{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-99773","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2026-09-25T10:25:14.433Z","datePublished":"2026-11-20T05:51:37.923Z","dateUpdated":"2026-11-22T11:10:05.721Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-11-22T11:10:05.721Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/drm_crtc: ensure dma_fence_ops remain valid during device unbind\n\nIn [1], sashiko reported the following issue:\n\n=== snip ===\nLooking at how these fences are managed, drm_crtc_create_fence()\ncreates a dma_fence without taking a reference to the drm_device or\ndrm_crtc. Because the sync_file framework exposes this fence to\nuserspace, the fence can outlive the CRTC.\n\nThe dma_fence contract requires that data accessed by dma_fence_ops\n(like get_driver_name) must remain valid for an RCU grace period after\nthe fence is signaled. However, drm_crtc_cleanup() and the subsequent\nfreeing of the device do not wait for an RCU grace period via\nsynchronize_rcu().\n\nIf userspace calls ioctl(SYNC_IOC_FILE_INFO) concurrently with a device\nhot-unplug:\n\nCPU1 (Userspace)\nsync_file_get_name()\n  ops = rcu_dereference(fence->ops);\n  if (!dma_fence_test_signaled_flag())\n    // Preempted or delayed here\n\nCPU2 (Driver Teardown)\nSignals the fence (setting fence->ops = NULL)\nDestroys and frees the CRTC without waiting for an RCU grace period\n\nCPU1 (Resumes)\n  ops->get_driver_name(fence) -> drm_crtc_fence_get_driver_name()\n  crtc = fence_to_crtc(fence); // Casts to the freed CRTC\n  return crtc->dev->driver->name; // Use-after-free\n\n...\n\nDoes the CRTC or DRM device need to be kept alive for the RCU grace\nperiod, or should the fence hold a proper reference to prevent the\nuse-after-free when get_driver_name() and get_timeline_name() access\nthe freed CRTC structure?\n=== snap ===\n\nI believe this to be a correct observation and this patch implements\nthe suggestion of waiting for an RCU grace period before proceeding\nwith destruction of the drm_crtc, so that get_driver_name() and\nget_timeline_name() can still work."}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H","baseScore":7.8,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:L - Local DRM_IOCTL_MODE_ATOMIC (drm_mode_atomic_ioctl -> prepare_signaling -> drm_crtc_create_fence) publishes the fence, and local SYNC_IOC_FILE_INFO uses it while drm_crtc_cleanup runs from drm_mode_config_cleanup on USBDEVFS_DISCONNECT or driver unbind. No network message supplies the bug; the missing synchronize_rcu() is in drm_crtc_cleanup.\nAC:L - A local DRM master creates the out-fence, then proc_ioctl(USBDEVFS_DISCONNECT) runs udl_usb_disconnect, gud_disconnect, or appletbdrm_disconnect. The last drm_release drops the drm_device ref so drm_mode_config_cleanup calls drm_crtc_cleanup. SYNC_IOC_FILE_INFO then runs; dma_fence_get_status locks the freed crtc->fence_lock with no condition the attacker cannot set up.\nPR:L - drm_master_open grants DRM_MASTER to the first opener of the DRM node with no CAP_SYS_ADMIN check, and drm_setclientcap enables atomic for that fd. USBDEVFS_DISCONNECT in proc_ioctl unbinds udl/gud/appletbdrm with no capable() check, and those USB device nodes are mode 0664, so a session user can tear the device down without init-namespace root.\nUI:N - The same local process opens the device, posts the atomic commit that returns the sync_file, disconnects the USB interface, closes the DRM fd, and calls SYNC_IOC_FILE_INFO. No other user has to mount media or accept a prompt.\nS:U - drm_crtc_fence_get_driver_name and dma_fence_get_status corrupt kernel memory of the host that owns the CRTC. This is not a KVM guest-to-host escape or an IOMMU boundary crossing.\nC:H - udl_usb_disconnect never signals the out-fence, so SYNC_IOC_FILE_INFO still calls drm_crtc_fence_get_driver_name and drm_crtc_fence_get_timeline_name. Those container_of fence->extern_lock back to the freed drm_crtc and return crtc->dev->driver->name and crtc->timeline_name into the ioctl buffer; reclaiming the slab turns that into a kernel string read.\nI:H - drm_crtc_create_fence stores &crtc->fence_lock as the dma-fence external lock. sync_file_ioctl_fence_info always calls dma_fence_get_status, which spin_lock()s that lock even after the fence is signaled. After drm_crtc_cleanup memsets the CRTC and the device is kfreed, that spin_lock writes reclaimed kernel memory.\nA:H - SYNC_IOC_FILE_INFO either NULL-dereferences crtc->dev after drm_crtc_cleanup's memset or spin_lock()s the freed crtc->fence_lock. Both fault in kernel context and oops the machine. The missing synchronize_rcu() in drm_crtc_cleanup is what lets the ioctl run against a CRTC that is already gone."}]}],"affected":[{"product":"Linux","vendor":"Linux","defaultStatus":"unaffected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/drm_crtc.c"],"versions":[{"version":"6d6003c4b613c93973e4e870d83f4bed2ad9ac34","lessThan":"dd9a96b2271e28340acf2e3764914fe5f32f6102","status":"affected","versionType":"git"},{"version":"6d6003c4b613c93973e4e870d83f4bed2ad9ac34","lessThan":"07b2d7a3f2b984c20f5f2630a94f43ede387a8dc","status":"affected","versionType":"git"},{"version":"6d6003c4b613c93973e4e870d83f4bed2ad9ac34","lessThan":"f8e6f4deb868fd9da3eda95db222609ab1edc18d","status":"affected","versionType":"git"},{"version":"6d6003c4b613c93973e4e870d83f4bed2ad9ac34","lessThan":"3be436cbd3d265fe4fcfbb5b650edad7b2108443","status":"affected","versionType":"git"}]},{"product":"Linux","vendor":"Linux","defaultStatus":"affected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/drm_crtc.c"],"versions":[{"version":"4.10","status":"affected"},{"version":"0","lessThan":"4.10","status":"unaffected","versionType":"semver"},{"version":"6.12.118","lessThanOrEqual":"6.12.*","status":"unaffected","versionType":"semver"},{"version":"6.18.64","lessThanOrEqual":"6.18.*","status":"unaffected","versionType":"semver"},{"version":"7.3.5","lessThanOrEqual":"7.3.*","status":"unaffected","versionType":"semver"},{"version":"7.4-rc2","lessThanOrEqual":"*","status":"unaffected","versionType":"original_commit_for_fix"}]}],"cpeApplicability":[{"nodes":[{"operator":"OR","negate":false,"cpeMatch":[{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"4.10","versionEndExcluding":"6.12.118"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"4.10","versionEndExcluding":"6.18.64"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"4.10","versionEndExcluding":"7.3.5"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"4.10","versionEndExcluding":"7.4-rc2"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/dd9a96b2271e28340acf2e3764914fe5f32f6102"},{"url":"https://git.kernel.org/stable/c/07b2d7a3f2b984c20f5f2630a94f43ede387a8dc"},{"url":"https://git.kernel.org/stable/c/f8e6f4deb868fd9da3eda95db222609ab1edc18d"},{"url":"https://git.kernel.org/stable/c/3be436cbd3d265fe4fcfbb5b650edad7b2108443"}],"title":"drm/drm_crtc: ensure dma_fence_ops remain valid during device unbind","x_generator":{"engine":"bippy-1.2.0"}}}}