Blame - Documentation/filesystems/nfs/nfs41-server.txt - android_kernel_oneplus_msm8996

Benny Halevy

3ef1728

2009-04-03 08:29:20 +0300

[diff] [blame]

1

NFSv4.1 Server Implementation

2

3

Server support for minorversion 1 can be controlled using the

4

/proc/fs/nfsd/versions control file. The string output returned

5

by reading this file will contain either "+4.1" or "-4.1"

6

correspondingly.

7

8

Currently, server support for minorversion 1 is disabled by default.

9

It can be enabled at run time by writing the string "+4.1" to

10

the /proc/fs/nfsd/versions control file. Note that to write this

11

control file, the nfsd service must be taken down. Use your user-mode

12

nfs-utils to set this up; see rpc.nfsd(8)

13

J. Bruce Fields

285a0f0

2009-09-20 17:01:33 -0400

[diff] [blame]

14

(Warning: older servers will interpret "+4.1" and "-4.1" as "+4" and

15

"-4", respectively. Therefore, code meant to work on both new and old

16

kernels must turn 4.1 on or off *before* turning support for version 4

17

on or off; rpc.nfsd does this correctly.)

18

Benny Halevy

3ef1728

2009-04-03 08:29:20 +0300

[diff] [blame]

19

The NFSv4 minorversion 1 (NFSv4.1) implementation in nfsd is based

J. Bruce Fields

73834d6

2010-01-20 17:17:04 -0500

[diff] [blame]

20

on RFC 5661.

Benny Halevy

3ef1728

2009-04-03 08:29:20 +0300

[diff] [blame]

21

22

From the many new features in NFSv4.1 the current implementation

23

focuses on the mandatory-to-implement NFSv4.1 Sessions, providing

24

"exactly once" semantics and better control and throttling of the

25

resources allocated for each client.

26

27

Other NFSv4.1 features, Parallel NFS operations in particular,

28

are still under development out of tree.

29

See http://wiki.linux-nfs.org/wiki/index.php/PNFS_prototype_design

30

for more information.

31

J. Bruce Fields

285a0f0

2009-09-20 17:01:33 -0400

[diff] [blame]

32

The current implementation is intended for developers only: while it

33

does support ordinary file operations on clients we have tested against

34

(including the linux client), it is incomplete in ways which may limit

35

features unexpectedly, cause known bugs in rare cases, or cause

36

interoperability problems with future clients. Known issues:

37

38

- gss support is questionable: currently mounts with kerberos

39

from a linux client are possible, but we aren't really

40

conformant with the spec (for example, we don't use kerberos

41

on the backchannel correctly).

42

- no trunking support: no clients currently take advantage of

J. Bruce Fields

03d6a74

2009-09-22 11:09:12 -0400

[diff] [blame]

43

trunking, but this is a mandatory feature, and its use is

J. Bruce Fields

285a0f0

2009-09-20 17:01:33 -0400

[diff] [blame]

44

recommended to clients in a number of places. (E.g. to ensure

45

timely renewal in case an existing connection's retry timeouts

J. Bruce Fields

73834d6

2010-01-20 17:17:04 -0500

[diff] [blame]

46

have gotten too long; see section 8.3 of the RFC.)

J. Bruce Fields

285a0f0

2009-09-20 17:01:33 -0400

[diff] [blame]

47

Therefore, lack of this feature may cause future clients to

48

fail.

49

- Incomplete backchannel support: incomplete backchannel gss

50

support and no support for BACKCHANNEL_CTL mean that

51

callbacks (hence delegations and layouts) may not be

52

available and clients confused by the incomplete

53

implementation may fail.

54

- Server reboot recovery is unsupported; if the server reboots,

55

clients may fail.

56

- We do not support SSV, which provides security for shared

57

client-server state (thus preventing unauthorized tampering

58

with locks and opens, for example). It is mandatory for

59

servers to support this, though no clients use it yet.

60

- Mandatory operations which we do not support, such as

61

DESTROY_CLIENTID, FREE_STATEID, SECINFO_NO_NAME, and

62

TEST_STATEID, are not currently used by clients, but will be

63

(and the spec recommends their uses in common cases), and

64

clients should not be expected to know how to recover from the

65

case where they are not supported. This will eventually cause

66

interoperability failures.

67

68

In addition, some limitations are inherited from the current NFSv4

69

implementation:

70

71

- Incomplete delegation enforcement: if a file is renamed or

72

unlinked, a client holding a delegation may continue to

73

indefinitely allow opens of the file under the old name.

74

Benny Halevy

3ef1728

2009-04-03 08:29:20 +0300

[diff] [blame]

75

The table below, taken from the NFSv4.1 document, lists

76

the operations that are mandatory to implement (REQ), optional

77

(OPT), and NFSv4.0 operations that are required not to implement (MNI)

78

in minor version 1. The first column indicates the operations that

79

are not supported yet by the linux server implementation.

80

81

The OPTIONAL features identified and their abbreviations are as follows:

82

pNFS Parallel NFS

83

FDELG File Delegations

84

DDELG Directory Delegations

85

86

The following abbreviations indicate the linux server implementation status.

87

I Implemented NFSv4.1 operations.

88

NS Not Supported.

89

NS* unimplemented optional feature.

90

P pNFS features implemented out of tree.

91

PNS pNFS features that are not supported yet (out of tree).

Operations

+----------------------+------------+--------------+----------------+

96

97

98

| | MNI | or OPT) | |

99

+----------------------+------------+--------------+----------------+

100

101

NS | BACKCHANNEL_CTL | REQ | | Section 18.33 |

102

NS | BIND_CONN_TO_SESSION | REQ | | Section 18.34 |

I | CREATE_SESSION | REQ | | Section 18.36 |

| | | (REQ) | |

NS | DESTROY_CLIENTID | REQ | | Section 18.50 |

112

I | DESTROY_SESSION | REQ | | Section 18.37 |

113

I | EXCHANGE_ID | REQ | | Section 18.35 |

114

NS | FREE_STATEID | REQ | | Section 18.38 |

NS*| OPENATTR | OPT | | Section 18.17 |

132

| OPEN_CONFIRM | MNI | | N/A |

J. Bruce Fields

2010-04-19 15:11:28 -0400

[diff] [blame]

140

Benny Halevy

3ef1728

2009-04-03 08:29:20 +0300

[diff] [blame]

141

| RELEASE_LOCKOWNER | MNI | | N/A |

142

143

144

| RENEW | MNI | | N/A |

I | SEQUENCE | REQ | | Section 18.46 |

151

152

| SETCLIENTID | MNI | | N/A |

153

| SETCLIENTID_CONFIRM | MNI | | N/A |

154

NS | SET_SSV | REQ | | Section 18.47 |

155

NS | TEST_STATEID | REQ | | Section 18.48 |

Callback Operations

+-------------------------+-----------+-------------+---------------+

163

164

165

| | MNI | or OPT) | |

166

+-------------------------+-----------+-------------+---------------+

NS*| CB_NOTIFY_LOCK | OPT | | Section 20.11 |

| | | (REQ) | |

| | | (REQ) | |

NS | CB_RECALL_SLOT | REQ | | Section 20.8 |

| | | (REQ) | |

| | | (REQ) | |

| | | (REQ) | |

+-------------------------+-----------+-------------+---------------+

189

190

Implementation notes:

191

J. Bruce Fields

285a0f0

2009-09-20 17:01:33 -0400

[diff] [blame]

192

DELEGPURGE:

193

* mandatory only for servers that support CLAIM_DELEGATE_PREV and/or

194

CLAIM_DELEG_PREV_FH (which allows clients to keep delegations that

195

persist across client reboots). Thus we need not implement this for

196

now.

197

Benny Halevy

3ef1728

2009-04-03 08:29:20 +0300

[diff] [blame]

198

EXCHANGE_ID:

199

* only SP4_NONE state protection supported

200

* implementation ids are ignored

201

202

CREATE_SESSION:

203

* backchannel attributes are ignored

204

* backchannel security parameters are ignored

205

206

SEQUENCE:

207

* no support for dynamic slot table renegotiation (optional)

208

209

nfsv4.1 COMPOUND rules:

210

The following cases aren't supported yet:

211

* Enforcing of NFS4ERR_NOT_ONLY_OP for: BIND_CONN_TO_SESSION, CREATE_SESSION,

212

DESTROY_CLIENTID, DESTROY_SESSION, EXCHANGE_ID.

213

* DESTROY_SESSION MUST be the final operation in the COMPOUND request.

214

Andy Adamson

ddc04fd

2009-09-23 21:32:21 -0400

[diff] [blame]

215

Nonstandard compound limitations:

216

* No support for a sessions fore channel RPC compound that requires both a

217

ca_maxrequestsize request and a ca_maxresponsesize reply, so we may

218

fail to live up to the promise we made in CREATE_SESSION fore channel

219

negotiation.

220

* No more than one IO operation (read, write, readdir) allowed per

221

compound.