diff options
Diffstat (limited to 'Documentation/filesystems')
28 files changed, 1482 insertions, 104 deletions
diff --git a/Documentation/filesystems/00-INDEX b/Documentation/filesystems/00-INDEX index 875d49696b6e..4303614b5add 100644 --- a/Documentation/filesystems/00-INDEX +++ b/Documentation/filesystems/00-INDEX @@ -16,6 +16,8 @@ befs.txt - information about the BeOS filesystem for Linux. bfs.txt - info for the SCO UnixWare Boot Filesystem (BFS). +ceph.txt + - info for the Ceph Distributed File System cifs.txt - description of the CIFS filesystem. coda.txt @@ -32,6 +34,8 @@ dlmfs.txt - info on the userspace interface to the OCFS2 DLM. dnotify.txt - info about directory notification in Linux. +dnotify_test.c + - example program for dnotify ecryptfs.txt - docs on eCryptfs: stacked cryptographic filesystem for Linux. exofs.txt @@ -62,6 +66,8 @@ jfs.txt - info and mount options for the JFS filesystem. locks.txt - info on file locking implementations, flock() vs. fcntl(), etc. +logfs.txt + - info on the LogFS flash filesystem. mandatory-locking.txt - info on the Linux implementation of Sys V mandatory file locking. ncpfs.txt diff --git a/Documentation/filesystems/9p.txt b/Documentation/filesystems/9p.txt index 57e0b80a5274..c0236e753bc8 100644 --- a/Documentation/filesystems/9p.txt +++ b/Documentation/filesystems/9p.txt @@ -37,6 +37,15 @@ For Plan 9 From User Space applications (http://swtch.com/plan9) mount -t 9p `namespace`/acme /mnt/9 -o trans=unix,uname=$USER +For server running on QEMU host with virtio transport: + + mount -t 9p -o trans=virtio <mount_tag> /mnt/9 + +where mount_tag is the tag associated by the server to each of the exported +mount points. Each 9P export is seen by the client as a virtio device with an +associated "mount_tag" property. Available mount tags can be +seen by reading /sys/bus/virtio/drivers/9pnet_virtio/virtio<n>/mount_tag files. + OPTIONS ======= @@ -47,7 +56,7 @@ OPTIONS fd - used passed file descriptors for connection (see rfdno and wfdno) virtio - connect to the next virtio channel available - (from lguest or KVM with trans_virtio module) + (from QEMU with trans_virtio module) rdma - connect to a specified RDMA channel uname=name user name to attempt mount as on the remote server. The @@ -85,7 +94,12 @@ OPTIONS port=n port to connect to on the remote server - noextend force legacy mode (no 9p2000.u semantics) + noextend force legacy mode (no 9p2000.u or 9p2000.L semantics) + + version=name Select 9P protocol version. Valid options are: + 9p2000 - Legacy mode (same as noextend) + 9p2000.u - Use 9P2000.u protocol + 9p2000.L - Use 9P2000.L protocol dfltuid attempt to mount as a particular uid diff --git a/Documentation/filesystems/Locking b/Documentation/filesystems/Locking index 18b9d0ca0630..96d4293607ec 100644 --- a/Documentation/filesystems/Locking +++ b/Documentation/filesystems/Locking @@ -178,7 +178,7 @@ prototypes: locking rules: All except set_page_dirty may block - BKL PageLocked(page) i_sem + BKL PageLocked(page) i_mutex writepage: no yes, unlocks (see below) readpage: no yes, unlocks sync_page: no maybe @@ -380,7 +380,7 @@ prototypes: int (*open) (struct inode *, struct file *); int (*flush) (struct file *); int (*release) (struct inode *, struct file *); - int (*fsync) (struct file *, struct dentry *, int datasync); + int (*fsync) (struct file *, int datasync); int (*aio_fsync) (struct kiocb *, int datasync); int (*fasync) (int, struct file *, int); int (*lock) (struct file *, int, struct file_lock *); @@ -429,8 +429,9 @@ check_flags: no implementations. If your fs is not using generic_file_llseek, you need to acquire and release the appropriate locks in your ->llseek(). For many filesystems, it is probably safe to acquire the inode -semaphore. Note some filesystems (i.e. remote ones) provide no -protection for i_size so you will need to use the BKL. +mutex or just to use i_size_read() instead. +Note: this does not protect the file->f_pos against concurrent modifications +since this is something the userspace has to take care about. Note: ext2_release() was *the* source of contention on fs-intensive loads and dropping BKL on ->release() helps to get rid of that (we still @@ -460,13 +461,6 @@ in sys_read() and friends. --------------------------- dquot_operations ------------------------------- prototypes: - int (*initialize) (struct inode *, int); - int (*drop) (struct inode *); - int (*alloc_space) (struct inode *, qsize_t, int); - int (*alloc_inode) (const struct inode *, unsigned long); - int (*free_space) (struct inode *, qsize_t); - int (*free_inode) (const struct inode *, unsigned long); - int (*transfer) (struct inode *, struct iattr *); int (*write_dquot) (struct dquot *); int (*acquire_dquot) (struct dquot *); int (*release_dquot) (struct dquot *); @@ -479,13 +473,6 @@ a proper locking wrt the filesystem and call the generic quota operations. What filesystem should expect from the generic quota functions: FS recursion Held locks when called -initialize: yes maybe dqonoff_sem -drop: yes - -alloc_space: ->mark_dirty() - -alloc_inode: ->mark_dirty() - -free_space: ->mark_dirty() - -free_inode: ->mark_dirty() - -transfer: yes - write_dquot: yes dqonoff_sem or dqptr_sem acquire_dquot: yes dqonoff_sem or dqptr_sem release_dquot: yes dqonoff_sem or dqptr_sem @@ -495,10 +482,6 @@ write_info: yes dqonoff_sem FS recursion means calling ->quota_read() and ->quota_write() from superblock operations. -->alloc_space(), ->alloc_inode(), ->free_space(), ->free_inode() are called -only directly by the filesystem and do not call any fs functions only -the ->mark_dirty() operation. - More details about quota locking can be found in fs/dquot.c. --------------------------- vm_operations_struct ----------------------------- diff --git a/Documentation/filesystems/Makefile b/Documentation/filesystems/Makefile new file mode 100644 index 000000000000..a5dd114da14f --- /dev/null +++ b/Documentation/filesystems/Makefile @@ -0,0 +1,8 @@ +# kbuild trick to avoid linker error. Can be omitted if a module is built. +obj- := dummy.o + +# List of programs to build +hostprogs-y := dnotify_test + +# Tell kbuild to always build the programs +always := $(hostprogs-y) diff --git a/Documentation/filesystems/autofs4-mount-control.txt b/Documentation/filesystems/autofs4-mount-control.txt index 8f78ded4b648..51986bf08a4d 100644 --- a/Documentation/filesystems/autofs4-mount-control.txt +++ b/Documentation/filesystems/autofs4-mount-control.txt @@ -146,7 +146,7 @@ found to be inadequate, in this case. The Generic Netlink system was used for this as raw Netlink would lead to a significant increase in complexity. There's no question that the Generic Netlink system is an elegant solution for common case ioctl functions but it's not a complete -replacement probably because it's primary purpose in life is to be a +replacement probably because its primary purpose in life is to be a message bus implementation rather than specifically an ioctl replacement. While it would be possible to work around this there is one concern that lead to the decision to not use it. This is that the autofs diff --git a/Documentation/filesystems/ceph.txt b/Documentation/filesystems/ceph.txt new file mode 100644 index 000000000000..763d8ebbbebd --- /dev/null +++ b/Documentation/filesystems/ceph.txt @@ -0,0 +1,140 @@ +Ceph Distributed File System +============================ + +Ceph is a distributed network file system designed to provide good +performance, reliability, and scalability. + +Basic features include: + + * POSIX semantics + * Seamless scaling from 1 to many thousands of nodes + * High availability and reliability. No single point of failure. + * N-way replication of data across storage nodes + * Fast recovery from node failures + * Automatic rebalancing of data on node addition/removal + * Easy deployment: most FS components are userspace daemons + +Also, + * Flexible snapshots (on any directory) + * Recursive accounting (nested files, directories, bytes) + +In contrast to cluster filesystems like GFS, OCFS2, and GPFS that rely +on symmetric access by all clients to shared block devices, Ceph +separates data and metadata management into independent server +clusters, similar to Lustre. Unlike Lustre, however, metadata and +storage nodes run entirely as user space daemons. Storage nodes +utilize btrfs to store data objects, leveraging its advanced features +(checksumming, metadata replication, etc.). File data is striped +across storage nodes in large chunks to distribute workload and +facilitate high throughputs. When storage nodes fail, data is +re-replicated in a distributed fashion by the storage nodes themselves +(with some minimal coordination from a cluster monitor), making the +system extremely efficient and scalable. + +Metadata servers effectively form a large, consistent, distributed +in-memory cache above the file namespace that is extremely scalable, +dynamically redistributes metadata in response to workload changes, +and can tolerate arbitrary (well, non-Byzantine) node failures. The +metadata server takes a somewhat unconventional approach to metadata +storage to significantly improve performance for common workloads. In +particular, inodes with only a single link are embedded in +directories, allowing entire directories of dentries and inodes to be +loaded into its cache with a single I/O operation. The contents of +extremely large directories can be fragmented and managed by +independent metadata servers, allowing scalable concurrent access. + +The system offers automatic data rebalancing/migration when scaling +from a small cluster of just a few nodes to many hundreds, without +requiring an administrator carve the data set into static volumes or +go through the tedious process of migrating data between servers. +When the file system approaches full, new nodes can be easily added +and things will "just work." + +Ceph includes flexible snapshot mechanism that allows a user to create +a snapshot on any subdirectory (and its nested contents) in the +system. Snapshot creation and deletion are as simple as 'mkdir +.snap/foo' and 'rmdir .snap/foo'. + +Ceph also provides some recursive accounting on directories for nested +files and bytes. That is, a 'getfattr -d foo' on any directory in the +system will reveal the total number of nested regular files and +subdirectories, and a summation of all nested file sizes. This makes +the identification of large disk space consumers relatively quick, as +no 'du' or similar recursive scan of the file system is required. + + +Mount Syntax +============ + +The basic mount syntax is: + + # mount -t ceph monip[:port][,monip2[:port]...]:/[subdir] mnt + +You only need to specify a single monitor, as the client will get the +full list when it connects. (However, if the monitor you specify +happens to be down, the mount won't succeed.) The port can be left +off if the monitor is using the default. So if the monitor is at +1.2.3.4, + + # mount -t ceph 1.2.3.4:/ /mnt/ceph + +is sufficient. If /sbin/mount.ceph is installed, a hostname can be +used instead of an IP address. + + + +Mount Options +============= + + ip=A.B.C.D[:N] + Specify the IP and/or port the client should bind to locally. + There is normally not much reason to do this. If the IP is not + specified, the client's IP address is determined by looking at the + address its connection to the monitor originates from. + + wsize=X + Specify the maximum write size in bytes. By default there is no + maximum. Ceph will normally size writes based on the file stripe + size. + + rsize=X + Specify the maximum readahead. + + mount_timeout=X + Specify the timeout value for mount (in seconds), in the case + of a non-responsive Ceph file system. The default is 30 + seconds. + + rbytes + When stat() is called on a directory, set st_size to 'rbytes', + the summation of file sizes over all files nested beneath that + directory. This is the default. + + norbytes + When stat() is called on a directory, set st_size to the + number of entries in that directory. + + nocrc + Disable CRC32C calculation for data writes. If set, the storage node + must rely on TCP's error correction to detect data corruption + in the data payload. + + noasyncreaddir + Disable client's use its local cache to satisfy readdir + requests. (This does not change correctness; the client uses + cached metadata only when a lease or capability ensures it is + valid.) + + +More Information +================ + +For more information on Ceph, see the home page at + http://ceph.newdream.net/ + +The Linux kernel client source tree is available at + git://ceph.newdream.net/git/ceph-client.git + git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client.git + +and the source for the full system is at + git://ceph.newdream.net/git/ceph.git diff --git a/Documentation/filesystems/dentry-locking.txt b/Documentation/filesystems/dentry-locking.txt index 4c0c575a4012..79334ed5daa7 100644 --- a/Documentation/filesystems/dentry-locking.txt +++ b/Documentation/filesystems/dentry-locking.txt @@ -62,7 +62,8 @@ changes are : 2. Insertion of a dentry into the hash table is done using hlist_add_head_rcu() which take care of ordering the writes - the writes to the dentry must be visible before the dentry is - inserted. This works in conjunction with hlist_for_each_rcu() while + inserted. This works in conjunction with hlist_for_each_rcu(), + which has since been replaced by hlist_for_each_entry_rcu(), while walking the hash chain. The only requirement is that all initialization to the dentry must be done before hlist_add_head_rcu() since we don't have dcache_lock protection diff --git a/Documentation/filesystems/dlmfs.txt b/Documentation/filesystems/dlmfs.txt index c50bbb2d52b4..1b528b2ad809 100644 --- a/Documentation/filesystems/dlmfs.txt +++ b/Documentation/filesystems/dlmfs.txt @@ -47,7 +47,7 @@ You'll want to start heartbeating on a volume which all the nodes in your lockspace can access. The easiest way to do this is via ocfs2_hb_ctl (distributed with ocfs2-tools). Right now it requires that an OCFS2 file system be in place so that it can automatically -find it's heartbeat area, though it will eventually support heartbeat +find its heartbeat area, though it will eventually support heartbeat against raw disks. Please see the ocfs2_hb_ctl and mkfs.ocfs2 manual pages distributed diff --git a/Documentation/filesystems/dnotify.txt b/Documentation/filesystems/dnotify.txt index 9f5d338ddbb8..6baf88f46859 100644 --- a/Documentation/filesystems/dnotify.txt +++ b/Documentation/filesystems/dnotify.txt @@ -62,38 +62,9 @@ disabled, fcntl(fd, F_NOTIFY, ...) will return -EINVAL. Example ------- +See Documentation/filesystems/dnotify_test.c for an example. - #define _GNU_SOURCE /* needed to get the defines */ - #include <fcntl.h> /* in glibc 2.2 this has the needed - values defined */ - #include <signal.h> - #include <stdio.h> - #include <unistd.h> - - static volatile int event_fd; - - static void handler(int sig, siginfo_t *si, void *data) - { - event_fd = si->si_fd; - } - - int main(void) - { - struct sigaction act; - int fd; - - act.sa_sigaction = handler; - sigemptyset(&act.sa_mask); - act.sa_flags = SA_SIGINFO; - sigaction(SIGRTMIN + 1, &act, NULL); - - fd = open(".", O_RDONLY); - fcntl(fd, F_SETSIG, SIGRTMIN + 1); - fcntl(fd, F_NOTIFY, DN_MODIFY|DN_CREATE|DN_MULTISHOT); - /* we will now be notified if any of the files - in "." is modified or new files are created */ - while (1) { - pause(); - printf("Got event on fd=%d\n", event_fd); - } - } +NOTE +---- +Beginning with Linux 2.6.13, dnotify has been replaced by inotify. +See Documentation/filesystems/inotify.txt for more information on it. diff --git a/Documentation/filesystems/dnotify_test.c b/Documentation/filesystems/dnotify_test.c new file mode 100644 index 000000000000..8b37b4a1e18d --- /dev/null +++ b/Documentation/filesystems/dnotify_test.c @@ -0,0 +1,34 @@ +#define _GNU_SOURCE /* needed to get the defines */ +#include <fcntl.h> /* in glibc 2.2 this has the needed + values defined */ +#include <signal.h> +#include <stdio.h> +#include <unistd.h> + +static volatile int event_fd; + +static void handler(int sig, siginfo_t *si, void *data) +{ + event_fd = si->si_fd; +} + +int main(void) +{ + struct sigaction act; + int fd; + + act.sa_sigaction = handler; + sigemptyset(&act.sa_mask); + act.sa_flags = SA_SIGINFO; + sigaction(SIGRTMIN + 1, &act, NULL); + + fd = open(".", O_RDONLY); + fcntl(fd, F_SETSIG, SIGRTMIN + 1); + fcntl(fd, F_NOTIFY, DN_MODIFY|DN_CREATE|DN_MULTISHOT); + /* we will now be notified if any of the files + in "." is modified or new files are created */ + while (1) { + pause(); + printf("Got event on fd=%d\n", event_fd); + } +} diff --git a/Documentation/filesystems/ext3.txt b/Documentation/filesystems/ext3.txt index 867c5b50cb42..272f80d5f966 100644 --- a/Documentation/filesystems/ext3.txt +++ b/Documentation/filesystems/ext3.txt @@ -59,8 +59,19 @@ commit=nrsec (*) Ext3 can be told to sync all its data and metadata Setting it to very large values will improve performance. -barrier=1 This enables/disables barriers. barrier=0 disables - it, barrier=1 enables it. +barrier=<0(*)|1> This enables/disables the use of write barriers in +barrier the jbd code. barrier=0 disables, barrier=1 enables. +nobarrier (*) This also requires an IO stack which can support + barriers, and if jbd gets an error on a barrier + write, it will disable again with a warning. + Write barriers enforce proper on-disk ordering + of journal commits, making volatile disk write caches + safe to use, at some performance penalty. If + your disks are battery-backed in one way or another, + disabling barriers may safely improve performance. + The mount options "barrier" and "nobarrier" can + also be used to enable or disable barriers, for + consistency with other ext3 mount options. orlov (*) This enables the new Orlov block allocator. It is enabled by default. diff --git a/Documentation/filesystems/fiemap.txt b/Documentation/filesystems/fiemap.txt index 606233cd4618..1b805a0efbb0 100644 --- a/Documentation/filesystems/fiemap.txt +++ b/Documentation/filesystems/fiemap.txt @@ -38,7 +38,7 @@ flags, it will return EBADR and the contents of fm_flags will contain the set of flags which caused the error. If the kernel is compatible with all flags passed, the contents of fm_flags will be unmodified. It is up to userspace to determine whether rejection of a particular -flag is fatal to it's operation. This scheme is intended to allow the +flag is fatal to its operation. This scheme is intended to allow the fiemap interface to grow in the future but without losing compatibility with old software. @@ -56,7 +56,7 @@ If this flag is set, the kernel will sync the file before mapping extents. * FIEMAP_FLAG_XATTR If this flag is set, the extents returned will describe the inodes -extended attribute lookup tree, instead of it's data tree. +extended attribute lookup tree, instead of its data tree. Extent Mapping @@ -89,7 +89,7 @@ struct fiemap_extent { }; All offsets and lengths are in bytes and mirror those on disk. It is valid -for an extents logical offset to start before the request or it's logical +for an extents logical offset to start before the request or its logical length to extend past the request. Unless FIEMAP_EXTENT_NOT_ALIGNED is returned, fe_logical, fe_physical, and fe_length will be aligned to the block size of the file system. With the exception of extents flagged as @@ -125,7 +125,7 @@ been allocated for the file yet. * FIEMAP_EXTENT_DELALLOC - This will also set FIEMAP_EXTENT_UNKNOWN. -Delayed allocation - while there is data for this extent, it's +Delayed allocation - while there is data for this extent, its physical location has not been allocated yet. * FIEMAP_EXTENT_ENCODED @@ -159,7 +159,7 @@ Data is located within a meta data block. Data is packed into a block with data from other files. * FIEMAP_EXTENT_UNWRITTEN -Unwritten extent - the extent is allocated but it's data has not been +Unwritten extent - the extent is allocated but its data has not been initialized. This indicates the extent's data will be all zero if read through the filesystem but the contents are undefined if read directly from the device. @@ -176,7 +176,7 @@ VFS -> File System Implementation File systems wishing to support fiemap must implement a ->fiemap callback on their inode_operations structure. The fs ->fiemap call is responsible for -defining it's set of supported fiemap flags, and calling a helper function on +defining its set of supported fiemap flags, and calling a helper function on each discovered extent: struct inode_operations { diff --git a/Documentation/filesystems/fuse.txt b/Documentation/filesystems/fuse.txt index 397a41adb4c3..13af4a49e7db 100644 --- a/Documentation/filesystems/fuse.txt +++ b/Documentation/filesystems/fuse.txt @@ -91,7 +91,7 @@ Mount options 'default_permissions' By default FUSE doesn't check file access permissions, the - filesystem is free to implement it's access policy or leave it to + filesystem is free to implement its access policy or leave it to the underlying file access mechanism (e.g. in case of network filesystems). This option enables permission checking, restricting access based on file mode. It is usually useful together with the @@ -171,7 +171,7 @@ or may honor them by sending a reply to the _original_ request, with the error set to EINTR. It is also possible that there's a race between processing the -original request and it's INTERRUPT request. There are two possibilities: +original request and its INTERRUPT request. There are two possibilities: 1) The INTERRUPT request is processed before the original request is processed diff --git a/Documentation/filesystems/gfs2.txt b/Documentation/filesystems/gfs2.txt index 5e3ab8f3beff..0b59c0200912 100644 --- a/Documentation/filesystems/gfs2.txt +++ b/Documentation/filesystems/gfs2.txt @@ -1,7 +1,7 @@ Global File System ------------------ -http://sources.redhat.com/cluster/ +http://sources.redhat.com/cluster/wiki/ GFS is a cluster file system. It allows a cluster of computers to simultaneously use a block device that is shared between them (with FC, @@ -36,11 +36,11 @@ GFS2 is not on-disk compatible with previous versions of GFS, but it is pretty close. The following man pages can be found at the URL above: - fsck.gfs2 to repair a filesystem - gfs2_grow to expand a filesystem online - gfs2_jadd to add journals to a filesystem online - gfs2_tool to manipulate, examine and tune a filesystem + fsck.gfs2 to repair a filesystem + gfs2_grow to expand a filesystem online + gfs2_jadd to add journals to a filesystem online + gfs2_tool to manipulate, examine and tune a filesystem gfs2_quota to examine and change quota values in a filesystem gfs2_convert to convert a gfs filesystem to gfs2 in-place mount.gfs2 to help mount(8) mount a filesystem - mkfs.gfs2 to make a filesystem + mkfs.gfs2 to make a filesystem diff --git a/Documentation/filesystems/hpfs.txt b/Documentation/filesystems/hpfs.txt index fa45c3baed98..74630bd504fb 100644 --- a/Documentation/filesystems/hpfs.txt +++ b/Documentation/filesystems/hpfs.txt @@ -103,7 +103,7 @@ to analyze or change OS2SYS.INI. Codepages HPFS can contain several uppercasing tables for several codepages and each -file has a pointer to codepage it's name is in. However OS/2 was created in +file has a pointer to codepage its name is in. However OS/2 was created in America where people don't care much about codepages and so multiple codepages support is quite buggy. I have Czech OS/2 working in codepage 852 on my disk. Once I booted English OS/2 working in cp 850 and I created a file on my 852 diff --git a/Documentation/filesystems/logfs.txt b/Documentation/filesystems/logfs.txt new file mode 100644 index 000000000000..bca42c22a143 --- /dev/null +++ b/Documentation/filesystems/logfs.txt @@ -0,0 +1,241 @@ + +The LogFS Flash Filesystem +========================== + +Specification +============= + +Superblocks +----------- + +Two superblocks exist at the beginning and end of the filesystem. +Each superblock is 256 Bytes large, with another 3840 Bytes reserved +for future purposes, making a total of 4096 Bytes. + +Superblock locations may differ for MTD and block devices. On MTD the +first non-bad block contains a superblock in the first 4096 Bytes and +the last non-bad block contains a superblock in the last 4096 Bytes. +On block devices, the first 4096 Bytes of the device contain the first +superblock and the last aligned 4096 Byte-block contains the second +superblock. + +For the most part, the superblocks can be considered read-only. They +are written only to correct errors detected within the superblocks, +move the journal and change the filesystem parameters through tunefs. +As a result, the superblock does not contain any fields that require +constant updates, like the amount of free space, etc. + +Segments +-------- + +The space in the device is split up into equal-sized segments. +Segments are the primary write unit of LogFS. Within each segments, +writes happen from front (low addresses) to back (high addresses. If +only a partial segment has been written, the segment number, the +current position within and optionally a write buffer are stored in +the journal. + +Segments are erased as a whole. Therefore Garbage Collection may be +required to completely free a segment before doing so. + +Journal +-------- + +The journal contains all global information about the filesystem that +is subject to frequent change. At mount time, it has to be scanned +for the most recent commit entry, which contains a list of pointers to +all currently valid entries. + +Object Store +------------ + +All space except for the superblocks and journal is part of the object +store. Each segment contains a segment header and a number of +objects, each consisting of the object header and the payload. +Objects are either inodes, directory entries (dentries), file data +blocks or indirect blocks. + +Levels +------ + +Garbage collection (GC) may fail if all data is written +indiscriminately. One requirement of GC is that data is separated +roughly according to the distance between the tree root and the data. +Effectively that means all file data is on level 0, indirect blocks +are on levels 1, 2, 3 4 or 5 for 1x, 2x, 3x, 4x or 5x indirect blocks, +respectively. Inode file data is on level 6 for the inodes and 7-11 +for indirect blocks. + +Each segment contains objects of a single level only. As a result, +each level requires its own separate segment to be open for writing. + +Inode File +---------- + +All inodes are stored in a special file, the inode file. Single +exception is the inode file's inode (master inode) which for obvious +reasons is stored in the journal instead. Instead of data blocks, the +leaf nodes of the inode files are inodes. + +Aliases +------- + +Writes in LogFS are done by means of a wandering tree. A naïve +implementation would require that for each write or a block, all +parent blocks are written as well, since the block pointers have +changed. Such an implementation would not be very efficient. + +In LogFS, the block pointer changes are cached in the journal by means +of alias entries. Each alias consists of its logical address - inode +number, block index, level and child number (index into block) - and +the changed data. Any 8-byte word can be changes in this manner. + +Currently aliases are used for block pointers, file size, file used +bytes and the height of an inodes indirect tree. + +Segment Aliases +--------------- + +Related to regular aliases, these are used to handle bad blocks. +Initially, bad blocks are handled by moving the affected segment +content to a spare segment and noting this move in the journal with a +segment alias, a simple (to, from) tupel. GC will later empty this +segment and the alias can be removed again. This is used on MTD only. + +Vim +--- + +By cleverly predicting the life time of data, it is possible to +separate long-living data from short-living data and thereby reduce +the GC overhead later. Each type of distinc life expectency (vim) can +have a separate segment open for writing. Each (level, vim) tupel can +be open just once. If an open segment with unknown vim is encountered +at mount time, it is closed and ignored henceforth. + +Indirect Tree +------------- + +Inodes in LogFS are similar to FFS-style filesystems with direct and +indirect block pointers. One difference is that LogFS uses a single +indirect pointer that can be either a 1x, 2x, etc. indirect pointer. +A height field in the inode defines the height of the indirect tree +and thereby the indirection of the pointer. + +Another difference is the addressing of indirect blocks. In LogFS, +the first 16 pointers in the first indirect block are left empty, +corresponding to the 16 direct pointers in the inode. In ext2 (maybe +others as well) the first pointer in the first indirect block +corresponds to logical block 12, skipping the 12 direct pointers. +So where ext2 is using arithmetic to better utilize space, LogFS keeps +arithmetic simple and uses compression to save space. + +Compression +----------- + +Both file data and metadata can be compressed. Compression for file +data can be enabled with chattr +c and disabled with chattr -c. Doing +so has no effect on existing data, but new data will be stored +accordingly. New inodes will inherit the compression flag of the +parent directory. + +Metadata is always compressed. However, the space accounting ignores +this and charges for the uncompressed size. Failing to do so could +result in GC failures when, after moving some data, indirect blocks +compress worse than previously. Even on a 100% full medium, GC may +not consume any extra space, so the compression gains are lost space +to the user. + +However, they are not lost space to the filesystem internals. By +cheating the user for those bytes, the filesystem gained some slack +space and GC will run less often and faster. + +Garbage Collection and Wear Leveling +------------------------------------ + +Garbage collection is invoked whenever the number of free segments +falls below a threshold. The best (known) candidate is picked based +on the least amount of valid data contained in the segment. All +remaining valid data is copied elsewhere, thereby invalidating it. + +The GC code also checks for aliases and writes then back if their +number gets too large. + +Wear leveling is done by occasionally picking a suboptimal segment for +garbage collection. If a stale segments erase count is significantly +lower than the active segments' erase counts, it will be picked. Wear +leveling is rate limited, so it will never monopolize the device for +more than one segment worth at a time. + +Values for "occasionally", "significantly lower" are compile time +constants. + +Hashed directories +------------------ + +To satisfy efficient lookup(), directory entries are hashed and +located based on the hash. In order to both support large directories +and not be overly inefficient for small directories, several hash +tables of increasing size are used. For each table, the hash value +modulo the table size gives the table index. + +Tables sizes are chosen to limit the number of indirect blocks with a +fully populated table to 0, 1, 2 or 3 respectively. So the first +table contains 16 entries, the second 512-16, etc. + +The last table is special in several ways. First its size depends on +the effective 32bit limit on telldir/seekdir cookies. Since logfs +uses the upper half of the address space for indirect blocks, the size +is limited to 2^31. Secondly the table contains hash buckets with 16 +entries each. + +Using single-entry buckets would result in birthday "attacks". At +just 2^16 used entries, hash collisions would be likely (P >= 0.5). +My math skills are insufficient to do the combinatorics for the 17x +collisions necessary to overflow a bucket, but testing showed that in +10,000 runs the lowest directory fill before a bucket overflow was +188,057,130 entries with an average of 315,149,915 entries. So for +directory sizes of up to a million, bucket overflows should be +virtually impossible under normal circumstances. + +With carefully chosen filenames, it is obviously possible to cause an +overflow with just 21 entries (4 higher tables + 16 entries + 1). So +there may be a security concern if a malicious user has write access +to a directory. + +Open For Discussion +=================== + +Device Address Space +-------------------- + +A device address space is used for caching. Both block devices and +MTD provide functions to either read a single page or write a segment. +Partial segments may be written for data integrity, but where possible +complete segments are written for performance on simple block device +flash media. + +Meta Inodes +----------- + +Inodes are stored in the inode file, which is just a regular file for +most purposes. At umount time, however, the inode file needs to +remain open until all dirty inodes are written. So +generic_shutdown_super() may not close this inode, but shouldn't +complain about remaining inodes due to the inode file either. Same +goes for mapping inode of the device address space. + +Currently logfs uses a hack that essentially copies part of fs/inode.c +code over. A general solution would be preferred. + +Indirect block mapping +---------------------- + +With compression, the block device (or mapping inode) cannot be used +to cache indirect blocks. Some other place is required. Currently +logfs uses the top half of each inode's address space. The low 8TB +(on 32bit) are filled with file data, the high 8TB are used for +indirect blocks. + +One problem is that 16TB files created on 64bit systems actually have +data in the top 8TB. But files >16TB would cause problems anyway, so +only the limit has changed. diff --git a/Documentation/filesystems/nfs/nfs41-server.txt b/Documentation/filesystems/nfs/nfs41-server.txt index 1bd0d0c05171..04884914a1c8 100644 --- a/Documentation/filesystems/nfs/nfs41-server.txt +++ b/Documentation/filesystems/nfs/nfs41-server.txt @@ -17,8 +17,7 @@ kernels must turn 4.1 on or off *before* turning support for version 4 on or off; rpc.nfsd does this correctly.) The NFSv4 minorversion 1 (NFSv4.1) implementation in nfsd is based -on the latest NFSv4.1 Internet Draft: -http://tools.ietf.org/html/draft-ietf-nfsv4-minorversion1-29 +on RFC 5661. From the many new features in NFSv4.1 the current implementation focuses on the mandatory-to-implement NFSv4.1 Sessions, providing @@ -44,7 +43,7 @@ interoperability problems with future clients. Known issues: trunking, but this is a mandatory feature, and its use is recommended to clients in a number of places. (E.g. to ensure timely renewal in case an existing connection's retry timeouts - have gotten too long; see section 8.3 of the draft.) + have gotten too long; see section 8.3 of the RFC.) Therefore, lack of this feature may cause future clients to fail. - Incomplete backchannel support: incomplete backchannel gss @@ -138,7 +137,7 @@ NS*| OPENATTR | OPT | | Section 18.17 | | READ | REQ | | Section 18.22 | | READDIR | REQ | | Section 18.23 | | READLINK | OPT | | Section 18.24 | -NS | RECLAIM_COMPLETE | REQ | | Section 18.51 | + | RECLAIM_COMPLETE | REQ | | Section 18.51 | | RELEASE_LOCKOWNER | MNI | | N/A | | REMOVE | REQ | | Section 18.25 | | RENAME | REQ | | Section 18.26 | diff --git a/Documentation/filesystems/nfs/rpc-cache.txt b/Documentation/filesystems/nfs/rpc-cache.txt index 8a382bea6808..ebcaaee21616 100644 --- a/Documentation/filesystems/nfs/rpc-cache.txt +++ b/Documentation/filesystems/nfs/rpc-cache.txt @@ -185,7 +185,7 @@ failed lookup meant a definite 'no'. request/response format ----------------------- -While each cache is free to use it's own format for requests +While each cache is free to use its own format for requests and responses over channel, the following is recommended as appropriate and support routines are available to help: Each request or response record should be printable ASCII diff --git a/Documentation/filesystems/nilfs2.txt b/Documentation/filesystems/nilfs2.txt index 839efd8a8a8c..d3e7673995eb 100644 --- a/Documentation/filesystems/nilfs2.txt +++ b/Documentation/filesystems/nilfs2.txt @@ -50,8 +50,8 @@ NILFS2 supports the following mount options: (*) == default nobarrier Disables barriers. -errors=continue(*) Keep going on a filesystem error. -errors=remount-ro Remount the filesystem read-only on an error. +errors=continue Keep going on a filesystem error. +errors=remount-ro(*) Remount the filesystem read-only on an error. errors=panic Panic and halt the machine if an error occurs. cp=n Specify the checkpoint-number of the snapshot to be mounted. Checkpoints and snapshots are listed by lscp @@ -74,6 +74,9 @@ norecovery Disable recovery of the filesystem on mount. This disables every write access on the device for read-only mounts or snapshots. This option will fail for r/w mounts on an unclean volume. +discard Issue discard/TRIM commands to the underlying block + device when blocks are freed. This is useful for SSD + devices and sparse/thinly-provisioned LUNs. NILFS2 usage ============ diff --git a/Documentation/filesystems/ocfs2.txt b/Documentation/filesystems/ocfs2.txt index c58b9f5ba002..1f7ae144f6d8 100644 --- a/Documentation/filesystems/ocfs2.txt +++ b/Documentation/filesystems/ocfs2.txt @@ -80,3 +80,10 @@ user_xattr (*) Enables Extended User Attributes. nouser_xattr Disables Extended User Attributes. acl Enables POSIX Access Control Lists support. noacl (*) Disables POSIX Access Control Lists support. +resv_level=2 (*) Set how agressive allocation reservations will be. + Valid values are between 0 (reservations off) to 8 + (maximum space for reservations). +dir_resv_level= (*) By default, directory reservations will scale with file + reservations - users should rarely need to change this + value. If allocation reservations are turned off, this + option will have no effect. diff --git a/Documentation/filesystems/proc.txt b/Documentation/filesystems/proc.txt index 0d07513a67a6..9fb6cbe70bde 100644 --- a/Documentation/filesystems/proc.txt +++ b/Documentation/filesystems/proc.txt @@ -164,6 +164,7 @@ read the file /proc/PID/status: VmExe: 68 kB VmLib: 1412 kB VmPTE: 20 kb + VmSwap: 0 kB Threads: 1 SigQ: 0/28578 SigPnd: 0000000000000000 @@ -188,7 +189,13 @@ memory usage. Its seven fields are explained in Table 1-3. The stat file contains details information about the process itself. Its fields are explained in Table 1-4. -Table 1-2: Contents of the statm files (as of 2.6.30-rc7) +(for SMP CONFIG users) +For making accounting scalable, RSS related information are handled in +asynchronous manner and the vaule may not be very precise. To see a precise +snapshot of a moment, you can see /proc/<pid>/smaps file and scan page table. +It's slow but very precise. + +Table 1-2: Contents of the status files (as of 2.6.30-rc7) .............................................................................. Field Content Name filename of the executable @@ -213,6 +220,7 @@ Table 1-2: Contents of the statm files (as of 2.6.30-rc7) VmExe size of text segment VmLib size of shared library code VmPTE size of page table entries + VmSwap size of swap usage (the number of referred swapents) Threads number of threads SigQ number of signals queued/max. number for queue SigPnd bitmap of pending signals for the thread @@ -297,7 +305,7 @@ Table 1-4: Contents of the stat files (as of 2.6.30-rc7) cgtime guest time of the task children in jiffies .............................................................................. -The /proc/PID/map file containing the currently mapped memory regions and +The /proc/PID/maps file containing the currently mapped memory regions and their access permissions. The format is: @@ -308,7 +316,7 @@ address perms offset dev inode pathname 08049000-0804a000 rw-p 00001000 03:00 8312 /opt/test 0804a000-0806b000 rw-p 00000000 00:00 0 [heap] a7cb1000-a7cb2000 ---p 00000000 00:00 0 -a7cb2000-a7eb2000 rw-p 00000000 00:00 0 [threadstack:001ff4b4] +a7cb2000-a7eb2000 rw-p 00000000 00:00 0 a7eb2000-a7eb3000 ---p 00000000 00:00 0 a7eb3000-a7ed5000 rw-p 00000000 00:00 0 a7ed5000-a8008000 r-xp 00000000 03:00 4222 /lib/libc.so.6 @@ -344,7 +352,6 @@ is not associated with a file: [stack] = the stack of the main process [vdso] = the "virtual dynamic shared object", the kernel system call handler - [threadstack:xxxxxxxx] = the stack of the thread, xxxxxxxx is the stack size or if empty, the mapping is anonymous. @@ -430,6 +437,7 @@ Table 1-5: Kernel info in /proc modules List of loaded modules mounts Mounted filesystems net Networking info (see text) + pagetypeinfo Additional page allocator information (see text) (2.5) partitions Table of partitions known to the system pci Deprecated info of PCI bus (new way -> /proc/bus/pci/, decoupled by lspci (2.4) @@ -557,6 +565,10 @@ The default_smp_affinity mask applies to all non-active IRQs, which are the IRQs which have not yet been allocated/activated, and hence which lack a /proc/irq/[0-9]* directory. +The node file on an SMP system shows the node to which the device using the IRQ +reports itself as being attached. This hardware locality information does not +include information about any possible driver locality preference. + prof_cpu_mask specifies which CPUs are to be profiled by the system wide profiler. Default value is ffffffff (all cpus). @@ -584,7 +596,7 @@ Node 0, zone DMA 0 4 5 4 4 3 ... Node 0, zone Normal 1 0 0 1 101 8 ... Node 0, zone HighMem 2 0 0 1 1 0 ... -Memory fragmentation is a problem under some workloads, and buddyinfo is a +External fragmentation is a problem under some workloads, and buddyinfo is a useful tool for helping diagnose these problems. Buddyinfo will give you a clue as to how big an area you can safely allocate, or why a previous allocation failed. @@ -594,6 +606,48 @@ available. In this case, there are 0 chunks of 2^0*PAGE_SIZE available in ZONE_DMA, 4 chunks of 2^1*PAGE_SIZE in ZONE_DMA, 101 chunks of 2^4*PAGE_SIZE available in ZONE_NORMAL, etc... +More information relevant to external fragmentation can be found in +pagetypeinfo. + +> cat /proc/pagetypeinfo +Page block order: 9 +Pages per block: 512 + +Free pages count per migrate type at order 0 1 2 3 4 5 6 7 8 9 10 +Node 0, zone DMA, type Unmovable 0 0 0 1 1 1 1 1 1 1 0 +Node 0, zone DMA, type Reclaimable 0 0 0 0 0 0 0 0 0 0 0 +Node 0, zone DMA, type Movable 1 1 2 1 2 1 1 0 1 0 2 +Node 0, zone DMA, type Reserve 0 0 0 0 0 0 0 0 0 1 0 +Node 0, zone DMA, type Isolate 0 0 0 0 0 0 0 0 0 0 0 +Node 0, zone DMA32, type Unmovable 103 54 77 1 1 1 11 8 7 1 9 +Node 0, zone DMA32, type Reclaimable 0 0 2 1 0 0 0 0 1 0 0 +Node 0, zone DMA32, type Movable 169 152 113 91 77 54 39 13 6 1 452 +Node 0, zone DMA32, type Reserve 1 2 2 2 2 0 1 1 1 1 0 +Node 0, zone DMA32, type Isolate 0 0 0 0 0 0 0 0 0 0 0 + +Number of blocks type Unmovable Reclaimable Movable Reserve Isolate +Node 0, zone DMA 2 0 5 1 0 +Node 0, zone DMA32 41 6 967 2 0 + +Fragmentation avoidance in the kernel works by grouping pages of different +migrate types into the same contiguous regions of memory called page blocks. +A page block is typically the size of the default hugepage size e.g. 2MB on +X86-64. By keeping pages grouped based on their ability to move, the kernel +can reclaim pages within a page block to satisfy a high-order allocation. + +The pagetypinfo begins with information on the size of a page block. It +then gives the same type of information as buddyinfo except broken down +by migrate-type and finishes with details on how many page blocks of each +type exist. + +If min_free_kbytes has been tuned correctly (recommendations made by hugeadm +from libhugetlbfs http://sourceforge.net/projects/libhugetlbfs/), one can +make an estimate of the likely number of huge pages that can be allocated +at a given point in time. All the "Movable" blocks should be allocatable +unless memory has been mlock()'d. Some of the Reclaimable blocks should +also be allocatable although a lot of filesystem metadata may have to be +reclaimed to achieve this. + .............................................................................. meminfo: @@ -914,7 +968,7 @@ your system and how much traffic was routed over those devices: ...] 1375103 17405 0 0 0 0 0 0 ...] 1703981 5535 0 0 0 3 0 0 -In addition, each Channel Bond interface has it's own directory. For +In addition, each Channel Bond interface has its own directory. For example, the bond0 device will have a directory called /proc/net/bond0/. It will contain information that is specific to that bond, such as the current slaves of the bond, the link status of the slaves, and how @@ -1311,7 +1365,7 @@ been accounted as having caused 1MB of write. In other words: The number of bytes which this process caused to not happen, by truncating pagecache. A task can cause "negative" IO too. If this task truncates some dirty pagecache, some IO which another task has been accounted -for (in it's write_bytes) will not be happening. We _could_ just subtract that +for (in its write_bytes) will not be happening. We _could_ just subtract that from the truncating task's write_bytes, but there is information loss in doing that. diff --git a/Documentation/filesystems/sharedsubtree.txt b/Documentation/filesystems/sharedsubtree.txt index 23a181074f94..fc0e39af43c3 100644 --- a/Documentation/filesystems/sharedsubtree.txt +++ b/Documentation/filesystems/sharedsubtree.txt @@ -837,6 +837,9 @@ replicas continue to be exactly same. individual lists does not affect propagation or the way propagation tree is modified by operations. + All vfsmounts in a peer group have the same ->mnt_master. If it is + non-NULL, they form a contiguous (ordered) segment of slave list. + A example propagation tree looks as shown in the figure below. [ NOTE: Though it looks like a forest, if we consider all the shared mounts as a conceptual entity called 'pnode', it becomes a tree] @@ -874,8 +877,19 @@ replicas continue to be exactly same. NOTE: The propagation tree is orthogonal to the mount tree. +8B Locking: + + ->mnt_share, ->mnt_slave, ->mnt_slave_list, ->mnt_master are protected + by namespace_sem (exclusive for modifications, shared for reading). + + Normally we have ->mnt_flags modifications serialized by vfsmount_lock. + There are two exceptions: do_add_mount() and clone_mnt(). + The former modifies a vfsmount that has not been visible in any shared + data structures yet. + The latter holds namespace_sem and the only references to vfsmount + are in lists that can't be traversed without namespace_sem. -8B Algorithm: +8C Algorithm: The crux of the implementation resides in rbind/move operation. diff --git a/Documentation/filesystems/smbfs.txt b/Documentation/filesystems/smbfs.txt index f673ef0de0f7..194fb0decd2c 100644 --- a/Documentation/filesystems/smbfs.txt +++ b/Documentation/filesystems/smbfs.txt @@ -3,6 +3,6 @@ protocol used by Windows for Workgroups, Windows 95 and Windows NT. Smbfs was inspired by Samba, the program written by Andrew Tridgell that turns any Unix host into a file server for DOS or Windows clients. -Smbfs is a SMB client, but uses parts of samba for it's operation. For +Smbfs is a SMB client, but uses parts of samba for its operation. For more info on samba, including documentation, please go to http://www.samba.org/ and then on to your nearest mirror. diff --git a/Documentation/filesystems/squashfs.txt b/Documentation/filesystems/squashfs.txt index b324c033035a..203f7202cc9e 100644 --- a/Documentation/filesystems/squashfs.txt +++ b/Documentation/filesystems/squashfs.txt @@ -38,7 +38,8 @@ Hard link support: yes no Real inode numbers: yes no 32-bit uids/gids: yes no File creation time: yes no -Xattr and ACL support: no no +Xattr support: yes no +ACL support: no no Squashfs compresses data, inodes and directories. In addition, inode and directory data are highly compacted, and packed on byte boundaries. Each @@ -58,7 +59,7 @@ obtained from this site also. 3. SQUASHFS FILESYSTEM DESIGN ----------------------------- -A squashfs filesystem consists of seven parts, packed together on a byte +A squashfs filesystem consists of a maximum of eight parts, packed together on a byte alignment: --------------- @@ -80,6 +81,9 @@ alignment: |---------------| | uid/gid | | lookup table | + |---------------| + | xattr | + | table | --------------- Compressed data blocks are written to the filesystem as files are read from @@ -192,6 +196,26 @@ This table is stored compressed into metadata blocks. A second index table is used to locate these. This second index table for speed of access (and because it is small) is read at mount time and cached in memory. +3.7 Xattr table +--------------- + +The xattr table contains extended attributes for each inode. The xattrs +for each inode are stored in a list, each list entry containing a type, +name and value field. The type field encodes the xattr prefix +("user.", "trusted." etc) and it also encodes how the name/value fields +should be interpreted. Currently the type indicates whether the value +is stored inline (in which case the value field contains the xattr value), +or if it is stored out of line (in which case the value field stores a +reference to where the actual value is stored). This allows large values +to be stored out of line improving scanning and lookup performance and it +also allows values to be de-duplicated, the value being stored once, and +all other occurences holding an out of line reference to that value. + +The xattr lists are packed into compressed 8K metadata blocks. +To reduce overhead in inodes, rather than storing the on-disk +location of the xattr list inside each inode, a 32-bit xattr id +is stored. This xattr id is mapped into the location of the xattr +list using a second xattr id lookup table. 4. TODOS AND OUTSTANDING ISSUES ------------------------------- @@ -199,9 +223,7 @@ it is small) is read at mount time and cached in memory. 4.1 Todo list ------------- -Implement Xattr and ACL support. The Squashfs 4.0 filesystem layout has hooks -for these but the code has not been written. Once the code has been written -the existing layout should not require modification. +Implement ACL support. 4.2 Squashfs internal cache --------------------------- diff --git a/Documentation/filesystems/sysfs-tagging.txt b/Documentation/filesystems/sysfs-tagging.txt new file mode 100644 index 000000000000..caaaf1266d8f --- /dev/null +++ b/Documentation/filesystems/sysfs-tagging.txt @@ -0,0 +1,42 @@ +Sysfs tagging +------------- + +(Taken almost verbatim from Eric Biederman's netns tagging patch +commit msg) + +The problem. Network devices show up in sysfs and with the network +namespace active multiple devices with the same name can show up in +the same directory, ouch! + +To avoid that problem and allow existing applications in network +namespaces to see the same interface that is currently presented in +sysfs, sysfs now has tagging directory support. + +By using the network namespace pointers as tags to separate out the +the sysfs directory entries we ensure that we don't have conflicts +in the directories and applications only see a limited set of +the network devices. + +Each sysfs directory entry may be tagged with zero or one +namespaces. A sysfs_dirent is augmented with a void *s_ns. If a +directory entry is tagged, then sysfs_dirent->s_flags will have a +flag between KOBJ_NS_TYPE_NONE and KOBJ_NS_TYPES, and s_ns will +point to the namespace to which it belongs. + +Each sysfs superblock's sysfs_super_info contains an array void +*ns[KOBJ_NS_TYPES]. When a a task in a tagging namespace +kobj_nstype first mounts sysfs, a new superblock is created. It +will be differentiated from other sysfs mounts by having its +s_fs_info->ns[kobj_nstype] set to the new namespace. Note that +through bind mounting and mounts propagation, a task can easily view +the contents of other namespaces' sysfs mounts. Therefore, when a +namespace exits, it will call kobj_ns_exit() to invalidate any +sysfs_dirent->s_ns pointers pointing to it. + +Users of this interface: +- define a type in the kobj_ns_type enumeration. +- call kobj_ns_type_register() with its kobj_ns_type_operations which has + - current_ns() which returns current's namespace + - netlink_ns() which returns a socket's namespace + - initial_ns() which returns the initial namesapce +- call kobj_ns_exit() when an individual tag is no longer valid diff --git a/Documentation/filesystems/tmpfs.txt b/Documentation/filesystems/tmpfs.txt index 3015da0c6b2a..98ef55124158 100644 --- a/Documentation/filesystems/tmpfs.txt +++ b/Documentation/filesystems/tmpfs.txt @@ -82,21 +82,31 @@ tmpfs has a mount option to set the NUMA memory allocation policy for all files in that instance (if CONFIG_NUMA is enabled) - which can be adjusted on the fly via 'mount -o remount ...' -mpol=default prefers to allocate memory from the local node +mpol=default use the process allocation policy + (see set_mempolicy(2)) mpol=prefer:Node prefers to allocate memory from the given Node mpol=bind:NodeList allocates memory only from nodes in NodeList mpol=interleave prefers to allocate from each node in turn mpol=interleave:NodeList allocates from each node of NodeList in turn +mpol=local prefers to allocate memory from the local node NodeList format is a comma-separated list of decimal numbers and ranges, a range being two hyphen-separated decimal numbers, the smallest and largest node numbers in the range. For example, mpol=bind:0-3,5,7,9-15 +A memory policy with a valid NodeList will be saved, as specified, for +use at file creation time. When a task allocates a file in the file +system, the mount option memory policy will be applied with a NodeList, +if any, modified by the calling task's cpuset constraints +[See Documentation/cgroups/cpusets.txt] and any optional flags, listed +below. If the resulting NodeLists is the empty set, the effective memory +policy for the file will revert to "default" policy. + NUMA memory allocation policies have optional flags that can be used in conjunction with their modes. These optional flags can be specified when tmpfs is mounted by appending them to the mode before the NodeList. See Documentation/vm/numa_memory_policy.txt for a list of all available -memory allocation policy mode flags. +memory allocation policy mode flags and their effect on memory policy. =static is equivalent to MPOL_F_STATIC_NODES =relative is equivalent to MPOL_F_RELATIVE_NODES @@ -134,3 +144,5 @@ Author: Christoph Rohland <cr@sap.com>, 1.12.01 Updated: Hugh Dickins, 4 June 2007 +Updated: + KOSAKI Motohiro, 16 Mar 2010 diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt index 3de2f32edd90..94677e7dcb13 100644 --- a/Documentation/filesystems/vfs.txt +++ b/Documentation/filesystems/vfs.txt @@ -72,7 +72,7 @@ structure (this is the kernel-side implementation of file descriptors). The freshly allocated file structure is initialized with a pointer to the dentry and a set of file operation member functions. These are taken from the inode data. The open() file method is then -called so the specific filesystem implementation can do it's work. You +called so the specific filesystem implementation can do its work. You can see that this is another switch performed by the VFS. The file structure is placed into the file descriptor table for the process. @@ -401,11 +401,16 @@ otherwise noted. started might not be in the page cache at the end of the walk). - truncate: called by the VFS to change the size of a file. The + truncate: Deprecated. This will not be called if ->setsize is defined. + Called by the VFS to change the size of a file. The i_size field of the inode is set to the desired size by the VFS before this method is called. This method is called by the truncate(2) system call and related functionality. + Note: ->truncate and vmtruncate are deprecated. Do not add new + instances/calls of these. Filesystems should be converted to do their + truncate sequence via ->setattr(). + permission: called by the VFS to check for access rights on a POSIX-like filesystem. @@ -729,7 +734,7 @@ struct file_operations { int (*open) (struct inode *, struct file *); int (*flush) (struct file *); int (*release) (struct inode *, struct file *); - int (*fsync) (struct file *, struct dentry *, int datasync); + int (*fsync) (struct file *, int datasync); int (*aio_fsync) (struct kiocb *, int datasync); int (*fasync) (int, struct file *, int); int (*lock) (struct file *, int, struct file_lock *); diff --git a/Documentation/filesystems/xfs-delayed-logging-design.txt b/Documentation/filesystems/xfs-delayed-logging-design.txt new file mode 100644 index 000000000000..96d0df28bed3 --- /dev/null +++ b/Documentation/filesystems/xfs-delayed-logging-design.txt @@ -0,0 +1,811 @@ +XFS Delayed Logging Design +-------------------------- + +Introduction to Re-logging in XFS +--------------------------------- + +XFS logging is a combination of logical and physical logging. Some objects, +such as inodes and dquots, are logged in logical format where the details +logged are made up of the changes to in-core structures rather than on-disk +structures. Other objects - typically buffers - have their physical changes +logged. The reason for these differences is to reduce the amount of log space +required for objects that are frequently logged. Some parts of inodes are more +frequently logged than others, and inodes are typically more frequently logged +than any other object (except maybe the superblock buffer) so keeping the +amount of metadata logged low is of prime importance. + +The reason that this is such a concern is that XFS allows multiple separate +modifications to a single object to be carried in the log at any given time. +This allows the log to avoid needing to flush each change to disk before +recording a new change to the object. XFS does this via a method called +"re-logging". Conceptually, this is quite simple - all it requires is that any +new change to the object is recorded with a *new copy* of all the existing +changes in the new transaction that is written to the log. + +That is, if we have a sequence of changes A through to F, and the object was +written to disk after change D, we would see in the log the following series +of transactions, their contents and the log sequence number (LSN) of the +transaction: + + Transaction Contents LSN + A A X + B A+B X+n + C A+B+C X+n+m + D A+B+C+D X+n+m+o + <object written to disk> + E E Y (> X+n+m+o) + F E+F Yٍ+p + +In other words, each time an object is relogged, the new transaction contains +the aggregation of all the previous changes currently held only in the log. + +This relogging technique also allows objects to be moved forward in the log so +that an object being relogged does not prevent the tail of the log from ever +moving forward. This can be seen in the table above by the changing +(increasing) LSN of each subsquent transaction - the LSN is effectively a +direct encoding of the location in the log of the transaction. + +This relogging is also used to implement long-running, multiple-commit +transactions. These transaction are known as rolling transactions, and require +a special log reservation known as a permanent transaction reservation. A +typical example of a rolling transaction is the removal of extents from an +inode which can only be done at a rate of two extents per transaction because +of reservation size limitations. Hence a rolling extent removal transaction +keeps relogging the inode and btree buffers as they get modified in each +removal operation. This keeps them moving forward in the log as the operation +progresses, ensuring that current operation never gets blocked by itself if the +log wraps around. + +Hence it can be seen that the relogging operation is fundamental to the correct +working of the XFS journalling subsystem. From the above description, most +people should be able to see why the XFS metadata operations writes so much to +the log - repeated operations to the same objects write the same changes to +the log over and over again. Worse is the fact that objects tend to get +dirtier as they get relogged, so each subsequent transaction is writing more +metadata into the log. + +Another feature of the XFS transaction subsystem is that most transactions are +asynchronous. That is, they don't commit to disk until either a log buffer is +filled (a log buffer can hold multiple transactions) or a synchronous operation +forces the log buffers holding the transactions to disk. This means that XFS is +doing aggregation of transactions in memory - batching them, if you like - to +minimise the impact of the log IO on transaction throughput. + +The limitation on asynchronous transaction throughput is the number and size of +log buffers made available by the log manager. By default there are 8 log +buffers available and the size of each is 32kB - the size can be increased up +to 256kB by use of a mount option. + +Effectively, this gives us the maximum bound of outstanding metadata changes +that can be made to the filesystem at any point in time - if all the log +buffers are full and under IO, then no more transactions can be committed until +the current batch completes. It is now common for a single current CPU core to +be to able to issue enough transactions to keep the log buffers full and under +IO permanently. Hence the XFS journalling subsystem can be considered to be IO +bound. + +Delayed Logging: Concepts +------------------------- + +The key thing to note about the asynchronous logging combined with the +relogging technique XFS uses is that we can be relogging changed objects +multiple times before they are committed to disk in the log buffers. If we +return to the previous relogging example, it is entirely possible that +transactions A through D are committed to disk in the same log buffer. + +That is, a single log buffer may contain multiple copies of the same object, +but only one of those copies needs to be there - the last one "D", as it +contains all the changes from the previous changes. In other words, we have one +necessary copy in the log buffer, and three stale copies that are simply +wasting space. When we are doing repeated operations on the same set of +objects, these "stale objects" can be over 90% of the space used in the log +buffers. It is clear that reducing the number of stale objects written to the +log would greatly reduce the amount of metadata we write to the log, and this +is the fundamental goal of delayed logging. + +From a conceptual point of view, XFS is already doing relogging in memory (where +memory == log buffer), only it is doing it extremely inefficiently. It is using +logical to physical formatting to do the relogging because there is no +infrastructure to keep track of logical changes in memory prior to physically +formatting the changes in a transaction to the log buffer. Hence we cannot avoid +accumulating stale objects in the log buffers. + +Delayed logging is the name we've given to keeping and tracking transactional +changes to objects in memory outside the log buffer infrastructure. Because of +the relogging concept fundamental to the XFS journalling subsystem, this is +actually relatively easy to do - all the changes to logged items are already +tracked in the current infrastructure. The big problem is how to accumulate +them and get them to the log in a consistent, recoverable manner. +Describing the problems and how they have been solved is the focus of this +document. + +One of the key changes that delayed logging makes to the operation of the +journalling subsystem is that it disassociates the amount of outstanding +metadata changes from the size and number of log buffers available. In other +words, instead of there only being a maximum of 2MB of transaction changes not +written to the log at any point in time, there may be a much greater amount +being accumulated in memory. Hence the potential for loss of metadata on a +crash is much greater than for the existing logging mechanism. + +It should be noted that this does not change the guarantee that log recovery +will result in a consistent filesystem. What it does mean is that as far as the +recovered filesystem is concerned, there may be many thousands of transactions +that simply did not occur as a result of the crash. This makes it even more +important that applications that care about their data use fsync() where they +need to ensure application level data integrity is maintained. + +It should be noted that delayed logging is not an innovative new concept that +warrants rigorous proofs to determine whether it is correct or not. The method +of accumulating changes in memory for some period before writing them to the +log is used effectively in many filesystems including ext3 and ext4. Hence +no time is spent in this document trying to convince the reader that the +concept is sound. Instead it is simply considered a "solved problem" and as +such implementing it in XFS is purely an exercise in software engineering. + +The fundamental requirements for delayed logging in XFS are simple: + + 1. Reduce the amount of metadata written to the log by at least + an order of magnitude. + 2. Supply sufficient statistics to validate Requirement #1. + 3. Supply sufficient new tracing infrastructure to be able to debug + problems with the new code. + 4. No on-disk format change (metadata or log format). + 5. Enable and disable with a mount option. + 6. No performance regressions for synchronous transaction workloads. + +Delayed Logging: Design +----------------------- + +Storing Changes + +The problem with accumulating changes at a logical level (i.e. just using the +existing log item dirty region tracking) is that when it comes to writing the +changes to the log buffers, we need to ensure that the object we are formatting +is not changing while we do this. This requires locking the object to prevent +concurrent modification. Hence flushing the logical changes to the log would +require us to lock every object, format them, and then unlock them again. + +This introduces lots of scope for deadlocks with transactions that are already +running. For example, a transaction has object A locked and modified, but needs +the delayed logging tracking lock to commit the transaction. However, the +flushing thread has the delayed logging tracking lock already held, and is +trying to get the lock on object A to flush it to the log buffer. This appears +to be an unsolvable deadlock condition, and it was solving this problem that +was the barrier to implementing delayed logging for so long. + +The solution is relatively simple - it just took a long time to recognise it. +Put simply, the current logging code formats the changes to each item into an +vector array that points to the changed regions in the item. The log write code +simply copies the memory these vectors point to into the log buffer during +transaction commit while the item is locked in the transaction. Instead of +using the log buffer as the destination of the formatting code, we can use an +allocated memory buffer big enough to fit the formatted vector. + +If we then copy the vector into the memory buffer and rewrite the vector to +point to the memory buffer rather than the object itself, we now have a copy of +the changes in a format that is compatible with the log buffer writing code. +that does not require us to lock the item to access. This formatting and +rewriting can all be done while the object is locked during transaction commit, +resulting in a vector that is transactionally consistent and can be accessed +without needing to lock the owning item. + +Hence we avoid the need to lock items when we need to flush outstanding +asynchronous transactions to the log. The differences between the existing +formatting method and the delayed logging formatting can be seen in the +diagram below. + +Current format log vector: + +Object +---------------------------------------------+ +Vector 1 +----+ +Vector 2 +----+ +Vector 3 +----------+ + +After formatting: + +Log Buffer +-V1-+-V2-+----V3----+ + +Delayed logging vector: + +Object +---------------------------------------------+ +Vector 1 +----+ +Vector 2 +----+ +Vector 3 +----------+ + +After formatting: + +Memory Buffer +-V1-+-V2-+----V3----+ +Vector 1 +----+ +Vector 2 +----+ +Vector 3 +----------+ + +The memory buffer and associated vector need to be passed as a single object, +but still need to be associated with the parent object so if the object is +relogged we can replace the current memory buffer with a new memory buffer that +contains the latest changes. + +The reason for keeping the vector around after we've formatted the memory +buffer is to support splitting vectors across log buffer boundaries correctly. +If we don't keep the vector around, we do not know where the region boundaries +are in the item, so we'd need a new encapsulation method for regions in the log +buffer writing (i.e. double encapsulation). This would be an on-disk format +change and as such is not desirable. It also means we'd have to write the log +region headers in the formatting stage, which is problematic as there is per +region state that needs to be placed into the headers during the log write. + +Hence we need to keep the vector, but by attaching the memory buffer to it and +rewriting the vector addresses to point at the memory buffer we end up with a +self-describing object that can be passed to the log buffer write code to be +handled in exactly the same manner as the existing log vectors are handled. +Hence we avoid needing a new on-disk format to handle items that have been +relogged in memory. + + +Tracking Changes + +Now that we can record transactional changes in memory in a form that allows +them to be used without limitations, we need to be able to track and accumulate +them so that they can be written to the log at some later point in time. The +log item is the natural place to store this vector and buffer, and also makes sense +to be the object that is used to track committed objects as it will always +exist once the object has been included in a transaction. + +The log item is already used to track the log items that have been written to +the log but not yet written to disk. Such log items are considered "active" +and as such are stored in the Active Item List (AIL) which is a LSN-ordered +double linked list. Items are inserted into this list during log buffer IO +completion, after which they are unpinned and can be written to disk. An object +that is in the AIL can be relogged, which causes the object to be pinned again +and then moved forward in the AIL when the log buffer IO completes for that +transaction. + +Essentially, this shows that an item that is in the AIL can still be modified +and relogged, so any tracking must be separate to the AIL infrastructure. As +such, we cannot reuse the AIL list pointers for tracking committed items, nor +can we store state in any field that is protected by the AIL lock. Hence the +committed item tracking needs it's own locks, lists and state fields in the log +item. + +Similar to the AIL, tracking of committed items is done through a new list +called the Committed Item List (CIL). The list tracks log items that have been +committed and have formatted memory buffers attached to them. It tracks objects +in transaction commit order, so when an object is relogged it is removed from +it's place in the list and re-inserted at the tail. This is entirely arbitrary +and done to make it easy for debugging - the last items in the list are the +ones that are most recently modified. Ordering of the CIL is not necessary for +transactional integrity (as discussed in the next section) so the ordering is +done for convenience/sanity of the developers. + + +Delayed Logging: Checkpoints + +When we have a log synchronisation event, commonly known as a "log force", +all the items in the CIL must be written into the log via the log buffers. +We need to write these items in the order that they exist in the CIL, and they +need to be written as an atomic transaction. The need for all the objects to be +written as an atomic transaction comes from the requirements of relogging and +log replay - all the changes in all the objects in a given transaction must +either be completely replayed during log recovery, or not replayed at all. If +a transaction is not replayed because it is not complete in the log, then +no later transactions should be replayed, either. + +To fulfill this requirement, we need to write the entire CIL in a single log +transaction. Fortunately, the XFS log code has no fixed limit on the size of a +transaction, nor does the log replay code. The only fundamental limit is that +the transaction cannot be larger than just under half the size of the log. The +reason for this limit is that to find the head and tail of the log, there must +be at least one complete transaction in the log at any given time. If a +transaction is larger than half the log, then there is the possibility that a +crash during the write of a such a transaction could partially overwrite the +only complete previous transaction in the log. This will result in a recovery +failure and an inconsistent filesystem and hence we must enforce the maximum +size of a checkpoint to be slightly less than a half the log. + +Apart from this size requirement, a checkpoint transaction looks no different +to any other transaction - it contains a transaction header, a series of +formatted log items and a commit record at the tail. From a recovery +perspective, the checkpoint transaction is also no different - just a lot +bigger with a lot more items in it. The worst case effect of this is that we +might need to tune the recovery transaction object hash size. + +Because the checkpoint is just another transaction and all the changes to log +items are stored as log vectors, we can use the existing log buffer writing +code to write the changes into the log. To do this efficiently, we need to +minimise the time we hold the CIL locked while writing the checkpoint +transaction. The current log write code enables us to do this easily with the +way it separates the writing of the transaction contents (the log vectors) from +the transaction commit record, but tracking this requires us to have a +per-checkpoint context that travels through the log write process through to +checkpoint completion. + +Hence a checkpoint has a context that tracks the state of the current +checkpoint from initiation to checkpoint completion. A new context is initiated +at the same time a checkpoint transaction is started. That is, when we remove +all the current items from the CIL during a checkpoint operation, we move all +those changes into the current checkpoint context. We then initialise a new +context and attach that to the CIL for aggregation of new transactions. + +This allows us to unlock the CIL immediately after transfer of all the +committed items and effectively allow new transactions to be issued while we +are formatting the checkpoint into the log. It also allows concurrent +checkpoints to be written into the log buffers in the case of log force heavy +workloads, just like the existing transaction commit code does. This, however, +requires that we strictly order the commit records in the log so that +checkpoint sequence order is maintained during log replay. + +To ensure that we can be writing an item into a checkpoint transaction at +the same time another transaction modifies the item and inserts the log item +into the new CIL, then checkpoint transaction commit code cannot use log items +to store the list of log vectors that need to be written into the transaction. +Hence log vectors need to be able to be chained together to allow them to be +detatched from the log items. That is, when the CIL is flushed the memory +buffer and log vector attached to each log item needs to be attached to the +checkpoint context so that the log item can be released. In diagrammatic form, +the CIL would look like this before the flush: + + CIL Head + | + V + Log Item <-> log vector 1 -> memory buffer + | -> vector array + V + Log Item <-> log vector 2 -> memory buffer + | -> vector array + V + ...... + | + V + Log Item <-> log vector N-1 -> memory buffer + | -> vector array + V + Log Item <-> log vector N -> memory buffer + -> vector array + +And after the flush the CIL head is empty, and the checkpoint context log +vector list would look like: + + Checkpoint Context + | + V + log vector 1 -> memory buffer + | -> vector array + | -> Log Item + V + log vector 2 -> memory buffer + | -> vector array + | -> Log Item + V + ...... + | + V + log vector N-1 -> memory buffer + | -> vector array + | -> Log Item + V + log vector N -> memory buffer + -> vector array + -> Log Item + +Once this transfer is done, the CIL can be unlocked and new transactions can +start, while the checkpoint flush code works over the log vector chain to +commit the checkpoint. + +Once the checkpoint is written into the log buffers, the checkpoint context is +attached to the log buffer that the commit record was written to along with a +completion callback. Log IO completion will call that callback, which can then +run transaction committed processing for the log items (i.e. insert into AIL +and unpin) in the log vector chain and then free the log vector chain and +checkpoint context. + +Discussion Point: I am uncertain as to whether the log item is the most +efficient way to track vectors, even though it seems like the natural way to do +it. The fact that we walk the log items (in the CIL) just to chain the log +vectors and break the link between the log item and the log vector means that +we take a cache line hit for the log item list modification, then another for +the log vector chaining. If we track by the log vectors, then we only need to +break the link between the log item and the log vector, which means we should +dirty only the log item cachelines. Normally I wouldn't be concerned about one +vs two dirty cachelines except for the fact I've seen upwards of 80,000 log +vectors in one checkpoint transaction. I'd guess this is a "measure and +compare" situation that can be done after a working and reviewed implementation +is in the dev tree.... + +Delayed Logging: Checkpoint Sequencing + +One of the key aspects of the XFS transaction subsystem is that it tags +committed transactions with the log sequence number of the transaction commit. +This allows transactions to be issued asynchronously even though there may be +future operations that cannot be completed until that transaction is fully +committed to the log. In the rare case that a dependent operation occurs (e.g. +re-using a freed metadata extent for a data extent), a special, optimised log +force can be issued to force the dependent transaction to disk immediately. + +To do this, transactions need to record the LSN of the commit record of the +transaction. This LSN comes directly from the log buffer the transaction is +written into. While this works just fine for the existing transaction +mechanism, it does not work for delayed logging because transactions are not +written directly into the log buffers. Hence some other method of sequencing +transactions is required. + +As discussed in the checkpoint section, delayed logging uses per-checkpoint +contexts, and as such it is simple to assign a sequence number to each +checkpoint. Because the switching of checkpoint contexts must be done +atomically, it is simple to ensure that each new context has a monotonically +increasing sequence number assigned to it without the need for an external +atomic counter - we can just take the current context sequence number and add +one to it for the new context. + +Then, instead of assigning a log buffer LSN to the transaction commit LSN +during the commit, we can assign the current checkpoint sequence. This allows +operations that track transactions that have not yet completed know what +checkpoint sequence needs to be committed before they can continue. As a +result, the code that forces the log to a specific LSN now needs to ensure that +the log forces to a specific checkpoint. + +To ensure that we can do this, we need to track all the checkpoint contexts +that are currently committing to the log. When we flush a checkpoint, the +context gets added to a "committing" list which can be searched. When a +checkpoint commit completes, it is removed from the committing list. Because +the checkpoint context records the LSN of the commit record for the checkpoint, +we can also wait on the log buffer that contains the commit record, thereby +using the existing log force mechanisms to execute synchronous forces. + +It should be noted that the synchronous forces may need to be extended with +mitigation algorithms similar to the current log buffer code to allow +aggregation of multiple synchronous transactions if there are already +synchronous transactions being flushed. Investigation of the performance of the +current design is needed before making any decisions here. + +The main concern with log forces is to ensure that all the previous checkpoints +are also committed to disk before the one we need to wait for. Therefore we +need to check that all the prior contexts in the committing list are also +complete before waiting on the one we need to complete. We do this +synchronisation in the log force code so that we don't need to wait anywhere +else for such serialisation - it only matters when we do a log force. + +The only remaining complexity is that a log force now also has to handle the +case where the forcing sequence number is the same as the current context. That +is, we need to flush the CIL and potentially wait for it to complete. This is a +simple addition to the existing log forcing code to check the sequence numbers +and push if required. Indeed, placing the current sequence checkpoint flush in +the log force code enables the current mechanism for issuing synchronous +transactions to remain untouched (i.e. commit an asynchronous transaction, then +force the log at the LSN of that transaction) and so the higher level code +behaves the same regardless of whether delayed logging is being used or not. + +Delayed Logging: Checkpoint Log Space Accounting + +The big issue for a checkpoint transaction is the log space reservation for the +transaction. We don't know how big a checkpoint transaction is going to be +ahead of time, nor how many log buffers it will take to write out, nor the +number of split log vector regions are going to be used. We can track the +amount of log space required as we add items to the commit item list, but we +still need to reserve the space in the log for the checkpoint. + +A typical transaction reserves enough space in the log for the worst case space +usage of the transaction. The reservation accounts for log record headers, +transaction and region headers, headers for split regions, buffer tail padding, +etc. as well as the actual space for all the changed metadata in the +transaction. While some of this is fixed overhead, much of it is dependent on +the size of the transaction and the number of regions being logged (the number +of log vectors in the transaction). + +An example of the differences would be logging directory changes versus logging +inode changes. If you modify lots of inode cores (e.g. chmod -R g+w *), then +there are lots of transactions that only contain an inode core and an inode log +format structure. That is, two vectors totaling roughly 150 bytes. If we modify +10,000 inodes, we have about 1.5MB of metadata to write in 20,000 vectors. Each +vector is 12 bytes, so the total to be logged is approximately 1.75MB. In +comparison, if we are logging full directory buffers, they are typically 4KB +each, so we in 1.5MB of directory buffers we'd have roughly 400 buffers and a +buffer format structure for each buffer - roughly 800 vectors or 1.51MB total +space. From this, it should be obvious that a static log space reservation is +not particularly flexible and is difficult to select the "optimal value" for +all workloads. + +Further, if we are going to use a static reservation, which bit of the entire +reservation does it cover? We account for space used by the transaction +reservation by tracking the space currently used by the object in the CIL and +then calculating the increase or decrease in space used as the object is +relogged. This allows for a checkpoint reservation to only have to account for +log buffer metadata used such as log header records. + +However, even using a static reservation for just the log metadata is +problematic. Typically log record headers use at least 16KB of log space per +1MB of log space consumed (512 bytes per 32k) and the reservation needs to be +large enough to handle arbitrary sized checkpoint transactions. This +reservation needs to be made before the checkpoint is started, and we need to +be able to reserve the space without sleeping. For a 8MB checkpoint, we need a +reservation of around 150KB, which is a non-trivial amount of space. + +A static reservation needs to manipulate the log grant counters - we can take a +permanent reservation on the space, but we still need to make sure we refresh +the write reservation (the actual space available to the transaction) after +every checkpoint transaction completion. Unfortunately, if this space is not +available when required, then the regrant code will sleep waiting for it. + +The problem with this is that it can lead to deadlocks as we may need to commit +checkpoints to be able to free up log space (refer back to the description of +rolling transactions for an example of this). Hence we *must* always have +space available in the log if we are to use static reservations, and that is +very difficult and complex to arrange. It is possible to do, but there is a +simpler way. + +The simpler way of doing this is tracking the entire log space used by the +items in the CIL and using this to dynamically calculate the amount of log +space required by the log metadata. If this log metadata space changes as a +result of a transaction commit inserting a new memory buffer into the CIL, then +the difference in space required is removed from the transaction that causes +the change. Transactions at this level will *always* have enough space +available in their reservation for this as they have already reserved the +maximal amount of log metadata space they require, and such a delta reservation +will always be less than or equal to the maximal amount in the reservation. + +Hence we can grow the checkpoint transaction reservation dynamically as items +are added to the CIL and avoid the need for reserving and regranting log space +up front. This avoids deadlocks and removes a blocking point from the +checkpoint flush code. + +As mentioned early, transactions can't grow to more than half the size of the +log. Hence as part of the reservation growing, we need to also check the size +of the reservation against the maximum allowed transaction size. If we reach +the maximum threshold, we need to push the CIL to the log. This is effectively +a "background flush" and is done on demand. This is identical to +a CIL push triggered by a log force, only that there is no waiting for the +checkpoint commit to complete. This background push is checked and executed by +transaction commit code. + +If the transaction subsystem goes idle while we still have items in the CIL, +they will be flushed by the periodic log force issued by the xfssyncd. This log +force will push the CIL to disk, and if the transaction subsystem stays idle, +allow the idle log to be covered (effectively marked clean) in exactly the same +manner that is done for the existing logging method. A discussion point is +whether this log force needs to be done more frequently than the current rate +which is once every 30s. + + +Delayed Logging: Log Item Pinning + +Currently log items are pinned during transaction commit while the items are +still locked. This happens just after the items are formatted, though it could +be done any time before the items are unlocked. The result of this mechanism is +that items get pinned once for every transaction that is committed to the log +buffers. Hence items that are relogged in the log buffers will have a pin count +for every outstanding transaction they were dirtied in. When each of these +transactions is completed, they will unpin the item once. As a result, the item +only becomes unpinned when all the transactions complete and there are no +pending transactions. Thus the pinning and unpinning of a log item is symmetric +as there is a 1:1 relationship with transaction commit and log item completion. + +For delayed logging, however, we have an assymetric transaction commit to +completion relationship. Every time an object is relogged in the CIL it goes +through the commit process without a corresponding completion being registered. +That is, we now have a many-to-one relationship between transaction commit and +log item completion. The result of this is that pinning and unpinning of the +log items becomes unbalanced if we retain the "pin on transaction commit, unpin +on transaction completion" model. + +To keep pin/unpin symmetry, the algorithm needs to change to a "pin on +insertion into the CIL, unpin on checkpoint completion". In other words, the +pinning and unpinning becomes symmetric around a checkpoint context. We have to +pin the object the first time it is inserted into the CIL - if it is already in +the CIL during a transaction commit, then we do not pin it again. Because there +can be multiple outstanding checkpoint contexts, we can still see elevated pin +counts, but as each checkpoint completes the pin count will retain the correct +value according to it's context. + +Just to make matters more slightly more complex, this checkpoint level context +for the pin count means that the pinning of an item must take place under the +CIL commit/flush lock. If we pin the object outside this lock, we cannot +guarantee which context the pin count is associated with. This is because of +the fact pinning the item is dependent on whether the item is present in the +current CIL or not. If we don't pin the CIL first before we check and pin the +object, we have a race with CIL being flushed between the check and the pin +(or not pinning, as the case may be). Hence we must hold the CIL flush/commit +lock to guarantee that we pin the items correctly. + +Delayed Logging: Concurrent Scalability + +A fundamental requirement for the CIL is that accesses through transaction +commits must scale to many concurrent commits. The current transaction commit +code does not break down even when there are transactions coming from 2048 +processors at once. The current transaction code does not go any faster than if +there was only one CPU using it, but it does not slow down either. + +As a result, the delayed logging transaction commit code needs to be designed +for concurrency from the ground up. It is obvious that there are serialisation +points in the design - the three important ones are: + + 1. Locking out new transaction commits while flushing the CIL + 2. Adding items to the CIL and updating item space accounting + 3. Checkpoint commit ordering + +Looking at the transaction commit and CIL flushing interactions, it is clear +that we have a many-to-one interaction here. That is, the only restriction on +the number of concurrent transactions that can be trying to commit at once is +the amount of space available in the log for their reservations. The practical +limit here is in the order of several hundred concurrent transactions for a +128MB log, which means that it is generally one per CPU in a machine. + +The amount of time a transaction commit needs to hold out a flush is a +relatively long period of time - the pinning of log items needs to be done +while we are holding out a CIL flush, so at the moment that means it is held +across the formatting of the objects into memory buffers (i.e. while memcpy()s +are in progress). Ultimately a two pass algorithm where the formatting is done +separately to the pinning of objects could be used to reduce the hold time of +the transaction commit side. + +Because of the number of potential transaction commit side holders, the lock +really needs to be a sleeping lock - if the CIL flush takes the lock, we do not +want every other CPU in the machine spinning on the CIL lock. Given that +flushing the CIL could involve walking a list of tens of thousands of log +items, it will get held for a significant time and so spin contention is a +significant concern. Preventing lots of CPUs spinning doing nothing is the +main reason for choosing a sleeping lock even though nothing in either the +transaction commit or CIL flush side sleeps with the lock held. + +It should also be noted that CIL flushing is also a relatively rare operation +compared to transaction commit for asynchronous transaction workloads - only +time will tell if using a read-write semaphore for exclusion will limit +transaction commit concurrency due to cache line bouncing of the lock on the +read side. + +The second serialisation point is on the transaction commit side where items +are inserted into the CIL. Because transactions can enter this code +concurrently, the CIL needs to be protected separately from the above +commit/flush exclusion. It also needs to be an exclusive lock but it is only +held for a very short time and so a spin lock is appropriate here. It is +possible that this lock will become a contention point, but given the short +hold time once per transaction I think that contention is unlikely. + +The final serialisation point is the checkpoint commit record ordering code +that is run as part of the checkpoint commit and log force sequencing. The code +path that triggers a CIL flush (i.e. whatever triggers the log force) will enter +an ordering loop after writing all the log vectors into the log buffers but +before writing the commit record. This loop walks the list of committing +checkpoints and needs to block waiting for checkpoints to complete their commit +record write. As a result it needs a lock and a wait variable. Log force +sequencing also requires the same lock, list walk, and blocking mechanism to +ensure completion of checkpoints. + +These two sequencing operations can use the mechanism even though the +events they are waiting for are different. The checkpoint commit record +sequencing needs to wait until checkpoint contexts contain a commit LSN +(obtained through completion of a commit record write) while log force +sequencing needs to wait until previous checkpoint contexts are removed from +the committing list (i.e. they've completed). A simple wait variable and +broadcast wakeups (thundering herds) has been used to implement these two +serialisation queues. They use the same lock as the CIL, too. If we see too +much contention on the CIL lock, or too many context switches as a result of +the broadcast wakeups these operations can be put under a new spinlock and +given separate wait lists to reduce lock contention and the number of processes +woken by the wrong event. + + +Lifecycle Changes + +The existing log item life cycle is as follows: + + 1. Transaction allocate + 2. Transaction reserve + 3. Lock item + 4. Join item to transaction + If not already attached, + Allocate log item + Attach log item to owner item + Attach log item to transaction + 5. Modify item + Record modifications in log item + 6. Transaction commit + Pin item in memory + Format item into log buffer + Write commit LSN into transaction + Unlock item + Attach transaction to log buffer + + <log buffer IO dispatched> + <log buffer IO completes> + + 7. Transaction completion + Mark log item committed + Insert log item into AIL + Write commit LSN into log item + Unpin log item + 8. AIL traversal + Lock item + Mark log item clean + Flush item to disk + + <item IO completion> + + 9. Log item removed from AIL + Moves log tail + Item unlocked + +Essentially, steps 1-6 operate independently from step 7, which is also +independent of steps 8-9. An item can be locked in steps 1-6 or steps 8-9 +at the same time step 7 is occurring, but only steps 1-6 or 8-9 can occur +at the same time. If the log item is in the AIL or between steps 6 and 7 +and steps 1-6 are re-entered, then the item is relogged. Only when steps 8-9 +are entered and completed is the object considered clean. + +With delayed logging, there are new steps inserted into the life cycle: + + 1. Transaction allocate + 2. Transaction reserve + 3. Lock item + 4. Join item to transaction + If not already attached, + Allocate log item + Attach log item to owner item + Attach log item to transaction + 5. Modify item + Record modifications in log item + 6. Transaction commit + Pin item in memory if not pinned in CIL + Format item into log vector + buffer + Attach log vector and buffer to log item + Insert log item into CIL + Write CIL context sequence into transaction + Unlock item + + <next log force> + + 7. CIL push + lock CIL flush + Chain log vectors and buffers together + Remove items from CIL + unlock CIL flush + write log vectors into log + sequence commit records + attach checkpoint context to log buffer + + <log buffer IO dispatched> + <log buffer IO completes> + + 8. Checkpoint completion + Mark log item committed + Insert item into AIL + Write commit LSN into log item + Unpin log item + 9. AIL traversal + Lock item + Mark log item clean + Flush item to disk + <item IO completion> + 10. Log item removed from AIL + Moves log tail + Item unlocked + +From this, it can be seen that the only life cycle differences between the two +logging methods are in the middle of the life cycle - they still have the same +beginning and end and execution constraints. The only differences are in the +commiting of the log items to the log itself and the completion processing. +Hence delayed logging should not introduce any constraints on log item +behaviour, allocation or freeing that don't already exist. + +As a result of this zero-impact "insertion" of delayed logging infrastructure +and the design of the internal structures to avoid on disk format changes, we +can basically switch between delayed logging and the existing mechanism with a +mount option. Fundamentally, there is no reason why the log manager would not +be able to swap methods automatically and transparently depending on load +characteristics, but this should not be necessary if delayed logging works as +designed. + +Roadmap: + +2.6.37 Remove experimental tag from mount option + => should be roughly 6 months after initial merge + => enough time to: + => gain confidence and fix problems reported by early + adopters (a.k.a. guinea pigs) + => address worst performance regressions and undesired + behaviours + => start tuning/optimising code for parallelism + => start tuning/optimising algorithms consuming + excessive CPU time + +2.6.39 Switch default mount option to use delayed logging + => should be roughly 12 months after initial merge + => enough time to shake out remaining problems before next round of + enterprise distro kernel rebases |
