Review: Concurrent Filestreams V4
Concurrent Multi-File Data Streams
In media spaces, video is often stored in a frame-per-file format. When dealing with uncompressed realtime HD video streams in this format, it is crucial that files do not get fragmented and that multiple files a placed contiguously on disk.
When multiple streams are being ingested and played out at the same time, it is critical that the filesystem does not cross the streams and interleave them together as this creates seek and readahead cache miss latency and prevents both ingest and playout from meeting frame rate targets.
This patches creates a "stream of files" concept into the allocator to place all the data from a single stream contiguously on disk so that RAID array readahead can be used effectively. Each additional stream gets placed in different allocation groups within the filesystem, thereby ensuring that we don't cross any streams. When an AG fills up, we select a new AG for the stream that is not in use.
The core of the functionality is the stream tracking - each inode that we create in a directory needs to be associated with the directories' stream. Hence every time we create a file, we look up the directories' stream object and associate the new file with that object.
Once we have a stream object for a file, we use the AG that the stream object point to for allocations. If we can't allocate in that AG (e.g. it is full) we move the entire stream to another AG. Other inodes in the same stream are moved to the new AG on their next allocation (i.e. lazy update).
Stream objects are kept in a cache and hold a reference on the inode. Hence the inode cannot be reclaimed while there is an outstanding stream reference. This means that on unlink we need to remove the stream association and we also need to flush all the associations on certain events that want to reclaim all unreferenced inodes (e.g. filesystem freeze).
Credits: The original filestream allocator on Irix was written by Glen Overby, the Linux port and rewrite by Nathan Scott and Sam Vaughan (none of whom work at SGI any more). I just picked the pieces and beat it repeatedly with a big stick until it passed XFSQA.
Version 4:
o cleanup code in xfs_bmap_btalloc
o add comments to xfs_bmap_btalloc
o moved comments from xfs_mru_cache.h to mxfs_mru_cache.c so functions are documented rather than their protoypes
o fixed use-after-free in tracing code
o fixed xfs_release merge screwup
o fixed ABBA deadlock on the directory inode in xfs_filestream_associate during xfs_freeze
Version 3:
o use proper define for mount args
o make filestreams inode flag mark child inodes correctly so that filestreams are applied to them even if they are not tagged
o split quota inode filestreams avoidance out into a separate patch.
o move xfs_close() hooks for stream destruction on unlink to xfs_release().
Version 2:
o fold xfs_bmap_filestream() into xfs_bmap_btalloc()
o use ktrace infrastructure for debug code in xfs_filestream.c
o wrap repeated filestream inode checks.
o rename per-AG filestream reference counting macros and convert to static inline
o remove debug from xfs_mru_cache.[ch]
o fix function call/error check formatting.
o removed unnecessary fstrm_mnt_data_t structure.
o cleaned up ASSERT checks
o cleaned up namespace-less globals in xfs_mru_cache.c
o removed unnecessary casts
--- fs/xfs/Makefile-linux-2.6 | 2 fs/xfs/linux-2.6/xfs_globals.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_sysctl.c | 11 fs/xfs/linux-2.6/xfs_sysctl.h | 2 fs/xfs/xfs.h | 1 fs/xfs/xfs_ag.h | 1 fs/xfs/xfs_bmap.c | 69 +++ fs/xfs/xfs_clnt.h | 2 fs/xfs/xfs_dinode.h | 4 fs/xfs/xfs_filestream.c | 771 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/xfs_filestream.h | 136 +++++++ fs/xfs/xfs_fs.h | 1 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_inode.c | 17 fs/xfs/xfs_inode.h | 1 fs/xfs/xfs_mount.h | 4 fs/xfs/xfs_mru_cache.c | 608 ++++++++++++++++++++++++++++++++ fs/xfs/xfs_mru_cache.h | 57 +++ fs/xfs/xfs_vfsops.c | 26 + fs/xfs/xfs_vnodeops.c | 25 + fs/xfs/xfsidbg.c | 188 +++++++++ 22 files changed, 1918 insertions(+), 12 deletions(-)
fs/xfs/Makefile-linux-2.6 +2 -0
@@ -54,6 +54,7 @@ xfs-y += xfs_alloc.o \ xfs_dir2_sf.o \ xfs_error.o \ xfs_extfree_item.o \ + xfs_filestream.o \ xfs_fsops.o \ xfs_ialloc.o \ xfs_ialloc_btree.o \
@@ -67,6 +68,7 @@ xfs-y += xfs_alloc.o \ xfs_log.o \ xfs_log_recover.o \ xfs_mount.o \ + xfs_mru_cache.o \ xfs_rename.o \ xfs_trans.o \ xfs_trans_ail.o \
fs/xfs/linux-2.6/xfs_globals.c +1 -0
@@ -49,6 +49,7 @@ xfs_param_t xfs_params = { .inherit_nosym = { 0, 0, 1 }, .rotorstep = { 1, 1, 255 }, .inherit_nodfrg = { 0, 1, 1 }, + .fstrm_timer = { 1, 50, 3600*100}, };/*
fs/xfs/linux-2.6/xfs_linux.h +1 -0
@@ -132,6 +132,7 @@#define xfs_inherit_nosymlinks xfs_params.inherit_nosym.val#define xfs_rotorstep xfs_params.rotorstep.val#define xfs_inherit_nodefrag xfs_params.inherit_nodfrg.val +#define xfs_fstrm_centisecs xfs_params.fstrm_timer.val#define current_cpu () (raw_smp_processor_id())#define current_pid () (current->pid)
fs/xfs/linux-2.6/xfs_sysctl.c +11 -0
@@ -243,6 +243,17 @@ static ctl_table xfs_table[] = { .extra1 = &xfs_params.inherit_nodfrg.min, .extra2 = &xfs_params.inherit_nodfrg.max }, + { + .ctl_name = XFS_FILESTREAM_TIMER, + .procname = "filestream_centisecs", + .data = &xfs_params.fstrm_timer.val, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = &proc_dointvec_minmax, + .strategy = &sysctl_intvec, + .extra1 = &xfs_params.fstrm_timer.min, + .extra2 = &xfs_params.fstrm_timer.max, + }, /* please keep this the last entry */ #ifdef CONFIG_PROC_FS {
fs/xfs/linux-2.6/xfs_sysctl.h +2 -0
@@ -50,6 +50,7 @@ typedef struct xfs_param {xfs_sysctl_val_t inherit_nosym;/* Inherit the "nosymlinks" flag. */ xfs_sysctl_val_t rotorstep;/* inode32 AG rotoring control knob */ xfs_sysctl_val_t inherit_nodfrg;/* Inherit the "nodefrag" inode flag. */ +xfs_sysctl_val_t fstrm_timer;/* Filestream dir-AG assoc'n timeout. */ } xfs_param_t;/*
@@ -89,6 +90,7 @@ enum {XFS_INHERIT_NOSYM =19 ,XFS_ROTORSTEP =20 ,XFS_INHERIT_NODFRG =21 , + XFS_FILESTREAM_TIMER = 22, };extern xfs_param_t xfs_params;
fs/xfs/xfs_ag.h +1 -0
@@ -196,6 +196,7 @@ typedef struct xfs_peraglock_t pagb_lock;/* lock for pagb_list */ #endif xfs_perag_busy_t *pagb_list;/* unstable blocks */ +atomic_t pagf_fstrms;/* # of filestreams active in this AG */ int pag_ici_init;/* incore inode cache initialised */ rwlock_t pag_ici_lock;/* incore inode lock */
fs/xfs/xfs_bmap.c +62 -7
104 lines not on this page · raw message
fs/xfs/xfs_clnt.h +2 -0
@@ -98,5 +98,7 @@ struct xfs_mount_args {*/ #define XFSMNT2_COMPAT_IOSIZE 0x00000001/* don't report large preferred * I/O size in stat(2) */ +#define XFSMNT2_FILESTREAMS 0x00000002/* enable the filestreams +* allocator */ #endif /* __XFS_CLNT_H__ */
fs/xfs/xfs_dinode.h +3 -1
@@ -257,6 +257,7 @@ typedef enum xfs_dinode_fmt#define XFS_DIFLAG_EXTSIZE_BIT 11/* inode extent size allocator hint */ #define XFS_DIFLAG_EXTSZINHERIT_BIT 12/* inherit inode extent size */ #define XFS_DIFLAG_NODEFRAG_BIT 13/* do not reorganize/defragment */ +#define XFS_DIFLAG_FILESTREAM_BIT 14/* use filestream allocator */ #define XFS_DIFLAG_REALTIME (1 << XFS_DIFLAG_REALTIME_BIT)#define XFS_DIFLAG_PREALLOC (1 << XFS_DIFLAG_PREALLOC_BIT)#define XFS_DIFLAG_NEWRTBM (1 << XFS_DIFLAG_NEWRTBM_BIT)
@@ -271,12 +272,13 @@ typedef enum xfs_dinode_fmt#define XFS_DIFLAG_EXTSIZE (1 << XFS_DIFLAG_EXTSIZE_BIT)#define XFS_DIFLAG_EXTSZINHERIT (1 << XFS_DIFLAG_EXTSZINHERIT_BIT)#define XFS_DIFLAG_NODEFRAG (1 << XFS_DIFLAG_NODEFRAG_BIT) +#define XFS_DIFLAG_FILESTREAM (1 << XFS_DIFLAG_FILESTREAM_BIT)#define XFS_DIFLAG_ANY \ (XFS_DIFLAG_REALTIME | XFS_DIFLAG_PREALLOC | XFS_DIFLAG_NEWRTBM | \ XFS_DIFLAG_IMMUTABLE | XFS_DIFLAG_APPEND | XFS_DIFLAG_SYNC | \ XFS_DIFLAG_NOATIME | XFS_DIFLAG_NODUMP | XFS_DIFLAG_RTINHERIT | \ XFS_DIFLAG_PROJINHERIT | XFS_DIFLAG_NOSYMLINKS | XFS_DIFLAG_EXTSIZE | \ - XFS_DIFLAG_EXTSZINHERIT | XFS_DIFLAG_NODEFRAG) + XFS_DIFLAG_EXTSZINHERIT | XFS_DIFLAG_NODEFRAG | XFS_DIFLAG_FILESTREAM)#endif /* __XFS_DINODE_H__ */
fs/xfs/xfs_filestream.c new file +771 -0
771 lines not on this page · raw message
fs/xfs/xfs_filestream.h new file +136 -0
136 lines not on this page · raw message
fs/xfs/xfs_fs.h +1 -0
@@ -66,6 +66,7 @@ struct fsxattr {#define XFS_XFLAG_EXTSIZE 0x00000800/* extent size allocator hint */ #define XFS_XFLAG_EXTSZINHERIT 0x00001000/* inherit inode extent size */ #define XFS_XFLAG_NODEFRAG 0x00002000/* do not defragment */ +#define XFS_XFLAG_FILESTREAM 0x00004000/* use filestream allocator */ #define XFS_XFLAG_HASATTR 0x80000000/* no DIFLAG for this */ /*
fs/xfs/xfs_fsops.c +2 -0
@@ -44,6 +44,7 @@#include "xfs_trans_space.h" #include "xfs_rtalloc.h" #include "xfs_rw.h" +#include "xfs_filestream.h" /* * File system operations
@@ -165,6 +166,7 @@ xfs_growfs_data_private( new = nb - mp->m_sb.sb_dblocks; oagcount = mp->m_sb.sb_agcount;if (nagcount > oagcount) { +xfs_filestream_flush (mp);down_write (&mp->m_peraglock); mp->m_perag = kmem_realloc(mp->m_perag, sizeof(xfs_perag_t) * nagcount,
fs/xfs/xfs_inode.c +15 -2
@@ -48,6 +48,7 @@#include "xfs_dir2_trace.h" #include "xfs_quota.h" #include "xfs_acl.h" +#include "xfs_filestream.h" #include <linux/log2.h>
@@ -818,6 +819,8 @@ _xfs_dic2xflags( flags |=XFS_XFLAG_EXTSZINHERIT ;if (di_flags &XFS_DIFLAG_NODEFRAG ) flags |=XFS_XFLAG_NODEFRAG ; +if (di_flags &XFS_DIFLAG_FILESTREAM ) + flags |=XFS_XFLAG_FILESTREAM ; }return flags;
@@ -1151,7 +1154,7 @@ xfs_ialloc(/* * Project ids won't be stored on disk if we are using a version 1 inode. */ -if ( (prid !=0 ) && (ip->i_d.di_version ==XFS_DINODE_VERSION_1 )) +if ((prid !=0 ) && (ip->i_d.di_version ==XFS_DINODE_VERSION_1 ))xfs_bump_ino_vers2 (tp, ip);if (XFS_INHERIT_GID (pip, vp->v_vfsp)) {
@@ -1196,8 +1199,16 @@ xfs_ialloc( flags |=XFS_ILOG_DEV ;break ;case S_IFREG : +if (xfs_inode_is_filestream (pip)) { + error =xfs_filestream_associate (pip, ip); +if (error <0 ) +return -error; +if (!error) +xfs_iflags_set (ip,XFS_IFILESTREAM ); + } +/* fall through */ case S_IFDIR : -if (unlikely (pip->i_d.di_flags &XFS_DIFLAG_ANY )) { +if (pip->i_d.di_flags &XFS_DIFLAG_ANY ) {uint di_flags =0 ;if ((mode &S_IFMT ) ==S_IFDIR ) {
@@ -1234,6 +1245,8 @@ xfs_ialloc(if ((pip->i_d.di_flags &XFS_DIFLAG_NODEFRAG ) && xfs_inherit_nodefrag) di_flags |=XFS_DIFLAG_NODEFRAG ; +if (pip->i_d.di_flags &XFS_DIFLAG_FILESTREAM ) + di_flags |=XFS_DIFLAG_FILESTREAM ; ip->i_d.di_flags |= di_flags; }/* FALLTHROUGH */
fs/xfs/xfs_mount.h +4 -0
@@ -63,6 +63,7 @@ struct xfs_bmbt_irec;struct xfs_bmap_free ;struct xfs_extdelta ;struct xfs_swapext ; +struct xfs_mru_cache ;extern struct bhv_vfsops xfs_vfsops;extern struct bhv_vnodeops xfs_vnodeops;
@@ -431,6 +432,7 @@ typedef struct xfs_mount {struct notifier_block m_icsb_notifier;/* hotplug cpu notifier */ struct mutex m_icsb_mutex;/* balancer sync lock */ #endif +struct xfs_mru_cache *m_filestream;/* per-mount filestream data */ } xfs_mount_t;/*
@@ -470,6 +472,8 @@ typedef struct xfs_mount {* I/O size in stat() */ #define XFS_MOUNT_NO_PERCPU_SB (1ULL << 23)/* don't use per-cpu superblock counters */ +#define XFS_MOUNT_FILESTREAMS (1ULL << 24)/* enable the filestreams +allocator */ /*
fs/xfs/xfs_mru_cache.c new file +608 -0
608 lines not on this page · raw message
fs/xfs/xfs_mru_cache.h new file +57 -0
57 lines not on this page · raw message
fs/xfs/xfs_vfsops.c +26 -0
@@ -51,6 +51,8 @@#include "xfs_acl.h" #include "xfs_attr.h" #include "xfs_clnt.h" +#include "xfs_mru_cache.h" +#include "xfs_filestream.h" #include "xfs_fsops.h" STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
@@ -81,6 +83,8 @@ xfs_init(void) xfs_dabuf_zone =kmem_zone_init (sizeof (xfs_dabuf_t),"xfs_dabuf" ); xfs_ifork_zone =kmem_zone_init (sizeof (xfs_ifork_t),"xfs_ifork" );xfs_acl_zone_init (xfs_acl_zone,"xfs_acl" ); +xfs_mru_cache_init (); +xfs_filestream_init ();/* * The size of the zone allocated buf log item is the maximum
@@ -164,6 +168,8 @@ xfs_cleanup(void)xfs_cleanup_procfs ();xfs_sysctl_unregister ();xfs_refcache_destroy (); +xfs_filestream_uninit (); +xfs_mru_cache_uninit ();xfs_acl_zone_destroy (xfs_acl_zone);#ifdef XFS_DIR2_TRACE
@@ -317,6 +323,9 @@ xfs_start_flags(else mp->m_flags &= ~XFS_MOUNT_BARRIER ; +if (ap->flags2 &XFSMNT2_FILESTREAMS ) + mp->m_flags |=XFS_MOUNT_FILESTREAMS ; +return 0 ; }
@@ -515,6 +524,9 @@ xfs_mount(if (mp->m_flags &XFS_MOUNT_BARRIER )xfs_mountfs_check_barriers (mp); +if ((error =xfs_filestream_mount (mp))) + goto error2; + error =XFS_IOINIT (vfsp, args, flags);if (error) goto error2;
@@ -572,6 +584,13 @@ xfs_unmount(*/ xfs_refcache_purge_mp (mp); +/* +* Blow away any referenced inode in the filestreams cache. +* This can and will cause log traffic as inodes go inactive +* here. +*/ +xfs_filestream_unmount (mp); +XFS_bflush (mp->m_ddev_targp); error =xfs_unmount_flush (mp,0 );if (error)
@@ -703,6 +722,7 @@ xfs_mntupdate( mp->m_flags &= ~XFS_MOUNT_BARRIER ; } }else if (!(vfsp->vfs_flag &VFS_RDONLY )) {/* rw -> ro */ +xfs_filestream_flush (mp);bhv_vfs_sync (vfsp,SYNC_DATA_QUIESCE ,NULL );xfs_attr_quiesce (mp); vfsp->vfs_flag |=VFS_RDONLY ;
@@ -927,6 +947,9 @@ xfs_sync( {xfs_mount_t *mp =XFS_BHVTOM (bdp); +if (flags &SYNC_IOWAIT ) +xfs_filestream_flush (mp); +return xfs_syncsub (mp, flags,NULL ); }
@@ -1676,6 +1699,7 @@ xfs_vget(* in stat(). */ #define MNTOPT_ATTR2 "attr2"/* do use attr2 attribute format */ #define MNTOPT_NOATTR2 "noattr2"/* do not use attr2 attribute format */ +#define MNTOPT_FILESTREAM "filestreams"/* use filestreams allocator */ STATIC unsigned long suffix_strtoul (char *s,char **endp,unsigned int base)
@@ -1853,6 +1877,8 @@ xfs_parseargs( args->flags |=XFSMNT_ATTR2 ; }else if (!strcmp (this_char,MNTOPT_NOATTR2 )) { args->flags &= ~XFSMNT_ATTR2 ; + }else if (!strcmp (this_char,MNTOPT_FILESTREAM )) { + args->flags2 |=XFSMNT2_FILESTREAMS ; }else if (!strcmp (this_char,"osyncisdsync" )) {/* no-op, this is now the default */ cmn_err(CE_WARN,
fs/xfs/xfs_vnodeops.c +23 -2
@@ -51,6 +51,7 @@#include "xfs_refcache.h" #include "xfs_trans_space.h" #include "xfs_log_priv.h" +#include "xfs_filestream.h" STATIC int xfs_open(
@@ -789,6 +790,8 @@ xfs_setattr( di_flags |=XFS_DIFLAG_PROJINHERIT ;if (vap->va_xflags &XFS_XFLAG_NODEFRAG ) di_flags |=XFS_DIFLAG_NODEFRAG ; +if (vap->va_xflags &XFS_XFLAG_FILESTREAM ) + di_flags |=XFS_DIFLAG_FILESTREAM ;if ((ip->i_d.di_mode &S_IFMT ) ==S_IFDIR ) {if (vap->va_xflags &XFS_XFLAG_RTINHERIT ) di_flags |=XFS_DIFLAG_RTINHERIT ;
@@ -1542,7 +1545,17 @@ xfs_release(if (vp->v_vfsp->vfs_flag &VFS_RDONLY )return 0 ; -if (!XFS_FORCED_SHUTDOWN (ip->i_mount)) { +if (!XFS_FORCED_SHUTDOWN (mp)) { +/* +* If we are using filestreams, and we have an unlinked +* file that we are processing the last close on, then nothing +* will be able to reopen and write to this file. Purge this +* inode from the filestreams cache so that it doesn't delay +* teardown of the inode. +*/ +if ((ip->i_d.di_nlink ==0 ) &&xfs_inode_is_filestream (ip)) +xfs_filestream_deassociate (ip); +/* * If we previously truncated this file and removed old data * in the process, we want to initiate "early" writeout on
@@ -1557,7 +1570,6 @@ xfs_release(bhv_vop_flush_pages (vp,0 ,-1 ,XFS_B_ASYNC ,FI_NONE ); } -#ifdef HAVE_REFCACHE /* If we are in the NFS reference cache then don't do this now */ if (ip->i_refcache)
@@ -2551,6 +2563,15 @@ xfs_remove(*/ xfs_refcache_purge_ip (ip); +/* +* If we are using filestreams, kill the stream association. +* If the file is still open it may get a new one but that +* will get killed on last close in xfs_close() so we don't +* have to worry about that. +*/ +if (link_zero &&xfs_inode_is_filestream (ip)) +xfs_filestream_deassociate (ip); +vn_trace_exit (XFS_ITOV (ip), __FUNCTION__, (inst_t *)__return_address);/*
fs/xfs/xfs.h +1 -0
@@ -38,6 +38,7 @@#define XFS_RW_TRACE 1#define XFS_BUF_TRACE 1#define XFS_VNODE_TRACE 1 +#define XFS_FILESTREAMS_TRACE 1#endif #include <linux/version.h>
fs/xfs/xfsidbg.c +188 -0
224 lines not on this page · raw message
fs/xfs/xfs_inode.h +1 -0
@@ -350,6 +350,7 @@ xfs_iflags_test(xfs_inode_t *ip, unsigne#define XFS_ISTALE 0x0010/* inode has been staled */ #define XFS_IRECLAIMABLE 0x0020/* inode can be reclaimed */ #define XFS_INEW 0x0040 +#define XFS_IFILESTREAM 0x0080/* inode is in a filestream directory */ /* * Flags for inode locking.