1OCFS2(7) OCFS2 Manual Pages OCFS2(7)
2
6 OCFS2 - A Shared-Disk Cluster File System for Linux
7
10 OCFS2 is a file system. It allows users to store and retrieve data. The
11 data is stored in files that are organized in a hierarchical directory
12 tree. It is a POSIX compliant file system that supports the standard
13 interfaces and the behavioral semantics as spelled out by that specifi‐
14 cation.
15 It is also a shared disk cluster file system, one that allows multiple
17 nodes to access the same disk at the same time. This is where the fun
18 begins as allowing a file system to be accessible on multiple nodes
19 opens a can of worms. What if the nodes are of different architectures?
20 What if a node dies while writing to the file system? What data consis‐
21 tency can one expect if processes on two nodes are reading and writing
22 concurrently? What if one node removes a file while it is still being
23 used on another node?
24 Unlike most shared file systems where the answer is fuzzy, the answer
26 in OCFS2 is very well defined. It behaves on all nodes exactly like a
27 local file system. If a file is removed, the directory entry is removed
28 but the inode is kept as long as it is in use across the cluster. When
29 the last user closes the descriptor, the inode is marked for deletion.
30 The data consistency model follows the same principle. It works as if
32 the two processes that are running on two different nodes are running
33 on the same node. A read on a node gets the last write irrespective of
34 the IO mode used. The modes can be buffered, direct, asynchronous,
35 splice or memory mapped IOs. It is fully cache coherent.
36 Take for example the REFLINK feature that allows a user to create mul‐
38 tiple write-able snapshots of a file. This feature, like all others, is
39 fully cluster-aware. A file being written to on multiple nodes can be
40 safely reflinked on another. The snapshot created is a point-in-time
41 image of the file that includes both the file data and all its
42 attributes (including extended attributes).
43 It is a journaling file system. When a node dies, a surviving node
45 transparently replays the journal of the dead node. This ensures that
46 the file system metadata is always consistent. It also defaults to
47 ordered data journaling to ensure the file data is flushed to disk
48 before the journal commit, to remove the small possibility of stale
49 data appearing in files after a crash.
50 It is architecture and endian neutral. It allows concurrent mounts on
52 nodes with different processors like x86, x86_64, IA64 and PPC64. It
53 handles little and big endian, 32-bit and 64-bit architectures.
54 It is feature rich. It supports indexed directories, metadata check‐
56 sums, extended attributes, POSIX ACLs, quotas, REFLINKs, sparse files,
57 unwritten extents and inline-data.
58 It is fully integrated with the mainline Linux kernel. The file system
60 was merged into Linux kernel 2.6.16 in early 2006.
61 It is quickly installed. It is available with almost all Linux distri‐
63 butions. The file system is on-disk compatible across all of them.
64 It is modular. The file system can be configured to operate with other
66 cluster stacks like Pacemaker and CMAN along with its own stack, O2CB.
67 It is easily configured. The O2CB cluster stack configuration involves
69 editing two files, one for cluster layout and the other for cluster
70 timeouts.
71 It is very efficient. The file system consumes very little resources.
73 It is used to store virtual machine images in limited memory environ‐
74 ments like Xen and KVM.
75 In summary, OCFS2 is an efficient, easily configured, modular, quickly
77 installed, fully integrated and compatible, feature-rich, architecture
78 and endian neutral, cache coherent, ordered data journaling, POSIX-com‐
79 pliant, shared disk cluster file system.
80
83 OCFS2 is a general-purpose shared-disk cluster file system for Linux
84 capable of providing both high performance and high availability.
85 As it provides local file system semantics, it can be used with almost
87 all applications. Cluster-aware applications can make use of cache-
88 coherent parallel I/Os from multiple nodes to scale out applications
89 easily. Other applications can make use of the clustering facilities to
90 fail-over running application in the event of a node failure.
91 The notable features of the file system are:
93 Tunable Block size
95 The file system supports block sizes of 512, 1K, 2K and 4K
96 bytes. 4KB is almost always recommended. This feature is avail‐
97 able in all releases of the file system.
98 Tunable Cluster size
101 A cluster size is also referred to as an allocation unit. The
102 file system supports cluster sizes of 4K, 8K, 16K, 32K, 64K,
103 128K, 256K, 512K and 1M bytes. For most use cases, 4KB is recom‐
104 mended. However, a larger value is recommended for volumes host‐
105 ing mostly very large files like database files, virtual machine
106 images, etc. A large cluster size allows the file system to
107 store large files more efficiently. This feature is available in
108 all releases of the file system.
109 Endian and Architecture neutral
112 The file system can be mounted concurrently on nodes having dif‐
113 ferent architectures. Like 32-bit, 64-bit, little-endian (x86,
114 x86_64, ia64) and big-endian (ppc64, s390x). This feature is
115 available in all releases of the file system.
116 Buffered, Direct, Asynchronous, Splice and Memory Mapped I/O modes
119 The file system supports all modes of I/O for maximum flexibil‐
120 ity and performance. It also supports cluster-wide shared
121 writeable mmap(2). The support for bufferred, direct and asyn‐
122 chronous I/O is available in all releases. The support for
123 splice I/O was added in Linux kernel 2.6.20 and for shared
124 writeable map(2) in 2.6.23.
125 Multiple Cluster Stacks
128 The file system includes a flexible framework to allow it to
129 function with userspace cluster stacks like Pacemaker (pcmk) and
130 CMAN (cman), its own in-kernel cluster stack o2cb and no cluster
131 stack.
132 The support for o2cb cluster stack is available in all releases.
134 The support for no cluster stack, or local mount, was added in
136 Linux kernel 2.6.20.
137 The support for userspace cluster stack was added in Linux ker‐
139 nel 2.6.26.
140 Journaling
143 The file system supports both ordered (default) and writeback
144 data journaling modes to provide file system consistency in the
145 event of power failure or system crash. It uses JBD2 in Linux
146 kernel 2.6.28 and later. It used JBD in earlier kernels.
147 Extent-based Allocations
150 The file system allocates and tracks space in ranges of clus‐
151 ters. This is unlike block based file systems that have to track
152 each and every block. This feature allows the file system to be
153 very efficient when dealing with both large volumes and large
154 files. This feature is available in all releases of the file
155 system.
156 Sparse files
159 Sparse files are files with holes. With this feature, the file
160 system delays allocating space until a write is issued to a
161 cluster. This feature was added in Linux kernel 2.6.22 and
162 requires enabling on-disk feature sparse.
163 Unwritten Extents
166 An unwritten extent is also referred to as user pre-allocation.
167 It allows an application to request a range of clusters to be
168 allocated, but not initialized, within a file. Pre-allocation
169 allows the file system to optimize the data layout with fewer,
170 larger extents. It also provides a performance boost, delaying
171 initialization until the user writes to the clusters. This fea‐
172 ture was added in Linux kernel 2.6.23 and requires enabling on-
173 disk feature unwritten.
174 Hole Punching
177 Hole punching allows an application to remove arbitrary allo‐
178 cated regions within a file. Creating holes, essentially. This
179 is more efficient than zeroing the same extents. This feature
180 is especially useful in virtualized environments as it allows a
181 block discard in a guest file system to be converted to a hole
182 punch in the host file system thus allowing users to reduce disk
183 space usage. This feature was added in Linux kernel 2.6.23 and
184 requires enabling on-disk features sparse and unwritten.
185 Inline-data
188 Inline data is also referred to as data-in-inode as it allows
189 storing small files and directories in the inode block. This not
190 only saves space but also has a positive impact on cold-cache
191 directory and file operations. The data is transparently moved
192 out to an extent when it no longer fits inside the inode block.
193 This feature was added in Linux kernel 2.6.24 and requires
194 enabling on-disk feature inline-data.
195 REFLINK
198 REFLINK is also referred to as fast copy. It allows users to
199 atomically (and instantly) copy regular files. In other words,
200 create multiple writeable snapshots of regular files. It is
201 called REFLINK because it looks and feels more like a (hard)
202 link(2) than a traditional snapshot. Like a link, it is a regu‐
203 lar user operation, subject to the security attributes of the
204 inode being reflinked and not to the super user privileges typi‐
205 cally required to create a snapshot. Like a link, it operates
206 within a file system. But unlike a link, it links the inodes at
207 the data extent level allowing each reflinked inode to grow
208 independently as and when written to. Up to four billion inodes
209 can share a data extent. This feature was added in Linux kernel
210 2.6.32 and requires enabling on-disk feature refcount.
211 Allocation Reservation
214 File contiguity plays an important role in file system perfor‐
215 mance. When a file is fragmented on disk, reading and writing to
216 the file involves many seeks, leading to lower throughput. Con‐
217 tiguous files, on the other hand, minimize seeks, allowing the
218 disks to perform IO at the maximum rate.
219 With allocation reservation, the file system reserves a window
221 in the bitmap for all extending files allowing each to grow as
222 contiguously as possible. As this extra space is not actually
223 allocated, it is available for use by other files if the need
224 arises. This feature was added in Linux kernel 2.6.35 and can
225 be tuned using the mount option resv_level.
226 Indexed Directories
229 An indexed directory allows users to perform quick lookups of a
230 file in very large directories. It also results in faster cre‐
231 ates and unlinks and thus provides better overall performance.
232 This feature was added in Linux kernel 2.6.30 and requires
233 enabling on-disk feature indexed-dirs.
234 File Attributes
237 This refers to EXT2-style file attributes, such as immutable,
238 modified using chattr(1) and queried using lsattr(1). This fea‐
239 ture was added in Linux kernel 2.6.19.
240 Extended Attributes
243 An extended attribute refers to a name:value pair than can be
244 associated with file system objects like regular files, directo‐
245 ries, symbolic links, etc. OCFS2 allows associating an unlimited
246 number of attributes per object. The attribute names can be up
247 to 255 bytes in length, terminated by the first NUL character.
248 While it is not required, printable names (ASCII) are recom‐
249 mended. The attribute values can be up to 64 KB of arbitrary
250 binary data. These attributes can be modified and listed using
251 standard Linux utilities setfattr(1) and getfattr(1). This fea‐
252 ture was added in Linux kernel 2.6.29 and requires enabling on-
253 disk feature xattr.
254 Metadata Checksums
257 This feature allows the file system to detect silent corruptions
258 in all metadata blocks like inodes and directories. This feature
259 was added in Linux kernel 2.6.29 and requires enabling on-disk
260 feature metaecc.
261 POSIX ACLs and Security Attributes
264 POSIX ACLs allows assigning fine-grained discretionary access
265 rights for files and directories. This security scheme is a lot
266 more flexible than the traditional file access permissions that
267 imposes a strict user-group-other model.
268 Security attributes allow the file system to support other secu‐
270 rity regimes like SELinux, SMACK, AppArmor, etc.
271 Both these security extensions were added in Linux kernel 2.6.29
273 and requires enabling on-disk feature xattr.
274 User and Group Quotas
277 This feature allows setting up usage quotas on user and group
278 basis by using the standard utilities like quota(1),
279 setquota(8), quotacheck(8), and quotaon(8). This feature was
280 added in Linux kernel 2.6.29 and requires enabling on-disk fea‐
281 tures usrquota and grpquota.
282 Unix File Locking
285 The Unix operating system has historically provided two system
286 calls to lock files. flock(2) or BSD locking and fcntl(2) or
287 POSIX locking. OCFS2 extends both file locks to the cluster.
288 File locks taken on one node interact with those taken on other
289 nodes.
290 The support for clustered flock(2) was added in Linux kernel
292 2.6.26. All flock(2) options are supported, including the ker‐
293 nels ability to cancel a lock request when an appropriate kill
294 signal is received by the user. This feature is supported with
295 all cluster-stacks including o2cb.
296 The support for clustered fcntl(2) was added in Linux kernel
298 2.6.28. But because it requires group communication to make the
299 locks coherent, it is only supported with userspace cluster
300 stacks, pcmk and cman and not with the default cluster stack
301 o2cb.
302 Comprehensive Tools Support
305 The file system has a comprehensive EXT3-style toolset that
306 tries to use similar parameters for ease-of-use. It includes
307 mkfs.ocfs2(8) (format), tunefs.ocfs2(8) (tune), fsck.ocfs2(8)
308 (check), debugfs.ocfs2(8) (debug), etc.
309 Online Resize
312 The file system can be dynamically grown using tunefs.ocfs2(8).
313 This feature was added in Linux kernel 2.6.25.
314
317 The O2CB cluster stack has a global heartbeat mode. It allows users to
318 specify heartbeat regions that are consistent across all nodes. The
319 cluster stack also allows online addition and removal of both nodes and
320 heartbeat regions.
321 o2cb(8) is the new cluster configuration utility. It is an easy to use
323 utility that allows users to create the cluster configuration on a node
324 that is not part of the cluster. It replaces the older utility
325 o2cb_ctl(8) which has being deprecated.
326 ocfs2console(8) has been obsoleted.
328 o2info(8) is a new utility that can be used to provide file system
330 information. It allows non-privileged users to see the enabled file
331 system features, block and cluster sizes, extended file stat, free
332 space fragmentation, etc.
333 o2hbmonitor(8) is a o2hb heartbeat monitor. It is an extremely light
335 weight utility that logs messages to the system logger once the heart‐
336 beat delay exceeds the warn threshold. This utility is useful in iden‐
337 tifying volumes encountering I/O delays.
338 debugfs.ocfs2(8) has some new commands. net_stats shows the o2net mes‐
340 sage times between various nodes. This is useful in identifying nodes
341 are that slowing down the cluster operations. stat_sysdir allows the
342 user to dump the entire system directory that can be used to debug
343 issues. grpextents dumps the complete free space fragmentation in the
344 cluster group allocator.
345 mkfs.ocfs2(8) now enables xattr, indexed-dirs, discontig-bg, refcount,
347 extended-slotmap and clusterinfo feature flags by default, in addition
348 to the older defaults, sparse, unwritten and inline-data.
349 mount.ocfs2(8) allows users to specify the level of cache coherency
351 between nodes. By default the file system operates in full coherency
352 mode that also serializes the direct I/Os. While this mode is techni‐
353 cally correct, it limits the I/O thruput in a clustered database. This
354 mount option allows the user to limit the cache coherency to only the
355 buffered I/Os to allow multiple nodes to do concurrent direct writes to
356 the same file. This feature works with Linux kernel 2.6.37 and later.
357
360 The OCFS2 development teams goes to great lengths to maintain compati‐
361 bility. It attempts to maintain both on-disk and network protocol com‐
362 patibility across all releases of the file system. It does so even
363 while adding new features that entail on-disk format and network proto‐
364 col changes. To do this successfully, it follows a few rules:
365 1. The on-disk format changes are managed by a set of feature flags
367 that can be turned on and off. The file system in kernel detects
368 these features during mount and continues only if it understands
369 all the features. Users encountering this have the option of either
370 disabling that feature or upgrading the file system to a newer
371 release.
372 2. The latest release of ocfs2-tools is compatible with all ver‐
374 sions of the file system. All utilities detect the features enabled
375 on disk and continue only if it understands all the features. Users
376 encountering this have to upgrade the tools to a newer release.
377 3. The network protocol version is negotiated by the nodes to
379 ensure all nodes understand the active protocol version.
380 FEATURE FLAGS
383 The feature flags are split into three categories, namely, Com‐
384 pat, Incompat and RO Compat.
385 Compat, or compatible, is a feature that the file system does
387 not need to fully understand to safely read/write to the volume.
388 An example of this is the backup-super feature that added the
389 capability to backup the super block in multiple locations in
390 the file system. As the backup super blocks are typically not
391 read nor written to by the file system, an older file system can
392 safely mount a volume with this feature enabled.
393 Incompat, or incompatible, is a feature that the file system
395 needs to fully understand to read/write to the volume. Most fea‐
396 tures fall under this category.
397 RO Compat, or read-only compatible, is a feature that the file
399 system needs to fully understand to write to the volume. Older
400 software can safely read a volume with this feature enabled. An
401 example of this would be user and group quotas. As quotas are
402 manipulated only when the file system is written to, older soft‐
403 ware can safely mount such volumes in read-only mode.
404 The list of feature flags, the version of the kernel it was
406 added in, the earliest version of the tools that understands it,
407 etc., is as follows:
408 ┌─────────────────────┬────────────────┬─────────────────┬───────────┬───────────┐
411 │Feature Flags │ Kernel Version │ Tools Version │ Category │ Hex Value │
412 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
413 │backup-super │ All │ ocfs2-tools 1.2 │ Compat │ 1 │
414 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
415 │strict-journal-super │ All │ All │ Compat │ 2 │
416 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
417 │local │ Linux 2.6.20 │ ocfs2-tools 1.2 │ Incompat │ 8 │
418 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
419 │sparse │ Linux 2.6.22 │ ocfs2-tools 1.4 │ Incompat │ 10 │
420 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
421 │inline-data │ Linux 2.6.24 │ ocfs2-tools 1.4 │ Incompat │ 40 │
422 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
423 │extended-slotmap │ Linux 2.6.27 │ ocfs2-tools 1.6 │ Incompat │ 100 │
424 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
425 │xattr │ Linux 2.6.29 │ ocfs2-tools 1.6 │ Incompat │ 200 │
426 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
427 │indexed-dirs │ Linux 2.6.30 │ ocfs2-tools 1.6 │ Incompat │ 400 │
428 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
429 │metaecc │ Linux 2.6.29 │ ocfs2-tools 1.6 │ Incompat │ 800 │
430 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
431 │refcount │ Linux 2.6.32 │ ocfs2-tools 1.6 │ Incompat │ 1000 │
432 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
433 │discontig-bg │ Linux 2.6.35 │ ocfs2-tools 1.6 │ Incompat │ 2000 │
434 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
435 │clusterinfo │ Linux 2.6.37 │ ocfs2-tools 1.8 │ Incompat │ 4000 │
436 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
437 │unwritten │ Linux 2.6.23 │ ocfs2-tools 1.4 │ RO Compat │ 1 │
438 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
439 │grpquota │ Linux 2.6.29 │ ocfs2-tools 1.6 │ RO Compat │ 2 │
440 ├─────────────────────┼────────────────┼─────────────────┼───────────┼───────────┤
441 │usrquota │ Linux 2.6.29 │ ocfs2-tools 1.6 │ RO Compat │ 4 │
442 └─────────────────────┴────────────────┴─────────────────┴───────────┴───────────┘
443 To query the features enabled on a volume, do:
445 $ o2info --fs-features /dev/sdf1
447 backup-super strict-journal-super sparse extended-slotmap inline-data xattr
448 indexed-dirs refcount discontig-bg clusterinfo unwritten
449 ENABLING AND DISABLING FEATURES
452 The format utility, mkfs.ocfs2(8), allows a user to enable and
454 disable specific features using the fs-features option. The fea‐
455 tures are provided as a comma separated list. The enabled fea‐
456 tures are listed as is. The disabled features are prefixed with
457 no. The example below shows the file system being formatted
458 with sparse disabled and inline-data enabled.
459 # mkfs.ocfs2 --fs-features=nosparse,inline-data /dev/sda1
461 After formatting, the users can toggle features using the tune
463 utility, tunefs.ocfs2(8). This is an offline operation. The
464 volume needs to be umounted across the cluster. The example
465 below shows the sparse feature being enabled and inline-data
466 disabled.
467 # tunefs.ocfs2 --fs-features=sparse,noinline-data /dev/sda1
469 Care should be taken before enabling and disabling features.
471 Users planning to use a volume with an older version of the file
472 system will be better of not enabling newer features as turning
473 disabling may not succeed.
474 An example would be disabling the sparse feature; this requires
476 filling every hole. The operation can only succeed if the file
477 system has enough free space.
478 DETECTING FEATURE INCOMPATIBILITY
481 Say one tries to mount a volume with an incompatible feature.
483 What happens then? How does one detect the problem? How does one
484 know the name of that incompatible feature?
485 To begin with, one should look for error messages in dmesg(8).
487 Mount failures that are due to an incompatible feature will
488 always result in an error message like the following:
489 ERROR: couldn't mount because of unsupported optional features (200).
491 Here the file system is unable to mount the volume due to an
493 unsupported optional feature. That means that that feature is an
494 Incompat feature. By referring to the table above, one can then
495 deduce that the user failed to mount a volume with the xattr
496 feature enabled. (The value in the error message is in hexadeci‐
497 mal.)
498 Another example of an error message due to incompatibility is as
500 follows:
501 ERROR: couldn't mount RDWR because of unsupported optional features (1).
503 Here the file system is unable to mount the volume in the RW
505 mode. That means that that feature is a RO Compat feature.
506 Another look at the table and it becomes apparent that the vol‐
507 ume had the unwritten feature enabled.
508 In both cases, the user has the option of disabling the feature.
510 In the second case, the user has the choice of mounting the vol‐
511 ume in the RO mode.
512
515 The OCFS2 software is split into two components, namely, kernel and
516 tools. The kernel component includes the core file system and the clus‐
517 ter stack, and is packaged along with the kernel. The tools component
518 is packaged as ocfs2-tools and needs to be specifically installed. It
519 provides utilities to format, tune, mount, debug and check the file
520 system.
521 To install ocfs2-tools, refer to the package handling utility in in
523 your distributions.
524 The next step is selecting a cluster stack. The options include:
526 A. No cluster stack, or local mount.
528 B. In-kernel o2cb cluster stack with local or global heartbeat.
530 C. Userspace cluster stacks pcmk or cman.
532 The file system allows changing cluster stacks easily using
534 tunefs.ocfs2(8). To list the cluster stacks stamped on the OCFS2 vol‐
535 umes, do:
536 # mounted.ocfs2 -d
538 Device Stack Cluster F UUID Label
539 /dev/sdb1 o2cb webcluster G DCDA2845177F4D59A0F2DCD8DE507CC3 hbvol1
540 /dev/sdc1 None 23878C320CF3478095D1318CB5C99EED localmount
541 /dev/sdd1 o2cb webcluster G 8AB016CD59FC4327A2CDAB69F08518E3 webvol
542 /dev/sdg1 o2cb webcluster G 77D95EF51C0149D2823674FCC162CF8B logsvol
543 /dev/sdh1 o2cb webcluster G BBA1DBD0F73F449384CE75197D9B7098 scratch
544 NON-CLUSTERED OR LOCAL MOUNT
547 To format a OCFS2 volume as a non-clustered (local) volume, do:
549 # mkfs.ocfs2 -L "mylabel" --fs-features=local /dev/sda1
551 To convert an existing clustered volume to a non-clustered vol‐
553 ume, do:
554 # tunefs.ocfs2 --fs-features=local /dev/sda1
556 Non-clustered volumes do not interact with the cluster stack.
558 One can have both clustered and non-clustered volumes mounted at
559 the same time.
560 While formatting a non-clustered volume, users should consider
562 the possibility of later converting that volume to a clustered
563 one. If there is a possibility of that, then the user should add
564 enough node-slots using the -N option. Adding node-slots during
565 format creates journals with large extents. If created later,
566 then the journals will be fragmented which is not good for per‐
567 formance.
568 CLUSTERED MOUNT WITH O2CB CLUSTER STACK
571 Only one of the two heartbeat mode can be active at any one
573 time. Changing heartbeat modes is an offline operation.
574 Both heartbeat modes require /etc/ocfs2/cluster.conf and
576 /etc/sysconfig/o2cb to be populated as described in ocfs2.clus‐
577 ter.conf(5) and o2cb.sysconfig(5) respectively. The only differ‐
578 ence in set up between the two modes is that global requires
579 heartbeat devices to be configured whereas local does not.
580 Refer o2cb(7) for more information.
582 LOCAL HEARTBEAT
585 This is the default heartbeat mode. The user needs to
586 populate the configuration files as described in
587 ocfs2.cluster.conf(5) and o2cb.sysconfig(5). In this
588 mode, the cluster stack heartbeats on all mounted vol‐
589 umes. Thus, one does not have to specify heartbeat
590 devices in cluster.conf.
591 Once configured, the o2cb cluster stack can be onlined
593 and offlined as follows:
594 # service o2cb online
596 Setting cluster stack "o2cb": OK
597 Registering O2CB cluster "webcluster": OK
598 Setting O2CB cluster timeouts : OK
599 # service o2cb offline
601 Clean userdlm domains: OK
602 Stopping O2CB cluster webcluster: OK
603 Unregistering O2CB cluster "webcluster": OK
604 GLOBAL HEARTBEAT
607 The configuration is similar to local heartbeat. The one
608 additional step in this mode is that it requires heart‐
609 beat devices to be also configured.
610 These heartbeat devices are OCFS2 formatted volumes with
612 global heartbeat enabled on disk. These volumes can later
613 be mounted and used as clustered file systems.
614 The steps to format a volume with global heartbeat
616 enabled is listed in o2cb(7). Also listed there is list‐
617 ing all volumes with the cluster stack stamped on disk.
618 In this mode, the heartbeat is started when the cluster
620 is onlined and stopped when the cluster is offlined.
621 # service o2cb online
623 Setting cluster stack "o2cb": OK
624 Registering O2CB cluster "webcluster": OK
625 Setting O2CB cluster timeouts : OK
626 Starting global heartbeat for cluster "webcluster": OK
627 # service o2cb offline
629 Clean userdlm domains: OK
630 Stopping global heartbeat on cluster "webcluster": OK
631 Stopping O2CB cluster webcluster: OK
632 Unregistering O2CB cluster "webcluster": OK
633 # service o2cb status
635 Driver for "configfs": Loaded
636 Filesystem "configfs": Mounted
637 Stack glue driver: Loaded
638 Stack plugin "o2cb": Loaded
639 Driver for "ocfs2_dlmfs": Loaded
640 Filesystem "ocfs2_dlmfs": Mounted
641 Checking O2CB cluster "webcluster": Online
642 Heartbeat dead threshold: 31
643 Network idle timeout: 30000
644 Network keepalive delay: 2000
645 Network reconnect delay: 2000
646 Heartbeat mode: Global
647 Checking O2CB heartbeat: Active
648 77D95EF51C0149D2823674FCC162CF8B /dev/sdg1
649 Nodes in O2CB cluster: 92 96
650 CLUSTERED MOUNT WITH USERSPACE CLUSTER STACK
654 Configure and online the userspace stack pcmk or cman before
656 using tunefs.ocfs2(8) to update the cluster stack on disk.
657 # tunefs.ocfs2 --update-cluster-stack /dev/sdd1
659 Updating on-disk cluster information to match the running cluster.
660 DANGER: YOU MUST BE ABSOLUTELY SURE THAT NO OTHER NODE IS USING THIS
661 FILESYSTEM BEFORE MODIFYING ITS CLUSTER CONFIGURATION.
662 Update the on-disk cluster information? y
663 Refer to the cluster stack documentation for information on
665 starting and stopping the cluster stack.
666
669 This sections lists the utilities that are used to manage the OCFS2
670 file systems. This includes tools to format, tune, check, mount, debug
671 the file system. Each utility has a man page that lists its capabili‐
672 ties in detail.
673 mkfs.ocfs2(8)
676 This is the file system format utility. All volumes have to be
677 formatted prior to its use. As this utility overwrites the vol‐
678 ume, use it with care. Double check to ensure the volume is not
679 in use on any node in the cluster.
680 As a precaution, the utility will abort if the volume is locally
682 mounted. It also detects use across the cluster if used by
683 OCFS2. But these checks are not comprehensive and can be over‐
684 ridden. So use it with care.
685 While it is not always required, the cluster should be online.
687 tunefs.ocfs2(8)
690 This is the file system tune utility. It allows users to change
691 certain on-disk parameters like label, uuid, number of node-
692 slots, volume size and the size of the journals. It also allows
693 turning on and off the file system features as listed above.
694 This utility requires the cluster to be online.
696 fsck.ocfs2(8)
699 This is the file system check utility. It detects and fixes on-
700 disk errors. All the check codes and their fixes are listed in
701 fsck.ocfs2.checks(8).
702 This utility requires the cluster to be online to ensure the
704 volume is not in use on another node and to prevent the volume
705 from being mounted for the duration of the check.
706 mount.ocfs2(8)
709 This is the file system mount utility. It is invoked indirectly
710 by the mount(8) utility.
711 This utility detects the cluster status and aborts if the clus‐
713 ter is offline or does not match the cluster stamped on disk.
714 o2cluster(8)
717 This is the file system cluster stack update utility. It allows
718 the users to update the on-disk cluster stack to the one pro‐
719 vided.
720 This utility only updates the disk if the utility is reasonably
722 assured that the file system is not in use on any node.
723 o2info(1)
726 This is the file system information utility. It provides infor‐
727 mation like the features enabled on disk, block size, cluster
728 size, free space fragmentation, etc.
729 It can be used by both privileged and non-privileged users.
731 Users having read permission on the device can provide the path
732 to the device. Other users can provide the path to a file on a
733 mounted file system.
734 debugfs.ocfs2(8)
737 This is the file system debug utility. It allows users to exam‐
738 ine all file system structures including walking directory
739 structures, displaying inodes, backing up files, etc., without
740 mounting the file system.
741 This utility requires the user to have read permission on the
743 device.
744 o2image(8)
747 This is the file system image utility. It allows users to copy
748 the file system metadata skeleton, including the inodes, direc‐
749 tories, bitmaps, etc. As it excludes data, it shrinks the size
750 of the file system tremendously.
751 The image file created can be used in debugging on-disk corrup‐
753 tions.
754 mounted.ocfs2(8)
757 This is the file system detect utility. It detects all OCFS2
758 volumes in the system and lists its label, uuid and cluster
759 stack.
760
763 This sections lists the utilities that are used to manage O2CB cluster
764 stack. Each utility has a man page that lists its capabilities in
765 detail.
766 o2cb(8)
768 This is the cluster configuration utility. It allows users to
769 update the cluster configuration by adding and removing nodes
770 and heartbeat regions. This utility is used by the o2cb init
771 script to online and offline the cluster.
772 This is a new utility and replaces o2cb_ctl(8) which has been
774 deprecated.
775 ocfs2_hb_ctl(8)
778 This is the cluster heartbeat utility. It allows users to start
779 and stop local heartbeat. This utility is invoked by
780 mount.ocfs2(8) and should not be invoked directly by the user.
781 o2hbmonitor(8)
784 This is the disk heartbeat monitor. It tracks the elapsed time
785 since the last heartbeat and logs warnings once that time
786 exceeds the warn threshold.
787
790 This section includes some useful notes that may prove helpful to the
791 user.
792 BALANCED CLUSTER
794 A cluster is a computer. This is a fact and not a slogan. What
795 this means is that an errant node in the cluster can affect the
796 behavior of other nodes. If one node is slow, the cluster opera‐
797 tions will slow down on all nodes. To prevent that, it is best
798 to have a balanced cluster. This is a cluster that has equally
799 powered and loaded nodes.
800 The standard recommendation for such clusters is to have identi‐
802 cal hardware and software across all the nodes. However, that is
803 not a hard and fast rule. After all, we have taken the effort to
804 ensure that OCFS2 works in a mixed architecture environment.
805 If one uses OCFS2 in a mixed architecture environment, try to
807 ensure that the nodes are equally powered and loaded. The use of
808 a load balancer can assist with the latter. Power refers to the
809 number of processors, speed, amount of memory, I/O throughput,
810 network bandwidth, etc. In reality, having equally powered het‐
811 erogeneous nodes is not always practical. In that case, make the
812 lower node numbers more powerful than the higher node numbers.
813 The O2CB cluster stack favors lower node numbers in all of its
814 tiebreaking logic.
815 This is not to suggest you should add a single core node in a
817 cluster of quad cores. No amount of node number juggling will
818 help you there.
819 FILE DELETION
822 In Linux, rm(1) removes the directory entry. It does not neces‐
823 sarily delete the corresponding inode. But by removing the
824 directory entry, it gives the illusion that the inode has been
825 deleted. This puzzles users when they do not see a correspond‐
826 ing up-tick in the reported free space. The reason is that
827 inode deletion has a few more hurdles to cross.
828 First is the hard link count, that indicates the number of
830 directory entries pointing to that inode. As long as an inode
831 has one or more directory entries pointing to it, it cannot be
832 deleted. The file system has to wait for the removal of all
833 those directory entries. In other words, wait for that count to
834 drop to zero.
835 The second hurdle is the POSIX semantics allowing files to be
837 unlinked even while they are in-use. In OCFS2, that translates
838 to in-use across the cluster. The file system has to wait for
839 all processes across the cluster to stop using the inode.
840 Once these conditions are met, the inode is deleted and the
842 freed space is visible after the next sync.
843 Now the amount of space freed depends on the allocation. Only
845 space that is actually allocated to that inode is freed. The
846 example below shows a sparsely allocated file of size 51TB of
847 which only 2.4GB is actually allocated.
848 $ ls -lsh largefile
850 2.4G -rw-r--r-- 1 mark mark 51T Sep 29 15:04 largefile
851 Furthermore, for reflinked files, only private extents are
853 freed. Shared extents are freed when the last inode accessing
854 it, is deleted. The example below shows a 4GB file that shares
855 3GB with other reflinked files. Deleting it will increase the
856 free space by 1GB. However, if it is the only remaining file
857 accessing the shared extents, the full 4G will be freed. (More
858 information on the shared-du(1) utility is provided below.)
859 $ shared-du -m -c --shared-size reflinkedfile
861 4000 (3000) reflinkedfile
862 The deletion itself is a multi-step process. Once the hard link
864 count falls to zero, the inode is moved to the orphan_dir system
865 directory where it remains until the last process, across the
866 cluster, stops using the inode. Then the file system frees the
867 extents and adds the freed space count to the truncate_log sys‐
868 tem file where it remains until the next sync. The freed space
869 is made visible to the user only after that sync.
870 DIRECTORY LISTING
873 ls(1) may be a simple command, but it is not cheap. What is
874 expensive is not the part where it reads the directory listing,
875 but the second part where it reads all the inodes, also referred
876 as an inode stat(2). If the inodes are not in cache, this can
877 entail disk I/O. Now, while a cold cache inode stat(2) is
878 expensive in all file systems, it is especially so in a clus‐
879 tered file system as it needs to take a cluster lock on each
880 inode.
881 A hot cache stat(2), on the other hand, has shown to perform on
883 OCFS2 like it does on EXT3.
884 In other words, the second ls(1) will be quicker than the first.
886 However, it is not guaranteed. Say you have a million files in a
887 file system and not enough kernel memory to cache all the
888 inodes. In that case, each ls(1) will involve some cold cache
889 stat(2)s.
890 ALLOCATION RESERVATION
893 Allocation reservation allows multiple concurrently extending
894 files to grow as contiguously as possible. One way to demon‐
895 strate its functioning is to run a script that extends multiple
896 files in a circular order. The script below does that by writing
897 one hundred 4KB chunks to four files, one after another.
898 $ for i in $(seq 0 99);
900 > do
901 > for j in $(seq 4);
902 > do
903 > dd if=/dev/zero of=file$j bs=4K count=1 seek=$i;
904 > done;
905 > done;
906 When run on a system running Linux kernel 2.6.34 or earlier, we
908 end up with files with 100 extents each. That is full fragmenta‐
909 tion. As the files are being extended one after another, the on-
910 disk allocations are fully interleaved.
911 $ filefrag file1 file2 file3 file4
913 file1: 100 extents found
914 file2: 100 extents found
915 file3: 100 extents found
916 file4: 100 extents found
917 When run on a system running Linux kernel 2.6.35 or later, we
919 see files with 7 extents each. That is a lot fewer than before.
920 Fewer extents mean more on-disk contiguity and that always leads
921 to better overall performance.
922 $ filefrag file1 file2 file3 file4
924 file1: 7 extents found
925 file2: 7 extents found
926 file3: 7 extents found
927 file4: 7 extents found
928 REFLINK OPERATION
931 This feature allows a user to create a writeable snapshot of a
932 regular file. In this operation, the file system creates a new
933 inode with the same extent pointers as the original inode. Mul‐
934 tiple inodes are thus able to share data extents. This adds a
935 twist in file system administration because none of the existing
936 file system utilities in Linux expect this behavior. du(1), a
937 utility to used to compute file space usage, simply adds the
938 blocks allocated to each inode. As it does not know about shared
939 extents, it over estimates the space used. Say, we have a 5GB
940 file in a volume having 42GB free.
941 $ ls -l
943 total 5120000
944 -rw-r--r-- 1 jeff jeff 5242880000 Sep 24 17:15 myfile
945 $ du -m myfile*
947 5000 myfile
948 $ df -h .
950 Filesystem Size Used Avail Use% Mounted on
951 /dev/sdd1 50G 8.2G 42G 17% /ocfs2
952 If we were to reflink it 4 times, we would expect the directory
954 listing to report five 5GB files, but the df(1) to report no
955 loss of available space. du(1), on the other hand, would report
956 the disk usage to climb to 25GB.
957 $ reflink myfile myfile-ref1
959 $ reflink myfile myfile-ref2
960 $ reflink myfile myfile-ref3
961 $ reflink myfile myfile-ref4
962 $ ls -l
964 total 25600000
965 -rw-r--r-- 1 jeff jeff 5242880000 Sep 24 17:15 myfile
966 -rw-r--r-- 1 jeff jeff 5242880000 Sep 24 17:16 myfile-ref1
967 -rw-r--r-- 1 jeff jeff 5242880000 Sep 24 17:16 myfile-ref2
968 -rw-r--r-- 1 jeff jeff 5242880000 Sep 24 17:16 myfile-ref3
969 -rw-r--r-- 1 jeff jeff 5242880000 Sep 24 17:16 myfile-ref4
970 $ df -h .
972 Filesystem Size Used Avail Use% Mounted on
973 /dev/sdd1 50G 8.2G 42G 17% /ocfs2
974 $ du -m myfile*
976 5000 myfile
977 5000 myfile-ref1
978 5000 myfile-ref2
979 5000 myfile-ref3
980 5000 myfile-ref4
981 25000 total
982 Enter shared-du(1), a shared extent-aware du. This utility
984 reports the shared extents per file in parenthesis and the over‐
985 all footprint. As expected, it lists the overall footprint at
986 5GB. One can view the details of the extents using shared-file‐
987 frag(1). Both these utilities are available at http://oss.ora‐
988 cle.com/~smushran/reflink-tools/. We are currently in the
989 process of pushing the changes to the upstream maintainers of
990 these utilities.
991 $ shared-du -m -c --shared-size myfile*
993 5000 (5000) myfile
994 5000 (5000) myfile-ref1
995 5000 (5000) myfile-ref2
996 5000 (5000) myfile-ref3
997 5000 (5000) myfile-ref4
998 25000 total
999 5000 footprint
1000 # shared-filefrag -v myfile
1002 Filesystem type is: 7461636f
1003 File size of myfile is 5242880000 (1280000 blocks, blocksize 4096)
1004 ext logical physical expected length flags
1005 0 0 2247937 8448
1006 1 8448 2257921 2256384 30720
1007 2 39168 2290177 2288640 30720
1008 3 69888 2322433 2320896 30720
1009 4 100608 2354689 2353152 30720
1010 7 192768 2451457 2449920 30720
1011 . . .
1012 37 1073408 2032129 2030592 30720 shared
1013 38 1104128 2064385 2062848 30720 shared
1014 39 1134848 2096641 2095104 30720 shared
1015 40 1165568 2128897 2127360 30720 shared
1016 41 1196288 2161153 2159616 30720 shared
1017 42 1227008 2193409 2191872 30720 shared
1018 43 1257728 2225665 2224128 22272 shared,eof
1019 myfile: 44 extents found
1020 DATA COHERENCY
1023 One of the challenges in a shared file system is data coherency
1024 when multiple nodes are writing to the same set of files. NFS,
1025 for example, provides close-to-open data coherency that results
1026 in the data being flushed to the server when the file is closed
1027 on the client. This leaves open a wide window for stale data
1028 being read on another node.
1029 A simple test to check the data coherency of a shared file sys‐
1031 tem involves concurrently appending the same file. Like running
1032 "uname -a >>/dir/file" using a parallel distributed shell like
1033 dsh or pconsole. If coherent, the file will contain the results
1034 from all nodes.
1035 # dsh -R ssh -w node32,node33,node34,node35 "uname -a >> /ocfs2/test"
1037 # cat /ocfs2/test
1038 Linux node32 2.6.32-10 #1 SMP Fri Sep 17 17:51:41 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
1039 Linux node35 2.6.32-10 #1 SMP Fri Sep 17 17:51:41 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
1040 Linux node33 2.6.32-10 #1 SMP Fri Sep 17 17:51:41 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
1041 Linux node34 2.6.32-10 #1 SMP Fri Sep 17 17:51:41 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
1042 OCFS2 is a fully cache coherent cluster file system.
1044 DISCONTIGUOUS BLOCK GROUP
1047 Most file systems pre-allocate space for inodes during format.
1048 OCFS2 dynamically allocates this space when required.
1049 However, this dynamic allocation has been problematic when the
1051 free space is very fragmented, because the file system required
1052 the inode and extent allocators to grow in contiguous fixed-size
1053 chunks.
1054 The discontiguous block group feature takes care of this problem
1056 by allowing the allocators to grow in smaller, variable-sized
1057 chunks.
1058 This feature was added in Linux kernel 2.6.35 and requires
1060 enabling on-disk feature discontig-bg.
1061 BACKUP SUPER BLOCKS
1064 A file system super block stores critical information that is
1065 hard to recreate. In OCFS2, it stores the block size, cluster
1066 size, and the locations of the root and system directories,
1067 among other things. As this block is close to the start of the
1068 disk, it is very susceptible to being overwritten by an errant
1069 write. Say, dd if=file of=/dev/sda1.
1070 Backup super blocks are copies of the super block. These blocks
1072 are dispersed in the volume to minimize the chances of being
1073 overwritten. On the small chance that the original gets cor‐
1074 rupted, the backups are available to scan and fix the corrup‐
1075 tion.
1076 mkfs.ocfs2(8) enables this feature by default. Users can disable
1078 this by specifying --fs-features=nobackup-super during format.
1079 o2info(1) can be used to view whether the feature has been
1081 enabled on a device.
1082 # o2info --fs-features /dev/sdb1
1084 backup-super strict-journal-super sparse extended-slotmap inline-data xattr
1085 indexed-dirs refcount discontig-bg clusterinfo unwritten
1086 In OCFS2, the super block is on the third block. The backups are
1088 located at the 1G, 4G, 16G, 64G, 256G and 1T byte offsets. The
1089 actual number of backup blocks depends on the size of the
1090 device. The super block is not backed up on devices smaller than
1091 1GB.
1092 fsck.ocfs2(8) refers to these six offsets by numbers, 1 to 6.
1094 Users can specify any backup with the -r option to recover the
1095 volume. The example below uses the second backup. If successful,
1096 fsck.ocfs2(8) overwrites the corrupted super block with the
1097 backup.
1098 # fsck.ocfs2 -f -r 2 /dev/sdb1
1100 fsck.ocfs2 1.8.0
1101 [RECOVER_BACKUP_SUPERBLOCK] Recover superblock information from backup block#1048576? <n> y
1102 Checking OCFS2 filesystem in /dev/sdb1:
1103 Label: webhome
1104 UUID: B3E021A2A12B4D0EB08E9E986CDC7947
1105 Number of blocks: 13107196
1106 Block size: 4096
1107 Number of clusters: 13107196
1108 Cluster size: 4096
1109 Number of slots: 8
1110 /dev/sdb1 was run with -f, check forced.
1112 Pass 0a: Checking cluster allocation chains
1113 Pass 0b: Checking inode allocation chains
1114 Pass 0c: Checking extent block allocation chains
1115 Pass 1: Checking inodes and blocks.
1116 Pass 2: Checking directory entries.
1117 Pass 3: Checking directory connectivity.
1118 Pass 4a: checking for orphaned inodes
1119 Pass 4b: Checking inodes link counts.
1120 All passes succeeded.
1121 SYNTHETIC FILE SYSTEMS
1124 The OCFS2 development effort included two synthetic file sys‐
1125 tems, configfs and dlmfs. It also makes use of a third, debugfs.
1126 configfs
1129 configfs has since been accepted as a generic kernel com‐
1130 ponent and is also used by netconsole and fs/dlm. OCFS2
1131 tools use it to communicate the list of nodes in the
1132 cluster, details of the heartbeat device, cluster time‐
1133 outs, and so on to the in-kernel node manager. The o2cb
1134 init script mounts this file system at /sys/kernel/con‐
1135 fig.
1136 dlmfs dlmfs exposes the in-kernel o2dlm to the user-space.
1139 While it was developed primarily for OCFS2 tools, it has
1140 seen usage by others looking to add a cluster locking
1141 dimension in their applications. Users interested in
1142 doing the same should look at the libo2dlm library pro‐
1143 vided by ocfs2-tools. The o2cb init script mounts this
1144 file system at /dlm.
1145 debugfs
1148 OCFS2 uses debugfs to expose its in-kernel information to
1149 user space. For example, listing the file system cluster
1150 locks, dlm locks, dlm state, o2net state, etc. Users can
1151 access the information by mounting the file system at
1152 /sys/kernel/debug. To automount, add the following to
1153 /etc/fstab: debugfs /sys/kernel/debug debugfs defaults 0
1154 0
1155 DISTRIBUTED LOCK MANAGER
1158 One of the key technologies in a cluster is the lock manager,
1159 which maintains the locking state of all resources across the
1160 cluster. An easy implementation of a lock manager involves des‐
1161 ignating one node to handle everything. In this model, if a node
1162 wanted to acquire a lock, it would send the request to the lock
1163 manager. However, this model has a weakness: lock manager’s
1164 death causes the cluster to seize up.
1165 A better model is one where all nodes manage a subset of the
1167 lock resources. Each node maintains enough information for all
1168 the lock resources it is interested in. On event of a node
1169 death, the remaining nodes pool in the information to recon‐
1170 struct the lock state maintained by the dead node. In this
1171 scheme, the locking overhead is distributed amongst all the
1172 nodes. Hence, the term distributed lock manager.
1173 O2DLM is a distributed lock manager. It is based on the specifi‐
1175 cation titled "Programming Locking Application" written by
1176 Kristin Thomas and is available at the following link.
1177 http://opendlm.sourceforge.net/cvsmirror/opendlm/docs/dlm‐
1178 book_final.pdf
1179 DLM DEBUGGING
1182 O2DLM has a rich debugging infrastructure that allows it to show
1183 the state of the lock manager, all the lock resources, among
1184 other things. The figure below shows the dlm state of a nine-
1185 node cluster that has just lost three nodes: 12, 32, and 35. It
1186 can be ascertained that node 7, the recovery master, is cur‐
1187 rently recovering node 12 and has received the lock states of
1188 the dead node from all other live nodes.
1189 # cat /sys/kernel/debug/o2dlm/45F81E3B6F2B48CCAAD1AE7945AB2001/dlm_state
1191 Domain: 45F81E3B6F2B48CCAAD1AE7945AB2001 Key: 0x10748e61
1192 Thread Pid: 24542 Node: 7 State: JOINED
1193 Number of Joins: 1 Joining Node: 255
1194 Domain Map: 7 31 33 34 40 50
1195 Live Map: 7 31 33 34 40 50
1196 Lock Resources: 48850 (439879)
1197 MLEs: 0 (1428625)
1198 Blocking: 0 (1066000)
1199 Mastery: 0 (362625)
1200 Migration: 0 (0)
1201 Lists: Dirty=Empty Purge=Empty PendingASTs=Empty PendingBASTs=Empty
1202 Purge Count: 0 Refs: 1
1203 Dead Node: 12
1204 Recovery Pid: 24543 Master: 7 State: ACTIVE
1205 Recovery Map: 12 32 35
1206 Recovery Node State:
1207 7 - DONE
1208 31 - DONE
1209 33 - DONE
1210 34 - DONE
1211 40 - DONE
1212 50 - DONE
1213 The figure below shows the state of a dlm lock resource that is
1215 mastered (owned) by node 25, with 6 locks in the granted queue
1216 and node 26 holding the EX (writelock) lock on that resource.
1217 # debugfs.ocfs2 -R "dlm_locks M000000000000000022d63c00000000" /dev/sda1
1219 Lockres: M000000000000000022d63c00000000 Owner: 25 State: 0x0
1220 Last Used: 0 ASTs Reserved: 0 Inflight: 0 Migration Pending: No
1221 Refs: 8 Locks: 6 On Lists: None
1222 Reference Map: 26 27 28 94 95
1223 Lock-Queue Node Level Conv Cookie Refs AST BAST Pending-Action
1224 Granted 94 NL -1 94:3169409 2 No No None
1225 Granted 28 NL -1 28:3213591 2 No No None
1226 Granted 27 NL -1 27:3216832 2 No No None
1227 Granted 95 NL -1 95:3178429 2 No No None
1228 Granted 25 NL -1 25:3513994 2 No No None
1229 Granted 26 EX -1 26:3512906 2 No No None
1230 The figure below shows a lock from the file system perspective.
1232 Specifically, it shows a lock that is in the process of being
1233 upconverted from a NL to EX. Locks in this state are are
1234 referred to in the file system as busy locks and can be listed
1235 using the debugfs.ocfs2 command, "fs_locks -B".
1236 # debugfs.ocfs2 -R "fs_locks -B" /dev/sda1
1238 Lockres: M000000000000000000000b9aba12ec Mode: No Lock
1239 Flags: Initialized Attached Busy
1240 RO Holders: 0 EX Holders: 0
1241 Pending Action: Convert Pending Unlock Action: None
1242 Requested Mode: Exclusive Blocking Mode: No Lock
1243 PR > Gets: 0 Fails: 0 Waits Total: 0us Max: 0us Avg: 0ns
1244 EX > Gets: 1 Fails: 0 Waits Total: 544us Max: 544us Avg: 544185ns
1245 Disk Refreshes: 1
1246 With this debugging infrastructure in place, users can debug
1248 hang issues as follows:
1249 * Dump the busy fs locks for all the OCFS2 volumes on the
1251 node with hanging processes. If no locks are found, then the
1252 problem is not related to O2DLM.
1253 * Dump the corresponding dlm lock for all the busy fs locks.
1255 Note down the owner (master) of all the locks.
1256 * Dump the dlm locks on the master node for each lock.
1258 At this stage, one should note that the hanging node is waiting
1260 to get an AST from the master. The master, on the other hand,
1261 cannot send the AST until the current holder has down converted
1262 that lock, which it will do upon receiving a Blocking AST. How‐
1263 ever, a node can only down convert if all the lock holders have
1264 stopped using that lock. After dumping the dlm lock on the mas‐
1265 ter node, identify the current lock holder and dump both the dlm
1266 and fs locks on that node.
1267 The trick here is to see whether the Blocking AST message has
1269 been relayed to file system. If not, the problem is in the dlm
1270 layer. If it has, then the most common reason would be a lock
1271 holder, the count for which is maintained in the fs lock.
1272 At this stage, printing the list of process helps.
1274 $ ps -e -o pid,stat,comm,wchan=WIDE-WCHAN-COLUMN
1276 Make a note of all D state processes. At least one of them is
1278 responsible for the hang on the first node.
1279 The challenge then is to figure out why those processes are
1281 hanging. Failing that, at least get enough information (like
1282 alt-sysrq t output) for the kernel developers to review. What
1283 to do next depends on where the process is hanging. If it is
1284 waiting for the I/O to complete, the problem could be anywhere
1285 in the I/O subsystem, from the block device layer through the
1286 drivers to the disk array. If the hang concerns a user lock
1287 (flock(2)), the problem could be in the user’s application. A
1288 possible solution could be to kill the holder. If the hang is
1289 due to tight or fragmented memory, free up some memory by
1290 killing non-essential processes.
1291 The thing to note is that the symptom for the problem was on one
1293 node but the cause is on another. The issue can only be resolved
1294 on the node holding the lock. Sometimes, the best solution will
1295 be to reset that node. Once killed, the O2DLM recovery process
1296 will clear all locks owned by the dead node and let the cluster
1297 continue to operate. As harsh as that sounds, at times it is the
1298 only solution. The good news is that, by following the trail,
1299 you now have enough information to file a bug and get the real
1300 issue resolved.
1301 NFS EXPORTING
1304 OCFS2 volumes can be exported as NFS volumes. This support is
1305 limited to NFS version 3, which translates to Linux kernel ver‐
1306 sion 2.4 or later.
1307 If the version of the Linux kernel on the system exporting the
1309 volume is older than 2.6.30, then the NFS clients must mount the
1310 volumes using the nordirplus mount option. This disables the
1311 READDIRPLUS RPC call to workaround a bug in NFSD, detailed in
1312 the following link:
1313 http://oss.oracle.com/pipermail/ocfs2-announce/2008-June/000025.html
1315 Users running NFS version 2 can export the volume after having
1317 disabled subtree checking (mount option no_subtree_check). Be
1318 warned, disabling the check has security implications (docu‐
1319 mented in the exports(5) man page) that users must evaluate on
1320 their own.
1321 FILE SYSTEM LIMITS
1324 OCFS2 has no intrinsic limit on the total number of files and
1325 directories in the file system. In general, it is only limited
1326 by the size of the device. But there is one limit imposed by the
1327 current filesystem. It can address at most four billion clus‐
1328 ters. A file system with 1MB cluster size can go up to 4PB,
1329 while a file system with a 4KB cluster size can address up to
1330 16TB.
1331 SYSTEM OBJECTS
1334 The OCFS2 file system stores its internal meta-data, including
1335 bitmaps, journals, etc., as system files. These are grouped in a
1336 system directory. These files and directories are not accessible
1337 via the file system interface but can be viewed using the
1338 debugfs.ocfs2(8) tool.
1339 To list the system directory (referred to as double-slash), do:
1341 # debugfs.ocfs2 -R "ls -l //" /dev/sde1
1343 66 drwxr-xr-x 10 0 0 3896 19-Jul-2011 13:36 .
1344 66 drwxr-xr-x 10 0 0 3896 19-Jul-2011 13:36 ..
1345 67 -rw-r--r-- 1 0 0 0 19-Jul-2011 13:36 bad_blocks
1346 68 -rw-r--r-- 1 0 0 1179648 19-Jul-2011 13:36 global_inode_alloc
1347 69 -rw-r--r-- 1 0 0 4096 19-Jul-2011 14:35 slot_map
1348 70 -rw-r--r-- 1 0 0 1048576 19-Jul-2011 13:36 heartbeat
1349 71 -rw-r--r-- 1 0 0 53686960128 19-Jul-2011 13:36 global_bitmap
1350 72 drwxr-xr-x 2 0 0 3896 25-Jul-2011 15:05 orphan_dir:0000
1351 73 drwxr-xr-x 2 0 0 3896 19-Jul-2011 13:36 orphan_dir:0001
1352 74 -rw-r--r-- 1 0 0 8388608 19-Jul-2011 13:36 extent_alloc:0000
1353 75 -rw-r--r-- 1 0 0 8388608 19-Jul-2011 13:36 extent_alloc:0001
1354 76 -rw-r--r-- 1 0 0 121634816 19-Jul-2011 13:36 inode_alloc:0000
1355 77 -rw-r--r-- 1 0 0 0 19-Jul-2011 13:36 inode_alloc:0001
1356 77 -rw-r--r-- 1 0 0 268435456 19-Jul-2011 13:36 journal:0000
1357 79 -rw-r--r-- 1 0 0 268435456 19-Jul-2011 13:37 journal:0001
1358 80 -rw-r--r-- 1 0 0 0 19-Jul-2011 13:36 local_alloc:0000
1359 81 -rw-r--r-- 1 0 0 0 19-Jul-2011 13:36 local_alloc:0001
1360 82 -rw-r--r-- 1 0 0 0 19-Jul-2011 13:36 truncate_log:0000
1361 83 -rw-r--r-- 1 0 0 0 19-Jul-2011 13:36 truncate_log:0001
1362 The file names that end with numbers are slot specific and are
1364 referred to as node-local system files. The set of node-local
1365 files used by a node can be determined from the slot map. To
1366 list the slot map, do:
1367 # debugfs.ocfs2 -R "slotmap" /dev/sde1
1369 Slot# Node#
1370 0 32
1371 1 35
1372 2 40
1373 3 31
1374 4 34
1375 5 33
1376 For more information, refer to the OCFS2 support guides avail‐
1378 able in the Documentation section at http://oss.ora‐
1379 cle.com/projects/ocfs2.
1380 HEARTBEAT, QUORUM, AND FENCING
1383 Heartbeat is an essential component in any cluster. It is
1384 charged with accurately designating nodes as dead or alive. A
1385 mistake here could lead to a cluster hang or a corruption.
1386 o2hb is the disk heartbeat component of o2cb. It periodically
1388 updates a timestamp on disk, indicating to others that this node
1389 is alive. It also reads all the timestamps to identify other
1390 live nodes. Other cluster components, like o2dlm and o2net, use
1391 the o2hb service to get node up and down events.
1392 The quorum is the group of nodes in a cluster that is allowed to
1394 operate on the shared storage. When there is a failure in the
1395 cluster, nodes may be split into groups that can communicate in
1396 their groups and with the shared storage but not between groups.
1397 o2quo determines which group is allowed to continue and initi‐
1398 ates fencing of the other group(s).
1399 Fencing is the act of forcefully removing a node from a cluster.
1401 A node with OCFS2 mounted will fence itself when it realizes
1402 that it does not have quorum in a degraded cluster. It does this
1403 so that other nodes won’t be stuck trying to access its
1404 resources.
1405 o2cb uses a machine reset to fence. This is the quickest route
1407 for the node to rejoin the cluster.
1408 PROCESSES
1411 [o2net]
1414 One per node. It is a work-queue thread started when the
1415 cluster is brought on-line and stopped when it is off-
1416 lined. It handles network communication for all mounts.
1417 It gets the list of active nodes from O2HB and sets up a
1418 TCP/IP communication channel with each live node. It
1419 sends regular keep-alive packets to detect any interrup‐
1420 tion on the channels.
1421 [user_dlm]
1424 One per node. It is a work-queue thread started when
1425 dlmfs is loaded and stopped when it is unloaded (dlmfs is
1426 a synthetic file system that allows user space processes
1427 to access the in-kernel dlm).
1428 [ocfs2_wq]
1431 One per node. It is a work-queue thread started when the
1432 OCFS2 module is loaded and stopped when it is unloaded.
1433 It is assigned background file system tasks that may take
1434 cluster locks like flushing the truncate log, orphan
1435 directory recovery and local alloc recovery. For example,
1436 orphan directory recovery runs in the background so that
1437 it does not affect recovery time.
1438 [o2hb-14C29A7392]
1441 One per heartbeat device. It is a kernel thread started
1442 when the heartbeat region is populated in configfs and
1443 stopped when it is removed. It writes every two seconds
1444 to a block in the heartbeat region, indicating that this
1445 node is alive. It also reads the region to maintain a map
1446 of live nodes. It notifies subscribers like o2net and
1447 o2dlm of any changes in the live node map.
1448 [ocfs2dc]
1451 One per mount. It is a kernel thread started when a vol‐
1452 ume is mounted and stopped when it is unmounted. It down‐
1453 grades locks in response to blocking ASTs (BASTs)
1454 requested by other nodes.
1455 [jbd2/sdf1-97]
1458 One per mount. It is part of JBD2, which OCFS2 uses for
1459 journaling.
1460 [ocfs2cmt]
1463 One per mount. It is a kernel thread started when a vol‐
1464 ume is mounted and stopped when it is unmounted. It works
1465 with kjournald2.
1466 [ocfs2rec]
1469 It is started whenever a node has to be recovered. This
1470 thread performs file system recovery by replaying the
1471 journal of the dead node. It is scheduled to run after
1472 dlm recovery has completed.
1473 [dlm_thread]
1476 One per dlm domain. It is a kernel thread started when a
1477 dlm domain is created and stopped when it is destroyed.
1478 This thread sends ASTs and blocking ASTs in response to
1479 lock level convert requests. It also frees unused lock
1480 resources.
1481 [dlm_reco_thread]
1484 One per dlm domain. It is a kernel thread that handles
1485 dlm recovery when another node dies. If this node is the
1486 dlm recovery master, it re-masters every lock resource
1487 owned by the dead node.
1488 [dlm_wq]
1491 One per dlm domain. It is a work-queue thread that o2dlm
1492 uses to queue blocking tasks.
1493 FUTURE WORK
1496 File system development is a never ending cycle. Faster and
1497 larger disks, faster and more number of processors, larger
1498 caches, etc. keep changing the sweet spot for performance forc‐
1499 ing developers to rethink long held beliefs. Add to that new use
1500 cases, which forces developers to be innovative in providing
1501 solutions that melds seamlessly with existing semantics.
1502 We are currently looking to add features like transparent com‐
1504 pression, transparent encryption, delayed allocation, multi-
1505 device support, etc. as well as work on improving performance on
1506 newer generation machines.
1507 If you are interested in contributing, email the development
1509 team at ocfs2-devel@oss.oracle.com.
1510
1513 The principal developers of the OCFS2 file system, its tools and the
1514 O2CB cluster stack, are Joel Becker, Zach Brown, Mark Fasheh, Jan Kara,
1515 Kurt Hackel, Tao Ma, Sunil Mushran, Tiger Yang and Tristan Ye.
1516 Other developers who have contributed to the file system via bug fixes,
1518 testing, etc. are Wim Coekaerts, Srinivas Eeda, Coly Li, Jeff Mahoney,
1519 Marcos Matsunaga, Goldwyn Rodrigues, Manish Singh and Wengang Wang.
1520 The members of the Linux Cluster community including Andrew Beekhof,
1522 Lars Marowsky-Bree, Fabio Massimo Di Nitto and David Teigland.
1523 The members of the Linux File system community including Christoph
1525 Hellwig and Chris Mason.
1526 The corporations that have contributed resources for this project
1528 including Oracle, SUSE Labs, EMC, Emulex, HP, IBM, Intel and Network
1529 Appliance.
1530
1533 debugfs.ocfs2(8) fsck.ocfs2(8) fsck.ocfs2.checks(8) mkfs.ocfs2(8)
1534 mount.ocfs2(8) mounted.ocfs2(8) o2cluster(8) o2image(8) o2info(1)
1535 o2cb(7) o2cb(8) o2cb.sysconfig(5) o2hbmonitor(8) ocfs2.cluster.conf(5)
1536 tunefs.ocfs2(8)
1537
1540 Oracle Corporation
1541
1544 Copyright © 2004, 2012 Oracle. All rights reserved.
1545Version 1.8.6 January 2012 OCFS2(7)