freebsd-nq/include/sys/arc_impl.h
Matthew Ahrens 64fc776208 OpenZFS 7968 - multi-threaded spa_sync()
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed by: Brad Lewis <brad.lewis@delphix.com>
Reviewed by: Saso Kiselkov <saso.kiselkov@nexenta.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Ported-by: Matthew Ahrens <mahrens@delphix.com>

spa_sync() iterates over all the dirty dnodes and processes each of them
by calling dnode_sync(). If there are many dirty dnodes (e.g. because we
created or removed a lot of files), the single thread of spa_sync()
calling dnode_sync() can become a bottleneck. Additionally, if many
dnodes are dirtied concurrently in open context (e.g. due to concurrent
file creation), the os_lock will experience lock contention via
dnode_setdirty().

The solution is to track dirty dnodes on a multilist_t, and for
spa_sync() to use separate threads to process each of the sublists in
the multilist.

OpenZFS-issue: https://www.illumos.org/issues/7968
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/4a2a54c
Closes #5752
2017-03-20 18:36:00 -07:00

246 lines
8.2 KiB
C

/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved.
* Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
* Copyright 2013 Nexenta Systems, Inc. All rights reserved.
*/
#ifndef _SYS_ARC_IMPL_H
#define _SYS_ARC_IMPL_H
#include <sys/arc.h>
#ifdef __cplusplus
extern "C" {
#endif
/*
* Note that buffers can be in one of 6 states:
* ARC_anon - anonymous (discussed below)
* ARC_mru - recently used, currently cached
* ARC_mru_ghost - recentely used, no longer in cache
* ARC_mfu - frequently used, currently cached
* ARC_mfu_ghost - frequently used, no longer in cache
* ARC_l2c_only - exists in L2ARC but not other states
* When there are no active references to the buffer, they are
* are linked onto a list in one of these arc states. These are
* the only buffers that can be evicted or deleted. Within each
* state there are multiple lists, one for meta-data and one for
* non-meta-data. Meta-data (indirect blocks, blocks of dnodes,
* etc.) is tracked separately so that it can be managed more
* explicitly: favored over data, limited explicitly.
*
* Anonymous buffers are buffers that are not associated with
* a DVA. These are buffers that hold dirty block copies
* before they are written to stable storage. By definition,
* they are "ref'd" and are considered part of arc_mru
* that cannot be freed. Generally, they will acquire a DVA
* as they are written and migrate onto the arc_mru list.
*
* The ARC_l2c_only state is for buffers that are in the second
* level ARC but no longer in any of the ARC_m* lists. The second
* level ARC itself may also contain buffers that are in any of
* the ARC_m* states - meaning that a buffer can exist in two
* places. The reason for the ARC_l2c_only state is to keep the
* buffer header in the hash table, so that reads that hit the
* second level ARC benefit from these fast lookups.
*/
typedef struct arc_state {
/*
* list of evictable buffers
*/
multilist_t *arcs_list[ARC_BUFC_NUMTYPES];
/*
* total amount of evictable data in this state
*/
refcount_t arcs_esize[ARC_BUFC_NUMTYPES];
/*
* total amount of data in this state; this includes: evictable,
* non-evictable, ARC_BUFC_DATA, and ARC_BUFC_METADATA.
*/
refcount_t arcs_size;
/*
* supports the "dbufs" kstat
*/
arc_state_type_t arcs_state;
} arc_state_t;
typedef struct arc_callback arc_callback_t;
struct arc_callback {
void *acb_private;
arc_done_func_t *acb_done;
arc_buf_t *acb_buf;
boolean_t acb_compressed;
zio_t *acb_zio_dummy;
arc_callback_t *acb_next;
};
typedef struct arc_write_callback arc_write_callback_t;
struct arc_write_callback {
void *awcb_private;
arc_done_func_t *awcb_ready;
arc_done_func_t *awcb_children_ready;
arc_done_func_t *awcb_physdone;
arc_done_func_t *awcb_done;
arc_buf_t *awcb_buf;
};
/*
* ARC buffers are separated into multiple structs as a memory saving measure:
* - Common fields struct, always defined, and embedded within it:
* - L2-only fields, always allocated but undefined when not in L2ARC
* - L1-only fields, only allocated when in L1ARC
*
* Buffer in L1 Buffer only in L2
* +------------------------+ +------------------------+
* | arc_buf_hdr_t | | arc_buf_hdr_t |
* | | | |
* | | | |
* | | | |
* +------------------------+ +------------------------+
* | l2arc_buf_hdr_t | | l2arc_buf_hdr_t |
* | (undefined if L1-only) | | |
* +------------------------+ +------------------------+
* | l1arc_buf_hdr_t |
* | |
* | |
* | |
* | |
* +------------------------+
*
* Because it's possible for the L2ARC to become extremely large, we can wind
* up eating a lot of memory in L2ARC buffer headers, so the size of a header
* is minimized by only allocating the fields necessary for an L1-cached buffer
* when a header is actually in the L1 cache. The sub-headers (l1arc_buf_hdr and
* l2arc_buf_hdr) are embedded rather than allocated separately to save a couple
* words in pointers. arc_hdr_realloc() is used to switch a header between
* these two allocation states.
*/
typedef struct l1arc_buf_hdr {
kmutex_t b_freeze_lock;
zio_cksum_t *b_freeze_cksum;
arc_buf_t *b_buf;
uint32_t b_bufcnt;
/* for waiting on writes to complete */
kcondvar_t b_cv;
uint8_t b_byteswap;
/* protected by arc state mutex */
arc_state_t *b_state;
multilist_node_t b_arc_node;
/* updated atomically */
clock_t b_arc_access;
uint32_t b_mru_hits;
uint32_t b_mru_ghost_hits;
uint32_t b_mfu_hits;
uint32_t b_mfu_ghost_hits;
uint32_t b_l2_hits;
/* self protecting */
refcount_t b_refcnt;
arc_callback_t *b_acb;
abd_t *b_pabd;
} l1arc_buf_hdr_t;
typedef struct l2arc_dev {
vdev_t *l2ad_vdev; /* vdev */
spa_t *l2ad_spa; /* spa */
uint64_t l2ad_hand; /* next write location */
uint64_t l2ad_start; /* first addr on device */
uint64_t l2ad_end; /* last addr on device */
boolean_t l2ad_first; /* first sweep through */
boolean_t l2ad_writing; /* currently writing */
kmutex_t l2ad_mtx; /* lock for buffer list */
list_t l2ad_buflist; /* buffer list */
list_node_t l2ad_node; /* device list node */
refcount_t l2ad_alloc; /* allocated bytes */
} l2arc_dev_t;
typedef struct l2arc_buf_hdr {
/* protected by arc_buf_hdr mutex */
l2arc_dev_t *b_dev; /* L2ARC device */
uint64_t b_daddr; /* disk address, offset byte */
uint32_t b_hits;
list_node_t b_l2node;
} l2arc_buf_hdr_t;
typedef struct l2arc_write_callback {
l2arc_dev_t *l2wcb_dev; /* device info */
arc_buf_hdr_t *l2wcb_head; /* head of write buflist */
} l2arc_write_callback_t;
struct arc_buf_hdr {
/* protected by hash lock */
dva_t b_dva;
uint64_t b_birth;
arc_buf_contents_t b_type;
arc_buf_hdr_t *b_hash_next;
arc_flags_t b_flags;
/*
* This field stores the size of the data buffer after
* compression, and is set in the arc's zio completion handlers.
* It is in units of SPA_MINBLOCKSIZE (e.g. 1 == 512 bytes).
*
* While the block pointers can store up to 32MB in their psize
* field, we can only store up to 32MB minus 512B. This is due
* to the bp using a bias of 1, whereas we use a bias of 0 (i.e.
* a field of zeros represents 512B in the bp). We can't use a
* bias of 1 since we need to reserve a psize of zero, here, to
* represent holes and embedded blocks.
*
* This isn't a problem in practice, since the maximum size of a
* buffer is limited to 16MB, so we never need to store 32MB in
* this field. Even in the upstream illumos code base, the
* maximum size of a buffer is limited to 16MB.
*/
uint16_t b_psize;
/*
* This field stores the size of the data buffer before
* compression, and cannot change once set. It is in units
* of SPA_MINBLOCKSIZE (e.g. 2 == 1024 bytes)
*/
uint16_t b_lsize; /* immutable */
uint64_t b_spa; /* immutable */
/* L2ARC fields. Undefined when not in L2ARC. */
l2arc_buf_hdr_t b_l2hdr;
/* L1ARC fields. Undefined when in l2arc_only state */
l1arc_buf_hdr_t b_l1hdr;
};
#ifdef __cplusplus
}
#endif
#endif /* _SYS_ARC_IMPL_H */