1 <?xml version='1.0' encoding='UTF-8'?>
2 <chapter xmlns="http://docbook.org/ns/docbook"
3 xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en-US"
4 xml:id="managingsecurity">
5 <title xml:id="managingsecurity.title">Managing Security in a Lustre File System</title>
6 <para>This chapter describes security features of the Lustre file system and
7 includes the following sections:</para>
10 <para><xref linkend="managingSecurity.acl"/></para>
13 <para><xref linkend="managingSecurity.root_squash"/></para>
16 <para><xref linkend="managingSecurity.isolation"/></para>
19 <para><xref linkend="managingSecurity.sepol"/></para>
22 <para><xref linkend="managingSecurity.clientencryption"/></para>
25 <section xml:id="managingSecurity.acl">
26 <title><indexterm><primary>Access Control List (ACL)</primary></indexterm>
28 <para>An access control list (ACL), is a set of data that informs an
29 operating system about permissions or access rights that each user or
30 group has to specific system objects, such as directories or files. Each
31 object has a unique security attribute that identifies users who have
32 access to it. The ACL lists each object and user access privileges such as
33 read, write or execute.</para>
34 <section xml:id="managingSecurity.acl.howItWorks" remap="h3">
35 <title><indexterm><primary>Access Control List (ACL)</primary><secondary>
36 how they work</secondary></indexterm>How ACLs Work</title>
37 <para>Implementing ACLs varies between operating systems. Systems that
38 support the Portable Operating System Interface (POSIX) family of
39 standards share a simple yet powerful file system permission model,
40 which should be well-known to the Linux/UNIX administrator. ACLs add
41 finer-grained permissions to this model, allowing for more complicated
42 permission schemes. For a detailed explanation of ACLs on a Linux
43 operating system, refer to the SUSE Labs article
44 <link xl:href="http://wiki.lustre.org/images/5/57/PosixAccessControlInLinux.pdf">
45 Posix Access Control Lists on Linux</link>.</para>
46 <para>We have implemented ACLs according to this model. The Lustre
47 software works with the standard Linux ACL tools, setfacl, getfacl, and
48 the historical chacl, normally installed with the ACL package.</para>
50 <para>ACL support is a system-range feature, meaning that all clients
51 have ACL enabled or not. You cannot specify which clients should
55 <section xml:id="managingSecurity.acl.using" remap="h3">
57 <primary>Access Control List (ACL)</primary>
58 <secondary>using</secondary>
59 </indexterm>Using ACLs with the Lustre Software</title>
60 <para>POSIX Access Control Lists (ACLs) can be used with the Lustre
61 software. An ACL consists of file entries representing permissions based
62 on standard POSIX file system object permissions that define three
63 classes of user (owner, group and other). Each class is associated with
64 a set of permissions [read (r), write (w) and execute (x)].</para>
67 <para>Owner class permissions define access privileges of the file
71 <para>Group class permissions define access privileges of the owning
75 <para>Other class permissions define access privileges of all users
76 not in the owner or group class.</para>
79 <para>The <literal>ls -l</literal> command displays the owner, group, and
80 other class permissions in the first column of its output (for example,
81 <literal>-rw-r- --</literal> for a regular file with read and write
82 access for the owner class, read access for the group class, and no
83 access for others).</para>
84 <para>Minimal ACLs have three entries. Extended ACLs have more than the
85 three entries. Extended ACLs also contain a mask entry and may contain
86 any number of named user and named group entries.</para>
87 <para>The MDS needs to be configured to enable ACLs. Use
88 <literal>--mountfsoptions</literal> to enable ACLs when creating your
90 <screen>$ mkfs.lustre --fsname spfs --mountfsoptions=acl --mdt -mgs /dev/sda</screen>
91 <para>Alternately, you can enable ACLs at run time by using the
92 <literal>--acl</literal> option with <literal>mkfs.lustre</literal>:
94 <screen>$ mount -t lustre -o acl /dev/sda /mnt/mdt</screen>
95 <para>To check ACLs on the MDS:</para>
96 <screen>$ lctl get_param -n mdc.home-MDT0000-mdc-*.connect_flags | grep acl acl</screen>
97 <para>To mount the client with no ACLs:</para>
98 <screen>$ mount -t lustre -o noacl ibmds2@o2ib:/home /home</screen>
99 <para>ACLs are enabled in a Lustre file system on a system-wide basis;
100 either all clients enable ACLs or none do. Activating ACLs is controlled
101 by MDS mount options <literal>acl</literal> / <literal>noacl</literal>
102 (enable/disable ACLs). Client-side mount options acl/noacl are ignored.
103 You do not need to change the client configuration, and the
104 'acl' string will not appear in the client /etc/mtab. The
105 client acl mount option is no longer needed. If a client is mounted with
106 that option, then this message appears in the MDS syslog:</para>
107 <screen>...MDS requires ACL support but client does not</screen>
108 <para>The message is harmless but indicates a configuration issue, which
109 should be corrected.</para>
110 <para>If ACLs are not enabled on the MDS, then any attempts to reference
111 an ACL on a client return an Operation not supported error.</para>
113 <section xml:id="managingSecurity.acl.examples" remap="h3">
115 <primary>Access Control List (ACL)</primary>
116 <secondary>examples</secondary>
117 </indexterm>Examples</title>
118 <para>These examples are taken directly from the POSIX paper referenced
119 above. ACLs on a Lustre file system work exactly like ACLs on any Linux
120 file system. They are manipulated with the standard tools in the
121 standard manner. Below, we create a directory and allow a specific user
123 <screen>[root@client lustre]# umask 027
124 [root@client lustre]# mkdir rain
125 [root@client lustre]# ls -ld rain
126 drwxr-x--- 2 root root 4096 Feb 20 06:50 rain
127 [root@client lustre]# getfacl rain
135 [root@client lustre]# setfacl -m user:chirag:rwx rain
136 [root@client lustre]# ls -ld rain
137 drwxrwx---+ 2 root root 4096 Feb 20 06:50 rain
138 [root@client lustre]# getfacl --omit-header rain
146 <section xml:id="managingSecurity.root_squash">
148 <primary>root squash</primary>
149 </indexterm>Using Root Squash</title>
150 <para>Root squash is a security feature which restricts super-user access
151 rights to a Lustre file system. Without the root squash feature enabled,
152 Lustre file system users on untrusted clients could access or modify files
153 owned by root on the file system, including deleting them. Using the root
154 squash feature restricts file access/modifications as the root user to
155 only the specified clients. Note, however, that this does
156 <emphasis>not</emphasis> prevent users on insecure clients from accessing
157 files owned by <emphasis>other</emphasis> users.</para>
158 <para>The root squash feature works by re-mapping the user ID (UID) and
159 group ID (GID) of the root user to a UID and GID specified by the system
160 administrator, via the Lustre configuration management server (MGS). The
161 root squash feature also enables the Lustre file system administrator to
162 specify a set of client for which UID/GID re-mapping does not apply.
164 <note><para>Nodemaps (<xref linkend="lustrenodemap.title" />) are an
165 alternative to root squash, since it also allows root squash on a per-client
166 basis. With UID maps, the clients can even have a local root UID without
167 actually having root access to the filesystem itself.</para></note>
168 <section xml:id="managingSecurity.root_squash.config" remap="h3">
170 <primary>root squash</primary>
171 <secondary>configuring</secondary>
172 </indexterm>Configuring Root Squash</title>
173 <para>Root squash functionality is managed by two configuration
174 parameters, <literal>root_squash</literal> and
175 <literal>nosquash_nids</literal>.</para>
178 <para>The <literal>root_squash</literal> parameter specifies the UID
179 and GID with which the root user accesses the Lustre file system.
183 <para>The <literal>nosquash_nids</literal> parameter specifies the set
184 of clients to which root squash does not apply. LNet NID range
185 syntax is used for this parameter (see the NID range syntax rules
186 described in <xref linkend="managingSecurity.root_squash"/>). For
190 <screen>nosquash_nids=172.16.245.[0-255/2]@tcp</screen>
191 <para>In this example, root squash does not apply to TCP clients on subnet
192 172.16.245.0 that have an even number as the last component of their IP
195 <section xml:id="managingSecurity.root_squash.tuning">
197 <primary>root squash</primary><secondary>enabling</secondary>
198 </indexterm>Enabling and Tuning Root Squash</title>
199 <para>The default value for <literal>nosquash_nids</literal> is NULL,
200 which means that root squashing applies to all clients. Setting the root
201 squash UID and GID to 0 turns root squash off.</para>
202 <para>Root squash parameters can be set when the MDT is created
203 (<literal>mkfs.lustre --mdt</literal>). For example:</para>
204 <screen>mds# mkfs.lustre --reformat --fsname=testfs --mdt --mgs \
205 --param "mdt.root_squash=500:501" \
206 --param "mdt.nosquash_nids='0@elan1 192.168.1.[10,11]'" /dev/sda1</screen>
207 <para>Root squash parameters can also be changed on an unmounted device
208 with <literal>tunefs.lustre</literal>. For example:</para>
209 <screen>tunefs.lustre --param "mdt.root_squash=65534:65534" \
210 --param "mdt.nosquash_nids=192.168.0.13@tcp0" /dev/sda1
212 <para>Root squash parameters can also be changed with the
213 <literal>lctl conf_param</literal> command. For example:</para>
214 <screen>mgs# lctl conf_param testfs.mdt.root_squash="1000:101"
215 mgs# lctl conf_param testfs.mdt.nosquash_nids="*@tcp"</screen>
216 <para>To retrieve the current root squash parameter settings, the
217 following <literal>lctl get_param</literal> commands can be used:</para>
218 <screen>mgs# lctl get_param mdt.*.root_squash
219 mgs# lctl get_param mdt.*.nosquash_nids</screen>
221 <para>When using the lctl conf_param command, keep in mind:</para>
224 <para><literal>lctl conf_param</literal> must be run on a live MGS
228 <para><literal>lctl conf_param</literal> causes the parameter to
229 change on all MDSs</para>
232 <para><literal>lctl conf_param</literal> is to be used once per a
237 <para>The root squash settings can also be changed temporarily with
238 <literal>lctl set_param</literal> or persistently with
239 <literal>lctl set_param -P</literal>. For example:</para>
240 <screen>mgs# lctl set_param mdt.testfs-MDT0000.root_squash="1:0"
241 mgs# lctl set_param -P mdt.testfs-MDT0000.root_squash="1:0"</screen>
242 <para>The <literal>nosquash_nids</literal> list can be cleared with:</para>
243 <screen>mgs# lctl conf_param testfs.mdt.nosquash_nids="NONE"</screen>
245 <screen>mgs# lctl conf_param testfs.mdt.nosquash_nids="clear"</screen>
246 <para>If the <literal>nosquash_nids</literal> value consists of several
247 NID ranges (e.g. <literal>0@elan</literal>, <literal>1@elan1</literal>),
248 the list of NID ranges must be quoted with single (') or double
249 ('') quotation marks. List elements must be separated with a
250 space. For example:</para>
251 <screen>mds# mkfs.lustre ... --param "mdt.nosquash_nids='0@elan1 1@elan2'" /dev/sda1
252 lctl conf_param testfs.mdt.nosquash_nids="24@elan 15@elan1"</screen>
253 <para>These are examples of incorrect syntax:</para>
254 <screen>mds# mkfs.lustre ... --param "mdt.nosquash_nids=0@elan1 1@elan2" /dev/sda1
255 lctl conf_param testfs.mdt.nosquash_nids=24@elan 15@elan1</screen>
256 <para>To check root squash parameters, use the lctl get_param command:
258 <screen>mds# lctl get_param mdt.testfs-MDT0000.root_squash
259 lctl get_param mdt.*.nosquash_nids</screen>
261 <para>An empty nosquash_nids list is reported as NONE.</para>
264 <section xml:id="managingSecurity.root_squash.tips" remap="h3">
266 <primary>root squash</primary>
267 <secondary>tips</secondary>
268 </indexterm>Tips on Using Root Squash</title>
269 <para>Lustre configuration management limits root squash in several ways.
273 <para>The <literal>lctl conf_param</literal> value overwrites the
274 parameter's previous value. If the new value uses an incorrect
275 syntax, then the system continues with the old parameters and the
276 previously-correct value is lost on remount. That is, be careful
277 doing root squash tuning.</para>
280 <para><literal>mkfs.lustre</literal> and
281 <literal>tunefs.lustre</literal> do not perform parameter syntax
282 checking. If the root squash parameters are incorrect, they are
283 ignored on mount and the default values are used instead.</para>
286 <para>Root squash parameters are parsed with rigorous syntax checking.
287 The root_squash parameter should be specified as
288 <literal><decnum>:<decnum></literal>. The
289 <literal>nosquash_nids</literal> parameter should follow LNet NID
290 range list syntax.</para>
293 <para>LNet NID range syntax:</para>
294 <screen><nidlist> :== <nidrange> [ ' ' <nidrange> ]
295 <nidrange> :== <addrrange> '@' <net>
296 <addrrange> :== '*' |
297 <ipaddr_range> |
298 <numaddr_range>
299 <ipaddr_range> :==
300 <numaddr_range>.<numaddr_range>.<numaddr_range>.<numaddr_range>
301 <numaddr_range> :== <number> |
303 <expr_list> :== '[' <range_expr> [ ',' <range_expr>] ']'
304 <range_expr> :== <number> |
305 <number> '-' <number> |
306 <number> '-' <number> '/' <number>
307 <net> :== <netname> | <netname><number>
308 <netname> :== "lo" | "tcp" | "o2ib"
309 | "ra" | "elan"
310 <number> :== <nonnegative decimal> | <hexadecimal></screen>
312 <para>For networks using numeric addresses (e.g. elan), the address
313 range must be specified in the
314 <literal><numaddr_range></literal> syntax. For networks using
315 IP addresses, the address range must be in the
316 <literal><ipaddr_range></literal>. For example, if elan is using
317 numeric addresses, <literal>1.2.3.4@elan</literal> is incorrect.
322 <section xml:id="managingSecurity.isolation">
323 <title><indexterm><primary>Isolation</primary></indexterm>
324 Isolating Clients to a Sub-directory Tree</title>
325 <para>Isolation is the Lustre implementation of the generic concept of
326 multi-tenancy, which aims at providing separated namespaces from a single
327 filesystem. Lustre Isolation enables different populations of users on
328 the same file system beyond normal Unix permissions/ACLs, even when users
329 on the clients may have root access. Those tenants share the same file
330 system, but they are isolated from each other: they cannot access or even
331 see each other’s files, and are not aware that they are sharing common
332 file system resources.</para>
333 <para>Lustre Isolation leverages the Fileset feature
334 (<xref linkend="SystemConfigurationUtilities.fileset" />)
335 to mount only a subdirectory of the filesystem rather than the root
337 In order to achieve isolation, the subdirectory mount, which presents to
338 tenants only their own fileset, has to be imposed to the clients. To that
339 extent, we make use of the nodemap feature
340 (<xref linkend="lustrenodemap.title" />). We group all clients used by a
341 tenant under a common nodemap entry, and we assign to this nodemap entry
342 the fileset to which the tenant is restricted.</para>
343 <section xml:id="managingSecurity.isolation.clientid" remap="h3">
344 <title><indexterm><primary>Isolation</primary><secondary>
345 client identification</secondary></indexterm>Identifying Clients</title>
346 <para>Enforcing multi-tenancy on Lustre relies on the ability to properly
347 identify the client nodes used by a tenant, and trust those identities.
348 This can be achieved by having physical hardware and/or network
349 security, so that client nodes have well-known NIDs. It is also possible
350 to make use of strong authentication with Kerberos or Shared-Secret Key
351 (see <xref linkend="lustressk" />).
352 Kerberos prevents NID spoofing, as every client needs its own
353 credentials, based on its NID, in order to connect to the servers.
354 Shared-Secret Key also prevents tenant impersonation, because keys
355 can be linked to a specific nodemap. See
356 <xref linkend="ssknodemaprole" /> for detailed explanations.
359 <section xml:id="managingSecurity.isolation.configuring" remap="h3">
360 <title><indexterm><primary>Isolation</primary><secondary>
361 configuring</secondary></indexterm>Configuring Isolation</title>
362 <para>Isolation on Lustre can be achieved by setting the
363 <literal>fileset</literal> parameter on a nodemap entry. All clients
364 belonging to this nodemap entry will automatically mount this fileset
365 instead of the root directory. For example:</para>
366 <screen>mgs# lctl nodemap_set_fileset --name tenant1 --fileset '/dir1'</screen>
367 <para>So all clients matching the <literal>tenant1</literal> nodemap will
368 be automatically presented the fileset <literal>/dir1</literal> when
369 mounting. This means these clients are doing an implicit subdirectory
370 mount on the subdirectory <literal>/dir1</literal>.
374 If subdirectory defined as fileset does not exist on the file system,
375 it will prevent any client belonging to the nodemap from mounting
379 <para>To delete the fileset parameter, just set it to an empty string:
381 <screen>mgs# lctl nodemap_set_fileset --name tenant1 --fileset ''</screen>
383 <section xml:id="managingSecurity.isolation.permanent" remap="h3">
384 <title><indexterm><primary>Isolation</primary><secondary>
385 making permanent</secondary></indexterm>Making Isolation Permanent
387 <para>In order to make isolation permanent, the fileset parameter on the
388 nodemap has to be set with <literal>lctl set_param</literal> with the
389 <literal>-P</literal> option.</para>
390 <screen>mgs# lctl set_param nodemap.tenant1.fileset=/dir1
391 mgs# lctl set_param -P nodemap.tenant1.fileset=/dir1</screen>
392 <para>This way the fileset parameter will be stored in the Lustre config
393 logs, letting the servers retrieve the information after a restart.
397 <section xml:id="managingSecurity.sepol" condition='l2D'>
398 <title><indexterm><primary>selinux policy check</primary></indexterm>
399 Checking SELinux Policy Enforced by Lustre Clients</title>
400 <para>SELinux provides a mechanism in Linux for supporting Mandatory Access
401 Control (MAC) policies. When a MAC policy is enforced, the operating
402 system’s (OS) kernel defines application rights, firewalling applications
403 from compromising the entire system. Regular users do not have the ability to
404 override the policy.</para>
405 <para>One purpose of SELinux is to protect the
406 <emphasis role="bold">OS</emphasis> from privilege escalation. To that
407 extent, SELinux defines confined and unconfined domains for processes and
408 users. Each process, user, file is assigned a security context, and
409 rules define the allowed operations by processes and users on files.
411 <para>Another purpose of SELinux can be to protect
412 <emphasis role="bold">data</emphasis> sensitivity, thanks to Multi-Level
413 Security (MLS). MLS works on top of SELinux, by defining the concept of
414 security levels in addition to domains. Each process, user and file is
415 assigned a security level, and the model states that processes and users
416 can read the same or lower security level, but can only write to their own
417 or higher security level.
419 <para>From a file system perspective, the security context of files must be
420 stored permanently. Lustre makes use of the
421 <literal>security.selinux</literal> extended attributes on files to hold
422 this information. Lustre supports SELinux on the client side. All you have
423 to do to have MAC and MLS on Lustre is to enforce the appropriate SELinux
424 policy (as provided by the Linux distribution) on all Lustre clients. No
425 SELinux is required on Lustre servers.
427 <para>Because Lustre is a distributed file system, the specificity when
428 using MLS is that Lustre really needs to make sure data is always accessed
429 by nodes with the SELinux MLS policy properly enforced. Otherwise, data is
430 not protected. This means Lustre has to check that SELinux is properly
431 enforced on client side, with the right, unaltered policy. And if SELinux
432 is not enforced as expected on a client, the server denies its access to
435 <section xml:id="managingSecurity.sepol.determining" remap="h3">
436 <title><indexterm><primary>selinux policy check</primary><secondary>
437 determining</secondary></indexterm>Determining SELinux Policy Info
439 <para>A string that represents the SELinux Status info will be used by
440 servers as a reference, to check if clients are enforcing SELinux
441 properly. This reference string can be obtained on a client node known
442 to enforce the right SELinux policy, by calling the
443 <literal>l_getsepol</literal> command line utility:</para>
444 <screen>client# l_getsepol
445 SELinux status info: 1:mls:31:40afb76d077c441b69af58cccaaa2ca63641ed6e21b0a887dc21a684f508b78f</screen>
446 <para>The string describing the SELinux policy has the following
448 <para><literal>mode:name:version:hash</literal></para>
452 <para><literal>mode</literal> is a digit telling if SELinux is in
453 Permissive mode (0) or Enforcing mode (1)</para>
456 <para><literal>name</literal> is the name of the SELinux policy
460 <para><literal>version</literal> is the version of the SELinux
464 <para><literal>hash</literal> is the computed hash of the binary
465 representation of the policy, as exported in
466 /etc/selinux/<literal>name</literal>/policy/policy.
467 <literal>version</literal></para>
471 <section xml:id="managingSecurity.sepol.configuring" remap="h3">
472 <title><indexterm><primary>selinux policy check</primary><secondary>
473 enforcing</secondary></indexterm>Enforcing SELinux Policy Check</title>
474 <para>SELinux policy check can be enforced by setting the
475 <literal>sepol</literal> parameter on a nodemap entry. All clients
476 belonging to this nodemap entry must enforce the SELinux policy
477 described by this parameter, otherwise they are denied access to the
478 Lustre file system. For example:</para>
479 <screen>mgs# lctl nodemap_set_sepol --name restricted
480 --sepol '1:mls:31:40afb76d077c441b69af58cccaaa2ca63641ed6e21b0a887dc21a684f508b78f'</screen>
481 <para>So all clients matching the <literal>restricted</literal> nodemap
482 must enforce the SELinux policy which description matches
483 <literal>1:mls:31:40afb76d077c441b69af58cccaaa2ca63641ed6e21b0a887dc21a684f508b78f</literal>.
484 If not, they will get Permission Denied when trying to mount or access
485 files on the Lustre file system.</para>
486 <para>To delete the <literal>sepol</literal> parameter, just set it to an
488 <screen>mgs# lctl nodemap_set_sepol --name restricted --sepol ''</screen>
489 <para>See <xref linkend="lustrenodemap.title" /> for more details about
490 the Nodemap feature.</para>
492 <section xml:id="managingSecurity.sepol.permanent" remap="h3">
493 <title><indexterm><primary>selinux policy check</primary><secondary>
494 making permanent</secondary></indexterm>Making SELinux Policy Check
496 <para>In order to make SELinux Policy check permanent, the sepol parameter
497 on the nodemap has to be set with <literal>lctl set_param</literal> with
498 the <literal>-P</literal> option.</para>
499 <screen>mgs# lctl set_param nodemap.restricted.sepol=1:mls:31:40afb76d077c441b69af58cccaaa2ca63641ed6e21b0a887dc21a684f508b78f
500 mgs# lctl set_param -P nodemap.restricted.sepol=1:mls:31:40afb76d077c441b69af58cccaaa2ca63641ed6e21b0a887dc21a684f508b78f</screen>
501 <para>This way the sepol parameter will be stored in the Lustre config
502 logs, letting the servers retrieve the information after a restart.
505 <section xml:id="managingSecurity.sepol.client" remap="h3">
506 <title><indexterm><primary>selinux policy check</primary><secondary>
507 sending client</secondary></indexterm>Sending SELinux Status Info from
509 <para>In order for Lustre clients to send their SELinux status
510 information, in case SELinux is enabled locally, the
511 <literal>send_sepol</literal> ptlrpc kernel module's parameter has to be
512 set to a non-zero value. <literal>send_sepol</literal> accepts various
516 <para>0: do not send SELinux policy info;</para>
519 <para>-1: fetch SELinux policy info for every request;</para>
522 <para>N > 0: only fetch SELinux policy info every N seconds. Use
523 <literal>N = 2^31-1</literal> to have SELinux policy info
524 fetched only at mount time.</para>
527 <para>Clients that are part of a nodemap on which
528 <literal>sepol</literal> is defined must send SELinux status info.
529 And the SELinux policy they enforce must match the representation
530 stored into the nodemap. Otherwise they will be denied access to the
531 Lustre file system.</para>
534 <section xml:id="managingSecurity.clientencryption" condition='l2E'>
535 <title><indexterm><primary>Client-side encryption</primary></indexterm>
536 Encrypting files and directories</title>
537 <para>The purpose that client-side encryption wants to serve is to be able
538 to provide a special directory for each user, to safely store sensitive
539 files. The goals are to protect data in transit between clients and
540 servers, and protect data at rest.</para>
541 <para>This feature is implemented directly at the Lustre client level.
542 Lustre client-side encryption relies on kernel <literal>fscrypt</literal>.
543 <literal>fscrypt</literal> is a library which filesystems can hook into to
544 support transparent encryption of files and directories. As a consequence,
545 the key points described below are extracted from
546 <literal>fscrypt</literal> documentation.</para>
547 <para>For full details, please refer to documentation available with the
548 Lustre sources, under the
549 <literal>Documentation/client_side_encryption</literal> directory.
551 <note><para>The client-side encryption feature is available on Lustre
552 clients running a Linux distribution with at least kernel 5.4, or have
553 backported the <literal>fscrypt</literal> v2 support, including:</para>
555 <listitem><para>CentOS/RHEL 8.1 and later;</para></listitem>
556 <listitem><para>Ubuntu 18.04 and later;</para></listitem>
557 <listitem><para>SLES 15 SP2 and later.</para></listitem>
560 <section xml:id="managingSecurity.clientencryption.semantics" remap="h3">
561 <title><indexterm><primary>encryption access semantics</primary>
562 </indexterm>Client-side encryption access semantics</title>
563 <para>Only Lustre clients need access to encryption master keys. Keys are
564 added to the filesystem-level encryption keyring on the Lustre client.
567 <para><emphasis role="bold">With the key</emphasis></para>
568 <para>With the encryption key, encrypted regular files, directories,
569 and symlinks behave very similarly to their unencrypted
570 counterparts --- after all, the encryption is intended to be
571 transparent. However, astute users may notice some differences in
575 <para>Unencrypted files, or files encrypted with a different
576 encryption policy (i.e. different key, modes, or flags),
577 cannot be renamed or linked into an encrypted directory.
578 However, encrypted files can be renamed within an encrypted
579 directory, or into an unencrypted directory.</para>
580 <note><para>"moving" an unencrypted file into an encrypted
581 directory, e.g. with the <literal>mv</literal> program, is
582 implemented in userspace by a copy followed by a delete. Be
583 aware the original unencrypted data may remain recoverable
584 from free space on the disk; it is best to keep all files
585 encrypted from the very beginning.</para></note>
587 <listitem><para>On Lustre, Direct I/O is supported for encrypted
590 <listitem><para>The <literal>fallocate()</literal> operations
591 <literal>FALLOC_FL_COLLAPSE_RANGE</literal>,
592 <literal>FALLOC_FL_INSERT_RANGE</literal>, and
593 <literal>FALLOC_FL_ZERO_RANGE</literal> are not
594 supported on encrypted files and will fail with
595 <literal>EOPNOTSUPP</literal>.
598 <listitem><para>DAX (Direct Access) is not supported on encrypted
601 <listitem><para><literal>mmap</literal> is supported. This is
602 possible because the pagecache for an encrypted file contains
603 the plaintext, not the ciphertext.</para>
608 <para><emphasis role="bold">Without the key</emphasis></para>
609 <para>Some filesystem operations may be performed on encrypted
610 regular files, directories, and symlinks even before their
611 encryption key has been added, or after their encryption key has
615 <para>File metadata may be read, e.g. using
616 <literal>stat()</literal>.</para>
619 <para>Directories may be listed, and the whole namespace tree
620 may be walked through.
624 <para>Files may be deleted. That is, nondirectory files may be
625 deleted with <literal>unlink()</literal> as usual, and empty
626 directories may be deleted with <literal>rmdir()</literal> as
627 usual. Therefore, <literal>rm</literal> and
628 <literal>rm -r</literal> will work as expected.</para>
631 <para>Symlink targets may be read and followed, but they will
632 be presented in encrypted form, similar to filenames in
633 directories. Hence, they are unlikely to point to anywhere
637 <para>Without the key, regular files cannot be opened or truncated.
638 Attempts to do so will fail with <literal>ENOKEY</literal>. This
639 implies that any regular file operations that require a file
640 descriptor, such as <literal>read()</literal>,
641 <literal>write()</literal>, <literal>mmap()</literal>,
642 <literal>fallocate()</literal>, and <literal>ioctl()</literal>,
643 are also forbidden.</para>
644 <para>Also without the key, files of any type (including
645 directories) cannot be created or linked into an encrypted
646 directory, nor can a name in an encrypted directory be the source
647 or target of a rename, nor can an <literal>O_TMPFILE</literal>
648 temporary file be created in an encrypted directory. All such
649 operations will fail with <literal>ENOKEY</literal>.</para>
650 <para>It is not currently possible to backup and restore encrypted
651 files without the encryption key. This would require special
652 APIs which have not yet been implemented.</para>
655 <para><emphasis role="bold">Encryption policy enforcement
657 <para>After an encryption policy has been set on a directory, all
658 regular files, directories, and symbolic links created in that
659 directory (recursively) will inherit that encryption policy.
660 Special files --- that is, named pipes, device nodes, and UNIX
661 domain sockets --- will not be encrypted.</para>
662 <para>Except for those special files, it is forbidden to have
663 unencrypted files, or files encrypted with a different encryption
664 policy, in an encrypted directory tree.</para>
669 <section xml:id="managingSecurity.clientencryption.keyhierarchy" remap="h3">
670 <title><indexterm><primary>encryption key hierarchy</primary>
671 </indexterm>Client-side encryption key hierarchy</title>
672 <para>Each encrypted directory tree is protected by a master key.</para>
673 <para>To "unlock" an encrypted directory tree, userspace must provide the
674 appropriate master key. There can be any number of master keys, each
675 of which protects any number of directory trees on any number of
678 <section xml:id="managingSecurity.clientencryption.modes" remap="h3">
679 <title><indexterm><primary>encryption modes usage</primary>
680 </indexterm>Client-side encryption modes and usage</title>
681 <para><literal>fscrypt</literal> allows one encryption mode to be
682 specified for file contents and one encryption mode to be specified for
683 filenames. Different directory trees are permitted to use different
684 encryption modes. Currently, the following pairs of encryption modes are
688 <para>AES-256-XTS for contents and AES-256-CTS-CBC for filenames
692 <para>AES-128-CBC for contents and AES-128-CTS-CBC for filenames
696 <para>If unsure, you should use the (AES-256-XTS, AES-256-CTS-CBC) pair.
698 <warning><para>In Lustre 2.14, client-side encryption only supports
699 content encryption, and not filename encryption. As a consequence, only
700 content encryption mode will be taken into account, and filename
701 encryption mode will be ignored to leave filenames in clear text.</para>
704 <section xml:id="managingSecurity.clientencryption.threatmodel" remap="h3">
705 <title><indexterm><primary>encryption threat model</primary>
706 </indexterm>Client-side encryption threat model</title>
709 <para><emphasis role="bold">Offline attacks</emphasis></para>
710 <para>For the Lustre case, block devices are Lustre targets attached
711 to the Lustre servers. Manipulating the filesystem offline means
712 accessing the filesystem on these targets while Lustre is offline.
714 <para>Provided that a strong encryption key is chosen,
715 <literal>fscrypt</literal> protects the confidentiality of file
716 contents in the event of a single point-in-time permanent offline
717 compromise of the block device content.
718 Lustre client-side encryption does not protect the confidentiality
719 of metadata, e.g. file names, file sizes, file permissions, file
720 timestamps, and extended attributes. Also, the existence and
721 location of holes (unallocated blocks which logically contain all
722 zeroes) in files is not protected.</para>
725 <para><emphasis role="bold">Online attacks</emphasis></para>
728 <para>On Lustre client</para>
729 <para>After an encryption key has been added,
730 <literal>fscrypt</literal> does not hide the plaintext file
731 contents or filenames from other users on the same node.
732 Instead, existing access control mechanisms such as file mode
733 bits, POSIX ACLs, LSMs, or namespaces should be used for this
735 <para>For the Lustre case, it means plaintext file contents or
736 filenames are not hidden from other users on the same Lustre
738 <para>An attacker who compromises the system enough to read from
739 arbitrary memory, e.g. by exploiting a kernel security
740 vulnerability, can compromise all encryption keys that are
742 However, <literal>fscrypt</literal> allows encryption keys to
743 be removed from the kernel, which may protect them from later
744 compromise. Key removal can be carried out by non-root users.
745 In more detail, the key removal will wipe the master encryption
746 key from kernel memory. Moreover, it will try to evict all
747 cached inodes which had been "unlocked" using the key, thereby
748 wiping their per-file keys and making them once again appear
749 "locked", i.e. in ciphertext or encrypted form.</para>
752 <para>On Lustre server</para>
753 <para>An attacker on a Lustre server who compromises the system
754 enough to read arbitrary memory, e.g. by exploiting a kernel
755 security vulnerability, cannot compromise Lustre files content.
756 Indeed, encryption keys are not forwarded to the Lustre servers,
757 and servers do not carry out decryption or encryption.
758 Moreover, bulk RPCs received by servers contain encrypted data,
759 which is written as-is to the underlying filesystem.</para>
765 <section xml:id="managingSecurity.clientencryption.fscrypt" remap="h3">
766 <title><indexterm><primary>encryption fscrypt policy</primary>
767 </indexterm>Manage encryption on directories</title>
768 <para>By default, Lustre client-side encryption is enabled, letting users
769 define encryption policies on a per-directory basis.</para>
770 <note><para>Administrators can decide to prevent a Lustre client
771 mount-point from using encryption by specifying the
772 <literal>noencrypt</literal> client mount option. This can be also
773 enforced from server side thanks to the
774 <literal>forbid_encryption</literal> property on nodemaps. See
775 <xref linkend="alteringproperties"/> for how to manage nodemaps.
777 <para><literal>fscrypt</literal> userspace tool can be used to manage
778 encryption policies. See https://github.com/google/fscrypt for
779 comprehensive explanations. Below are examples on how to use this tool
780 with Lustre. If not told otherwise, commands must be run on Lustre
784 <para>Two preliminary steps are required before actually deciding
785 which directories to encrypt, and this is the only
786 functionality which requires root privileges. Administrator has to
788 <screen># fscrypt setup
789 Customizing passphrase hashing difficulty for this system...
790 Created global config file at "/etc/fscrypt.conf".
791 Metadata directories created at "/.fscrypt".</screen>
792 <para>This first command has to be run on all clients that want to use
793 encryption, as it sets up global fscrypt parameters outside of
795 <screen># fscrypt setup /mnt/lustre
796 Metadata directories created at "/mnt/lustre/.fscrypt"</screen>
797 <para>This second command has to be run on just one Lustre
799 <note><para>The file <literal>/etc/fscrypt.conf</literal> can be
800 edited. It is strongly recommended to set
801 <literal>policy_version</literal> to 2, so that
802 <literal>fscrypt</literal> wipes files from memory when the
803 encryption key is removed.</para></note>
806 <para>Now a regular user is able to select a directory to
808 <screen>$ fscrypt encrypt /mnt/lustre/vault
809 The following protector sources are available:
810 1 - Your login passphrase (pam_passphrase)
811 2 - A custom passphrase (custom_passphrase)
812 3 - A raw 256-bit key (raw_key)
813 Enter the source number for the new protector [2 - custom_passphrase]: 2
814 Enter a name for the new protector: shield
815 Enter custom passphrase for protector "shield":
817 "/mnt/lustre/vault" is now encrypted, unlocked, and ready for use.</screen>
818 <para>Starting from here, all files and directories created under
819 <literal>/mnt/lustre/vault</literal> will be encrypted, according
820 to the policy defined at the previsous step.</para>
821 <note><para>The encryption policy is inherited by all subdirectories.
822 It is not possible to change the policy for a subdirectory.</para>
826 <para>Another user can decide to encrypt a different directory with
827 its own protector:</para>
828 <screen>$ fscrypt encrypt /mnt/lustre/private
829 Should we create a new protector? [y/N] Y
830 The following protector sources are available:
831 1 - Your login passphrase (pam_passphrase)
832 2 - A custom passphrase (custom_passphrase)
833 3 - A raw 256-bit key (raw_key)
834 Enter the source number for the new protector [2 - custom_passphrase]: 2
835 Enter a name for the new protector: armor
836 Enter custom passphrase for protector "armor":
838 "/mnt/lustre/private" is now encrypted, unlocked, and ready for use.</screen>
841 <para>Users can decide to lock an encrypted directory at any
843 <screen>$ fscrypt lock /mnt/lustre/vault
844 "/mnt/lustre/vault" is now locked.</screen>
845 <para>This action prevents access to encrypted content, and by
846 removing the key from memory, it also wipes files from memory if
847 they are not still open.</para>
850 <para>Users regain access to the encrypted directory with the command:
852 <screen>$ fscrypt unlock /mnt/lustre/vault
853 Enter custom passphrase for protector "shield":
854 "/mnt/lustre/vault" is now unlocked and ready for use.</screen>
857 <para>Actually, <literal>fscrypt</literal> does not give direct access
858 to master keys, but to protectors that are used to encrypt them.
859 This mechanism gives the ability to change a passphrase:</para>
860 <screen>$ fscrypt status /mnt/lustre
861 lustre filesystem "/mnt/lustre" has 2 protectors and 2 policies
863 PROTECTOR LINKED DESCRIPTION
864 deacab807bf0e788 No custom protector "shield"
865 e691ae7a1990fc2a No custom protector "armor"
867 POLICY UNLOCKED PROTECTORS
868 52b2b5aff0e59d8e0d58f962e715862e No deacab807bf0e788
869 374e8944e4294b527e50363d86fc9411 No e691ae7a1990fc2a
871 $ fscrypt metadata change-passphrase --protector=/mnt/lustre:deacab807bf0e788
872 Enter old custom passphrase for protector "shield":
873 Enter new custom passphrase for protector "shield":
875 Passphrase for protector deacab807bf0e788 successfully changed.</screen>
876 <para>It makes also possible to have multiple protectors for the same
877 policy. This is really useful when several users share an encrypted
878 directory, because it avoids the need to share any secret between
880 <screen>$ fscrypt status /mnt/lustre/vault
881 "/mnt/lustre/vault" is encrypted with fscrypt.
883 Policy: 52b2b5aff0e59d8e0d58f962e715862e
884 Options: padding:32 contents:AES_256_XTS filenames:AES_256_CTS policy_version:2
887 Protected with 1 protector:
888 PROTECTOR LINKED DESCRIPTION
889 deacab807bf0e788 No custom protector "shield"
891 $ fscrypt metadata create protector /mnt/lustre
892 Create new protector on "/mnt/lustre" [Y/n] Y
893 The following protector sources are available:
894 1 - Your login passphrase (pam_passphrase)
895 2 - A custom passphrase (custom_passphrase)
896 3 - A raw 256-bit key (raw_key)
897 Enter the source number for the new protector [2 - custom_passphrase]: 2
898 Enter a name for the new protector: bunker
899 Enter custom passphrase for protector "bunker":
901 Protector f3cc1b5cf9b8f41c created on filesystem "/mnt/lustre".
903 $ fscrypt metadata add-protector-to-policy
904 --protector=/mnt/lustre:f3cc1b5cf9b8f41c
905 --policy=/mnt/lustre:52b2b5aff0e59d8e0d58f962e715862e
906 WARNING: All files using this policy will be accessible with this protector!!
907 Protect policy 52b2b5aff0e59d8e0d58f962e715862e with protector f3cc1b5cf9b8f41c? [Y/n] Y
908 Enter custom passphrase for protector "bunker":
909 Enter custom passphrase for protector "shield":
910 Protector f3cc1b5cf9b8f41c now protecting policy 52b2b5aff0e59d8e0d58f962e715862e.
912 $ fscrypt status /mnt/lustre/vault
913 "/mnt/lustre/vault" is encrypted with fscrypt.
915 Policy: 52b2b5aff0e59d8e0d58f962e715862e
916 Options: padding:32 contents:AES_256_XTS filenames:AES_256_CTS policy_version:2
919 Protected with 2 protectors:
920 PROTECTOR LINKED DESCRIPTION
921 deacab807bf0e788 No custom protector "shield"
922 f3cc1b5cf9b8f41c No custom protector "bunker"</screen>
929 vim:expandtab:shiftwidth=2:tabstop=8: