linux/fs/xfs/libxfs/xfs_group.c
Christoph Hellwig 59655147ec xfs: improve the xg_active_ref check in xfs_group_free
Split up the XFS_IS_CORRUPT statement so that it immediately shows
if the reference counter overflowed or underflowed.

I ran into this quite a bit when developing the zoned allocator, and had
to reapply the patch for some work recently.  We might as well just apply
it upstream given that freeing group is far removed from performance
critical code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Carlos Maiolino <cem@kernel.org>
2025-07-24 17:30:14 +02:00

230 lines
5 KiB
C

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2018 Red Hat, Inc.
*/
#include "xfs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_error.h"
#include "xfs_trace.h"
#include "xfs_extent_busy.h"
#include "xfs_group.h"
/*
* Groups can have passive and active references.
*
* For passive references the code freeing a group is responsible for cleaning
* up objects that hold the passive references (e.g. cached buffers).
* Routines manipulating passive references are xfs_group_get, xfs_group_hold
* and xfs_group_put.
*
* Active references are for short term access to the group for walking trees or
* accessing state. If a group is being shrunk or offlined, the lookup will fail
* to find that group and return NULL instead.
* Routines manipulating active references are xfs_group_grab and
* xfs_group_rele.
*/
struct xfs_group *
xfs_group_get(
struct xfs_mount *mp,
uint32_t index,
enum xfs_group_type type)
{
struct xfs_group *xg;
rcu_read_lock();
xg = xa_load(&mp->m_groups[type].xa, index);
if (xg) {
trace_xfs_group_get(xg, _RET_IP_);
ASSERT(atomic_read(&xg->xg_ref) >= 0);
atomic_inc(&xg->xg_ref);
}
rcu_read_unlock();
return xg;
}
struct xfs_group *
xfs_group_hold(
struct xfs_group *xg)
{
ASSERT(atomic_read(&xg->xg_ref) > 0 ||
atomic_read(&xg->xg_active_ref) > 0);
trace_xfs_group_hold(xg, _RET_IP_);
atomic_inc(&xg->xg_ref);
return xg;
}
void
xfs_group_put(
struct xfs_group *xg)
{
trace_xfs_group_put(xg, _RET_IP_);
ASSERT(atomic_read(&xg->xg_ref) > 0);
atomic_dec(&xg->xg_ref);
}
struct xfs_group *
xfs_group_grab(
struct xfs_mount *mp,
uint32_t index,
enum xfs_group_type type)
{
struct xfs_group *xg;
rcu_read_lock();
xg = xa_load(&mp->m_groups[type].xa, index);
if (xg) {
trace_xfs_group_grab(xg, _RET_IP_);
if (!atomic_inc_not_zero(&xg->xg_active_ref))
xg = NULL;
}
rcu_read_unlock();
return xg;
}
/*
* Iterate to the next group. To start the iteration at @start_index, a %NULL
* @xg is passed, else the previous group returned from this function. The
* caller should break out of the loop when this returns %NULL. If the caller
* wants to break out of a loop that did not finish it needs to release the
* active reference to @xg using xfs_group_rele() itself.
*/
struct xfs_group *
xfs_group_next_range(
struct xfs_mount *mp,
struct xfs_group *xg,
uint32_t start_index,
uint32_t end_index,
enum xfs_group_type type)
{
uint32_t index = start_index;
if (xg) {
index = xg->xg_gno + 1;
xfs_group_rele(xg);
}
if (index > end_index)
return NULL;
return xfs_group_grab(mp, index, type);
}
/*
* Find the next group after @xg, or the first group if @xg is NULL.
*/
struct xfs_group *
xfs_group_grab_next_mark(
struct xfs_mount *mp,
struct xfs_group *xg,
xa_mark_t mark,
enum xfs_group_type type)
{
unsigned long index = 0;
if (xg) {
index = xg->xg_gno + 1;
xfs_group_rele(xg);
}
rcu_read_lock();
xg = xa_find(&mp->m_groups[type].xa, &index, ULONG_MAX, mark);
if (xg) {
trace_xfs_group_grab_next_tag(xg, _RET_IP_);
if (!atomic_inc_not_zero(&xg->xg_active_ref))
xg = NULL;
}
rcu_read_unlock();
return xg;
}
void
xfs_group_rele(
struct xfs_group *xg)
{
trace_xfs_group_rele(xg, _RET_IP_);
atomic_dec(&xg->xg_active_ref);
}
void
xfs_group_free(
struct xfs_mount *mp,
uint32_t index,
enum xfs_group_type type,
void (*uninit)(struct xfs_group *xg))
{
struct xfs_group *xg = xa_erase(&mp->m_groups[type].xa, index);
XFS_IS_CORRUPT(mp, atomic_read(&xg->xg_ref) != 0);
xfs_defer_drain_free(&xg->xg_intents_drain);
#ifdef __KERNEL__
if (xfs_group_has_extent_busy(xg->xg_mount, xg->xg_type))
kfree(xg->xg_busy_extents);
#endif
if (uninit)
uninit(xg);
/* drop the mount's active reference */
xfs_group_rele(xg);
XFS_IS_CORRUPT(mp, atomic_read(&xg->xg_active_ref) > 0);
XFS_IS_CORRUPT(mp, atomic_read(&xg->xg_active_ref) < 0);
kfree_rcu_mightsleep(xg);
}
int
xfs_group_insert(
struct xfs_mount *mp,
struct xfs_group *xg,
uint32_t index,
enum xfs_group_type type)
{
int error;
xg->xg_mount = mp;
xg->xg_gno = index;
xg->xg_type = type;
#ifdef __KERNEL__
if (xfs_group_has_extent_busy(mp, type)) {
xg->xg_busy_extents = xfs_extent_busy_alloc();
if (!xg->xg_busy_extents)
return -ENOMEM;
}
spin_lock_init(&xg->xg_state_lock);
xfs_hooks_init(&xg->xg_rmap_update_hooks);
#endif
xfs_defer_drain_init(&xg->xg_intents_drain);
/* Active ref owned by mount indicates group is online. */
atomic_set(&xg->xg_active_ref, 1);
error = xa_insert(&mp->m_groups[type].xa, index, xg, GFP_KERNEL);
if (error) {
WARN_ON_ONCE(error == -EBUSY);
goto out_drain;
}
return 0;
out_drain:
xfs_defer_drain_free(&xg->xg_intents_drain);
#ifdef __KERNEL__
if (xfs_group_has_extent_busy(xg->xg_mount, xg->xg_type))
kfree(xg->xg_busy_extents);
#endif
return error;
}
struct xfs_group *
xfs_group_get_by_fsb(
struct xfs_mount *mp,
xfs_fsblock_t fsbno,
enum xfs_group_type type)
{
return xfs_group_get(mp, xfs_fsb_to_gno(mp, fsbno, type), type);
}