linux/fs/xfs/xfs_mount.h

505 lines
18 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* All Rights Reserved.
*/
#ifndef __XFS_MOUNT_H__
#define __XFS_MOUNT_H__
struct xlog;
struct xfs_inode;
[XFS] Concurrent Multi-File Data Streams In media spaces, video is often stored in a frame-per-file format. When dealing with uncompressed realtime HD video streams in this format, it is crucial that files do not get fragmented and that multiple files a placed contiguously on disk. When multiple streams are being ingested and played out at the same time, it is critical that the filesystem does not cross the streams and interleave them together as this creates seek and readahead cache miss latency and prevents both ingest and playout from meeting frame rate targets. This patch set creates a "stream of files" concept into the allocator to place all the data from a single stream contiguously on disk so that RAID array readahead can be used effectively. Each additional stream gets placed in different allocation groups within the filesystem, thereby ensuring that we don't cross any streams. When an AG fills up, we select a new AG for the stream that is not in use. The core of the functionality is the stream tracking - each inode that we create in a directory needs to be associated with the directories' stream. Hence every time we create a file, we look up the directories' stream object and associate the new file with that object. Once we have a stream object for a file, we use the AG that the stream object point to for allocations. If we can't allocate in that AG (e.g. it is full) we move the entire stream to another AG. Other inodes in the same stream are moved to the new AG on their next allocation (i.e. lazy update). Stream objects are kept in a cache and hold a reference on the inode. Hence the inode cannot be reclaimed while there is an outstanding stream reference. This means that on unlink we need to remove the stream association and we also need to flush all the associations on certain events that want to reclaim all unreferenced inodes (e.g. filesystem freeze). SGI-PV: 964469 SGI-Modid: xfs-linux-melb:xfs-kern:29096a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Barry Naujok <bnaujok@sgi.com> Signed-off-by: Donald Douwsma <donaldd@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com> Signed-off-by: Vlad Apostolov <vapo@sgi.com>
2007-07-11 05:09:12 +04:00
struct xfs_mru_cache;
struct xfs_ail;
struct xfs_quotainfo;
struct xfs_da_geometry;
struct xfs_perag;
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 03:35:03 +03:00
/* dynamic preallocation free space thresholds, 5% down to 1% */
enum {
XFS_LOWSP_1_PCNT = 0,
XFS_LOWSP_2_PCNT,
XFS_LOWSP_3_PCNT,
XFS_LOWSP_4_PCNT,
XFS_LOWSP_5_PCNT,
XFS_LOWSP_MAX,
};
/*
* Error Configuration
*
* Error classes define the subsystem the configuration belongs to.
* Error numbers define the errors that are configurable.
*/
enum {
XFS_ERR_METADATA,
XFS_ERR_CLASS_MAX,
};
enum {
XFS_ERR_DEFAULT,
XFS_ERR_EIO,
XFS_ERR_ENOSPC,
XFS_ERR_ENODEV,
XFS_ERR_ERRNO_MAX,
};
#define XFS_ERR_RETRY_FOREVER -1
/*
* Although retry_timeout is in jiffies which is normally an unsigned long,
* we limit the retry timeout to 86400 seconds, or one day. So even a
* signed 32-bit long is sufficient for a HZ value up to 24855. Making it
* signed lets us store the special "-1" value, meaning retry forever.
*/
struct xfs_error_cfg {
struct xfs_kobj kobj;
int max_retries;
long retry_timeout; /* in jiffies, -1 = infinite */
};
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
/*
* Per-cpu deferred inode inactivation GC lists.
*/
struct xfs_inodegc {
struct llist_head list;
struct work_struct work;
/* approximate count of inodes in the list */
unsigned int items;
unsigned int shrinker_hits;
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
};
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
/*
* The struct xfsmount layout is optimised to separate read-mostly variables
* from variables that are frequently modified. We put the read-mostly variables
* first, then place all the other variables at the end.
*
* Typically, read-mostly variables are those that are set at mount time and
* never changed again, or only change rarely as a result of things like sysfs
* knobs being tweaked.
*/
typedef struct xfs_mount {
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
struct xfs_sb m_sb; /* copy of fs superblock */
struct super_block *m_super;
struct xfs_ail *m_ail; /* fs active log item list */
struct xfs_buf *m_sb_bp; /* buffer for superblock */
char *m_rtname; /* realtime device name */
char *m_logname; /* external log device name */
struct xfs_da_geometry *m_dir_geo; /* directory block geometry */
struct xfs_da_geometry *m_attr_geo; /* attribute block geometry */
struct xlog *m_log; /* log specific stuff */
struct xfs_inode *m_rbmip; /* pointer to bitmap inode */
struct xfs_inode *m_rsumip; /* pointer to summary inode */
struct xfs_inode *m_rootip; /* pointer to root directory */
struct xfs_quotainfo *m_quotainfo; /* disk quota information */
xfs_buftarg_t *m_ddev_targp; /* saves taking the address */
xfs_buftarg_t *m_logdev_targp;/* ptr to log device */
xfs_buftarg_t *m_rtdev_targp; /* ptr to rt device */
struct list_head m_mount_list; /* global mount list */
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
void __percpu *m_inodegc; /* percpu inodegc structures */
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
/*
* Optional cache of rt summary level per bitmap block with the
* invariant that m_rsum_cache[bbno] <= the minimum i for which
* rsum[i][bbno] != 0. Reads and writes are serialized by the rsumip
* inode lock.
*/
uint8_t *m_rsum_cache;
struct xfs_mru_cache *m_filestream; /* per-mount filestream data */
struct workqueue_struct *m_buf_workqueue;
struct workqueue_struct *m_unwritten_workqueue;
struct workqueue_struct *m_reclaim_workqueue;
struct workqueue_struct *m_sync_workqueue;
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
struct workqueue_struct *m_blockgc_wq;
struct workqueue_struct *m_inodegc_wq;
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
int m_bsize; /* fs logical block size */
uint8_t m_blkbit_log; /* blocklog + NBBY */
uint8_t m_blkbb_log; /* blocklog - BBSHIFT */
uint8_t m_agno_log; /* log #ag's */
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
uint8_t m_sectbb_log; /* sectlog - BBSHIFT */
uint m_blockmask; /* sb_blocksize-1 */
uint m_blockwsize; /* sb_blocksize in words */
uint m_blockwmask; /* blockwsize-1 */
uint m_alloc_mxr[2]; /* max alloc btree records */
uint m_alloc_mnr[2]; /* min alloc btree records */
uint m_bmap_dmxr[2]; /* max bmap btree records */
uint m_bmap_dmnr[2]; /* min bmap btree records */
uint m_rmap_mxr[2]; /* max rmap btree records */
uint m_rmap_mnr[2]; /* min rmap btree records */
uint m_refc_mxr[2]; /* max refc btree records */
uint m_refc_mnr[2]; /* min refc btree records */
uint m_ag_maxlevels; /* XFS_AG_MAXLEVELS */
uint m_bm_maxlevels[2]; /* XFS_BM_MAXLEVELS */
uint m_rmap_maxlevels; /* max rmap btree levels */
uint m_refc_maxlevels; /* max refcount btree level */
xfs_extlen_t m_ag_prealloc_blocks; /* reserved ag blocks */
uint m_alloc_set_aside; /* space we can't use */
uint m_ag_max_usable; /* max space per AG */
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
int m_dalign; /* stripe unit */
int m_swidth; /* stripe width */
xfs_agnumber_t m_maxagi; /* highest inode alloc group */
uint m_allocsize_log;/* min write size log bytes */
uint m_allocsize_blocks; /* min write size blocks */
int m_logbufs; /* number of log buffers */
int m_logbsize; /* size of each log buffer */
uint m_rsumlevels; /* rt summary levels */
uint m_rsumsize; /* size of rt summary, bytes */
int m_fixedfsid[2]; /* unchanged for life of FS */
uint m_qflags; /* quota status flags */
uint64_t m_features; /* active filesystem features */
uint64_t m_low_space[XFS_LOWSP_MAX];
uint64_t m_low_rtexts[XFS_LOWSP_MAX];
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
struct xfs_ino_geometry m_ino_geo; /* inode geometry */
struct xfs_trans_resv m_resv; /* precomputed res values */
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
/* low free space thresholds */
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
unsigned long m_opstate; /* dynamic state flags */
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
bool m_always_cow;
bool m_fail_unmount;
bool m_finobt_nores; /* no per-AG finobt resv. */
bool m_update_sb; /* sb needs update in mount */
/*
* Bitsets of per-fs metadata that have been checked and/or are sick.
* Callers must hold m_sb_lock to access these two fields.
*/
uint8_t m_fs_checked;
uint8_t m_fs_sick;
/*
* Bitsets of rt metadata that have been checked and/or are sick.
* Callers must hold m_sb_lock to access this field.
*/
uint8_t m_rt_checked;
uint8_t m_rt_sick;
/*
* End of read-mostly variables. Frequently written variables and locks
* should be placed below this comment from now on. The first variable
* here is marked as cacheline aligned so they it is separated from
* the read-mostly variables.
*/
spinlock_t ____cacheline_aligned m_sb_lock; /* sb counter lock */
struct percpu_counter m_icount; /* allocated inodes counter */
struct percpu_counter m_ifree; /* free inodes counter */
struct percpu_counter m_fdblocks; /* free block counter */
/*
* Count of data device blocks reserved for delayed allocations,
* including indlen blocks. Does not include allocated CoW staging
* extents or anything related to the rt device.
*/
struct percpu_counter m_delalloc_blks;
/*
* Global count of allocation btree blocks in use across all AGs. Only
* used when perag reservation is enabled. Helps prevent block
* reservation from attempting to reserve allocation btree blocks.
*/
atomic64_t m_allocbt_blks;
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
struct radix_tree_root m_perag_tree; /* per-ag accounting info */
spinlock_t m_perag_lock; /* lock for m_perag_tree */
uint64_t m_resblks; /* total reserved blocks */
uint64_t m_resblks_avail;/* available reserved blocks */
uint64_t m_resblks_save; /* reserved blks @ remount,ro */
xfs: introduce background inode reclaim work Background inode reclaim needs to run more frequently that the XFS syncd work is run as 30s is too long between optimal reclaim runs. Add a new periodic work item to the xfs syncd workqueue to run a fast, non-blocking inode reclaim scan. Background inode reclaim is kicked by the act of marking inodes for reclaim. When an AG is first marked as having reclaimable inodes, the background reclaim work is kicked. It will continue to run periodically untill it detects that there are no more reclaimable inodes. It will be kicked again when the first inode is queued for reclaim. To ensure shrinker based inode reclaim throttles to the inode cleaning and reclaim rate but still reclaim inodes efficiently, make it kick the background inode reclaim so that when we are low on memory we are trying to reclaim inodes as efficiently as possible. This kick shoul d not be necessary, but it will protect against failures to kick the background reclaim when inodes are first dirtied. To provide the rate throttling, make the shrinker pass do synchronous inode reclaim so that it blocks on inodes under IO. This means that the shrinker will reclaim inodes rather than just skipping over them, but it does not adversely affect the rate of reclaim because most dirty inodes are already under IO due to the background reclaim work the shrinker kicked. These two modifications solve one of the two OOM killer invocations Chris Mason reported recently when running a stress testing script. The particular workload trigger for the OOM killer invocation is where there are more threads than CPUs all unlinking files in an extremely memory constrained environment. Unlike other solutions, this one does not have a performance impact on performance when memory is not constrained or the number of concurrent threads operating is <= to the number of CPUs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Alex Elder <aelder@sgi.com>
2011-04-08 06:45:07 +04:00
struct delayed_work m_reclaim_work; /* background inode reclaim */
struct xfs_kobj m_kobj;
struct xfs_kobj m_error_kobj;
struct xfs_kobj m_error_meta_kobj;
struct xfs_error_cfg m_error_cfg[XFS_ERR_CLASS_MAX][XFS_ERR_ERRNO_MAX];
struct xstats m_stats; /* per-fs stats */
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
xfs_agnumber_t m_agfrotor; /* last ag where space found */
xfs_agnumber_t m_agirotor; /* last ag dir inode alloced */
spinlock_t m_agirotor_lock;/* .. and lock protecting it */
/* Memory shrinker to throttle and reprioritize inodegc */
struct shrinker m_inodegc_shrinker;
/*
* Workqueue item so that we can coalesce multiple inode flush attempts
* into a single flush.
*/
struct work_struct m_flush_inodes_work;
/*
* Generation of the filesysyem layout. This is incremented by each
* growfs, and used by the pNFS server to ensure the client updates
* its view of the block device once it gets a layout that might
* reference the newly added blocks. Does not need to be persistent
* as long as we only allow file system size increments, but if we
* ever support shrinks it would have to be persisted in addition
* to various other kinds of pain inflicted on the pNFS server.
*/
uint32_t m_generation;
xfs: separate read-only variables in struct xfs_mount Seeing massive cpu usage from xfs_agino_range() on one machine; instruction level profiles look similar to another machine running the same workload, only one machine is consuming 10x as much CPU as the other and going much slower. The only real difference between the two machines is core count per socket. Both are running identical 16p/16GB virtual machine configurations Machine A: 25.83% [k] xfs_agino_range 12.68% [k] __xfs_dir3_data_check 6.95% [k] xfs_verify_ino 6.78% [k] xfs_dir2_data_entry_tag_p 3.56% [k] xfs_buf_find 2.31% [k] xfs_verify_dir_ino 2.02% [k] xfs_dabuf_map.constprop.0 1.65% [k] xfs_ag_block_count And takes around 13 minutes to remove 50 million inodes. Machine B: 13.90% [k] __pv_queued_spin_lock_slowpath 3.76% [k] do_raw_spin_lock 2.83% [k] xfs_dir3_leaf_check_int 2.75% [k] xfs_agino_range 2.51% [k] __raw_callee_save___pv_queued_spin_unlock 2.18% [k] __xfs_dir3_data_check 2.02% [k] xfs_log_commit_cil And takes around 5m30s to remove 50 million inodes. Suspect is cacheline contention on m_sectbb_log which is used in one of the macros in xfs_agino_range. This is a read-only variable but shares a cacheline with m_active_trans which is a global atomic that gets bounced all around the machine. The workload is trying to run hundreds of thousands of transactions per second and hence cacheline contention will be occurring on this atomic counter. Hence xfs_agino_range() is likely just be an innocent bystander as the cache coherency protocol fights over the cacheline between CPU cores and sockets. On machine A, this rearrangement of the struct xfs_mount results in the profile changing to: 9.77% [kernel] [k] xfs_agino_range 6.27% [kernel] [k] __xfs_dir3_data_check 5.31% [kernel] [k] __pv_queued_spin_lock_slowpath 4.54% [kernel] [k] xfs_buf_find 3.79% [kernel] [k] do_raw_spin_lock 3.39% [kernel] [k] xfs_verify_ino 2.73% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock Vastly less CPU usage in xfs_agino_range(), but still 3x the amount of machine B and still runs substantially slower than it should. Current rm -rf of 50 million files: vanilla patched machine A 13m20s 6m42s machine B 5m30s 5m02s It's an improvement, hence indicating that separation and further optimisation of read-only global filesystem data is worthwhile, but it clearly isn't the underlying issue causing this specific performance degradation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-05-20 23:17:11 +03:00
struct mutex m_growlock; /* growfs mutex */
#ifdef DEBUG
/*
* Frequency with which errors are injected. Replaces xfs_etest; the
* value stored in here is the inverse of the frequency with which the
* error triggers. 1 = always, 2 = half the time, etc.
*/
unsigned int *m_errortag;
struct xfs_kobj m_errortag_kobj;
#endif
} xfs_mount_t;
#define M_IGEO(mp) (&(mp)->m_ino_geo)
/*
* Flags for m_features.
*
* These are all the active features in the filesystem, regardless of how
* they are configured.
*/
#define XFS_FEAT_ATTR (1ULL << 0) /* xattrs present in fs */
#define XFS_FEAT_NLINK (1ULL << 1) /* 32 bit link counts */
#define XFS_FEAT_QUOTA (1ULL << 2) /* quota active */
#define XFS_FEAT_ALIGN (1ULL << 3) /* inode alignment */
#define XFS_FEAT_DALIGN (1ULL << 4) /* data alignment */
#define XFS_FEAT_LOGV2 (1ULL << 5) /* version 2 logs */
#define XFS_FEAT_SECTOR (1ULL << 6) /* sector size > 512 bytes */
#define XFS_FEAT_EXTFLG (1ULL << 7) /* unwritten extents */
#define XFS_FEAT_ASCIICI (1ULL << 8) /* ASCII only case-insens. */
#define XFS_FEAT_LAZYSBCOUNT (1ULL << 9) /* Superblk counters */
#define XFS_FEAT_ATTR2 (1ULL << 10) /* dynamic attr fork */
#define XFS_FEAT_PARENT (1ULL << 11) /* parent pointers */
#define XFS_FEAT_PROJID32 (1ULL << 12) /* 32 bit project id */
#define XFS_FEAT_CRC (1ULL << 13) /* metadata CRCs */
#define XFS_FEAT_V3INODES (1ULL << 14) /* Version 3 inodes */
#define XFS_FEAT_PQUOTINO (1ULL << 15) /* non-shared proj/grp quotas */
#define XFS_FEAT_FTYPE (1ULL << 16) /* inode type in dir */
#define XFS_FEAT_FINOBT (1ULL << 17) /* free inode btree */
#define XFS_FEAT_RMAPBT (1ULL << 18) /* reverse map btree */
#define XFS_FEAT_REFLINK (1ULL << 19) /* reflinked files */
#define XFS_FEAT_SPINODES (1ULL << 20) /* sparse inode chunks */
#define XFS_FEAT_META_UUID (1ULL << 21) /* metadata UUID */
#define XFS_FEAT_REALTIME (1ULL << 22) /* realtime device present */
#define XFS_FEAT_INOBTCNT (1ULL << 23) /* inobt block counts */
#define XFS_FEAT_BIGTIME (1ULL << 24) /* large timestamps */
#define XFS_FEAT_NEEDSREPAIR (1ULL << 25) /* needs xfs_repair */
/* Mount features */
#define XFS_FEAT_NOATTR2 (1ULL << 48) /* disable attr2 creation */
#define XFS_FEAT_NOALIGN (1ULL << 49) /* ignore alignment */
#define XFS_FEAT_ALLOCSIZE (1ULL << 50) /* user specified allocation size */
#define XFS_FEAT_LARGE_IOSIZE (1ULL << 51) /* report large preferred
* I/O size in stat() */
#define XFS_FEAT_WSYNC (1ULL << 52) /* synchronous metadata ops */
#define XFS_FEAT_DIRSYNC (1ULL << 53) /* synchronous directory ops */
#define XFS_FEAT_DISCARD (1ULL << 54) /* discard unused blocks */
#define XFS_FEAT_GRPID (1ULL << 55) /* group-ID assigned from directory */
#define XFS_FEAT_SMALL_INUMS (1ULL << 56) /* user wants 32bit inodes */
#define XFS_FEAT_IKEEP (1ULL << 57) /* keep empty inode clusters*/
#define XFS_FEAT_SWALLOC (1ULL << 58) /* stripe width allocation */
#define XFS_FEAT_FILESTREAMS (1ULL << 59) /* use filestreams allocator */
#define XFS_FEAT_DAX_ALWAYS (1ULL << 60) /* DAX always enabled */
#define XFS_FEAT_DAX_NEVER (1ULL << 61) /* DAX never enabled */
#define XFS_FEAT_NORECOVERY (1ULL << 62) /* no recovery - dirty fs */
#define XFS_FEAT_NOUUID (1ULL << 63) /* ignore uuid during mount */
#define __XFS_HAS_FEAT(name, NAME) \
static inline bool xfs_has_ ## name (struct xfs_mount *mp) \
{ \
return mp->m_features & XFS_FEAT_ ## NAME; \
}
/* Some features can be added dynamically so they need a set wrapper, too. */
#define __XFS_ADD_FEAT(name, NAME) \
__XFS_HAS_FEAT(name, NAME); \
static inline void xfs_add_ ## name (struct xfs_mount *mp) \
{ \
mp->m_features |= XFS_FEAT_ ## NAME; \
xfs_sb_version_add ## name(&mp->m_sb); \
}
/* Superblock features */
__XFS_ADD_FEAT(attr, ATTR)
__XFS_HAS_FEAT(nlink, NLINK)
__XFS_ADD_FEAT(quota, QUOTA)
__XFS_HAS_FEAT(align, ALIGN)
__XFS_HAS_FEAT(dalign, DALIGN)
__XFS_HAS_FEAT(logv2, LOGV2)
__XFS_HAS_FEAT(sector, SECTOR)
__XFS_HAS_FEAT(extflg, EXTFLG)
__XFS_HAS_FEAT(asciici, ASCIICI)
__XFS_HAS_FEAT(lazysbcount, LAZYSBCOUNT)
__XFS_ADD_FEAT(attr2, ATTR2)
__XFS_HAS_FEAT(parent, PARENT)
__XFS_ADD_FEAT(projid32, PROJID32)
__XFS_HAS_FEAT(crc, CRC)
__XFS_HAS_FEAT(v3inodes, V3INODES)
__XFS_HAS_FEAT(pquotino, PQUOTINO)
__XFS_HAS_FEAT(ftype, FTYPE)
__XFS_HAS_FEAT(finobt, FINOBT)
__XFS_HAS_FEAT(rmapbt, RMAPBT)
__XFS_HAS_FEAT(reflink, REFLINK)
__XFS_HAS_FEAT(sparseinodes, SPINODES)
__XFS_HAS_FEAT(metauuid, META_UUID)
__XFS_HAS_FEAT(realtime, REALTIME)
__XFS_HAS_FEAT(inobtcounts, INOBTCNT)
__XFS_HAS_FEAT(bigtime, BIGTIME)
__XFS_HAS_FEAT(needsrepair, NEEDSREPAIR)
/*
* Mount features
*
* These do not change dynamically - features that can come and go, such as 32
* bit inodes and read-only state, are kept as operational state rather than
* features.
*/
__XFS_HAS_FEAT(noattr2, NOATTR2)
__XFS_HAS_FEAT(noalign, NOALIGN)
__XFS_HAS_FEAT(allocsize, ALLOCSIZE)
__XFS_HAS_FEAT(large_iosize, LARGE_IOSIZE)
__XFS_HAS_FEAT(wsync, WSYNC)
__XFS_HAS_FEAT(dirsync, DIRSYNC)
__XFS_HAS_FEAT(discard, DISCARD)
__XFS_HAS_FEAT(grpid, GRPID)
__XFS_HAS_FEAT(small_inums, SMALL_INUMS)
__XFS_HAS_FEAT(ikeep, IKEEP)
__XFS_HAS_FEAT(swalloc, SWALLOC)
__XFS_HAS_FEAT(filestreams, FILESTREAMS)
__XFS_HAS_FEAT(dax_always, DAX_ALWAYS)
__XFS_HAS_FEAT(dax_never, DAX_NEVER)
__XFS_HAS_FEAT(norecovery, NORECOVERY)
__XFS_HAS_FEAT(nouuid, NOUUID)
/*
* Operational mount state flags
*
* Use these with atomic bit ops only!
*/
#define XFS_OPSTATE_UNMOUNTING 0 /* filesystem is unmounting */
#define XFS_OPSTATE_CLEAN 1 /* mount was clean */
#define XFS_OPSTATE_SHUTDOWN 2 /* stop all fs operations */
#define XFS_OPSTATE_INODE32 3 /* inode32 allocator active */
#define XFS_OPSTATE_READONLY 4 /* read-only fs */
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
/*
* If set, inactivation worker threads will be scheduled to process queued
* inodegc work. If not, queued inodes remain in memory waiting to be
* processed.
*/
#define XFS_OPSTATE_INODEGC_ENABLED 5
/*
* If set, background speculative prealloc gc worker threads will be scheduled
* to process queued blockgc work. If not, inodes retain their preallocations
* until explicitly deleted.
*/
#define XFS_OPSTATE_BLOCKGC_ENABLED 6
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
#define __XFS_IS_OPSTATE(name, NAME) \
static inline bool xfs_is_ ## name (struct xfs_mount *mp) \
{ \
return test_bit(XFS_OPSTATE_ ## NAME, &mp->m_opstate); \
} \
static inline bool xfs_clear_ ## name (struct xfs_mount *mp) \
{ \
return test_and_clear_bit(XFS_OPSTATE_ ## NAME, &mp->m_opstate); \
} \
static inline bool xfs_set_ ## name (struct xfs_mount *mp) \
{ \
return test_and_set_bit(XFS_OPSTATE_ ## NAME, &mp->m_opstate); \
}
__XFS_IS_OPSTATE(unmounting, UNMOUNTING)
__XFS_IS_OPSTATE(clean, CLEAN)
__XFS_IS_OPSTATE(shutdown, SHUTDOWN)
__XFS_IS_OPSTATE(inode32, INODE32)
__XFS_IS_OPSTATE(readonly, READONLY)
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
__XFS_IS_OPSTATE(inodegc_enabled, INODEGC_ENABLED)
__XFS_IS_OPSTATE(blockgc_enabled, BLOCKGC_ENABLED)
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
#define XFS_OPSTATE_STRINGS \
{ (1UL << XFS_OPSTATE_UNMOUNTING), "unmounting" }, \
{ (1UL << XFS_OPSTATE_CLEAN), "clean" }, \
{ (1UL << XFS_OPSTATE_SHUTDOWN), "shutdown" }, \
{ (1UL << XFS_OPSTATE_INODE32), "inode32" }, \
{ (1UL << XFS_OPSTATE_READONLY), "read_only" }, \
{ (1UL << XFS_OPSTATE_INODEGC_ENABLED), "inodegc" }, \
{ (1UL << XFS_OPSTATE_BLOCKGC_ENABLED), "blockgc" }
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-06 21:05:39 +03:00
/*
* Max and min values for mount-option defined I/O
* preallocation sizes.
*/
#define XFS_MAX_IO_LOG 30 /* 1G */
#define XFS_MIN_IO_LOG PAGE_SHIFT
#define XFS_FORCED_SHUTDOWN(mp) xfs_is_shutdown(mp)
void xfs_do_force_shutdown(struct xfs_mount *mp, int flags, char *fname,
int lnnum);
#define xfs_force_shutdown(m,f) \
xfs_do_force_shutdown(m, f, __FILE__, __LINE__)
#define SHUTDOWN_META_IO_ERROR 0x0001 /* write attempt to metadata failed */
#define SHUTDOWN_LOG_IO_ERROR 0x0002 /* write attempt to the log failed */
#define SHUTDOWN_FORCE_UMOUNT 0x0004 /* shutdown from a forced unmount */
#define SHUTDOWN_CORRUPT_INCORE 0x0008 /* corrupt in-memory data structures */
#define XFS_SHUTDOWN_STRINGS \
{ SHUTDOWN_META_IO_ERROR, "metadata_io" }, \
{ SHUTDOWN_LOG_IO_ERROR, "log_io" }, \
{ SHUTDOWN_FORCE_UMOUNT, "force_umount" }, \
{ SHUTDOWN_CORRUPT_INCORE, "corruption" }
/*
* Flags for xfs_mountfs
*/
#define XFS_MFSI_QUIET 0x40 /* Be silent if mount errors found */
static inline xfs_agnumber_t
xfs_daddr_to_agno(struct xfs_mount *mp, xfs_daddr_t d)
{
xfs_rfsblock_t ld = XFS_BB_TO_FSBT(mp, d);
do_div(ld, mp->m_sb.sb_agblocks);
return (xfs_agnumber_t) ld;
}
static inline xfs_agblock_t
xfs_daddr_to_agbno(struct xfs_mount *mp, xfs_daddr_t d)
{
xfs_rfsblock_t ld = XFS_BB_TO_FSBT(mp, d);
return (xfs_agblock_t) do_div(ld, mp->m_sb.sb_agblocks);
}
int xfs_buf_hash_init(struct xfs_perag *pag);
void xfs_buf_hash_destroy(struct xfs_perag *pag);
extern void xfs_uuid_table_free(void);
extern uint64_t xfs_default_resblks(xfs_mount_t *mp);
extern int xfs_mountfs(xfs_mount_t *mp);
extern void xfs_unmountfs(xfs_mount_t *);
/*
* Deltas for the block count can vary from 1 to very large, but lock contention
* only occurs on frequent small block count updates such as in the delayed
* allocation path for buffered writes (page a time updates). Hence we set
* a large batch count (1024) to minimise global counter updates except when
* we get near to ENOSPC and we have to be very accurate with our updates.
*/
#define XFS_FDBLOCKS_BATCH 1024
extern int xfs_mod_fdblocks(struct xfs_mount *mp, int64_t delta,
bool reserved);
extern int xfs_mod_frextents(struct xfs_mount *mp, int64_t delta);
extern int xfs_readsb(xfs_mount_t *, int);
extern void xfs_freesb(xfs_mount_t *);
extern bool xfs_fs_writable(struct xfs_mount *mp, int level);
extern int xfs_sb_validate_fsb_count(struct xfs_sb *, uint64_t);
extern int xfs_dev_is_read_only(struct xfs_mount *, char *);
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 03:35:03 +03:00
extern void xfs_set_low_space_thresholds(struct xfs_mount *);
int xfs_zero_extent(struct xfs_inode *ip, xfs_fsblock_t start_fsb,
xfs_off_t count_fsb);
struct xfs_error_cfg * xfs_error_get_cfg(struct xfs_mount *mp,
int error_class, int error);
void xfs_force_summary_recalc(struct xfs_mount *mp);
int xfs_add_incompat_log_feature(struct xfs_mount *mp, uint32_t feature);
bool xfs_clear_incompat_log_features(struct xfs_mount *mp);
void xfs_mod_delalloc(struct xfs_mount *mp, int64_t delta);
#endif /* __XFS_MOUNT_H__ */