linux_dsm_epyc7002/fs/xfs/xfs_mount.c

2041 lines
51 KiB
C
Raw Normal View History

/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* All Rights Reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it would be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_bit.h"
#include "xfs_sb.h"
#include "xfs_mount.h"
#include "xfs_da_format.h"
#include "xfs_da_btree.h"
#include "xfs_inode.h"
#include "xfs_dir2.h"
#include "xfs_ialloc.h"
#include "xfs_alloc.h"
#include "xfs_rtalloc.h"
#include "xfs_bmap.h"
#include "xfs_trans.h"
#include "xfs_trans_priv.h"
#include "xfs_log.h"
#include "xfs_error.h"
#include "xfs_quota.h"
#include "xfs_fsops.h"
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
#include "xfs_trace.h"
#include "xfs_icache.h"
#include "xfs_sysfs.h"
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
#ifdef HAVE_PERCPU_SB
STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
int);
STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
int);
STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
#else
#define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
#define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
#endif
static DEFINE_MUTEX(xfs_uuid_table_mutex);
static int xfs_uuid_table_size;
static uuid_t *xfs_uuid_table;
/*
* See if the UUID is unique among mounted XFS filesystems.
* Mount fails if UUID is nil or a FS with the same UUID is already mounted.
*/
STATIC int
xfs_uuid_mount(
struct xfs_mount *mp)
{
uuid_t *uuid = &mp->m_sb.sb_uuid;
int hole, i;
if (mp->m_flags & XFS_MOUNT_NOUUID)
return 0;
if (uuid_is_nil(uuid)) {
xfs_warn(mp, "Filesystem has nil UUID - can't mount");
return -EINVAL;
}
mutex_lock(&xfs_uuid_table_mutex);
for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
if (uuid_is_nil(&xfs_uuid_table[i])) {
hole = i;
continue;
}
if (uuid_equal(uuid, &xfs_uuid_table[i]))
goto out_duplicate;
}
if (hole < 0) {
xfs_uuid_table = kmem_realloc(xfs_uuid_table,
(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
xfs_uuid_table_size * sizeof(*xfs_uuid_table),
KM_SLEEP);
hole = xfs_uuid_table_size++;
}
xfs_uuid_table[hole] = *uuid;
mutex_unlock(&xfs_uuid_table_mutex);
return 0;
out_duplicate:
mutex_unlock(&xfs_uuid_table_mutex);
xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
return -EINVAL;
}
STATIC void
xfs_uuid_unmount(
struct xfs_mount *mp)
{
uuid_t *uuid = &mp->m_sb.sb_uuid;
int i;
if (mp->m_flags & XFS_MOUNT_NOUUID)
return;
mutex_lock(&xfs_uuid_table_mutex);
for (i = 0; i < xfs_uuid_table_size; i++) {
if (uuid_is_nil(&xfs_uuid_table[i]))
continue;
if (!uuid_equal(uuid, &xfs_uuid_table[i]))
continue;
memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
break;
}
ASSERT(i < xfs_uuid_table_size);
mutex_unlock(&xfs_uuid_table_mutex);
}
STATIC void
__xfs_free_perag(
struct rcu_head *head)
{
struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
ASSERT(atomic_read(&pag->pag_ref) == 0);
kmem_free(pag);
}
/*
* Free up the per-ag resources associated with the mount structure.
*/
STATIC void
xfs_free_perag(
xfs_mount_t *mp)
{
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
xfs_agnumber_t agno;
struct xfs_perag *pag;
for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
spin_lock(&mp->m_perag_lock);
pag = radix_tree_delete(&mp->m_perag_tree, agno);
spin_unlock(&mp->m_perag_lock);
ASSERT(pag);
ASSERT(atomic_read(&pag->pag_ref) == 0);
call_rcu(&pag->rcu_head, __xfs_free_perag);
}
}
/*
* Check size of device based on the (data/realtime) block count.
* Note: this check is used by the growfs code as well as mount.
*/
int
xfs_sb_validate_fsb_count(
xfs_sb_t *sbp,
__uint64_t nblocks)
{
ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
ASSERT(sbp->sb_blocklog >= BBSHIFT);
/* Limited by ULONG_MAX of page cache index */
if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
return -EFBIG;
return 0;
}
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
int
xfs_initialize_perag(
xfs_mount_t *mp,
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
xfs_agnumber_t agcount,
xfs_agnumber_t *maxagi)
{
xfs_agnumber_t index;
xfs_agnumber_t first_initialised = 0;
xfs_perag_t *pag;
xfs_agino_t agino;
xfs_ino_t ino;
xfs_sb_t *sbp = &mp->m_sb;
int error = -ENOMEM;
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
/*
* Walk the current per-ag tree so we don't try to initialise AGs
* that already exist (growfs case). Allocate and insert all the
* AGs we don't find ready for initialisation.
*/
for (index = 0; index < agcount; index++) {
pag = xfs_perag_get(mp, index);
if (pag) {
xfs_perag_put(pag);
continue;
}
if (!first_initialised)
first_initialised = index;
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
if (!pag)
goto out_unwind;
pag->pag_agno = index;
pag->pag_mount = mp;
spin_lock_init(&pag->pag_ici_lock);
mutex_init(&pag->pag_ici_reclaim_lock);
INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
spin_lock_init(&pag->pag_buf_lock);
pag->pag_buf_tree = RB_ROOT;
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
if (radix_tree_preload(GFP_NOFS))
goto out_unwind;
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
spin_lock(&mp->m_perag_lock);
if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
BUG();
spin_unlock(&mp->m_perag_lock);
radix_tree_preload_end();
error = -EEXIST;
goto out_unwind;
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
}
spin_unlock(&mp->m_perag_lock);
radix_tree_preload_end();
}
/*
* If we mount with the inode64 option, or no inode overflows
* the legacy 32-bit address space clear the inode32 option.
*/
agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
mp->m_flags |= XFS_MOUNT_32BITINODES;
else
mp->m_flags &= ~XFS_MOUNT_32BITINODES;
if (mp->m_flags & XFS_MOUNT_32BITINODES)
index = xfs_set_inode32(mp, agcount);
else
index = xfs_set_inode64(mp, agcount);
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
if (maxagi)
*maxagi = index;
return 0;
out_unwind:
kmem_free(pag);
for (; index > first_initialised; index--) {
pag = radix_tree_delete(&mp->m_perag_tree, index);
kmem_free(pag);
}
return error;
}
/*
* xfs_readsb
*
* Does the initial read of the superblock.
*/
int
xfs_readsb(
struct xfs_mount *mp,
int flags)
{
unsigned int sector_size;
struct xfs_buf *bp;
struct xfs_sb *sbp = &mp->m_sb;
int error;
int loud = !(flags & XFS_MFSI_QUIET);
const struct xfs_buf_ops *buf_ops;
ASSERT(mp->m_sb_bp == NULL);
ASSERT(mp->m_ddev_targp != NULL);
/*
* For the initial read, we must guess at the sector
* size based on the block device. It's enough to
* get the sb_sectsize out of the superblock and
* then reread with the proper length.
* We don't verify it yet, because it may not be complete.
*/
sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
buf_ops = NULL;
/*
* Allocate a (locked) buffer to hold the superblock.
* This will be kept around at all times to optimize
* access to the superblock.
*/
reread:
error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
BTOBB(sector_size), 0, &bp, buf_ops);
if (error) {
if (loud)
xfs_warn(mp, "SB validate failed with error %d.", error);
/* bad CRC means corrupted metadata */
if (error == -EFSBADCRC)
error = -EFSCORRUPTED;
return error;
}
/*
* Initialize the mount structure from the superblock.
*/
xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
/*
* If we haven't validated the superblock, do so now before we try
* to check the sector size and reread the superblock appropriately.
*/
if (sbp->sb_magicnum != XFS_SB_MAGIC) {
if (loud)
xfs_warn(mp, "Invalid superblock magic number");
error = -EINVAL;
goto release_buf;
}
/*
* We must be able to do sector-sized and sector-aligned IO.
*/
if (sector_size > sbp->sb_sectsize) {
if (loud)
xfs_warn(mp, "device supports %u byte sectors (not %u)",
sector_size, sbp->sb_sectsize);
error = -ENOSYS;
goto release_buf;
}
if (buf_ops == NULL) {
/*
* Re-read the superblock so the buffer is correctly sized,
* and properly verified.
*/
xfs_buf_relse(bp);
sector_size = sbp->sb_sectsize;
buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
goto reread;
}
/* Initialize per-cpu counters */
xfs_icsb_reinit_counters(mp);
/* no need to be quiet anymore, so reset the buf ops */
bp->b_ops = &xfs_sb_buf_ops;
mp->m_sb_bp = bp;
xfs_buf_unlock(bp);
return 0;
release_buf:
xfs_buf_relse(bp);
return error;
}
/*
* Update alignment values based on mount options and sb values
*/
STATIC int
xfs_update_alignment(xfs_mount_t *mp)
{
xfs_sb_t *sbp = &(mp->m_sb);
if (mp->m_dalign) {
/*
* If stripe unit and stripe width are not multiples
* of the fs blocksize turn off alignment.
*/
if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
(BBTOB(mp->m_swidth) & mp->m_blockmask)) {
xfs_warn(mp,
"alignment check failed: sunit/swidth vs. blocksize(%d)",
sbp->sb_blocksize);
return -EINVAL;
} else {
/*
* Convert the stripe unit and width to FSBs.
*/
mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
xfs_warn(mp,
"alignment check failed: sunit/swidth vs. agsize(%d)",
sbp->sb_agblocks);
return -EINVAL;
} else if (mp->m_dalign) {
mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
} else {
xfs_warn(mp,
"alignment check failed: sunit(%d) less than bsize(%d)",
mp->m_dalign, sbp->sb_blocksize);
return -EINVAL;
}
}
/*
* Update superblock with new values
* and log changes
*/
if (xfs_sb_version_hasdalign(sbp)) {
if (sbp->sb_unit != mp->m_dalign) {
sbp->sb_unit = mp->m_dalign;
mp->m_update_flags |= XFS_SB_UNIT;
}
if (sbp->sb_width != mp->m_swidth) {
sbp->sb_width = mp->m_swidth;
mp->m_update_flags |= XFS_SB_WIDTH;
}
xfs: Don't keep silent if sunit/swidth can not be changed via mount As per the mount man page, sunit and swidth can be changed via mount options. For XFS, on the face of it, those options seems works if the specified alignments is properly, e.g. # mount -o sunit=4096,swidth=8192 /dev/sdb1 /mnt # mount | grep sdb1 /dev/sdb1 on /mnt type xfs (rw,sunit=4096,swidth=8192) However, neither sunit nor swidth is shown from the xfs_info output. # xfs_info /mnt meta-data=/dev/sdb1 isize=256 agcount=4, agsize=262144 blks = sectsz=512 attr=2 data = bsize=4096 blocks=1048576, imaxpct=25 = sunit=0 swidth=0 blks ^^^^^^^^^^^^^^^^^^^^^^^^^^ naming =version 2 bsize=4096 ascii-ci=0 log =internal bsize=4096 blocks=2560, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =none extsz=4096 blocks=0, rtextents=0 The reason is that the alignment can only be changed if the relevant super block is already configured with alignments, otherwise, the given value is silently ignored. With this fix, the attempt to mount a storage without strip alignment setup on a super block will get an error with a warning in syslog to indicate the true cause, e.g. # mount -o sunit=4096,swidth=8192 /dev/sdb1 /mnt mount: wrong fs type, bad option, bad superblock on /dev/sdb1, missing codepage or helper program, or other error In some cases useful info is found in syslog - try dmesg | tail or so ....... XFS (sdb1): cannot change alignment: superblock does not support data alignment Signed-off-by: Jie Liu <jeff.liu@oracle.com> Cc: Mark Tinguely <tinguely@sgi.com> Cc: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-05-02 18:27:53 +07:00
} else {
xfs_warn(mp,
"cannot change alignment: superblock does not support data alignment");
return -EINVAL;
}
} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
xfs_sb_version_hasdalign(&mp->m_sb)) {
mp->m_dalign = sbp->sb_unit;
mp->m_swidth = sbp->sb_width;
}
return 0;
}
/*
* Set the maximum inode count for this filesystem
*/
STATIC void
xfs_set_maxicount(xfs_mount_t *mp)
{
xfs_sb_t *sbp = &(mp->m_sb);
__uint64_t icount;
if (sbp->sb_imax_pct) {
/*
* Make sure the maximum inode count is a multiple
* of the units we allocate inodes in.
*/
icount = sbp->sb_dblocks * sbp->sb_imax_pct;
do_div(icount, 100);
do_div(icount, mp->m_ialloc_blks);
mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
sbp->sb_inopblog;
} else {
mp->m_maxicount = 0;
}
}
/*
* Set the default minimum read and write sizes unless
* already specified in a mount option.
* We use smaller I/O sizes when the file system
* is being used for NFS service (wsync mount option).
*/
STATIC void
xfs_set_rw_sizes(xfs_mount_t *mp)
{
xfs_sb_t *sbp = &(mp->m_sb);
int readio_log, writeio_log;
if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
if (mp->m_flags & XFS_MOUNT_WSYNC) {
readio_log = XFS_WSYNC_READIO_LOG;
writeio_log = XFS_WSYNC_WRITEIO_LOG;
} else {
readio_log = XFS_READIO_LOG_LARGE;
writeio_log = XFS_WRITEIO_LOG_LARGE;
}
} else {
readio_log = mp->m_readio_log;
writeio_log = mp->m_writeio_log;
}
if (sbp->sb_blocklog > readio_log) {
mp->m_readio_log = sbp->sb_blocklog;
} else {
mp->m_readio_log = readio_log;
}
mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
if (sbp->sb_blocklog > writeio_log) {
mp->m_writeio_log = sbp->sb_blocklog;
} else {
mp->m_writeio_log = writeio_log;
}
mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
}
xfs: dynamic speculative EOF preallocation Currently the size of the speculative preallocation during delayed allocation is fixed by either the allocsize mount option of a default size. We are seeing a lot of cases where we need to recommend using the allocsize mount option to prevent fragmentation when buffered writes land in the same AG. Rather than using a fixed preallocation size by default (up to 64k), make it dynamic by basing it on the current inode size. That way the EOF preallocation will increase as the file size increases. Hence for streaming writes we are much more likely to get large preallocations exactly when we need it to reduce fragementation. For default settings, the size of the initial extents is determined by the number of parallel writers and the amount of memory in the machine. For 4GB RAM and 4 concurrent 32GB file writes: EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET TOTAL 0: [0..1048575]: 1048672..2097247 0 (1048672..2097247) 1048576 1: [1048576..2097151]: 5242976..6291551 0 (5242976..6291551) 1048576 2: [2097152..4194303]: 12583008..14680159 0 (12583008..14680159) 2097152 3: [4194304..8388607]: 25165920..29360223 0 (25165920..29360223) 4194304 4: [8388608..16777215]: 58720352..67108959 0 (58720352..67108959) 8388608 5: [16777216..33554423]: 117440584..134217791 0 (117440584..134217791) 16777208 6: [33554424..50331511]: 184549056..201326143 0 (184549056..201326143) 16777088 7: [50331512..67108599]: 251657408..268434495 0 (251657408..268434495) 16777088 and for 16 concurrent 16GB file writes: EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET TOTAL 0: [0..262143]: 2490472..2752615 0 (2490472..2752615) 262144 1: [262144..524287]: 6291560..6553703 0 (6291560..6553703) 262144 2: [524288..1048575]: 13631592..14155879 0 (13631592..14155879) 524288 3: [1048576..2097151]: 30408808..31457383 0 (30408808..31457383) 1048576 4: [2097152..4194303]: 52428904..54526055 0 (52428904..54526055) 2097152 5: [4194304..8388607]: 104857704..109052007 0 (104857704..109052007) 4194304 6: [8388608..16777215]: 209715304..218103911 0 (209715304..218103911) 8388608 7: [16777216..33554423]: 452984848..469762055 0 (452984848..469762055) 16777208 Because it is hard to take back specualtive preallocation, cases where there are large slow growing log files on a nearly full filesystem may cause premature ENOSPC. Hence as the filesystem nears full, the maximum dynamic prealloc size іs reduced according to this table (based on 4k block size): freespace max prealloc size >5% full extent (8GB) 4-5% 2GB (8GB >> 2) 3-4% 1GB (8GB >> 3) 2-3% 512MB (8GB >> 4) 1-2% 256MB (8GB >> 5) <1% 128MB (8GB >> 6) This should reduce the amount of space held in speculative preallocation for such cases. The allocsize mount option turns off the dynamic behaviour and fixes the prealloc size to whatever the mount option specifies. i.e. the behaviour is unchanged. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2011-01-04 07:35:03 +07:00
/*
* precalculate the low space thresholds for dynamic speculative preallocation.
*/
void
xfs_set_low_space_thresholds(
struct xfs_mount *mp)
{
int i;
for (i = 0; i < XFS_LOWSP_MAX; i++) {
__uint64_t space = mp->m_sb.sb_dblocks;
do_div(space, 100);
mp->m_low_space[i] = space * (i + 1);
}
}
/*
* Set whether we're using inode alignment.
*/
STATIC void
xfs_set_inoalignment(xfs_mount_t *mp)
{
if (xfs_sb_version_hasalign(&mp->m_sb) &&
mp->m_sb.sb_inoalignmt >=
XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
else
mp->m_inoalign_mask = 0;
/*
* If we are using stripe alignment, check whether
* the stripe unit is a multiple of the inode alignment
*/
if (mp->m_dalign && mp->m_inoalign_mask &&
!(mp->m_dalign & mp->m_inoalign_mask))
mp->m_sinoalign = mp->m_dalign;
else
mp->m_sinoalign = 0;
}
/*
* Check that the data (and log if separate) is an ok size.
*/
STATIC int
xfs_check_sizes(
struct xfs_mount *mp)
{
struct xfs_buf *bp;
xfs_daddr_t d;
int error;
d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
xfs_warn(mp, "filesystem size mismatch detected");
return -EFBIG;
}
error = xfs_buf_read_uncached(mp->m_ddev_targp,
d - XFS_FSS_TO_BB(mp, 1),
XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
if (error) {
xfs_warn(mp, "last sector read failed");
return error;
}
xfs_buf_relse(bp);
if (mp->m_logdev_targp == mp->m_ddev_targp)
return 0;
d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
xfs_warn(mp, "log size mismatch detected");
return -EFBIG;
}
error = xfs_buf_read_uncached(mp->m_logdev_targp,
d - XFS_FSB_TO_BB(mp, 1),
XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
if (error) {
xfs_warn(mp, "log device read failed");
return error;
}
xfs_buf_relse(bp);
return 0;
}
/*
* Clear the quotaflags in memory and in the superblock.
*/
int
xfs_mount_reset_sbqflags(
struct xfs_mount *mp)
{
int error;
struct xfs_trans *tp;
mp->m_qflags = 0;
/*
* It is OK to look at sb_qflags here in mount path,
* without m_sb_lock.
*/
if (mp->m_sb.sb_qflags == 0)
return 0;
spin_lock(&mp->m_sb_lock);
mp->m_sb.sb_qflags = 0;
spin_unlock(&mp->m_sb_lock);
/*
* If the fs is readonly, let the incore superblock run
* with quotas off but don't flush the update out to disk
*/
if (mp->m_flags & XFS_MOUNT_RDONLY)
return 0;
tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
if (error) {
xfs_trans_cancel(tp, 0);
xfs_alert(mp, "%s: Superblock update failed!", __func__);
return error;
}
xfs_mod_sb(tp, XFS_SB_QFLAGS);
return xfs_trans_commit(tp, 0);
}
__uint64_t
xfs_default_resblks(xfs_mount_t *mp)
{
__uint64_t resblks;
/*
* We default to 5% or 8192 fsbs of space reserved, whichever is
* smaller. This is intended to cover concurrent allocation
* transactions when we initially hit enospc. These each require a 4
* block reservation. Hence by default we cover roughly 2000 concurrent
* allocation reservations.
*/
resblks = mp->m_sb.sb_dblocks;
do_div(resblks, 20);
resblks = min_t(__uint64_t, resblks, 8192);
return resblks;
}
/*
* This function does the following on an initial mount of a file system:
* - reads the superblock from disk and init the mount struct
* - if we're a 32-bit kernel, do a size check on the superblock
* so we don't mount terabyte filesystems
* - init mount struct realtime fields
* - allocate inode hash table for fs
* - init directory manager
* - perform recovery and init the log manager
*/
int
xfs_mountfs(
xfs_mount_t *mp)
{
xfs_sb_t *sbp = &(mp->m_sb);
xfs_inode_t *rip;
__uint64_t resblks;
uint quotamount = 0;
uint quotaflags = 0;
int error = 0;
xfs_sb_mount_common(mp, sbp);
/*
* Check for a mismatched features2 values. Older kernels
* read & wrote into the wrong sb offset for sb_features2
* on some platforms due to xfs_sb_t not being 64bit size aligned
* when sb_features2 was added, which made older superblock
* reading/writing routines swap it as a 64-bit value.
*
* For backwards compatibility, we make both slots equal.
*
* If we detect a mismatched field, we OR the set bits into the
* existing features2 field in case it has already been modified; we
* don't want to lose any features. We then update the bad location
* with the ORed value so that older kernels will see any features2
* flags, and mark the two fields as needing updates once the
* transaction subsystem is online.
*/
if (xfs_sb_has_mismatched_features2(sbp)) {
xfs_warn(mp, "correcting sb_features alignment problem");
sbp->sb_features2 |= sbp->sb_bad_features2;
sbp->sb_bad_features2 = sbp->sb_features2;
mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
/*
* Re-check for ATTR2 in case it was found in bad_features2
* slot.
*/
if (xfs_sb_version_hasattr2(&mp->m_sb) &&
!(mp->m_flags & XFS_MOUNT_NOATTR2))
mp->m_flags |= XFS_MOUNT_ATTR2;
}
if (xfs_sb_version_hasattr2(&mp->m_sb) &&
(mp->m_flags & XFS_MOUNT_NOATTR2)) {
xfs_sb_version_removeattr2(&mp->m_sb);
mp->m_update_flags |= XFS_SB_FEATURES2;
/* update sb_versionnum for the clearing of the morebits */
if (!sbp->sb_features2)
mp->m_update_flags |= XFS_SB_VERSIONNUM;
}
/* always use v2 inodes by default now */
if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
mp->m_update_flags |= XFS_SB_VERSIONNUM;
}
/*
* Check if sb_agblocks is aligned at stripe boundary
* If sb_agblocks is NOT aligned turn off m_dalign since
* allocator alignment is within an ag, therefore ag has
* to be aligned at stripe boundary.
*/
error = xfs_update_alignment(mp);
if (error)
goto out;
xfs_alloc_compute_maxlevels(mp);
xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
xfs_ialloc_compute_maxlevels(mp);
xfs_set_maxicount(mp);
error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
if (error)
goto out;
error = xfs_uuid_mount(mp);
if (error)
goto out_remove_sysfs;
/*
* Set the minimum read and write sizes
*/
xfs_set_rw_sizes(mp);
xfs: dynamic speculative EOF preallocation Currently the size of the speculative preallocation during delayed allocation is fixed by either the allocsize mount option of a default size. We are seeing a lot of cases where we need to recommend using the allocsize mount option to prevent fragmentation when buffered writes land in the same AG. Rather than using a fixed preallocation size by default (up to 64k), make it dynamic by basing it on the current inode size. That way the EOF preallocation will increase as the file size increases. Hence for streaming writes we are much more likely to get large preallocations exactly when we need it to reduce fragementation. For default settings, the size of the initial extents is determined by the number of parallel writers and the amount of memory in the machine. For 4GB RAM and 4 concurrent 32GB file writes: EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET TOTAL 0: [0..1048575]: 1048672..2097247 0 (1048672..2097247) 1048576 1: [1048576..2097151]: 5242976..6291551 0 (5242976..6291551) 1048576 2: [2097152..4194303]: 12583008..14680159 0 (12583008..14680159) 2097152 3: [4194304..8388607]: 25165920..29360223 0 (25165920..29360223) 4194304 4: [8388608..16777215]: 58720352..67108959 0 (58720352..67108959) 8388608 5: [16777216..33554423]: 117440584..134217791 0 (117440584..134217791) 16777208 6: [33554424..50331511]: 184549056..201326143 0 (184549056..201326143) 16777088 7: [50331512..67108599]: 251657408..268434495 0 (251657408..268434495) 16777088 and for 16 concurrent 16GB file writes: EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET TOTAL 0: [0..262143]: 2490472..2752615 0 (2490472..2752615) 262144 1: [262144..524287]: 6291560..6553703 0 (6291560..6553703) 262144 2: [524288..1048575]: 13631592..14155879 0 (13631592..14155879) 524288 3: [1048576..2097151]: 30408808..31457383 0 (30408808..31457383) 1048576 4: [2097152..4194303]: 52428904..54526055 0 (52428904..54526055) 2097152 5: [4194304..8388607]: 104857704..109052007 0 (104857704..109052007) 4194304 6: [8388608..16777215]: 209715304..218103911 0 (209715304..218103911) 8388608 7: [16777216..33554423]: 452984848..469762055 0 (452984848..469762055) 16777208 Because it is hard to take back specualtive preallocation, cases where there are large slow growing log files on a nearly full filesystem may cause premature ENOSPC. Hence as the filesystem nears full, the maximum dynamic prealloc size іs reduced according to this table (based on 4k block size): freespace max prealloc size >5% full extent (8GB) 4-5% 2GB (8GB >> 2) 3-4% 1GB (8GB >> 3) 2-3% 512MB (8GB >> 4) 1-2% 256MB (8GB >> 5) <1% 128MB (8GB >> 6) This should reduce the amount of space held in speculative preallocation for such cases. The allocsize mount option turns off the dynamic behaviour and fixes the prealloc size to whatever the mount option specifies. i.e. the behaviour is unchanged. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2011-01-04 07:35:03 +07:00
/* set the low space thresholds for dynamic preallocation */
xfs_set_low_space_thresholds(mp);
/*
* Set the inode cluster size.
* This may still be overridden by the file system
* block size if it is larger than the chosen cluster size.
xfs: increase inode cluster size for v5 filesystems v5 filesystems use 512 byte inodes as a minimum, so read inodes in clusters that are effectively half the size of a v4 filesystem with 256 byte inodes. For v5 fielsystems, scale the inode cluster size with the size of the inode so that we keep a constant 32 inodes per cluster ratio for all inode IO. This only works if mkfs.xfs sets the inode alignment appropriately for larger inode clusters, so this functionality is made conditional on mkfs doing the right thing. xfs_repair needs to know about the inode alignment changes, too. Wall time: create bulkstat find+stat ls -R unlink v4 237s 161s 173s 201s 299s v5 235s 163s 205s 31s 356s patched 234s 160s 182s 29s 317s System time: create bulkstat find+stat ls -R unlink v4 2601s 2490s 1653s 1656s 2960s v5 2637s 2497s 1681s 20s 3216s patched 2613s 2451s 1658s 20s 3007s So, wall time same or down across the board, system time same or down across the board, and cache hit rates all improve except for the ls -R case which is a pure cold cache directory read workload on v5 filesystems... So, this patch removes most of the performance and CPU usage differential between v4 and v5 filesystems on traversal related workloads. Note: while this patch is currently for v5 filesystems only, there is no reason it can't be ported back to v4 filesystems. This hasn't been done here because bringing the code back to v4 requires forwards and backwards kernel compatibility testing. i.e. to deterine if older kernels(*) do the right thing with larger inode alignments but still only using 8k inode cluster sizes. None of this testing and validation on v4 filesystems has been done, so for the moment larger inode clusters is limited to v5 superblocks. (*) a current default config v4 filesystem should mount just fine on 2.6.23 (when lazy-count support was introduced), and so if we change the alignment emitted by mkfs without a feature bit then we have to make sure it works properly on all kernels since 2.6.23. And if we allow it to be changed when the lazy-count bit is not set, then it's all kernels since v2 logs were introduced that need to be tested for compatibility... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-11-01 11:27:20 +07:00
*
* For v5 filesystems, scale the cluster size with the inode size to
* keep a constant ratio of inode per cluster buffer, but only if mkfs
* has set the inode alignment value appropriately for larger cluster
* sizes.
*/
mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
xfs: increase inode cluster size for v5 filesystems v5 filesystems use 512 byte inodes as a minimum, so read inodes in clusters that are effectively half the size of a v4 filesystem with 256 byte inodes. For v5 fielsystems, scale the inode cluster size with the size of the inode so that we keep a constant 32 inodes per cluster ratio for all inode IO. This only works if mkfs.xfs sets the inode alignment appropriately for larger inode clusters, so this functionality is made conditional on mkfs doing the right thing. xfs_repair needs to know about the inode alignment changes, too. Wall time: create bulkstat find+stat ls -R unlink v4 237s 161s 173s 201s 299s v5 235s 163s 205s 31s 356s patched 234s 160s 182s 29s 317s System time: create bulkstat find+stat ls -R unlink v4 2601s 2490s 1653s 1656s 2960s v5 2637s 2497s 1681s 20s 3216s patched 2613s 2451s 1658s 20s 3007s So, wall time same or down across the board, system time same or down across the board, and cache hit rates all improve except for the ls -R case which is a pure cold cache directory read workload on v5 filesystems... So, this patch removes most of the performance and CPU usage differential between v4 and v5 filesystems on traversal related workloads. Note: while this patch is currently for v5 filesystems only, there is no reason it can't be ported back to v4 filesystems. This hasn't been done here because bringing the code back to v4 requires forwards and backwards kernel compatibility testing. i.e. to deterine if older kernels(*) do the right thing with larger inode alignments but still only using 8k inode cluster sizes. None of this testing and validation on v4 filesystems has been done, so for the moment larger inode clusters is limited to v5 superblocks. (*) a current default config v4 filesystem should mount just fine on 2.6.23 (when lazy-count support was introduced), and so if we change the alignment emitted by mkfs without a feature bit then we have to make sure it works properly on all kernels since 2.6.23. And if we allow it to be changed when the lazy-count bit is not set, then it's all kernels since v2 logs were introduced that need to be tested for compatibility... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-11-01 11:27:20 +07:00
if (xfs_sb_version_hascrc(&mp->m_sb)) {
int new_size = mp->m_inode_cluster_size;
new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
mp->m_inode_cluster_size = new_size;
}
/*
* Set inode alignment fields
*/
xfs_set_inoalignment(mp);
/*
* Check that the data (and log if separate) is an ok size.
*/
error = xfs_check_sizes(mp);
if (error)
goto out_remove_uuid;
/*
* Initialize realtime fields in the mount structure
*/
error = xfs_rtmount_init(mp);
if (error) {
xfs_warn(mp, "RT mount failed");
goto out_remove_uuid;
}
/*
* Copies the low order bits of the timestamp and the randomly
* set "sequence" number out of a UUID.
*/
uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
mp->m_dmevmask = 0; /* not persistent; set after each mount */
error = xfs_da_mount(mp);
if (error) {
xfs_warn(mp, "Failed dir/attr init: %d", error);
goto out_remove_uuid;
}
/*
* Initialize the precomputed transaction reservations values.
*/
xfs_trans_init(mp);
/*
* Allocate and initialize the per-ag data.
*/
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
spin_lock_init(&mp->m_perag_lock);
INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
if (error) {
xfs_warn(mp, "Failed per-ag init: %d", error);
goto out_free_dir;
xfs: Replace per-ag array with a radix tree The use of an array for the per-ag structures requires reallocation of the array when growing the filesystem. This requires locking access to the array to avoid use after free situations, and the locking is difficult to get right. To avoid needing to reallocate an array, change the per-ag structures to an allocated object per ag and index them using a tree structure. The AGs are always densely indexed (hence the use of an array), but the number supported is 2^32 and lookups tend to be random and hence indexing needs to scale. A simple choice is a radix tree - it works well with this sort of index. This change also removes another large contiguous allocation from the mount/growfs path in XFS. The growing process now needs to change to only initialise the new AGs required for the extra space, and as such only needs to exclusively lock the tree for inserts. The rest of the code only needs to lock the tree while doing lookups, and hence this will remove all the deadlocks that currently occur on the m_perag_lock as it is now an innermost lock. The lock is also changed to a spinlock from a read/write lock as the hold time is now extremely short. To complete the picture, the per-ag structures will need to be reference counted to ensure that we don't free/modify them while they are still in use. This will be done in subsequent patch. Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-01-11 18:47:44 +07:00
}
if (!sbp->sb_logblocks) {
xfs_warn(mp, "no log defined");
XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
error = -EFSCORRUPTED;
goto out_free_perag;
}
/*
* log's mount-time initialization. Perform 1st part recovery if needed
*/
error = xfs_log_mount(mp, mp->m_logdev_targp,
XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
if (error) {
xfs_warn(mp, "log mount failed");
goto out_fail_wait;
}
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
/*
* Now the log is mounted, we know if it was an unclean shutdown or
* not. If it was, with the first phase of recovery has completed, we
* have consistent AG blocks on disk. We have not recovered EFIs yet,
* but they are recovered transactionally in the second recovery phase
* later.
*
* Hence we can safely re-initialise incore superblock counters from
* the per-ag data. These may not be correct if the filesystem was not
* cleanly unmounted, so we need to wait for recovery to finish before
* doing this.
*
* If the filesystem was cleanly unmounted, then we can trust the
* values in the superblock to be correct and we don't need to do
* anything here.
*
* If we are currently making the filesystem, the initialisation will
* fail as the perag data is in an undefined state.
*/
if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
!XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
!mp->m_sb.sb_inprogress) {
error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
if (error)
goto out_log_dealloc;
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
}
/*
* Get and sanity-check the root inode.
* Save the pointer to it in the mount structure.
*/
error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
if (error) {
xfs_warn(mp, "failed to read root inode");
goto out_log_dealloc;
}
ASSERT(rip != NULL);
if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
xfs_warn(mp, "corrupted root inode %llu: not a directory",
(unsigned long long)rip->i_ino);
xfs_iunlock(rip, XFS_ILOCK_EXCL);
XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
mp);
error = -EFSCORRUPTED;
goto out_rele_rip;
}
mp->m_rootip = rip; /* save it */
xfs_iunlock(rip, XFS_ILOCK_EXCL);
/*
* Initialize realtime inode pointers in the mount structure
*/
error = xfs_rtmount_inodes(mp);
if (error) {
/*
* Free up the root inode.
*/
xfs_warn(mp, "failed to read RT inodes");
goto out_rele_rip;
}
/*
* If this is a read-only mount defer the superblock updates until
* the next remount into writeable mode. Otherwise we would never
* perform the update e.g. for the root filesystem.
*/
if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
error = xfs_mount_log_sb(mp, mp->m_update_flags);
if (error) {
xfs_warn(mp, "failed to write sb changes");
goto out_rtunmount;
}
}
/*
* Initialise the XFS quota management subsystem for this mount
*/
if (XFS_IS_QUOTA_RUNNING(mp)) {
error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
if (error)
goto out_rtunmount;
} else {
ASSERT(!XFS_IS_QUOTA_ON(mp));
/*
* If a file system had quotas running earlier, but decided to
* mount without -o uquota/pquota/gquota options, revoke the
* quotachecked license.
*/
if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
xfs_notice(mp, "resetting quota flags");
error = xfs_mount_reset_sbqflags(mp);
if (error)
goto out_rtunmount;
}
}
/*
* Finish recovering the file system. This part needed to be
* delayed until after the root and real-time bitmap inodes
* were consistently read in.
*/
error = xfs_log_mount_finish(mp);
if (error) {
xfs_warn(mp, "log mount finish failed");
goto out_rtunmount;
}
/*
* Complete the quota initialisation, post-log-replay component.
*/
if (quotamount) {
ASSERT(mp->m_qflags == 0);
mp->m_qflags = quotaflags;
xfs_qm_mount_quotas(mp);
}
/*
* Now we are mounted, reserve a small amount of unused space for
* privileged transactions. This is needed so that transaction
* space required for critical operations can dip into this pool
* when at ENOSPC. This is needed for operations like create with
* attr, unwritten extent conversion at ENOSPC, etc. Data allocations
* are not allowed to use this reserved space.
*
* This may drive us straight to ENOSPC on mount, but that implies
* we were already there on the last unmount. Warn if this occurs.
*/
if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
resblks = xfs_default_resblks(mp);
error = xfs_reserve_blocks(mp, &resblks, NULL);
if (error)
xfs_warn(mp,
"Unable to allocate reserve blocks. Continuing without reserve pool.");
}
return 0;
out_rtunmount:
xfs_rtunmount_inodes(mp);
out_rele_rip:
IRELE(rip);
out_log_dealloc:
xfs_log_unmount(mp);
out_fail_wait:
if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
xfs_wait_buftarg(mp->m_logdev_targp);
xfs_wait_buftarg(mp->m_ddev_targp);
out_free_perag:
xfs_free_perag(mp);
out_free_dir:
xfs_da_unmount(mp);
out_remove_uuid:
xfs_uuid_unmount(mp);
out_remove_sysfs:
xfs_sysfs_del(&mp->m_kobj);
out:
return error;
}
/*
* This flushes out the inodes,dquots and the superblock, unmounts the
* log and makes sure that incore structures are freed.
*/
void
xfs_unmountfs(
struct xfs_mount *mp)
{
__uint64_t resblks;
int error;
cancel_delayed_work_sync(&mp->m_eofblocks_work);
xfs_qm_unmount_quotas(mp);
xfs_rtunmount_inodes(mp);
IRELE(mp->m_rootip);
/*
* We can potentially deadlock here if we have an inode cluster
* that has been freed has its buffer still pinned in memory because
* the transaction is still sitting in a iclog. The stale inodes
* on that buffer will have their flush locks held until the
* transaction hits the disk and the callbacks run. the inode
* flush takes the flush lock unconditionally and with nothing to
* push out the iclog we will never get that unlocked. hence we
* need to force the log first.
*/
xfs_log_force(mp, XFS_LOG_SYNC);
xfs: Use delayed write for inodes rather than async V2 We currently do background inode flush asynchronously, resulting in inodes being written in whatever order the background writeback issues them. Not only that, there are also blocking and non-blocking asynchronous inode flushes, depending on where the flush comes from. This patch completely removes asynchronous inode writeback. It removes all the strange writeback modes and replaces them with either a synchronous flush or a non-blocking delayed write flush. That is, inode flushes will only issue IO directly if they are synchronous, and background flushing may do nothing if the operation would block (e.g. on a pinned inode or buffer lock). Delayed write flushes will now result in the inode buffer sitting in the delwri queue of the buffer cache to be flushed by either an AIL push or by the xfsbufd timing out the buffer. This will allow accumulation of dirty inode buffers in memory and allow optimisation of inode cluster writeback at the xfsbufd level where we have much greater queue depths than the block layer elevators. We will also get adjacent inode cluster buffer IO merging for free when a later patch in the series allows sorting of the delayed write buffers before dispatch. This effectively means that any inode that is written back by background writeback will be seen as flush locked during AIL pushing, and will result in the buffers being pushed from there. This writeback path is currently non-optimal, but the next patch in the series will fix that problem. A side effect of this delayed write mechanism is that background inode reclaim will no longer directly flush inodes, nor can it wait on the flush lock. The result is that inode reclaim must leave the inode in the reclaimable state until it is clean. Hence attempts to reclaim a dirty inode in the background will simply skip the inode until it is clean and this allows other mechanisms (i.e. xfsbufd) to do more optimal writeback of the dirty buffers. As a result, the inode reclaim code has been rewritten so that it no longer relies on the ambiguous return values of xfs_iflush() to determine whether it is safe to reclaim an inode. Portions of this patch are derived from patches by Christoph Hellwig. Version 2: - cleanup reclaim code as suggested by Christoph - log background reclaim inode flush errors - just pass sync flags to xfs_iflush Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-02-06 08:39:36 +07:00
/*
* Flush all pending changes from the AIL.
*/
xfs_ail_push_all_sync(mp->m_ail);
/*
* And reclaim all inodes. At this point there should be no dirty
* inodes and none should be pinned or locked, but use synchronous
* reclaim just to be sure. We can stop background inode reclaim
* here as well if it is still running.
xfs: Use delayed write for inodes rather than async V2 We currently do background inode flush asynchronously, resulting in inodes being written in whatever order the background writeback issues them. Not only that, there are also blocking and non-blocking asynchronous inode flushes, depending on where the flush comes from. This patch completely removes asynchronous inode writeback. It removes all the strange writeback modes and replaces them with either a synchronous flush or a non-blocking delayed write flush. That is, inode flushes will only issue IO directly if they are synchronous, and background flushing may do nothing if the operation would block (e.g. on a pinned inode or buffer lock). Delayed write flushes will now result in the inode buffer sitting in the delwri queue of the buffer cache to be flushed by either an AIL push or by the xfsbufd timing out the buffer. This will allow accumulation of dirty inode buffers in memory and allow optimisation of inode cluster writeback at the xfsbufd level where we have much greater queue depths than the block layer elevators. We will also get adjacent inode cluster buffer IO merging for free when a later patch in the series allows sorting of the delayed write buffers before dispatch. This effectively means that any inode that is written back by background writeback will be seen as flush locked during AIL pushing, and will result in the buffers being pushed from there. This writeback path is currently non-optimal, but the next patch in the series will fix that problem. A side effect of this delayed write mechanism is that background inode reclaim will no longer directly flush inodes, nor can it wait on the flush lock. The result is that inode reclaim must leave the inode in the reclaimable state until it is clean. Hence attempts to reclaim a dirty inode in the background will simply skip the inode until it is clean and this allows other mechanisms (i.e. xfsbufd) to do more optimal writeback of the dirty buffers. As a result, the inode reclaim code has been rewritten so that it no longer relies on the ambiguous return values of xfs_iflush() to determine whether it is safe to reclaim an inode. Portions of this patch are derived from patches by Christoph Hellwig. Version 2: - cleanup reclaim code as suggested by Christoph - log background reclaim inode flush errors - just pass sync flags to xfs_iflush Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-02-06 08:39:36 +07:00
*/
cancel_delayed_work_sync(&mp->m_reclaim_work);
xfs: Use delayed write for inodes rather than async V2 We currently do background inode flush asynchronously, resulting in inodes being written in whatever order the background writeback issues them. Not only that, there are also blocking and non-blocking asynchronous inode flushes, depending on where the flush comes from. This patch completely removes asynchronous inode writeback. It removes all the strange writeback modes and replaces them with either a synchronous flush or a non-blocking delayed write flush. That is, inode flushes will only issue IO directly if they are synchronous, and background flushing may do nothing if the operation would block (e.g. on a pinned inode or buffer lock). Delayed write flushes will now result in the inode buffer sitting in the delwri queue of the buffer cache to be flushed by either an AIL push or by the xfsbufd timing out the buffer. This will allow accumulation of dirty inode buffers in memory and allow optimisation of inode cluster writeback at the xfsbufd level where we have much greater queue depths than the block layer elevators. We will also get adjacent inode cluster buffer IO merging for free when a later patch in the series allows sorting of the delayed write buffers before dispatch. This effectively means that any inode that is written back by background writeback will be seen as flush locked during AIL pushing, and will result in the buffers being pushed from there. This writeback path is currently non-optimal, but the next patch in the series will fix that problem. A side effect of this delayed write mechanism is that background inode reclaim will no longer directly flush inodes, nor can it wait on the flush lock. The result is that inode reclaim must leave the inode in the reclaimable state until it is clean. Hence attempts to reclaim a dirty inode in the background will simply skip the inode until it is clean and this allows other mechanisms (i.e. xfsbufd) to do more optimal writeback of the dirty buffers. As a result, the inode reclaim code has been rewritten so that it no longer relies on the ambiguous return values of xfs_iflush() to determine whether it is safe to reclaim an inode. Portions of this patch are derived from patches by Christoph Hellwig. Version 2: - cleanup reclaim code as suggested by Christoph - log background reclaim inode flush errors - just pass sync flags to xfs_iflush Signed-off-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-02-06 08:39:36 +07:00
xfs_reclaim_inodes(mp, SYNC_WAIT);
xfs_qm_unmount(mp);
/*
* Unreserve any blocks we have so that when we unmount we don't account
* the reserved free space as used. This is really only necessary for
* lazy superblock counting because it trusts the incore superblock
* counters to be absolutely correct on clean unmount.
*
* We don't bother correcting this elsewhere for lazy superblock
* counting because on mount of an unclean filesystem we reconstruct the
* correct counter value and this is irrelevant.
*
* For non-lazy counter filesystems, this doesn't matter at all because
* we only every apply deltas to the superblock and hence the incore
* value does not matter....
*/
resblks = 0;
error = xfs_reserve_blocks(mp, &resblks, NULL);
if (error)
xfs_warn(mp, "Unable to free reserved block pool. "
"Freespace may not be correct on next mount.");
error = xfs_log_sbcount(mp);
if (error)
xfs_warn(mp, "Unable to update superblock counters. "
"Freespace may not be correct on next mount.");
xfs_log_unmount(mp);
xfs_da_unmount(mp);
xfs_uuid_unmount(mp);
#if defined(DEBUG)
xfs_errortag_clearall(mp, 0);
#endif
xfs_free_perag(mp);
xfs_sysfs_del(&mp->m_kobj);
}
/*
* Determine whether modifications can proceed. The caller specifies the minimum
* freeze level for which modifications should not be allowed. This allows
* certain operations to proceed while the freeze sequence is in progress, if
* necessary.
*/
bool
xfs_fs_writable(
struct xfs_mount *mp,
int level)
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
{
ASSERT(level > SB_UNFROZEN);
if ((mp->m_super->s_writers.frozen >= level) ||
XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
return false;
return true;
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
}
/*
* xfs_log_sbcount
*
* Sync the superblock counters to disk.
*
* Note this code can be called during the process of freezing, so we use the
* transaction allocator that does not block when the transaction subsystem is
* in its frozen state.
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
*/
int
xfs_log_sbcount(xfs_mount_t *mp)
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
{
xfs_trans_t *tp;
int error;
/* allow this to proceed during the freeze sequence... */
if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
return 0;
xfs_icsb_sync_counters(mp, 0);
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
/*
* we don't need to do this if we are updating the superblock
* counters on every modification.
*/
if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
return 0;
tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
if (error) {
xfs_trans_cancel(tp, 0);
return error;
}
xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
xfs_trans_set_sync(tp);
error = xfs_trans_commit(tp, 0);
return error;
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 12:26:31 +07:00
}
/*
* xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
* a delta to a specified field in the in-core superblock. Simply
* switch on the field indicated and apply the delta to that field.
* Fields are not allowed to dip below zero, so if the delta would
* do this do not apply it and return EINVAL.
*
* The m_sb_lock must be held when this routine is called.
*/
STATIC int
xfs_mod_incore_sb_unlocked(
xfs_mount_t *mp,
xfs_sb_field_t field,
int64_t delta,
int rsvd)
{
int scounter; /* short counter for 32 bit fields */
long long lcounter; /* long counter for 64 bit fields */
long long res_used, rem;
/*
* With the in-core superblock spin lock held, switch
* on the indicated field. Apply the delta to the
* proper field. If the fields value would dip below
* 0, then do not apply the delta and return EINVAL.
*/
switch (field) {
case XFS_SBS_ICOUNT:
lcounter = (long long)mp->m_sb.sb_icount;
lcounter += delta;
if (lcounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_icount = lcounter;
return 0;
case XFS_SBS_IFREE:
lcounter = (long long)mp->m_sb.sb_ifree;
lcounter += delta;
if (lcounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_ifree = lcounter;
return 0;
case XFS_SBS_FDBLOCKS:
lcounter = (long long)
mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
if (delta > 0) { /* Putting blocks back */
if (res_used > delta) {
mp->m_resblks_avail += delta;
} else {
rem = delta - res_used;
mp->m_resblks_avail = mp->m_resblks;
lcounter += rem;
}
} else { /* Taking blocks away */
lcounter += delta;
if (lcounter >= 0) {
mp->m_sb.sb_fdblocks = lcounter +
XFS_ALLOC_SET_ASIDE(mp);
return 0;
}
/*
* We are out of blocks, use any available reserved
* blocks if were allowed to.
*/
if (!rsvd)
return -ENOSPC;
lcounter = (long long)mp->m_resblks_avail + delta;
if (lcounter >= 0) {
mp->m_resblks_avail = lcounter;
return 0;
}
printk_once(KERN_WARNING
"Filesystem \"%s\": reserve blocks depleted! "
"Consider increasing reserve pool size.",
mp->m_fsname);
return -ENOSPC;
}
mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
return 0;
case XFS_SBS_FREXTENTS:
lcounter = (long long)mp->m_sb.sb_frextents;
lcounter += delta;
if (lcounter < 0) {
return -ENOSPC;
}
mp->m_sb.sb_frextents = lcounter;
return 0;
case XFS_SBS_DBLOCKS:
lcounter = (long long)mp->m_sb.sb_dblocks;
lcounter += delta;
if (lcounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_dblocks = lcounter;
return 0;
case XFS_SBS_AGCOUNT:
scounter = mp->m_sb.sb_agcount;
scounter += delta;
if (scounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_agcount = scounter;
return 0;
case XFS_SBS_IMAX_PCT:
scounter = mp->m_sb.sb_imax_pct;
scounter += delta;
if (scounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_imax_pct = scounter;
return 0;
case XFS_SBS_REXTSIZE:
scounter = mp->m_sb.sb_rextsize;
scounter += delta;
if (scounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_rextsize = scounter;
return 0;
case XFS_SBS_RBMBLOCKS:
scounter = mp->m_sb.sb_rbmblocks;
scounter += delta;
if (scounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_rbmblocks = scounter;
return 0;
case XFS_SBS_RBLOCKS:
lcounter = (long long)mp->m_sb.sb_rblocks;
lcounter += delta;
if (lcounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_rblocks = lcounter;
return 0;
case XFS_SBS_REXTENTS:
lcounter = (long long)mp->m_sb.sb_rextents;
lcounter += delta;
if (lcounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_rextents = lcounter;
return 0;
case XFS_SBS_REXTSLOG:
scounter = mp->m_sb.sb_rextslog;
scounter += delta;
if (scounter < 0) {
ASSERT(0);
return -EINVAL;
}
mp->m_sb.sb_rextslog = scounter;
return 0;
default:
ASSERT(0);
return -EINVAL;
}
}
/*
* xfs_mod_incore_sb() is used to change a field in the in-core
* superblock structure by the specified delta. This modification
* is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
* routine to do the work.
*/
int
xfs_mod_incore_sb(
struct xfs_mount *mp,
xfs_sb_field_t field,
int64_t delta,
int rsvd)
{
int status;
#ifdef HAVE_PERCPU_SB
ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
#endif
spin_lock(&mp->m_sb_lock);
status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
spin_unlock(&mp->m_sb_lock);
return status;
}
/*
* Change more than one field in the in-core superblock structure at a time.
*
* The fields and changes to those fields are specified in the array of
* xfs_mod_sb structures passed in. Either all of the specified deltas
* will be applied or none of them will. If any modified field dips below 0,
* then all modifications will be backed out and EINVAL will be returned.
*
* Note that this function may not be used for the superblock values that
* are tracked with the in-memory per-cpu counters - a direct call to
* xfs_icsb_modify_counters is required for these.
*/
int
xfs_mod_incore_sb_batch(
struct xfs_mount *mp,
xfs_mod_sb_t *msb,
uint nmsb,
int rsvd)
{
xfs_mod_sb_t *msbp;
int error = 0;
/*
* Loop through the array of mod structures and apply each individually.
* If any fail, then back out all those which have already been applied.
* Do all of this within the scope of the m_sb_lock so that all of the
* changes will be atomic.
*/
spin_lock(&mp->m_sb_lock);
for (msbp = msb; msbp < (msb + nmsb); msbp++) {
ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
msbp->msb_field > XFS_SBS_FDBLOCKS);
error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
msbp->msb_delta, rsvd);
if (error)
goto unwind;
}
spin_unlock(&mp->m_sb_lock);
return 0;
unwind:
while (--msbp >= msb) {
error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
-msbp->msb_delta, rsvd);
ASSERT(error == 0);
}
spin_unlock(&mp->m_sb_lock);
return error;
}
/*
* xfs_getsb() is called to obtain the buffer for the superblock.
* The buffer is returned locked and read in from disk.
* The buffer should be released with a call to xfs_brelse().
*
* If the flags parameter is BUF_TRYLOCK, then we'll only return
* the superblock buffer if it can be locked without sleeping.
* If it can't then we'll return NULL.
*/
struct xfs_buf *
xfs_getsb(
struct xfs_mount *mp,
int flags)
{
struct xfs_buf *bp = mp->m_sb_bp;
if (!xfs_buf_trylock(bp)) {
if (flags & XBF_TRYLOCK)
return NULL;
xfs_buf_lock(bp);
}
xfs_buf_hold(bp);
ASSERT(XFS_BUF_ISDONE(bp));
return bp;
}
/*
* Used to free the superblock along various error paths.
*/
void
xfs_freesb(
struct xfs_mount *mp)
{
struct xfs_buf *bp = mp->m_sb_bp;
xfs_buf_lock(bp);
mp->m_sb_bp = NULL;
xfs_buf_relse(bp);
}
/*
* Used to log changes to the superblock unit and width fields which could
* be altered by the mount options, as well as any potential sb_features2
* fixup. Only the first superblock is updated.
*/
int
xfs_mount_log_sb(
xfs_mount_t *mp,
__int64_t fields)
{
xfs_trans_t *tp;
int error;
ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
XFS_SB_VERSIONNUM));
tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
if (error) {
xfs_trans_cancel(tp, 0);
return error;
}
xfs_mod_sb(tp, fields);
error = xfs_trans_commit(tp, 0);
return error;
}
/*
* If the underlying (data/log/rt) device is readonly, there are some
* operations that cannot proceed.
*/
int
xfs_dev_is_read_only(
struct xfs_mount *mp,
char *message)
{
if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
xfs_readonly_buftarg(mp->m_logdev_targp) ||
(mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
xfs_notice(mp, "%s required on read-only device.", message);
xfs_notice(mp, "write access unavailable, cannot proceed.");
return -EROFS;
}
return 0;
}
#ifdef HAVE_PERCPU_SB
/*
* Per-cpu incore superblock counters
*
* Simple concept, difficult implementation
*
* Basically, replace the incore superblock counters with a distributed per cpu
* counter for contended fields (e.g. free block count).
*
* Difficulties arise in that the incore sb is used for ENOSPC checking, and
* hence needs to be accurately read when we are running low on space. Hence
* there is a method to enable and disable the per-cpu counters based on how
* much "stuff" is available in them.
*
* Basically, a counter is enabled if there is enough free resource to justify
* running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
* ENOSPC), then we disable the counters to synchronise all callers and
* re-distribute the available resources.
*
* If, once we redistributed the available resources, we still get a failure,
* we disable the per-cpu counter and go through the slow path.
*
* The slow path is the current xfs_mod_incore_sb() function. This means that
* when we disable a per-cpu counter, we need to drain its resources back to
* the global superblock. We do this after disabling the counter to prevent
* more threads from queueing up on the counter.
*
* Essentially, this means that we still need a lock in the fast path to enable
* synchronisation between the global counters and the per-cpu counters. This
* is not a problem because the lock will be local to a CPU almost all the time
* and have little contention except when we get to ENOSPC conditions.
*
* Basically, this lock becomes a barrier that enables us to lock out the fast
* path while we do things like enabling and disabling counters and
* synchronising the counters.
*
* Locking rules:
*
* 1. m_sb_lock before picking up per-cpu locks
* 2. per-cpu locks always picked up via for_each_online_cpu() order
* 3. accurate counter sync requires m_sb_lock + per cpu locks
* 4. modifying per-cpu counters requires holding per-cpu lock
* 5. modifying global counters requires holding m_sb_lock
* 6. enabling or disabling a counter requires holding the m_sb_lock
* and _none_ of the per-cpu locks.
*
* Disabled counters are only ever re-enabled by a balance operation
* that results in more free resources per CPU than a given threshold.
* To ensure counters don't remain disabled, they are rebalanced when
* the global resource goes above a higher threshold (i.e. some hysteresis
* is present to prevent thrashing).
*/
#ifdef CONFIG_HOTPLUG_CPU
/*
* hot-plug CPU notifier support.
*
* We need a notifier per filesystem as we need to be able to identify
* the filesystem to balance the counters out. This is achieved by
* having a notifier block embedded in the xfs_mount_t and doing pointer
* magic to get the mount pointer from the notifier block address.
*/
STATIC int
xfs_icsb_cpu_notify(
struct notifier_block *nfb,
unsigned long action,
void *hcpu)
{
xfs_icsb_cnts_t *cntp;
xfs_mount_t *mp;
mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
cntp = (xfs_icsb_cnts_t *)
per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
switch (action) {
case CPU_UP_PREPARE:
case CPU_UP_PREPARE_FROZEN:
/* Easy Case - initialize the area and locks, and
* then rebalance when online does everything else for us. */
memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
break;
case CPU_ONLINE:
case CPU_ONLINE_FROZEN:
xfs_icsb_lock(mp);
xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
xfs_icsb_unlock(mp);
break;
case CPU_DEAD:
case CPU_DEAD_FROZEN:
/* Disable all the counters, then fold the dead cpu's
* count into the total on the global superblock and
* re-enable the counters. */
xfs_icsb_lock(mp);
spin_lock(&mp->m_sb_lock);
xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
mp->m_sb.sb_icount += cntp->icsb_icount;
mp->m_sb.sb_ifree += cntp->icsb_ifree;
mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
spin_unlock(&mp->m_sb_lock);
xfs_icsb_unlock(mp);
break;
}
return NOTIFY_OK;
}
#endif /* CONFIG_HOTPLUG_CPU */
int
xfs_icsb_init_counters(
xfs_mount_t *mp)
{
xfs_icsb_cnts_t *cntp;
int i;
mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
if (mp->m_sb_cnts == NULL)
return -ENOMEM;
for_each_online_cpu(i) {
cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
}
mutex_init(&mp->m_icsb_mutex);
/*
* start with all counters disabled so that the
* initial balance kicks us off correctly
*/
mp->m_icsb_counters = -1;
#ifdef CONFIG_HOTPLUG_CPU
mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
mp->m_icsb_notifier.priority = 0;
register_hotcpu_notifier(&mp->m_icsb_notifier);
#endif /* CONFIG_HOTPLUG_CPU */
return 0;
}
void
xfs_icsb_reinit_counters(
xfs_mount_t *mp)
{
xfs_icsb_lock(mp);
/*
* start with all counters disabled so that the
* initial balance kicks us off correctly
*/
mp->m_icsb_counters = -1;
xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
xfs_icsb_unlock(mp);
}
void
xfs_icsb_destroy_counters(
xfs_mount_t *mp)
{
if (mp->m_sb_cnts) {
unregister_hotcpu_notifier(&mp->m_icsb_notifier);
free_percpu(mp->m_sb_cnts);
}
mutex_destroy(&mp->m_icsb_mutex);
}
STATIC void
xfs_icsb_lock_cntr(
xfs_icsb_cnts_t *icsbp)
{
while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
ndelay(1000);
}
}
STATIC void
xfs_icsb_unlock_cntr(
xfs_icsb_cnts_t *icsbp)
{
clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
}
STATIC void
xfs_icsb_lock_all_counters(
xfs_mount_t *mp)
{
xfs_icsb_cnts_t *cntp;
int i;
for_each_online_cpu(i) {
cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
xfs_icsb_lock_cntr(cntp);
}
}
STATIC void
xfs_icsb_unlock_all_counters(
xfs_mount_t *mp)
{
xfs_icsb_cnts_t *cntp;
int i;
for_each_online_cpu(i) {
cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
xfs_icsb_unlock_cntr(cntp);
}
}
STATIC void
xfs_icsb_count(
xfs_mount_t *mp,
xfs_icsb_cnts_t *cnt,
int flags)
{
xfs_icsb_cnts_t *cntp;
int i;
memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
if (!(flags & XFS_ICSB_LAZY_COUNT))
xfs_icsb_lock_all_counters(mp);
for_each_online_cpu(i) {
cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
cnt->icsb_icount += cntp->icsb_icount;
cnt->icsb_ifree += cntp->icsb_ifree;
cnt->icsb_fdblocks += cntp->icsb_fdblocks;
}
if (!(flags & XFS_ICSB_LAZY_COUNT))
xfs_icsb_unlock_all_counters(mp);
}
STATIC int
xfs_icsb_counter_disabled(
xfs_mount_t *mp,
xfs_sb_field_t field)
{
ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
return test_bit(field, &mp->m_icsb_counters);
}
STATIC void
xfs_icsb_disable_counter(
xfs_mount_t *mp,
xfs_sb_field_t field)
{
xfs_icsb_cnts_t cnt;
ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
/*
* If we are already disabled, then there is nothing to do
* here. We check before locking all the counters to avoid
* the expensive lock operation when being called in the
* slow path and the counter is already disabled. This is
* safe because the only time we set or clear this state is under
* the m_icsb_mutex.
*/
if (xfs_icsb_counter_disabled(mp, field))
return;
xfs_icsb_lock_all_counters(mp);
if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
/* drain back to superblock */
xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
switch(field) {
case XFS_SBS_ICOUNT:
mp->m_sb.sb_icount = cnt.icsb_icount;
break;
case XFS_SBS_IFREE:
mp->m_sb.sb_ifree = cnt.icsb_ifree;
break;
case XFS_SBS_FDBLOCKS:
mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
break;
default:
BUG();
}
}
xfs_icsb_unlock_all_counters(mp);
}
STATIC void
xfs_icsb_enable_counter(
xfs_mount_t *mp,
xfs_sb_field_t field,
uint64_t count,
uint64_t resid)
{
xfs_icsb_cnts_t *cntp;
int i;
ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
xfs_icsb_lock_all_counters(mp);
for_each_online_cpu(i) {
cntp = per_cpu_ptr(mp->m_sb_cnts, i);
switch (field) {
case XFS_SBS_ICOUNT:
cntp->icsb_icount = count + resid;
break;
case XFS_SBS_IFREE:
cntp->icsb_ifree = count + resid;
break;
case XFS_SBS_FDBLOCKS:
cntp->icsb_fdblocks = count + resid;
break;
default:
BUG();
break;
}
resid = 0;
}
clear_bit(field, &mp->m_icsb_counters);
xfs_icsb_unlock_all_counters(mp);
}
void
xfs_icsb_sync_counters_locked(
xfs_mount_t *mp,
int flags)
{
xfs_icsb_cnts_t cnt;
xfs_icsb_count(mp, &cnt, flags);
if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
mp->m_sb.sb_icount = cnt.icsb_icount;
if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
mp->m_sb.sb_ifree = cnt.icsb_ifree;
if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
}
/*
* Accurate update of per-cpu counters to incore superblock
*/
void
xfs_icsb_sync_counters(
xfs_mount_t *mp,
int flags)
{
spin_lock(&mp->m_sb_lock);
xfs_icsb_sync_counters_locked(mp, flags);
spin_unlock(&mp->m_sb_lock);
}
/*
* Balance and enable/disable counters as necessary.
*
* Thresholds for re-enabling counters are somewhat magic. inode counts are
* chosen to be the same number as single on disk allocation chunk per CPU, and
* free blocks is something far enough zero that we aren't going thrash when we
* get near ENOSPC. We also need to supply a minimum we require per cpu to
* prevent looping endlessly when xfs_alloc_space asks for more than will
* be distributed to a single CPU but each CPU has enough blocks to be
* reenabled.
*
* Note that we can be called when counters are already disabled.
* xfs_icsb_disable_counter() optimises the counter locking in this case to
* prevent locking every per-cpu counter needlessly.
*/
#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
(uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
STATIC void
xfs_icsb_balance_counter_locked(
xfs_mount_t *mp,
xfs_sb_field_t field,
int min_per_cpu)
{
uint64_t count, resid;
int weight = num_online_cpus();
uint64_t min = (uint64_t)min_per_cpu;
/* disable counter and sync counter */
xfs_icsb_disable_counter(mp, field);
/* update counters - first CPU gets residual*/
switch (field) {
case XFS_SBS_ICOUNT:
count = mp->m_sb.sb_icount;
resid = do_div(count, weight);
if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
return;
break;
case XFS_SBS_IFREE:
count = mp->m_sb.sb_ifree;
resid = do_div(count, weight);
if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
return;
break;
case XFS_SBS_FDBLOCKS:
count = mp->m_sb.sb_fdblocks;
resid = do_div(count, weight);
if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
return;
break;
default:
BUG();
count = resid = 0; /* quiet, gcc */
break;
}
xfs_icsb_enable_counter(mp, field, count, resid);
}
STATIC void
xfs_icsb_balance_counter(
xfs_mount_t *mp,
xfs_sb_field_t fields,
int min_per_cpu)
{
spin_lock(&mp->m_sb_lock);
xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
spin_unlock(&mp->m_sb_lock);
}
int
xfs_icsb_modify_counters(
xfs_mount_t *mp,
xfs_sb_field_t field,
int64_t delta,
int rsvd)
{
xfs_icsb_cnts_t *icsbp;
long long lcounter; /* long counter for 64 bit fields */
int ret = 0;
might_sleep();
again:
preempt_disable();
icsbp = this_cpu_ptr(mp->m_sb_cnts);
/*
* if the counter is disabled, go to slow path
*/
if (unlikely(xfs_icsb_counter_disabled(mp, field)))
goto slow_path;
xfs_icsb_lock_cntr(icsbp);
if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
xfs_icsb_unlock_cntr(icsbp);
goto slow_path;
}
switch (field) {
case XFS_SBS_ICOUNT:
lcounter = icsbp->icsb_icount;
lcounter += delta;
if (unlikely(lcounter < 0))
goto balance_counter;
icsbp->icsb_icount = lcounter;
break;
case XFS_SBS_IFREE:
lcounter = icsbp->icsb_ifree;
lcounter += delta;
if (unlikely(lcounter < 0))
goto balance_counter;
icsbp->icsb_ifree = lcounter;
break;
case XFS_SBS_FDBLOCKS:
BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
lcounter += delta;
if (unlikely(lcounter < 0))
goto balance_counter;
icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
break;
default:
BUG();
break;
}
xfs_icsb_unlock_cntr(icsbp);
preempt_enable();
return 0;
slow_path:
preempt_enable();
/*
* serialise with a mutex so we don't burn lots of cpu on
* the superblock lock. We still need to hold the superblock
* lock, however, when we modify the global structures.
*/
xfs_icsb_lock(mp);
/*
* Now running atomically.
*
* If the counter is enabled, someone has beaten us to rebalancing.
* Drop the lock and try again in the fast path....
*/
if (!(xfs_icsb_counter_disabled(mp, field))) {
xfs_icsb_unlock(mp);
goto again;
}
/*
* The counter is currently disabled. Because we are
* running atomically here, we know a rebalance cannot
* be in progress. Hence we can go straight to operating
* on the global superblock. We do not call xfs_mod_incore_sb()
* here even though we need to get the m_sb_lock. Doing so
* will cause us to re-enter this function and deadlock.
* Hence we get the m_sb_lock ourselves and then call
* xfs_mod_incore_sb_unlocked() as the unlocked path operates
* directly on the global counters.
*/
spin_lock(&mp->m_sb_lock);
ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
spin_unlock(&mp->m_sb_lock);
/*
* Now that we've modified the global superblock, we
* may be able to re-enable the distributed counters
* (e.g. lots of space just got freed). After that
* we are done.
*/
if (ret != -ENOSPC)
xfs_icsb_balance_counter(mp, field, 0);
xfs_icsb_unlock(mp);
return ret;
balance_counter:
xfs_icsb_unlock_cntr(icsbp);
preempt_enable();
/*
* We may have multiple threads here if multiple per-cpu
* counters run dry at the same time. This will mean we can
* do more balances than strictly necessary but it is not
* the common slowpath case.
*/
xfs_icsb_lock(mp);
/*
* running atomically.
*
* This will leave the counter in the correct state for future
* accesses. After the rebalance, we simply try again and our retry
* will either succeed through the fast path or slow path without
* another balance operation being required.
*/
xfs_icsb_balance_counter(mp, field, delta);
xfs_icsb_unlock(mp);
goto again;
}
#endif