{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2025-39758","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2025-04-16T07:20:57.125Z","datePublished":"2025-09-11T16:52:27.598Z","dateUpdated":"2026-08-23T12:45:23.784Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-08-23T12:45:23.784Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\nRDMA/siw: Fix the sendmsg byte count in siw_tcp_sendpages\n\nEver since commit c2ff29e99a76 (\"siw: Inline do_tcp_sendpages()\"),\nwe have been doing this:\n\nstatic int siw_tcp_sendpages(struct socket *s, struct page **page, int offset,\n                             size_t size)\n[...]\n        /* Calculate the number of bytes we need to push, for this page\n         * specifically */\n        size_t bytes = min_t(size_t, PAGE_SIZE - offset, size);\n        /* If we can't splice it, then copy it in, as normal */\n        if (!sendpage_ok(page[i]))\n                msg.msg_flags &= ~MSG_SPLICE_PAGES;\n        /* Set the bvec pointing to the page, with len $bytes */\n        bvec_set_page(&bvec, page[i], bytes, offset);\n        /* Set the iter to $size, aka the size of the whole sendpages (!!!) */\n        iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);\ntry_page_again:\n        lock_sock(sk);\n        /* Sendmsg with $size size (!!!) */\n        rv = tcp_sendmsg_locked(sk, &msg, size);\n\nThis means we've been sending oversized iov_iters and tcp_sendmsg calls\nfor a while. This has a been a benign bug because sendpage_ok() always\nreturned true. With the recent slab allocator changes being slowly\nintroduced into next (that disallow sendpage on large kmalloc\nallocations), we have recently hit out-of-bounds crashes, due to slight\ndifferences in iov_iter behavior between the MSG_SPLICE_PAGES and\n\"regular\" copy paths:\n\n(MSG_SPLICE_PAGES)\nskb_splice_from_iter\n  iov_iter_extract_pages\n    iov_iter_extract_bvec_pages\n      uses i->nr_segs to correctly stop in its tracks before OoB'ing everywhere\n  skb_splice_from_iter gets a \"short\" read\n\n(!MSG_SPLICE_PAGES)\nskb_copy_to_page_nocache copy=iov_iter_count\n [...]\n   copy_from_iter\n        /* this doesn't help */\n        if (unlikely(iter->count < len))\n                len = iter->count;\n          iterate_bvec\n            ... and we run off the bvecs\n\nFix this by properly setting the iov_iter's byte count, plus sending the\ncorrect byte count to tcp_sendmsg_locked."}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H","baseScore":9.8,"baseSeverity":"CRITICAL"},"scenarios":[{"lang":"en","value":"AV:N - siw is an iWARP/RDMA transport running over TCP, and the overrun parameters (`read_size`, `source_to`/page offset, and GSO/FPDU segment size) come straight from a remote peer's RDMA READ Request via `siw_init_rresp()`; the same TX path serves network-facing kernel ULPs (nfsd/RDMA, ksmbd SMB Direct, nvmet-rdma) layered on soft-iWARP, and the out-of-bounds data is copied into the TCP skb and sent back over the network.\nAC:L - The attacker deterministically creates the overrun by choosing a non-page-aligned local offset with a transfer length ≥ PAGE_SIZE, which forces `size > bytes` on the very first iteration; `zcopy_tx` is on by default and READ.Responses are always non-signalled, so no race or unpredictable state is involved.\nPR:N - A remote iWARP peer needs only an established MPA/TCP connection to a soft-iWARP target — nvmet-rdma and NFS/RDMA targets accept connections with no transport-level authentication, and the ≥4 KB kmalloc-backed reply/READ-Response path is exercised immediately after connect.\nUI:N - The overrun happens entirely inside the kernel's RDMA transmit path while servicing a peer-initiated request; no local user has to open, mount, or click anything.\nS:U - The out-of-bounds access and its consequences are confined to the kernel's own memory and the affected RDMA connection, with no crossing of a VM, IOMMU, or other security-authority boundary.\nC:H - `iterate_bvec()` fabricates `struct page *` values from stack residue and copies memory from those arbitrary kernel addresses into the outgoing TCP stream, so kernel memory — up to ~64 KB per FPDU — is disclosed directly to the remote peer, an effectively arbitrary kernel read primitive.\nI:H - Kernel memory is substituted for legitimate RDMA payload delivered to the peer, silently corrupting data that NFS/RDMA, SMB Direct, and NVMe-oF clients commit to files and block devices, and the oversized return value underflows the `size_t bytes` counter, corrupting siw's send-length accounting and retry loop.\nA:H - Dereferencing fabricated `struct page` pointers via `kmap_local_page()` and memcpy'ing from unmapped addresses reliably oopses or panics the kernel — the crashes the fix commit reports — and the `bytes` underflow can wedge the transmit loop; the attacker can repeat this at will."}]}],"affected":[{"product":"Linux","vendor":"Linux","defaultStatus":"unaffected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/infiniband/sw/siw/siw_qp_tx.c"],"versions":[{"version":"a19a58ab12695abb91a4e8ecf62cc0bc5badefbe","lessThan":"b47c4b7cb78b6382c6d1af0bfc0d8218e4f445cc","status":"affected","versionType":"git"},{"version":"c2ff29e99a764769eb2ce3a1a5585013633ee9a6","lessThan":"5661fdd218c2799001b88c17acd19f4395e4488e","status":"affected","versionType":"git"},{"version":"c2ff29e99a764769eb2ce3a1a5585013633ee9a6","lessThan":"673cf582fd788af12cdacfb62a6a593083542481","status":"affected","versionType":"git"},{"version":"c2ff29e99a764769eb2ce3a1a5585013633ee9a6","lessThan":"42ebc16d9d2563f1a1ce0f05b643ee68d54fabf8","status":"affected","versionType":"git"},{"version":"c2ff29e99a764769eb2ce3a1a5585013633ee9a6","lessThan":"edf82bc8150570167a33a7d54627d66614cbf841","status":"affected","versionType":"git"},{"version":"c2ff29e99a764769eb2ce3a1a5585013633ee9a6","lessThan":"c18646248fed07683d4cee8a8af933fc4fe83c0d","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/infiniband/sw/siw/siw_qp_tx.c"],"versions":[{"version":"6.5","status":"affected"},{"version":"0","lessThan":"6.5","status":"unaffected","versionType":"semver"},{"version":"6.6.103","lessThanOrEqual":"6.6.*","status":"unaffected","versionType":"semver"},{"version":"6.12.43","lessThanOrEqual":"6.12.*","status":"unaffected","versionType":"semver"},{"version":"6.15.11","lessThanOrEqual":"6.15.*","status":"unaffected","versionType":"semver"},{"version":"6.16.2","lessThanOrEqual":"6.16.*","status":"unaffected","versionType":"semver"},{"version":"6.17","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":"6.5","versionEndExcluding":"6.6.103"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.5","versionEndExcluding":"6.12.43"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.5","versionEndExcluding":"6.15.11"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.5","versionEndExcluding":"6.16.2"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.5","versionEndExcluding":"6.17"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/b47c4b7cb78b6382c6d1af0bfc0d8218e4f445cc"},{"url":"https://git.kernel.org/stable/c/5661fdd218c2799001b88c17acd19f4395e4488e"},{"url":"https://git.kernel.org/stable/c/673cf582fd788af12cdacfb62a6a593083542481"},{"url":"https://git.kernel.org/stable/c/42ebc16d9d2563f1a1ce0f05b643ee68d54fabf8"},{"url":"https://git.kernel.org/stable/c/edf82bc8150570167a33a7d54627d66614cbf841"},{"url":"https://git.kernel.org/stable/c/c18646248fed07683d4cee8a8af933fc4fe83c0d"}],"title":"RDMA/siw: Fix the sendmsg byte count in siw_tcp_sendpages","x_generator":{"engine":"bippy-1.2.0"}}}}