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 && (((ip->i_d.di_mode & S_IFREG) && pip &&
(pip->i_d.di_mode & S_IFDIR)) ||
((ip->i_d.di_mode & S_IFDIR) && !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;
}
item = kmem_zone_zalloc(item_zone, KM_MAYFAIL);
if (!item)
return ENOMEM;
item->ag = ag;
item->ip = ip;
item->pip = pip;
err = xfs_mru_cache_insert(cache, ip->i_ino, item);
if (err) {
kmem_zone_free(item_zone, item);
return err;
}
/* 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;
}
int
xfs_filestream_new_ag(
xfs_bmalloca_t *ap,
xfs_agnumber_t *agp)
{
int flags, err;
xfs_inode_t *ip, *pip = NULL;
xfs_mount_t *mp;
xfs_mru_cache_t *cache;
xfs_extlen_t minlen;
fstrm_item_t *dir, *file;
xfs_agnumber_t ag = NULLAGNUMBER;
ip = ap->ip;
mp = ip->i_mount;
cache = mp->m_filestream;
minlen = ap->alen;
*agp = NULLAGNUMBER;
/*
* Look for the file in the cache, removing it if it's found. Doing
* this allows it to be held across the dir lookup that follows.
*/
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. */
pip = file->pip;
ag = file->ag;
/* Look for the file's directory in the cache. */
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.
*/
if (pip)
xfs_ilock(pip, XFS_IOLOCK_EXCL);
/*
* 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) |
(ap->low ? XFS_PICK_LOWSPACE : 0);
err = _xfs_filestream_pick_ag(mp, ag, agp, flags, minlen);
if (err || *agp == NULLAGNUMBER)
goto exit;
/*
* If the file wasn't found in the file cache, then its parent directory
* 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) {
TRACE_ORPHAN(mp, ip, *agp);
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));
exit:
/*
* 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;
if (pip)
xfs_iunlock(pip, XFS_IOLOCK_EXCL);
return err;
}
static int
_xfs_filestream_pick_ag(
xfs_mount_t *mp,
xfs_agnumber_t startag,
xfs_agnumber_t *agp,
int flags,
xfs_extlen_t minlen)
{
int streams, max_streams;
int err, trylock, nscan;
xfs_extlen_t longest, free, minfree, maxfree = 0;
xfs_agnumber_t ag, max_ag = NULLAGNUMBER;
struct xfs_perag *pag;
/* 2% of an AG's blocks must be free for it to be chosen. */
minfree = mp->m_sb.sb_agblocks / 50;
ag = startag;
*agp = NULLAGNUMBER;
/* For the first pass, don't sleep trying to init the per-AG. */
trylock = XFS_ALLOC_FLAG_TRYLOCK;
for (nscan = 0; 1; nscan++) {
pag = xfs_perag_get(mp, ag);
TRACE_AG_SCAN(mp, ag, atomic_read(&pag->pagf_fstrms));
if (!pag->pagf_init) {
err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
if (err && !trylock) {
xfs_perag_put(pag);
return err;
}
}
/* Might fail sometimes during the 1st pass with trylock set. */
if (!pag->pagf_init)
goto next_ag;
/* Keep track of the AG with the most free blocks. */
if (pag->pagf_freeblks > maxfree) {
maxfree = pag->pagf_freeblks;
max_streams = atomic_read(&pag->pagf_fstrms);
max_ag = ag;
}
/*
* The AG reference count does two things: it enforces mutual
* exclusion when examining the suitability of an AG in this
* loop, and it guards against two filestreams being established
* in the same AG as each other.
*/
if (xfs_filestream_get_ag(mp, ag) > 1) {
xfs_filestream_put_ag(mp, ag);
goto next_ag;
}
longest = xfs_alloc_longest_free_extent(mp, pag);
if (((minlen && longest >= minlen) ||
(!minlen && pag->pagf_freeblks >= minfree)) &&
(!pag->pagf_metadata || !(flags & XFS_PICK_USERDATA) ||
(flags & XFS_PICK_LOWSPACE))) {
/* Break out, retaining the reference on the AG. */
free = pag->pagf_freeblks;
streams = atomic_read(&pag->pagf_fstrms);
xfs_perag_put(pag);
*agp = ag;
break;
}
/* Drop the reference on this AG, it's not usable. */
xfs_filestream_put_ag(mp, ag);
next_ag:
xfs_perag_put(pag);
/* Move to the next AG, wrapping to AG 0 if necessary. */
if (++ag >= mp->m_sb.sb_agcount)
ag = 0;
/* If a full pass of the AGs hasn't been done yet, continue. */
if (ag != startag)
continue;
/* Allow sleeping in xfs_alloc_pagf_init() on the 2nd pass. */
if (trylock != 0) {
trylock = 0;
continue;
}
/* Finally, if lowspace wasn't set, set it for the 3rd pass. */
if (!(flags & XFS_PICK_LOWSPACE)) {
flags |= XFS_PICK_LOWSPACE;
continue;
}
/*
* Take the AG with the most free space, regardless of whether
* it's already in use by another filestream.
*/
if (max_ag != NULLAGNUMBER) {
xfs_filestream_get_ag(mp, max_ag);
TRACE_AG_PICK1(mp, max_ag, maxfree);
streams = max_streams;
free = maxfree;
*agp = max_ag;
break;
}
/* take AG 0 if none matched */
TRACE_AG_PICK1(mp, max_ag, maxfree);
*agp = 0;
return 0;
}
TRACE_AG_PICK2(mp, startag, *agp, streams, free, nscan, flags);
return 0;
}
/*
* Scan the AGs starting at startag looking for an AG that isn't in use and has
* at least minlen blocks free.
*/
static int
xfs_filestream_pick_ag(
struct xfs_inode *ip,
xfs_agnumber_t startag,
xfs_agnumber_t *agp,
int flags,
xfs_extlen_t minlen)
{
struct xfs_mount *mp = ip->i_mount;
struct xfs_fstrm_item *item;
struct xfs_perag *pag;
xfs_extlen_t longest, free = 0, minfree, maxfree = 0;
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. */
minfree = mp->m_sb.sb_agblocks / 50;
ag = startag;
*agp = NULLAGNUMBER;
/* For the first pass, don't sleep trying to init the per-AG. */
trylock = XFS_ALLOC_FLAG_TRYLOCK;
for (nscan = 0; 1; nscan++) {
trace_xfs_filestream_scan(ip, ag);
pag = xfs_perag_get(mp, ag);
if (!pag->pagf_init) {
err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
if (err && !trylock) {
xfs_perag_put(pag);
return err;
}
}
/* Might fail sometimes during the 1st pass with trylock set. */
if (!pag->pagf_init)
goto next_ag;
/* Keep track of the AG with the most free blocks. */
if (pag->pagf_freeblks > maxfree) {
maxfree = pag->pagf_freeblks;
max_ag = ag;
}
/*
* The AG reference count does two things: it enforces mutual
* exclusion when examining the suitability of an AG in this
* loop, and it guards against two filestreams being established
* in the same AG as each other.
*/
if (xfs_filestream_get_ag(mp, ag) > 1) {
xfs_filestream_put_ag(mp, ag);
goto next_ag;
}
longest = xfs_alloc_longest_free_extent(mp, pag);
if (((minlen && longest >= minlen) ||
(!minlen && pag->pagf_freeblks >= minfree)) &&
(!pag->pagf_metadata || !(flags & XFS_PICK_USERDATA) ||
(flags & XFS_PICK_LOWSPACE))) {
/* Break out, retaining the reference on the AG. */
free = pag->pagf_freeblks;
xfs_perag_put(pag);
*agp = ag;
break;
}
/* Drop the reference on this AG, it's not usable. */
xfs_filestream_put_ag(mp, ag);
next_ag:
xfs_perag_put(pag);
/* Move to the next AG, wrapping to AG 0 if necessary. */
if (++ag >= mp->m_sb.sb_agcount)
ag = 0;
/* If a full pass of the AGs hasn't been done yet, continue. */
if (ag != startag)
continue;
/* Allow sleeping in xfs_alloc_pagf_init() on the 2nd pass. */
if (trylock != 0) {
trylock = 0;
continue;
}
/* Finally, if lowspace wasn't set, set it for the 3rd pass. */
if (!(flags & XFS_PICK_LOWSPACE)) {
flags |= XFS_PICK_LOWSPACE;
continue;
}
/*
* Take the AG with the most free space, regardless of whether
* it's already in use by another filestream.
*/
if (max_ag != NULLAGNUMBER) {
xfs_filestream_get_ag(mp, max_ag);
free = maxfree;
*agp = max_ag;
break;
}
/* take AG 0 if none matched */
trace_xfs_filestream_pick(ip, *agp, free, nscan);
*agp = 0;
return 0;
}
trace_xfs_filestream_pick(ip, *agp, free, nscan);
if (*agp == NULLAGNUMBER)
return 0;
err = ENOMEM;
item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
if (!item)
goto out_put_ag;
item->ag = *agp;
item->ip = ip;
err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);
if (err) {
if (err == EEXIST)
err = 0;
goto out_free_item;
}
return 0;
out_free_item:
kmem_free(item);
out_put_ag:
xfs_filestream_put_ag(mp, *agp);
return err;
}
int
xfs_filestream_new_ag(
xfs_bmalloca_t *ap,
xfs_agnumber_t *agp)
{
int flags, err;
xfs_inode_t *ip, *pip = NULL;
xfs_mount_t *mp;
xfs_mru_cache_t *cache;
xfs_extlen_t minlen;
fstrm_item_t *dir, *file;
xfs_agnumber_t ag = NULLAGNUMBER;
ip = ap->ip;
mp = ip->i_mount;
cache = mp->m_filestream;
minlen = ap->length;
*agp = NULLAGNUMBER;
file = xfs_mru_cache_remove(cache, ip->i_ino);
if (file) {
ASSERT(ip == file->ip);
pip = file->pip;
ag = file->ag;
dir = xfs_mru_cache_lookup(cache, pip->i_ino);
if (dir) {
ASSERT(pip == dir->ip);
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);
}
err = xfs_mru_cache_insert(cache, ip->i_ino, file);
if (err) {
xfs_fstrm_free_func(ip->i_ino, file);
return err;
}
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 (pip)
xfs_ilock(pip, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
ag = (ag == NULLAGNUMBER) ? 0 : (ag + 1) % mp->m_sb.sb_agcount;
flags = (ap->userdata ? XFS_PICK_USERDATA : 0) |
(ap->flist->xbf_low ? XFS_PICK_LOWSPACE : 0);
err = _xfs_filestream_pick_ag(mp, ag, agp, flags, minlen);
if (err || *agp == NULLAGNUMBER)
goto exit;
if (!pip) {
TRACE_ORPHAN(mp, ip, *agp);
goto exit;
}
err = _xfs_filestream_update_ag(pip, NULL, *agp);
if (err)
goto exit;
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));
exit:
if (*agp != NULLAGNUMBER)
xfs_filestream_put_ag(mp, *agp);
else
*agp = 0;
if (pip)
xfs_iunlock(pip, XFS_IOLOCK_EXCL);
return err;
}
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;
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 (ag != old_ag) {
xfs_filestream_put_ag(mp, old_ag);
xfs_filestream_get_ag(mp, ag);
}
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;
}
item = kmem_zone_zalloc(item_zone, KM_MAYFAIL);
if (!item)
return ENOMEM;
item->ag = ag;
item->ip = ip;
item->pip = pip;
err = xfs_mru_cache_insert(cache, ip->i_ino, item);
if (err) {
kmem_zone_free(item_zone, item);
return err;
}
xfs_filestream_get_ag(mp, ag);
IHOLD(ip);
if (pip)
IHOLD(pip);
TRACE_UPDATE(mp, ip, ag, xfs_filestream_peek_ag(mp, ag),
ag, xfs_filestream_peek_ag(mp, ag));
return 0;
}
static int
_xfs_filestream_pick_ag(
xfs_mount_t *mp,
xfs_agnumber_t startag,
xfs_agnumber_t *agp,
int flags,
xfs_extlen_t minlen)
{
int streams, max_streams;
int err, trylock, nscan;
xfs_extlen_t longest, free, minfree, maxfree = 0;
xfs_agnumber_t ag, max_ag = NULLAGNUMBER;
struct xfs_perag *pag;
minfree = mp->m_sb.sb_agblocks / 50;
ag = startag;
*agp = NULLAGNUMBER;
trylock = XFS_ALLOC_FLAG_TRYLOCK;
for (nscan = 0; 1; nscan++) {
pag = xfs_perag_get(mp, ag);
TRACE_AG_SCAN(mp, ag, atomic_read(&pag->pagf_fstrms));
if (!pag->pagf_init) {
err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
if (err && !trylock) {
xfs_perag_put(pag);
return err;
}
}
if (!pag->pagf_init)
goto next_ag;
if (pag->pagf_freeblks > maxfree) {
maxfree = pag->pagf_freeblks;
max_streams = atomic_read(&pag->pagf_fstrms);
max_ag = ag;
}
if (xfs_filestream_get_ag(mp, ag) > 1) {
xfs_filestream_put_ag(mp, ag);
goto next_ag;
}
longest = xfs_alloc_longest_free_extent(mp, pag);
if (((minlen && longest >= minlen) ||
(!minlen && pag->pagf_freeblks >= minfree)) &&
(!pag->pagf_metadata || !(flags & XFS_PICK_USERDATA) ||
(flags & XFS_PICK_LOWSPACE))) {
free = pag->pagf_freeblks;
streams = atomic_read(&pag->pagf_fstrms);
xfs_perag_put(pag);
*agp = ag;
break;
}
xfs_filestream_put_ag(mp, ag);
next_ag:
xfs_perag_put(pag);
if (++ag >= mp->m_sb.sb_agcount)
ag = 0;
if (ag != startag)
continue;
if (trylock != 0) {
trylock = 0;
continue;
}
if (!(flags & XFS_PICK_LOWSPACE)) {
flags |= XFS_PICK_LOWSPACE;
continue;
}
if (max_ag != NULLAGNUMBER) {
xfs_filestream_get_ag(mp, max_ag);
TRACE_AG_PICK1(mp, max_ag, maxfree);
streams = max_streams;
free = maxfree;
*agp = max_ag;
break;
}
TRACE_AG_PICK1(mp, max_ag, maxfree);
*agp = 0;
return 0;
}
TRACE_AG_PICK2(mp, startag, *agp, streams, free, nscan, flags);
return 0;
}
