1.. SPDX-License-Identifier: GPL-2.0 2 3======================= 4NFSv4 client identifier 5======================= 6 7This document explains how the NFSv4 protocol identifies client 8instances in order to maintain file open and lock state during 9system restarts. A special identifier and principal are maintained 10on each client. These can be set by administrators, scripts 11provided by site administrators, or tools provided by Linux 12distributors. 13 14There are risks if a client's NFSv4 identifier and its principal 15are not chosen carefully. 16 17 18Introduction 19------------ 20 21The NFSv4 protocol uses "lease-based file locking". Leases help 22NFSv4 servers provide file lock guarantees and manage their 23resources. 24 25Simply put, an NFSv4 server creates a lease for each NFSv4 client. 26The server collects each client's file open and lock state under 27the lease for that client. 28 29The client is responsible for periodically renewing its leases. 30While a lease remains valid, the server holding that lease 31guarantees the file locks the client has created remain in place. 32 33If a client stops renewing its lease (for example, if it crashes), 34the NFSv4 protocol allows the server to remove the client's open 35and lock state after a certain period of time. When a client 36restarts, it indicates to servers that open and lock state 37associated with its previous leases is no longer valid and can be 38destroyed immediately. 39 40In addition, each NFSv4 server manages a persistent list of client 41leases. When the server restarts and clients attempt to recover 42their state, the server uses this list to distinguish amongst 43clients that held state before the server restarted and clients 44sending fresh OPEN and LOCK requests. This enables file locks to 45persist safely across server restarts. 46 47NFSv4 client identifiers 48------------------------ 49 50Each NFSv4 client presents an identifier to NFSv4 servers so that 51they can associate the client with its lease. Each client's 52identifier consists of two elements: 53 54 - co_ownerid: An arbitrary but fixed string. 55 56 - boot verifier: A 64-bit incarnation verifier that enables a 57 server to distinguish successive boot epochs of the same client. 58 59The NFSv4.0 specification refers to these two items as an 60"nfs_client_id4". The NFSv4.1 specification refers to these two 61items as a "client_owner4". 62 63NFSv4 servers tie this identifier to the principal and security 64flavor that the client used when presenting it. Servers use this 65principal to authorize subsequent lease modification operations 66sent by the client. Effectively this principal is a third element of 67the identifier. 68 69As part of the identity presented to servers, a good 70"co_ownerid" string has several important properties: 71 72 - The "co_ownerid" string identifies the client during reboot 73 recovery, therefore the string is persistent across client 74 reboots. 75 - The "co_ownerid" string helps servers distinguish the client 76 from others, therefore the string is globally unique. Note 77 that there is no central authority that assigns "co_ownerid" 78 strings. 79 - Because it often appears on the network in the clear, the 80 "co_ownerid" string does not reveal private information about 81 the client itself. 82 - The content of the "co_ownerid" string is set and unchanging 83 before the client attempts NFSv4 mounts after a restart. 84 - The NFSv4 protocol places a 1024-byte limit on the size of the 85 "co_ownerid" string. 86 87Protecting NFSv4 lease state 88---------------------------- 89 90NFSv4 servers utilize the "client_owner4" as described above to 91assign a unique lease to each client. Under this scheme, there are 92circumstances where clients can interfere with each other. This is 93referred to as "lease stealing". 94 95If distinct clients present the same "co_ownerid" string and use 96the same principal (for example, AUTH_SYS and UID 0), a server is 97unable to tell that the clients are not the same. Each distinct 98client presents a different boot verifier, so it appears to the 99server as if there is one client that is rebooting frequently. 100Neither client can maintain open or lock state in this scenario. 101 102If distinct clients present the same "co_ownerid" string and use 103distinct principals, the server is likely to allow the first client 104to operate normally but reject subsequent clients with the same 105"co_ownerid" string. 106 107If a client's "co_ownerid" string or principal are not stable, 108state recovery after a server or client reboot is not guaranteed. 109If a client unexpectedly restarts but presents a different 110"co_ownerid" string or principal to the server, the server orphans 111the client's previous open and lock state. This blocks access to 112locked files until the server removes the orphaned state. 113 114If the server restarts and a client presents a changed "co_ownerid" 115string or principal to the server, the server will not allow the 116client to reclaim its open and lock state, and may give those locks 117to other clients in the meantime. This is referred to as "lock 118stealing". 119 120Lease stealing and lock stealing increase the potential for denial 121of service and in rare cases even data corruption. 122 123Selecting an appropriate client identifier 124------------------------------------------ 125 126By default, the Linux NFSv4 client implementation constructs its 127"co_ownerid" string starting with the words "Linux NFS" followed by 128the client's UTS node name (the same node name, incidentally, that 129is used as the "machine name" in an AUTH_SYS credential). In small 130deployments, this construction is usually adequate. Often, however, 131the node name by itself is not adequately unique, and can change 132unexpectedly. Problematic situations include: 133 134 - NFS-root (diskless) clients, where the local DCHP server (or 135 equivalent) does not provide a unique host name. 136 137 - "Containers" within a single Linux host. If each container has 138 a separate network namespace, but does not use the UTS namespace 139 to provide a unique host name, then there can be multiple NFS 140 client instances with the same host name. 141 142 - Clients across multiple administrative domains that access a 143 common NFS server. If hostnames are not assigned centrally 144 then uniqueness cannot be guaranteed unless a domain name is 145 included in the hostname. 146 147Linux provides two mechanisms to add uniqueness to its "co_ownerid" 148string: 149 150 nfs.nfs4_unique_id 151 This module parameter can set an arbitrary uniquifier string 152 via the kernel command line, or when the "nfs" module is 153 loaded. 154 155 /sys/fs/nfs/net/nfs_client/identifier 156 This virtual file, available since Linux 5.3, is local to the 157 network namespace in which it is accessed and so can provide 158 distinction between network namespaces (containers) when the 159 hostname remains uniform. 160 161Note that this file is empty on name-space creation. If the 162container system has access to some sort of per-container identity 163then that uniquifier can be used. For example, a uniquifier might 164be formed at boot using the container's internal identifier: 165 166 sha256sum /etc/machine-id | awk '{print $1}' \\ 167 > /sys/fs/nfs/net/nfs_client/identifier 168 169Security considerations 170----------------------- 171 172The use of cryptographic security for lease management operations 173is strongly encouraged. 174 175If NFS with Kerberos is not configured, a Linux NFSv4 client uses 176AUTH_SYS and UID 0 as the principal part of its client identity. 177This configuration is not only insecure, it increases the risk of 178lease and lock stealing. However, it might be the only choice for 179client configurations that have no local persistent storage. 180"co_ownerid" string uniqueness and persistence is critical in this 181case. 182 183When a Kerberos keytab is present on a Linux NFS client, the client 184attempts to use one of the principals in that keytab when 185identifying itself to servers. The "sec=" mount option does not 186control this behavior. Alternately, a single-user client with a 187Kerberos principal can use that principal in place of the client's 188host principal. 189 190Using Kerberos for this purpose enables the client and server to 191use the same lease for operations covered by all "sec=" settings. 192Additionally, the Linux NFS client uses the RPCSEC_GSS security 193flavor with Kerberos and the integrity QOS to prevent in-transit 194modification of lease modification requests. 195 196Additional notes 197---------------- 198The Linux NFSv4 client establishes a single lease on each NFSv4 199server it accesses. NFSv4 mounts from a Linux NFSv4 client of a 200particular server then share that lease. 201 202Once a client establishes open and lock state, the NFSv4 protocol 203enables lease state to transition to other servers, following data 204that has been migrated. This hides data migration completely from 205running applications. The Linux NFSv4 client facilitates state 206migration by presenting the same "client_owner4" to all servers it 207encounters. 208 209======== 210See Also 211======== 212 213 - nfs(5) 214 - kerberos(7) 215 - RFC 7530 for the NFSv4.0 specification 216 - RFC 8881 for the NFSv4.1 specification. 217