The internet is broken right now.
49 critical CVEs without a fix in the last 48 hours. Somewhere a staging server is public. It always is. 119 more already have a fix that someone still has to apply.
-
CVE-2026-99383
In the Linux kernel, the following vulnerability has been resolved: smb: client: reclaim credits for decrypted reads with no matching mid When a decrypted SMB2 READ response carries a MessageId that matches no outstanding request, both the inline receive_encrypted_read() and the offloaded smb2_decrypt_offload() paths drop the frame without accounting the credits the server granted in it. The plaintext unknown-frame path in cifs_demultiplex_thread() already reclaims them with smb2_add_credits_from_hdr(), so encrypted reads leak credits that the plaintext path would not. This also happens for legitimate frames - e.g. a reply that arrives after the client has timed out and dropped the mid - so over a long-lived connection the client's credit count drifts down and can eventually starve it of the ability to issue requests. The frame is AEAD-authenticated, so the granted credits in the decrypted header come from the real server and are safe to apply. Reclaim the credits from the decrypted header before dropping the frame, and export smb2_add_credits_from_hdr() so the SMB2 receive paths can use it.
-
CVE-2026-99384
In the Linux kernel, the following vulnerability has been resolved: smb: client: fix OOB read of the resume name sent in TRANS2_FIND_NEXT2 cifs_fill_dirent() reads the fixed part of a directory entry and the name length it holds, or for SMB_FIND_FILE_UNIX scans the name for its terminating NUL, without knowing where the response ends. cifs_save_resume_key() calls it for the entry at the LastNameOffset the server chose, after every FindNext and after the FindFirst of a search rewind, and records the name it returns as the resume name of the search without bounding it; the next CIFSFindNext() copies up to PATH_MAX - 1 bytes from that name into its request and sends them to the server. cifs_filldir() bounds the name against the end of the response since commit f8cf09a53a0d ("smb: client: bound dirent name against end of SMB response in cifs_filldir"), but only after cifs_fill_dirent() has read the entry, and nxt_dir_entry() checks the entries it walks only against sizeof(FILE_DIRECTORY_INFO), for every level but SMB_FIND_FILE_INFO_STANDARD, and never sees the first entry of a response. Bounding where last_entry starts, as the preceding patch does, does not bound the read. A last entry that starts within the final bytes of the response still has its fixed part read past the end of the received data, and the name length it declares, a __le32 for every level but SMB_FIND_FILE_INFO_STANDARD and SMB_FIND_FILE_UNIX, is recorded without being checked against anything. CIFSFindNext() rejects only a length of PATH_MAX or more, then copies the name, up to 4095 bytes, into its request and sends it, and as the report below shows that copy can run past the end of the cifs_request allocation that holds the response. For SMB_FIND_FILE_UNIX the name is in addition scanned for its terminating NUL over up to PATH_MAX + 1 UTF-16 units, or PATH_MAX bytes, before any check. The FindNext that carries the name is only sent by the SMB1 dialect, which is not negotiated by default; reaching it requires an explicit vers=1.0 mount. With KASAN enabled, a response whose last entry declares a name longer than the bytes that follow it gives: ================================================================== BUG: KASAN: slab-out-of-bounds in CIFSFindNext+0x8ca/0x14c0 Read of size 4080 at addr ffff88807c823f98 by task ls/83 CPU: 0 UID: 0 PID: 83 Comm: ls Not tainted 7.3.0-rc4-00457-gf14572c203d5 #69 PREEMPT(lazy) Call Trace: <TASK> kasan_report+0xdf/0x1a0 kasan_check_range+0x10f/0x1e0 __asan_memcpy+0x23/0x60 CIFSFindNext+0x8ca/0x14c0 cifs_readdir+0xf51/0x29c0 iterate_dir+0x1c0/0x570 __x64_sys_getdents64+0x133/0x270 do_syscall_64+0x109/0x5d0 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> Allocated by task 83 on cpu 1 at 13.497119s: cifs_buf_get+0x36/0x90 smb_init+0x4f/0x110 CIFSFindNext+0xf7/0x14c0 cifs_readdir+0xf51/0x29c0 The buggy address belongs to the object at ffff88807c820000 which belongs to the cache cifs_request of size 16588 The buggy address is located 16280 bytes inside of allocated 16588-byte region [ffff88807c820000, ffff88807c8240cc) ================================================================== The read is the memcpy() that copies the recorded name into the next request, so those bytes go to the server. Pass the end of the response to cifs_fill_dirent() and have it reject an entry whose fixed part does not fit or whose name extends past the end, before either is used, so that both callers are covered; cifs_save_resume_key() computes the end the same way cifs_readdir() does. The name bound compares the bytes left between the name and the end with the name length rather than adding the length to the name pointer, since the length is a 32-bit value taken from the entry and the sum could wrap on 32-bit systems. cifs_fill_dirent() is also used on the SMB2 readdir path; there smb2_parse_query_directory() has already checked every entry against the end of the response, so a vali ---truncated---
-
CVE-2026-99385
In the Linux kernel, the following vulnerability has been resolved: smb: client: fix buffer overflow in SMB2_tcon() In SMB2_tcon(), unc_path is allocated with a fixed buffer size of MAX_SHARENAME_LENGTH * 2 bytes (672 bytes). However, the user-controlled UNC path (tree) length is passed to cifs_strtoUTF16(), which assumes the destination buffer is large enough and writes strlen(tree) + 1 UTF-16 code units. Commit 62a1a439e0fd ("cifs: clean up handling of unc= option") removed the length check on UNC mount options, allowing mount paths to exceed MAX_SHARENAME_LENGTH while retaining the fixed-size buffer allocation in SMB2_tcon(). Consequently, an oversized UNC string causes cifs_strtoUTF16() to write past the end of the heap object, resulting in a slab-out-of-bounds write. Furthermore, if a tree exceeds MAX_TREE_SIZE characters, strscpy() truncates it in cifs_tcon::tree_name, causing subsequent reconnects or failovers to fail. Fix this by first verifying that the length of tree does not exceed MAX_TREE_SIZE, returning -EINVAL if it does. Then, dynamically allocate unc_path based on the length of tree so that it safely accommodates all UTF-16 code units for paths within MAX_TREE_SIZE. [pc: remove unused MAX_SHARENAME_LENGTH]
-
CVE-2026-99386
In the Linux kernel, the following vulnerability has been resolved: smb: client: don't resurrect an exiting tcon from cached dir workers cfids_laundromat_worker() bumps tcon->tc_count for every expired open cfid and queues cached_dir_offload_close() without checking whether the tcon is already being torn down. Racing with the final cifs_put_tcon(), the tcon comes back from zero and gets torn down twice: T1 T2 cifs_put_tcon() --tc_count == 0, TID_EXITING tconInfoFree() free_cached_dirs() cancel_delayed_work_sync() cfids_laundromat_worker() ++tc_count == 1 queue_work(serverclose_wq) kfree(tcon->origin_fullpath) kfree(tcon) cached_dir_offload_close() close_cached_dir() [UAF] cifs_put_tcon() tconInfoFree() [double free] cached_dir_lease_break() also bumps tc_count without tc_lock, racing with the laundromat's locked update. Take the tcon reference through a helper that refuses once the tcon is TID_EXITING or its refcount has dropped to zero. Queued put_work and close_work then always hold a tcon reference, so tconInfoFree() can't run before they finish and free_cached_dirs() doesn't need to flush any workqueue. If the laundromat can't get a reference, put the cfid back on the dying list for free_cached_dirs() to free; its handle goes away with the tree connection. The lease break path finds the tcon on ses->tcon_list under cifs_tcp_ses_lock, so it can't see TID_EXITING, but it now takes tc_lock.
-
CVE-2026-99387
In the Linux kernel, the following vulnerability has been resolved: smb: client: bound POSIX lock data area by response length CIFSSMBPosixLock() receives a TRANS2 response whose server-provided DataOffset and DataCount are not required by validate_t2() to describe an interval inside the response. A server can therefore make the client read a cifs_posix_lock beyond the response buffer and return those kernel bytes through F_GETLK. Use smbCalcSize() to obtain the full response length after any multipart TRANS2 reassembly, then reject data areas that extend past it before dereferencing them.
-
CVE-2026-99372
In the Linux kernel, the following vulnerability has been resolved: NFSD: cap the number of listeners accepted in listener_set nfsd_nl_listener_set_doit() matches each requested listener against the existing set. The nested loop that does this is O(N * M), where N is the requested count and M is the existing count. The loop runs under sv_lock with bottom halves disabled. A userland request with a very large listener list can therefore spin in atomic context for a long time. Reject a request that carries more than NFSD_NL_LISTENER_MAX (1024) entries. The check goes in nfsd_nl_validate_listeners(), before the code takes any lock. The limit is far above any realistic configuration. This patch does not cap M. Only the message size bounded N; real sockets bound M. A listener_set result set is the requested set, so that path also holds M at the cap, but __write_ports_addxprt() adds two listeners per call and removes none, so repeated calls can push M past it. The worst case under sv_lock is therefore 1024 * M, plus 1024 nla_parse_nested() calls. Both interfaces require CAP_NET_ADMIN.
-
CVE-2026-99373
In the Linux kernel, the following vulnerability has been resolved: NFS/localio: detect a short read or write before the iterator has moved nfs_local_call_read() and nfs_local_call_write() issue a misaligned DIO as up to three segments and must stop at the first short one, or the next segment lands at the wrong file offset. They test for that by comparing the bytes transferred against iov_iter_count() of the segment, read after read_iter() or write_iter() has advanced the iterator, when the count is the residual rather than the request: before the call: count == expected after the call: count == expected - status The test is therefore status < expected - status, and it fires only when less than half of the segment was transferred. A short I/O that covers between 50% and 99% of a segment slips through: the loop moves on with ki_pos advanced by the short amount, the next segment is issued at the wrong offset, and the header's byte count is over-reported to the caller. A write that slips through also never sets NFS_CONTEXT_WRITE_SYNC, which is what makes the writes that follow a short one synchronous. Take the segment's count before issuing it and compare against that. This is the same defect that commit 250ec14932d5 ("nfsd: fix partial-write detection in nfsd_direct_write") fixed in NFSD's version of this loop.
-
CVE-2026-99374
In the Linux kernel, the following vulnerability has been resolved: NFSv4/flexfiles: Use the full 64-bit stripe_unit ff_layout_alloc_lseg() decodes stripe_unit as the 64-bit value the protocol defines and struct nfs4_ff_layout_segment stores it as one, but both consumers narrow it back to 32 bits: nfs4_ff_layout_calc_dss_id() divides with do_div(), which casts the divisor, and ff_layout_pg_test() copies it into a u32 first. A stripe_unit that does not fit is silently truncated, so the client stripes on a unit the server did not specify -- or divides by zero, if the low 32 bits happen to be clear. Divide by the full value, with div64_u64() and div64_u64_rem().
-
CVE-2026-99375
In the Linux kernel, the following vulnerability has been resolved: NFSv4/flexfiles: fail I/O to the data server once a lock is lost When a client loses a byte-range lock, because its lease expired and it may not recover the lock (nfs.recover_lost_locks=0) or because reclaim failed, it marks the lock state NFS_LOCK_LOST and fails further I/O by that lock owner. Otherwise an application that believes it holds a lock over a range is never told that the lock is gone, and can overwrite data that another client has since written under its own lock on that range. The NFSv4 READ/WRITE and file layout I/O paths apply this in rpc_call_prepare, where nfs4_set_rw_stateid() returns -EIO for a lost lock. The flexfiles data server I/O path does not: the NFSv3 data server path never checked, and the check on the NFSv4 path was removed when the data server stateid began to be taken from the layout. Fail read and write RPCs to the data server in rpc_call_prepare when the lock is lost. The data server path has no use for the stateid that nfs4_set_rw_stateid() selects, and selecting it looks up a delegation or open stateid on every I/O. Instead add nfs4_lock_context_lost(), which only tests NFS_LOCK_LOST, and check it alongside the open state. Its lookup, __nfs4_lookup_lock_state(), is split out of __nfs4_find_lock_state() and takes no reference: a lock state is only unlinked from state->lock_states under state->state_lock, so one found while holding that lock cannot be freed before it is released. Use it in nfs4_copy_lock_stateid() as well, which only reads the lock state under that lock. That drops a reference get and put from stateid selection for all NFSv4 I/O without changing the stateid selected. Terminating the RPC in rpc_call_prepare is not enough on its own. The done callback reads the -EIO as a data server connection error, forgets the device ID, and resets the I/O to pNFS (or, for an NFSv4 data server, possibly to the MDS). After a reset to pNFS a new LAYOUTGET is granted and rpc_call_prepare fails the I/O again, without end. The existing NFS_CONTEXT_BAD check does not loop, since the open state is then invalid and pnfs_update_layout() fails the reset I/O, but it still forgets the device ID. So when an RPC that was never sent fails and the open context is bad, the open state is invalid, or the lock is lost, complete it with -EIO without error handling. An RPC that was sent and failed is handled as before, so a real data server error is still reported to the MDS.
-
CVE-2026-99376
In the Linux kernel, the following vulnerability has been resolved: NFSv4/pnfs: Free the netid when draining a data-server address list nfs4_pnfs_ds_addr_free() releases all three of an address entry's allocations, but the four sites that drain a decoded list before it reaches a data server open-code the loop and free only da_remotestr and the entry, leaking da_netid. Both layout drivers drain that list when nfs4_pnfs_ds_add() returned a data server already in the cache, and again on their error exits. Lift the drain into nfs4_pnfs_ds_addr_list_free() so an address list has one owner. A striping mount reaches the cached case on nearly every GETDEVICEINFO, since many deviceIDs name the same data servers.
-
CVE-2026-99377
In the Linux kernel, the following vulnerability has been resolved: NFSv4/pnfs: bound the CB_NOTIFY_DEVICEID array count before allocating decode_devicenotify_args() hands the server's notification count straight to kmalloc_objs() as the array length, without checking it against the message that carried it. Bound it the way nfs4xdr.c bounds an attribute length, against xdr_stream_remaining().
-
CVE-2026-99378
In the Linux kernel, the following vulnerability has been resolved: pnfs/blocklayout: Complete a device upcall when the pipe is closed Once blkmapd has read a whole upcall, rpc_pipe_read() unlinks the message from every pipe list, so the purges in rpc_pipe_release() and rpc_close_pipes() no longer reach it. bl_upcall_ops supplies no .release_pipe callback, and bl_pipe_destroy_msg() returns without completing the upcall because a downcall is expected to follow. A daemon that exits between the read and the write therefore strands its waiter: bl_resolve_deviceid() sleeps in wait_for_completion() and never returns, and it holds nn->bl_mutex across that wait, so every later device resolution in the net namespace blocks behind it. Add a .release_pipe that completes an upcall blkmapd has consumed and left unanswered. Completing an upcall retires it, so a message the framework has already purged is no longer in flight and is not completed a second time here.
-
CVE-2026-99379
In the Linux kernel, the following vulnerability has been resolved: smb: client: validate SMB1 EA list bounds before parsing CIFSSMBQAllEAs() checks the server-provided EA list length by adding it to the response pointer. On 32-bit systems, a positive length can wrap the pointer and pass the comparison. The parser then walks outside the received response while reading the next FEA header. A response with a list length of 0x40000000 reproduces this as a native i386 page fault and kernel panic from listxattr() against an SMB1 server. Check that the EA list starts within the SMB byte area before reading its header. Keep its length unsigned and compare it with the bytes left in the response without pointer addition. SMB1 is not negotiated by default and requires an explicit vers=1.0 mount.
-
CVE-2026-99364
In the Linux kernel, the following vulnerability has been resolved: NFSD: Fix out-of-bounds read in the rpc_status dump nfsd_nl_rpc_status_get_dumpit() loads args->opcnt and args->ops separately, then walks ops[] before rechecking rq_status_counter. A COMPOUND that completes between the two loads runs nfsd4_release_compoundargs(), which zeroes opcnt and points ops back at the eight-entry inline array. A dump that already sampled an opcnt of 200 clamps it to the sixteen slots in rq_opnum, then indexes iops[0..15]. iops is the last member of struct nfsd4_compoundargs and rq_argp is allocated at exactly that size, so the walk runs off the end of the allocation. The trailing recheck discards the sampled data, but the read has already happened. NFSD_CMD_RPC_STATUS_GET carries no GENL_ADMIN_PERM, so an unprivileged local user can repeat the dump against a busy server until it lands in the window. Sample opcnt and ops into locals, then finish the counter recheck before dereferencing ops. An unchanged counter means both came from the same COMPOUND, where opcnt cannot exceed what ops holds.
-
CVE-2026-99365
In the Linux kernel, the following vulnerability has been resolved: nfsd: Fix id-to-name cache entry leak in idtoname_parse() idtoname_lookup() returns the id-to-name cache entry with a new reference. The name is parsed between the lookup and the update call, and when the parsed name is malformed (negative length or longer than IDMAP_NAMESZ), idtoname_parse() jumps to the error label without dropping the reference returned by idtoname_lookup(), leaking the cache entry. Release the reference with cache_put() before taking the error exit.
-
CVE-2026-99366
In the Linux kernel, the following vulnerability has been resolved: nfsd: don't modify a session slot when replaying its cached reply nfsd4_sequence() claims a session slot by setting NFSD4_SLOT_INUSE under nn->client_lock. A reply served from the slot's reply cache does not claim it, and that distinction lived only in cstate->status, which nfsd4_sequence() set to nfserr_replay_cache. Commit cc028a10a48c ("NFSD: Hoist status code encoding into XDR encoder functions") moved nfsd4_proc_compound()'s cstate->status assignment below the out: label, and the replay path's goto out was the one path that relied on skipping it. The test in nfsd4_sequence_done() therefore no longer identifies a replay, and every replay now stores its reply and clears NFSD4_SLOT_INUSE as though it owned the slot. NFSD4_SLOT_INUSE is the interlock: check_slot_seqid() rejects every sequence id for a slot that is in use, before replay_matches_cache() runs. A replay never sets it, so two replays can be in flight on the same slot at once. One can read slot->sl_cred under nn->client_lock while the other's completion frees and rebuilds it from the XDR encoder without that lock. Two completions can also collide with each other and release the same group_info twice. free_svc_cred() leaves cr_uid, cr_gid and cr_flavor intact, so the reader passes every earlier test in same_creds() and dereferences a NULL cr_group_info. From a 6.12.91 production server: BUG: kernel NULL pointer dereference, address: 0000000000000004 CPU: 63 UID: 0 PID: 39015 Comm: nfsd RIP: 0010:same_creds+0x38/0xa0 [nfsd] RDX: 0000000000000000 Call Trace: nfsd4_sequence+0x6a8/0x910 [nfsd] nfsd4_proc_compound+0x345/0x670 [nfsd] nfsd_dispatch+0x100/0x220 [nfsd] svc_process_common+0x311/0x700 [sunrpc] svc_process+0x131/0x1c0 [sunrpc] svc_recv+0x7ef/0x9c0 [sunrpc] nfsd+0xa3/0x100 [nfsd] Kernel panic - not syncing: Fatal exception Since v6.14 a live session's slot table can shrink, and the unowned store becomes a use-after-free write. nfsd4_sequence() defers the shrink while a slot is in use, but it tests NFSD4_SLOT_INUSE, which a replay does not set, so free_session_slots() can kfree() the slot the replay still holds in cstate->slot. Record the claim in cstate->slot_owned when nfsd4_sequence() accepts a request and test that in nfsd4_sequence_done(), so only the request that claimed the slot updates its cached reply and clears NFSD4_SLOT_INUSE. svc_generic_init_request() zeroes the compound response before each request, so the flag starts clear.
-
CVE-2026-99367
In the Linux kernel, the following vulnerability has been resolved: NFSD: Don't complete a cld upcall the daemon has not read cld_pipe_downcall() matches a reply to its upcall on the xid alone. A write that arrives before the daemon has read that upcall completes __cld_pipe_upcall() while its struct rpc_pipe_msg is still queued on the pipe. The waiter returns, its stack frame goes away, and the pipe keeps a list_head that points into it. The next rpc_queue_upcall() walks that pointer: BUG: KASAN: vmalloc-out-of-bounds in __list_add_valid_or_report Read of size 8 at addr ffffc90007027950 by task syz.0.96/13066 __list_add_valid_or_report+0x6a/0x130 rpc_queue_upcall cld_pipe_upcall nfsd4_cld_grace_start nfsd4_cld_tracking_init The pipe is mode 0600 in rpc_pipefs, so the trigger is a broken or hostile nfsdcld rather than an unprivileged task. Record whether the daemon has consumed the whole message and complete only when it has. cld_pipe_destroy_msg() runs on every path by which the pipe releases a message, so it can maintain that flag under cn_lock. nfsdcld sends its -EINPROGRESS downcalls only after reading the upcall, so the new check does not break the GraceStart record stream.
-
CVE-2026-99368
In the Linux kernel, the following vulnerability has been resolved: NFSD: Do not send CB_RECALL_ANY to NFSv4.0 clients deleg_reaper() sends CB_RECALL_ANY to every ACTIVE client holding delegations, but CB_RECALL_ANY is an NFSv4.1 operation. An NFSv4.0 client's callback service accepts only CB_GETATTR and CB_RECALL, so it replies OP_ILLEGAL. The decoder maps the unexpected opnum to -EIO, and nfsd4_cb_done() marks the client's callback channel down. Nothing brings the channel back. nfsd4_run_cb_work() sets NFSD4_CB_UP only for a minor version above zero, and the only nfsd4_probe_callback() call site an NFSv4.0 client reaches is nfsd4_setclientid_confirm(). One visit from the reaper therefore leaves the channel marked down until the client re-establishes its clientid. RENEW then returns NFS4ERR_CB_PATH_DOWN for as long as the client holds delegations. nfsd4_cb_channel_good() stops returning true, so the client is granted no further delegations. Skip clients at minor version zero.
-
CVE-2026-99369
In the Linux kernel, the following vulnerability has been resolved: NFSD: Destroy a recalled delegation the client does not hold A client that answers CB_RECALL with NFS4ERR_BADHANDLE or NFS4ERR_BAD_STATEID has no record of the delegation, so the FREE_STATEID that clears it from cl_revoked never arrives. Every later SEQUENCE reply carries SEQ4_STATUS_RECALLABLE_STATE_REVOKED, and the client loops issuing TEST_STATEID. Destroy such a delegation when it is reaped rather than revoking it onto cl_revoked. RFC 8881 Section 20.2.4 completes the recall at the reply when its status is neither NFS4_OK nor NFS4ERR_DELAY, so a rejected recall leaves nothing to revoke. An administrative revoke keeps that path, since NFS4ERR_ADMIN_REVOKED reports it. A destroyed stateid returns NFS4ERR_BAD_STATEID instead of NFS4ERR_DELEG_REVOKED. A client that rejects the recall but still holds the delegation gets no notice that its state was revoked. CB_RECALL can outrun the reply that granted the delegation, so honor a rejection only once the client has seen that grant. Per RFC 8881 Section 2.10.6.3, retirement of the slot that carried the grant is that proof; retry until then, and revoke when the retries lapse.
-
CVE-2026-99370
In the Linux kernel, the following vulnerability has been resolved: NFSD: copy SETXATTR data from all XDR buffer segments nfsd4_vbuf_from_vector() manually copies the head followed by whole pages starting at offset zero. This assumes that all remaining data starts at the beginning of the page array and that no data is stored in the tail. However, nfsd4_decode_setxattr() passes an xdr_buf returned by xdr_stream_subsegment(). Such a sub-buffer can have a nonzero page_base and can span its page and tail segments. Ignoring that layout can copy the wrong bytes into the value passed to vfs_setxattr(). Use read_bytes_from_xdr_buf(), which observes the complete xdr_buf layout.
-
CVE-2026-99371
In the Linux kernel, the following vulnerability has been resolved: nfsd: clear XDR padding in GETXATTR replies nfsd4_vbuf_to_stream() uses a raw memcpy() for the first fragment of an attribute value. xdr_reserve_space() aligns the reservation but does not initialize the extra bytes. If the complete value fits in that fragment, the loop that clears padding for the final fragment is never entered. A remote client reading an attribute whose length is not a multiple of four can therefore receive one to three stale bytes from the response page. Encode the first fragment with xdr_encode_opaque_fixed(), matching the final-fragment path and ensuring that any XDR padding is zeroed.
-
CVE-2026-99356
In the Linux kernel, the following vulnerability has been resolved: SUNRPC: Do not credit control-record octets to the RPC stream svc_tcp_sock_recv_cmsg() receives up to two octets into a local buffer, and returns that count for any record type other than TLS_RECORD_TYPE_ALERT. Nothing reached the caller's buffer, but svc_tcp_read_marker() adds the count to sk_tcplen and svc_tcp_read_msg()'s caller adds it to sk_datalen. The RPC stream advances over octets it never received. The fragment marker is assembled from stale sk_marker octets. The message body comes from pages nothing wrote. A conforming client reaches this. RFC 8446 Section 4.6.3 lets either peer send KeyUpdate once it has sent its Finished, and svcsock has no rekey path. kTLS leaves the partially consumed record on ctx->rx_list, so the body drains two octets per svc_tcp_recvfrom() call. Each pair is credited the same way. Return -EAGAIN for a record that is not an alert. That is what svc_tcp_sock_process_cmsg()'s default arm returned before the receive moved into a local buffer.
-
CVE-2026-99357
In the Linux kernel, the following vulnerability has been resolved: SUNRPC: Copy the deferred RPC Call from the head buffer svc_defer() snapshots the RPC Call for replay once a pending cache upcall resolves. It copies the Call starting (rq_arg.len - head[0].iov_len) bytes before head[0].iov_base, a back-up that once covered a transport header. No transport has had one since commit 983084b2672c ("SUNRPC: Remove svc_rqst::rq_xprt_hlen"), so the two lengths are meant to be equal. An RPCSEC_GSS Call breaks that assumption. Unwrapping the integrity service truncates rq_arg.len by the length of the trailing checksum, and unwrapping the privacy service leaves rq_arg.len short by an amount the client sets with trailing bytes after the wrap token. Neither adjusts head[0].iov_len. Both lengths are size_t, so the subtraction underflows and the copy starts past head[0].iov_base. Every deferred integrity Call replays a shifted snapshot. A deferred replay skips GSS verification, so a shifted snapshot runs as an authenticated Call. With the privacy service the offset is client-chosen. A forged Call placed in that window, naming another user's GSS context handle, runs under that identity. Copy the Call from head[0].iov_base. Refuse to defer a Call whose length runs past the head buffer, so the copy cannot read out of bounds.
-
CVE-2026-99358
In the Linux kernel, the following vulnerability has been resolved: nfsd: zero NFSv4 COMPOUND tag padding nfs4svc_encode_compoundres() copies the tag bytes into reserved XDR space and skips directly to the aligned end of the field. xdr_reserve_space() rounds the reservation up but does not initialize the padding bytes. A remote client can choose a tag length that is not a multiple of four, causing one to three stale bytes from the response page to be returned in the COMPOUND reply. Use xdr_encode_opaque_fixed() to copy the tag and clear its XDR padding.
-
CVE-2026-99359
In the Linux kernel, the following vulnerability has been resolved: NFSD: validate transport name in listener_set before serv creation nfsd_nl_listener_set_doit() holds nfsd_mutex for the whole listener teardown and rebuild. The code checks NFSD_A_SOCK_TRANSPORT_NAME for presence only, and not for content. An arbitrary name therefore reaches svc_xprt_create_from_sa(). There, a name that matches no registered class calls request_module("svc%s", name). That call is a TASK_KILLABLE usermode helper upcall, and it runs under nfsd_mutex. Check the name against the classes that NFSD can create: tcp, udp and rdma. The check goes in nfsd_nl_validate_listeners(), which runs before the code takes nfsd_mutex. The rejection names the offending attribute through extack, since -EPROTONOSUPPORT on its own does not say which entry carried the bad name. This narrows the upcall. It does not remove it. NFSD accepts "rdma" without a condition, so on a kernel that does not build svcrdma the name still reaches request_module("svcrdma") under nfsd_mutex. That is necessary for the modular case, where the autoload is legitimate.
-
CVE-2026-99360
In the Linux kernel, the following vulnerability has been resolved: NFSD: Set nn->cld_net before registering the cld pipe rpc_mkpipe_dentry() instantiates the pipe's dentry before it returns, so the cld pipe can be opened as soon as nfsd4_cld_register_net() has run. __nfsd4_init_cld_pipe() does not store the new cld_net in nn->cld_net until after that call. cld_pipe_downcall() reaches the cld_net only through that pointer and takes cn_lock without checking it, so a write() that lands in the window oopses. Store nn->cld_net once @cn is fully initialized and before the pipe is registered, and clear it again when registration fails.
-
CVE-2026-99361
In the Linux kernel, the following vulnerability has been resolved: nfsd: propagate SETXATTR value decode errors nfsd4_decode_setxattr() stores the status returned while assembling a non-empty attribute value, but then unconditionally returns nfs_ok. An allocation failure can therefore leave a NULL value buffer paired with a non-zero length and allow compound processing to continue. Return the decoder status so that the operation stops on the original error.
-
CVE-2026-99362
In the Linux kernel, the following vulnerability has been resolved: nfsd: preflight SEQUENCE replies before accepting a slot nfsd4_sequence() narrows the reply buffer to the session's cached reply limit before it accepts the slot seqid. A client may negotiate ca_maxresponsesize_cached down to NFSD_MIN_HDR_SEQ_SZ, and nfsd4_alloc_slot() then gives every slot a zero-length sl_data[]. A COMPOUND tag can fill that narrowed buffer until it holds the SEQUENCE opcode but not the status word that follows. nfsd4_encode_operation() returns without running nfsd4_encode_sequence(), so cstate.data_offset stays zero. It leaves op->status at nfs_ok as well, so the COMPOUND is treated as having succeeded. nfsd4_store_cache_entry() declines to cache a lone SEQUENCE that returned an error. That test reads the status the operation reported, so it passes here. The copy starts at offset zero and takes the whole reply, RPC and COMPOUND headers included, into the zero-length sl_data[]. The COMPOUND tag is copied along with it, so the client picks most of the bytes written past the end of the slot: BUG: KASAN: slab-out-of-bounds in read_bytes_from_xdr_buf+0x1bc/0x390 Write of size 80 at addr ffff888003a549cd by task kunit_try_catch/24 __asan_memcpy+0x38/0x60 read_bytes_from_xdr_buf+0x1bc/0x390 nfsd4_sequence_done+0x5b0/0x810 nfs4svc_encode_compoundres+0x1bf/0x240 Check that the fixed-size SEQUENCE result, plus room for a following operation's error status, fits the negotiated limit before narrowing the buffer and consuming the slot seqid. The slot and its reply cache are left unchanged, as RFC 8881 Section 2.10.6.1.2 requires of an error returned from SEQUENCE.
-
CVE-2026-99363
In the Linux kernel, the following vulnerability has been resolved: NFSD: Map flex file layout IDs through the request's user namespace nfsd4_ff_proc_layoutget() and nfsd4_ff_encode_layoutget() translate the file's owner and group with init_user_ns, but every other identity nfsd places on the wire goes through nfsd_user_namespace(). When the transport carries a credential from a non-initial user namespace, the flex file layout reports host-global IDs. The client copies those IDs into the AUTH_SYS credential it presents to the data server, and svcauth_unix_accept() resolves that credential in the transport's namespace, so data server I/O runs under an identity unrelated to the file's owner. Switching to the request's namespace introduces a second hazard. from_kuid() returns (uid_t)-1 when the target namespace has no mapping for the owner, and the IOMODE_READ arm adds one to that result to derive an identity for which the data server denies writes. The addition would wrap to zero, handing the client uid 0 instead of an identity distinct from the owner. Translate both IDs in nfsd4_ff_proc_layoutget(), which has the svc_rqst, and carry the wire values in struct pnfs_ff_layout. from_kuid_munged() substitutes overflowuid for an unmapped owner and thus never returns (uid_t)-1, so the increment cannot wrap to zero.
-
CVE-2026-99348
In the Linux kernel, the following vulnerability has been resolved: afs: Fix afs to abort the rxrpc call on send error Fix afs_send_empty_reply() and afs_send_simple_reply() to always try to abort the rxrpc call rather than just aborting on -ENOMEM and otherwise abandoning it. If the call is already complete due to network failure or a received abort, this will do nothing. Also make afs_make_call() always abort on send error; again, it does nothing if the rxrpc call is already dead.
-
CVE-2026-99349
In the Linux kernel, the following vulnerability has been resolved: rxrpc: Fix lack of short-send handling in rxrpc_kernel_send_data() Fix rxrpc_kernel_send_data() to loop around if it detects a short send. David Laight suggested doing it here rather than wrapping all the calls in loops. Further, remove the len argument and use the iterator count instead and return 0 on success, not the amount copied. Note this is also a prerequisite for changing the way rxrpc_send_data() works to return a short send rather than an error if some data was buffered.
-
CVE-2026-99350
In the Linux kernel, the following vulnerability has been resolved: rxrpc: Fix update of call->tx_pending without holding lock Currently, rxrpc_send_data() updates call->tx_pending just before it returns - but it won't be holding the call->user_mutex when it does this if a wait was interrupted by a signal. This would allow a parallel sendmsg() to race. Further, both the callers of rxrpc_send_data() call it with the lock held, and then it returns an indication through the parameter list to say whether it has dropped the lock or not - after which the callers both just drop the lock if it's still held. Fix this by: (1) Moving the release of call->user_mutex down into rxrpc_send_data() and get rid of the indicator parameter. This makes it easier to see where the lock is held. (2) After waiting, if the attempt to reacquire the mutex is interrupted, just return directly there rather than going to out_unlock (3) Restricting the txb variable to inside the buffering loop and leaving ->tx_pending set until we've queued the buffer. Note that there's a slight change in behaviour in that wait_for_space failure now doesn't check for completion because it doesn't hold the call user_mutex. The caller, however, should re-issue the send and pick up any error at a second attempt.
-
CVE-2026-99351
In the Linux kernel, the following vulnerability has been resolved: sunrpc: treat empty auth.unix.gid replies as negative entries When rpc.mountd cannot resolve a uid (getpwuid() or getgrouplist() failure, e.g. while winbind or sssd is briefly unreachable), it answers the auth.unix.gid upcall with zero groups. unix_gid_parse() installs that as a valid positive entry, and svcauth_unix_set_client() then replaces the credential's group list with the empty one on every request, RPCSEC_GSS included via svcauth_gss_set_client(). One failed lookup strips that uid of all supplementary groups on every export for up to mountd's configured TTL (30 minutes by default), long after the NSS backend has recovered. mountd cannot send an empty list for a successful lookup, since getgrouplist(3) always includes at least the user's primary group, so a zero-group reply can only mean the lookup failed. Record it as a negative entry: unix_gid_find() then returns -ENOENT and svcauth_unix_set_client() keeps the groups the RPC credential already carries. This is the fallback that commit 3fc605a2aa38 ("[PATCH] knfsd: allow the server to provide a gid list when using AUTH_UNIX authentication") promised when no answer is available, and the same state try_to_negate_entry() already creates when no listener holds the channel open.
-
CVE-2026-99352
In the Linux kernel, the following vulnerability has been resolved: SUNRPC: trace an accepted transport before publishing it svc_handle_xprt() publishes a newly accepted transport before tracing it. svc_add_new_temp_xprt() places the transport on sv_tempsocks and clears XPT_BUSY, so another nfsd thread can dequeue it at once. If the peer has already closed the connection, that thread runs svc_delete_xprt(), which drops the list reference, the only reference the new transport holds. trace_svc_xprt_accept() then reads the transport's addresses and network namespace from freed memory. Before publication the accepting thread owns the transport exclusively: XPT_BUSY is still set from svc_xprt_init() and the transport is on no list. Emit the tracepoint there. [ cel: trace before publication instead of taking an extra reference ]
-
CVE-2026-99353
In the Linux kernel, the following vulnerability has been resolved: SUNRPC: Reject a socket that already has an svc_sock attached Writing the same socket descriptor to /proc/fs/nfsd/portlist twice attaches a second svc_sock to one socket. svc_setup_socket() saves the socket's callbacks before installing its own, so the second attach records svc_write_space() as the old write_space callback. svc_udp_init() invokes that callback by way of svc_sock_setbufsize(), and svc_write_space() then calls itself until the kernel stack is exhausted: BUG: TASK stack guard page was hit at ffffc900037d7ff8 svc_write_space+0x90/0x2b0 net/sunrpc/svcsock.c:429 svc_write_space+0xe6/0x2b0 net/sunrpc/svcsock.c:430 ... 700 more ... svc_sock_setbufsize+0x18d/0x220 net/sunrpc/svcsock.c:386 svc_udp_init net/sunrpc/svcsock.c:854 [inline] svc_setup_socket+0xb2f/0x1090 net/sunrpc/svcsock.c:1498 svc_addsock+0x2fd/0x760 net/sunrpc/svcsock.c:1547 __write_ports_addfd fs/nfsd/nfsctl.c:742 [inline] write_ports+0xa5b/0xcc0 fs/nfsd/nfsctl.c:861 nfsctl_transaction_write+0x106/0x1a0 fs/nfsd/nfsctl.c:112 svc_data_ready() and svc_tcp_state_change() chain through their saved callbacks the same way, so a TCP descriptor added twice recurses on the next incoming segment instead. Reaching any of this takes a writer on portlist, and the nfsd filesystem sets no FS_USERNS_MOUNT, so the reproducer needs CAP_SYS_ADMIN in the initial user namespace. Reject a socket that already carries sk_user_data. svc_setup_socket() overwrites that field unconditionally, so a socket some other consumer has claimed is one NFSD would corrupt whether or not the callbacks recurse.
-
CVE-2026-99354
In the Linux kernel, the following vulnerability has been resolved: sunrpc: pin gss module across auth_domain RCU free A gss auth_domain stores a raw pointer to svcauthops_gss in dom->flavour, and its release queues svcauth_gss_domain_release_rcu() via call_rcu(). Both live in auth_rpcgss.ko text, and nothing pins the module while a domain referencing it is reachable. rmmod auth_rpcgss can therefore unmap that text while an NFSD export still holds a reference to the domain. The export cache is one such holder. When the export is finally released, the domain reference drops from module text: svc_export_put() auth_domain_put(exp->ex_client) auth_domain_release() dom->flavour->domain_release(dom) /* into module text */ svcauth_gss_domain_release() call_rcu(&dom->rcu_head, svcauth_gss_domain_release_rcu) Once auth_rpcgss is unloaded, the domain_release call lands in unmapped memory. The rcu_barrier() in exit_rpcsec_gss() does not help: it waits only for callbacks already queued, and this one is not queued until the export goes away. NFSD's NFSv4 objects import symbols from auth_rpcgss.ko, so only a server built without CONFIG_NFSD_V4 can unload auth_rpcgss while exports still name a gss domain. Take a module reference in svcauth_gss_register_pseudoflavor() before publishing the domain and drop it at the end of svcauth_gss_domain_release_rcu(). The rcu_barrier() then holds the module image until this callback has returned. [ cel: rewrote description ]
-
CVE-2026-99355
In the Linux kernel, the following vulnerability has been resolved: SUNRPC: fix oversized GSS proxy token copy gss_read_proxy_verf() allocates one page per token chunk, but it copies the entire linear head of an RPC request into in_token->pages[0]. A UDP datagram can leave more than PAGE_SIZE bytes in the linear head, so a large RPCSEC_GSS INIT token overwrites following pages. Copy the linear head in page-sized chunks before switching to the request page array. The linear head itself can span multiple pages.
-
CVE-2026-99340
In the Linux kernel, the following vulnerability has been resolved: KVM: x86: Take SRCU in kvm_zap_gfn_range() kvm_zap_gfn_range() walks memslots and rmaps and may drop mmu_lock to reschedule. Callers that do not already hold kvm->srcu (or slots_lock) can race with memslot deletion: synchronize_srcu_expedited() does not wait, kvm_free_memslot() frees the old slot and its rmap, and the zap resumes on freed memory. The VFIO noncoherent-DMA path hits this by zapping the entire GPA space without SRCU. KASAN reported: BUG: KASAN: vmalloc-out-of-bounds in slot_rmap_walk_next+0x82/0x1c0 Read of size 8 at addr ffffc900005c1008 Call Trace: slot_rmap_walk_next+0x82/0x1c0 __kvm_rmap_zap_gfn_range+0x17a/0x280 kvm_zap_gfn_range+0x2a6/0x6a0 kvm_vfio_set_attr+0x576/0x770 kvm_device_ioctl+0x1ff/0x3b0 Take SRCU inside kvm_zap_gfn_range() so every caller is covered. Nesting with an existing kvm->srcu critical section is fine; the helper uses a local index. Drop the now-redundant SRCU pair from __kvm_set_or_clear_apicv_inhibit().
-
CVE-2026-99341
In the Linux kernel, the following vulnerability has been resolved: KVM: x86: Request event processing for exception VM-Exits Request event processing whenever kvm_queue_exception_vmexit() queues an exception VM-Exit. Most callers reach the helper through kvm_multiple_exception(), which already sets KVM_REQ_EVENT, but nested asynchronous page faults call it directly. Without another event request, vcpu_enter_guest() skips kvm_check_and_inject_events() and re-enters L2. A subsequent VM-Exit can clear the pending exception while reconstructing vectoring state, losing the synthetic #PF VM-Exit. The APF reason remains PAGE_NOT_PRESENT and a later regular #PF can consume the stale value. Setting KVM_REQ_EVENT in the common helper ensures that L1 receives the queued VM-Exit before KVM re-enters L2.
-
CVE-2026-99342
In the Linux kernel, the following vulnerability has been resolved: KVM: x86/xen: Enforce 4-byte alignment of vcpu_info registration Xen's map_guest_area() enforces that vcpu_info is aligned to sizeof(xen_ulong_t). KVM has no such check, allowing a guest to register vcpu_info at an arbitrary byte alignment. Enforce unconditional 4-byte alignment regardless of the current shinfo mode. This is sufficient because subsequent commits ensure that all locked atomic operations on vcpu_info fields use at most 32-bit accesses. Return -ENXIO on failure, matching Xen's map_guest_area() behaviour for unaligned requests. Originally observed in review of an unrelated patch: https://lore.kernel.org/all/[email protected]/
-
CVE-2026-99343
In the Linux kernel, the following vulnerability has been resolved: KVM: x86/mmu: Bug the VM if KVM attempts to walk more levels than the MMU has Extend the "EFER.LMA && !CR4.PAE" check, which exists largely to guard against KVM configuring a paging32 MMU with more than 2 levels of paging, with a very explicit check for exactly that: that KVM isn't trying to walk more levels of paging than the MMU template provides. I.e. harden KVM against all bugs that would cause KVM to generates accesses beyond the bounds of guest_walker's arrays, regardless of how KVM ended up with the misconfigured MMU. Note, don't use w->cpu_role.base.level directly as the paging64 template only provides two levels of page tables for PAE paging on 32-bit hosts, and handles the third level by manually emulating the PDPTR access.
-
CVE-2026-99344
In the Linux kernel, the following vulnerability has been resolved: KVM: nSVM: Ignore EFER.LMA if EFER.LME=0 when preparing L2 state Force EFER.LMA=0 if EFER.LME=0 when preparing L2 state for VMRUN, i.e. mimic real hardware's behavior of ignoring EFER.LMA if EFER.LME=0. VMRUN unfortunately allows the nonsensical combination, i.e. doesn't fail, but KVM itself has an invariant EFER.LMA can be set et if and only if EFER.LME is set. Breaking that invariant can lead to a variety of issue, particularly in MMU code that keys off EFER.LMA when determining whether to emulate/virtualization 4/5-level paging versus PAE paging.
-
CVE-2026-99345
In the Linux kernel, the following vulnerability has been resolved: KVM: nSVM: Reject KVM_SET_NESTED_STATE if L1 has EFER.LMA=1 && EFER.LME=0 Reject KVM_SET_NESTED_STATE if the incoming L1 host state has what is effectively an impossible EFER combination of LMA=1 but LME=0, i.e. if the state says long mode is active but not enabled. Unlike VMX, SVM doesn't have an explicit consistent check for the illegal combination; presumably hardware simply ignores EFER.LMA if EFER.LME=0. Unfortunately, KVM doesn't ignore EFER.LMA in this case and consumes the illegal state when constructing the shadow MMU for L2. E.g. if userspace also clears CR4.PAE, then kvm_calc_cpu_role() will compute a role with 4 or 5 levels of paging, but shadow_mmu_init_context() will wire up the MMU to use the paging32 template, which maxes out its levels at 2. Note, the "real badness" is effectively the same as what happened with the nVMX bug fixed by commit 112e66017bff ("KVM: nVMX: add missing consistency checks for CR0 and CR4"). Unfortunately, the sanity check added by commit 72e2fb24a0b0 ("KVM: x86/mmu: Bug the VM if a vCPU ends up in long mode without PAE enabled") doesn't work for this case, since L2 state is active at the time of the page fault, but it's L1 that has the bad state.
-
CVE-2026-99346
In the Linux kernel, the following vulnerability has been resolved: KVM: SVM: Trigger new ASID allocation in both VMCBs on pCPU switch When the pCPU where the VMCB was most recently used is switched, reset the ASID generation in both VMCBs, triggering new ASID allocation for the immediate VMRUN as well as the next VMRUN on the other VMCB. The ASID is shared between vmcb01 and vmcb02, and gets flushed on every nested transition. However, since pCPU tracking is done per VMCB, it is possible for one VMCB to allocate a new ASID when migrated to a new pCPU, and then the other VMCB reuses that ASID on the old pCPU. This can result in the same ASID being used by multiple vCPUs on the old pCPU. Example scenario: - vCPU runs on pCPU A, vmcb01 is active, asid=1. - vCPU migrates to pCPU B, vmcb01 pCPU changes, new asid=2. - Another vCPU runs on pCPU A and allocates asid=2 as well. - vCPU migrates back to pCPU A, and then switches to vmcb02 before it runs again with vmcb01. - No pCPU switch is detected for vmcb02, so VMRUN is done with asid=2. - Two vCPUs end up using asid=2 on the same pCPU. Keep the VMCB dirtying to the active VMCB only. Clean bits are tracked by a pCPU for each VMCB, so do not unnecessarily dirty a VMCB if its pCPU does not change. Additionally, initialize the tracked pCPU for vmcb02 to -1 on nested enablement as hardening, so that a new ASID allocation is always triggered when nested is disabled and re-enabled. No performance regression was noticed when overcommitting L1 vCPUs in L0 (to force rescheduling), pinning L1 <-> L2 vCPUs, and running CPUID in a tight loop bouncing between 2 vCPUs in L2. An alternative (and perhaps more proper) fix would be tracking the ASID per-VMCB instead (e.g. [1]). However, that's a more involved change, and it would result in having different ASIDs for L1 and L2 without actually properly maintaining them. It would probably work because all TLB flushes target the current VMCB, and the other VMCB is always flushed on nested transitions, but the code ends up in an arguably more fragile state. Punt a proper clean fix to an incoming (and overdue) overhaul of SVM's ASID usage [2]. [1] https://lore.kernel.org/lkml/[email protected] [2] https://lore.kernel.org/kvm/[email protected]
-
CVE-2026-99347
In the Linux kernel, the following vulnerability has been resolved: ALSA: x86: hdmi: fix chmap array missing terminator entry had_build_channel_allocation_map() allocates a single struct snd_pcm_chmap_elem via kzalloc() and stores it as intelhaddata->chmap->chmap. struct snd_pcm_chmap.chmap is expected to be a zero-terminated array (see snd_pcm_std_chmaps[]), but had_register_chmap_ctls() never overrides the TLV callback installed by snd_pcm_add_chmap_ctls(), so the generic pcm_chmap_ctl_tlv() from ALSA core is left in place. pcm_chmap_ctl_tlv() is reachable by any local user via SNDRV_CTL_IOCTL_TLV_READ on the "Playback Channel Map" control and walks the array with "for (map = info->chmap; map->channels; map++)". Since the driver's single element is never terminated, this walks past the allocation. The same pointer is also freed on hot-unplug and rebuilt on hot-plug under intelhaddata->mutex, while the TLV callback runs under card->controls_rwsem without that mutex, so the walk can also race a concurrent free of the array. Fix this by allocating a second, zeroed element so the array is properly terminated, and by dropping the inherited TLV callback entirely at registration time, since the driver never implements one of its own.
-
CVE-2026-99332
In the Linux kernel, the following vulnerability has been resolved: HID: core: Fix out-of-bounds left shift by 32 in s32ton() Syzbot reported an out-of-bounds shift (32-bit integer shifted left by 32) in the s32ton() routine in the HID core. This routine converts a signed 32-bit integer to a signed n-bit integer, and it caps the value of n at 32. But when n == 32 there's no need to do anything; it can simply return the input value. This will avoid calculating (1 << n) when n is 32, which was the cause of the invalid shift.
-
CVE-2026-99333
In the Linux kernel, the following vulnerability has been resolved: HID: amd_sfh: Use report type from request node Use report type from request node while sending feature and input reports to HID core. The cli_data report type array is never assigned and feature reports are sent as input reports.
-
CVE-2026-99334
In the Linux kernel, the following vulnerability has been resolved: KVM: PPC: Book3S HV: Add preempt_disable() around virtual-mode HPTE bit-lock users kvmppc_hv_find_lock_hpte() requires virtual-mode callers to run with preemption disabled, because it can return with HPTE_V_HVLOCK still held until the caller later unlocks the HPTE. Existing virtual-mode callers in book3s_64_mmu_hv.c already follow that rule. Two virtual-mode call paths do not: 1. kvmppc_handle_exit_hv() calls kvmppc_hpte_hv_fault() for hash-mode data-side and instruction-side faults after guest exit, with preemption already enabled. kvmppc_hpte_hv_fault() calls kvmppc_hv_find_lock_hpte() directly. 2. kvmppc_pseries_do_hcall() executes virtual-mode handlers for HPT hcalls via kvmppc_pseries_do_hpt_hcall() with preemption enabled. The handlers for H_ENTER, H_REMOVE, H_READ, H_CLEAR_MOD, H_CLEAR_REF, H_PROTECT, and H_BULK_REMOVE spin on try_lock_hpte() or lock_rmap(), and H_ENTER also reaches kvmppc_do_h_enter(), which uses arch_spin_lock() on kvm->mmu_lock. That raw lock choice is intentional because kvmppc_do_h_enter() is also used by real-mode callers, so the correct fix is to establish the proper preemption context at the virtual-mode caller boundary. If a vCPU thread is preempted while holding HPTE_V_HVLOCK, any other thread on the same CPU trying to acquire the same bit-lock can spin indefinitely. The same applies to lock_rmap() users and to the H_ENTER path while holding the raw mmu_lock. Fix this by adding preempt_disable()/preempt_enable() pairs around the two kvmppc_hpte_hv_fault() call sites in kvmppc_handle_exit_hv(), and around the kvmppc_pseries_do_hpt_hcall() invocation in kvmppc_pseries_do_hcall(). Targeting only the HPT hcall helper avoids wrapping non-HPT hcalls (such as H_RTAS, H_CONFER, H_REGISTER_VPA) that perform page faults, blocking memory allocations, or context switches.
-
CVE-2026-99335
In the Linux kernel, the following vulnerability has been resolved: KVM: guest_memfd: Establish memslot<=>guest_memfd bindings *after* memslot is ready Wait to bind a memslot to a guest_memfd instance until *after* the memslot is fully prepared, as creating the binding in guest_memfd will effectively expose the memslot to readers. As pointed out by Sashiko, binding the memslot before it's ready to be exposed to the rest of the world can break various memslot assumption and rules. E.g. x86 could observe a NULL rmap pointer if a PUNCH_HOLE hit the guest_memfd after the binding was created, but before KVM made it through kvm_prepare_memory_region().
-
CVE-2026-99336
In the Linux kernel, the following vulnerability has been resolved: KVM: guest_memfd: take the invalidate lock when unbinding a dying file kvm_gmem_unbind() takes mapping->invalidate_lock only if the get_file_active() succeeds. When the guest_memfd file is dying, its reference count is already zero so get_file_active fails and the binding is removed without holding the invalidate lock. kvm_gmem_invalidate_{start,end}() checks f->bindings independently to decide whether to begin or end KVM MMU invalidation. So, the bindings must therefore remain stable between the two calls. Otherwise, unbind can remove a binding after start increments mmu_invalidate_in_progress, but before end finds the binding and decrements it. Example, unbind race with memory failure: CPU 0: memory failure CPU 1: memslot delete ---------------------------------- --------------------------- (guest_memfd file is dying) kvm_gmem_error_folio() kvm_gmem_invalidate_start() finds binding mmu_invalidate_in_progress++ kvm_gmem_unbind() get_file_active() fails removes binding kvm_gmem_invalidate_end() no binding found counter stays elevated mmu_invalidate_retry() then returns 1 forever, so guest page faults retry without ever installing a mapping and the guest hangs. Remove the distinction between live and dying files in the unbind path. Always use slot->gmem.file and take the invalidate lock when it's non-NULL. Normal unbind callers hold slots_lock, which prevents release from clearing the pointer or freeing the file state until unbind completes. Final VM teardown can only see a NULL pointer because guest_memfd pins the KVM until release has cleared all bindings.