{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-93105","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2026-09-17T16:02:15.085Z","datePublished":"2026-09-17T16:11:16.442Z","dateUpdated":"2026-09-18T17:55:56.276Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-09-18T17:55:56.276Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\nesp: do not unref managed frag pages in esp_ssg_unref()\n\nesp_ssg_unref() releases the page references held on the source\nscatterlist after the AEAD operation completes.  It calls\nskb_page_unref() on every frag page for an out-of-place transform\n(req->src != req->dst), and in the error path of esp_output_tail()\n(already_unref == true) on the request's own scatterlist.\n\nThis is wrong when the skb carries managed frags\n(SKBFL_MANAGED_FRAG_REFS).  Managed frags are owned by a zerocopy ubuf\nand the skb does not hold a per-frag page reference; io_uring SEND_ZC\nwith a registered buffer attaches the bvec pages this way via\nio_sg_from_iter().  The rest of the stack honours this invariant:\nskb_release_data() skips the per-frag unref when SKBFL_MANAGED_FRAG_REFS\nis set, and skb_zcopy_managed() is the guard used at the other unref\nsites.\n\nesp_ssg_unref() is missing that guard, so for a managed-frag skb it\ndrops a page reference the skb never acquired.  This can underflow the\npage reference count and free a page that is still in use.\n\nGuard the function with skb_zcopy_managed() so both unref paths are\nskipped for managed-frag skbs, matching skb_release_data()."}],"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 - The extra unref is reached from IORING_OP_SEND_ZC/SENDMSG_ZC with IORING_RECVSEND_FIXED_BUF: io_sendmsg_zc() -> sock_sendmsg() -> tcp_sendmsg_locked() -> xfrm_output() -> esp_output_tail() -> esp_ssg_unref(). io_sg_from_iter() is the only site that sets SKBFL_MANAGED_FRAG_REFS; esp_input() never calls esp_ssg_unref(), so a remote ESP packet cannot supply the buggy skb.\nAC:L - tcp_stream_alloc_skb() allocates MAX_TCP_HEADER then skb_reserve()s all of it, so a SEND_ZC TCP skb has ~0 tailroom. esp_output_head() therefore takes inplace=false (tailen > tailroom, nr_frags < MAX_SKB_FRAGS, data_len+tailen <= PAGE_SIZE), after which esp_output_tail() replaces the frags and esp_ssg_unref() always drops the managed pages. The attacker chooses the send size.\nPR:L - io_uring_setup() allows IORING_OP_SEND_ZC when sysctl_io_uring_disabled==0 (default); no capability is checked on that send path. An ESP SA is also required: the attacker can add one in a user+net namespace because xfrm_user_rcv_msg() only tests netlink_net_capable(CAP_NET_ADMIN) against the netns user_ns, or can send through an already-installed IPsec policy.\nUI:N - The attacker registers the buffer with IORING_REGISTER_BUFFERS, submits SEND_ZC, and if needed installs the XFRM state/policy; no other user must open a file, mount a volume, or otherwise act.\nS:U - esp_ssg_unref()'s extra put_page() corrupts host kernel page refcounts. That is a conventional local privilege escalation inside the same kernel, not a KVM/Xen guest-to-host or IOMMU-scope change.\nC:H - After the out-of-place transform, esp_ssg_unref() calls skb_page_unref() on every source sg page past skb->data. Those pages were attached by io_sg_from_iter() without a per-frag get_page(), so the extra put underflows the refcount of still-mapped IORING_REGISTER_BUFFERS pages and frees memory the attacker can read.\nI:H - The same extra skb_page_unref() frees registered-buffer pages that remain in the io_uring bvec and the attacker's userspace mapping, so the page can be reallocated and overwritten, giving an arbitrary write and control-flow hijack.\nA:H - The extra put_page() on managed frag pages underflows the page refcount and frees a page still in use, which oopses with bad page state even when the use-after-free is not turned into a full memory-corruption exploit."}]}],"affected":[{"product":"Linux","vendor":"Linux","defaultStatus":"unaffected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["net/ipv4/esp4.c","net/ipv6/esp6.c"],"versions":[{"version":"cac2661c53f35cbe651bef9b07026a5a05ab8ce0","lessThan":"26b6b14c7a0368e317a1e9fb5144ebe6f8d495cf","status":"affected","versionType":"git"},{"version":"cac2661c53f35cbe651bef9b07026a5a05ab8ce0","lessThan":"21697720ff43b8dfa25b8e8d9ca7f56f4597fc80","status":"affected","versionType":"git"}]},{"product":"Linux","vendor":"Linux","defaultStatus":"affected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["net/ipv4/esp4.c","net/ipv6/esp6.c"],"versions":[{"version":"4.11","status":"affected"},{"version":"0","lessThan":"4.11","status":"unaffected","versionType":"semver"},{"version":"7.2.6","lessThanOrEqual":"7.2.*","status":"unaffected","versionType":"semver"},{"version":"7.3-rc1","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":"4.11","versionEndExcluding":"7.2.6"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"4.11","versionEndExcluding":"7.3-rc1"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/26b6b14c7a0368e317a1e9fb5144ebe6f8d495cf"},{"url":"https://git.kernel.org/stable/c/21697720ff43b8dfa25b8e8d9ca7f56f4597fc80"}],"title":"esp: do not unref managed frag pages in esp_ssg_unref()","x_generator":{"engine":"bippy-1.2.0"}}}}