mirrors/linux-stable
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git synced 2024-10-03 07:38:10 +00:00
Code Issues Releases Wiki Activity

Merge branch 'xfs-filestreams-lookup' into for-next

Browse source
This commit is contained in:
Dave Chinner 2014-05-15 09:36:59 +10:00
commit 232c2f5c65
10 changed files with 407 additions and 810 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

204
fs/xfs/xfs_bmap.c
View file

@ -3515,6 +3515,67 @@ xfs_bmap_adjacent(
#undef ISVALID #undef ISVALID
} }
static int
xfs_bmap_longest_free_extent(
struct xfs_trans *tp,
xfs_agnumber_t ag,
xfs_extlen_t *blen,
int *notinit)
{
struct xfs_mount *mp = tp->t_mountp;
struct xfs_perag *pag;
xfs_extlen_t longest;
int error = 0;
pag = xfs_perag_get(mp, ag);
if (!pag->pagf_init) {
error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
if (error)
goto out;
if (!pag->pagf_init) {
*notinit = 1;
goto out;
}
}
longest = xfs_alloc_longest_free_extent(mp, pag);
if (*blen < longest)
*blen = longest;
out:
xfs_perag_put(pag);
return error;
}
static void
xfs_bmap_select_minlen(
struct xfs_bmalloca *ap,
struct xfs_alloc_arg *args,
xfs_extlen_t *blen,
int notinit)
{
if (notinit || *blen < ap->minlen) {
/*
* Since we did a BUF_TRYLOCK above, it is possible that
* there is space for this request.
*/
args->minlen = ap->minlen;
} else if (*blen < args->maxlen) {
/*
* If the best seen length is less than the request length,
* use the best as the minimum.
*/
args->minlen = *blen;
} else {
/*
* Otherwise we've seen an extent as big as maxlen, use that
* as the minimum.
*/
args->minlen = args->maxlen;
}
}
STATIC int STATIC int
xfs_bmap_btalloc_nullfb( xfs_bmap_btalloc_nullfb(
struct xfs_bmalloca *ap, struct xfs_bmalloca *ap,
@ -3522,111 +3583,74 @@ xfs_bmap_btalloc_nullfb(
xfs_extlen_t *blen) xfs_extlen_t *blen)
{ {
struct xfs_mount *mp = ap->ip->i_mount; struct xfs_mount *mp = ap->ip->i_mount;
struct xfs_perag *pag;
xfs_agnumber_t ag, startag; xfs_agnumber_t ag, startag;
int notinit = 0; int notinit = 0;
int error; int error;
if (ap->userdata && xfs_inode_is_filestream(ap->ip)) args->type = XFS_ALLOCTYPE_START_BNO;
args->type = XFS_ALLOCTYPE_NEAR_BNO;
else
args->type = XFS_ALLOCTYPE_START_BNO;
args->total = ap->total; args->total = ap->total;
/*
* Search for an allocation group with a single extent large enough
* for the request. If one isn't found, then adjust the minimum
* allocation size to the largest space found.
*/
startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
if (startag == NULLAGNUMBER) if (startag == NULLAGNUMBER)
startag = ag = 0; startag = ag = 0;
pag = xfs_perag_get(mp, ag);
while (*blen < args->maxlen) { while (*blen < args->maxlen) {
if (!pag->pagf_init) { error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
error = xfs_alloc_pagf_init(mp, args->tp, ag, &notinit);
XFS_ALLOC_FLAG_TRYLOCK); if (error)
if (error) { return error;
xfs_perag_put(pag);
return error;
}
}
/*
* See xfs_alloc_fix_freelist...
*/
if (pag->pagf_init) {
xfs_extlen_t longest;
longest = xfs_alloc_longest_free_extent(mp, pag);
if (*blen < longest)
*blen = longest;
} else
notinit = 1;
if (xfs_inode_is_filestream(ap->ip)) {
if (*blen >= args->maxlen)
break;
if (ap->userdata) {
/*
* If startag is an invalid AG, we've
* come here once before and
* xfs_filestream_new_ag picked the
* best currently available.
*
* Don't continue looping, since we
* could loop forever.
*/
if (startag == NULLAGNUMBER)
break;
error = xfs_filestream_new_ag(ap, &ag);
xfs_perag_put(pag);
if (error)
return error;
/* loop again to set 'blen'*/
startag = NULLAGNUMBER;
pag = xfs_perag_get(mp, ag);
continue;
}
}
if (++ag == mp->m_sb.sb_agcount) if (++ag == mp->m_sb.sb_agcount)
ag = 0; ag = 0;
if (ag == startag) if (ag == startag)
break; break;
xfs_perag_put(pag);
pag = xfs_perag_get(mp, ag);
} }
xfs_perag_put(pag);
xfs_bmap_select_minlen(ap, args, blen, notinit);
return 0;
}
STATIC int
xfs_bmap_btalloc_filestreams(
struct xfs_bmalloca *ap,
struct xfs_alloc_arg *args,
xfs_extlen_t *blen)
{
struct xfs_mount *mp = ap->ip->i_mount;
xfs_agnumber_t ag;
int notinit = 0;
int error;
args->type = XFS_ALLOCTYPE_NEAR_BNO;
args->total = ap->total;
ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
if (ag == NULLAGNUMBER)
ag = 0;
error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
if (error)
return error;
if (*blen < args->maxlen) {
error = xfs_filestream_new_ag(ap, &ag);
if (error)
return error;
error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
&notinit);
if (error)
return error;
}
xfs_bmap_select_minlen(ap, args, blen, notinit);
/* /*
* Since the above loop did a BUF_TRYLOCK, it is * Set the failure fallback case to look in the selected AG as stream
* possible that there is space for this request. * may have moved.
*/ */
if (notinit || *blen < ap->minlen) ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
args->minlen = ap->minlen;
/*
* If the best seen length is less than the request
* length, use the best as the minimum.
*/
else if (*blen < args->maxlen)
args->minlen = *blen;
/*
* Otherwise we've seen an extent as big as maxlen,
* use that as the minimum.
*/
else
args->minlen = args->maxlen;
/*
* set the failure fallback case to look in the selected
* AG as the stream may have moved.
*/
if (xfs_inode_is_filestream(ap->ip))
ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
return 0; return 0;
} }
@ -3706,7 +3730,15 @@ xfs_bmap_btalloc(
args.firstblock = *ap->firstblock; args.firstblock = *ap->firstblock;
blen = 0; blen = 0;
if (nullfb) { if (nullfb) {
error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); /*
* Search for an allocation group with a single extent large
* enough for the request. If one isn't found, then adjust
* the minimum allocation size to the largest space found.
*/
if (ap->userdata && xfs_inode_is_filestream(ap->ip))
error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
else
error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
if (error) if (error)
return error; return error;
} else if (ap->flist->xbf_low) { } else if (ap->flist->xbf_low) {

690
fs/xfs/xfs_filestream.c
View file

@ -1,5 +1,6 @@
/* /*
* Copyright (c) 2006-2007 Silicon Graphics, Inc. * Copyright (c) 2006-2007 Silicon Graphics, Inc.
* Copyright (c) 2014 Christoph Hellwig.
* All Rights Reserved. * All Rights Reserved.
* *
* This program is free software; you can redistribute it and/or * This program is free software; you can redistribute it and/or
@ -32,100 +33,20 @@
#include "xfs_filestream.h" #include "xfs_filestream.h"
#include "xfs_trace.h" #include "xfs_trace.h"
#ifdef XFS_FILESTREAMS_TRACE struct xfs_fstrm_item {
struct xfs_mru_cache_elem mru;
struct xfs_inode *ip;
xfs_agnumber_t ag; /* AG in use for this directory */
};
ktrace_t *xfs_filestreams_trace_buf; enum xfs_fstrm_alloc {
XFS_PICK_USERDATA = 1,
STATIC void XFS_PICK_LOWSPACE = 2,
xfs_filestreams_trace( };
xfs_mount_t *mp, /* mount point */
int type, /* type of trace */
const char *func, /* source function */
int line, /* source line number */
__psunsigned_t arg0,
__psunsigned_t arg1,
__psunsigned_t arg2,
__psunsigned_t arg3,
__psunsigned_t arg4,
__psunsigned_t arg5)
{
ktrace_enter(xfs_filestreams_trace_buf,
(void *)(__psint_t)(type | (line << 16)),
(void *)func,
(void *)(__psunsigned_t)current_pid(),
(void *)mp,
(void *)(__psunsigned_t)arg0,
(void *)(__psunsigned_t)arg1,
(void *)(__psunsigned_t)arg2,
(void *)(__psunsigned_t)arg3,
(void *)(__psunsigned_t)arg4,
(void *)(__psunsigned_t)arg5,
NULL, NULL, NULL, NULL, NULL, NULL);
}
#define TRACE0(mp,t) TRACE6(mp,t,0,0,0,0,0,0)
#define TRACE1(mp,t,a0) TRACE6(mp,t,a0,0,0,0,0,0)
#define TRACE2(mp,t,a0,a1) TRACE6(mp,t,a0,a1,0,0,0,0)
#define TRACE3(mp,t,a0,a1,a2) TRACE6(mp,t,a0,a1,a2,0,0,0)
#define TRACE4(mp,t,a0,a1,a2,a3) TRACE6(mp,t,a0,a1,a2,a3,0,0)
#define TRACE5(mp,t,a0,a1,a2,a3,a4) TRACE6(mp,t,a0,a1,a2,a3,a4,0)
#define TRACE6(mp,t,a0,a1,a2,a3,a4,a5) \
xfs_filestreams_trace(mp, t, __func__, __LINE__, \
(__psunsigned_t)a0, (__psunsigned_t)a1, \
(__psunsigned_t)a2, (__psunsigned_t)a3, \
(__psunsigned_t)a4, (__psunsigned_t)a5)
#define TRACE_AG_SCAN(mp, ag, ag2) \
TRACE2(mp, XFS_FSTRM_KTRACE_AGSCAN, ag, ag2);
#define TRACE_AG_PICK1(mp, max_ag, maxfree) \
TRACE2(mp, XFS_FSTRM_KTRACE_AGPICK1, max_ag, maxfree);
#define TRACE_AG_PICK2(mp, ag, ag2, cnt, free, scan, flag) \
TRACE6(mp, XFS_FSTRM_KTRACE_AGPICK2, ag, ag2, \
cnt, free, scan, flag)
#define TRACE_UPDATE(mp, ip, ag, cnt, ag2, cnt2) \
TRACE5(mp, XFS_FSTRM_KTRACE_UPDATE, ip, ag, cnt, ag2, cnt2)
#define TRACE_FREE(mp, ip, pip, ag, cnt) \
TRACE4(mp, XFS_FSTRM_KTRACE_FREE, ip, pip, ag, cnt)
#define TRACE_LOOKUP(mp, ip, pip, ag, cnt) \
TRACE4(mp, XFS_FSTRM_KTRACE_ITEM_LOOKUP, ip, pip, ag, cnt)
#define TRACE_ASSOCIATE(mp, ip, pip, ag, cnt) \
TRACE4(mp, XFS_FSTRM_KTRACE_ASSOCIATE, ip, pip, ag, cnt)
#define TRACE_MOVEAG(mp, ip, pip, oag, ocnt, nag, ncnt) \
TRACE6(mp, XFS_FSTRM_KTRACE_MOVEAG, ip, pip, oag, ocnt, nag, ncnt)
#define TRACE_ORPHAN(mp, ip, ag) \
TRACE2(mp, XFS_FSTRM_KTRACE_ORPHAN, ip, ag);
#else
#define TRACE_AG_SCAN(mp, ag, ag2)
#define TRACE_AG_PICK1(mp, max_ag, maxfree)
#define TRACE_AG_PICK2(mp, ag, ag2, cnt, free, scan, flag)
#define TRACE_UPDATE(mp, ip, ag, cnt, ag2, cnt2)
#define TRACE_FREE(mp, ip, pip, ag, cnt)
#define TRACE_LOOKUP(mp, ip, pip, ag, cnt)
#define TRACE_ASSOCIATE(mp, ip, pip, ag, cnt)
#define TRACE_MOVEAG(mp, ip, pip, oag, ocnt, nag, ncnt)
#define TRACE_ORPHAN(mp, ip, ag)
#endif
static kmem_zone_t *item_zone;
/*
* Structure for associating a file or a directory with an allocation group.
* The parent directory pointer is only needed for files, but since there will
* generally be vastly more files than directories in the cache, using the same
* data structure simplifies the code with very little memory overhead.
*/
typedef struct fstrm_item
{
xfs_agnumber_t ag; /* AG currently in use for the file/directory. */
xfs_inode_t *ip; /* inode self-pointer. */
xfs_inode_t *pip; /* Parent directory inode pointer. */
} fstrm_item_t;
/* /*
* Allocation group filestream associations are tracked with per-ag atomic * Allocation group filestream associations are tracked with per-ag atomic
* counters. These counters allow _xfs_filestream_pick_ag() to tell whether a * counters. These counters allow xfs_filestream_pick_ag() to tell whether a
* particular AG already has active filestreams associated with it. The mount * particular AG already has active filestreams associated with it. The mount
* point's m_peraglock is used to protect these counters from per-ag array * point's m_peraglock is used to protect these counters from per-ag array
* re-allocation during a growfs operation. When xfs_growfs_data_private() is * re-allocation during a growfs operation. When xfs_growfs_data_private() is
@ -160,7 +81,7 @@ typedef struct fstrm_item
* the cache that reference per-ag array elements that have since been * the cache that reference per-ag array elements that have since been
* reallocated. * reallocated.
*/ */
static int int
xfs_filestream_peek_ag( xfs_filestream_peek_ag(
xfs_mount_t *mp, xfs_mount_t *mp,
xfs_agnumber_t agno) xfs_agnumber_t agno)
@ -200,23 +121,40 @@ xfs_filestream_put_ag(
xfs_perag_put(pag); xfs_perag_put(pag);
} }
static void
xfs_fstrm_free_func(
struct xfs_mru_cache_elem *mru)
{
struct xfs_fstrm_item *item =
container_of(mru, struct xfs_fstrm_item, mru);
xfs_filestream_put_ag(item->ip->i_mount, item->ag);
trace_xfs_filestream_free(item->ip, item->ag);
kmem_free(item);
}
/* /*
* Scan the AGs starting at startag looking for an AG that isn't in use and has * Scan the AGs starting at startag looking for an AG that isn't in use and has
* at least minlen blocks free. * at least minlen blocks free.
*/ */
static int static int
_xfs_filestream_pick_ag( xfs_filestream_pick_ag(
xfs_mount_t *mp, struct xfs_inode *ip,
xfs_agnumber_t startag, xfs_agnumber_t startag,
xfs_agnumber_t *agp, xfs_agnumber_t *agp,
int flags, int flags,
xfs_extlen_t minlen) xfs_extlen_t minlen)
{ {
int streams, max_streams; struct xfs_mount *mp = ip->i_mount;
int err, trylock, nscan; struct xfs_fstrm_item *item;
xfs_extlen_t longest, free, minfree, maxfree = 0; struct xfs_perag *pag;
xfs_agnumber_t ag, max_ag = NULLAGNUMBER; xfs_extlen_t longest, free = 0, minfree, maxfree = 0;
struct xfs_perag *pag; xfs_agnumber_t ag, max_ag = NULLAGNUMBER;
int err, trylock, nscan;
ASSERT(S_ISDIR(ip->i_d.di_mode));
/* 2% of an AG's blocks must be free for it to be chosen. */ /* 2% of an AG's blocks must be free for it to be chosen. */
minfree = mp->m_sb.sb_agblocks / 50; minfree = mp->m_sb.sb_agblocks / 50;
@ -228,8 +166,9 @@ _xfs_filestream_pick_ag(
trylock = XFS_ALLOC_FLAG_TRYLOCK; trylock = XFS_ALLOC_FLAG_TRYLOCK;
for (nscan = 0; 1; nscan++) { for (nscan = 0; 1; nscan++) {
trace_xfs_filestream_scan(ip, ag);
pag = xfs_perag_get(mp, ag); pag = xfs_perag_get(mp, ag);
TRACE_AG_SCAN(mp, ag, atomic_read(&pag->pagf_fstrms));
if (!pag->pagf_init) { if (!pag->pagf_init) {
err = xfs_alloc_pagf_init(mp, NULL, ag, trylock); err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
@ -246,7 +185,6 @@ _xfs_filestream_pick_ag(
/* Keep track of the AG with the most free blocks. */ /* Keep track of the AG with the most free blocks. */
if (pag->pagf_freeblks > maxfree) { if (pag->pagf_freeblks > maxfree) {
maxfree = pag->pagf_freeblks; maxfree = pag->pagf_freeblks;
max_streams = atomic_read(&pag->pagf_fstrms);
max_ag = ag; max_ag = ag;
} }
@ -269,7 +207,6 @@ _xfs_filestream_pick_ag(
/* Break out, retaining the reference on the AG. */ /* Break out, retaining the reference on the AG. */
free = pag->pagf_freeblks; free = pag->pagf_freeblks;
streams = atomic_read(&pag->pagf_fstrms);
xfs_perag_put(pag); xfs_perag_put(pag);
*agp = ag; *agp = ag;
break; break;
@ -305,317 +242,98 @@ _xfs_filestream_pick_ag(
*/ */
if (max_ag != NULLAGNUMBER) { if (max_ag != NULLAGNUMBER) {
xfs_filestream_get_ag(mp, max_ag); xfs_filestream_get_ag(mp, max_ag);
TRACE_AG_PICK1(mp, max_ag, maxfree);
streams = max_streams;
free = maxfree; free = maxfree;
*agp = max_ag; *agp = max_ag;
break; break;
} }
/* take AG 0 if none matched */ /* take AG 0 if none matched */
TRACE_AG_PICK1(mp, max_ag, maxfree); trace_xfs_filestream_pick(ip, *agp, free, nscan);
*agp = 0; *agp = 0;
return 0; return 0;
} }
TRACE_AG_PICK2(mp, startag, *agp, streams, free, nscan, flags); trace_xfs_filestream_pick(ip, *agp, free, nscan);
return 0; if (*agp == NULLAGNUMBER)
}
/*
* Set the allocation group number for a file or a directory, updating inode
* references and per-AG references as appropriate.
*/
static int
_xfs_filestream_update_ag(
xfs_inode_t *ip,
xfs_inode_t *pip,
xfs_agnumber_t ag)
{
int err = 0;
xfs_mount_t *mp;
xfs_mru_cache_t *cache;
fstrm_item_t *item;
xfs_agnumber_t old_ag;
xfs_inode_t *old_pip;
/*
* Either ip is a regular file and pip is a directory, or ip is a
* directory and pip is NULL.
*/
ASSERT(ip && ((S_ISREG(ip->i_d.di_mode) && pip &&
S_ISDIR(pip->i_d.di_mode)) ||
(S_ISDIR(ip->i_d.di_mode) && !pip)));
mp = ip->i_mount;
cache = mp->m_filestream;
item = xfs_mru_cache_lookup(cache, ip->i_ino);
if (item) {
ASSERT(item->ip == ip);
old_ag = item->ag;
item->ag = ag;
old_pip = item->pip;
item->pip = pip;
xfs_mru_cache_done(cache);
/*
* If the AG has changed, drop the old ref and take a new one,
* effectively transferring the reference from old to new AG.
*/
if (ag != old_ag) {
xfs_filestream_put_ag(mp, old_ag);
xfs_filestream_get_ag(mp, ag);
}
/*
* If ip is a file and its pip has changed, drop the old ref and
* take a new one.
*/
if (pip && pip != old_pip) {
IRELE(old_pip);
IHOLD(pip);
}
TRACE_UPDATE(mp, ip, old_ag, xfs_filestream_peek_ag(mp, old_ag),
ag, xfs_filestream_peek_ag(mp, ag));
return 0; return 0;
}
item = kmem_zone_zalloc(item_zone, KM_MAYFAIL); err = ENOMEM;
item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
if (!item) if (!item)
return ENOMEM; goto out_put_ag;
item->ag = ag; item->ag = *agp;
item->ip = ip; item->ip = ip;
item->pip = pip;
err = xfs_mru_cache_insert(cache, ip->i_ino, item); err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);
if (err) { if (err) {
kmem_zone_free(item_zone, item); if (err == EEXIST)
return err; err = 0;
goto out_free_item;
} }
/* Take a reference on the AG. */
xfs_filestream_get_ag(mp, ag);
/*
* Take a reference on the inode itself regardless of whether it's a
* regular file or a directory.
*/
IHOLD(ip);
/*
* In the case of a regular file, take a reference on the parent inode
* as well to ensure it remains in-core.
*/
if (pip)
IHOLD(pip);
TRACE_UPDATE(mp, ip, ag, xfs_filestream_peek_ag(mp, ag),
ag, xfs_filestream_peek_ag(mp, ag));
return 0; return 0;
}
/* xfs_fstrm_free_func(): callback for freeing cached stream items. */
STATIC void
xfs_fstrm_free_func(
unsigned long ino,
void *data)
{
fstrm_item_t *item = (fstrm_item_t *)data;
xfs_inode_t *ip = item->ip;
ASSERT(ip->i_ino == ino);
xfs_iflags_clear(ip, XFS_IFILESTREAM);
/* Drop the reference taken on the AG when the item was added. */
xfs_filestream_put_ag(ip->i_mount, item->ag);
TRACE_FREE(ip->i_mount, ip, item->pip, item->ag,
xfs_filestream_peek_ag(ip->i_mount, item->ag));
/*
* _xfs_filestream_update_ag() always takes a reference on the inode
* itself, whether it's a file or a directory. Release it here.
* This can result in the inode being freed and so we must
* not hold any inode locks when freeing filesstreams objects
* otherwise we can deadlock here.
*/
IRELE(ip);
/*
* In the case of a regular file, _xfs_filestream_update_ag() also
* takes a ref on the parent inode to keep it in-core. Release that
* too.
*/
if (item->pip)
IRELE(item->pip);
/* Finally, free the memory allocated for the item. */
kmem_zone_free(item_zone, item);
}
/*
* xfs_filestream_init() is called at xfs initialisation time to set up the
* memory zone that will be used for filestream data structure allocation.
*/
int
xfs_filestream_init(void)
{
item_zone = kmem_zone_init(sizeof(fstrm_item_t), "fstrm_item");
if (!item_zone)
return -ENOMEM;
return 0;
}
/*
* xfs_filestream_uninit() is called at xfs termination time to destroy the
* memory zone that was used for filestream data structure allocation.
*/
void
xfs_filestream_uninit(void)
{
kmem_zone_destroy(item_zone);
}
/*
* xfs_filestream_mount() is called when a file system is mounted with the
* filestream option. It is responsible for allocating the data structures
* needed to track the new file system's file streams.
*/
int
xfs_filestream_mount(
xfs_mount_t *mp)
{
int err;
unsigned int lifetime, grp_count;
/*
* The filestream timer tunable is currently fixed within the range of
* one second to four minutes, with five seconds being the default. The
* group count is somewhat arbitrary, but it'd be nice to adhere to the
* timer tunable to within about 10 percent. This requires at least 10
* groups.
*/
lifetime = xfs_fstrm_centisecs * 10;
grp_count = 10;
err = xfs_mru_cache_create(&mp->m_filestream, lifetime, grp_count,
xfs_fstrm_free_func);
out_free_item:
kmem_free(item);
out_put_ag:
xfs_filestream_put_ag(mp, *agp);
return err; return err;
} }
/* static struct xfs_inode *
* xfs_filestream_unmount() is called when a file system that was mounted with xfs_filestream_get_parent(
* the filestream option is unmounted. It drains the data structures created struct xfs_inode *ip)
* to track the file system's file streams and frees all the memory that was
* allocated.
*/
void
xfs_filestream_unmount(
xfs_mount_t *mp)
{ {
xfs_mru_cache_destroy(mp->m_filestream); struct inode *inode = VFS_I(ip), *dir = NULL;
struct dentry *dentry, *parent;
dentry = d_find_alias(inode);
if (!dentry)
goto out;
parent = dget_parent(dentry);
if (!parent)
goto out_dput;
dir = igrab(parent->d_inode);
dput(parent);
out_dput:
dput(dentry);
out:
return dir ? XFS_I(dir) : NULL;
} }
/* /*
* Return the AG of the filestream the file or directory belongs to, or * Find the right allocation group for a file, either by finding an
* NULLAGNUMBER otherwise. * existing file stream or creating a new one.
*
* Returns NULLAGNUMBER in case of an error.
*/ */
xfs_agnumber_t xfs_agnumber_t
xfs_filestream_lookup_ag( xfs_filestream_lookup_ag(
xfs_inode_t *ip) struct xfs_inode *ip)
{ {
xfs_mru_cache_t *cache; struct xfs_mount *mp = ip->i_mount;
fstrm_item_t *item; struct xfs_inode *pip = NULL;
xfs_agnumber_t ag; xfs_agnumber_t startag, ag = NULLAGNUMBER;
int ref; struct xfs_mru_cache_elem *mru;
if (!S_ISREG(ip->i_d.di_mode) && !S_ISDIR(ip->i_d.di_mode)) {
ASSERT(0);
return NULLAGNUMBER;
}
cache = ip->i_mount->m_filestream;
item = xfs_mru_cache_lookup(cache, ip->i_ino);
if (!item) {
TRACE_LOOKUP(ip->i_mount, ip, NULL, NULLAGNUMBER, 0);
return NULLAGNUMBER;
}
ASSERT(ip == item->ip);
ag = item->ag;
ref = xfs_filestream_peek_ag(ip->i_mount, ag);
xfs_mru_cache_done(cache);
TRACE_LOOKUP(ip->i_mount, ip, item->pip, ag, ref);
return ag;
}
/*
* xfs_filestream_associate() should only be called to associate a regular file
* with its parent directory. Calling it with a child directory isn't
* appropriate because filestreams don't apply to entire directory hierarchies.
* Creating a file in a child directory of an existing filestream directory
* starts a new filestream with its own allocation group association.
*
* Returns < 0 on error, 0 if successful association occurred, > 0 if
* we failed to get an association because of locking issues.
*/
int
xfs_filestream_associate(
xfs_inode_t *pip,
xfs_inode_t *ip)
{
xfs_mount_t *mp;
xfs_mru_cache_t *cache;
fstrm_item_t *item;
xfs_agnumber_t ag, rotorstep, startag;
int err = 0;
ASSERT(S_ISDIR(pip->i_d.di_mode));
ASSERT(S_ISREG(ip->i_d.di_mode)); ASSERT(S_ISREG(ip->i_d.di_mode));
if (!S_ISDIR(pip->i_d.di_mode) || !S_ISREG(ip->i_d.di_mode))
return -EINVAL;
mp = pip->i_mount; pip = xfs_filestream_get_parent(ip);
cache = mp->m_filestream; if (!pip)
goto out;
/* mru = xfs_mru_cache_lookup(mp->m_filestream, pip->i_ino);
* We have a problem, Houston. if (mru) {
* ag = container_of(mru, struct xfs_fstrm_item, mru)->ag;
* Taking the iolock here violates inode locking order - we already xfs_mru_cache_done(mp->m_filestream);
* hold the ilock. Hence if we block getting this lock we may never
* wake. Unfortunately, that means if we can't get the lock, we're
* screwed in terms of getting a stream association - we can't spin
* waiting for the lock because someone else is waiting on the lock we
* hold and we cannot drop that as we are in a transaction here.
*
* Lucky for us, this inversion is not a problem because it's a
* directory inode that we are trying to lock here.
*
* So, if we can't get the iolock without sleeping then just give up
*/
if (!xfs_ilock_nowait(pip, XFS_IOLOCK_EXCL))
return 1;
/* If the parent directory is already in the cache, use its AG. */ trace_xfs_filestream_lookup(ip, ag);
item = xfs_mru_cache_lookup(cache, pip->i_ino); goto out;
if (item) {
ASSERT(item->ip == pip);
ag = item->ag;
xfs_mru_cache_done(cache);
TRACE_LOOKUP(mp, pip, pip, ag, xfs_filestream_peek_ag(mp, ag));
err = _xfs_filestream_update_ag(ip, pip, ag);
goto exit;
} }
/* /*
@ -623,202 +341,94 @@ xfs_filestream_associate(
* use the directory inode's AG. * use the directory inode's AG.
*/ */
if (mp->m_flags & XFS_MOUNT_32BITINODES) { if (mp->m_flags & XFS_MOUNT_32BITINODES) {
rotorstep = xfs_rotorstep; xfs_agnumber_t rotorstep = xfs_rotorstep;
startag = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount; startag = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
mp->m_agfrotor = (mp->m_agfrotor + 1) % mp->m_agfrotor = (mp->m_agfrotor + 1) %
(mp->m_sb.sb_agcount * rotorstep); (mp->m_sb.sb_agcount * rotorstep);
} else } else
startag = XFS_INO_TO_AGNO(mp, pip->i_ino); startag = XFS_INO_TO_AGNO(mp, pip->i_ino);
/* Pick a new AG for the parent inode starting at startag. */ if (xfs_filestream_pick_ag(pip, startag, &ag, 0, 0))
err = _xfs_filestream_pick_ag(mp, startag, &ag, 0, 0); ag = NULLAGNUMBER;
if (err || ag == NULLAGNUMBER) out:
goto exit_did_pick; IRELE(pip);
return ag;
/* Associate the parent inode with the AG. */
err = _xfs_filestream_update_ag(pip, NULL, ag);
if (err)
goto exit_did_pick;
/* Associate the file inode with the AG. */
err = _xfs_filestream_update_ag(ip, pip, ag);
if (err)
goto exit_did_pick;
TRACE_ASSOCIATE(mp, ip, pip, ag, xfs_filestream_peek_ag(mp, ag));
exit_did_pick:
/*
* If _xfs_filestream_pick_ag() returned a valid AG, remove the
* reference it took on it, since the file and directory will have taken
* their own now if they were successfully cached.
*/
if (ag != NULLAGNUMBER)
xfs_filestream_put_ag(mp, ag);
exit:
xfs_iunlock(pip, XFS_IOLOCK_EXCL);
return -err;
} }
/* /*
* Pick a new allocation group for the current file and its file stream. This * Pick a new allocation group for the current file and its file stream.
* function is called by xfs_bmap_filestreams() with the mount point's per-ag *
* lock held. * This is called when the allocator can't find a suitable extent in the
* current AG, and we have to move the stream into a new AG with more space.
*/ */
int int
xfs_filestream_new_ag( xfs_filestream_new_ag(
struct xfs_bmalloca *ap, struct xfs_bmalloca *ap,
xfs_agnumber_t *agp) xfs_agnumber_t *agp)
{ {
int flags, err; struct xfs_inode *ip = ap->ip, *pip;
xfs_inode_t *ip, *pip = NULL; struct xfs_mount *mp = ip->i_mount;
xfs_mount_t *mp; xfs_extlen_t minlen = ap->length;
xfs_mru_cache_t *cache; xfs_agnumber_t startag = 0;
xfs_extlen_t minlen; int flags, err = 0;
fstrm_item_t *dir, *file; struct xfs_mru_cache_elem *mru;
xfs_agnumber_t ag = NULLAGNUMBER;
ip = ap->ip;
mp = ip->i_mount;
cache = mp->m_filestream;
minlen = ap->length;
*agp = NULLAGNUMBER; *agp = NULLAGNUMBER;
/* pip = xfs_filestream_get_parent(ip);
* Look for the file in the cache, removing it if it's found. Doing if (!pip)
* this allows it to be held across the dir lookup that follows. goto exit;
*/
file = xfs_mru_cache_remove(cache, ip->i_ino);
if (file) {
ASSERT(ip == file->ip);
/* Save the file's parent inode and old AG number for later. */ mru = xfs_mru_cache_remove(mp->m_filestream, pip->i_ino);
pip = file->pip; if (mru) {
ag = file->ag; struct xfs_fstrm_item *item =
container_of(mru, struct xfs_fstrm_item, mru);
/* Look for the file's directory in the cache. */ startag = (item->ag + 1) % mp->m_sb.sb_agcount;
dir = xfs_mru_cache_lookup(cache, pip->i_ino);
if (dir) {
ASSERT(pip == dir->ip);
/*
* If the directory has already moved on to a new AG,
* use that AG as the new AG for the file. Don't
* forget to twiddle the AG refcounts to match the
* movement.
*/
if (dir->ag != file->ag) {
xfs_filestream_put_ag(mp, file->ag);
xfs_filestream_get_ag(mp, dir->ag);
*agp = file->ag = dir->ag;
}
xfs_mru_cache_done(cache);
}
/*
* Put the file back in the cache. If this fails, the free
* function needs to be called to tidy up in the same way as if
* the item had simply expired from the cache.
*/
err = xfs_mru_cache_insert(cache, ip->i_ino, file);
if (err) {
xfs_fstrm_free_func(ip->i_ino, file);
return err;
}
/*
* If the file's AG was moved to the directory's new AG, there's
* nothing more to be done.
*/
if (*agp != NULLAGNUMBER) {
TRACE_MOVEAG(mp, ip, pip,
ag, xfs_filestream_peek_ag(mp, ag),
*agp, xfs_filestream_peek_ag(mp, *agp));
return 0;
}
} }
/*
* If the file's parent directory is known, take its iolock in exclusive
* mode to prevent two sibling files from racing each other to migrate
* themselves and their parent to different AGs.
*
* Note that we lock the parent directory iolock inside the child
* iolock here. That's fine as we never hold both parent and child
* iolock in any other place. This is different from the ilock,
* which requires locking of the child after the parent for namespace
* operations.
*/
if (pip)
xfs_ilock(pip, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
/*
* A new AG needs to be found for the file. If the file's parent
* directory is also known, it will be moved to the new AG as well to
* ensure that files created inside it in future use the new AG.
*/
ag = (ag == NULLAGNUMBER) ? 0 : (ag + 1) % mp->m_sb.sb_agcount;
flags = (ap->userdata ? XFS_PICK_USERDATA : 0) | flags = (ap->userdata ? XFS_PICK_USERDATA : 0) |
(ap->flist->xbf_low ? XFS_PICK_LOWSPACE : 0); (ap->flist->xbf_low ? XFS_PICK_LOWSPACE : 0);
err = _xfs_filestream_pick_ag(mp, ag, agp, flags, minlen); err = xfs_filestream_pick_ag(pip, startag, agp, flags, minlen);
if (err || *agp == NULLAGNUMBER)
goto exit;
/* /*
* If the file wasn't found in the file cache, then its parent directory * Only free the item here so we skip over the old AG earlier.
* inode isn't known. For this to have happened, the file must either
* be pre-existing, or it was created long enough ago that its cache
* entry has expired. This isn't the sort of usage that the filestreams
* allocator is trying to optimise, so there's no point trying to track
* its new AG somehow in the filestream data structures.
*/ */
if (!pip) { if (mru)
TRACE_ORPHAN(mp, ip, *agp); xfs_fstrm_free_func(mru);
goto exit;
}
/* Associate the parent inode with the AG. */
err = _xfs_filestream_update_ag(pip, NULL, *agp);
if (err)
goto exit;
/* Associate the file inode with the AG. */
err = _xfs_filestream_update_ag(ip, pip, *agp);
if (err)
goto exit;
TRACE_MOVEAG(mp, ip, pip, NULLAGNUMBER, 0,
*agp, xfs_filestream_peek_ag(mp, *agp));
IRELE(pip);
exit: exit:
/* if (*agp == NULLAGNUMBER)
* If _xfs_filestream_pick_ag() returned a valid AG, remove the
* reference it took on it, since the file and directory will have taken
* their own now if they were successfully cached.
*/
if (*agp != NULLAGNUMBER)
xfs_filestream_put_ag(mp, *agp);
else
*agp = 0; *agp = 0;
if (pip)
xfs_iunlock(pip, XFS_IOLOCK_EXCL);
return err; return err;
} }
/*
* Remove an association between an inode and a filestream object.
* Typically this is done on last close of an unlinked file.
*/
void void
xfs_filestream_deassociate( xfs_filestream_deassociate(
xfs_inode_t *ip) struct xfs_inode *ip)
{ {
xfs_mru_cache_t *cache = ip->i_mount->m_filestream; xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino);
}
xfs_mru_cache_delete(cache, ip->i_ino);
int
xfs_filestream_mount(
xfs_mount_t *mp)
{
/*
* The filestream timer tunable is currently fixed within the range of
* one second to four minutes, with five seconds being the default. The
* group count is somewhat arbitrary, but it'd be nice to adhere to the
* timer tunable to within about 10 percent. This requires at least 10
* groups.
*/
return xfs_mru_cache_create(&mp->m_filestream, xfs_fstrm_centisecs * 10,
10, xfs_fstrm_free_func);
}
void
xfs_filestream_unmount(
xfs_mount_t *mp)
{
xfs_mru_cache_destroy(mp->m_filestream);
} }

34
fs/xfs/xfs_filestream.h
View file

@ -20,50 +20,20 @@
struct xfs_mount; struct xfs_mount;
struct xfs_inode; struct xfs_inode;
struct xfs_perag;
struct xfs_bmalloca; struct xfs_bmalloca;
#ifdef XFS_FILESTREAMS_TRACE
#define XFS_FSTRM_KTRACE_INFO 1
#define XFS_FSTRM_KTRACE_AGSCAN 2
#define XFS_FSTRM_KTRACE_AGPICK1 3
#define XFS_FSTRM_KTRACE_AGPICK2 4
#define XFS_FSTRM_KTRACE_UPDATE 5
#define XFS_FSTRM_KTRACE_FREE 6
#define XFS_FSTRM_KTRACE_ITEM_LOOKUP 7
#define XFS_FSTRM_KTRACE_ASSOCIATE 8
#define XFS_FSTRM_KTRACE_MOVEAG 9
#define XFS_FSTRM_KTRACE_ORPHAN 10
#define XFS_FSTRM_KTRACE_SIZE 16384
extern ktrace_t *xfs_filestreams_trace_buf;
#endif
/* allocation selection flags */
typedef enum xfs_fstrm_alloc {
XFS_PICK_USERDATA = 1,
XFS_PICK_LOWSPACE = 2,
} xfs_fstrm_alloc_t;
/* prototypes for filestream.c */
int xfs_filestream_init(void);
void xfs_filestream_uninit(void);
int xfs_filestream_mount(struct xfs_mount *mp); int xfs_filestream_mount(struct xfs_mount *mp);
void xfs_filestream_unmount(struct xfs_mount *mp); void xfs_filestream_unmount(struct xfs_mount *mp);
xfs_agnumber_t xfs_filestream_lookup_ag(struct xfs_inode *ip);
int xfs_filestream_associate(struct xfs_inode *dip, struct xfs_inode *ip);
void xfs_filestream_deassociate(struct xfs_inode *ip); void xfs_filestream_deassociate(struct xfs_inode *ip);
xfs_agnumber_t xfs_filestream_lookup_ag(struct xfs_inode *ip);
int xfs_filestream_new_ag(struct xfs_bmalloca *ap, xfs_agnumber_t *agp); int xfs_filestream_new_ag(struct xfs_bmalloca *ap, xfs_agnumber_t *agp);
int xfs_filestream_peek_ag(struct xfs_mount *mp, xfs_agnumber_t agno);
/* filestreams for the inode? */
static inline int static inline int
xfs_inode_is_filestream( xfs_inode_is_filestream(
struct xfs_inode *ip) struct xfs_inode *ip)
{ {
return (ip->i_mount->m_flags & XFS_MOUNT_FILESTREAMS) || return (ip->i_mount->m_flags & XFS_MOUNT_FILESTREAMS) ||
xfs_iflags_test(ip, XFS_IFILESTREAM) ||
(ip->i_d.di_flags & XFS_DIFLAG_FILESTREAM); (ip->i_d.di_flags & XFS_DIFLAG_FILESTREAM);
} }

35
fs/xfs/xfs_inode.c
View file

@ -655,7 +655,6 @@ xfs_ialloc(
uint flags; uint flags;
int error; int error;
timespec_t tv; timespec_t tv;
int filestreams = 0;
/* /*
* Call the space management code to pick * Call the space management code to pick
@ -772,13 +771,6 @@ xfs_ialloc(
flags |= XFS_ILOG_DEV; flags |= XFS_ILOG_DEV;
break; break;
case S_IFREG: case S_IFREG:
/*
* we can't set up filestreams until after the VFS inode
* is set up properly.
*/
if (pip && xfs_inode_is_filestream(pip))
filestreams = 1;
/* fall through */
case S_IFDIR: case S_IFDIR:
if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
uint di_flags = 0; uint di_flags = 0;
@ -844,15 +836,6 @@ xfs_ialloc(
/* now that we have an i_mode we can setup inode ops and unlock */ /* now that we have an i_mode we can setup inode ops and unlock */
xfs_setup_inode(ip); xfs_setup_inode(ip);
/* now we have set up the vfs inode we can associate the filestream */
if (filestreams) {
error = xfs_filestream_associate(pip, ip);
if (error < 0)
return -error;
if (!error)
xfs_iflags_set(ip, XFS_IFILESTREAM);
}
*ipp = ip; *ipp = ip;
return 0; return 0;
} }
@ -1698,16 +1681,6 @@ xfs_release(
if (!XFS_FORCED_SHUTDOWN(mp)) { if (!XFS_FORCED_SHUTDOWN(mp)) {
int truncated; int truncated;
/*
* 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 * If we previously truncated this file and removed old data
* in the process, we want to initiate "early" writeout on * in the process, we want to initiate "early" writeout on
@ -2664,13 +2637,7 @@ xfs_remove(
if (error) if (error)
goto std_return; goto std_return;
/* if (is_dir && xfs_inode_is_filestream(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 (!is_dir && link_zero && xfs_inode_is_filestream(ip))
xfs_filestream_deassociate(ip); xfs_filestream_deassociate(ip);
return 0; return 0;

4
fs/xfs/xfs_inode.h
View file

@ -209,7 +209,6 @@ xfs_get_initial_prid(struct xfs_inode *dp)
#define XFS_ISTALE (1 << 1) /* inode has been staled */ #define XFS_ISTALE (1 << 1) /* inode has been staled */
#define XFS_IRECLAIMABLE (1 << 2) /* inode can be reclaimed */ #define XFS_IRECLAIMABLE (1 << 2) /* inode can be reclaimed */
#define XFS_INEW (1 << 3) /* inode has just been allocated */ #define XFS_INEW (1 << 3) /* inode has just been allocated */
#define XFS_IFILESTREAM (1 << 4) /* inode is in a filestream dir. */
#define XFS_ITRUNCATED (1 << 5) /* truncated down so flush-on-close */ #define XFS_ITRUNCATED (1 << 5) /* truncated down so flush-on-close */
#define XFS_IDIRTY_RELEASE (1 << 6) /* dirty release already seen */ #define XFS_IDIRTY_RELEASE (1 << 6) /* dirty release already seen */
#define __XFS_IFLOCK_BIT 7 /* inode is being flushed right now */ #define __XFS_IFLOCK_BIT 7 /* inode is being flushed right now */
@ -225,8 +224,7 @@ xfs_get_initial_prid(struct xfs_inode *dp)
*/ */
#define XFS_IRECLAIM_RESET_FLAGS \ #define XFS_IRECLAIM_RESET_FLAGS \
(XFS_IRECLAIMABLE | XFS_IRECLAIM | \ (XFS_IRECLAIMABLE | XFS_IRECLAIM | \
XFS_IDIRTY_RELEASE | XFS_ITRUNCATED | \ XFS_IDIRTY_RELEASE | XFS_ITRUNCATED)
XFS_IFILESTREAM);
/* /*
* Synchronize processes attempting to flush the in-core inode back to disk. * Synchronize processes attempting to flush the in-core inode back to disk.

151
fs/xfs/xfs_mru_cache.c
View file

@ -100,14 +100,20 @@
* likely result in a loop in one of the lists. That's a sure-fire recipe for * likely result in a loop in one of the lists. That's a sure-fire recipe for
* an infinite loop in the code. * an infinite loop in the code.
*/ */
typedef struct xfs_mru_cache_elem struct xfs_mru_cache {
{ struct radix_tree_root store; /* Core storage data structure. */
struct list_head list_node; struct list_head *lists; /* Array of lists, one per grp. */
unsigned long key; struct list_head reap_list; /* Elements overdue for reaping. */
void *value; spinlock_t lock; /* Lock to protect this struct. */
} xfs_mru_cache_elem_t; unsigned int grp_count; /* Number of discrete groups. */
unsigned int grp_time; /* Time period spanned by grps. */
unsigned int lru_grp; /* Group containing time zero. */
unsigned long time_zero; /* Time first element was added. */
xfs_mru_cache_free_func_t free_func; /* Function pointer for freeing. */
struct delayed_work work; /* Workqueue data for reaping. */
unsigned int queued; /* work has been queued */
};
static kmem_zone_t *xfs_mru_elem_zone;
static struct workqueue_struct *xfs_mru_reap_wq; static struct workqueue_struct *xfs_mru_reap_wq;
/* /*
@ -129,12 +135,12 @@ static struct workqueue_struct *xfs_mru_reap_wq;
*/ */
STATIC unsigned long STATIC unsigned long
_xfs_mru_cache_migrate( _xfs_mru_cache_migrate(
xfs_mru_cache_t *mru, struct xfs_mru_cache *mru,
unsigned long now) unsigned long now)
{ {
unsigned int grp; unsigned int grp;
unsigned int migrated = 0; unsigned int migrated = 0;
struct list_head *lru_list; struct list_head *lru_list;
/* Nothing to do if the data store is empty. */ /* Nothing to do if the data store is empty. */
if (!mru->time_zero) if (!mru->time_zero)
@ -193,11 +199,11 @@ _xfs_mru_cache_migrate(
*/ */
STATIC void STATIC void
_xfs_mru_cache_list_insert( _xfs_mru_cache_list_insert(
xfs_mru_cache_t *mru, struct xfs_mru_cache *mru,
xfs_mru_cache_elem_t *elem) struct xfs_mru_cache_elem *elem)
{ {
unsigned int grp = 0; unsigned int grp = 0;
unsigned long now = jiffies; unsigned long now = jiffies;
/* /*
* If the data store is empty, initialise time zero, leave grp set to * If the data store is empty, initialise time zero, leave grp set to
@ -231,10 +237,10 @@ _xfs_mru_cache_list_insert(
*/ */
STATIC void STATIC void
_xfs_mru_cache_clear_reap_list( _xfs_mru_cache_clear_reap_list(
xfs_mru_cache_t *mru) __releases(mru->lock) __acquires(mru->lock) struct xfs_mru_cache *mru)
__releases(mru->lock) __acquires(mru->lock)
{ {
xfs_mru_cache_elem_t *elem, *next; struct xfs_mru_cache_elem *elem, *next;
struct list_head tmp; struct list_head tmp;
INIT_LIST_HEAD(&tmp); INIT_LIST_HEAD(&tmp);
@ -252,15 +258,8 @@ _xfs_mru_cache_clear_reap_list(
spin_unlock(&mru->lock); spin_unlock(&mru->lock);
list_for_each_entry_safe(elem, next, &tmp, list_node) { list_for_each_entry_safe(elem, next, &tmp, list_node) {
/* Remove the element from the reap list. */
list_del_init(&elem->list_node); list_del_init(&elem->list_node);
mru->free_func(elem);
/* Call the client's free function with the key and value pointer. */
mru->free_func(elem->key, elem->value);
/* Free the element structure. */
kmem_zone_free(xfs_mru_elem_zone, elem);
} }
spin_lock(&mru->lock); spin_lock(&mru->lock);
@ -277,7 +276,8 @@ STATIC void
_xfs_mru_cache_reap( _xfs_mru_cache_reap(
struct work_struct *work) struct work_struct *work)
{ {
xfs_mru_cache_t *mru = container_of(work, xfs_mru_cache_t, work.work); struct xfs_mru_cache *mru =
container_of(work, struct xfs_mru_cache, work.work);
unsigned long now, next; unsigned long now, next;
ASSERT(mru && mru->lists); ASSERT(mru && mru->lists);
@ -304,28 +304,16 @@ _xfs_mru_cache_reap(
int int
xfs_mru_cache_init(void) xfs_mru_cache_init(void)
{ {
xfs_mru_elem_zone = kmem_zone_init(sizeof(xfs_mru_cache_elem_t),
"xfs_mru_cache_elem");
if (!xfs_mru_elem_zone)
goto out;
xfs_mru_reap_wq = alloc_workqueue("xfs_mru_cache", WQ_MEM_RECLAIM, 1); xfs_mru_reap_wq = alloc_workqueue("xfs_mru_cache", WQ_MEM_RECLAIM, 1);
if (!xfs_mru_reap_wq) if (!xfs_mru_reap_wq)
goto out_destroy_mru_elem_zone; return -ENOMEM;
return 0; return 0;
out_destroy_mru_elem_zone:
kmem_zone_destroy(xfs_mru_elem_zone);
out:
return -ENOMEM;
} }
void void
xfs_mru_cache_uninit(void) xfs_mru_cache_uninit(void)
{ {
destroy_workqueue(xfs_mru_reap_wq); destroy_workqueue(xfs_mru_reap_wq);
kmem_zone_destroy(xfs_mru_elem_zone);
} }
/* /*
@ -336,14 +324,14 @@ xfs_mru_cache_uninit(void)
*/ */
int int
xfs_mru_cache_create( xfs_mru_cache_create(
xfs_mru_cache_t **mrup, struct xfs_mru_cache **mrup,
unsigned int lifetime_ms, unsigned int lifetime_ms,
unsigned int grp_count, unsigned int grp_count,
xfs_mru_cache_free_func_t free_func) xfs_mru_cache_free_func_t free_func)
{ {
xfs_mru_cache_t *mru = NULL; struct xfs_mru_cache *mru = NULL;
int err = 0, grp; int err = 0, grp;
unsigned int grp_time; unsigned int grp_time;
if (mrup) if (mrup)
*mrup = NULL; *mrup = NULL;
@ -400,7 +388,7 @@ xfs_mru_cache_create(
*/ */
static void static void
xfs_mru_cache_flush( xfs_mru_cache_flush(
xfs_mru_cache_t *mru) struct xfs_mru_cache *mru)
{ {
if (!mru || !mru->lists) if (!mru || !mru->lists)
return; return;
@ -420,7 +408,7 @@ xfs_mru_cache_flush(
void void
xfs_mru_cache_destroy( xfs_mru_cache_destroy(
xfs_mru_cache_t *mru) struct xfs_mru_cache *mru)
{ {
if (!mru || !mru->lists) if (!mru || !mru->lists)
return; return;
@ -438,38 +426,30 @@ xfs_mru_cache_destroy(
*/ */
int int
xfs_mru_cache_insert( xfs_mru_cache_insert(
xfs_mru_cache_t *mru, struct xfs_mru_cache *mru,
unsigned long key, unsigned long key,
void *value) struct xfs_mru_cache_elem *elem)
{ {
xfs_mru_cache_elem_t *elem; int error;
ASSERT(mru && mru->lists); ASSERT(mru && mru->lists);
if (!mru || !mru->lists) if (!mru || !mru->lists)
return EINVAL; return EINVAL;
elem = kmem_zone_zalloc(xfs_mru_elem_zone, KM_SLEEP); if (radix_tree_preload(GFP_KERNEL))
if (!elem)
return ENOMEM; return ENOMEM;
if (radix_tree_preload(GFP_KERNEL)) {
kmem_zone_free(xfs_mru_elem_zone, elem);
return ENOMEM;
}
INIT_LIST_HEAD(&elem->list_node); INIT_LIST_HEAD(&elem->list_node);
elem->key = key; elem->key = key;
elem->value = value;
spin_lock(&mru->lock); spin_lock(&mru->lock);
error = -radix_tree_insert(&mru->store, key, elem);
radix_tree_insert(&mru->store, key, elem);
radix_tree_preload_end(); radix_tree_preload_end();
_xfs_mru_cache_list_insert(mru, elem); if (!error)
_xfs_mru_cache_list_insert(mru, elem);
spin_unlock(&mru->lock); spin_unlock(&mru->lock);
return 0; return error;
} }
/* /*
@ -478,13 +458,12 @@ xfs_mru_cache_insert(
* the client data pointer for the removed element is returned, otherwise this * the client data pointer for the removed element is returned, otherwise this
* function will return a NULL pointer. * function will return a NULL pointer.
*/ */
void * struct xfs_mru_cache_elem *
xfs_mru_cache_remove( xfs_mru_cache_remove(
xfs_mru_cache_t *mru, struct xfs_mru_cache *mru,
unsigned long key) unsigned long key)
{ {
xfs_mru_cache_elem_t *elem; struct xfs_mru_cache_elem *elem;
void *value = NULL;
ASSERT(mru && mru->lists); ASSERT(mru && mru->lists);
if (!mru || !mru->lists) if (!mru || !mru->lists)
@ -492,17 +471,11 @@ xfs_mru_cache_remove(
spin_lock(&mru->lock); spin_lock(&mru->lock);
elem = radix_tree_delete(&mru->store, key); elem = radix_tree_delete(&mru->store, key);
if (elem) { if (elem)
value = elem->value;
list_del(&elem->list_node); list_del(&elem->list_node);
}
spin_unlock(&mru->lock); spin_unlock(&mru->lock);
if (elem) return elem;
kmem_zone_free(xfs_mru_elem_zone, elem);
return value;
} }
/* /*
@ -511,13 +484,14 @@ xfs_mru_cache_remove(
*/ */
void void
xfs_mru_cache_delete( xfs_mru_cache_delete(
xfs_mru_cache_t *mru, struct xfs_mru_cache *mru,
unsigned long key) unsigned long key)
{ {
void *value = xfs_mru_cache_remove(mru, key); struct xfs_mru_cache_elem *elem;
if (value) elem = xfs_mru_cache_remove(mru, key);
mru->free_func(key, value); if (elem)
mru->free_func(elem);
} }
/* /*
@ -540,12 +514,12 @@ xfs_mru_cache_delete(
* status, we need to help it get it right by annotating the path that does * status, we need to help it get it right by annotating the path that does
* not release the lock. * not release the lock.
*/ */
void * struct xfs_mru_cache_elem *
xfs_mru_cache_lookup( xfs_mru_cache_lookup(
xfs_mru_cache_t *mru, struct xfs_mru_cache *mru,
unsigned long key) unsigned long key)
{ {
xfs_mru_cache_elem_t *elem; struct xfs_mru_cache_elem *elem;
ASSERT(mru && mru->lists); ASSERT(mru && mru->lists);
if (!mru || !mru->lists) if (!mru || !mru->lists)
@ -560,7 +534,7 @@ xfs_mru_cache_lookup(
} else } else
spin_unlock(&mru->lock); spin_unlock(&mru->lock);
return elem ? elem->value : NULL; return elem;
} }
/* /*
@ -570,7 +544,8 @@ xfs_mru_cache_lookup(
*/ */
void void
xfs_mru_cache_done( xfs_mru_cache_done(
xfs_mru_cache_t *mru) __releases(mru->lock) struct xfs_mru_cache *mru)
__releases(mru->lock)
{ {
spin_unlock(&mru->lock); spin_unlock(&mru->lock);
} }

31
fs/xfs/xfs_mru_cache.h
View file

@ -18,24 +18,15 @@
#ifndef __XFS_MRU_CACHE_H__ #ifndef __XFS_MRU_CACHE_H__
#define __XFS_MRU_CACHE_H__ #define __XFS_MRU_CACHE_H__
struct xfs_mru_cache;
struct xfs_mru_cache_elem {
struct list_head list_node;
unsigned long key;
};
/* Function pointer type for callback to free a client's data pointer. */ /* Function pointer type for callback to free a client's data pointer. */
typedef void (*xfs_mru_cache_free_func_t)(unsigned long, void*); typedef void (*xfs_mru_cache_free_func_t)(struct xfs_mru_cache_elem *elem);
typedef struct xfs_mru_cache
{
struct radix_tree_root store; /* Core storage data structure. */
struct list_head *lists; /* Array of lists, one per grp. */
struct list_head reap_list; /* Elements overdue for reaping. */
spinlock_t lock; /* Lock to protect this struct. */
unsigned int grp_count; /* Number of discrete groups. */
unsigned int grp_time; /* Time period spanned by grps. */
unsigned int lru_grp; /* Group containing time zero. */
unsigned long time_zero; /* Time first element was added. */
xfs_mru_cache_free_func_t free_func; /* Function pointer for freeing. */
struct delayed_work work; /* Workqueue data for reaping. */
unsigned int queued; /* work has been queued */
} xfs_mru_cache_t;
int xfs_mru_cache_init(void); int xfs_mru_cache_init(void);
void xfs_mru_cache_uninit(void); void xfs_mru_cache_uninit(void);
@ -44,10 +35,12 @@ int xfs_mru_cache_create(struct xfs_mru_cache **mrup, unsigned int lifetime_ms,
xfs_mru_cache_free_func_t free_func); xfs_mru_cache_free_func_t free_func);
void xfs_mru_cache_destroy(struct xfs_mru_cache *mru); void xfs_mru_cache_destroy(struct xfs_mru_cache *mru);
int xfs_mru_cache_insert(struct xfs_mru_cache *mru, unsigned long key, int xfs_mru_cache_insert(struct xfs_mru_cache *mru, unsigned long key,
void *value); struct xfs_mru_cache_elem *elem);
void * xfs_mru_cache_remove(struct xfs_mru_cache *mru, unsigned long key); struct xfs_mru_cache_elem *
xfs_mru_cache_remove(struct xfs_mru_cache *mru, unsigned long key);
void xfs_mru_cache_delete(struct xfs_mru_cache *mru, unsigned long key); void xfs_mru_cache_delete(struct xfs_mru_cache *mru, unsigned long key);
void *xfs_mru_cache_lookup(struct xfs_mru_cache *mru, unsigned long key); struct xfs_mru_cache_elem *
xfs_mru_cache_lookup(struct xfs_mru_cache *mru, unsigned long key);
void xfs_mru_cache_done(struct xfs_mru_cache *mru); void xfs_mru_cache_done(struct xfs_mru_cache *mru);
#endif /* __XFS_MRU_CACHE_H__ */ #endif /* __XFS_MRU_CACHE_H__ */

9
fs/xfs/xfs_super.c
View file

@ -1749,13 +1749,9 @@ init_xfs_fs(void)
if (error) if (error)
goto out_destroy_wq; goto out_destroy_wq;
error = xfs_filestream_init();
if (error)
goto out_mru_cache_uninit;
error = xfs_buf_init(); error = xfs_buf_init();
if (error) if (error)
goto out_filestream_uninit; goto out_mru_cache_uninit;
error = xfs_init_procfs(); error = xfs_init_procfs();
if (error) if (error)
@ -1782,8 +1778,6 @@ init_xfs_fs(void)
xfs_cleanup_procfs(); xfs_cleanup_procfs();
out_buf_terminate: out_buf_terminate:
xfs_buf_terminate(); xfs_buf_terminate();
out_filestream_uninit:
xfs_filestream_uninit();
out_mru_cache_uninit: out_mru_cache_uninit:
xfs_mru_cache_uninit(); xfs_mru_cache_uninit();
out_destroy_wq: out_destroy_wq:
@ -1802,7 +1796,6 @@ exit_xfs_fs(void)
xfs_sysctl_unregister(); xfs_sysctl_unregister();
xfs_cleanup_procfs(); xfs_cleanup_procfs();
xfs_buf_terminate(); xfs_buf_terminate();
xfs_filestream_uninit();
xfs_mru_cache_uninit(); xfs_mru_cache_uninit();
xfs_destroy_workqueues(); xfs_destroy_workqueues();
xfs_destroy_zones(); xfs_destroy_zones();

1
fs/xfs/xfs_trace.c
View file

@ -46,6 +46,7 @@
#include "xfs_log_recover.h" #include "xfs_log_recover.h"
#include "xfs_inode_item.h" #include "xfs_inode_item.h"
#include "xfs_bmap_btree.h" #include "xfs_bmap_btree.h"
#include "xfs_filestream.h"
/* /*
* We include this last to have the helpers above available for the trace * We include this last to have the helpers above available for the trace

58
fs/xfs/xfs_trace.h
View file

@ -538,6 +538,64 @@ DEFINE_BUF_ITEM_EVENT(xfs_trans_bhold_release);
DEFINE_BUF_ITEM_EVENT(xfs_trans_binval); DEFINE_BUF_ITEM_EVENT(xfs_trans_binval);
DEFINE_BUF_ITEM_EVENT(xfs_trans_buf_ordered); DEFINE_BUF_ITEM_EVENT(xfs_trans_buf_ordered);
DECLARE_EVENT_CLASS(xfs_filestream_class,
TP_PROTO(struct xfs_inode *ip, xfs_agnumber_t agno),
TP_ARGS(ip, agno),
TP_STRUCT__entry(
__field(dev_t, dev)
__field(xfs_ino_t, ino)
__field(xfs_agnumber_t, agno)
__field(int, streams)
),
TP_fast_assign(
__entry->dev = VFS_I(ip)->i_sb->s_dev;
__entry->ino = ip->i_ino;
__entry->agno = agno;
__entry->streams = xfs_filestream_peek_ag(ip->i_mount, agno);
),
TP_printk("dev %d:%d ino 0x%llx agno %u streams %d",
MAJOR(__entry->dev), MINOR(__entry->dev),
__entry->ino,
__entry->agno,
__entry->streams)
)
#define DEFINE_FILESTREAM_EVENT(name) \
DEFINE_EVENT(xfs_filestream_class, name, \
TP_PROTO(struct xfs_inode *ip, xfs_agnumber_t agno), \
TP_ARGS(ip, agno))
DEFINE_FILESTREAM_EVENT(xfs_filestream_free);
DEFINE_FILESTREAM_EVENT(xfs_filestream_lookup);
DEFINE_FILESTREAM_EVENT(xfs_filestream_scan);
TRACE_EVENT(xfs_filestream_pick,
TP_PROTO(struct xfs_inode *ip, xfs_agnumber_t agno,
xfs_extlen_t free, int nscan),
TP_ARGS(ip, agno, free, nscan),
TP_STRUCT__entry(
__field(dev_t, dev)
__field(xfs_ino_t, ino)
__field(xfs_agnumber_t, agno)
__field(int, streams)
__field(xfs_extlen_t, free)
__field(int, nscan)
),
TP_fast_assign(
__entry->dev = VFS_I(ip)->i_sb->s_dev;
__entry->ino = ip->i_ino;
__entry->agno = agno;
__entry->streams = xfs_filestream_peek_ag(ip->i_mount, agno);
__entry->free = free;
__entry->nscan = nscan;
),
TP_printk("dev %d:%d ino 0x%llx agno %u streams %d free %d nscan %d",
MAJOR(__entry->dev), MINOR(__entry->dev),
__entry->ino,
__entry->agno,
__entry->streams,
__entry->free,
__entry->nscan)
);
DECLARE_EVENT_CLASS(xfs_lock_class, DECLARE_EVENT_CLASS(xfs_lock_class,
TP_PROTO(struct xfs_inode *ip, unsigned lock_flags, TP_PROTO(struct xfs_inode *ip, unsigned lock_flags,
unsigned long caller_ip), unsigned long caller_ip),