linux-stable/fs/xfs/xfs_filestream.c
Dave Chinner 1e473279f4 xfs: fix double xfs_perag_rele() in xfs_filestream_pick_ag()
xfs_bmap_longest_free_extent() can return an error when accessing
the AGF fails. In this case, the behaviour of
xfs_filestream_pick_ag() is conditional on the error. We may
continue the loop, or break out of it. The error handling after the
loop cleans up the perag reference held when the break occurs. If we
continue, the next loop iteration handles cleaning up the perag
reference.

EIther way, we don't need to release the active perag reference when
xfs_bmap_longest_free_extent() fails. Doing so means we do a double
decrement on the active reference count, and this causes tha active
reference count to fall to zero. At this point, new active
references will fail.

This leads to unmount hanging because it tries to grab active
references to that perag, only for it to fail. This happens inside a
loop that retries until a inode tree radix tree tag is cleared,
which cannot happen because we can't get an active reference to the
perag.

The unmount livelocks in this path:

  xfs_reclaim_inodes+0x80/0xc0
  xfs_unmount_flush_inodes+0x5b/0x70
  xfs_unmountfs+0x5b/0x1a0
  xfs_fs_put_super+0x49/0x110
  generic_shutdown_super+0x7c/0x1a0
  kill_block_super+0x27/0x50
  deactivate_locked_super+0x30/0x90
  deactivate_super+0x3c/0x50
  cleanup_mnt+0xc2/0x160
  __cleanup_mnt+0x12/0x20
  task_work_run+0x5e/0xa0
  exit_to_user_mode_prepare+0x1bc/0x1c0
  syscall_exit_to_user_mode+0x16/0x40
  do_syscall_64+0x40/0x80
  entry_SYSCALL_64_after_hwframe+0x63/0xcd

Reported-by: Pengfei Xu <pengfei.xu@intel.com>
Fixes: eb70aa2d8e ("xfs: use for_each_perag_wrap in xfs_filestream_pick_ag")
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2023-06-05 14:48:15 +10:00

407 lines
10 KiB
C

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2006-2007 Silicon Graphics, Inc.
* Copyright (c) 2014 Christoph Hellwig.
* All Rights Reserved.
*/
#include "xfs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_inode.h"
#include "xfs_bmap.h"
#include "xfs_bmap_util.h"
#include "xfs_alloc.h"
#include "xfs_mru_cache.h"
#include "xfs_trace.h"
#include "xfs_ag.h"
#include "xfs_ag_resv.h"
#include "xfs_trans.h"
#include "xfs_filestream.h"
struct xfs_fstrm_item {
struct xfs_mru_cache_elem mru;
struct xfs_perag *pag; /* AG in use for this directory */
};
enum xfs_fstrm_alloc {
XFS_PICK_USERDATA = 1,
XFS_PICK_LOWSPACE = 2,
};
static void
xfs_fstrm_free_func(
void *data,
struct xfs_mru_cache_elem *mru)
{
struct xfs_fstrm_item *item =
container_of(mru, struct xfs_fstrm_item, mru);
struct xfs_perag *pag = item->pag;
trace_xfs_filestream_free(pag, mru->key);
atomic_dec(&pag->pagf_fstrms);
xfs_perag_rele(pag);
kmem_free(item);
}
/*
* Scan the AGs starting at start_agno looking for an AG that isn't in use and
* has at least minlen blocks free. If no AG is found to match the allocation
* requirements, pick the AG with the most free space in it.
*/
static int
xfs_filestream_pick_ag(
struct xfs_alloc_arg *args,
xfs_ino_t pino,
xfs_agnumber_t start_agno,
int flags,
xfs_extlen_t *longest)
{
struct xfs_mount *mp = args->mp;
struct xfs_perag *pag;
struct xfs_perag *max_pag = NULL;
xfs_extlen_t minlen = *longest;
xfs_extlen_t free = 0, minfree, maxfree = 0;
xfs_agnumber_t agno;
bool first_pass = true;
int err;
/* 2% of an AG's blocks must be free for it to be chosen. */
minfree = mp->m_sb.sb_agblocks / 50;
restart:
for_each_perag_wrap(mp, start_agno, agno, pag) {
trace_xfs_filestream_scan(pag, pino);
*longest = 0;
err = xfs_bmap_longest_free_extent(pag, NULL, longest);
if (err) {
if (err != -EAGAIN)
break;
/* Couldn't lock the AGF, skip this AG. */
err = 0;
continue;
}
/* Keep track of the AG with the most free blocks. */
if (pag->pagf_freeblks > maxfree) {
maxfree = pag->pagf_freeblks;
if (max_pag)
xfs_perag_rele(max_pag);
atomic_inc(&pag->pag_active_ref);
max_pag = pag;
}
/*
* 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 (atomic_inc_return(&pag->pagf_fstrms) <= 1) {
if (((minlen && *longest >= minlen) ||
(!minlen && pag->pagf_freeblks >= minfree)) &&
(!xfs_perag_prefers_metadata(pag) ||
!(flags & XFS_PICK_USERDATA) ||
(flags & XFS_PICK_LOWSPACE))) {
/* Break out, retaining the reference on the AG. */
free = pag->pagf_freeblks;
break;
}
}
/* Drop the reference on this AG, it's not usable. */
atomic_dec(&pag->pagf_fstrms);
}
if (err) {
xfs_perag_rele(pag);
if (max_pag)
xfs_perag_rele(max_pag);
return err;
}
if (!pag) {
/*
* Allow a second pass to give xfs_bmap_longest_free_extent()
* another attempt at locking AGFs that it might have skipped
* over before we fail.
*/
if (first_pass) {
first_pass = false;
goto restart;
}
/*
* We must be low on data space, so run a final lowspace
* optimised selection pass if we haven't already.
*/
if (!(flags & XFS_PICK_LOWSPACE)) {
flags |= XFS_PICK_LOWSPACE;
goto restart;
}
/*
* No unassociated AGs are available, so select the AG with the
* most free space, regardless of whether it's already in use by
* another filestream. It none suit, just use whatever AG we can
* grab.
*/
if (!max_pag) {
for_each_perag_wrap(args->mp, 0, start_agno, args->pag)
break;
atomic_inc(&args->pag->pagf_fstrms);
*longest = 0;
} else {
pag = max_pag;
free = maxfree;
atomic_inc(&pag->pagf_fstrms);
}
} else if (max_pag) {
xfs_perag_rele(max_pag);
}
trace_xfs_filestream_pick(pag, pino, free);
args->pag = pag;
return 0;
}
static struct xfs_inode *
xfs_filestream_get_parent(
struct xfs_inode *ip)
{
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(d_inode(parent));
dput(parent);
out_dput:
dput(dentry);
out:
return dir ? XFS_I(dir) : NULL;
}
/*
* Lookup the mru cache for an existing association. If one exists and we can
* use it, return with an active perag reference indicating that the allocation
* will proceed with that association.
*
* If we have no association, or we cannot use the current one and have to
* destroy it, return with longest = 0 to tell the caller to create a new
* association.
*/
static int
xfs_filestream_lookup_association(
struct xfs_bmalloca *ap,
struct xfs_alloc_arg *args,
xfs_ino_t pino,
xfs_extlen_t *longest)
{
struct xfs_mount *mp = args->mp;
struct xfs_perag *pag;
struct xfs_mru_cache_elem *mru;
int error = 0;
*longest = 0;
mru = xfs_mru_cache_lookup(mp->m_filestream, pino);
if (!mru)
return 0;
/*
* Grab the pag and take an extra active reference for the caller whilst
* the mru item cannot go away. This means we'll pin the perag with
* the reference we get here even if the filestreams association is torn
* down immediately after we mark the lookup as done.
*/
pag = container_of(mru, struct xfs_fstrm_item, mru)->pag;
atomic_inc(&pag->pag_active_ref);
xfs_mru_cache_done(mp->m_filestream);
trace_xfs_filestream_lookup(pag, ap->ip->i_ino);
ap->blkno = XFS_AGB_TO_FSB(args->mp, pag->pag_agno, 0);
xfs_bmap_adjacent(ap);
/*
* If there is very little free space before we start a filestreams
* allocation, we're almost guaranteed to fail to find a large enough
* free space available so just use the cached AG.
*/
if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
*longest = 1;
goto out_done;
}
error = xfs_bmap_longest_free_extent(pag, args->tp, longest);
if (error == -EAGAIN)
error = 0;
if (error || *longest < args->maxlen) {
/* We aren't going to use this perag */
*longest = 0;
xfs_perag_rele(pag);
return error;
}
out_done:
args->pag = pag;
return 0;
}
static int
xfs_filestream_create_association(
struct xfs_bmalloca *ap,
struct xfs_alloc_arg *args,
xfs_ino_t pino,
xfs_extlen_t *longest)
{
struct xfs_mount *mp = args->mp;
struct xfs_mru_cache_elem *mru;
struct xfs_fstrm_item *item;
xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, pino);
int flags = 0;
int error;
/* Changing parent AG association now, so remove the existing one. */
mru = xfs_mru_cache_remove(mp->m_filestream, pino);
if (mru) {
struct xfs_fstrm_item *item =
container_of(mru, struct xfs_fstrm_item, mru);
agno = (item->pag->pag_agno + 1) % mp->m_sb.sb_agcount;
xfs_fstrm_free_func(mp, mru);
} else if (xfs_is_inode32(mp)) {
xfs_agnumber_t rotorstep = xfs_rotorstep;
agno = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
mp->m_agfrotor = (mp->m_agfrotor + 1) %
(mp->m_sb.sb_agcount * rotorstep);
}
ap->blkno = XFS_AGB_TO_FSB(args->mp, agno, 0);
xfs_bmap_adjacent(ap);
if (ap->datatype & XFS_ALLOC_USERDATA)
flags |= XFS_PICK_USERDATA;
if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
flags |= XFS_PICK_LOWSPACE;
*longest = ap->length;
error = xfs_filestream_pick_ag(args, pino, agno, flags, longest);
if (error)
return error;
/*
* We are going to use this perag now, so create an assoication for it.
* xfs_filestream_pick_ag() has already bumped the perag fstrms counter
* for us, so all we need to do here is take another active reference to
* the perag for the cached association.
*
* If we fail to store the association, we need to drop the fstrms
* counter as well as drop the perag reference we take here for the
* item. We do not need to return an error for this failure - as long as
* we return a referenced AG, the allocation can still go ahead just
* fine.
*/
item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
if (!item)
goto out_put_fstrms;
atomic_inc(&args->pag->pag_active_ref);
item->pag = args->pag;
error = xfs_mru_cache_insert(mp->m_filestream, pino, &item->mru);
if (error)
goto out_free_item;
return 0;
out_free_item:
xfs_perag_rele(item->pag);
kmem_free(item);
out_put_fstrms:
atomic_dec(&args->pag->pagf_fstrms);
return 0;
}
/*
* Search for an allocation group with a single extent large enough for
* the request. First we look for an existing association and use that if it
* is found. Otherwise, we create a new association by selecting an AG that fits
* the allocation criteria.
*
* We return with a referenced perag in args->pag to indicate which AG we are
* allocating into or an error with no references held.
*/
int
xfs_filestream_select_ag(
struct xfs_bmalloca *ap,
struct xfs_alloc_arg *args,
xfs_extlen_t *longest)
{
struct xfs_mount *mp = args->mp;
struct xfs_inode *pip;
xfs_ino_t ino = 0;
int error = 0;
*longest = 0;
args->total = ap->total;
pip = xfs_filestream_get_parent(ap->ip);
if (pip) {
ino = pip->i_ino;
error = xfs_filestream_lookup_association(ap, args, ino,
longest);
xfs_irele(pip);
if (error)
return error;
if (*longest >= args->maxlen)
goto out_select;
if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
goto out_select;
}
error = xfs_filestream_create_association(ap, args, ino, longest);
if (error)
return error;
out_select:
ap->blkno = XFS_AGB_TO_FSB(mp, args->pag->pag_agno, 0);
return 0;
}
void
xfs_filestream_deassociate(
struct xfs_inode *ip)
{
xfs_mru_cache_delete(ip->i_mount->m_filestream, 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, mp,
xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);
}
void
xfs_filestream_unmount(
xfs_mount_t *mp)
{
xfs_mru_cache_destroy(mp->m_filestream);
}