorPixelData(self.lock());
+    auto  shmAttrs = current.buffer->buffer->shm();
+
+    if (!shmAttrs.success) {
+        LOGM(TRACE, "updateCursorShm: ignoring, not a shm buffer");
+        return;
+    }
+
+    // no need to end, shm.
+    auto [pixelData, fmt, bufLen] = current.buffer->buffer->beginDataPtr(0);
+
+    shmData.resize(bufLen);
+    memcpy(shmData.data(), pixelData, bufLen);
+}
+
 void CWLSurfaceResource::presentFeedback(timespec* when, CMonitor* pMonitor, bool needsExplicitSync) {
     frame(when);
     auto FEEDBACK = makeShared<CQueuedPresentationData>(self.lock());

src/protocols/core/Compositor.hpp

@@ -87,10 +87,10 @@ class CWLSurfaceResource {
         CRegion                opaque, input = CBox{{}, {INT32_MAX, INT32_MAX}}, damage, bufferDamage = CBox{{}, {INT32_MAX, INT32_MAX}} /* initial damage */;
         wl_output_transform    transform = WL_OUTPUT_TRANSFORM_NORMAL;
         int                    scale     = 1;
-        SP<IHLBuffer>       buffer;
+        SP<CHLBufferReference> buffer; // buffer ref will be released once the buffer is no longer locked. For checking if a buffer is attached to this state, check texture.
         SP<CTexture>           texture;
         Vector2D               offset;
-        Vector2D            size;
+        Vector2D               size, bufferSize;
         struct {
             bool     hasDestination = false;
             bool     hasSource      = false;
@@ -116,7 +116,7 @@ class CWLSurfaceResource {
     std::vector<WP<CMonitor>>              enteredOutputs;
     bool                                   mapped = false;
     std::vector<WP<CWLSubsurfaceResource>> subsurfaces;
-    WP<ISurfaceRole>                       role;
+    SP<ISurfaceRole>                       role;
     WP<CViewportResource>                  viewportResource;
     WP<CDRMSyncobjSurfaceResource>         syncobj; // may not be present
 
@@ -134,12 +134,21 @@ class CWLSurfaceResource {
     SP<CWlSurface> resource;
     wl_client*     pClient = nullptr;
 
+    // this is for cursor dumb copy. Due to our (and wayland's...) architecture,
+    // this stupid-ass hack is used
+    WP<IHLBuffer> lastBuffer;
+
     int           stateLocks = 0;
 
     void          destroy();
     void          releaseBuffers(bool onlyCurrent = true);
+    void          dropPendingBuffer();
+    void          dropCurrentBuffer();
     void          commitPendingState();
     void          bfHelper(std::vector<SP<CWLSurfaceResource>> nodes, std::function<void(SP<CWLSurfaceResource>, const Vector2D&, void*)> fn, void* data);
+    void          updateCursorShm();
+
+    friend class CWLPointerResource;
 };
 
 class CWLCompositorResource {

src/protocols/core/DataDevice.cpp

@@ -469,12 +469,12 @@ void CWLDataDeviceProtocol::initiateDrag(WP<CWLDataSourceResource> currentSource
     if (dragSurface) {
         dnd.dndSurfaceDestroy = dragSurface->events.destroy.registerListener([this](std::any d) { abortDrag(); });
         dnd.dndSurfaceCommit  = dragSurface->events.commit.registerListener([this](std::any d) {
-            if (dnd.dndSurface->current.buffer && !dnd.dndSurface->mapped) {
+            if (dnd.dndSurface->current.texture && !dnd.dndSurface->mapped) {
                 dnd.dndSurface->map();
                 return;
             }
 
-            if (dnd.dndSurface->current.buffer <= 0 && dnd.dndSurface->mapped) {
+            if (dnd.dndSurface->current.texture <= 0 && dnd.dndSurface->mapped) {
                 dnd.dndSurface->unmap();
                 return;
             }
@@ -660,13 +660,13 @@ void CWLDataDeviceProtocol::abortDrag() {
 }
 
 void CWLDataDeviceProtocol::renderDND(CMonitor* pMonitor, timespec* when) {
-    if (!dnd.dndSurface || !dnd.dndSurface->current.buffer || !dnd.dndSurface->current.buffer->texture)
+    if (!dnd.dndSurface || !dnd.dndSurface->current.texture)
         return;
 
     const auto POS = g_pInputManager->getMouseCoordsInternal();
 
     CBox       box = CBox{POS, dnd.dndSurface->current.size}.translate(-pMonitor->vecPosition + g_pPointerManager->cursorSizeLogical() / 2.F).scale(pMonitor->scale);
-    g_pHyprOpenGL->renderTexture(dnd.dndSurface->current.buffer->texture, &box, 1.F);
+    g_pHyprOpenGL->renderTexture(dnd.dndSurface->current.texture, &box, 1.F);
 
     box = CBox{POS, dnd.dndSurface->current.size}.translate(g_pPointerManager->cursorSizeLogical() / 2.F);
     g_pHyprRenderer->damageBox(&box);

src/protocols/core/Seat.cpp

@@ -119,7 +119,19 @@ CWLPointerResource::CWLPointerResource(SP<CWlPointer> resource_, SP<CWLSeatResou
             return;
         }
 
-        g_pSeatManager->onSetCursor(owner.lock(), serial, surf ? CWLSurfaceResource::fromResource(surf) : nullptr, {hotX, hotY});
+        auto surfResource = surf ? CWLSurfaceResource::fromResource(surf) : nullptr;
+
+        if (surfResource && surfResource->role->role() != SURFACE_ROLE_CURSOR && surfResource->role->role() != SURFACE_ROLE_UNASSIGNED) {
+            r->error(-1, "Cursor surface already has a different role");
+            return;
+        }
+
+        if (surfResource) {
+            surfResource->role = makeShared<CCursorSurfaceRole>();
+            surfResource->updateCursorShm();
+        }
+
+        g_pSeatManager->onSetCursor(owner.lock(), serial, surfResource, {hotX, hotY});
     });
 
     if (g_pSeatManager->state.pointerFocus && g_pSeatManager->state.pointerFocus->client() == resource->client())
@@ -546,3 +558,10 @@ SP<CWLSeatResource> CWLSeatProtocol::seatResourceForClient(wl_client* client) {
 
     return nullptr;
 }
+
+std::vector<uint8_t>& CCursorSurfaceRole::cursorPixelData(SP<CWLSurfaceResource> surface) {
+    RASSERT(surface->role->role() == SURFACE_ROLE_CURSOR, "cursorPixelData called on a non-cursor surface");
+
+    auto role = (CCursorSurfaceRole*)surface->role.get();
+    return role->cursorShmPixelData;
+}

src/protocols/core/Seat.hpp

@@ -16,6 +16,7 @@
 #include "wayland.hpp"
 #include "../../helpers/signal/Signal.hpp"
 #include "../../helpers/math/Math.hpp"
+#include "../types/SurfaceRole.hpp"
 
 constexpr const char* HL_SEAT_NAME = "Hyprland";
 
@@ -27,6 +28,20 @@ class CWLKeyboardResource;
 class CWLTouchResource;
 class CWLSeatResource;
 
+class CCursorSurfaceRole : public ISurfaceRole {
+  public:
+    virtual eSurfaceRole role() {
+        return SURFACE_ROLE_CURSOR;
+    }
+
+    // gets the current pixel data from a shm surface
+    // will assert if the surface is not a cursor
+    static std::vector<uint8_t>& cursorPixelData(SP<CWLSurfaceResource> surface);
+
+  private:
+    std::vector<uint8_t> cursorShmPixelData;
+};
+
 class CWLTouchResource {
   public:
     CWLTouchResource(SP<CWlTouch> resource_, SP<CWLSeatResource> owner_);

src/protocols/core/Subcompositor.cpp

@@ -80,13 +80,13 @@ CWLSubsurfaceResource::CWLSubsurfaceResource(SP<CWlSubsurface> resource_, SP<CWL
     });
 
     listeners.commitSurface = surface->events.commit.registerListener([this](std::any d) {
-        if (surface->current.buffer && !surface->mapped) {
+        if (surface->current.texture && !surface->mapped) {
             surface->map();
             surface->events.map.emit();
             return;
         }
 
-        if (!surface->current.buffer && surface->mapped) {
+        if (!surface->current.texture && surface->mapped) {
             surface->events.unmap.emit();
             surface->unmap();
             return;

src/protocols/types/Buffer.cpp

@@ -1,5 +1,10 @@
 #include "Buffer.hpp"
 
+IHLBuffer::~IHLBuffer() {
+    if (locked() && resource)
+        sendRelease();
+}
+
 void IHLBuffer::sendRelease() {
     resource->sendRelease();
 }
@@ -8,3 +13,54 @@ void IHLBuffer::sendReleaseWithSurface(SP<CWLSurfaceResource> surf) {
     if (resource && resource->good())
         resource->sendReleaseWithSurface(surf);
 }
+
+void IHLBuffer::lock() {
+    nLocks++;
+}
+
+void IHLBuffer::unlock() {
+    nLocks--;
+
+    ASSERT(nLocks >= 0);
+
+    if (nLocks == 0)
+        sendRelease();
+}
+
+void IHLBuffer::unlockWithSurface(SP<CWLSurfaceResource> surf) {
+    nLocks--;
+
+    ASSERT(nLocks >= 0);
+
+    if (nLocks == 0)
+        sendReleaseWithSurface(surf);
+}
+
+bool IHLBuffer::locked() {
+    return nLocks > 0;
+}
+
+void IHLBuffer::unlockOnBufferRelease(WP<CWLSurfaceResource> surf) {
+    unlockSurface           = surf;
+    hlEvents.backendRelease = events.backendRelease.registerListener([this](std::any data) {
+        if (unlockSurface.expired())
+            unlock();
+        else
+            unlockWithSurface(unlockSurface.lock());
+        hlEvents.backendRelease.reset();
+    });
+}
+
+CHLBufferReference::CHLBufferReference(SP<IHLBuffer> buffer_, SP<CWLSurfaceResource> surface_) : buffer(buffer_), surface(surface_) {
+    buffer->lock();
+}
+
+CHLBufferReference::~CHLBufferReference() {
+    if (buffer.expired())
+        return;
+
+    if (surface)
+        buffer->unlockWithSurface(surface.lock());
+    else
+        buffer->unlock();
+}

src/protocols/types/Buffer.hpp

@@ -8,9 +8,7 @@
 
 class IHLBuffer : public Aquamarine::IBuffer {
   public:
-    virtual ~IHLBuffer() {
+    virtual ~IHLBuffer();
-        ;
-    }
     virtual Aquamarine::eBufferCapability caps()                        = 0;
     virtual Aquamarine::eBufferType       type()                        = 0;
     virtual void                          update(const CRegion& damage) = 0;
@@ -18,6 +16,12 @@ class IHLBuffer : public Aquamarine::IBuffer {
     virtual bool                          good()                        = 0;
     virtual void                          sendRelease();
     virtual void                          sendReleaseWithSurface(SP<CWLSurfaceResource>);
+    virtual void                          lock();
+    virtual void                          unlock();
+    virtual void                          unlockWithSurface(SP<CWLSurfaceResource> surf);
+    virtual bool                          locked();
+
+    void                                  unlockOnBufferRelease(WP<CWLSurfaceResource> surf /* optional */);
 
     SP<CTexture>                          texture;
     bool                                  opaque = false;
@@ -26,4 +30,22 @@ class IHLBuffer : public Aquamarine::IBuffer {
     struct {
         CHyprSignalListener backendRelease;
     } hlEvents;
+
+  private:
+    int                    nLocks = 0;
+
+    WP<CWLSurfaceResource> unlockSurface;
+};
+
+// for ref-counting. Releases in ~dtor
+// surface optional
+class CHLBufferReference {
+  public:
+    CHLBufferReference(SP<IHLBuffer> buffer, SP<CWLSurfaceResource> surface);
+    ~CHLBufferReference();
+
+    WP<IHLBuffer> buffer;
+
+  private:
+    WP<CWLSurfaceResource> surface;
 };

src/protocols/types/SurfaceRole.hpp

@@ -6,6 +6,7 @@ enum eSurfaceRole {
     SURFACE_ROLE_LAYER_SHELL,
     SURFACE_ROLE_EASTER_EGG,
     SURFACE_ROLE_SUBSURFACE,
+    SURFACE_ROLE_CURSOR,
 };
 
 class ISurfaceRole {

src/protocols/types/WLBuffer.cpp

@@ -29,7 +29,6 @@ bool CWLBufferResource::good() {
 }
 
 void CWLBufferResource::sendRelease() {
-    released = true;
     resource->sendRelease();
 }
 

src/protocols/types/WLBuffer.hpp

@@ -24,8 +24,6 @@ class CWLBufferResource {
 
     WP<CWLBufferResource>        self;
 
-    bool                         released = false;
-
   private:
     CWLBufferResource(SP<CWlBuffer> resource_);
 

src/render/Renderer.cpp

@@ -102,10 +102,10 @@ CHyprRenderer::~CHyprRenderer() {
 }
 
 static void renderSurface(SP<CWLSurfaceResource> surface, int x, int y, void* data) {
-    if (!surface->current.buffer || !surface->current.buffer->texture)
+    if (!surface->current.texture)
         return;
 
-    const auto& TEXTURE                     = surface->current.buffer->texture;
+    const auto& TEXTURE                     = surface->current.texture;
     const auto  RDATA                       = (SRenderData*)data;
     const auto  INTERACTIVERESIZEINPROGRESS = RDATA->pWindow && g_pInputManager->currentlyDraggedWindow.lock() == RDATA->pWindow && g_pInputManager->dragMode == MBIND_RESIZE;
 
@@ -192,8 +192,8 @@ static void renderSurface(SP<CWLSurfaceResource> surface, int x, int y, void* da
     windowBox.round();
 
     const bool MISALIGNEDFSV1 = std::floor(RDATA->pMonitor->scale) != RDATA->pMonitor->scale /* Fractional */ && surface->current.scale == 1 /* fs protocol */ &&
-        windowBox.size() != surface->current.buffer->size /* misaligned */ && DELTALESSTHAN(windowBox.width, surface->current.buffer->size.x, 3) &&
+        windowBox.size() != surface->current.bufferSize /* misaligned */ && DELTALESSTHAN(windowBox.width, surface->current.bufferSize.x, 3) &&
-        DELTALESSTHAN(windowBox.height, surface->current.buffer->size.y, 3) /* off by one-or-two */ &&
+        DELTALESSTHAN(windowBox.height, surface->current.bufferSize.y, 3) /* off by one-or-two */ &&
         (!RDATA->pWindow || (!RDATA->pWindow->m_vRealSize.isBeingAnimated() && !INTERACTIVERESIZEINPROGRESS)) /* not window or not animated/resizing */;
 
     g_pHyprRenderer->calculateUVForSurface(RDATA->pWindow, surface, RDATA->surface == surface, windowBox.size(), MISALIGNEDFSV1);
@@ -1014,7 +1014,7 @@ void CHyprRenderer::calculateUVForSurface(PHLWINDOW pWindow, SP<CWLSurfaceResour
 
         if (pSurface->current.viewport.hasSource) {
             // we stretch it to dest. if no dest, to 1,1
-            Vector2D bufferSize   = pSurface->current.buffer->size;
+            Vector2D bufferSize   = pSurface->current.bufferSize;
             auto     bufferSource = pSurface->current.viewport.source;
 
             // calculate UV for the basic src_box. Assume dest == size. Scale to dest later
@@ -1030,8 +1030,8 @@ void CHyprRenderer::calculateUVForSurface(PHLWINDOW pWindow, SP<CWLSurfaceResour
         if (projSize != Vector2D{} && fixMisalignedFSV1) {
             // instead of nearest_neighbor (we will repeat / skip)
             // just cut off / expand surface
-            const Vector2D PIXELASUV    = Vector2D{1, 1} / pSurface->current.buffer->size;
+            const Vector2D PIXELASUV    = Vector2D{1, 1} / pSurface->current.bufferSize;
-            const Vector2D MISALIGNMENT = pSurface->current.buffer->size - projSize;
+            const Vector2D MISALIGNMENT = pSurface->current.bufferSize - projSize;
             if (MISALIGNMENT != Vector2D{})
                 uvBR -= MISALIGNMENT * PIXELASUV;
         }

src/render/Texture.cpp

@@ -28,6 +28,8 @@ CTexture::CTexture(const SP<Aquamarine::IBuffer> buffer) {
     if (!buffer)
         return;
 
+    m_bOpaque = buffer->opaque;
+
     auto attrs = buffer->dmabuf();
 
     if (!attrs.success) {
@@ -86,6 +88,7 @@ void CTexture::createFromDma(const Aquamarine::SDMABUFAttrs& attrs, void* image)
         return;
     }
 
+    m_bOpaque = FormatUtils::isFormatOpaque(attrs.format);
     m_iTarget = GL_TEXTURE_2D;
     m_iType   = TEXTURE_RGBA;
     m_vSize   = attrs.size;

src/render/Texture.hpp

@@ -41,6 +41,7 @@ class CTexture {
     Vector2D    m_vSize      = {};
     void*       m_pEglImage  = nullptr;
     eTransform  m_eTransform = HYPRUTILS_TRANSFORM_NORMAL;
+    bool        m_bOpaque    = false;
 
   private:
     void createFromShm(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const Vector2D& size);

src/xwayland/XSurface.cpp

@@ -62,12 +62,12 @@ void CXWaylandSurface::ensureListeners() {
         });
 
         listeners.commitSurface = surface->events.commit.registerListener([this](std::any d) {
-            if (surface->pending.buffer && !mapped) {
+            if (surface->pending.texture && !mapped) {
                 map();
                 return;
             }
 
-            if (!surface->pending.buffer && mapped) {
+            if (!surface->pending.texture && mapped) {
                 unmap();
                 return;
             }
@@ -131,7 +131,7 @@ void CXWaylandSurface::considerMap() {
         return;
     }
 
-    if (surface->pending.buffer) {
+    if (surface->pending.texture) {
         Debug::log(LOG, "XWayland surface: considerMap, sure, we have a buffer");
         map();
         return;