{"id":"CVE-2026-98518","state":"PUBLISHED","title":"drm/nouveau/dmem: pin VRAM for the whole registered range","description":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/nouveau/dmem: pin VRAM for the whole registered range\n\nCommit c32287471077 (\"gpu/drm/nouveau: enable THP support for GPU memory\nmigration\") grew the device-private region that\nnouveau_dmem_chunk_alloc() registers from DMEM_CHUNK_SIZE to\nDMEM_CHUNK_SIZE * NR_CHUNKS, but left the VRAM buffer object backing that\nregion at DMEM_CHUNK_SIZE.\n\nnouveau_dmem_page_addr() returns chunk->bo->offset plus the page's offset\nwithin the registered region, so every page past the first chunk resolves\nto VRAM outside the buffer object.\n\nSize the buffer object to the region it backs.","publishedAt":"2026-10-18T15:17:36.371Z","updatedAt":"2026-10-24T14:09:15.821Z","severity":{"score":7,"label":"High","version":"3.1","vector":"CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H","source":"Linux"},"assigner":"Linux","products":["Linux Linux"],"affected":[{"vendor":"Linux","product":"Linux","defaultStatus":"unaffected","versions":[{"version":"c3228747107705d47c7e9a03598a434a0380cb73","status":"affected","lessThan":"06c69c64686a31783c8bec6495dc22822d38c4ce","versionType":"git"},{"version":"c3228747107705d47c7e9a03598a434a0380cb73","status":"affected","lessThan":"64ca4cdd1031206424e6455f0ae4fb0560d8f46a","versionType":"git"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/nouveau/nouveau_dmem.c"]},{"vendor":"Linux","product":"Linux","defaultStatus":"affected","versions":[{"version":"6.19","status":"affected"},{"version":"0","status":"unaffected","lessThan":"6.19","versionType":"semver"},{"version":"7.2.9","status":"unaffected","lessThanOrEqual":"7.2.*","versionType":"semver"},{"version":"7.3","status":"unaffected","lessThanOrEqual":"*","versionType":"original_commit_for_fix"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/nouveau/nouveau_dmem.c"]}],"references":["https://git.kernel.org/stable/c/06c69c64686a31783c8bec6495dc22822d38c4ce","https://git.kernel.org/stable/c/64ca4cdd1031206424e6455f0ae4fb0560d8f46a"],"weaknesses":[],"ingestedAt":"2026-10-29T20:13:43.249Z","source":{"repository":"CVEProject/cvelistV5","path":"cves/2026/98xxx/CVE-2026-98518.json","commit":"556c9d3a8cd17e218bb8e452ba7cd619a384ff47"},"raw":{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-98518","state":"PUBLISHED","datePublished":"2026-10-18T15:17:36.371Z","dateUpdated":"2026-10-24T14:09:15.821Z","assignerShortName":"Linux","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","dateReserved":"2026-09-25T10:25:14.354Z"},"containers":{"cna":{"title":"drm/nouveau/dmem: pin VRAM for the whole registered range","descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/nouveau/dmem: pin VRAM for the whole registered range\n\nCommit c32287471077 (\"gpu/drm/nouveau: enable THP support for GPU memory\nmigration\") grew the device-private region that\nnouveau_dmem_chunk_alloc() registers from DMEM_CHUNK_SIZE to\nDMEM_CHUNK_SIZE * NR_CHUNKS, but left the VRAM buffer object backing that\nregion at DMEM_CHUNK_SIZE.\n\nnouveau_dmem_page_addr() returns chunk->bo->offset plus the page's offset\nwithin the registered region, so every page past the first chunk resolves\nto VRAM outside the buffer object.\n\nSize the buffer object to the region it backs."}],"affected":[{"vendor":"Linux","product":"Linux","defaultStatus":"unaffected","versions":[{"version":"c3228747107705d47c7e9a03598a434a0380cb73","status":"affected","lessThan":"06c69c64686a31783c8bec6495dc22822d38c4ce","versionType":"git"},{"version":"c3228747107705d47c7e9a03598a434a0380cb73","status":"affected","lessThan":"64ca4cdd1031206424e6455f0ae4fb0560d8f46a","versionType":"git"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/nouveau/nouveau_dmem.c"]},{"vendor":"Linux","product":"Linux","defaultStatus":"affected","versions":[{"version":"6.19","status":"affected"},{"version":"0","status":"unaffected","lessThan":"6.19","versionType":"semver"},{"version":"7.2.9","status":"unaffected","lessThanOrEqual":"7.2.*","versionType":"semver"},{"version":"7.3","status":"unaffected","lessThanOrEqual":"*","versionType":"original_commit_for_fix"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/nouveau/nouveau_dmem.c"]}],"references":[{"url":"https://git.kernel.org/stable/c/06c69c64686a31783c8bec6495dc22822d38c4ce"},{"url":"https://git.kernel.org/stable/c/64ca4cdd1031206424e6455f0ae4fb0560d8f46a"}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H","baseScore":7,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:L - ioctl(DRM_IOCTL_NOUVEAU_SVM_BIND) reaches nouveau_svmm_bind(), nouveau_dmem_migrate_vma(), nouveau_dmem_migrate_chunk(), nouveau_dmem_migrate_copy_one(), nouveau_dmem_page_alloc_locked(), then patched nouveau_dmem_chunk_alloc(); the triggering VMA and request come from a local process.\nAC:H - The path needs CONFIG_DRM_NOUVEAU_SVM, an EXPERIMENTAL default-n option depending on STAGING; nouveau_svm_init() also rejects GPUs newer than Pascal while nouveau_dmem_init() rejects older ones. The attacker cannot create these host build and hardware conditions.\nPR:L - NOUVEAU_SVM_INIT and NOUVEAU_SVM_BIND are registered with DRM_RENDER_ALLOW and neither DRM_ROOT_ONLY nor DRM_AUTH; an ordinary user able to open the Nouveau render node can initialize SVM and migrate their own VMA without capabilities.\nUI:N - The attacking process itself issues SVM_INIT and SVM_BIND and accesses its migrated VMA to trigger migration or readback; no victim must mount, open, or approve anything.\nS:U - nouveau_dmem_page_addr() supplies an invalid LOCAL_FB VRAM address to the same host's Nouveau copy engine and VMM; the code demonstrates a VRAM BO boundary violation, not a VM escape or IOMMU/DMA boundary bypass.\nC:H - nouveau_pfns_map() installs the out-of-BO VRAM PFN in the caller's SVM VMM, and a CPU fault invokes nouveau_dmem_migrate_to_ram() and nouveau_dmem_copy_folio() to copy from that physical VRAM address into a user page, permitting disclosure from overlapping VRAM allocations.\nI:H - nouveau_dmem_migrate_copy_one() passes the out-of-BO paddr to nvc0b5_migrate_copy() as a physical LOCAL_FB destination and sets a writable PFN for writable VMAs, enabling attacker-controlled writes into other VRAM allocations.\nA:H - The unchecked nvc0b5_migrate_copy() physical destination can overwrite active VRAM BOs or fall beyond installed VRAM, faulting or hanging the GPU and denying graphics and compute service to all clients."}]}],"providerMetadata":{"shortName":"Linux","orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","dateUpdated":"2026-10-24T14:09:15.821Z"},"cpeApplicability":[{"nodes":[{"operator":"OR","negate":false,"cpeMatch":[{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.19","versionEndExcluding":"7.2.9"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.19","versionEndExcluding":"7.3"}]}]}],"x_generator":{"engine":"bippy-1.2.0"}}}}}