2018-06-06 05:42:14 +03:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
2007-07-11 05:09:12 +04:00
|
|
|
/*
|
|
|
|
* Copyright (c) 2006-2007 Silicon Graphics, Inc.
|
2014-04-23 01:11:51 +04:00
|
|
|
* Copyright (c) 2014 Christoph Hellwig.
|
2007-07-11 05:09:12 +04:00
|
|
|
* All Rights Reserved.
|
|
|
|
*/
|
|
|
|
#include "xfs.h"
|
2019-06-29 05:25:35 +03:00
|
|
|
#include "xfs_shared.h"
|
2013-10-23 03:51:50 +04:00
|
|
|
#include "xfs_format.h"
|
2013-10-23 03:50:10 +04:00
|
|
|
#include "xfs_log_format.h"
|
|
|
|
#include "xfs_trans_resv.h"
|
|
|
|
#include "xfs_mount.h"
|
2007-07-11 05:09:12 +04:00
|
|
|
#include "xfs_inode.h"
|
|
|
|
#include "xfs_bmap.h"
|
|
|
|
#include "xfs_alloc.h"
|
|
|
|
#include "xfs_mru_cache.h"
|
2009-12-15 02:14:59 +03:00
|
|
|
#include "xfs_trace.h"
|
2021-06-02 03:48:24 +03:00
|
|
|
#include "xfs_ag.h"
|
xfs: set up per-AG free space reservations
One unfortunate quirk of the reference count and reverse mapping
btrees -- they can expand in size when blocks are written to *other*
allocation groups if, say, one large extent becomes a lot of tiny
extents. Since we don't want to start throwing errors in the middle
of CoWing, we need to reserve some blocks to handle future expansion.
The transaction block reservation counters aren't sufficient here
because we have to have a reserve of blocks in every AG, not just
somewhere in the filesystem.
Therefore, create two per-AG block reservation pools. One feeds the
AGFL so that rmapbt expansion always succeeds, and the other feeds all
other metadata so that refcountbt expansion never fails.
Use the count of how many reserved blocks we need to have on hand to
create a virtual reservation in the AG. Through selective clamping of
the maximum length of allocation requests and of the length of the
longest free extent, we can make it look like there's less free space
in the AG unless the reservation owner is asking for blocks.
In other words, play some accounting tricks in-core to make sure that
we always have blocks available. On the plus side, there's nothing to
clean up if we crash, which is contrast to the strategy that the rough
draft used (actually removing extents from the freespace btrees).
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-19 03:30:52 +03:00
|
|
|
#include "xfs_ag_resv.h"
|
2018-07-12 08:26:14 +03:00
|
|
|
#include "xfs_trans.h"
|
2019-11-13 07:40:00 +03:00
|
|
|
#include "xfs_filestream.h"
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_fstrm_item {
|
|
|
|
struct xfs_mru_cache_elem mru;
|
|
|
|
xfs_agnumber_t ag; /* AG in use for this directory */
|
|
|
|
};
|
|
|
|
|
|
|
|
enum xfs_fstrm_alloc {
|
|
|
|
XFS_PICK_USERDATA = 1,
|
|
|
|
XFS_PICK_LOWSPACE = 2,
|
|
|
|
};
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2010-07-20 11:31:01 +04:00
|
|
|
/*
|
|
|
|
* Allocation group filestream associations are tracked with per-ag atomic
|
2014-04-23 01:11:51 +04:00
|
|
|
* counters. These counters allow xfs_filestream_pick_ag() to tell whether a
|
2020-09-23 19:26:31 +03:00
|
|
|
* particular AG already has active filestreams associated with it.
|
2010-07-20 11:31:01 +04:00
|
|
|
*/
|
2014-04-23 01:11:52 +04:00
|
|
|
int
|
2010-07-20 11:31:01 +04:00
|
|
|
xfs_filestream_peek_ag(
|
|
|
|
xfs_mount_t *mp,
|
|
|
|
xfs_agnumber_t agno)
|
|
|
|
{
|
|
|
|
struct xfs_perag *pag;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
pag = xfs_perag_get(mp, agno);
|
|
|
|
ret = atomic_read(&pag->pagf_fstrms);
|
|
|
|
xfs_perag_put(pag);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
xfs_filestream_get_ag(
|
|
|
|
xfs_mount_t *mp,
|
|
|
|
xfs_agnumber_t agno)
|
|
|
|
{
|
|
|
|
struct xfs_perag *pag;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
pag = xfs_perag_get(mp, agno);
|
|
|
|
ret = atomic_inc_return(&pag->pagf_fstrms);
|
|
|
|
xfs_perag_put(pag);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
xfs_filestream_put_ag(
|
|
|
|
xfs_mount_t *mp,
|
|
|
|
xfs_agnumber_t agno)
|
|
|
|
{
|
|
|
|
struct xfs_perag *pag;
|
|
|
|
|
|
|
|
pag = xfs_perag_get(mp, agno);
|
|
|
|
atomic_dec(&pag->pagf_fstrms);
|
|
|
|
xfs_perag_put(pag);
|
|
|
|
}
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
static void
|
|
|
|
xfs_fstrm_free_func(
|
2018-04-09 20:23:39 +03:00
|
|
|
void *data,
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_mru_cache_elem *mru)
|
|
|
|
{
|
2018-04-09 20:23:39 +03:00
|
|
|
struct xfs_mount *mp = data;
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_fstrm_item *item =
|
|
|
|
container_of(mru, struct xfs_fstrm_item, mru);
|
|
|
|
|
2018-04-09 20:23:39 +03:00
|
|
|
xfs_filestream_put_ag(mp, item->ag);
|
|
|
|
trace_xfs_filestream_free(mp, mru->key, item->ag);
|
2014-04-23 01:11:51 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
kmem_free(item);
|
2014-04-23 01:11:51 +04:00
|
|
|
}
|
|
|
|
|
2007-07-11 05:09:12 +04:00
|
|
|
/*
|
|
|
|
* Scan the AGs starting at startag looking for an AG that isn't in use and has
|
|
|
|
* at least minlen blocks free.
|
|
|
|
*/
|
|
|
|
static int
|
2014-04-23 01:11:51 +04:00
|
|
|
xfs_filestream_pick_ag(
|
|
|
|
struct xfs_inode *ip,
|
|
|
|
xfs_agnumber_t startag,
|
|
|
|
xfs_agnumber_t *agp,
|
|
|
|
int flags,
|
|
|
|
xfs_extlen_t minlen)
|
2007-07-11 05:09:12 +04:00
|
|
|
{
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_mount *mp = ip->i_mount;
|
|
|
|
struct xfs_fstrm_item *item;
|
|
|
|
struct xfs_perag *pag;
|
2014-04-23 01:11:52 +04:00
|
|
|
xfs_extlen_t longest, free = 0, minfree, maxfree = 0;
|
2014-04-23 01:11:51 +04:00
|
|
|
xfs_agnumber_t ag, max_ag = NULLAGNUMBER;
|
|
|
|
int err, trylock, nscan;
|
|
|
|
|
2016-02-09 08:54:58 +03:00
|
|
|
ASSERT(S_ISDIR(VFS_I(ip)->i_mode));
|
2007-07-11 05:09:12 +04:00
|
|
|
|
|
|
|
/* 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++) {
|
2018-04-09 20:23:39 +03:00
|
|
|
trace_xfs_filestream_scan(mp, ip->i_ino, ag);
|
2014-04-23 01:11:52 +04:00
|
|
|
|
2010-01-11 14:47:42 +03:00
|
|
|
pag = xfs_perag_get(mp, ag);
|
2007-07-11 05:09:12 +04:00
|
|
|
|
|
|
|
if (!pag->pagf_init) {
|
|
|
|
err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
|
2020-01-24 04:01:19 +03:00
|
|
|
if (err) {
|
2010-01-11 14:47:42 +03:00
|
|
|
xfs_perag_put(pag);
|
2020-01-24 04:01:19 +03:00
|
|
|
if (err != -EAGAIN)
|
|
|
|
return err;
|
|
|
|
/* Couldn't lock the AGF, skip this AG. */
|
|
|
|
continue;
|
2010-01-11 14:47:42 +03:00
|
|
|
}
|
2007-07-11 05:09:12 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/* 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;
|
|
|
|
}
|
|
|
|
|
2018-04-06 20:09:42 +03:00
|
|
|
longest = xfs_alloc_longest_free_extent(pag,
|
xfs: set up per-AG free space reservations
One unfortunate quirk of the reference count and reverse mapping
btrees -- they can expand in size when blocks are written to *other*
allocation groups if, say, one large extent becomes a lot of tiny
extents. Since we don't want to start throwing errors in the middle
of CoWing, we need to reserve some blocks to handle future expansion.
The transaction block reservation counters aren't sufficient here
because we have to have a reserve of blocks in every AG, not just
somewhere in the filesystem.
Therefore, create two per-AG block reservation pools. One feeds the
AGFL so that rmapbt expansion always succeeds, and the other feeds all
other metadata so that refcountbt expansion never fails.
Use the count of how many reserved blocks we need to have on hand to
create a virtual reservation in the AG. Through selective clamping of
the maximum length of allocation requests and of the length of the
longest free extent, we can make it look like there's less free space
in the AG unless the reservation owner is asking for blocks.
In other words, play some accounting tricks in-core to make sure that
we always have blocks available. On the plus side, there's nothing to
clean up if we crash, which is contrast to the strategy that the rough
draft used (actually removing extents from the freespace btrees).
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-19 03:30:52 +03:00
|
|
|
xfs_alloc_min_freelist(mp, pag),
|
|
|
|
xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
|
2007-07-11 05:09:12 +04:00
|
|
|
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;
|
2010-01-11 14:47:42 +03:00
|
|
|
xfs_perag_put(pag);
|
2007-07-11 05:09:12 +04:00
|
|
|
*agp = ag;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Drop the reference on this AG, it's not usable. */
|
|
|
|
xfs_filestream_put_ag(mp, ag);
|
|
|
|
next_ag:
|
2010-01-11 14:47:42 +03:00
|
|
|
xfs_perag_put(pag);
|
2007-07-11 05:09:12 +04:00
|
|
|
/* 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 */
|
2014-04-23 01:11:52 +04:00
|
|
|
trace_xfs_filestream_pick(ip, *agp, free, nscan);
|
2007-07-11 05:09:12 +04:00
|
|
|
*agp = 0;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-04-23 01:11:52 +04:00
|
|
|
trace_xfs_filestream_pick(ip, *agp, free, nscan);
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
if (*agp == NULLAGNUMBER)
|
2007-07-11 05:09:12 +04:00
|
|
|
return 0;
|
|
|
|
|
2014-06-25 08:58:08 +04:00
|
|
|
err = -ENOMEM;
|
2014-04-23 01:11:51 +04:00
|
|
|
item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
|
2007-07-11 05:09:12 +04:00
|
|
|
if (!item)
|
2014-04-23 01:11:51 +04:00
|
|
|
goto out_put_ag;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
item->ag = *agp;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);
|
2007-07-11 05:09:12 +04:00
|
|
|
if (err) {
|
2014-06-25 08:58:08 +04:00
|
|
|
if (err == -EEXIST)
|
2014-04-23 01:11:51 +04:00
|
|
|
err = 0;
|
|
|
|
goto out_free_item;
|
2007-07-11 05:09:12 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
out_free_item:
|
2014-04-23 01:11:51 +04:00
|
|
|
kmem_free(item);
|
2014-04-23 01:11:51 +04:00
|
|
|
out_put_ag:
|
|
|
|
xfs_filestream_put_ag(mp, *agp);
|
|
|
|
return err;
|
2007-07-11 05:09:12 +04:00
|
|
|
}
|
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
static struct xfs_inode *
|
|
|
|
xfs_filestream_get_parent(
|
|
|
|
struct xfs_inode *ip)
|
2007-07-11 05:09:12 +04:00
|
|
|
{
|
2014-04-23 01:11:51 +04:00
|
|
|
struct inode *inode = VFS_I(ip), *dir = NULL;
|
|
|
|
struct dentry *dentry, *parent;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
dentry = d_find_alias(inode);
|
|
|
|
if (!dentry)
|
|
|
|
goto out;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
parent = dget_parent(dentry);
|
|
|
|
if (!parent)
|
|
|
|
goto out_dput;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2015-03-18 01:25:59 +03:00
|
|
|
dir = igrab(d_inode(parent));
|
2014-04-23 01:11:51 +04:00
|
|
|
dput(parent);
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
out_dput:
|
|
|
|
dput(dentry);
|
|
|
|
out:
|
|
|
|
return dir ? XFS_I(dir) : NULL;
|
2007-07-11 05:09:12 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2014-04-23 01:11:52 +04:00
|
|
|
* Find the right allocation group for a file, either by finding an
|
|
|
|
* existing file stream or creating a new one.
|
|
|
|
*
|
|
|
|
* Returns NULLAGNUMBER in case of an error.
|
2007-07-11 05:09:12 +04:00
|
|
|
*/
|
|
|
|
xfs_agnumber_t
|
|
|
|
xfs_filestream_lookup_ag(
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_inode *ip)
|
2007-07-11 05:09:12 +04:00
|
|
|
{
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_mount *mp = ip->i_mount;
|
|
|
|
struct xfs_inode *pip = NULL;
|
2014-04-23 01:11:52 +04:00
|
|
|
xfs_agnumber_t startag, ag = NULLAGNUMBER;
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_mru_cache_elem *mru;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2016-02-09 08:54:58 +03:00
|
|
|
ASSERT(S_ISREG(VFS_I(ip)->i_mode));
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
pip = xfs_filestream_get_parent(ip);
|
|
|
|
if (!pip)
|
2015-03-25 06:54:25 +03:00
|
|
|
return NULLAGNUMBER;
|
2014-04-23 01:11:51 +04:00
|
|
|
|
|
|
|
mru = xfs_mru_cache_lookup(mp->m_filestream, pip->i_ino);
|
2014-04-23 01:11:51 +04:00
|
|
|
if (mru) {
|
2014-04-23 01:11:52 +04:00
|
|
|
ag = container_of(mru, struct xfs_fstrm_item, mru)->ag;
|
2014-04-23 01:11:51 +04:00
|
|
|
xfs_mru_cache_done(mp->m_filestream);
|
2014-04-23 01:11:52 +04:00
|
|
|
|
2018-04-09 20:23:39 +03:00
|
|
|
trace_xfs_filestream_lookup(mp, ip->i_ino, ag);
|
2014-04-23 01:11:52 +04:00
|
|
|
goto out;
|
2007-07-11 05:09:12 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Set the starting AG using the rotor for inode32, otherwise
|
|
|
|
* use the directory inode's AG.
|
|
|
|
*/
|
2021-08-19 04:46:52 +03:00
|
|
|
if (xfs_is_inode32(mp)) {
|
2014-04-23 01:11:51 +04:00
|
|
|
xfs_agnumber_t rotorstep = xfs_rotorstep;
|
2007-07-11 05:09:12 +04:00
|
|
|
startag = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
|
|
|
|
mp->m_agfrotor = (mp->m_agfrotor + 1) %
|
|
|
|
(mp->m_sb.sb_agcount * rotorstep);
|
|
|
|
} else
|
|
|
|
startag = XFS_INO_TO_AGNO(mp, pip->i_ino);
|
|
|
|
|
2014-04-23 01:11:52 +04:00
|
|
|
if (xfs_filestream_pick_ag(pip, startag, &ag, 0, 0))
|
|
|
|
ag = NULLAGNUMBER;
|
|
|
|
out:
|
2018-07-25 22:52:32 +03:00
|
|
|
xfs_irele(pip);
|
2014-04-23 01:11:52 +04:00
|
|
|
return ag;
|
2007-07-11 05:09:12 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2014-04-23 01:11:51 +04:00
|
|
|
* Pick a new allocation group for the current file and its file stream.
|
|
|
|
*
|
|
|
|
* 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.
|
2007-07-11 05:09:12 +04:00
|
|
|
*/
|
|
|
|
int
|
|
|
|
xfs_filestream_new_ag(
|
2013-08-12 14:49:42 +04:00
|
|
|
struct xfs_bmalloca *ap,
|
|
|
|
xfs_agnumber_t *agp)
|
2007-07-11 05:09:12 +04:00
|
|
|
{
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_inode *ip = ap->ip, *pip;
|
|
|
|
struct xfs_mount *mp = ip->i_mount;
|
|
|
|
xfs_extlen_t minlen = ap->length;
|
|
|
|
xfs_agnumber_t startag = 0;
|
xfs: remote attribute blocks aren't really userdata
When adding a new remote attribute, we write the attribute to the
new extent before the allocation transaction is committed. This
means we cannot reuse busy extents as that violates crash
consistency semantics. Hence we currently treat remote attribute
extent allocation like userdata because it has the same overwrite
ordering constraints as userdata.
Unfortunately, this also allows the allocator to incorrectly apply
extent size hints to the remote attribute extent allocation. This
results in interesting failures, such as transaction block
reservation overruns and in-memory inode attribute fork corruption.
To fix this, we need to separate the busy extent reuse configuration
from the userdata configuration. This changes the definition of
XFS_BMAPI_METADATA slightly - it now means that allocation is
metadata and reuse of busy extents is acceptible due to the metadata
ordering semantics of the journal. If this flag is not set, it
means the allocation is that has unordered data writeback, and hence
busy extent reuse is not allowed. It no longer implies the
allocation is for user data, just that the data write will not be
strictly ordered. This matches the semantics for both user data
and remote attribute block allocation.
As such, This patch changes the "userdata" field to a "datatype"
field, and adds a "no busy reuse" flag to the field.
When we detect an unordered data extent allocation, we immediately set
the no reuse flag. We then set the "user data" flags based on the
inode fork we are allocating the extent to. Hence we only set
userdata flags on data fork allocations now and consider attribute
fork remote extents to be an unordered metadata extent.
The result is that remote attribute extents now have the expected
allocation semantics, and the data fork allocation behaviour is
completely unchanged.
It should be noted that there may be other ways to fix this (e.g.
use ordered metadata buffers for the remote attribute extent data
write) but they are more invasive and difficult to validate both
from a design and implementation POV. Hence this patch takes the
simple, obvious route to fixing the problem...
Reported-and-tested-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-26 01:21:28 +03:00
|
|
|
int flags = 0;
|
|
|
|
int err = 0;
|
2014-04-23 01:11:51 +04:00
|
|
|
struct xfs_mru_cache_elem *mru;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
*agp = NULLAGNUMBER;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
pip = xfs_filestream_get_parent(ip);
|
|
|
|
if (!pip)
|
|
|
|
goto exit;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
mru = xfs_mru_cache_remove(mp->m_filestream, pip->i_ino);
|
|
|
|
if (mru) {
|
|
|
|
struct xfs_fstrm_item *item =
|
|
|
|
container_of(mru, struct xfs_fstrm_item, mru);
|
|
|
|
startag = (item->ag + 1) % mp->m_sb.sb_agcount;
|
2007-07-11 05:09:12 +04:00
|
|
|
}
|
|
|
|
|
2019-10-30 22:25:00 +03:00
|
|
|
if (ap->datatype & XFS_ALLOC_USERDATA)
|
xfs: remote attribute blocks aren't really userdata
When adding a new remote attribute, we write the attribute to the
new extent before the allocation transaction is committed. This
means we cannot reuse busy extents as that violates crash
consistency semantics. Hence we currently treat remote attribute
extent allocation like userdata because it has the same overwrite
ordering constraints as userdata.
Unfortunately, this also allows the allocator to incorrectly apply
extent size hints to the remote attribute extent allocation. This
results in interesting failures, such as transaction block
reservation overruns and in-memory inode attribute fork corruption.
To fix this, we need to separate the busy extent reuse configuration
from the userdata configuration. This changes the definition of
XFS_BMAPI_METADATA slightly - it now means that allocation is
metadata and reuse of busy extents is acceptible due to the metadata
ordering semantics of the journal. If this flag is not set, it
means the allocation is that has unordered data writeback, and hence
busy extent reuse is not allowed. It no longer implies the
allocation is for user data, just that the data write will not be
strictly ordered. This matches the semantics for both user data
and remote attribute block allocation.
As such, This patch changes the "userdata" field to a "datatype"
field, and adds a "no busy reuse" flag to the field.
When we detect an unordered data extent allocation, we immediately set
the no reuse flag. We then set the "user data" flags based on the
inode fork we are allocating the extent to. Hence we only set
userdata flags on data fork allocations now and consider attribute
fork remote extents to be an unordered metadata extent.
The result is that remote attribute extents now have the expected
allocation semantics, and the data fork allocation behaviour is
completely unchanged.
It should be noted that there may be other ways to fix this (e.g.
use ordered metadata buffers for the remote attribute extent data
write) but they are more invasive and difficult to validate both
from a design and implementation POV. Hence this patch takes the
simple, obvious route to fixing the problem...
Reported-and-tested-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-26 01:21:28 +03:00
|
|
|
flags |= XFS_PICK_USERDATA;
|
2018-08-01 17:20:31 +03:00
|
|
|
if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
|
xfs: remote attribute blocks aren't really userdata
When adding a new remote attribute, we write the attribute to the
new extent before the allocation transaction is committed. This
means we cannot reuse busy extents as that violates crash
consistency semantics. Hence we currently treat remote attribute
extent allocation like userdata because it has the same overwrite
ordering constraints as userdata.
Unfortunately, this also allows the allocator to incorrectly apply
extent size hints to the remote attribute extent allocation. This
results in interesting failures, such as transaction block
reservation overruns and in-memory inode attribute fork corruption.
To fix this, we need to separate the busy extent reuse configuration
from the userdata configuration. This changes the definition of
XFS_BMAPI_METADATA slightly - it now means that allocation is
metadata and reuse of busy extents is acceptible due to the metadata
ordering semantics of the journal. If this flag is not set, it
means the allocation is that has unordered data writeback, and hence
busy extent reuse is not allowed. It no longer implies the
allocation is for user data, just that the data write will not be
strictly ordered. This matches the semantics for both user data
and remote attribute block allocation.
As such, This patch changes the "userdata" field to a "datatype"
field, and adds a "no busy reuse" flag to the field.
When we detect an unordered data extent allocation, we immediately set
the no reuse flag. We then set the "user data" flags based on the
inode fork we are allocating the extent to. Hence we only set
userdata flags on data fork allocations now and consider attribute
fork remote extents to be an unordered metadata extent.
The result is that remote attribute extents now have the expected
allocation semantics, and the data fork allocation behaviour is
completely unchanged.
It should be noted that there may be other ways to fix this (e.g.
use ordered metadata buffers for the remote attribute extent data
write) but they are more invasive and difficult to validate both
from a design and implementation POV. Hence this patch takes the
simple, obvious route to fixing the problem...
Reported-and-tested-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-26 01:21:28 +03:00
|
|
|
flags |= XFS_PICK_LOWSPACE;
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
err = xfs_filestream_pick_ag(pip, startag, agp, flags, minlen);
|
2007-07-11 05:09:12 +04:00
|
|
|
|
|
|
|
/*
|
2014-04-23 01:11:51 +04:00
|
|
|
* Only free the item here so we skip over the old AG earlier.
|
2007-07-11 05:09:12 +04:00
|
|
|
*/
|
2014-04-23 01:11:51 +04:00
|
|
|
if (mru)
|
2018-04-09 20:23:39 +03:00
|
|
|
xfs_fstrm_free_func(mp, mru);
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2018-07-25 22:52:32 +03:00
|
|
|
xfs_irele(pip);
|
2014-04-23 01:11:51 +04:00
|
|
|
exit:
|
|
|
|
if (*agp == NULLAGNUMBER)
|
|
|
|
*agp = 0;
|
|
|
|
return err;
|
|
|
|
}
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
void
|
|
|
|
xfs_filestream_deassociate(
|
|
|
|
struct xfs_inode *ip)
|
|
|
|
{
|
|
|
|
xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino);
|
|
|
|
}
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
int
|
|
|
|
xfs_filestream_mount(
|
|
|
|
xfs_mount_t *mp)
|
|
|
|
{
|
2007-07-11 05:09:12 +04:00
|
|
|
/*
|
2014-04-23 01:11:51 +04:00
|
|
|
* 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.
|
2007-07-11 05:09:12 +04:00
|
|
|
*/
|
2018-04-09 20:23:39 +03:00
|
|
|
return xfs_mru_cache_create(&mp->m_filestream, mp,
|
|
|
|
xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);
|
2014-04-23 01:11:51 +04:00
|
|
|
}
|
2007-07-11 05:09:12 +04:00
|
|
|
|
2014-04-23 01:11:51 +04:00
|
|
|
void
|
|
|
|
xfs_filestream_unmount(
|
|
|
|
xfs_mount_t *mp)
|
|
|
|
{
|
|
|
|
xfs_mru_cache_destroy(mp->m_filestream);
|
|
|
|
}
|