The Red Hat GFS2 file system is included in the Resilient Storage Add-On. It is a native file system that interfaces directly with the Linux kernel file system interface (VFS layer). When implemented as a cluster file system, GFS2 employs distributed metadata and multiple journals. Red Hat supports the use of GFS2 file systems only as implemented in the High Availability Add-On.
Although a GFS2 file system can be implemented in a standalone system or as part of a cluster configuration, for the Red Hat Enterprise Linux 6 release Red Hat does not support the use of GFS2 as a single-node file system. Red Hat does support a number of high-performance single node file systems which are optimized for single node and thus have generally lower overhead than a cluster file system. Red Hat recommends using these file systems in preference to GFS2 in cases where only a single node needs to mount the file system.
Red Hat will continue to support single-node GFS2 file systems for mounting snapshots of cluster file systems (for example, for backup purposes).
Red Hat does not support using GFS2 for cluster file system deployments greater than 16 nodes.
GFS2 is based on a 64-bit architecture, which can theoretically accommodate an 8 EB file system. However, the current supported maximum size of a GFS2 file system for 64-bit hardware is 100 TB. The current supported maximum size of a GFS2 file system for 32-bit hardware is 16 TB. If your system requires larger GFS2 file systems, contact your Red Hat service representative.
When determining the size of your file system, you should consider your recovery needs. Running the
fsck.gfs2 command on a very large file system can take a long time and consume a large amount of memory. Additionally, in the event of a disk or disk-subsystem failure, recovery time is limited by the speed of your backup media. For information on the amount of memory the
fsck.gfs2 command requires, see
Section 3.11, “Repairing a File System”.
When configured in a cluster, Red Hat GFS2 nodes can be configured and managed with High Availability Add-On configuration and management tools. Red Hat GFS2 then provides data sharing among GFS2 nodes in a cluster, with a single, consistent view of the file system name space across the GFS2 nodes. This allows processes on different nodes to share GFS2 files in the same way that processes on the same node can share files on a local file system, with no discernible difference. For information about the High Availability Add-On refer to Configuring and Managing a Red Hat Cluster.
While a GFS2 file system may be used outside of LVM, Red Hat supports only GFS2 file systems that are created on a CLVM logical volume. CLVM is included in the Resilient Storage Add-On. It is a cluster-wide implementation of LVM, enabled by the CLVM daemon clvmd, which manages LVM logical volumes in a cluster. The daemon makes it possible to use LVM2 to manage logical volumes across a cluster, allowing all nodes in the cluster to share the logical volumes. For information on the LVM volume manager, see Logical Volume Manager Administration
The gfs2.ko kernel module implements the GFS2 file system and is loaded on GFS2 cluster nodes.
When you configure a GFS2 file system as a cluster file system, you must ensure that all nodes in the cluster have access to the shared storage. Asymmetric cluster configurations in which some nodes have access to the shared storage and others do not are not supported. This does not require that all nodes actually mount the GFS2 file system itself.
This chapter provides some basic, abbreviated information as background to help you understand GFS2. It contains the following sections:
1.1. New and Changed Features
This section lists new and changed features of the GFS2 file system and the GFS2 documentation that are included with the initial and subsequent releases of Red Hat Enterprise Linux 6.
1.1.1. New and Changed Features for Red Hat Enterprise Linux 6.0
Red Hat Enterprise Linux 6.0 includes the following documentation and feature updates and changes.
1.1.2. New and Changed Features for Red Hat Enterprise Linux 6.1
Red Hat Enterprise Linux 6.1 includes the following documentation and feature updates and changes.
1.1.3. New and Changed Features for Red Hat Enterprise Linux 6.2
Red Hat Enterprise Linux 6.2 includes the following documentation and feature updates and changes.
As of the Red Hat Enterprise Linux 6.2 release, GFS2 supports the tunegfs2 command, which replaces some of the features of the gfs2_tool command. For further information, refer to the tunegfs2 man page.
The following sections have been updated to provide administrative procedures that do not require the use of the gfs2_tool command:
This document includes a new appendix,
Appendix C, GFS2 tracepoints and the debugfs glocks File. This appendix describes the glock
debugfs interface and the GFS2 tracepoints. It is intended for advanced users who are familiar with file system internals who would like to learn more about the design of GFS2 and how to debug GFS2-specific issues.
1.2. Before Setting Up GFS2
Before you install and set up GFS2, note the following key characteristics of your GFS2 file systems:
- GFS2 nodes
Determine which nodes in the cluster will mount the GFS2 file systems.
- Number of file systems
Determine how many GFS2 file systems to create initially. (More file systems can be added later.)
- File system name
Determine a unique name for each file system. The name must be unique for all lock_dlm file systems over the cluster. Each file system name is required in the form of a parameter variable. For example, this book uses file system names mydata1 and mydata2 in some example procedures.
- Journals
Determine the number of journals for your GFS2 file systems. One journal is required for each node that mounts a GFS2 file system. GFS2 allows you to add journals dynamically at a later point as additional servers mount a file system. For information on adding journals to a GFS2 file system, see
Section 3.7, “Adding Journals to a File System”.
- Storage devices and partitions
Determine the storage devices and partitions to be used for creating logical volumes (via CLVM) in the file systems.
You may see performance problems with GFS2 when many create and delete operations are issued from more than one node in the same directory at the same time. If this causes performance problems in your system, you should localize file creation and deletions by a node to directories specific to that node as much as possible.
1.3. Differences between GFS and GFS2
This section lists the improvements and changes that GFS2 offers over GFS.
1.3.1. GFS2 Command Names
In general, the functionality of GFS2 is identical to GFS. The names of the file system commands, however, specify GFS2 instead of GFS.
Table 1.1, “GFS and GFS2 Commands” shows the equivalent GFS and GFS2 commands and functionality.
Table 1.1. GFS and GFS2 Commands
|
GFS Command
|
GFS2 Command
|
Description
|
|---|
mount
|
mount
|
Mount a file system. The system can determine whether the file system is a GFS or GFS2 file system type. For information on the GFS2 mount options see the gfs2_mount(8) man page.
|
umount
|
umount
|
Unmount a file system.
|
|
|
|
Check and repair an unmounted file system.
|
gfs_grow
|
gfs2_grow
|
Grow a mounted file system.
|
gfs_jadd
|
gfs2_jadd
|
Add a journal to a mounted file system.
|
|
|
|
Create a file system on a storage device.
|
gfs_quota
|
gfs2_quota
|
Manage quotas on a mounted file system. As of the Red Hat Enterprise Linux 6.1 release, GFS2 supports the standard Linux quota facilities. For further information on quota management in GFS2, refer to Section 3.5, “GFS2 Quota Management”.
|
gfs_tool
|
tunegfs2
mount parameters
dmsetup suspend
|
Configure, tune, or gather information about a file system. The tunegfs2 command is supported as of the Red Hat Enterprise Linux 6.2 release. There is also a gfs2_tool command.
|
gfs_edit
|
gfs2_edit
|
Display, print, or edit file system internal structures. The gfs2_edit command can be used for GFS file systems as well as GFS2 file system.
|
gfs_tool setflag jdata/inherit_jdata
|
chattr +j (preferred)
|
Enable journaling on a file or directory.
|
setfacl/getfacl
|
setfacl/getfacl
|
Set or get file access control list for a file or directory.
|
setfattr/getfattr
|
setfattr/getfattr
|
Set or get the extended attributes of a file.
|
For a full listing of the supported options for the GFS2 file system commands, see the man pages for those commands.
1.3.2. Additional Differences Between GFS and GFS2
Context-Dependent Path Names
GFS2 file systems do not provide support for context-dependent path names, which allow you to create symbolic links that point to variable destination files or directories. For this functionality in GFS2, you can use the
bind option of the
mount command. For information on bind mounts and context-dependent pathnames in GFS2, see
Section 3.12, “Bind Mounts and Context-Dependent Path Names”.
The kernel module that implements the GFS file system is gfs.ko. The kernel module that implements the GFS2 file system is gfs2.ko.
Enabling Quota Enforcement in GFS2
In GFS2 file systems, quota enforcement is disabled by default and must be explicitly enabled. For information on enabling and disabling quota enforcement, see
Section 3.5, “GFS2 Quota Management”.
GFS2 file systems support the use of the chattr command to set and clear the j flag on a file or directory. Setting the +j flag on a file enables data journaling on that file. Setting the +j flag on a directory means "inherit jdata", which indicates that all files and directories subsequently created in that directory are journaled. Using the chattr command is the preferred way to enable and disable data journaling on a file.
Adding Journals Dynamically
In GFS file systems, journals are embedded metadata that exists outside of the file system, making it necessary to extend the size of the logical volume that contains the file system before adding journals. In GFS2 file systems, journals are plain (though hidden) files. This means that for GFS2 file systems, journals can be dynamically added as additional servers mount a file system, as long as space remains on the file system for the additional journals. For information on adding journals to a GFS2 file system, see
Section 3.7, “Adding Journals to a File System”.
atime_quantum parameter removed
The GFS2 file system does not support the atime_quantum tunable parameter, which can be used by the GFS file system to specify how often atime updates occur. In its place GFS2 supports the relatime and noatime mount options. The relatime mount option is recommended to achieve similar behavior to setting the atime_quantum parameter in GFS.
The data= option of the mount command
When mounting GFS2 file systems, you can specify the data=ordered or data=writeback option of the mount. When data=ordered is set, the user data modified by a transaction is flushed to the disk before the transaction is committed to disk. This should prevent the user from seeing uninitialized blocks in a file after a crash. When data=writeback is set, the user data is written to the disk at any time after it is dirtied. This does not provide the same consistency guarantee as ordered mode, but it should be slightly faster for some workloads. The default is ordered mode.
The gfs2_edit command supports a different set of options for GFS2 than the gfs_edit command supports for GFS. For information on the specific options each version of the command supports, see the gfs2_edit and gfs_edit man pages.
1.3.3. GFS2 Performance Improvements
There are many features of GFS2 file systems that do not result in a difference in the user interface from GFS file systems but which improve file system performance.
A GFS2 file system provides improved file system performance in the following ways:
Better performance for heavy usage in a single directory
Faster synchronous I/O operations
Faster cached reads (no locking overhead)
Faster direct I/O with preallocated files (provided I/O size is reasonably large, such as 4M blocks)
Faster I/O operations in general
Faster execution of the df command, because of faster statfs calls
Improved atime mode to reduce the number of write I/O operations generated by atime when compared with GFS
GFS2 file systems provide broader and more mainstream support in the following ways:
A GFS2 file system provides the following improvements to the internal efficiency of the file system.
GFS2 uses less kernel memory.
GFS2 requires no metadata generation numbers.
Allocating GFS2 metadata does not require reads. Copies of metadata blocks in multiple journals are managed by revoking blocks from the journal before lock release.
GFS2 includes a much simpler log manager that knows nothing about unlinked inodes or quota changes.
The gfs2_grow and gfs2_jadd commands use locking to prevent multiple instances running at the same time.
The ACL code has been simplified for calls like creat() and mkdir().
Unlinked inodes, quota changes, and statfs changes are recovered without remounting the journal.
In order to get the best performance from a GFS2 file system, it is very important to understand some of the basic theory of its operation. A single node file system is implemented alongside a cache, the purpose of which is to eliminate latency of disk accesses when using frequently requested data. In Linux the page cache (and historically the buffer cache) provide this caching function.
With GFS2, each node has its own page cache which may contain some portion of the on-disk data. GFS2 uses a locking mechanism called glocks (pronounced gee-locks) to maintain the integrity of the cache between nodes. The glock subsystem provides a cache management function which is implemented using the distributed lock manager (DLM) as the underlying communication layer.
The glocks provide protection for the cache on a per-inode basis, so there is one lock per inode which is used for controlling the caching layer. If that glock is granted in shared mode (DLM lock mode: PR) then the data under that glock may be cached upon one or more nodes at the same time, so that all the nodes may have local access to the data.
If the glock is granted in exclusive mode (DLM lock mode: EX) then only a single node may cache the data under that glock. This mode is used by all operations which modify the data (such as the write system call).
If another node requests a glock which cannot be granted immediately, then the DLM sends a message to the node or nodes which currently hold the glocks blocking the new request to ask them to drop their locks. Dropping glocks can be (by the standards of most file system operations) a long process. Dropping a shared glock requires only that the cache be invalidated, which is relatively quick and proportional to the amount of cached data.
Dropping an exclusive glock requires a log flush, and writing back any changed data to disk, followed by the invalidation as per the shared glock.
The different between a single node file system and GFS2 then, is that a single node file system has a single cache and GFS2 has a separate cache on each node. In both cases, latency to access to cached data is of a similar order of magnitude, but the latency to access uncached data is much greater in GFS2 if another node has previously cached that same data.
Due to the way in which GFS2's caching is implemented the best performance is obtained when either of the following takes place:
Note that inserting and removing entries from a directory during file creation and deletion counts as writing to the directory inode.
It is possible to break this rule provided that it is broken relatively infrequently. Ignoring this rule too often will result in a severe performance penalty.
If you mmap() a file on GFS2 with a read/write mapping, but only read from it, this only counts as a read. On GFS though, it counts as a write, so GFS2 is much more scalable with mmap() I/O.
If you do not set the noatime mount parameter, then reads will also result in writes to update the file timestamps. We recommend that all GFS2 users should mount with noatime unless they have a specific requirement for atime.
It is usually possible to alter the way in which a troublesome application stores its data in order to gain a considerable performance advantage.
A typical example of a troublesome application is an email server. These are often laid out with a spool directory containing files for each user (mbox), or with a directory for each user containing a file for each message (maildir). When requests arrive over IMAP, the ideal arrangement is to give each user an affinity to a particular node. That way their requests to view and delete email messages will tend to be served from the cache on that one node. Obviously if that node fails, then the session can be restarted on a different node.
When mail arrives via SMTP, then again the individual nodes can be set up so as to pass a certain user's mail to a particular node by default. If the default node is not up, then the message can be saved directly into the user's mail spool by the receiving node. Again this design is intended to keep particular sets of files cached on just one node in the normal case, but to allow direct access in the case of node failure.
This setup allows the best use of GFS2's page cache and also makes failures transparent to the application, whether imap or smtp.
Backup is often another tricky area. Again, if it is possible it is greatly preferable to back up the working set of each node directly from the node which is caching that particular set of inodes. If you have a backup script which runs at a regular point in time, and that seems to coincide with a spike in the response time of an application running on GFS2, then there is a good chance that the cluster may not be making the most efficient use of the page cache.
Obviously, if you are in the (enviable) position of being able to stop the application in order to perform a backup, then this won't be a problem. On the other hand, if a backup is run from just one node, then after it has completed a large portion of the file system will be cached on that node, with a performance penalty for subsequent accesses from other nodes. This can be mitigated to a certain extent by dropping the VFS page cache on the backup node after the backup has completed with following command:
echo -n 3 >/proc/sys/vm/drop_caches
However this is not as good a solution as taking care to ensure the working set on each node is either shared, mostly read only across the cluster, or accessed largely from a single node.
If your cluster performance is suffering because of inefficient use of GFS2 caching, you may see large and increasing I/O wait times. You can make use of GFS2's lock dump information to determine the cause of the problem.
The GFS2 lock dump information can be gathered from the debugfs file which can be found at the following path name, assuming that debugfs is mounted on /sys/kernel/debug/:
/sys/kernel/debug/gfs2/fsname/glocks
The content of the file is a series of lines. Each line starting with G: represents one glock, and the following lines, indented by a single space, represent an item of information relating to the glock immediately before them in the file.
The best way to use the debugfs file is to use the cat command to take a copy of the complete content of the file (it might take a long time if you have a large amount of RAM and a lot of cached inodes) while the application is experiencing problems, and then looking through the resulting data at a later date.
It can be useful to make two copies of the debugfs file, one a few seconds or even a minute or two after the other. By comparing the holder information in the two traces relating to the same glock number, you can tell whether the workload is making progress (that is, it is just slow) or whether it has become stuck (which is always a bug and should be reported to Red Hat support immediately).
Lines in the debugfs file starting with H: (holders) represent lock requests either granted or waiting to be granted. The flags field on the holders line f: shows which: The 'W' flag refers to a waiting request, the 'H' flag refers to a granted request. The glocks which have large numbers of waiting requests are likely to be those which are experiencing particular contention.
Table 1.2. Glock flags
|
Flag
|
Name
|
Meaning
|
|---|
|
d
|
Pending demote
|
A deferred (remote) demote request
|
|
D
|
Demote
|
A demote request (local or remote)
|
|
f
|
Log flush
|
The log needs to be committed before releasing this glock
|
|
F
|
Frozen
|
Replies from remote nodes ignored - recovery is in progress
|
|
i
|
Invalidate in progress
|
In the process of invalidating pages under this glock
|
|
I
|
Initial
|
Set when DLM lock is associated with this glock
|
|
l
|
Locked
|
The glock is in the process of changing state
|
|
p
|
Demote in progress
|
The glock is in the process of responding to a demote request
|
|
r
|
Reply pending
|
Reply received from remote node is awaiting processing
|
|
y
|
Dirty
|
Data needs flushing to disk before releasing this glock
|
Table 1.3. Glock holder flags
|
Flag
|
Name
|
Meaning
|
|---|
|
a
|
Async
|
Do not wait for glock result (will poll for result later)
|
|
A
|
Any
|
Any compatible lock mode is acceptable
|
|
c
|
No cache
|
When unlocked, demote DLM lock immediately
|
|
e
|
No expire
|
Ignore subsequent lock cancel requests
|
|
E
|
exact
|
Must have exact lock mode
|
|
F
|
First
|
Set when holder is the first to be granted for this lock
|
|
H
|
Holder
|
Indicates that requested lock is granted
|
|
p
|
Priority
|
Enqueue holder at the head of the queue
|
|
t
|
Try
|
A "try" lock
|
|
T
|
Try 1CB
|
A "try" lock that sends a callback
|
|
W
|
Wait
|
Set while waiting for request to complete
|
Having identified a glock which is causing a problem, the next step is to find out which inode it relates to. The glock number (n: on the G: line) indicates this. It is of the form type/number and if type is 2, then the glock is an inode glock and the number is an inode number. To track down the inode, you can then run find -inum number where number is the inode number converted from the hex format in the glocks file into decimal.
If you run the find on a file system when it is experiencing lock contention, you are likely to make the problem worse. It is a good idea to stop the application before running the find when you are looking for contended inodes.
Table 1.4. Glock types
|
Type number
|
Lock type
|
Use
|
|---|
|
1
|
Trans
|
Transaction lock
|
|
2
|
Inode
|
Inode metadata and data
|
|
3
|
Rgrp
|
Resource group metadata
|
|
4
|
Meta
|
The superblock
|
|
5
|
Iopen
|
Inode last closer detection
|
|
6
|
Flock
|
flock(2) syscall
|
|
8
|
Quota
|
Quota operations
|
|
9
|
Journal
|
Journal mutex
|
If the glock that was identified was of a different type, then it is most likely to be of type 3: (resource group). If you see significant numbers of processes waiting for other types of glock under normal loads, then please report this to Red Hat support.
If you do see a number of waiting requests queued on a resource group lock there may be a number of reason for this. One is that there are a large number of nodes compared to the number of resource groups in the file system. Another is that the file system may be very nearly full (requiring, on average, longer searches for free blocks). The situation in both cases can be improved by adding more storage and using the gfs2_grow command to expand the file system.
This chapter describes the tasks and commands for managing GFS2 and consists of the following sections:
3.1. Making a File System
You create a GFS2 file system with the mkfs.gfs2 command. You can also use the mkfs command with the -t gfs2 option specified. A file system is created on an activated LVM volume. The following information is required to run the mkfs.gfs2 command:
Lock protocol/module name (the lock protocol for a cluster is lock_dlm)
Cluster name (when running as part of a cluster configuration)
Number of journals (one journal required for each node that may be mounting the file system)
When creating a GFS2 file system, you can use the mkfs.gfs2 command directly, or you can use the mkfs command with the -t parameter specifying a file system of type gfs2, followed by the gfs2 file system options.
Once you have created a GFS2 file system with the
mkfs.gfs2 command, you cannot decrease the size of the file system. You can, however, increase the size of an existing file system with the
gfs2_grow command, as described in
Section 3.6, “Growing a File System”.
When creating a clustered GFS2 file system, you can use either of the following formats:
mkfs.gfs2 -p LockProtoName -t LockTableName -j NumberJournals BlockDevice
mkfs -t gfs2 -p LockProtoName -t LockTableName -j NumberJournals BlockDevice
When creating a local GFS2 file system, you can use either of the following formats:
For the Red Hat Enterprise Linux 6 release, Red Hat does not support the use of GFS2 as a single-node file system.
mkfs.gfs2 -p LockProtoName -j NumberJournals BlockDevice
mkfs -t gfs2 -p LockProtoName -j NumberJournals BlockDevice
Make sure that you are very familiar with using the LockProtoName and LockTableName parameters. Improper use of the LockProtoName and LockTableName parameters may cause file system or lock space corruption.
LockProtoName
Specifies the name of the locking protocol to use. The lock protocol for a cluster is lock_dlm.
LockTableName
This parameter is specified for GFS2 file system in a cluster configuration. It has two parts separated by a colon (no spaces) as follows: ClusterName:FSName
ClusterName, the name of the cluster for which the GFS2 file system is being created.
FSName, the file system name, can be 1 to 16 characters long. The name must be unique for all lock_dlm file systems over the cluster, and for all file systems (lock_dlm and lock_nolock) on each local node.
Number
Specifies the number of journals to be created by the
mkfs.gfs2 command. One journal is required for each node that mounts the file system. For GFS2 file systems, more journals can be added later without growing the file system, as described in
Section 3.7, “Adding Journals to a File System”.
BlockDevice
Specifies a logical or physical volume.
In these example, lock_dlm is the locking protocol that the file system uses, since this is a clustered file system. The cluster name is alpha, and the file system name is mydata1. The file system contains eight journals and is created on /dev/vg01/lvol0.
mkfs.gfs2 -p lock_dlm -t alpha:mydata1 -j 8 /dev/vg01/lvol0
mkfs -t gfs2 -p lock_dlm -t alpha:mydata1 -j 8 /dev/vg01/lvol0
In these examples, a second lock_dlm file system is made, which can be used in cluster alpha. The file system name is mydata2. The file system contains eight journals and is created on /dev/vg01/lvol1.
mkfs.gfs2 -p lock_dlm -t alpha:mydata2 -j 8 /dev/vg01/lvol1
mkfs -t gfs2 -p lock_dlm -t alpha:mydata2 -j 8 /dev/vg01/lvol1
Table 3.1. Command Options: mkfs.gfs2
|
Flag
|
Parameter
|
Description
|
|---|
-c
|
Megabytes
|
Sets the initial size of each journal's quota change file to Megabytes.
|
-D
|
|
Enables debugging output.
|
-h
|
|
Help. Displays available options.
|
-J
|
MegaBytes
|
Specifies the size of the journal in megabytes. Default journal size is 128 megabytes. The minimum size is 8 megabytes. Larger journals improve performance, although they use more memory than smaller journals.
|
-j
|
Number
|
Specifies the number of journals to be created by the mkfs.gfs2 command. One journal is required for each node that mounts the file system. If this option is not specified, one journal will be created. For GFS2 file systems, you can add additional journals at a later time without growing the file system.
|
-O
|
|
Prevents the mkfs.gfs2 command from asking for confirmation before writing the file system.
|
-p
|
LockProtoName
|
| Specifies the name of the locking protocol to use. Recognized locking protocols include: | lock_dlm — The standard locking module, required for a clustered file system. | lock_nolock — Used when GFS2 is acting as a local file system (one node only). |
|
-q
|
|
Quiet. Do not display anything.
|
-r
|
MegaBytes
|
Specifies the size of the resource groups in megabytes. The minimum resource group size is 32 MB. The maximum resource group size is 2048 MB. A large resource group size may increase performance on very large file systems. If this is not specified, mkfs.gfs2 chooses the resource group size based on the size of the file system: average size file systems will have 256 MB resource groups, and bigger file systems will have bigger RGs for better performance.
|
-t
|
LockTableName
|
A unique identifier that specifies the lock table field when you use the lock_dlm protocol; the lock_nolock protocol does not use this parameter. | | This parameter has two parts separated by a colon (no spaces) as follows: ClusterName:FSName. | | ClusterName is the name of the cluster for which the GFS2 file system is being created; only members of this cluster are permitted to use this file system. The cluster name is set in the /etc/cluster/cluster.conf file via the Cluster Configuration Tool and displayed at the Cluster Status Tool in the Red Hat Cluster Suite cluster management GUI. | | FSName, the file system name, can be 1 to 16 characters in length, and the name must be unique among all file systems in the cluster. |
|
-u
|
MegaBytes
|
Specifies the initial size of each journal's unlinked tag file.
|
-V
|
|
Displays command version information.
|
3.2. Mounting a File System
Before you can mount a GFS2 file system, the file system must exist (refer to
Section 3.1, “Making a File System”), the volume where the file system exists must be activated, and the supporting clustering and locking systems must be started (refer to
Configuring and Managing a Red Hat Cluster). After those requirements have been met, you can mount the GFS2 file system as you would any Linux file system.
Attempting to mount a GFS2 file system when the Cluster Manager (cman) has not been started produces the following error message:
[root@gfs-a24c-01 ~]# mount -t gfs2 -o noatime /dev/mapper/mpathap1 /mnt
gfs_controld join connect error: Connection refused
error mounting lockproto lock_dlm
To manipulate file ACLs, you must mount the file system with the -o acl mount option. If a file system is mounted without the -o acl mount option, users are allowed to view ACLs (with getfacl), but are not allowed to set them (with setfacl).
Mounting Without ACL Manipulation
mount BlockDevice MountPoint
Mounting With ACL Manipulation
mount -o acl BlockDevice MountPoint
-o acl
GFS2-specific option to allow manipulating file ACLs.
BlockDevice
Specifies the block device where the GFS2 file system resides.
MountPoint
Specifies the directory where the GFS2 file system should be mounted.
In this example, the GFS2 file system on /dev/vg01/lvol0 is mounted on the /mygfs2 directory.
mount /dev/vg01/lvol0 /mygfs2
mount BlockDevice MountPoint -o option
The
-o option argument consists of GFS2-specific options (refer to
Table 3.2, “GFS2-Specific Mount Options”) or acceptable standard Linux
mount -o options, or a combination of both. Multiple
option parameters are separated by a comma and no spaces.
The mount command is a Linux system command. In addition to using GFS2-specific options described in this section, you can use other, standard, mount command options (for example, -r). For information about other Linux mount command options, see the Linux mount man page.
This table includes descriptions of options that are used with local file systems only. Note, however, that for the Red Hat Enterprise Linux 6 release, Red Hat does not support the use of GFS2 as a single-node file system. Red Hat will continue to support single-node GFS2 file systems for mounting snapshots of cluster file systems (for example, for backup purposes).
Table 3.2. GFS2-Specific Mount Options
|
Option
|
Description
|
|---|
acl
|
Allows manipulating file ACLs. If a file system is mounted without the acl mount option, users are allowed to view ACLs (with getfacl), but are not allowed to set them (with setfacl).
|
data=[ordered|writeback]
|
When data=ordered is set, the user data modified by a transaction is flushed to the disk before the transaction is committed to disk. This should prevent the user from seeing uninitialized blocks in a file after a crash. When data=writeback mode is set, the user data is written to the disk at any time after it is dirtied; this does not provide the same consistency guarantee as ordered mode, but it should be slightly faster for some workloads. The default value is ordered mode.
|
ignore_local_fs | | Caution: This option should not be used when GFS2 file systems are shared. |
|
Forces GFS2 to treat the file system as a multihost file system. By default, using lock_nolock automatically turns on the localflocks flag.
|
localflocks | | Caution: This option should not be used when GFS2 file systems are shared. |
|
Tells GFS2 to let the VFS (virtual file system) layer do all flock and fcntl. The localflocks flag is automatically turned on by lock_nolock.
|
lockproto=LockModuleName
|
Allows the user to specify which locking protocol to use with the file system. If LockModuleName is not specified, the locking protocol name is read from the file system superblock.
|
locktable=LockTableName
|
Allows the user to specify which locking table to use with the file system.
|
quota=[off/account/on]
|
Turns quotas on or off for a file system. Setting the quotas to be in the account state causes the per UID/GID usage statistics to be correctly maintained by the file system; limit and warn values are ignored. The default value is off.
|
errors=panic|withdraw
|
When errors=panic is specified, file system errors will cause a kernel panic. The default behavior, which is the same as specifying errors=withdraw, is for the system to withdraw from the file system and make it inaccessible until the next reboot; in some cases the system may remain running. For information on the GFS2 withdraw function, see Section 3.14, “The GFS2 Withdraw Function”.
|
discard/nodiscard
|
Causes GFS2 to generate "discard" I/O requests for blocks that have been freed. These can be used by suitable hardware to implement thin provisioning and similar schemes.
|
barrier/nobarrier
|
Causes GFS2 to send I/O barriers when flushing the journal. The default value is on. This option is automatically turned off if the underlying device does not support I/O barriers. Use of I/O barriers with GFS2 is highly recommended at all times unless the block device is designed so that it cannot lose its write cache content (for example, if it is on a UPS or it does not have a write cache).
|
quota_quantum=secs
|
Sets the number of seconds for which a change in the quota information may sit on one node before being written to the quota file. This is the preferred way to set this parameter. The value is an integer number of seconds greater than zero. The default is 60 seconds. Shorter settings result in faster updates of the lazy quota information and less likelihood of someone exceeding their quota. Longer settings make file system operations involving quotas faster and more efficient.
|
statfs_quantum=secs
|
Setting statfs_quantum to 0 is the preferred way to set the slow version of statfs. The default value is 30 secs which sets the maximum time period before statfs changes will be synced to the master statfs file. This can be adjusted to allow for faster, less accurate statfs values or slower more accurate values. When this option is set to 0, statfs will always report the true values.
|
statfs_percent=value
|
Provides a bound on the maximum percentage change in the statfs information on a local basis before it is synced back to the master statfs file, even if the time period has not expired. If the setting of statfs_quantum is 0, then this setting is ignored.
|
3.3. Unmounting a File System
The GFS2 file system can be unmounted the same way as any Linux file system — by using the umount command.
The umount command is a Linux system command. Information about this command can be found in the Linux umount command man pages.
umount MountPoint
MountPoint
Specifies the directory where the GFS2 file system is currently mounted.
3.4. Special Considerations when Mounting GFS2 File Systems
GFS2 file systems that have been mounted manually rather than automatically through an entry in the fstab file will not be known to the system when file systems are unmounted at system shutdown. As a result, the GFS2 script will not unmount the GFS2 file system. After the GFS2 shutdown script is run, the standard shutdown process kills off all remaining user processes, including the cluster infrastructure, and tries to unmount the file system. This unmount will fail without the cluster infrastructure and the system will hang.
To prevent the system from hanging when the GFS2 file systems are unmounted, you should do one of the following:
Always use an entry in the fstab file to mount the GFS2 file system.
If a GFS2 file system has been mounted manually with the mount command, be sure to unmount the file system manually with the umount command before rebooting or shutting down the system.
If your file system hangs while it is being unmounted during system shutdown under these circumstances, perform a hardware reboot. It is unlikely that any data will be lost since the file system is synced earlier in the shutdown process.
3.5. GFS2 Quota Management
File-system quotas are used to limit the amount of file system space a user or group can use. A user or group does not have a quota limit until one is set. When a GFS2 file system is mounted with the quota=on or quota=account option, GFS2 keeps track of the space used by each user and group even when there are no limits in place. GFS2 updates quota information in a transactional way so system crashes do not require quota usages to be reconstructed.
To prevent a performance slowdown, a GFS2 node synchronizes updates to the quota file only periodically. The fuzzy quota accounting can allow users or groups to slightly exceed the set limit. To minimize this, GFS2 dynamically reduces the synchronization period as a hard quota limit is approached.
As of the Red Hat Enterprise Linux 6.1 release, GFS2 supports the standard Linux quota facilities. In order to use this you will need to install the quota RPM. This is the preferred way to administer quotas on GFS2 and should be used for all new deployments of GFS2 using quotas. This section documents GFS2 quota management using these facilities.
3.5.1. Configuring Disk Quotas
To implement disk quotas, use the following steps:
Set up quotas in enforcement or accounting mode.
Initialize the quota database file with current block usage information.
Assign quota policies. (In accounting mode, these policies are not enforced.)
Each of these steps is discussed in detail in the following sections.
3.5.1.1. Setting Up Quotas in Enforcement or Accounting Mode
In GFS2 file systems, quotas are disabled by default. To enable quotas for a file system, mount the file system with the quota=on option specified.
It is possible to keep track of disk usage and maintain quota accounting for every user and group without enforcing the limit and warn values. To do this, mount the file system with the quota=account option specified.
To mount a file system with quotas enabled, mount the file system with the quota=on option specified.
mount -o quota=on BlockDevice MountPoint
To mount a file system with quota accounting maintained, even though the quota limits are not enforced, mount the file system with the quota=account option specified.
mount -o quota=account BlockDevice MountPoint
To mount a file system with quotas disabled, mount the file system with the quota=off option specified. This is the default setting.
mount -o quota=off BlockDevice MountPoint
quota={on|off|account}
on - Specifies that quotas are enabled when the file system is mounted.
off - Specifies that quotas are disabled when the file system is mounted.
account - Specifies that user and group usage statistics are maintained by the file system, even though the quota limits are not enforced.
BlockDevice
Specifies the block device where the GFS2 file system resides.
MountPoint
Specifies the directory where the GFS2 file system should be mounted.
In this example, the GFS2 file system on /dev/vg01/lvol0 is mounted on the /mygfs2 directory with quotas enabled.
mount -o quota=on /dev/vg01/lvol0 /mygfs2
In this example, the GFS2 file system on /dev/vg01/lvol0 is mounted on the /mygfs2 directory with quota accounting maintained, but not enforced.
mount -o quota=account /dev/vg01/lvol0 /mygfs2
3.5.1.2. Creating the Quota Database Files
After each quota-enabled file system is mounted, the system is capable of working with disk quotas. However, the file system itself is not yet ready to support quotas. The next step is to run the quotacheck command.
The quotacheck command examines quota-enabled file systems and builds a table of the current disk usage per file system. The table is then used to update the operating system's copy of disk usage. In addition, the file system's disk quota files are updated.
To create the quota files on the file system, use the -u and the -g options of the quotacheck command; both of these options must be specified for user and group quotas to be initialized. For example, if quotas are enabled for the /home file system, create the files in the /home directory:
quotacheck -ug /home
3.5.1.3. Assigning Quotas per User
The last step is assigning the disk quotas with the edquota command. Note that if you have mounted your file system in accounting mode (with the quota=account option specified), the quotas are not enforced.
To configure the quota for a user, as root in a shell prompt, execute the command:
edquota username
Perform this step for each user who needs a quota. For example, if a quota is enabled in /etc/fstab for the /home partition (/dev/VolGroup00/LogVol02 in the example below) and the command edquota testuser is executed, the following is shown in the editor configured as the default for the system:
Disk quotas for user testuser (uid 501):
Filesystem blocks soft hard inodes soft hard
/dev/VolGroup00/LogVol02 440436 0 0
The text editor defined by the EDITOR environment variable is used by edquota. To change the editor, set the EDITOR environment variable in your ~/.bash_profile file to the full path of the editor of your choice.
The first column is the name of the file system that has a quota enabled for it. The second column shows how many blocks the user is currently using. The next two columns are used to set soft and hard block limits for the user on the file system.
The soft block limit defines the maximum amount of disk space that can be used.
The hard block limit is the absolute maximum amount of disk space that a user or group can use. Once this limit is reached, no further disk space can be used.
The GFS2 file system does not maintain quotas for inodes, so these columns do not apply to GFS2 file systems and will be blank.
If any of the values are set to 0, that limit is not set. In the text editor, change the desired limits. For example:
Disk quotas for user testuser (uid 501):
Filesystem blocks soft hard inodes soft hard
/dev/VolGroup00/LogVol02 440436 500000 550000
To verify that the quota for the user has been set, use the command:
quota testuser
3.5.1.4. Assigning Quotas per Group
Quotas can also be assigned on a per-group basis. Note that if you have mounted your file system in accounting mode (with the account=on option specified), the quotas are not enforced.
To set a group quota for the devel group (the group must exist prior to setting the group quota), use the following command:
edquota -g devel
This command displays the existing quota for the group in the text editor:
Disk quotas for group devel (gid 505):
Filesystem blocks soft hard inodes soft hard
/dev/VolGroup00/LogVol02 440400 0 0
The GFS2 file system does not maintain quotas for inodes, so these columns do not apply to GFS2 file systems and will be blank. Modify the limits, then save the file.
To verify that the group quota has been set, use the following command:
quota -g devel
3.5.2. Managing Disk Quotas
If quotas are implemented, they need some maintenance — mostly in the form of watching to see if the quotas are exceeded and making sure the quotas are accurate.
Of course, if users repeatedly exceed their quotas or consistently reach their soft limits, a system administrator has a few choices to make depending on what type of users they are and how much disk space impacts their work. The administrator can either help the user determine how to use less disk space or increase the user's disk quota.
You can create a disk usage report by running the repquota utility. For example, the command repquota /home produces this output:
*** Report for user quotas on device /dev/mapper/VolGroup00-LogVol02
Block grace time: 7days; Inode grace time: 7days
Block limits File limits
User used soft hard grace used soft hard grace
----------------------------------------------------------------------
root -- 36 0 0 4 0 0
kristin -- 540 0 0 125 0 0
testuser -- 440400 500000 550000 37418 0 0
To view the disk usage report for all (option -a) quota-enabled file systems, use the command:
repquota -a
While the report is easy to read, a few points should be explained. The -- displayed after each user is a quick way to determine whether the block limits have been exceeded. If the block soft limit is exceeded, a + appears in place of the the first - in the output. The second - indicates the inode limit, but GFS2 file systems do not support inode limits so that character will remain as -. GFS2 file systems do not support a grace period, so the grace column will remain blank.
Note that the repquota command is not supported over NFS, irrespective of the underlying file system.
3.5.3. Keeping Quotas Accurate
If you enable quotas on your file system after a period of time when you have been running with quotas disabled, you should run the quotacheck command to create, check, and repair quota files. Additionally, you may want to run the quotacheck if you think your quota files may not be accurate, as may occur when a file system is not unmounted cleanly after a system crash.
For more information about the quotacheck command, see the quotacheck man page.
Run quotacheck when the file system is relatively idle on all nodes because disk activity may affect the computed quota values.
3.5.4. Synchronizing Quotas with the quotasync Command
GFS2 stores all quota information in its own internal file on disk. A GFS2 node does not update this quota file for every file system write; rather, by default it updates the quota file once every 60 seconds. This is necessary to avoid contention among nodes writing to the quota file, which would cause a slowdown in performance.
As a user or group approaches their quota limit, GFS2 dynamically reduces the time between its quota-file updates to prevent the limit from being exceeded. The normal time period between quota synchronizations is a tunable parameter,
quota_quantum. You can change this from its default value of 60 seconds using the
quota_quantum= mount option, as described in
Table 3.2, “GFS2-Specific Mount Options”. The
quota_quantum parameter must be set on each node and each time the file system is mounted. Changes to the
quota_quantum parameter are not persistent across unmounts. You can update the
quota_quantum value with the
mount -o remount.
You can use the quotasync command to synchronize the quota information from a node to the on-disk quota file between the automatic updates performed by GFS2.
Synchronizing Quota Information
quotasync [-ug] -a|mntpnt...
u
Sync the user quota files.
g
Sync the group quota files
a
Sync all file systems that are currently quota-enabled and support sync. When -a is absent, a file system mountpoint should be specified.
mntpnt
Specifies the GFS2 file system to which the actions apply.
Tuning the Time Between Synchronizations
mount -o quota_quantum=secs,remount BlockDevice MountPoint
MountPoint
Specifies the GFS2 file system to which the actions apply.
secs
Specifies the new time period between regular quota-file synchronizations by GFS2. Smaller values may increase contention and slow down performance.
This example synchronizes all the cached dirty quotas from the node it is run on to the ondisk quota file for the file system /mnt/mygfs2.
# quotasync -ug /mnt/mygfs2
This example changes the default time period between regular quota-file updates to one hour (3600 seconds) for file system /mnt/mygfs2 when remounting that file system on logical volume /dev/volgroup/logical_volume.
# mount -o quota_quantum=3600,remount /dev/volgroup/logical_volume /mnt/mygfs2
For more information on disk quotas, refer to the man pages of the following commands:
quotacheck
edquota
repquota
quota
3.6. Growing a File System
The gfs2_grow command is used to expand a GFS2 file system after the device where the file system resides has been expanded. Running a gfs2_grow command on an existing GFS2 file system fills all spare space between the current end of the file system and the end of the device with a newly initialized GFS2 file system extension. When the fill operation is completed, the resource index for the file system is updated. All nodes in the cluster can then use the extra storage space that has been added.
The gfs2_grow command must be run on a mounted file system, but only needs to be run on one node in a cluster. All the other nodes sense that the expansion has occurred and automatically start using the new space.
Once you have created a GFS2 file system with the mkfs.gfs2 command, you cannot decrease the size of the file system.
gfs2_grow MountPoint
MountPoint
Specifies the GFS2 file system to which the actions apply.
Before running the gfs2_grow command:
Back up important data on the file system.
Determine the volume that is used by the file system to be expanded by running a df MountPoint command.
Expand the underlying cluster volume with LVM. For information on administering LVM volumes, see Logical Volume Manager Administration.
After running the gfs2_grow command, run a df command to check that the new space is now available in the file system.
In this example, the file system on the /mygfs2fs directory is expanded.
[root@dash-01 ~]# gfs2_grow /mygfs2fs
FS: Mount Point: /mygfs2fs
FS: Device: /dev/mapper/gfs2testvg-gfs2testlv
FS: Size: 524288 (0x80000)
FS: RG size: 65533 (0xfffd)
DEV: Size: 655360 (0xa0000)
The file system grew by 512MB.
gfs2_grow complete.
gfs2_grow [Options] {MountPoint | Device} [MountPoint | Device]
MountPoint
Specifies the directory where the GFS2 file system is mounted.
Device
Specifies the device node of the file system.
Table 3.3. GFS2-specific Options Available While Expanding A File System
|
Option
|
Description
|
|---|
-h
|
Help. Displays a short usage message.
|
-q
|
Quiet. Turns down the verbosity level.
|
-r MegaBytes
|
Specifies the size of the new resource group. The default size is 256MB.
|
-T
|
Test. Do all calculations, but do not write any data to the disk and do not expand the file system.
|
-V
|
Displays command version information.
|
3.7. Adding Journals to a File System
The gfs2_jadd command is used to add journals to a GFS2 file system. You can add journals to a GFS2 file system dynamically at any point without expanding the underlying logical volume. The gfs2_jadd command must be run on a mounted file system, but it needs to be run on only one node in the cluster. All the other nodes sense that the expansion has occurred.
If a GFS2 file system is full, the gfs2_jadd will fail, even if the logical volume containing the file system has been extended and is larger than the file system. This is because in a GFS2 file system, journals are plain files rather than embedded metadata, so simply extending the underlying logical volume will not provide space for the journals.
Before adding journals to a GFS file system, you can use the journals option of the gfs2_tool to find out how many journals the GFS2 file system currently contains. The following example displays the number and size of the journals in the file system mounted at /mnt/gfs2.
[root@roth-01 ../cluster/gfs2]# gfs2_tool journals /mnt/gfs2
journal2 - 128MB
journal1 - 128MB
journal0 - 128MB
3 journal(s) found.
gfs2_jadd -j Number MountPoint
Number
Specifies the number of new journals to be added.
MountPoint
Specifies the directory where the GFS2 file system is mounted.
In this example, one journal is added to the file system on the /mygfs2 directory.
gfs2_jadd -j1 /mygfs2
In this example, two journals are added to the file system on the /mygfs2 directory.
gfs2_jadd -j2 /mygfs2
gfs2_jadd [Options] {MountPoint | Device} [MountPoint | Device]
MountPoint
Specifies the directory where the GFS2 file system is mounted.
Device
Specifies the device node of the file system.
Table 3.4. GFS2-specific Options Available When Adding Journals
|
Flag
|
Parameter
|
Description
|
|---|
-h
|
|
Help. Displays short usage message.
|
-J
|
MegaBytes
|
Specifies the size of the new journals in megabytes. Default journal size is 128 megabytes. The minimum size is 32 megabytes. To add journals of different sizes to the file system, the gfs2_jadd command must be run for each size journal. The size specified is rounded down so that it is a multiple of the journal-segment size that was specified when the file system was created.
|
-j
|
Number
|
Specifies the number of new journals to be added by the gfs2_jadd command. The default value is 1.
|
-q
|
|
Quiet. Turns down the verbosity level.
|
-V
|
|
Displays command version information.
|
Ordinarily, GFS2 writes only metadata to its journal. File contents are subsequently written to disk by the kernel's periodic sync that flushes file system buffers. An fsync() call on a file causes the file's data to be written to disk immediately. The call returns when the disk reports that all data is safely written.
Data journaling can result in a reduced fsync() time for very small files because the file data is written to the journal in addition to the metadata. This advantage rapidly reduces as the file size increases. Writing to medium and larger files will be much slower with data journaling turned on.
Applications that rely on fsync() to sync file data may see improved performance by using data journaling. Data journaling can be enabled automatically for any GFS2 files created in a flagged directory (and all its subdirectories). Existing files with zero length can also have data journaling turned on or off.
Enabling data journaling on a directory sets the directory to "inherit jdata", which indicates that all files and directories subsequently created in that directory are journaled. You can enable and disable data journaling on a file with the chattr command.
The following commands enable data journaling on the /mnt/gfs2/gfs2_dir/newfile file and then check whether the flag has been set properly.
[root@roth-01 ~]# chattr +j /mnt/gfs2/gfs2_dir/newfile
[root@roth-01 ~]# lsattr /mnt/gfs2/gfs2_dir
---------j--- /mnt/gfs2/gfs2_dir/newfile
The following commands disable data journaling on the /mnt/gfs2/gfs2_dir/newfile file and then check whether the flag has been set properly.
[root@roth-01 ~]# chattr -j /mnt/gfs2/gfs2_dir/newfile
[root@roth-01 ~]# lsattr /mnt/gfs2/gfs2_dir
------------- /mnt/gfs2/gfs2_dir/newfile
You can also use the chattr command to set the j flag on a directory. When you set this flag for a directory, all files and directories subsequently created in that directory are journaled. The following set of commands sets the j flag on the gfs2_dir directory, then checks whether the flag has been set properly. After this, the commands create a new file called newfile in the /mnt/gfs2/gfs2_dir directory and then check whether the j flag has been set for the file. Since the j flag is set for the directory, then newfile should also have journaling enabled.
[root@roth-01 ~]# chattr -j /mnt/gfs2/gfs2_dir
[root@roth-01 ~]# lsattr /mnt/gfs2
---------j--- /mnt/gfs2/gfs2_dir
[root@roth-01 ~]# touch /mnt/gfs2/gfs2_dir/newfile
[root@roth-01 ~]# lsattr /mnt/gfs2/gfs2_dir
---------j--- /mnt/gfs2/gfs2_dir/newfile
3.9. Configuring atime Updates
Each file inode and directory inode has three time stamps associated with it:
ctime — The last time the inode status was changed
mtime — The last time the file (or directory) data was modified
atime — The last time the file (or directory) data was accessed
If atime updates are enabled as they are by default on GFS2 and other Linux file systems then every time a file is read, its inode needs to be updated.
Because few applications use the information provided by atime, those updates can require a significant amount of unnecessary write traffic and file locking traffic. That traffic can degrade performance; therefore, it may be preferable to turn off or reduce the frequency of atime updates.
Two methods of reducing the effects of atime updating are available:
Mount with relatime (relative atime), which updates the atime if the previous atime update is older than the mtime or ctime update.
Mount with noatime, which disables atime updates on that file system.
3.9.1. Mount with relatime
The relatime (relative atime) Linux mount option can be specified when the file system is mounted. This specifies that the atime is updated if the previous atime update is older than the mtime or ctime update.
mount BlockDevice MountPoint -o relatime
BlockDevice
Specifies the block device where the GFS2 file system resides.
MountPoint
Specifies the directory where the GFS2 file system should be mounted.
In this example, the GFS2 file system resides on the /dev/vg01/lvol0 and is mounted on directory /mygfs2. The atime updates take place only if the previous atime update is older than the mtime or ctime update.
mount /dev/vg01/lvol0 /mygfs2 -o relatime
3.9.2. Mount with noatime
The noatime Linux mount option can be specified when the file system is mounted, which disables atime updates on that file system.
mount BlockDevice MountPoint -o noatime
BlockDevice
Specifies the block device where the GFS2 file system resides.
MountPoint
Specifies the directory where the GFS2 file system should be mounted.
In this example, the GFS2 file system resides on the /dev/vg01/lvol0 and is mounted on directory /mygfs2 with atime updates turned off.
mount /dev/vg01/lvol0 /mygfs2 -o noatime
3.10. Suspending Activity on a File System
You can suspend write activity to a file system by using the dmsetup suspend command. Suspending write activity allows hardware-based device snapshots to be used to capture the file system in a consistent state. The dmsetup resume command ends the suspension.
Start Suspension
dmsetup suspend MountPoint
End Suspension
dmsetup resume MountPoint
MountPoint
Specifies the file system.
This example suspends writes to file system /mygfs2.
# dmsetup suspend /mygfs2
This example ends suspension of writes to file system /mygfs2.
# dmsetup resume /mygfs2
3.11. Repairing a File System
When nodes fail with the file system mounted, file system journaling allows fast recovery. However, if a storage device loses power or is physically disconnected, file system corruption may occur. (Journaling cannot be used to recover from storage subsystem failures.) When that type of corruption occurs, you can recover the GFS2 file system by using the fsck.gfs2 command.
The fsck.gfs2 command must be run only on a file system that is unmounted from all nodes.
You should not check a GFS2 file system at boot time with the fsck.gfs2 command. The fsck.gfs2 command can not determine at boot time whether the file system is mounted by another node in the cluster. You should run the fsck.gfs2 command manually only after the system boots.
To ensure that the fsck.gfs2 command does not run on a GFS2 file system at boot time, modify the /etc/fstab file so that the final two columns for a GFS2 file system mount point show "0 0" rather than "1 1" (or any other numbers), as in the following example:
/dev/VG12/lv_svr_home /svr_home gfs2 defaults,noatime,nodiratime,noquota 0 0
If you have previous experience using the gfs_fsck command on GFS file systems, note that the fsck.gfs2 command differs from some earlier releases of gfs_fsck in the in the following ways:
Pressing Ctrl+C while running the fsck.gfs2 interrupts processing and displays a prompt asking whether you would like to abort the command, skip the rest of the current pass, or continue processing.
You can increase the level of verbosity by using the -v flag. Adding a second -v flag increases the level again.
You can decrease the level of verbosity by using the -q flag. Adding a second -q flag decreases the level again.
The -n option opens a file system as read-only and answers no to any queries automatically. The option provides a way of trying the command to reveal errors without actually allowing the fsck.gfs2 command to take effect.
Refer to the fsck.gfs2 man page for additional information about other command options.
Running the fsck.gfs2 command requires system memory above and beyond the memory used for the operating system and kernel. Each block of memory in the GFS2 file system itself requires approximately five bits of additional memory, or 5/8 of a byte. So to estimate how many bytes of memory you will need to run the fsck.gfs2 command on your file system, determine how many blocks the file system contains and multiply that number by 5/8.
For example, to determine approximately how much memory is required to run the fsck.gfs2 command on a GFS2 file system that is 16TB with a block size of 4K, first determine how many blocks of memory the file system contains by dividing 16Tb by 4K:
17592186044416 / 4096 = 4294967296
Since this file system contains 4294967296 blocks, multiply that number by 5/8 to determine how many bytes of memory are required:
4294967296 * 5/8 = 2684354560
This file system requires approximately 2.6GB of free memory to run the fsck.gfs2 command. Note that if the block size was 1K, running the fsck.gfs2 command would require four times the memory, or approximately 11GB.
fsck.gfs2 -y BlockDevice
-y
The -y flag causes all questions to be answered with yes. With the -y flag specified, the fsck.gfs2 command does not prompt you for an answer before making changes.
BlockDevice
Specifies the block device where the GFS2 file system resides.
In this example, the GFS2 file system residing on block device /dev/testvol/testlv is repaired. All queries to repair are automatically answered with yes.
[root@dash-01 ~]# fsck.gfs2 -y /dev/testvg/testlv
Initializing fsck
Validating Resource Group index.
Level 1 RG check.
(level 1 passed)
Clearing journals (this may take a while)...
Journals cleared.
Starting pass1
Pass1 complete
Starting pass1b
Pass1b complete
Starting pass1c
Pass1c complete
Starting pass2
Pass2 complete
Starting pass3
Pass3 complete
Starting pass4
Pass4 complete
Starting pass5
Pass5 complete
Writing changes to disk
fsck.gfs2 complete
3.12. Bind Mounts and Context-Dependent Path Names
GFS2 file systems do not provide support for Context-Dependent Path Names (CDPNs), which allow you to create symbolic links that point to variable destination files or directories. For this functionality in GFS2, you can use the bind option of the mount command.
The bind option of the mount command allows you to remount part of a file hierarchy at a different location while it is still available at the original location. The format of this command is as follows.
mount --bind olddir newdir
After executing this command, the contents of the olddir directory are available at two locations: olddir and newdir. You can also use this option to make an individual file available at two locations.
For example, after executing the following commands the contents of /root/tmp will be identical to the contents of the previously mounted /var/log directory.
[root@menscryfa ~]# cd ~root
[root@menscryfa ~]# mkdir ./tmp
[root@menscryfa ~]# mount --bind /var/log /root/tmp
Alternately, you can use an entry in the /etc/fstab file to achieve the same results at mount time. The following /etc/fstab entry will result in the contents of /root/tmp being identical to the contents of the /var/log directory.
/var/log /root/tmp none bind 0 0
After you have mounted the file system, you can use the mount command to see that the file system has been mounted, as in the following example.
[root@menscryfa ~]# mount | grep /tmp
/var/log on /root/tmp type none (rw,bind)
With a file system that supports Context-Dependent Path Names, you might have defined the /bin directory as a Context-Dependent Path Name that would resolve to one of the following paths, depending on the system architecture.
/usr/i386-bin
/usr/x86_64-bin
/usr/ppc64-bin
You can achieve this same functionality by creating an empty /bin directory. Then, using a script or an entry in the /etc/fstab file, you can mount each of the individual architecture directories onto the /bin directory with a mount -bind command. For example, you can use the following command as a line in a script.
mount --bind /usr/i386-bin /bin
Alternately, you can use the following entry in the /etc/fstab file.
/usr/1386-bin /bin none bind 0 0
A bind mount can provide greater flexibility than a Context-Dependent Path Name, since you can use this feature to mount different directories according to any criteria you define (such as the value of %fill for the file system). Context-Dependent Path Names are more limited in what they can encompass. Note, however, that you will need to write your own script to mount according to a criteria such as the value of %fill.
When you mount a file system with the bind option and the original file system was mounted rw, the new file system will also be mounted rw even if you use the ro flag; the ro flag is silently ignored. In this case, the new file system might be marked as ro in the /proc/mounts directory, which may be misleading.
3.13. Bind Mounts and File System Mount Order
When you use the bind option of the mount command, you must be sure that the file systems are mounted in the correct order. In the following example, the /var/log directory must be mounted before executing the bind mount on the /tmp directory:
# mount --bind /var/log /tmp
The ordering of file system mounts is determined as follows:
In general, file system mount order is determined by the order in which the file systems appear in the fstab file. The exceptions to this ordering are file systems mounted with the _netdev flag or file systems that have their own init scripts.
A file system with its own init script is mounted later in the initialization process, after the file systems in the fstab file.
File systems mounted with the _netdev flag are mounted when the network has been enabled on the system.
If your configuration requires that you create a bind mount on which to mount a GFS2 file system, you can order your fstab file as follows:
Mount local file systems that are required for the bind mount.
Bind mount the directory on which to mount the GFS2 file system.
Mount the GFS2 file system.
If your configuration requires that you bind mount a local directory or file system onto a GFS2 file system, listing the file systems in the correct order in the fstab file will not mount the file systems correctly since the GFS2 file system will not be mounted until the GFS2 init script is run. In this case, you should write an init script to execute the bind mount so that the bind mount will not take place until after the GFS2 file system is mounted.
The following script is an example of a custom init script. This script performs a bind mount of two directories onto two directories of a GFS2 file system. In this example, there is an existing GFS2 mount point at /mnt/gfs2a, which is mounted when the GFS2 init script runs, after cluster startup.
In this example script, the values of the chkconfig statement indicate the following:
345 indicates the run levels that the script will be started in
29 is the start priority, which in this case indicates that the script will run at startup time after the GFS2 init script, which has a start priority of 26
73 is the stop priority, which in this case indicates that the script will be stopped during shutdown before the GFS2 script, which has a stop priority of 74
The start and stop values indicate that you can manually perform the indicated action by executing a service start and a service stop command. For example, if the script is named fredwilma, then you can execute service fredwilma start.
This script should be put in the /etc/init.d directory with the same permissions as the other scripts in that directory. You can then execute a chkconfig on command to link the script to the indicated run levels. For example, if the script is named fredwilma, then you can execute chkconfig fredwilma on.
#!/bin/bash
#
# chkconfig: 345 29 73
# description: mount/unmount my custom bind mounts onto a gfs2 subdirectory
#
#
### BEGIN INIT INFO
# Provides:
### END INIT INFO
. /etc/init.d/functions
case "$1" in
start)
# In this example, fred and wilma want their home directories
# bind-mounted over the gfs2 directory /mnt/gfs2a, which has
# been mounted as /mnt/gfs2a
mkdir -p /mnt/gfs2a/home/fred &> /dev/null
mkdir -p /mnt/gfs2a/home/wilma &> /dev/null
/bin/mount --bind /mnt/gfs2a/home/fred /home/fred
/bin/mount --bind /mnt/gfs2a/home/wilma /home/wilma
;;
stop)
/bin/umount /mnt/gfs2a/home/fred
/bin/umount /mnt/gfs2a/home/wilma
;;
status)
;;
restart)
$0 stop
$0 start
;;
reload)
$0 start
;;
*)
echo $"Usage: $0 {start|stop|restart|reload|status}"
exit 1
esac
exit 0
3.14. The GFS2 Withdraw Function
The GFS2 withdraw function is a data integrity feature of GFS2 file systems in a cluster. If the GFS2 kernel module detects an inconsistency in a GFS2 file system following an I/O operation, the file system becomes unavailable to the cluster. The I/O operation stops and the system waits for further I/O operations to stop with an error, preventing further damage. When this occurs, you can stop any other services or applications manually, after which you can reboot and remount the GFS2 file system to replay the journals. If the problem persists, you can unmount the file system from all nodes in the cluster and perform file system recovery with the fsck.gfs2 command. The GFS withdraw function is less severe than a kernel panic, which would cause another node to fence the node.
If your system is configured with the gfs2 startup script enabled and the GFS2 file system is included in the /etc/fstab file, the GFS2 file system will be remounted when you reboot. If the GFS2 file system withdrew because of perceived file system corruption, it is recommended that you run the fsck.gfs2 command before remounting the file system. In this case, in order to prevent your file system from remounting at boot time, you can perform the following procedure:
Temporarily disable the startup script on the affected node with the following command:
# chkconfig gfs2 off
Reboot the affected node, starting the cluster software. The GFS2 file system will not be mounted.
Unmount the file system from every node in the cluster.
Run the fsck.gfs2 on the file system from one node only to ensure there is no file system corruption.
Re-enable the startup script on the affected node by running the following command:
# chkconfig gfs2 on
Remount the GFS2 file system from all nodes in the cluster.
An example of an inconsistency that would yield a GFS2 withdraw is an incorrect block count. When the GFS kernel deletes a file from a file system, it systematically removes all the data and metadata blocks associated with that file. When it is done, it checks the block count. If the block count is not one (meaning all that is left is the disk inode itself), that indicates a file system inconsistency since the block count did not match the list of blocks found.
You can override the GFS2 withdraw function by mounting the file system with the -o errors=panic option specified. When this option is specified, any errors that would normally cause the system to withdraw cause the system to panic instead. This stops the node's cluster communications, which causes the node to be fenced.
Internally, the GFS2 withdraw function works by having the kernel send a message to the gfs_controld daemon requesting withdraw. The gfs_controld daemon runs the dmsetup program to place the device mapper error target underneath the file system preventing further access to the block device. It then tells the kernel that this has been completed. This is the reason for the GFS2 support requirement to always use a CLVM device under GFS2, since otherwise it is not possible to insert a device mapper target.
The purpose of the device mapper error target is to ensure that all future I/O operations will result in an I/O error that will allow the file system to be unmounted in an orderly fashion. As a result, when the withdraw occurs, it is normal to see a number of I/O errors from the device mapper device reported in the system logs.
Occasionally, the withdraw may fail if it is not possible for the dmsetup program to insert the error target as requested. This can happen if there is a shortage of memory at the point of the withdraw and memory cannot be reclaimed due to the problem that triggered the withdraw in the first place.
A withdraw does not always mean that there is an error in GFS2. Sometimes the withdraw function can be triggered by device I/O errors relating to the underlying block device. It is highly recommended to check the logs to see if that is the case if a withdraw occurs.
GFS2 tracepoints and the debugfs glocks File
This appendix describes both the glock debugfs interface and the GFS2 tracepoints. It is intended for advanced users who are familiar with file system internals who would like to learn more about the design of GFS2 and how to debug GFS2-specific issues.
C.1. GFS2 tracepoint Types
There are currently three types of GFS2 tracepoints: glock (pronounced "gee-lock") tracepoints, bmap tracepoints and log tracepoints. These can be used to monitor a running GFS2 file system and give additional information to that which can be obtained with the debugging options supported in previous releases of Red Hat Enterprise Linux. Tracepoints are particularly useful when a problem, such as a hang or performance issue, is reproducible and thus the tracepoint output can be obtained during the problematic operation. In GFS2, glocks are the primary cache control mechanism and they are the key to understanding the performance of the core of GFS2. The bmap (block map) tracepoints can be used to monitor block allocations and block mapping (lookup of already allocated blocks in the on-disk metadata tree) as they happen and check for any issues relating to locality of access. The log tracepoints keep track of the data being written to and released from the journal and can provide useful information on that part of GFS2.
The tracepoints are designed to be as generic as possible. This should mean that it will not be necessary to change the API during the course of Red Hat Enterprise Linux 6. On the other hand, users of this interface should be aware that this is a debugging interface and not part of the normal Red Hat Enterprise Linux 6 API set, and as such Red Hat makes no guarantees that changes in the GFS2 tracepoints interface will not occur.
Tracepoints are a generic feature of Red Hat Enterprise Linux 6 and their scope goes well beyond GFS2. In particular they are used to implement the blktrace infrastructure and the blktrace tracepoints can be used in combination with those of GFS2 to gain a fuller picture of the system performance. Due to the level at which the tracepoints operate, they can produce large volumes of data in a very short period of time. They are designed to put a minimum load on the system when they are enabled, but it is inevitable that they will have some effect. Filtering events via a variety of means can help reduce the volume of data and help focus on obtaining just the information which is useful for understanding any particular situation.
The tracepoints can be found under /sys/kernel/debug/tracing/ directory assuming that debugfs is mounted in the standard place at the /sys/kernel/debug directory. The events subdirectory contains all the tracing events that may be specified and, provided the gfs2 module is loaded, there will be a gfs2 subdirectory containing further subdirectories, one for each GFS2 event. The contents of the /sys/kernel/debug/tracing/events/gfs2 directory should look roughly like the following:
[root@chywoon gfs2]# ls
enable gfs2_bmap gfs2_glock_queue gfs2_log_flush
filter gfs2_demote_rq gfs2_glock_state_change gfs2_pin
gfs2_block_alloc gfs2_glock_put gfs2_log_blocks gfs2_promote
To enable all the GFS2 tracepoints, run the following command:
[root@chywoon gfs2]# echo -n 1 >/sys/kernel/debug/tracing/events/gfs2/enable
To enable a specific tracepoint, there is an enable file in each of the individual event subdirectories. The same is true of the filter file which can be used to set an event filter for each event or set of events. The meaning of the individual events is explained in more detail below.
The output from the tracepoints is available in ASCII or binary format. This appendix does not currently cover the binary interface. The ASCII interface is available in two ways. To list the current content of the ring buffer, you can run the following command:
[root@chywoon gfs2]# cat /sys/kernel/debug/tracing/trace
This interface is useful in cases where you are using a long-running process for a certain period of time and, after some event, want to look back at the latest captured information in the buffer. An alternative interface, /sys/kernel/debug/tracing/trace_pipe, can be used when all the output is required. Events are read from this file as they occur; there is no historical information available via this interface. The format of the output is the same from both interfaces and is described for each of the GFS2 events in the later sections of this appendix.
A utility called
trace-cmd is available for reading tracepoint data. For more information on this utility, refer to the link in
Section C.9, “References”. The
trace-cmd utility can be used in a similar way to the
strace utility, for example to run a command while gathering trace data from various sources.
To understand GFS2, the most important concept to understand, and the one which sets it aside from other file systems, is the concept of glocks. In terms of the source code, a glock is a data structure that brings together the DLM and caching into a single state machine. Each glock has a 1:1 relationship with a single DLM lock, and provides caching for that lock state so that repetitive operations carried out from a single node of the file system do not have to repeatedly call the DLM, and thus they help avoid unnecessary network traffic. There are two broad categories of glocks, those which cache metadata and those which do not. The inode glocks and the resource group glocks both cache metadata, other types of glocks do not cache metadata. The inode glock is also involved in the caching of data in addition to metadata and has the most complex logic of all glocks.
Table C.1. Glock Modes and DLM Lock Modes
|
Glock mode
|
DLM lock mode
|
Notes
|
|---|
|
UN
|
IV/NL
|
Unlocked (no DLM lock associated with glock or NL lock depending on I flag)
|
|
SH
|
PR
|
Shared (protected read) lock
|
|
EX
|
EX
|
Exclusive lock
|
|
DF
|
CW
|
Deferred (concurrent write) used for Direct I/O and file system freeze
|
Glocks remain in memory until either they are unlocked (at the request of another node or at the request of the VM) and there are no local users. At that point they are removed from the glock hash table and freed. When a glock is created, the DLM lock is not associated with the glock immediately. The DLM lock becomes associated with the glock upon the first request to the DLM, and if this request is successful then the 'I' (initial) flag will be set on the glock.
Table C.4, “Glock flags” shows the meanings of the different glock flags. Once the DLM has been associated with the glock, the DLM lock will always remain at least at NL (Null) lock mode until the glock is to be freed. A demotion of the DLM lock from NL to unlocked is always the last operation in the life of a glock.
This particular aspect of DLM lock behavior has changed since Red Hat Enterprise Linux 5, which does sometimes unlock the DLM locks attached to glocks completely, and thus Red Hat Enterprise Linux 5 has a different mechanism to ensure that LVBs (lock value blocks) are preserved where required. The new scheme that Red Hat Enterprise Linux 6 uses was made possible due to the merging of the lock_dlm lock module (not to be confused with the DLM itself) into GFS2.
Each glock can have a number of "holders" associated with it, each of which represents one lock request from the higher layers. System calls relating to GFS2 queue and dequeue holders from the glock to protect the critical section of code.
The glock state machine is based on a workqueue. For performance reasons, tasklets would be preferable; however, in the current implementation we need to submit I/O from that context which prohibits their use.
Workqueues have their own tracepoints which can be used in combination with the GFS2 tracepoints if desired
Table C.2, “Glock Modes and Data Types” shows what state may be cached under each of the glock modes and whether that cached state may be dirty. This applies to both inode and resource group locks, although there is no data component for the resource group locks, only metadata.
Table C.2. Glock Modes and Data Types
|
Glock mode
|
Cache Data
|
Cache Metadata
|
Dirty Data
|
Dirty Metadata
|
|---|
|
UN
|
No
|
No
|
No
|
No
|
|
SH
|
Yes
|
Yes
|
No
|
No
|
|
DF
|
No
|
Yes
|
No
|
No
|
|
EX
|
Yes
|
Yes
|
Yes
|
Yes
|
C.4. The glock debugfs Interface
The glock debugfs interface allows the visualization of the internal state of the glocks and the holders and it also includes some summary details of the objects being locked in some cases. Each line of the file either begins G: with no indentation (which refers to the glock itself) or it begins with a different letter, indented with a single space, and refers to the structures associated with the glock immediately above it in the file (H: is a holder, I: an inode, and R: a resource group) . Here is an example of what the content of this file might look like:
G: s:SH n:5/75320 f:I t:SH d:EX/0 a:0 r:3
H: s:SH f:EH e:0 p:4466 [postmark] gfs2_inode_lookup+0x14e/0x260 [gfs2]
G: s:EX n:3/258028 f:yI t:EX d:EX/0 a:3 r:4
H: s:EX f:tH e:0 p:4466 [postmark] gfs2_inplace_reserve_i+0x177/0x780 [gfs2]
R: n:258028 f:05 b:22256/22256 i:16800
G: s:EX n:2/219916 f:yfI t:EX d:EX/0 a:0 r:3
I: n:75661/219916 t:8 f:0x10 d:0x00000000 s:7522/7522
G: s:SH n:5/127205 f:I t:SH d:EX/0 a:0 r:3
H: s:SH f:EH e:0 p:4466 [postmark] gfs2_inode_lookup+0x14e/0x260 [gfs2]
G: s:EX n:2/50382 f:yfI t:EX d:EX/0 a:0 r:2
G: s:SH n:5/302519 f:I t:SH d:EX/0 a:0 r:3
H: s:SH f:EH e:0 p:4466 [postmark] gfs2_inode_lookup+0x14e/0x260 [gfs2]
G: s:SH n:5/313874 f:I t:SH d:EX/0 a:0 r:3
H: s:SH f:EH e:0 p:4466 [postmark] gfs2_inode_lookup+0x14e/0x260 [gfs2]
G: s:SH n:5/271916 f:I t:SH d:EX/0 a:0 r:3
H: s:SH f:EH e:0 p:4466 [postmark] gfs2_inode_lookup+0x14e/0x260 [gfs2]
G: s:SH n:5/312732 f:I t:SH d:EX/0 a:0 r:3
H: s:SH f:EH e:0 p:4466 [postmark] gfs2_inode_lookup+0x14e/0x260 [gfs2]
The above example is a series of excerpts (from an approximately 18MB file) generated by the command cat /sys/kernel/debug/gfs2/unity:myfs/glocks >my.lock during a run of the postmark benchmark on a single node GFS2 file system. The glocks in the figure have been selected in order to show some of the more interesting features of the glock dumps.
The glock states are either EX (exclusive), DF (deferred), SH (shared) or UN (unlocked). These states correspond directly with DLM lock modes except for UN which may represent either the DLM null lock state, or that GFS2 does not hold a DLM lock (depending on the I flag as explained above). The s: field of the glock indicates the current state of the lock and the same field in the holder indicates the requested mode. If the lock is granted, the holder will have the H bit set in its flags (f: field). Otherwise, it will have the W wait bit set.
The n: field (number) indicates the number associated with each item. For glocks, that is the type number followed by the glock number so that in the above example, the first glock is n:5/75320; that is, an iopen glock which relates to inode 75320. In the case of inode and iopen glocks, the glock number is always identical to the inode's disk block number.
The glock numbers (n: field) in the debugfs glocks file are in hexadecimal, whereas the tracepoints output lists them in decimal. This is for historical reasons; glock numbers were always written in hex, but decimal was chosen for the tracepoints so that the numbers could easily be compared with the other tracepoint output (from blktrace for example) and with output from stat(1).
Table C.3. Glock types
|
Type number
|
Lock type
|
Use
|
|---|
|
1
|
trans
|
Transaction lock
|
|
2
|
inode
|
Inode metadata and data
|
|
3
|
rgrp
|
Resource group metadata
|
|
4
|
meta
|
The superblock
|
|
5
|
iopen
|
Inode last closer detection
|
|
6
|
flock
|
flock(2) syscall
|
|
8
|
quota
|
Quota operations
|
|
9
|
journal
|
Journal mutex
|
One of the more important glock flags is the l (locked) flag. This is the bit lock that is used to arbitrate access to the glock state when a state change is to be performed. It is set when the state machine is about to send a remote lock request via the DLM, and only cleared when the complete operation has been performed. Sometimes this can mean that more than one lock request will have been sent, with various invalidations occurring between times.
Table C.4. Glock flags
|
Flag
|
Name
|
Meaning
|
|---|
|
d
|
Pending demote
|
A deferred (remote) demote request
|
|
D
|
Demote
|
A demote request (local or remote)
|
|
f
|
Log flush
|
The log needs to be committed before releasing this glock
|
|
F
|
Frozen
|
Replies from remote nodes ignored - recovery is in progress.
|
|
i
|
Invalidate in progress
|
In the process of invalidating pages under this glock
|
|
I
|
Initial
|
Set when DLM lock is associated with this glock
|
|
l
|
Locked
|
The glock is in the process of changing state
|
|
p
|
Demote in progress
|
The glock is in the process of responding to a demote request
|
|
r
|
Reply pending
|
Reply received from remote node is awaiting processing
|
|
y
|
Dirty
|
Data needs flushing to disk before releasing this glock
|
When a remote callback is received from a node that wants to get a lock in a mode that conflicts with that being held on the local node, then one or other of the two flags D (demote) or d (demote pending) is set. In order to prevent starvation conditions when there is contention on a particular lock, each lock is assigned a minimum hold time. A node which has not yet had the lock for the minimum hold time is allowed to retain that lock until the time interval has expired.
If the time interval has expired, then the D (demote) flag will be set and the state required will be recorded. In that case the next time there are no granted locks on the holders queue, the lock will be demoted. If the time interval has not expired, then the d (demote pending) flag is set instead. This also schedules the state machine to clear d (demote pending) and set D (demote) when the minimum hold time has expired.
The I (initial) flag is set when the glock has been assigned a DLM lock. This happens when the glock is first used and the I flag will then remain set until the glock is finally freed (which the DLM lock is unlocked).
Table C.5. Glock holder flags
|
Flag
|
Name
|
Meaning
|
|---|
|
a
|
Async
|
Do not wait for glock result (will poll for result later)
|
|
A
|
Any
|
Any compatible lock mode is acceptable
|
|
c
|
No cache
|
When unlocked, demote DLM lock immediately
|
|
e
|
No expire
|
Ignore subsequent lock cancel requests
|
|
E
|
Exact
|
Must have exact lock mode
|
|
F
|
First
|
Set when holder is the first to be granted for this lock
|
|
H
|
Holder
|
Indicates that requested lock is granted
|
|
p
|
Priority
|
Enqueue holder at the head of the queue
|
|
t
|
Try
|
A "try" lock
|
|
T
|
Try 1CB
|
A "try" lock that sends a callback
|
|
W
|
Wait
|
Set while waiting for request to complete
|
The most important holder flags are H (holder) and W (wait) as mentioned earlier, since they are set on granted lock requests and queued lock requests respectively. The ordering of the holders in the list is important. If there are any granted holders, they will always be at the head of the queue, followed by any queued holders.
If there are no granted holders, then the first holder in the list will be the one that triggers the next state change. Since demote requests are always considered higher priority than requests from the file system, that might not always directly result in a change to the state requested.
The glock subsystem supports two kinds of "try" lock. These are useful both because they allow the taking of locks out of the normal order (with suitable back-off and retry) and because they can be used to help avoid resources in use by other nodes. The normal t (try) lock is basically just what its name indicates; it is a "try" lock that does not do anything special. The T (try 1CB) lock, on the other hand, is identical to the t lock except that the DLM will send a single callback to current incompatible lock holders. One use of the T (try 1CB) lock is with the iopen locks, which are used to arbitrate among the nodes when an inode's i_nlink count is zero, and determine which of the nodes will be responsible for deallocating the inode. The iopen glock is normally held in the shared state, but when the i_nlink count becomes zero and ->delete_inode() is called, it will request an exclusive lock with T (try 1CB) set. It will continue to deallocate the inode if the lock is granted. If the lock is not granted it will result in the node(s) which were preventing the grant of the lock marking their glock(s) with the D (demote) flag, which is checked at ->drop_inode() time in order to ensure that the deallocation is not forgotten.
This means that inodes that have zero link count but are still open will be deallocated by the node on which the final close() occurs. Also, at the same time as the inode's link count is decremented to zero the inode is marked as being in the special state of having zero link count but still in use in the resource group bitmap. This functions like the ext3 file system3's orphan list in that it allows any subsequent reader of the bitmap to know that there is potentially space that might be reclaimed, and to attempt to reclaim it.
The tracepoints are also designed to be able to confirm the correctness of the cache control by combining them with the blktrace output and with knowledge of the on-disk layout. It is then possible to check that any given I/O has been issued and completed under the correct lock, and that no races are present.
The gfs2_glock_state_change tracepoint is the most important one to understand. It tracks every state change of the glock from initial creation right through to the final demotion which ends with gfs2_glock_put and the final NL to unlocked transition. The l (locked) glock flag is always set before a state change occurs and will not be cleared until after it has finished. There are never any granted holders (the H glock holder flag) during a state change. If there are any queued holders, they will always be in the W (waiting) state. When the state change is complete then the holders may be granted which is the final operation before the l glock flag is cleared.
The gfs2_demote_rq tracepoint keeps track of demote requests, both local and remote. Assuming that there is enough memory on the node, the local demote requests will rarely be seen, and most often they will be created by umount or by occasional memory reclaim. The number of remote demote requests is a measure of the contention between nodes for a particular inode or resource group.
When a holder is granted a lock, gfs2_promote is called, this occurs as the final stages of a state change or when a lock is requested which can be granted immediately due to the glock state already caching a lock of a suitable mode. If the holder is the first one to be granted for this glock, then the f (first) flag is set on that holder. This is currently used only by resource groups.
Block mapping is a task central to any file system. GFS2 uses a traditional bitmap-based system with two bits per block. The main purpose of the tracepoints in this subsystem is to allow monitoring of the time taken to allocate and map blocks.
The gfs2_bmap tracepoint is called twice for each bmap operation: once at the start to display the bmap request, and once at the end to display the result. This makes it easy to match the requests and results together and measure the time taken to map blocks n different parts of the file system, different file offsets, or even of different files. It is also possible to see what the average extent sizes being returned are in comparison to those being requested.
To keep track of allocated blocks, gfs2_block_alloc is called not only on allocations, but also on freeing of blocks. Since the allocations are all referenced according to the inode for which the block is intended, this can be used to track which physical blocks belong to which files in a live file system. This is particularly useful when combined with blktrace, which will show problematic I/O patterns that may then be referred back to the relevant inodes using the mapping gained via this tracepoint.
The tracepoints in this subsystem track blocks being added to and removed from the journal (gfs2_pin), as well as the time taken to commit the transactions to the log (gfs2_log_flush). This can be very useful when trying to debug journaling performance issues.
The gfs2_log_blocks tracepoint keeps track of the reserved blocks in the log, which can help show if the log is too small for the workload, for example.
The gfs2_ail_flush tracepoint (Red Hat Enterprise Linux 6.2 and later) is similar to the gfs2_log_flush tracepoint in that it keeps track of the start and end of flushes of the AIL list. The AIL list contains buffers which have been through the log, but have not yet been written back in place and this is periodically flushed in order to release more log space for use by the filesystem, or when a process requests a sync or fsync.
For more information about tracepoints and the GFS2 glocks file, refer to the following resources:
