Index: head/sys/cddl/compat/opensolaris/sys/kmem.h =================================================================== --- head/sys/cddl/compat/opensolaris/sys/kmem.h (revision 327484) +++ head/sys/cddl/compat/opensolaris/sys/kmem.h (revision 327485) @@ -1,90 +1,91 @@ /*- * Copyright (c) 2007 Pawel Jakub Dawidek * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _OPENSOLARIS_SYS_KMEM_H_ #define _OPENSOLARIS_SYS_KMEM_H_ #include #include #include #include #include #include #include #include MALLOC_DECLARE(M_SOLARIS); #define POINTER_IS_VALID(p) (!((uintptr_t)(p) & 0x3)) #define POINTER_INVALIDATE(pp) (*(pp) = (void *)((uintptr_t)(*(pp)) | 0x1)) #define KM_SLEEP M_WAITOK #define KM_PUSHPAGE M_WAITOK #define KM_NOSLEEP M_NOWAIT #define KM_NODEBUG M_NODUMP #define KM_NORMALPRI 0 #define KMC_NODEBUG UMA_ZONE_NODUMP #define KMC_NOTOUCH 0 typedef struct kmem_cache { char kc_name[32]; #if defined(_KERNEL) && !defined(KMEM_DEBUG) uma_zone_t kc_zone; #else size_t kc_size; #endif int (*kc_constructor)(void *, void *, int); void (*kc_destructor)(void *, void *); void *kc_private; } kmem_cache_t; void *zfs_kmem_alloc(size_t size, int kmflags); void zfs_kmem_free(void *buf, size_t size); uint64_t kmem_size(void); kmem_cache_t *kmem_cache_create(char *name, size_t bufsize, size_t align, int (*constructor)(void *, void *, int), void (*destructor)(void *, void *), void (*reclaim)(void *) __unused, void *private, vmem_t *vmp, int cflags); void kmem_cache_destroy(kmem_cache_t *cache); void *kmem_cache_alloc(kmem_cache_t *cache, int flags); void kmem_cache_free(kmem_cache_t *cache, void *buf); void kmem_cache_reap_now(kmem_cache_t *cache); void kmem_reap(void); int kmem_debugging(void); void *calloc(size_t n, size_t s); #define freemem vm_cnt.v_free_count #define minfree vm_cnt.v_free_min -#define heap_arena kmem_arena +#define heap_arena kernel_arena +#define zio_arena NULL #define kmem_alloc(size, kmflags) zfs_kmem_alloc((size), (kmflags)) #define kmem_zalloc(size, kmflags) zfs_kmem_alloc((size), (kmflags) | M_ZERO) #define kmem_free(buf, size) zfs_kmem_free((buf), (size)) #define kmem_cache_set_move(cache, movefunc) do { } while (0) #endif /* _OPENSOLARIS_SYS_KMEM_H_ */ Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c (revision 327484) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c (revision 327485) @@ -1,7834 +1,7818 @@ /* * 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) 2012, Joyent, Inc. All rights reserved. * Copyright (c) 2011, 2017 by Delphix. All rights reserved. * Copyright (c) 2014 by Saso Kiselkov. All rights reserved. * Copyright 2015 Nexenta Systems, Inc. All rights reserved. */ /* * DVA-based Adjustable Replacement Cache * * While much of the theory of operation used here is * based on the self-tuning, low overhead replacement cache * presented by Megiddo and Modha at FAST 2003, there are some * significant differences: * * 1. The Megiddo and Modha model assumes any page is evictable. * Pages in its cache cannot be "locked" into memory. This makes * the eviction algorithm simple: evict the last page in the list. * This also make the performance characteristics easy to reason * about. Our cache is not so simple. At any given moment, some * subset of the blocks in the cache are un-evictable because we * have handed out a reference to them. Blocks are only evictable * when there are no external references active. This makes * eviction far more problematic: we choose to evict the evictable * blocks that are the "lowest" in the list. * * There are times when it is not possible to evict the requested * space. In these circumstances we are unable to adjust the cache * size. To prevent the cache growing unbounded at these times we * implement a "cache throttle" that slows the flow of new data * into the cache until we can make space available. * * 2. The Megiddo and Modha model assumes a fixed cache size. * Pages are evicted when the cache is full and there is a cache * miss. Our model has a variable sized cache. It grows with * high use, but also tries to react to memory pressure from the * operating system: decreasing its size when system memory is * tight. * * 3. The Megiddo and Modha model assumes a fixed page size. All * elements of the cache are therefore exactly the same size. So * when adjusting the cache size following a cache miss, its simply * a matter of choosing a single page to evict. In our model, we * have variable sized cache blocks (rangeing from 512 bytes to * 128K bytes). We therefore choose a set of blocks to evict to make * space for a cache miss that approximates as closely as possible * the space used by the new block. * * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache" * by N. Megiddo & D. Modha, FAST 2003 */ /* * The locking model: * * A new reference to a cache buffer can be obtained in two * ways: 1) via a hash table lookup using the DVA as a key, * or 2) via one of the ARC lists. The arc_read() interface * uses method 1, while the internal ARC algorithms for * adjusting the cache use method 2. We therefore provide two * types of locks: 1) the hash table lock array, and 2) the * ARC list locks. * * Buffers do not have their own mutexes, rather they rely on the * hash table mutexes for the bulk of their protection (i.e. most * fields in the arc_buf_hdr_t are protected by these mutexes). * * buf_hash_find() returns the appropriate mutex (held) when it * locates the requested buffer in the hash table. It returns * NULL for the mutex if the buffer was not in the table. * * buf_hash_remove() expects the appropriate hash mutex to be * already held before it is invoked. * * Each ARC state also has a mutex which is used to protect the * buffer list associated with the state. When attempting to * obtain a hash table lock while holding an ARC list lock you * must use: mutex_tryenter() to avoid deadlock. Also note that * the active state mutex must be held before the ghost state mutex. * * Note that the majority of the performance stats are manipulated * with atomic operations. * * The L2ARC uses the l2ad_mtx on each vdev for the following: * * - L2ARC buflist creation * - L2ARC buflist eviction * - L2ARC write completion, which walks L2ARC buflists * - ARC header destruction, as it removes from L2ARC buflists * - ARC header release, as it removes from L2ARC buflists */ /* * ARC operation: * * Every block that is in the ARC is tracked by an arc_buf_hdr_t structure. * This structure can point either to a block that is still in the cache or to * one that is only accessible in an L2 ARC device, or it can provide * information about a block that was recently evicted. If a block is * only accessible in the L2ARC, then the arc_buf_hdr_t only has enough * information to retrieve it from the L2ARC device. This information is * stored in the l2arc_buf_hdr_t sub-structure of the arc_buf_hdr_t. A block * that is in this state cannot access the data directly. * * Blocks that are actively being referenced or have not been evicted * are cached in the L1ARC. The L1ARC (l1arc_buf_hdr_t) is a structure within * the arc_buf_hdr_t that will point to the data block in memory. A block can * only be read by a consumer if it has an l1arc_buf_hdr_t. The L1ARC * caches data in two ways -- in a list of ARC buffers (arc_buf_t) and * also in the arc_buf_hdr_t's private physical data block pointer (b_pabd). * * The L1ARC's data pointer may or may not be uncompressed. The ARC has the * ability to store the physical data (b_pabd) associated with the DVA of the * arc_buf_hdr_t. Since the b_pabd is a copy of the on-disk physical block, * it will match its on-disk compression characteristics. This behavior can be * disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the * compressed ARC functionality is disabled, the b_pabd will point to an * uncompressed version of the on-disk data. * * Data in the L1ARC is not accessed by consumers of the ARC directly. Each * arc_buf_hdr_t can have multiple ARC buffers (arc_buf_t) which reference it. * Each ARC buffer (arc_buf_t) is being actively accessed by a specific ARC * consumer. The ARC will provide references to this data and will keep it * cached until it is no longer in use. The ARC caches only the L1ARC's physical * data block and will evict any arc_buf_t that is no longer referenced. The * amount of memory consumed by the arc_buf_ts' data buffers can be seen via the * "overhead_size" kstat. * * Depending on the consumer, an arc_buf_t can be requested in uncompressed or * compressed form. The typical case is that consumers will want uncompressed * data, and when that happens a new data buffer is allocated where the data is * decompressed for them to use. Currently the only consumer who wants * compressed arc_buf_t's is "zfs send", when it streams data exactly as it * exists on disk. When this happens, the arc_buf_t's data buffer is shared * with the arc_buf_hdr_t. * * Here is a diagram showing an arc_buf_hdr_t referenced by two arc_buf_t's. The * first one is owned by a compressed send consumer (and therefore references * the same compressed data buffer as the arc_buf_hdr_t) and the second could be * used by any other consumer (and has its own uncompressed copy of the data * buffer). * * arc_buf_hdr_t * +-----------+ * | fields | * | common to | * | L1- and | * | L2ARC | * +-----------+ * | l2arc_buf_hdr_t * | | * +-----------+ * | l1arc_buf_hdr_t * | | arc_buf_t * | b_buf +------------>+-----------+ arc_buf_t * | b_pabd +-+ |b_next +---->+-----------+ * +-----------+ | |-----------| |b_next +-->NULL * | |b_comp = T | +-----------+ * | |b_data +-+ |b_comp = F | * | +-----------+ | |b_data +-+ * +->+------+ | +-----------+ | * compressed | | | | * data | |<--------------+ | uncompressed * +------+ compressed, | data * shared +-->+------+ * data | | * | | * +------+ * * When a consumer reads a block, the ARC must first look to see if the * arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new * arc_buf_t and either copies uncompressed data into a new data buffer from an * existing uncompressed arc_buf_t, decompresses the hdr's b_pabd buffer into a * new data buffer, or shares the hdr's b_pabd buffer, depending on whether the * hdr is compressed and the desired compression characteristics of the * arc_buf_t consumer. If the arc_buf_t ends up sharing data with the * arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be * the last buffer in the hdr's b_buf list, however a shared compressed buf can * be anywhere in the hdr's list. * * The diagram below shows an example of an uncompressed ARC hdr that is * sharing its data with an arc_buf_t (note that the shared uncompressed buf is * the last element in the buf list): * * arc_buf_hdr_t * +-----------+ * | | * | | * | | * +-----------+ * l2arc_buf_hdr_t| | * | | * +-----------+ * l1arc_buf_hdr_t| | * | | arc_buf_t (shared) * | b_buf +------------>+---------+ arc_buf_t * | | |b_next +---->+---------+ * | b_pabd +-+ |---------| |b_next +-->NULL * +-----------+ | | | +---------+ * | |b_data +-+ | | * | +---------+ | |b_data +-+ * +->+------+ | +---------+ | * | | | | * uncompressed | | | | * data +------+ | | * ^ +->+------+ | * | uncompressed | | | * | data | | | * | +------+ | * +---------------------------------+ * * Writing to the ARC requires that the ARC first discard the hdr's b_pabd * since the physical block is about to be rewritten. The new data contents * will be contained in the arc_buf_t. As the I/O pipeline performs the write, * it may compress the data before writing it to disk. The ARC will be called * with the transformed data and will bcopy the transformed on-disk block into * a newly allocated b_pabd. Writes are always done into buffers which have * either been loaned (and hence are new and don't have other readers) or * buffers which have been released (and hence have their own hdr, if there * were originally other readers of the buf's original hdr). This ensures that * the ARC only needs to update a single buf and its hdr after a write occurs. * * When the L2ARC is in use, it will also take advantage of the b_pabd. The * L2ARC will always write the contents of b_pabd to the L2ARC. This means * that when compressed ARC is enabled that the L2ARC blocks are identical * to the on-disk block in the main data pool. This provides a significant * advantage since the ARC can leverage the bp's checksum when reading from the * L2ARC to determine if the contents are valid. However, if the compressed * ARC is disabled, then the L2ARC's block must be transformed to look * like the physical block in the main data pool before comparing the * checksum and determining its validity. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #include #endif #include #include #include #include #include #include #ifdef illumos #ifndef _KERNEL /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ boolean_t arc_watch = B_FALSE; int arc_procfd; #endif #endif /* illumos */ static kmutex_t arc_reclaim_lock; static kcondvar_t arc_reclaim_thread_cv; static boolean_t arc_reclaim_thread_exit; static kcondvar_t arc_reclaim_waiters_cv; static kmutex_t arc_dnlc_evicts_lock; static kcondvar_t arc_dnlc_evicts_cv; static boolean_t arc_dnlc_evicts_thread_exit; uint_t arc_reduce_dnlc_percent = 3; /* * The number of headers to evict in arc_evict_state_impl() before * dropping the sublist lock and evicting from another sublist. A lower * value means we're more likely to evict the "correct" header (i.e. the * oldest header in the arc state), but comes with higher overhead * (i.e. more invocations of arc_evict_state_impl()). */ int zfs_arc_evict_batch_limit = 10; /* number of seconds before growing cache again */ static int arc_grow_retry = 60; /* shift of arc_c for calculating overflow limit in arc_get_data_impl */ int zfs_arc_overflow_shift = 8; /* shift of arc_c for calculating both min and max arc_p */ static int arc_p_min_shift = 4; /* log2(fraction of arc to reclaim) */ static int arc_shrink_shift = 7; /* * log2(fraction of ARC which must be free to allow growing). * I.e. If there is less than arc_c >> arc_no_grow_shift free memory, * when reading a new block into the ARC, we will evict an equal-sized block * from the ARC. * * This must be less than arc_shrink_shift, so that when we shrink the ARC, * we will still not allow it to grow. */ int arc_no_grow_shift = 5; /* * minimum lifespan of a prefetch block in clock ticks * (initialized in arc_init()) */ static int arc_min_prefetch_lifespan; /* * If this percent of memory is free, don't throttle. */ int arc_lotsfree_percent = 10; static int arc_dead; extern boolean_t zfs_prefetch_disable; /* * The arc has filled available memory and has now warmed up. */ static boolean_t arc_warm; /* * These tunables are for performance analysis. */ uint64_t zfs_arc_max; uint64_t zfs_arc_min; uint64_t zfs_arc_meta_limit = 0; uint64_t zfs_arc_meta_min = 0; int zfs_arc_grow_retry = 0; int zfs_arc_shrink_shift = 0; int zfs_arc_no_grow_shift = 0; int zfs_arc_p_min_shift = 0; uint64_t zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ u_int zfs_arc_free_target = 0; /* Absolute min for arc min / max is 16MB. */ static uint64_t arc_abs_min = 16 << 20; boolean_t zfs_compressed_arc_enabled = B_TRUE; static int sysctl_vfs_zfs_arc_free_target(SYSCTL_HANDLER_ARGS); static int sysctl_vfs_zfs_arc_meta_limit(SYSCTL_HANDLER_ARGS); static int sysctl_vfs_zfs_arc_max(SYSCTL_HANDLER_ARGS); static int sysctl_vfs_zfs_arc_min(SYSCTL_HANDLER_ARGS); static int sysctl_vfs_zfs_arc_no_grow_shift(SYSCTL_HANDLER_ARGS); #if defined(__FreeBSD__) && defined(_KERNEL) static void arc_free_target_init(void *unused __unused) { zfs_arc_free_target = vm_pageout_wakeup_thresh; } SYSINIT(arc_free_target_init, SI_SUB_KTHREAD_PAGE, SI_ORDER_ANY, arc_free_target_init, NULL); TUNABLE_QUAD("vfs.zfs.arc_meta_limit", &zfs_arc_meta_limit); TUNABLE_QUAD("vfs.zfs.arc_meta_min", &zfs_arc_meta_min); TUNABLE_INT("vfs.zfs.arc_shrink_shift", &zfs_arc_shrink_shift); TUNABLE_INT("vfs.zfs.arc_grow_retry", &zfs_arc_grow_retry); TUNABLE_INT("vfs.zfs.arc_no_grow_shift", &zfs_arc_no_grow_shift); SYSCTL_DECL(_vfs_zfs); SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_max, CTLTYPE_U64 | CTLFLAG_RWTUN, 0, sizeof(uint64_t), sysctl_vfs_zfs_arc_max, "QU", "Maximum ARC size"); SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_min, CTLTYPE_U64 | CTLFLAG_RWTUN, 0, sizeof(uint64_t), sysctl_vfs_zfs_arc_min, "QU", "Minimum ARC size"); SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_no_grow_shift, CTLTYPE_U32 | CTLFLAG_RWTUN, 0, sizeof(uint32_t), sysctl_vfs_zfs_arc_no_grow_shift, "U", "log2(fraction of ARC which must be free to allow growing)"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, arc_average_blocksize, CTLFLAG_RDTUN, &zfs_arc_average_blocksize, 0, "ARC average blocksize"); SYSCTL_INT(_vfs_zfs, OID_AUTO, arc_shrink_shift, CTLFLAG_RW, &arc_shrink_shift, 0, "log2(fraction of arc to reclaim)"); SYSCTL_INT(_vfs_zfs, OID_AUTO, arc_grow_retry, CTLFLAG_RW, &arc_grow_retry, 0, "Wait in seconds before considering growing ARC"); SYSCTL_INT(_vfs_zfs, OID_AUTO, compressed_arc_enabled, CTLFLAG_RDTUN, &zfs_compressed_arc_enabled, 0, "Enable compressed ARC"); /* * We don't have a tunable for arc_free_target due to the dependency on * pagedaemon initialisation. */ SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_free_target, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(u_int), sysctl_vfs_zfs_arc_free_target, "IU", "Desired number of free pages below which ARC triggers reclaim"); static int sysctl_vfs_zfs_arc_free_target(SYSCTL_HANDLER_ARGS) { u_int val; int err; val = zfs_arc_free_target; err = sysctl_handle_int(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val < minfree) return (EINVAL); if (val > vm_cnt.v_page_count) return (EINVAL); zfs_arc_free_target = val; return (0); } /* * Must be declared here, before the definition of corresponding kstat * macro which uses the same names will confuse the compiler. */ SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_meta_limit, CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), sysctl_vfs_zfs_arc_meta_limit, "QU", "ARC metadata limit"); #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 aquire 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; } arc_state_t; /* The 6 states: */ static arc_state_t ARC_anon; static arc_state_t ARC_mru; static arc_state_t ARC_mru_ghost; static arc_state_t ARC_mfu; static arc_state_t ARC_mfu_ghost; static arc_state_t ARC_l2c_only; typedef struct arc_stats { kstat_named_t arcstat_hits; kstat_named_t arcstat_misses; kstat_named_t arcstat_demand_data_hits; kstat_named_t arcstat_demand_data_misses; kstat_named_t arcstat_demand_metadata_hits; kstat_named_t arcstat_demand_metadata_misses; kstat_named_t arcstat_prefetch_data_hits; kstat_named_t arcstat_prefetch_data_misses; kstat_named_t arcstat_prefetch_metadata_hits; kstat_named_t arcstat_prefetch_metadata_misses; kstat_named_t arcstat_mru_hits; kstat_named_t arcstat_mru_ghost_hits; kstat_named_t arcstat_mfu_hits; kstat_named_t arcstat_mfu_ghost_hits; kstat_named_t arcstat_allocated; kstat_named_t arcstat_deleted; /* * Number of buffers that could not be evicted because the hash lock * was held by another thread. The lock may not necessarily be held * by something using the same buffer, since hash locks are shared * by multiple buffers. */ kstat_named_t arcstat_mutex_miss; /* * Number of buffers skipped because they have I/O in progress, are * indrect prefetch buffers that have not lived long enough, or are * not from the spa we're trying to evict from. */ kstat_named_t arcstat_evict_skip; /* * Number of times arc_evict_state() was unable to evict enough * buffers to reach it's target amount. */ kstat_named_t arcstat_evict_not_enough; kstat_named_t arcstat_evict_l2_cached; kstat_named_t arcstat_evict_l2_eligible; kstat_named_t arcstat_evict_l2_ineligible; kstat_named_t arcstat_evict_l2_skip; kstat_named_t arcstat_hash_elements; kstat_named_t arcstat_hash_elements_max; kstat_named_t arcstat_hash_collisions; kstat_named_t arcstat_hash_chains; kstat_named_t arcstat_hash_chain_max; kstat_named_t arcstat_p; kstat_named_t arcstat_c; kstat_named_t arcstat_c_min; kstat_named_t arcstat_c_max; kstat_named_t arcstat_size; /* * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd. * Note that the compressed bytes may match the uncompressed bytes * if the block is either not compressed or compressed arc is disabled. */ kstat_named_t arcstat_compressed_size; /* * Uncompressed size of the data stored in b_pabd. If compressed * arc is disabled then this value will be identical to the stat * above. */ kstat_named_t arcstat_uncompressed_size; /* * Number of bytes stored in all the arc_buf_t's. This is classified * as "overhead" since this data is typically short-lived and will * be evicted from the arc when it becomes unreferenced unless the * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level * values have been set (see comment in dbuf.c for more information). */ kstat_named_t arcstat_overhead_size; /* * Number of bytes consumed by internal ARC structures necessary * for tracking purposes; these structures are not actually * backed by ARC buffers. This includes arc_buf_hdr_t structures * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only * caches), and arc_buf_t structures (allocated via arc_buf_t * cache). */ kstat_named_t arcstat_hdr_size; /* * Number of bytes consumed by ARC buffers of type equal to * ARC_BUFC_DATA. This is generally consumed by buffers backing * on disk user data (e.g. plain file contents). */ kstat_named_t arcstat_data_size; /* * Number of bytes consumed by ARC buffers of type equal to * ARC_BUFC_METADATA. This is generally consumed by buffers * backing on disk data that is used for internal ZFS * structures (e.g. ZAP, dnode, indirect blocks, etc). */ kstat_named_t arcstat_metadata_size; /* * Number of bytes consumed by various buffers and structures * not actually backed with ARC buffers. This includes bonus * buffers (allocated directly via zio_buf_* functions), * dmu_buf_impl_t structures (allocated via dmu_buf_impl_t * cache), and dnode_t structures (allocated via dnode_t cache). */ kstat_named_t arcstat_other_size; /* * Total number of bytes consumed by ARC buffers residing in the * arc_anon state. This includes *all* buffers in the arc_anon * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. */ kstat_named_t arcstat_anon_size; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_DATA, * residing in the arc_anon state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_anon_evictable_data; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_METADATA, * residing in the arc_anon state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_anon_evictable_metadata; /* * Total number of bytes consumed by ARC buffers residing in the * arc_mru state. This includes *all* buffers in the arc_mru * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. */ kstat_named_t arcstat_mru_size; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_DATA, * residing in the arc_mru state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_mru_evictable_data; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_METADATA, * residing in the arc_mru state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_mru_evictable_metadata; /* * Total number of bytes that *would have been* consumed by ARC * buffers in the arc_mru_ghost state. The key thing to note * here, is the fact that this size doesn't actually indicate * RAM consumption. The ghost lists only consist of headers and * don't actually have ARC buffers linked off of these headers. * Thus, *if* the headers had associated ARC buffers, these * buffers *would have* consumed this number of bytes. */ kstat_named_t arcstat_mru_ghost_size; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. */ kstat_named_t arcstat_mru_ghost_evictable_data; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. */ kstat_named_t arcstat_mru_ghost_evictable_metadata; /* * Total number of bytes consumed by ARC buffers residing in the * arc_mfu state. This includes *all* buffers in the arc_mfu * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. */ kstat_named_t arcstat_mfu_size; /* * Number of bytes consumed by ARC buffers that are eligible for * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu * state. */ kstat_named_t arcstat_mfu_evictable_data; /* * Number of bytes consumed by ARC buffers that are eligible for * eviction, of type ARC_BUFC_METADATA, and reside in the * arc_mfu state. */ kstat_named_t arcstat_mfu_evictable_metadata; /* * Total number of bytes that *would have been* consumed by ARC * buffers in the arc_mfu_ghost state. See the comment above * arcstat_mru_ghost_size for more details. */ kstat_named_t arcstat_mfu_ghost_size; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. */ kstat_named_t arcstat_mfu_ghost_evictable_data; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. */ kstat_named_t arcstat_mfu_ghost_evictable_metadata; kstat_named_t arcstat_l2_hits; kstat_named_t arcstat_l2_misses; kstat_named_t arcstat_l2_feeds; kstat_named_t arcstat_l2_rw_clash; kstat_named_t arcstat_l2_read_bytes; kstat_named_t arcstat_l2_write_bytes; kstat_named_t arcstat_l2_writes_sent; kstat_named_t arcstat_l2_writes_done; kstat_named_t arcstat_l2_writes_error; kstat_named_t arcstat_l2_writes_lock_retry; kstat_named_t arcstat_l2_evict_lock_retry; kstat_named_t arcstat_l2_evict_reading; kstat_named_t arcstat_l2_evict_l1cached; kstat_named_t arcstat_l2_free_on_write; kstat_named_t arcstat_l2_abort_lowmem; kstat_named_t arcstat_l2_cksum_bad; kstat_named_t arcstat_l2_io_error; kstat_named_t arcstat_l2_lsize; kstat_named_t arcstat_l2_psize; kstat_named_t arcstat_l2_hdr_size; kstat_named_t arcstat_l2_write_trylock_fail; kstat_named_t arcstat_l2_write_passed_headroom; kstat_named_t arcstat_l2_write_spa_mismatch; kstat_named_t arcstat_l2_write_in_l2; kstat_named_t arcstat_l2_write_hdr_io_in_progress; kstat_named_t arcstat_l2_write_not_cacheable; kstat_named_t arcstat_l2_write_full; kstat_named_t arcstat_l2_write_buffer_iter; kstat_named_t arcstat_l2_write_pios; kstat_named_t arcstat_l2_write_buffer_bytes_scanned; kstat_named_t arcstat_l2_write_buffer_list_iter; kstat_named_t arcstat_l2_write_buffer_list_null_iter; kstat_named_t arcstat_memory_throttle_count; kstat_named_t arcstat_meta_used; kstat_named_t arcstat_meta_limit; kstat_named_t arcstat_meta_max; kstat_named_t arcstat_meta_min; kstat_named_t arcstat_sync_wait_for_async; kstat_named_t arcstat_demand_hit_predictive_prefetch; } arc_stats_t; static arc_stats_t arc_stats = { { "hits", KSTAT_DATA_UINT64 }, { "misses", KSTAT_DATA_UINT64 }, { "demand_data_hits", KSTAT_DATA_UINT64 }, { "demand_data_misses", KSTAT_DATA_UINT64 }, { "demand_metadata_hits", KSTAT_DATA_UINT64 }, { "demand_metadata_misses", KSTAT_DATA_UINT64 }, { "prefetch_data_hits", KSTAT_DATA_UINT64 }, { "prefetch_data_misses", KSTAT_DATA_UINT64 }, { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, { "mru_hits", KSTAT_DATA_UINT64 }, { "mru_ghost_hits", KSTAT_DATA_UINT64 }, { "mfu_hits", KSTAT_DATA_UINT64 }, { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, { "allocated", KSTAT_DATA_UINT64 }, { "deleted", KSTAT_DATA_UINT64 }, { "mutex_miss", KSTAT_DATA_UINT64 }, { "evict_skip", KSTAT_DATA_UINT64 }, { "evict_not_enough", KSTAT_DATA_UINT64 }, { "evict_l2_cached", KSTAT_DATA_UINT64 }, { "evict_l2_eligible", KSTAT_DATA_UINT64 }, { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, { "evict_l2_skip", KSTAT_DATA_UINT64 }, { "hash_elements", KSTAT_DATA_UINT64 }, { "hash_elements_max", KSTAT_DATA_UINT64 }, { "hash_collisions", KSTAT_DATA_UINT64 }, { "hash_chains", KSTAT_DATA_UINT64 }, { "hash_chain_max", KSTAT_DATA_UINT64 }, { "p", KSTAT_DATA_UINT64 }, { "c", KSTAT_DATA_UINT64 }, { "c_min", KSTAT_DATA_UINT64 }, { "c_max", KSTAT_DATA_UINT64 }, { "size", KSTAT_DATA_UINT64 }, { "compressed_size", KSTAT_DATA_UINT64 }, { "uncompressed_size", KSTAT_DATA_UINT64 }, { "overhead_size", KSTAT_DATA_UINT64 }, { "hdr_size", KSTAT_DATA_UINT64 }, { "data_size", KSTAT_DATA_UINT64 }, { "metadata_size", KSTAT_DATA_UINT64 }, { "other_size", KSTAT_DATA_UINT64 }, { "anon_size", KSTAT_DATA_UINT64 }, { "anon_evictable_data", KSTAT_DATA_UINT64 }, { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, { "mru_size", KSTAT_DATA_UINT64 }, { "mru_evictable_data", KSTAT_DATA_UINT64 }, { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, { "mru_ghost_size", KSTAT_DATA_UINT64 }, { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, { "mfu_size", KSTAT_DATA_UINT64 }, { "mfu_evictable_data", KSTAT_DATA_UINT64 }, { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, { "mfu_ghost_size", KSTAT_DATA_UINT64 }, { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, { "l2_hits", KSTAT_DATA_UINT64 }, { "l2_misses", KSTAT_DATA_UINT64 }, { "l2_feeds", KSTAT_DATA_UINT64 }, { "l2_rw_clash", KSTAT_DATA_UINT64 }, { "l2_read_bytes", KSTAT_DATA_UINT64 }, { "l2_write_bytes", KSTAT_DATA_UINT64 }, { "l2_writes_sent", KSTAT_DATA_UINT64 }, { "l2_writes_done", KSTAT_DATA_UINT64 }, { "l2_writes_error", KSTAT_DATA_UINT64 }, { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, { "l2_evict_reading", KSTAT_DATA_UINT64 }, { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, { "l2_free_on_write", KSTAT_DATA_UINT64 }, { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, { "l2_cksum_bad", KSTAT_DATA_UINT64 }, { "l2_io_error", KSTAT_DATA_UINT64 }, { "l2_size", KSTAT_DATA_UINT64 }, { "l2_asize", KSTAT_DATA_UINT64 }, { "l2_hdr_size", KSTAT_DATA_UINT64 }, { "l2_write_trylock_fail", KSTAT_DATA_UINT64 }, { "l2_write_passed_headroom", KSTAT_DATA_UINT64 }, { "l2_write_spa_mismatch", KSTAT_DATA_UINT64 }, { "l2_write_in_l2", KSTAT_DATA_UINT64 }, { "l2_write_io_in_progress", KSTAT_DATA_UINT64 }, { "l2_write_not_cacheable", KSTAT_DATA_UINT64 }, { "l2_write_full", KSTAT_DATA_UINT64 }, { "l2_write_buffer_iter", KSTAT_DATA_UINT64 }, { "l2_write_pios", KSTAT_DATA_UINT64 }, { "l2_write_buffer_bytes_scanned", KSTAT_DATA_UINT64 }, { "l2_write_buffer_list_iter", KSTAT_DATA_UINT64 }, { "l2_write_buffer_list_null_iter", KSTAT_DATA_UINT64 }, { "memory_throttle_count", KSTAT_DATA_UINT64 }, { "arc_meta_used", KSTAT_DATA_UINT64 }, { "arc_meta_limit", KSTAT_DATA_UINT64 }, { "arc_meta_max", KSTAT_DATA_UINT64 }, { "arc_meta_min", KSTAT_DATA_UINT64 }, { "sync_wait_for_async", KSTAT_DATA_UINT64 }, { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64 }, }; #define ARCSTAT(stat) (arc_stats.stat.value.ui64) #define ARCSTAT_INCR(stat, val) \ atomic_add_64(&arc_stats.stat.value.ui64, (val)) #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) #define ARCSTAT_MAX(stat, val) { \ uint64_t m; \ while ((val) > (m = arc_stats.stat.value.ui64) && \ (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ continue; \ } #define ARCSTAT_MAXSTAT(stat) \ ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) /* * We define a macro to allow ARC hits/misses to be easily broken down by * two separate conditions, giving a total of four different subtypes for * each of hits and misses (so eight statistics total). */ #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ if (cond1) { \ if (cond2) { \ ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ } else { \ ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ } \ } else { \ if (cond2) { \ ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ } else { \ ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ } \ } kstat_t *arc_ksp; static arc_state_t *arc_anon; static arc_state_t *arc_mru; static arc_state_t *arc_mru_ghost; static arc_state_t *arc_mfu; static arc_state_t *arc_mfu_ghost; static arc_state_t *arc_l2c_only; /* * There are several ARC variables that are critical to export as kstats -- * but we don't want to have to grovel around in the kstat whenever we wish to * manipulate them. For these variables, we therefore define them to be in * terms of the statistic variable. This assures that we are not introducing * the possibility of inconsistency by having shadow copies of the variables, * while still allowing the code to be readable. */ #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */ #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */ #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ /* compressed size of entire arc */ #define arc_compressed_size ARCSTAT(arcstat_compressed_size) /* uncompressed size of entire arc */ #define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size) /* number of bytes in the arc from arc_buf_t's */ #define arc_overhead_size ARCSTAT(arcstat_overhead_size) static int arc_no_grow; /* Don't try to grow cache size */ static uint64_t arc_tempreserve; static uint64_t arc_loaned_bytes; 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; #ifdef ZFS_DEBUG /* * Used for debugging with kmem_flags - by allocating and freeing * b_thawed when the buffer is thawed, we get a record of the stack * trace that thawed it. */ void *b_thawed; #endif 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; /* self protecting */ refcount_t b_refcnt; arc_callback_t *b_acb; abd_t *b_pabd; } l1arc_buf_hdr_t; typedef struct l2arc_dev 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 */ list_node_t b_l2node; } l2arc_buf_hdr_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; }; #if defined(__FreeBSD__) && defined(_KERNEL) static int sysctl_vfs_zfs_arc_meta_limit(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = arc_meta_limit; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val <= 0 || val > arc_c_max) return (EINVAL); arc_meta_limit = val; return (0); } static int sysctl_vfs_zfs_arc_no_grow_shift(SYSCTL_HANDLER_ARGS) { uint32_t val; int err; val = arc_no_grow_shift; err = sysctl_handle_32(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val >= arc_shrink_shift) return (EINVAL); arc_no_grow_shift = val; return (0); } static int sysctl_vfs_zfs_arc_max(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = zfs_arc_max; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (zfs_arc_max == 0) { /* Loader tunable so blindly set */ zfs_arc_max = val; return (0); } if (val < arc_abs_min || val > kmem_size()) return (EINVAL); if (val < arc_c_min) return (EINVAL); if (zfs_arc_meta_limit > 0 && val < zfs_arc_meta_limit) return (EINVAL); arc_c_max = val; arc_c = arc_c_max; arc_p = (arc_c >> 1); if (zfs_arc_meta_limit == 0) { /* limit meta-data to 1/4 of the arc capacity */ arc_meta_limit = arc_c_max / 4; } /* if kmem_flags are set, lets try to use less memory */ if (kmem_debugging()) arc_c = arc_c / 2; zfs_arc_max = arc_c; return (0); } static int sysctl_vfs_zfs_arc_min(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = zfs_arc_min; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (zfs_arc_min == 0) { /* Loader tunable so blindly set */ zfs_arc_min = val; return (0); } if (val < arc_abs_min || val > arc_c_max) return (EINVAL); arc_c_min = val; if (zfs_arc_meta_min == 0) arc_meta_min = arc_c_min / 2; if (arc_c < arc_c_min) arc_c = arc_c_min; zfs_arc_min = arc_c_min; return (0); } #endif #define GHOST_STATE(state) \ ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ (state) == arc_l2c_only) #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) #define HDR_COMPRESSION_ENABLED(hdr) \ ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC) #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) #define HDR_L2_READING(hdr) \ (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA) #define HDR_ISTYPE_METADATA(hdr) \ ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) /* For storing compression mode in b_flags */ #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1) #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \ HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS)) #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \ HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp)); #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL) #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED) #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED) /* * Other sizes */ #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) /* * Hash table routines */ #define HT_LOCK_PAD CACHE_LINE_SIZE struct ht_lock { kmutex_t ht_lock; #ifdef _KERNEL unsigned char pad[(HT_LOCK_PAD - sizeof (kmutex_t))]; #endif }; #define BUF_LOCKS 256 typedef struct buf_hash_table { uint64_t ht_mask; arc_buf_hdr_t **ht_table; struct ht_lock ht_locks[BUF_LOCKS] __aligned(CACHE_LINE_SIZE); } buf_hash_table_t; static buf_hash_table_t buf_hash_table; #define BUF_HASH_INDEX(spa, dva, birth) \ (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) #define HDR_LOCK(hdr) \ (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) uint64_t zfs_crc64_table[256]; /* * Level 2 ARC */ #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ #define L2ARC_HEADROOM 2 /* num of writes */ /* * If we discover during ARC scan any buffers to be compressed, we boost * our headroom for the next scanning cycle by this percentage multiple. */ #define L2ARC_HEADROOM_BOOST 200 #define L2ARC_FEED_SECS 1 /* caching interval secs */ #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) /* L2ARC Performance Tunables */ uint64_t l2arc_write_max = L2ARC_WRITE_SIZE; /* default max write size */ uint64_t l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra write during warmup */ uint64_t l2arc_headroom = L2ARC_HEADROOM; /* number of dev writes */ uint64_t l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; uint64_t l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ uint64_t l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval milliseconds */ boolean_t l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ boolean_t l2arc_feed_again = B_TRUE; /* turbo warmup */ boolean_t l2arc_norw = B_TRUE; /* no reads during writes */ SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_write_max, CTLFLAG_RW, &l2arc_write_max, 0, "max write size"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_write_boost, CTLFLAG_RW, &l2arc_write_boost, 0, "extra write during warmup"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_headroom, CTLFLAG_RW, &l2arc_headroom, 0, "number of dev writes"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_feed_secs, CTLFLAG_RW, &l2arc_feed_secs, 0, "interval seconds"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_feed_min_ms, CTLFLAG_RW, &l2arc_feed_min_ms, 0, "min interval milliseconds"); SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_noprefetch, CTLFLAG_RW, &l2arc_noprefetch, 0, "don't cache prefetch bufs"); SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_feed_again, CTLFLAG_RW, &l2arc_feed_again, 0, "turbo warmup"); SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_norw, CTLFLAG_RW, &l2arc_norw, 0, "no reads during writes"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_size, CTLFLAG_RD, &ARC_anon.arcs_size.rc_count, 0, "size of anonymous state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_metadata_esize, CTLFLAG_RD, &ARC_anon.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, "size of anonymous state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_data_esize, CTLFLAG_RD, &ARC_anon.arcs_esize[ARC_BUFC_DATA].rc_count, 0, "size of anonymous state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_size, CTLFLAG_RD, &ARC_mru.arcs_size.rc_count, 0, "size of mru state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_metadata_esize, CTLFLAG_RD, &ARC_mru.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, "size of metadata in mru state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_data_esize, CTLFLAG_RD, &ARC_mru.arcs_esize[ARC_BUFC_DATA].rc_count, 0, "size of data in mru state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_size, CTLFLAG_RD, &ARC_mru_ghost.arcs_size.rc_count, 0, "size of mru ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_metadata_esize, CTLFLAG_RD, &ARC_mru_ghost.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, "size of metadata in mru ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_data_esize, CTLFLAG_RD, &ARC_mru_ghost.arcs_esize[ARC_BUFC_DATA].rc_count, 0, "size of data in mru ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_size, CTLFLAG_RD, &ARC_mfu.arcs_size.rc_count, 0, "size of mfu state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_metadata_esize, CTLFLAG_RD, &ARC_mfu.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, "size of metadata in mfu state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_data_esize, CTLFLAG_RD, &ARC_mfu.arcs_esize[ARC_BUFC_DATA].rc_count, 0, "size of data in mfu state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_size, CTLFLAG_RD, &ARC_mfu_ghost.arcs_size.rc_count, 0, "size of mfu ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_metadata_esize, CTLFLAG_RD, &ARC_mfu_ghost.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, "size of metadata in mfu ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_data_esize, CTLFLAG_RD, &ARC_mfu_ghost.arcs_esize[ARC_BUFC_DATA].rc_count, 0, "size of data in mfu ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2c_only_size, CTLFLAG_RD, &ARC_l2c_only.arcs_size.rc_count, 0, "size of mru state"); /* * L2ARC Internals */ 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 */ }; static list_t L2ARC_dev_list; /* device list */ static list_t *l2arc_dev_list; /* device list pointer */ static kmutex_t l2arc_dev_mtx; /* device list mutex */ static l2arc_dev_t *l2arc_dev_last; /* last device used */ static list_t L2ARC_free_on_write; /* free after write buf list */ static list_t *l2arc_free_on_write; /* free after write list ptr */ static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ static uint64_t l2arc_ndev; /* number of devices */ typedef struct l2arc_read_callback { arc_buf_hdr_t *l2rcb_hdr; /* read header */ blkptr_t l2rcb_bp; /* original blkptr */ zbookmark_phys_t l2rcb_zb; /* original bookmark */ int l2rcb_flags; /* original flags */ abd_t *l2rcb_abd; /* temporary buffer */ } l2arc_read_callback_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; typedef struct l2arc_data_free { /* protected by l2arc_free_on_write_mtx */ abd_t *l2df_abd; size_t l2df_size; arc_buf_contents_t l2df_type; list_node_t l2df_list_node; } l2arc_data_free_t; static kmutex_t l2arc_feed_thr_lock; static kcondvar_t l2arc_feed_thr_cv; static uint8_t l2arc_thread_exit; static abd_t *arc_get_data_abd(arc_buf_hdr_t *, uint64_t, void *); static void *arc_get_data_buf(arc_buf_hdr_t *, uint64_t, void *); static void arc_get_data_impl(arc_buf_hdr_t *, uint64_t, void *); static void arc_free_data_abd(arc_buf_hdr_t *, abd_t *, uint64_t, void *); static void arc_free_data_buf(arc_buf_hdr_t *, void *, uint64_t, void *); static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag); static void arc_hdr_free_pabd(arc_buf_hdr_t *); static void arc_hdr_alloc_pabd(arc_buf_hdr_t *); static void arc_access(arc_buf_hdr_t *, kmutex_t *); static boolean_t arc_is_overflowing(); static void arc_buf_watch(arc_buf_t *); static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); static uint32_t arc_bufc_to_flags(arc_buf_contents_t); static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); static void l2arc_read_done(zio_t *); static void l2arc_trim(const arc_buf_hdr_t *hdr) { l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; ASSERT(HDR_HAS_L2HDR(hdr)); ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); if (HDR_GET_PSIZE(hdr) != 0) { trim_map_free(dev->l2ad_vdev, hdr->b_l2hdr.b_daddr, HDR_GET_PSIZE(hdr), 0); } } static uint64_t buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) { uint8_t *vdva = (uint8_t *)dva; uint64_t crc = -1ULL; int i; ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); for (i = 0; i < sizeof (dva_t); i++) crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ vdva[i]) & 0xFF]; crc ^= (spa>>8) ^ birth; return (crc); } #define HDR_EMPTY(hdr) \ ((hdr)->b_dva.dva_word[0] == 0 && \ (hdr)->b_dva.dva_word[1] == 0) #define HDR_EQUAL(spa, dva, birth, hdr) \ ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa) static void buf_discard_identity(arc_buf_hdr_t *hdr) { hdr->b_dva.dva_word[0] = 0; hdr->b_dva.dva_word[1] = 0; hdr->b_birth = 0; } static arc_buf_hdr_t * buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) { const dva_t *dva = BP_IDENTITY(bp); uint64_t birth = BP_PHYSICAL_BIRTH(bp); uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); kmutex_t *hash_lock = BUF_HASH_LOCK(idx); arc_buf_hdr_t *hdr; mutex_enter(hash_lock); for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; hdr = hdr->b_hash_next) { if (HDR_EQUAL(spa, dva, birth, hdr)) { *lockp = hash_lock; return (hdr); } } mutex_exit(hash_lock); *lockp = NULL; return (NULL); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. * If lockp == NULL, the caller is assumed to already hold the hash lock. */ static arc_buf_hdr_t * buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) { uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); kmutex_t *hash_lock = BUF_HASH_LOCK(idx); arc_buf_hdr_t *fhdr; uint32_t i; ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); ASSERT(hdr->b_birth != 0); ASSERT(!HDR_IN_HASH_TABLE(hdr)); if (lockp != NULL) { *lockp = hash_lock; mutex_enter(hash_lock); } else { ASSERT(MUTEX_HELD(hash_lock)); } for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; fhdr = fhdr->b_hash_next, i++) { if (HDR_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) return (fhdr); } hdr->b_hash_next = buf_hash_table.ht_table[idx]; buf_hash_table.ht_table[idx] = hdr; arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); /* collect some hash table performance data */ if (i > 0) { ARCSTAT_BUMP(arcstat_hash_collisions); if (i == 1) ARCSTAT_BUMP(arcstat_hash_chains); ARCSTAT_MAX(arcstat_hash_chain_max, i); } ARCSTAT_BUMP(arcstat_hash_elements); ARCSTAT_MAXSTAT(arcstat_hash_elements); return (NULL); } static void buf_hash_remove(arc_buf_hdr_t *hdr) { arc_buf_hdr_t *fhdr, **hdrp; uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); ASSERT(HDR_IN_HASH_TABLE(hdr)); hdrp = &buf_hash_table.ht_table[idx]; while ((fhdr = *hdrp) != hdr) { ASSERT3P(fhdr, !=, NULL); hdrp = &fhdr->b_hash_next; } *hdrp = hdr->b_hash_next; hdr->b_hash_next = NULL; arc_hdr_clear_flags(hdr, ARC_FLAG_IN_HASH_TABLE); /* collect some hash table performance data */ ARCSTAT_BUMPDOWN(arcstat_hash_elements); if (buf_hash_table.ht_table[idx] && buf_hash_table.ht_table[idx]->b_hash_next == NULL) ARCSTAT_BUMPDOWN(arcstat_hash_chains); } /* * Global data structures and functions for the buf kmem cache. */ static kmem_cache_t *hdr_full_cache; static kmem_cache_t *hdr_l2only_cache; static kmem_cache_t *buf_cache; static void buf_fini(void) { int i; kmem_free(buf_hash_table.ht_table, (buf_hash_table.ht_mask + 1) * sizeof (void *)); for (i = 0; i < BUF_LOCKS; i++) mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); kmem_cache_destroy(hdr_full_cache); kmem_cache_destroy(hdr_l2only_cache); kmem_cache_destroy(buf_cache); } /* * Constructor callback - called when the cache is empty * and a new buf is requested. */ /* ARGSUSED */ static int hdr_full_cons(void *vbuf, void *unused, int kmflag) { arc_buf_hdr_t *hdr = vbuf; bzero(hdr, HDR_FULL_SIZE); cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); refcount_create(&hdr->b_l1hdr.b_refcnt); mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); multilist_link_init(&hdr->b_l1hdr.b_arc_node); arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); return (0); } /* ARGSUSED */ static int hdr_l2only_cons(void *vbuf, void *unused, int kmflag) { arc_buf_hdr_t *hdr = vbuf; bzero(hdr, HDR_L2ONLY_SIZE); arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); return (0); } /* ARGSUSED */ static int buf_cons(void *vbuf, void *unused, int kmflag) { arc_buf_t *buf = vbuf; bzero(buf, sizeof (arc_buf_t)); mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); return (0); } /* * Destructor callback - called when a cached buf is * no longer required. */ /* ARGSUSED */ static void hdr_full_dest(void *vbuf, void *unused) { arc_buf_hdr_t *hdr = vbuf; ASSERT(HDR_EMPTY(hdr)); cv_destroy(&hdr->b_l1hdr.b_cv); refcount_destroy(&hdr->b_l1hdr.b_refcnt); mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); } /* ARGSUSED */ static void hdr_l2only_dest(void *vbuf, void *unused) { arc_buf_hdr_t *hdr = vbuf; ASSERT(HDR_EMPTY(hdr)); arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); } /* ARGSUSED */ static void buf_dest(void *vbuf, void *unused) { arc_buf_t *buf = vbuf; mutex_destroy(&buf->b_evict_lock); arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); } /* * Reclaim callback -- invoked when memory is low. */ /* ARGSUSED */ static void hdr_recl(void *unused) { dprintf("hdr_recl called\n"); /* * umem calls the reclaim func when we destroy the buf cache, * which is after we do arc_fini(). */ if (!arc_dead) cv_signal(&arc_reclaim_thread_cv); } static void buf_init(void) { uint64_t *ct; uint64_t hsize = 1ULL << 12; int i, j; /* * The hash table is big enough to fill all of physical memory * with an average block size of zfs_arc_average_blocksize (default 8K). * By default, the table will take up * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). */ while (hsize * zfs_arc_average_blocksize < (uint64_t)physmem * PAGESIZE) hsize <<= 1; retry: buf_hash_table.ht_mask = hsize - 1; buf_hash_table.ht_table = kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); if (buf_hash_table.ht_table == NULL) { ASSERT(hsize > (1ULL << 8)); hsize >>= 1; goto retry; } hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, 0, hdr_full_cons, hdr_full_dest, hdr_recl, NULL, NULL, 0); hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, hdr_recl, NULL, NULL, 0); buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); for (i = 0; i < 256; i++) for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); for (i = 0; i < BUF_LOCKS; i++) { mutex_init(&buf_hash_table.ht_locks[i].ht_lock, NULL, MUTEX_DEFAULT, NULL); } } /* * This is the size that the buf occupies in memory. If the buf is compressed, * it will correspond to the compressed size. You should use this method of * getting the buf size unless you explicitly need the logical size. */ int32_t arc_buf_size(arc_buf_t *buf) { return (ARC_BUF_COMPRESSED(buf) ? HDR_GET_PSIZE(buf->b_hdr) : HDR_GET_LSIZE(buf->b_hdr)); } int32_t arc_buf_lsize(arc_buf_t *buf) { return (HDR_GET_LSIZE(buf->b_hdr)); } enum zio_compress arc_get_compression(arc_buf_t *buf) { return (ARC_BUF_COMPRESSED(buf) ? HDR_GET_COMPRESS(buf->b_hdr) : ZIO_COMPRESS_OFF); } #define ARC_MINTIME (hz>>4) /* 62 ms */ static inline boolean_t arc_buf_is_shared(arc_buf_t *buf) { boolean_t shared = (buf->b_data != NULL && buf->b_hdr->b_l1hdr.b_pabd != NULL && abd_is_linear(buf->b_hdr->b_l1hdr.b_pabd) && buf->b_data == abd_to_buf(buf->b_hdr->b_l1hdr.b_pabd)); IMPLY(shared, HDR_SHARED_DATA(buf->b_hdr)); IMPLY(shared, ARC_BUF_SHARED(buf)); IMPLY(shared, ARC_BUF_COMPRESSED(buf) || ARC_BUF_LAST(buf)); /* * It would be nice to assert arc_can_share() too, but the "hdr isn't * already being shared" requirement prevents us from doing that. */ return (shared); } /* * Free the checksum associated with this header. If there is no checksum, this * is a no-op. */ static inline void arc_cksum_free(arc_buf_hdr_t *hdr) { ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum != NULL) { kmem_free(hdr->b_l1hdr.b_freeze_cksum, sizeof (zio_cksum_t)); hdr->b_l1hdr.b_freeze_cksum = NULL; } mutex_exit(&hdr->b_l1hdr.b_freeze_lock); } /* * Return true iff at least one of the bufs on hdr is not compressed. */ static boolean_t arc_hdr_has_uncompressed_buf(arc_buf_hdr_t *hdr) { for (arc_buf_t *b = hdr->b_l1hdr.b_buf; b != NULL; b = b->b_next) { if (!ARC_BUF_COMPRESSED(b)) { return (B_TRUE); } } return (B_FALSE); } /* * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data * matches the checksum that is stored in the hdr. If there is no checksum, * or if the buf is compressed, this is a no-op. */ static void arc_cksum_verify(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; zio_cksum_t zc; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; if (ARC_BUF_COMPRESSED(buf)) { ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || arc_hdr_has_uncompressed_buf(hdr)); return; } ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum == NULL || HDR_IO_ERROR(hdr)) { mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, &zc); if (!ZIO_CHECKSUM_EQUAL(*hdr->b_l1hdr.b_freeze_cksum, zc)) panic("buffer modified while frozen!"); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); } static boolean_t arc_cksum_is_equal(arc_buf_hdr_t *hdr, zio_t *zio) { enum zio_compress compress = BP_GET_COMPRESS(zio->io_bp); boolean_t valid_cksum; ASSERT(!BP_IS_EMBEDDED(zio->io_bp)); VERIFY3U(BP_GET_PSIZE(zio->io_bp), ==, HDR_GET_PSIZE(hdr)); /* * We rely on the blkptr's checksum to determine if the block * is valid or not. When compressed arc is enabled, the l2arc * writes the block to the l2arc just as it appears in the pool. * This allows us to use the blkptr's checksum to validate the * data that we just read off of the l2arc without having to store * a separate checksum in the arc_buf_hdr_t. However, if compressed * arc is disabled, then the data written to the l2arc is always * uncompressed and won't match the block as it exists in the main * pool. When this is the case, we must first compress it if it is * compressed on the main pool before we can validate the checksum. */ if (!HDR_COMPRESSION_ENABLED(hdr) && compress != ZIO_COMPRESS_OFF) { ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); uint64_t lsize = HDR_GET_LSIZE(hdr); uint64_t csize; void *cbuf = zio_buf_alloc(HDR_GET_PSIZE(hdr)); csize = zio_compress_data(compress, zio->io_abd, cbuf, lsize); ASSERT3U(csize, <=, HDR_GET_PSIZE(hdr)); if (csize < HDR_GET_PSIZE(hdr)) { /* * Compressed blocks are always a multiple of the * smallest ashift in the pool. Ideally, we would * like to round up the csize to the next * spa_min_ashift but that value may have changed * since the block was last written. Instead, * we rely on the fact that the hdr's psize * was set to the psize of the block when it was * last written. We set the csize to that value * and zero out any part that should not contain * data. */ bzero((char *)cbuf + csize, HDR_GET_PSIZE(hdr) - csize); csize = HDR_GET_PSIZE(hdr); } zio_push_transform(zio, cbuf, csize, HDR_GET_PSIZE(hdr), NULL); } /* * Block pointers always store the checksum for the logical data. * If the block pointer has the gang bit set, then the checksum * it represents is for the reconstituted data and not for an * individual gang member. The zio pipeline, however, must be able to * determine the checksum of each of the gang constituents so it * treats the checksum comparison differently than what we need * for l2arc blocks. This prevents us from using the * zio_checksum_error() interface directly. Instead we must call the * zio_checksum_error_impl() so that we can ensure the checksum is * generated using the correct checksum algorithm and accounts for the * logical I/O size and not just a gang fragment. */ valid_cksum = (zio_checksum_error_impl(zio->io_spa, zio->io_bp, BP_GET_CHECKSUM(zio->io_bp), zio->io_abd, zio->io_size, zio->io_offset, NULL) == 0); zio_pop_transforms(zio); return (valid_cksum); } /* * Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a * checksum and attaches it to the buf's hdr so that we can ensure that the buf * isn't modified later on. If buf is compressed or there is already a checksum * on the hdr, this is a no-op (we only checksum uncompressed bufs). */ static void arc_cksum_compute(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum != NULL) { ASSERT(arc_hdr_has_uncompressed_buf(hdr)); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } else if (ARC_BUF_COMPRESSED(buf)) { mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } ASSERT(!ARC_BUF_COMPRESSED(buf)); hdr->b_l1hdr.b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), KM_SLEEP); fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, hdr->b_l1hdr.b_freeze_cksum); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); #ifdef illumos arc_buf_watch(buf); #endif } #ifdef illumos #ifndef _KERNEL typedef struct procctl { long cmd; prwatch_t prwatch; } procctl_t; #endif /* ARGSUSED */ static void arc_buf_unwatch(arc_buf_t *buf) { #ifndef _KERNEL if (arc_watch) { int result; procctl_t ctl; ctl.cmd = PCWATCH; ctl.prwatch.pr_vaddr = (uintptr_t)buf->b_data; ctl.prwatch.pr_size = 0; ctl.prwatch.pr_wflags = 0; result = write(arc_procfd, &ctl, sizeof (ctl)); ASSERT3U(result, ==, sizeof (ctl)); } #endif } /* ARGSUSED */ static void arc_buf_watch(arc_buf_t *buf) { #ifndef _KERNEL if (arc_watch) { int result; procctl_t ctl; ctl.cmd = PCWATCH; ctl.prwatch.pr_vaddr = (uintptr_t)buf->b_data; ctl.prwatch.pr_size = arc_buf_size(buf); ctl.prwatch.pr_wflags = WA_WRITE; result = write(arc_procfd, &ctl, sizeof (ctl)); ASSERT3U(result, ==, sizeof (ctl)); } #endif } #endif /* illumos */ static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *hdr) { arc_buf_contents_t type; if (HDR_ISTYPE_METADATA(hdr)) { type = ARC_BUFC_METADATA; } else { type = ARC_BUFC_DATA; } VERIFY3U(hdr->b_type, ==, type); return (type); } boolean_t arc_is_metadata(arc_buf_t *buf) { return (HDR_ISTYPE_METADATA(buf->b_hdr) != 0); } static uint32_t arc_bufc_to_flags(arc_buf_contents_t type) { switch (type) { case ARC_BUFC_DATA: /* metadata field is 0 if buffer contains normal data */ return (0); case ARC_BUFC_METADATA: return (ARC_FLAG_BUFC_METADATA); default: break; } panic("undefined ARC buffer type!"); return ((uint32_t)-1); } void arc_buf_thaw(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); arc_cksum_verify(buf); /* * Compressed buffers do not manipulate the b_freeze_cksum or * allocate b_thawed. */ if (ARC_BUF_COMPRESSED(buf)) { ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || arc_hdr_has_uncompressed_buf(hdr)); return; } ASSERT(HDR_HAS_L1HDR(hdr)); arc_cksum_free(hdr); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); #ifdef ZFS_DEBUG if (zfs_flags & ZFS_DEBUG_MODIFY) { if (hdr->b_l1hdr.b_thawed != NULL) kmem_free(hdr->b_l1hdr.b_thawed, 1); hdr->b_l1hdr.b_thawed = kmem_alloc(1, KM_SLEEP); } #endif mutex_exit(&hdr->b_l1hdr.b_freeze_lock); #ifdef illumos arc_buf_unwatch(buf); #endif } void arc_buf_freeze(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; kmutex_t *hash_lock; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; if (ARC_BUF_COMPRESSED(buf)) { ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || arc_hdr_has_uncompressed_buf(hdr)); return; } hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(hdr->b_l1hdr.b_freeze_cksum != NULL || hdr->b_l1hdr.b_state == arc_anon); arc_cksum_compute(buf); mutex_exit(hash_lock); } /* * The arc_buf_hdr_t's b_flags should never be modified directly. Instead, * the following functions should be used to ensure that the flags are * updated in a thread-safe way. When manipulating the flags either * the hash_lock must be held or the hdr must be undiscoverable. This * ensures that we're not racing with any other threads when updating * the flags. */ static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) { ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); hdr->b_flags |= flags; } static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) { ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); hdr->b_flags &= ~flags; } /* * Setting the compression bits in the arc_buf_hdr_t's b_flags is * done in a special way since we have to clear and set bits * at the same time. Consumers that wish to set the compression bits * must use this function to ensure that the flags are updated in * thread-safe manner. */ static void arc_hdr_set_compress(arc_buf_hdr_t *hdr, enum zio_compress cmp) { ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * Holes and embedded blocks will always have a psize = 0 so * we ignore the compression of the blkptr and set the * arc_buf_hdr_t's compression to ZIO_COMPRESS_OFF. * Holes and embedded blocks remain anonymous so we don't * want to uncompress them. Mark them as uncompressed. */ if (!zfs_compressed_arc_enabled || HDR_GET_PSIZE(hdr) == 0) { arc_hdr_clear_flags(hdr, ARC_FLAG_COMPRESSED_ARC); HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF); ASSERT(!HDR_COMPRESSION_ENABLED(hdr)); ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); } else { arc_hdr_set_flags(hdr, ARC_FLAG_COMPRESSED_ARC); HDR_SET_COMPRESS(hdr, cmp); ASSERT3U(HDR_GET_COMPRESS(hdr), ==, cmp); ASSERT(HDR_COMPRESSION_ENABLED(hdr)); } } /* * Looks for another buf on the same hdr which has the data decompressed, copies * from it, and returns true. If no such buf exists, returns false. */ static boolean_t arc_buf_try_copy_decompressed_data(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; boolean_t copied = B_FALSE; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3P(buf->b_data, !=, NULL); ASSERT(!ARC_BUF_COMPRESSED(buf)); for (arc_buf_t *from = hdr->b_l1hdr.b_buf; from != NULL; from = from->b_next) { /* can't use our own data buffer */ if (from == buf) { continue; } if (!ARC_BUF_COMPRESSED(from)) { bcopy(from->b_data, buf->b_data, arc_buf_size(buf)); copied = B_TRUE; break; } } /* * There were no decompressed bufs, so there should not be a * checksum on the hdr either. */ EQUIV(!copied, hdr->b_l1hdr.b_freeze_cksum == NULL); return (copied); } /* * Given a buf that has a data buffer attached to it, this function will * efficiently fill the buf with data of the specified compression setting from * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr * are already sharing a data buf, no copy is performed. * * If the buf is marked as compressed but uncompressed data was requested, this * will allocate a new data buffer for the buf, remove that flag, and fill the * buf with uncompressed data. You can't request a compressed buf on a hdr with * uncompressed data, and (since we haven't added support for it yet) if you * want compressed data your buf must already be marked as compressed and have * the correct-sized data buffer. */ static int arc_buf_fill(arc_buf_t *buf, boolean_t compressed) { arc_buf_hdr_t *hdr = buf->b_hdr; boolean_t hdr_compressed = (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); dmu_object_byteswap_t bswap = hdr->b_l1hdr.b_byteswap; ASSERT3P(buf->b_data, !=, NULL); IMPLY(compressed, hdr_compressed); IMPLY(compressed, ARC_BUF_COMPRESSED(buf)); if (hdr_compressed == compressed) { if (!arc_buf_is_shared(buf)) { abd_copy_to_buf(buf->b_data, hdr->b_l1hdr.b_pabd, arc_buf_size(buf)); } } else { ASSERT(hdr_compressed); ASSERT(!compressed); ASSERT3U(HDR_GET_LSIZE(hdr), !=, HDR_GET_PSIZE(hdr)); /* * If the buf is sharing its data with the hdr, unlink it and * allocate a new data buffer for the buf. */ if (arc_buf_is_shared(buf)) { ASSERT(ARC_BUF_COMPRESSED(buf)); /* We need to give the buf it's own b_data */ buf->b_flags &= ~ARC_BUF_FLAG_SHARED; buf->b_data = arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); /* Previously overhead was 0; just add new overhead */ ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr)); } else if (ARC_BUF_COMPRESSED(buf)) { /* We need to reallocate the buf's b_data */ arc_free_data_buf(hdr, buf->b_data, HDR_GET_PSIZE(hdr), buf); buf->b_data = arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); /* We increased the size of b_data; update overhead */ ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr) - HDR_GET_PSIZE(hdr)); } /* * Regardless of the buf's previous compression settings, it * should not be compressed at the end of this function. */ buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; /* * Try copying the data from another buf which already has a * decompressed version. If that's not possible, it's time to * bite the bullet and decompress the data from the hdr. */ if (arc_buf_try_copy_decompressed_data(buf)) { /* Skip byteswapping and checksumming (already done) */ ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, !=, NULL); return (0); } else { int error = zio_decompress_data(HDR_GET_COMPRESS(hdr), hdr->b_l1hdr.b_pabd, buf->b_data, HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); /* * Absent hardware errors or software bugs, this should * be impossible, but log it anyway so we can debug it. */ if (error != 0) { zfs_dbgmsg( "hdr %p, compress %d, psize %d, lsize %d", hdr, HDR_GET_COMPRESS(hdr), HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); return (SET_ERROR(EIO)); } } } /* Byteswap the buf's data if necessary */ if (bswap != DMU_BSWAP_NUMFUNCS) { ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT3U(bswap, <, DMU_BSWAP_NUMFUNCS); dmu_ot_byteswap[bswap].ob_func(buf->b_data, HDR_GET_LSIZE(hdr)); } /* Compute the hdr's checksum if necessary */ arc_cksum_compute(buf); return (0); } int arc_decompress(arc_buf_t *buf) { return (arc_buf_fill(buf, B_FALSE)); } /* * Return the size of the block, b_pabd, that is stored in the arc_buf_hdr_t. */ static uint64_t arc_hdr_size(arc_buf_hdr_t *hdr) { uint64_t size; if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && HDR_GET_PSIZE(hdr) > 0) { size = HDR_GET_PSIZE(hdr); } else { ASSERT3U(HDR_GET_LSIZE(hdr), !=, 0); size = HDR_GET_LSIZE(hdr); } return (size); } /* * Increment the amount of evictable space in the arc_state_t's refcount. * We account for the space used by the hdr and the arc buf individually * so that we can add and remove them from the refcount individually. */ static void arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state) { arc_buf_contents_t type = arc_buf_type(hdr); ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(state)) { ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); (void) refcount_add_many(&state->arcs_esize[type], HDR_GET_LSIZE(hdr), hdr); return; } ASSERT(!GHOST_STATE(state)); if (hdr->b_l1hdr.b_pabd != NULL) { (void) refcount_add_many(&state->arcs_esize[type], arc_hdr_size(hdr), hdr); } for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { if (arc_buf_is_shared(buf)) continue; (void) refcount_add_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } } /* * Decrement the amount of evictable space in the arc_state_t's refcount. * We account for the space used by the hdr and the arc buf individually * so that we can add and remove them from the refcount individually. */ static void arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state) { arc_buf_contents_t type = arc_buf_type(hdr); ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(state)) { ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); (void) refcount_remove_many(&state->arcs_esize[type], HDR_GET_LSIZE(hdr), hdr); return; } ASSERT(!GHOST_STATE(state)); if (hdr->b_l1hdr.b_pabd != NULL) { (void) refcount_remove_many(&state->arcs_esize[type], arc_hdr_size(hdr), hdr); } for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { if (arc_buf_is_shared(buf)) continue; (void) refcount_remove_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } } /* * Add a reference to this hdr indicating that someone is actively * referencing that memory. When the refcount transitions from 0 to 1, * we remove it from the respective arc_state_t list to indicate that * it is not evictable. */ static void add_reference(arc_buf_hdr_t *hdr, void *tag) { ASSERT(HDR_HAS_L1HDR(hdr)); if (!MUTEX_HELD(HDR_LOCK(hdr))) { ASSERT(hdr->b_l1hdr.b_state == arc_anon); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); } arc_state_t *state = hdr->b_l1hdr.b_state; if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && (state != arc_anon)) { /* We don't use the L2-only state list. */ if (state != arc_l2c_only) { multilist_remove(state->arcs_list[arc_buf_type(hdr)], hdr); arc_evictable_space_decrement(hdr, state); } /* remove the prefetch flag if we get a reference */ arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); } } /* * Remove a reference from this hdr. When the reference transitions from * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's * list making it eligible for eviction. */ static int remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) { int cnt; arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); ASSERT(!GHOST_STATE(state)); /* * arc_l2c_only counts as a ghost state so we don't need to explicitly * check to prevent usage of the arc_l2c_only list. */ if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && (state != arc_anon)) { multilist_insert(state->arcs_list[arc_buf_type(hdr)], hdr); ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); arc_evictable_space_increment(hdr, state); } return (cnt); } /* * Move the supplied buffer to the indicated state. The hash lock * for the buffer must be held by the caller. */ static void arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { arc_state_t *old_state; int64_t refcnt; uint32_t bufcnt; boolean_t update_old, update_new; arc_buf_contents_t buftype = arc_buf_type(hdr); /* * We almost always have an L1 hdr here, since we call arc_hdr_realloc() * in arc_read() when bringing a buffer out of the L2ARC. However, the * L1 hdr doesn't always exist when we change state to arc_anon before * destroying a header, in which case reallocating to add the L1 hdr is * pointless. */ if (HDR_HAS_L1HDR(hdr)) { old_state = hdr->b_l1hdr.b_state; refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt); bufcnt = hdr->b_l1hdr.b_bufcnt; update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pabd != NULL); } else { old_state = arc_l2c_only; refcnt = 0; bufcnt = 0; update_old = B_FALSE; } update_new = update_old; ASSERT(MUTEX_HELD(hash_lock)); ASSERT3P(new_state, !=, old_state); ASSERT(!GHOST_STATE(new_state) || bufcnt == 0); ASSERT(old_state != arc_anon || bufcnt <= 1); /* * If this buffer is evictable, transfer it from the * old state list to the new state list. */ if (refcnt == 0) { if (old_state != arc_anon && old_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); multilist_remove(old_state->arcs_list[buftype], hdr); if (GHOST_STATE(old_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); update_old = B_TRUE; } arc_evictable_space_decrement(hdr, old_state); } if (new_state != arc_anon && new_state != arc_l2c_only) { /* * An L1 header always exists here, since if we're * moving to some L1-cached state (i.e. not l2c_only or * anonymous), we realloc the header to add an L1hdr * beforehand. */ ASSERT(HDR_HAS_L1HDR(hdr)); multilist_insert(new_state->arcs_list[buftype], hdr); if (GHOST_STATE(new_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); update_new = B_TRUE; } arc_evictable_space_increment(hdr, new_state); } } ASSERT(!HDR_EMPTY(hdr)); if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) buf_hash_remove(hdr); /* adjust state sizes (ignore arc_l2c_only) */ if (update_new && new_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(new_state)) { ASSERT0(bufcnt); /* * When moving a header to a ghost state, we first * remove all arc buffers. Thus, we'll have a * bufcnt of zero, and no arc buffer to use for * the reference. As a result, we use the arc * header pointer for the reference. */ (void) refcount_add_many(&new_state->arcs_size, HDR_GET_LSIZE(hdr), hdr); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); } else { uint32_t buffers = 0; /* * Each individual buffer holds a unique reference, * thus we must remove each of these references one * at a time. */ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { ASSERT3U(bufcnt, !=, 0); buffers++; /* * When the arc_buf_t is sharing the data * block with the hdr, the owner of the * reference belongs to the hdr. Only * add to the refcount if the arc_buf_t is * not shared. */ if (arc_buf_is_shared(buf)) continue; (void) refcount_add_many(&new_state->arcs_size, arc_buf_size(buf), buf); } ASSERT3U(bufcnt, ==, buffers); if (hdr->b_l1hdr.b_pabd != NULL) { (void) refcount_add_many(&new_state->arcs_size, arc_hdr_size(hdr), hdr); } else { ASSERT(GHOST_STATE(old_state)); } } } if (update_old && old_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(old_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); /* * When moving a header off of a ghost state, * the header will not contain any arc buffers. * We use the arc header pointer for the reference * which is exactly what we did when we put the * header on the ghost state. */ (void) refcount_remove_many(&old_state->arcs_size, HDR_GET_LSIZE(hdr), hdr); } else { uint32_t buffers = 0; /* * Each individual buffer holds a unique reference, * thus we must remove each of these references one * at a time. */ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { ASSERT3U(bufcnt, !=, 0); buffers++; /* * When the arc_buf_t is sharing the data * block with the hdr, the owner of the * reference belongs to the hdr. Only * add to the refcount if the arc_buf_t is * not shared. */ if (arc_buf_is_shared(buf)) continue; (void) refcount_remove_many( &old_state->arcs_size, arc_buf_size(buf), buf); } ASSERT3U(bufcnt, ==, buffers); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); (void) refcount_remove_many( &old_state->arcs_size, arc_hdr_size(hdr), hdr); } } if (HDR_HAS_L1HDR(hdr)) hdr->b_l1hdr.b_state = new_state; /* * L2 headers should never be on the L2 state list since they don't * have L1 headers allocated. */ ASSERT(multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); } void arc_space_consume(uint64_t space, arc_space_type_t type) { ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); switch (type) { case ARC_SPACE_DATA: ARCSTAT_INCR(arcstat_data_size, space); break; case ARC_SPACE_META: ARCSTAT_INCR(arcstat_metadata_size, space); break; case ARC_SPACE_OTHER: ARCSTAT_INCR(arcstat_other_size, space); break; case ARC_SPACE_HDRS: ARCSTAT_INCR(arcstat_hdr_size, space); break; case ARC_SPACE_L2HDRS: ARCSTAT_INCR(arcstat_l2_hdr_size, space); break; } if (type != ARC_SPACE_DATA) ARCSTAT_INCR(arcstat_meta_used, space); atomic_add_64(&arc_size, space); } void arc_space_return(uint64_t space, arc_space_type_t type) { ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); switch (type) { case ARC_SPACE_DATA: ARCSTAT_INCR(arcstat_data_size, -space); break; case ARC_SPACE_META: ARCSTAT_INCR(arcstat_metadata_size, -space); break; case ARC_SPACE_OTHER: ARCSTAT_INCR(arcstat_other_size, -space); break; case ARC_SPACE_HDRS: ARCSTAT_INCR(arcstat_hdr_size, -space); break; case ARC_SPACE_L2HDRS: ARCSTAT_INCR(arcstat_l2_hdr_size, -space); break; } if (type != ARC_SPACE_DATA) { ASSERT(arc_meta_used >= space); if (arc_meta_max < arc_meta_used) arc_meta_max = arc_meta_used; ARCSTAT_INCR(arcstat_meta_used, -space); } ASSERT(arc_size >= space); atomic_add_64(&arc_size, -space); } /* * Given a hdr and a buf, returns whether that buf can share its b_data buffer * with the hdr's b_pabd. */ static boolean_t arc_can_share(arc_buf_hdr_t *hdr, arc_buf_t *buf) { /* * The criteria for sharing a hdr's data are: * 1. the hdr's compression matches the buf's compression * 2. the hdr doesn't need to be byteswapped * 3. the hdr isn't already being shared * 4. the buf is either compressed or it is the last buf in the hdr list * * Criterion #4 maintains the invariant that shared uncompressed * bufs must be the final buf in the hdr's b_buf list. Reading this, you * might ask, "if a compressed buf is allocated first, won't that be the * last thing in the list?", but in that case it's impossible to create * a shared uncompressed buf anyway (because the hdr must be compressed * to have the compressed buf). You might also think that #3 is * sufficient to make this guarantee, however it's possible * (specifically in the rare L2ARC write race mentioned in * arc_buf_alloc_impl()) there will be an existing uncompressed buf that * is sharable, but wasn't at the time of its allocation. Rather than * allow a new shared uncompressed buf to be created and then shuffle * the list around to make it the last element, this simply disallows * sharing if the new buf isn't the first to be added. */ ASSERT3P(buf->b_hdr, ==, hdr); boolean_t hdr_compressed = HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF; boolean_t buf_compressed = ARC_BUF_COMPRESSED(buf) != 0; return (buf_compressed == hdr_compressed && hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS && !HDR_SHARED_DATA(hdr) && (ARC_BUF_LAST(buf) || ARC_BUF_COMPRESSED(buf))); } /* * Allocate a buf for this hdr. If you care about the data that's in the hdr, * or if you want a compressed buffer, pass those flags in. Returns 0 if the * copy was made successfully, or an error code otherwise. */ static int arc_buf_alloc_impl(arc_buf_hdr_t *hdr, void *tag, boolean_t compressed, boolean_t fill, arc_buf_t **ret) { arc_buf_t *buf; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); VERIFY(hdr->b_type == ARC_BUFC_DATA || hdr->b_type == ARC_BUFC_METADATA); ASSERT3P(ret, !=, NULL); ASSERT3P(*ret, ==, NULL); buf = *ret = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); buf->b_hdr = hdr; buf->b_data = NULL; buf->b_next = hdr->b_l1hdr.b_buf; buf->b_flags = 0; add_reference(hdr, tag); /* * We're about to change the hdr's b_flags. We must either * hold the hash_lock or be undiscoverable. */ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * Only honor requests for compressed bufs if the hdr is actually * compressed. */ if (compressed && HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) buf->b_flags |= ARC_BUF_FLAG_COMPRESSED; /* * If the hdr's data can be shared then we share the data buffer and * set the appropriate bit in the hdr's b_flags to indicate the hdr is * sharing it's b_pabd with the arc_buf_t. Otherwise, we allocate a new * buffer to store the buf's data. * * There are two additional restrictions here because we're sharing * hdr -> buf instead of the usual buf -> hdr. First, the hdr can't be * actively involved in an L2ARC write, because if this buf is used by * an arc_write() then the hdr's data buffer will be released when the * write completes, even though the L2ARC write might still be using it. * Second, the hdr's ABD must be linear so that the buf's user doesn't * need to be ABD-aware. */ boolean_t can_share = arc_can_share(hdr, buf) && !HDR_L2_WRITING(hdr) && abd_is_linear(hdr->b_l1hdr.b_pabd); /* Set up b_data and sharing */ if (can_share) { buf->b_data = abd_to_buf(hdr->b_l1hdr.b_pabd); buf->b_flags |= ARC_BUF_FLAG_SHARED; arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); } else { buf->b_data = arc_get_data_buf(hdr, arc_buf_size(buf), buf); ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); } VERIFY3P(buf->b_data, !=, NULL); hdr->b_l1hdr.b_buf = buf; hdr->b_l1hdr.b_bufcnt += 1; /* * If the user wants the data from the hdr, we need to either copy or * decompress the data. */ if (fill) { return (arc_buf_fill(buf, ARC_BUF_COMPRESSED(buf) != 0)); } return (0); } static char *arc_onloan_tag = "onloan"; static inline void arc_loaned_bytes_update(int64_t delta) { atomic_add_64(&arc_loaned_bytes, delta); /* assert that it did not wrap around */ ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); } /* * Loan out an anonymous arc buffer. Loaned buffers are not counted as in * flight data by arc_tempreserve_space() until they are "returned". Loaned * buffers must be returned to the arc before they can be used by the DMU or * freed. */ arc_buf_t * arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size) { arc_buf_t *buf = arc_alloc_buf(spa, arc_onloan_tag, is_metadata ? ARC_BUFC_METADATA : ARC_BUFC_DATA, size); arc_loaned_bytes_update(size); return (buf); } arc_buf_t * arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize, enum zio_compress compression_type) { arc_buf_t *buf = arc_alloc_compressed_buf(spa, arc_onloan_tag, psize, lsize, compression_type); arc_loaned_bytes_update(psize); return (buf); } /* * Return a loaned arc buffer to the arc. */ void arc_return_buf(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(buf->b_data, !=, NULL); ASSERT(HDR_HAS_L1HDR(hdr)); (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); arc_loaned_bytes_update(-arc_buf_size(buf)); } /* Detach an arc_buf from a dbuf (tag) */ void arc_loan_inuse_buf(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(buf->b_data, !=, NULL); ASSERT(HDR_HAS_L1HDR(hdr)); (void) refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, tag); arc_loaned_bytes_update(arc_buf_size(buf)); } static void l2arc_free_abd_on_write(abd_t *abd, size_t size, arc_buf_contents_t type) { l2arc_data_free_t *df = kmem_alloc(sizeof (*df), KM_SLEEP); df->l2df_abd = abd; df->l2df_size = size; df->l2df_type = type; mutex_enter(&l2arc_free_on_write_mtx); list_insert_head(l2arc_free_on_write, df); mutex_exit(&l2arc_free_on_write_mtx); } static void arc_hdr_free_on_write(arc_buf_hdr_t *hdr) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); uint64_t size = arc_hdr_size(hdr); /* protected by hash lock, if in the hash table */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT(state != arc_anon && state != arc_l2c_only); (void) refcount_remove_many(&state->arcs_esize[type], size, hdr); } (void) refcount_remove_many(&state->arcs_size, size, hdr); if (type == ARC_BUFC_METADATA) { arc_space_return(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); arc_space_return(size, ARC_SPACE_DATA); } l2arc_free_abd_on_write(hdr->b_l1hdr.b_pabd, size, type); } /* * Share the arc_buf_t's data with the hdr. Whenever we are sharing the * data buffer, we transfer the refcount ownership to the hdr and update * the appropriate kstats. */ static void arc_share_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) { arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT(arc_can_share(hdr, buf)); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * Start sharing the data buffer. We transfer the * refcount ownership to the hdr since it always owns * the refcount whenever an arc_buf_t is shared. */ refcount_transfer_ownership(&state->arcs_size, buf, hdr); hdr->b_l1hdr.b_pabd = abd_get_from_buf(buf->b_data, arc_buf_size(buf)); abd_take_ownership_of_buf(hdr->b_l1hdr.b_pabd, HDR_ISTYPE_METADATA(hdr)); arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); buf->b_flags |= ARC_BUF_FLAG_SHARED; /* * Since we've transferred ownership to the hdr we need * to increment its compressed and uncompressed kstats and * decrement the overhead size. */ ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); ARCSTAT_INCR(arcstat_overhead_size, -arc_buf_size(buf)); } static void arc_unshare_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) { arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT(arc_buf_is_shared(buf)); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * We are no longer sharing this buffer so we need * to transfer its ownership to the rightful owner. */ refcount_transfer_ownership(&state->arcs_size, hdr, buf); arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); abd_release_ownership_of_buf(hdr->b_l1hdr.b_pabd); abd_put(hdr->b_l1hdr.b_pabd); hdr->b_l1hdr.b_pabd = NULL; buf->b_flags &= ~ARC_BUF_FLAG_SHARED; /* * Since the buffer is no longer shared between * the arc buf and the hdr, count it as overhead. */ ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); } /* * Remove an arc_buf_t from the hdr's buf list and return the last * arc_buf_t on the list. If no buffers remain on the list then return * NULL. */ static arc_buf_t * arc_buf_remove(arc_buf_hdr_t *hdr, arc_buf_t *buf) { ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); arc_buf_t **bufp = &hdr->b_l1hdr.b_buf; arc_buf_t *lastbuf = NULL; /* * Remove the buf from the hdr list and locate the last * remaining buffer on the list. */ while (*bufp != NULL) { if (*bufp == buf) *bufp = buf->b_next; /* * If we've removed a buffer in the middle of * the list then update the lastbuf and update * bufp. */ if (*bufp != NULL) { lastbuf = *bufp; bufp = &(*bufp)->b_next; } } buf->b_next = NULL; ASSERT3P(lastbuf, !=, buf); IMPLY(hdr->b_l1hdr.b_bufcnt > 0, lastbuf != NULL); IMPLY(hdr->b_l1hdr.b_bufcnt > 0, hdr->b_l1hdr.b_buf != NULL); IMPLY(lastbuf != NULL, ARC_BUF_LAST(lastbuf)); return (lastbuf); } /* * Free up buf->b_data and pull the arc_buf_t off of the the arc_buf_hdr_t's * list and free it. */ static void arc_buf_destroy_impl(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; /* * Free up the data associated with the buf but only if we're not * sharing this with the hdr. If we are sharing it with the hdr, the * hdr is responsible for doing the free. */ if (buf->b_data != NULL) { /* * We're about to change the hdr's b_flags. We must either * hold the hash_lock or be undiscoverable. */ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); arc_cksum_verify(buf); #ifdef illumos arc_buf_unwatch(buf); #endif if (arc_buf_is_shared(buf)) { arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); } else { uint64_t size = arc_buf_size(buf); arc_free_data_buf(hdr, buf->b_data, size, buf); ARCSTAT_INCR(arcstat_overhead_size, -size); } buf->b_data = NULL; ASSERT(hdr->b_l1hdr.b_bufcnt > 0); hdr->b_l1hdr.b_bufcnt -= 1; } arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); if (ARC_BUF_SHARED(buf) && !ARC_BUF_COMPRESSED(buf)) { /* * If the current arc_buf_t is sharing its data buffer with the * hdr, then reassign the hdr's b_pabd to share it with the new * buffer at the end of the list. The shared buffer is always * the last one on the hdr's buffer list. * * There is an equivalent case for compressed bufs, but since * they aren't guaranteed to be the last buf in the list and * that is an exceedingly rare case, we just allow that space be * wasted temporarily. */ if (lastbuf != NULL) { /* Only one buf can be shared at once */ VERIFY(!arc_buf_is_shared(lastbuf)); /* hdr is uncompressed so can't have compressed buf */ VERIFY(!ARC_BUF_COMPRESSED(lastbuf)); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); arc_hdr_free_pabd(hdr); /* * We must setup a new shared block between the * last buffer and the hdr. The data would have * been allocated by the arc buf so we need to transfer * ownership to the hdr since it's now being shared. */ arc_share_buf(hdr, lastbuf); } } else if (HDR_SHARED_DATA(hdr)) { /* * Uncompressed shared buffers are always at the end * of the list. Compressed buffers don't have the * same requirements. This makes it hard to * simply assert that the lastbuf is shared so * we rely on the hdr's compression flags to determine * if we have a compressed, shared buffer. */ ASSERT3P(lastbuf, !=, NULL); ASSERT(arc_buf_is_shared(lastbuf) || HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); } /* * Free the checksum if we're removing the last uncompressed buf from * this hdr. */ if (!arc_hdr_has_uncompressed_buf(hdr)) { arc_cksum_free(hdr); } /* clean up the buf */ buf->b_hdr = NULL; kmem_cache_free(buf_cache, buf); } static void arc_hdr_alloc_pabd(arc_buf_hdr_t *hdr) { ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr); hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); } static void arc_hdr_free_pabd(arc_buf_hdr_t *hdr) { ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); /* * If the hdr is currently being written to the l2arc then * we defer freeing the data by adding it to the l2arc_free_on_write * list. The l2arc will free the data once it's finished * writing it to the l2arc device. */ if (HDR_L2_WRITING(hdr)) { arc_hdr_free_on_write(hdr); ARCSTAT_BUMP(arcstat_l2_free_on_write); } else { arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, arc_hdr_size(hdr), hdr); } hdr->b_l1hdr.b_pabd = NULL; hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); } static arc_buf_hdr_t * arc_hdr_alloc(uint64_t spa, int32_t psize, int32_t lsize, enum zio_compress compression_type, arc_buf_contents_t type) { arc_buf_hdr_t *hdr; VERIFY(type == ARC_BUFC_DATA || type == ARC_BUFC_METADATA); hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); ASSERT(HDR_EMPTY(hdr)); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_thawed, ==, NULL); HDR_SET_PSIZE(hdr, psize); HDR_SET_LSIZE(hdr, lsize); hdr->b_spa = spa; hdr->b_type = type; hdr->b_flags = 0; arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L1HDR); arc_hdr_set_compress(hdr, compression_type); hdr->b_l1hdr.b_state = arc_anon; hdr->b_l1hdr.b_arc_access = 0; hdr->b_l1hdr.b_bufcnt = 0; hdr->b_l1hdr.b_buf = NULL; /* * Allocate the hdr's buffer. This will contain either * the compressed or uncompressed data depending on the block * it references and compressed arc enablement. */ arc_hdr_alloc_pabd(hdr); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); return (hdr); } /* * Transition between the two allocation states for the arc_buf_hdr struct. * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller * version is used when a cache buffer is only in the L2ARC in order to reduce * memory usage. */ static arc_buf_hdr_t * arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) { ASSERT(HDR_HAS_L2HDR(hdr)); arc_buf_hdr_t *nhdr; l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || (old == hdr_l2only_cache && new == hdr_full_cache)); nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); buf_hash_remove(hdr); bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); if (new == hdr_full_cache) { arc_hdr_set_flags(nhdr, ARC_FLAG_HAS_L1HDR); /* * arc_access and arc_change_state need to be aware that a * header has just come out of L2ARC, so we set its state to * l2c_only even though it's about to change. */ nhdr->b_l1hdr.b_state = arc_l2c_only; /* Verify previous threads set to NULL before freeing */ ASSERT3P(nhdr->b_l1hdr.b_pabd, ==, NULL); } else { ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); /* * If we've reached here, We must have been called from * arc_evict_hdr(), as such we should have already been * removed from any ghost list we were previously on * (which protects us from racing with arc_evict_state), * thus no locking is needed during this check. */ ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); /* * A buffer must not be moved into the arc_l2c_only * state if it's not finished being written out to the * l2arc device. Otherwise, the b_l1hdr.b_pabd field * might try to be accessed, even though it was removed. */ VERIFY(!HDR_L2_WRITING(hdr)); VERIFY3P(hdr->b_l1hdr.b_pabd, ==, NULL); #ifdef ZFS_DEBUG if (hdr->b_l1hdr.b_thawed != NULL) { kmem_free(hdr->b_l1hdr.b_thawed, 1); hdr->b_l1hdr.b_thawed = NULL; } #endif arc_hdr_clear_flags(nhdr, ARC_FLAG_HAS_L1HDR); } /* * The header has been reallocated so we need to re-insert it into any * lists it was on. */ (void) buf_hash_insert(nhdr, NULL); ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); mutex_enter(&dev->l2ad_mtx); /* * We must place the realloc'ed header back into the list at * the same spot. Otherwise, if it's placed earlier in the list, * l2arc_write_buffers() could find it during the function's * write phase, and try to write it out to the l2arc. */ list_insert_after(&dev->l2ad_buflist, hdr, nhdr); list_remove(&dev->l2ad_buflist, hdr); mutex_exit(&dev->l2ad_mtx); /* * Since we're using the pointer address as the tag when * incrementing and decrementing the l2ad_alloc refcount, we * must remove the old pointer (that we're about to destroy) and * add the new pointer to the refcount. Otherwise we'd remove * the wrong pointer address when calling arc_hdr_destroy() later. */ (void) refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); (void) refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(nhdr), nhdr); buf_discard_identity(hdr); kmem_cache_free(old, hdr); return (nhdr); } /* * Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller. * The buf is returned thawed since we expect the consumer to modify it. */ arc_buf_t * arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type, int32_t size) { arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), size, size, ZIO_COMPRESS_OFF, type); ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); arc_buf_t *buf = NULL; VERIFY0(arc_buf_alloc_impl(hdr, tag, B_FALSE, B_FALSE, &buf)); arc_buf_thaw(buf); return (buf); } /* * Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this * for bufs containing metadata. */ arc_buf_t * arc_alloc_compressed_buf(spa_t *spa, void *tag, uint64_t psize, uint64_t lsize, enum zio_compress compression_type) { ASSERT3U(lsize, >, 0); ASSERT3U(lsize, >=, psize); ASSERT(compression_type > ZIO_COMPRESS_OFF); ASSERT(compression_type < ZIO_COMPRESS_FUNCTIONS); arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, compression_type, ARC_BUFC_DATA); ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); arc_buf_t *buf = NULL; VERIFY0(arc_buf_alloc_impl(hdr, tag, B_TRUE, B_FALSE, &buf)); arc_buf_thaw(buf); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); if (!arc_buf_is_shared(buf)) { /* * To ensure that the hdr has the correct data in it if we call * arc_decompress() on this buf before it's been written to * disk, it's easiest if we just set up sharing between the * buf and the hdr. */ ASSERT(!abd_is_linear(hdr->b_l1hdr.b_pabd)); arc_hdr_free_pabd(hdr); arc_share_buf(hdr, buf); } return (buf); } static void arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr) { l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; l2arc_dev_t *dev = l2hdr->b_dev; uint64_t psize = arc_hdr_size(hdr); ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); ASSERT(HDR_HAS_L2HDR(hdr)); list_remove(&dev->l2ad_buflist, hdr); ARCSTAT_INCR(arcstat_l2_psize, -psize); ARCSTAT_INCR(arcstat_l2_lsize, -HDR_GET_LSIZE(hdr)); vdev_space_update(dev->l2ad_vdev, -psize, 0, 0); (void) refcount_remove_many(&dev->l2ad_alloc, psize, hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); } static void arc_hdr_destroy(arc_buf_hdr_t *hdr) { if (HDR_HAS_L1HDR(hdr)) { ASSERT(hdr->b_l1hdr.b_buf == NULL || hdr->b_l1hdr.b_bufcnt > 0); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); } ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(!HDR_IN_HASH_TABLE(hdr)); if (!HDR_EMPTY(hdr)) buf_discard_identity(hdr); if (HDR_HAS_L2HDR(hdr)) { l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); if (!buflist_held) mutex_enter(&dev->l2ad_mtx); /* * Even though we checked this conditional above, we * need to check this again now that we have the * l2ad_mtx. This is because we could be racing with * another thread calling l2arc_evict() which might have * destroyed this header's L2 portion as we were waiting * to acquire the l2ad_mtx. If that happens, we don't * want to re-destroy the header's L2 portion. */ if (HDR_HAS_L2HDR(hdr)) { l2arc_trim(hdr); arc_hdr_l2hdr_destroy(hdr); } if (!buflist_held) mutex_exit(&dev->l2ad_mtx); } if (HDR_HAS_L1HDR(hdr)) { arc_cksum_free(hdr); while (hdr->b_l1hdr.b_buf != NULL) arc_buf_destroy_impl(hdr->b_l1hdr.b_buf); #ifdef ZFS_DEBUG if (hdr->b_l1hdr.b_thawed != NULL) { kmem_free(hdr->b_l1hdr.b_thawed, 1); hdr->b_l1hdr.b_thawed = NULL; } #endif if (hdr->b_l1hdr.b_pabd != NULL) { arc_hdr_free_pabd(hdr); } } ASSERT3P(hdr->b_hash_next, ==, NULL); if (HDR_HAS_L1HDR(hdr)) { ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); kmem_cache_free(hdr_full_cache, hdr); } else { kmem_cache_free(hdr_l2only_cache, hdr); } } void arc_buf_destroy(arc_buf_t *buf, void* tag) { arc_buf_hdr_t *hdr = buf->b_hdr; kmutex_t *hash_lock = HDR_LOCK(hdr); if (hdr->b_l1hdr.b_state == arc_anon) { ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); VERIFY0(remove_reference(hdr, NULL, tag)); arc_hdr_destroy(hdr); return; } mutex_enter(hash_lock); ASSERT3P(hdr, ==, buf->b_hdr); ASSERT(hdr->b_l1hdr.b_bufcnt > 0); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3P(hdr->b_l1hdr.b_state, !=, arc_anon); ASSERT3P(buf->b_data, !=, NULL); (void) remove_reference(hdr, hash_lock, tag); arc_buf_destroy_impl(buf); mutex_exit(hash_lock); } /* * Evict the arc_buf_hdr that is provided as a parameter. The resultant * state of the header is dependent on its state prior to entering this * function. The following transitions are possible: * * - arc_mru -> arc_mru_ghost * - arc_mfu -> arc_mfu_ghost * - arc_mru_ghost -> arc_l2c_only * - arc_mru_ghost -> deleted * - arc_mfu_ghost -> arc_l2c_only * - arc_mfu_ghost -> deleted */ static int64_t arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { arc_state_t *evicted_state, *state; int64_t bytes_evicted = 0; ASSERT(MUTEX_HELD(hash_lock)); ASSERT(HDR_HAS_L1HDR(hdr)); state = hdr->b_l1hdr.b_state; if (GHOST_STATE(state)) { ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); /* * l2arc_write_buffers() relies on a header's L1 portion * (i.e. its b_pabd field) during it's write phase. * Thus, we cannot push a header onto the arc_l2c_only * state (removing it's L1 piece) until the header is * done being written to the l2arc. */ if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { ARCSTAT_BUMP(arcstat_evict_l2_skip); return (bytes_evicted); } ARCSTAT_BUMP(arcstat_deleted); bytes_evicted += HDR_GET_LSIZE(hdr); DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); if (HDR_HAS_L2HDR(hdr)) { /* * This buffer is cached on the 2nd Level ARC; * don't destroy the header. */ arc_change_state(arc_l2c_only, hdr, hash_lock); /* * dropping from L1+L2 cached to L2-only, * realloc to remove the L1 header. */ hdr = arc_hdr_realloc(hdr, hdr_full_cache, hdr_l2only_cache); } else { arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); } return (bytes_evicted); } ASSERT(state == arc_mru || state == arc_mfu); evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; /* prefetch buffers have a minimum lifespan */ if (HDR_IO_IN_PROGRESS(hdr) || ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < arc_min_prefetch_lifespan)) { ARCSTAT_BUMP(arcstat_evict_skip); return (bytes_evicted); } ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); while (hdr->b_l1hdr.b_buf) { arc_buf_t *buf = hdr->b_l1hdr.b_buf; if (!mutex_tryenter(&buf->b_evict_lock)) { ARCSTAT_BUMP(arcstat_mutex_miss); break; } if (buf->b_data != NULL) bytes_evicted += HDR_GET_LSIZE(hdr); mutex_exit(&buf->b_evict_lock); arc_buf_destroy_impl(buf); } if (HDR_HAS_L2HDR(hdr)) { ARCSTAT_INCR(arcstat_evict_l2_cached, HDR_GET_LSIZE(hdr)); } else { if (l2arc_write_eligible(hdr->b_spa, hdr)) { ARCSTAT_INCR(arcstat_evict_l2_eligible, HDR_GET_LSIZE(hdr)); } else { ARCSTAT_INCR(arcstat_evict_l2_ineligible, HDR_GET_LSIZE(hdr)); } } if (hdr->b_l1hdr.b_bufcnt == 0) { arc_cksum_free(hdr); bytes_evicted += arc_hdr_size(hdr); /* * If this hdr is being evicted and has a compressed * buffer then we discard it here before we change states. * This ensures that the accounting is updated correctly * in arc_free_data_impl(). */ arc_hdr_free_pabd(hdr); arc_change_state(evicted_state, hdr, hash_lock); ASSERT(HDR_IN_HASH_TABLE(hdr)); arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); } return (bytes_evicted); } static uint64_t arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, uint64_t spa, int64_t bytes) { multilist_sublist_t *mls; uint64_t bytes_evicted = 0; arc_buf_hdr_t *hdr; kmutex_t *hash_lock; int evict_count = 0; ASSERT3P(marker, !=, NULL); IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); mls = multilist_sublist_lock(ml, idx); for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; hdr = multilist_sublist_prev(mls, marker)) { if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || (evict_count >= zfs_arc_evict_batch_limit)) break; /* * To keep our iteration location, move the marker * forward. Since we're not holding hdr's hash lock, we * must be very careful and not remove 'hdr' from the * sublist. Otherwise, other consumers might mistake the * 'hdr' as not being on a sublist when they call the * multilist_link_active() function (they all rely on * the hash lock protecting concurrent insertions and * removals). multilist_sublist_move_forward() was * specifically implemented to ensure this is the case * (only 'marker' will be removed and re-inserted). */ multilist_sublist_move_forward(mls, marker); /* * The only case where the b_spa field should ever be * zero, is the marker headers inserted by * arc_evict_state(). It's possible for multiple threads * to be calling arc_evict_state() concurrently (e.g. * dsl_pool_close() and zio_inject_fault()), so we must * skip any markers we see from these other threads. */ if (hdr->b_spa == 0) continue; /* we're only interested in evicting buffers of a certain spa */ if (spa != 0 && hdr->b_spa != spa) { ARCSTAT_BUMP(arcstat_evict_skip); continue; } hash_lock = HDR_LOCK(hdr); /* * We aren't calling this function from any code path * that would already be holding a hash lock, so we're * asserting on this assumption to be defensive in case * this ever changes. Without this check, it would be * possible to incorrectly increment arcstat_mutex_miss * below (e.g. if the code changed such that we called * this function with a hash lock held). */ ASSERT(!MUTEX_HELD(hash_lock)); if (mutex_tryenter(hash_lock)) { uint64_t evicted = arc_evict_hdr(hdr, hash_lock); mutex_exit(hash_lock); bytes_evicted += evicted; /* * If evicted is zero, arc_evict_hdr() must have * decided to skip this header, don't increment * evict_count in this case. */ if (evicted != 0) evict_count++; /* * If arc_size isn't overflowing, signal any * threads that might happen to be waiting. * * For each header evicted, we wake up a single * thread. If we used cv_broadcast, we could * wake up "too many" threads causing arc_size * to significantly overflow arc_c; since * arc_get_data_impl() doesn't check for overflow * when it's woken up (it doesn't because it's * possible for the ARC to be overflowing while * full of un-evictable buffers, and the * function should proceed in this case). * * If threads are left sleeping, due to not * using cv_broadcast, they will be woken up * just before arc_reclaim_thread() sleeps. */ mutex_enter(&arc_reclaim_lock); if (!arc_is_overflowing()) cv_signal(&arc_reclaim_waiters_cv); mutex_exit(&arc_reclaim_lock); } else { ARCSTAT_BUMP(arcstat_mutex_miss); } } multilist_sublist_unlock(mls); return (bytes_evicted); } /* * Evict buffers from the given arc state, until we've removed the * specified number of bytes. Move the removed buffers to the * appropriate evict state. * * This function makes a "best effort". It skips over any buffers * it can't get a hash_lock on, and so, may not catch all candidates. * It may also return without evicting as much space as requested. * * If bytes is specified using the special value ARC_EVICT_ALL, this * will evict all available (i.e. unlocked and evictable) buffers from * the given arc state; which is used by arc_flush(). */ static uint64_t arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, arc_buf_contents_t type) { uint64_t total_evicted = 0; multilist_t *ml = state->arcs_list[type]; int num_sublists; arc_buf_hdr_t **markers; IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); num_sublists = multilist_get_num_sublists(ml); /* * If we've tried to evict from each sublist, made some * progress, but still have not hit the target number of bytes * to evict, we want to keep trying. The markers allow us to * pick up where we left off for each individual sublist, rather * than starting from the tail each time. */ markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); for (int i = 0; i < num_sublists; i++) { markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); /* * A b_spa of 0 is used to indicate that this header is * a marker. This fact is used in arc_adjust_type() and * arc_evict_state_impl(). */ markers[i]->b_spa = 0; multilist_sublist_t *mls = multilist_sublist_lock(ml, i); multilist_sublist_insert_tail(mls, markers[i]); multilist_sublist_unlock(mls); } /* * While we haven't hit our target number of bytes to evict, or * we're evicting all available buffers. */ while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { /* * Start eviction using a randomly selected sublist, * this is to try and evenly balance eviction across all * sublists. Always starting at the same sublist * (e.g. index 0) would cause evictions to favor certain * sublists over others. */ int sublist_idx = multilist_get_random_index(ml); uint64_t scan_evicted = 0; for (int i = 0; i < num_sublists; i++) { uint64_t bytes_remaining; uint64_t bytes_evicted; if (bytes == ARC_EVICT_ALL) bytes_remaining = ARC_EVICT_ALL; else if (total_evicted < bytes) bytes_remaining = bytes - total_evicted; else break; bytes_evicted = arc_evict_state_impl(ml, sublist_idx, markers[sublist_idx], spa, bytes_remaining); scan_evicted += bytes_evicted; total_evicted += bytes_evicted; /* we've reached the end, wrap to the beginning */ if (++sublist_idx >= num_sublists) sublist_idx = 0; } /* * If we didn't evict anything during this scan, we have * no reason to believe we'll evict more during another * scan, so break the loop. */ if (scan_evicted == 0) { /* This isn't possible, let's make that obvious */ ASSERT3S(bytes, !=, 0); /* * When bytes is ARC_EVICT_ALL, the only way to * break the loop is when scan_evicted is zero. * In that case, we actually have evicted enough, * so we don't want to increment the kstat. */ if (bytes != ARC_EVICT_ALL) { ASSERT3S(total_evicted, <, bytes); ARCSTAT_BUMP(arcstat_evict_not_enough); } break; } } for (int i = 0; i < num_sublists; i++) { multilist_sublist_t *mls = multilist_sublist_lock(ml, i); multilist_sublist_remove(mls, markers[i]); multilist_sublist_unlock(mls); kmem_cache_free(hdr_full_cache, markers[i]); } kmem_free(markers, sizeof (*markers) * num_sublists); return (total_evicted); } /* * Flush all "evictable" data of the given type from the arc state * specified. This will not evict any "active" buffers (i.e. referenced). * * When 'retry' is set to B_FALSE, the function will make a single pass * over the state and evict any buffers that it can. Since it doesn't * continually retry the eviction, it might end up leaving some buffers * in the ARC due to lock misses. * * When 'retry' is set to B_TRUE, the function will continually retry the * eviction until *all* evictable buffers have been removed from the * state. As a result, if concurrent insertions into the state are * allowed (e.g. if the ARC isn't shutting down), this function might * wind up in an infinite loop, continually trying to evict buffers. */ static uint64_t arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, boolean_t retry) { uint64_t evicted = 0; while (refcount_count(&state->arcs_esize[type]) != 0) { evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); if (!retry) break; } return (evicted); } /* * Evict the specified number of bytes from the state specified, * restricting eviction to the spa and type given. This function * prevents us from trying to evict more from a state's list than * is "evictable", and to skip evicting altogether when passed a * negative value for "bytes". In contrast, arc_evict_state() will * evict everything it can, when passed a negative value for "bytes". */ static uint64_t arc_adjust_impl(arc_state_t *state, uint64_t spa, int64_t bytes, arc_buf_contents_t type) { int64_t delta; if (bytes > 0 && refcount_count(&state->arcs_esize[type]) > 0) { delta = MIN(refcount_count(&state->arcs_esize[type]), bytes); return (arc_evict_state(state, spa, delta, type)); } return (0); } /* * Evict metadata buffers from the cache, such that arc_meta_used is * capped by the arc_meta_limit tunable. */ static uint64_t arc_adjust_meta(void) { uint64_t total_evicted = 0; int64_t target; /* * If we're over the meta limit, we want to evict enough * metadata to get back under the meta limit. We don't want to * evict so much that we drop the MRU below arc_p, though. If * we're over the meta limit more than we're over arc_p, we * evict some from the MRU here, and some from the MFU below. */ target = MIN((int64_t)(arc_meta_used - arc_meta_limit), (int64_t)(refcount_count(&arc_anon->arcs_size) + refcount_count(&arc_mru->arcs_size) - arc_p)); total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); /* * Similar to the above, we want to evict enough bytes to get us * below the meta limit, but not so much as to drop us below the * space allotted to the MFU (which is defined as arc_c - arc_p). */ target = MIN((int64_t)(arc_meta_used - arc_meta_limit), (int64_t)(refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); return (total_evicted); } /* * Return the type of the oldest buffer in the given arc state * * This function will select a random sublist of type ARC_BUFC_DATA and * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist * is compared, and the type which contains the "older" buffer will be * returned. */ static arc_buf_contents_t arc_adjust_type(arc_state_t *state) { multilist_t *data_ml = state->arcs_list[ARC_BUFC_DATA]; multilist_t *meta_ml = state->arcs_list[ARC_BUFC_METADATA]; int data_idx = multilist_get_random_index(data_ml); int meta_idx = multilist_get_random_index(meta_ml); multilist_sublist_t *data_mls; multilist_sublist_t *meta_mls; arc_buf_contents_t type; arc_buf_hdr_t *data_hdr; arc_buf_hdr_t *meta_hdr; /* * We keep the sublist lock until we're finished, to prevent * the headers from being destroyed via arc_evict_state(). */ data_mls = multilist_sublist_lock(data_ml, data_idx); meta_mls = multilist_sublist_lock(meta_ml, meta_idx); /* * These two loops are to ensure we skip any markers that * might be at the tail of the lists due to arc_evict_state(). */ for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { if (data_hdr->b_spa != 0) break; } for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { if (meta_hdr->b_spa != 0) break; } if (data_hdr == NULL && meta_hdr == NULL) { type = ARC_BUFC_DATA; } else if (data_hdr == NULL) { ASSERT3P(meta_hdr, !=, NULL); type = ARC_BUFC_METADATA; } else if (meta_hdr == NULL) { ASSERT3P(data_hdr, !=, NULL); type = ARC_BUFC_DATA; } else { ASSERT3P(data_hdr, !=, NULL); ASSERT3P(meta_hdr, !=, NULL); /* The headers can't be on the sublist without an L1 header */ ASSERT(HDR_HAS_L1HDR(data_hdr)); ASSERT(HDR_HAS_L1HDR(meta_hdr)); if (data_hdr->b_l1hdr.b_arc_access < meta_hdr->b_l1hdr.b_arc_access) { type = ARC_BUFC_DATA; } else { type = ARC_BUFC_METADATA; } } multilist_sublist_unlock(meta_mls); multilist_sublist_unlock(data_mls); return (type); } /* * Evict buffers from the cache, such that arc_size is capped by arc_c. */ static uint64_t arc_adjust(void) { uint64_t total_evicted = 0; uint64_t bytes; int64_t target; /* * If we're over arc_meta_limit, we want to correct that before * potentially evicting data buffers below. */ total_evicted += arc_adjust_meta(); /* * Adjust MRU size * * If we're over the target cache size, we want to evict enough * from the list to get back to our target size. We don't want * to evict too much from the MRU, such that it drops below * arc_p. So, if we're over our target cache size more than * the MRU is over arc_p, we'll evict enough to get back to * arc_p here, and then evict more from the MFU below. */ target = MIN((int64_t)(arc_size - arc_c), (int64_t)(refcount_count(&arc_anon->arcs_size) + refcount_count(&arc_mru->arcs_size) + arc_meta_used - arc_p)); /* * If we're below arc_meta_min, always prefer to evict data. * Otherwise, try to satisfy the requested number of bytes to * evict from the type which contains older buffers; in an * effort to keep newer buffers in the cache regardless of their * type. If we cannot satisfy the number of bytes from this * type, spill over into the next type. */ if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && arc_meta_used > arc_meta_min) { bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * metadata, we try to get the rest from data. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); } else { bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * data, we try to get the rest from metadata. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); } /* * Adjust MFU size * * Now that we've tried to evict enough from the MRU to get its * size back to arc_p, if we're still above the target cache * size, we evict the rest from the MFU. */ target = arc_size - arc_c; if (arc_adjust_type(arc_mfu) == ARC_BUFC_METADATA && arc_meta_used > arc_meta_min) { bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * metadata, we try to get the rest from data. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); } else { bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * data, we try to get the rest from data. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); } /* * Adjust ghost lists * * In addition to the above, the ARC also defines target values * for the ghost lists. The sum of the mru list and mru ghost * list should never exceed the target size of the cache, and * the sum of the mru list, mfu list, mru ghost list, and mfu * ghost list should never exceed twice the target size of the * cache. The following logic enforces these limits on the ghost * caches, and evicts from them as needed. */ target = refcount_count(&arc_mru->arcs_size) + refcount_count(&arc_mru_ghost->arcs_size) - arc_c; bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); total_evicted += bytes; target -= bytes; total_evicted += arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); /* * We assume the sum of the mru list and mfu list is less than * or equal to arc_c (we enforced this above), which means we * can use the simpler of the two equations below: * * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c * mru ghost + mfu ghost <= arc_c */ target = refcount_count(&arc_mru_ghost->arcs_size) + refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); total_evicted += bytes; target -= bytes; total_evicted += arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); return (total_evicted); } void arc_flush(spa_t *spa, boolean_t retry) { uint64_t guid = 0; /* * If retry is B_TRUE, a spa must not be specified since we have * no good way to determine if all of a spa's buffers have been * evicted from an arc state. */ ASSERT(!retry || spa == 0); if (spa != NULL) guid = spa_load_guid(spa); (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); } void arc_shrink(int64_t to_free) { if (arc_c > arc_c_min) { DTRACE_PROBE4(arc__shrink, uint64_t, arc_c, uint64_t, arc_c_min, uint64_t, arc_p, uint64_t, to_free); if (arc_c > arc_c_min + to_free) atomic_add_64(&arc_c, -to_free); else arc_c = arc_c_min; atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); if (arc_c > arc_size) arc_c = MAX(arc_size, arc_c_min); if (arc_p > arc_c) arc_p = (arc_c >> 1); DTRACE_PROBE2(arc__shrunk, uint64_t, arc_c, uint64_t, arc_p); ASSERT(arc_c >= arc_c_min); ASSERT((int64_t)arc_p >= 0); } if (arc_size > arc_c) { DTRACE_PROBE2(arc__shrink_adjust, uint64_t, arc_size, uint64_t, arc_c); (void) arc_adjust(); } } typedef enum free_memory_reason_t { FMR_UNKNOWN, FMR_NEEDFREE, FMR_LOTSFREE, FMR_SWAPFS_MINFREE, FMR_PAGES_PP_MAXIMUM, FMR_HEAP_ARENA, FMR_ZIO_ARENA, - FMR_ZIO_FRAG, } free_memory_reason_t; int64_t last_free_memory; free_memory_reason_t last_free_reason; /* * Additional reserve of pages for pp_reserve. */ int64_t arc_pages_pp_reserve = 64; /* * Additional reserve of pages for swapfs. */ int64_t arc_swapfs_reserve = 64; /* * Return the amount of memory that can be consumed before reclaim will be * needed. Positive if there is sufficient free memory, negative indicates * the amount of memory that needs to be freed up. */ static int64_t arc_available_memory(void) { int64_t lowest = INT64_MAX; int64_t n; free_memory_reason_t r = FMR_UNKNOWN; #ifdef _KERNEL /* * Cooperate with pagedaemon when it's time for it to scan * and reclaim some pages. */ n = PAGESIZE * ((int64_t)freemem - zfs_arc_free_target); if (n < lowest) { lowest = n; r = FMR_LOTSFREE; } #ifdef illumos /* * check that we're out of range of the pageout scanner. It starts to * schedule paging if freemem is less than lotsfree and needfree. * lotsfree is the high-water mark for pageout, and needfree is the * number of needed free pages. We add extra pages here to make sure * the scanner doesn't start up while we're freeing memory. */ n = PAGESIZE * (freemem - lotsfree - needfree - desfree); if (n < lowest) { lowest = n; r = FMR_LOTSFREE; } /* * check to make sure that swapfs has enough space so that anon * reservations can still succeed. anon_resvmem() checks that the * availrmem is greater than swapfs_minfree, and the number of reserved * swap pages. We also add a bit of extra here just to prevent * circumstances from getting really dire. */ n = PAGESIZE * (availrmem - swapfs_minfree - swapfs_reserve - desfree - arc_swapfs_reserve); if (n < lowest) { lowest = n; r = FMR_SWAPFS_MINFREE; } /* * Check that we have enough availrmem that memory locking (e.g., via * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum * stores the number of pages that cannot be locked; when availrmem * drops below pages_pp_maximum, page locking mechanisms such as * page_pp_lock() will fail.) */ n = PAGESIZE * (availrmem - pages_pp_maximum - arc_pages_pp_reserve); if (n < lowest) { lowest = n; r = FMR_PAGES_PP_MAXIMUM; } #endif /* illumos */ #if defined(__i386) || !defined(UMA_MD_SMALL_ALLOC) /* * If we're on an i386 platform, it's possible that we'll exhaust the * kernel heap space before we ever run out of available physical * memory. Most checks of the size of the heap_area compare against * tune.t_minarmem, which is the minimum available real memory that we * can have in the system. However, this is generally fixed at 25 pages * which is so low that it's useless. In this comparison, we seek to * calculate the total heap-size, and reclaim if more than 3/4ths of the * heap is allocated. (Or, in the calculation, if less than 1/4th is * free) */ - n = (int64_t)vmem_size(heap_arena, VMEM_FREE) - - (vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC) >> 2); + n = uma_avail() - (long)(uma_limit() / 4); if (n < lowest) { lowest = n; r = FMR_HEAP_ARENA; } -#define zio_arena NULL -#else -#define zio_arena heap_arena #endif /* * If zio data pages are being allocated out of a separate heap segment, * then enforce that the size of available vmem for this arena remains * above about 1/16th free. * * Note: The 1/16th arena free requirement was put in place * to aggressively evict memory from the arc in order to avoid * memory fragmentation issues. */ if (zio_arena != NULL) { n = (int64_t)vmem_size(zio_arena, VMEM_FREE) - (vmem_size(zio_arena, VMEM_ALLOC) >> 4); if (n < lowest) { lowest = n; r = FMR_ZIO_ARENA; } } - /* - * Above limits know nothing about real level of KVA fragmentation. - * Start aggressive reclamation if too little sequential KVA left. - */ - if (lowest > 0) { - n = (vmem_size(heap_arena, VMEM_MAXFREE) < SPA_MAXBLOCKSIZE) ? - -((int64_t)vmem_size(heap_arena, VMEM_ALLOC) >> 4) : - INT64_MAX; - if (n < lowest) { - lowest = n; - r = FMR_ZIO_FRAG; - } - } - #else /* _KERNEL */ /* Every 100 calls, free a small amount */ if (spa_get_random(100) == 0) lowest = -1024; #endif /* _KERNEL */ last_free_memory = lowest; last_free_reason = r; DTRACE_PROBE2(arc__available_memory, int64_t, lowest, int, r); return (lowest); } /* * Determine if the system is under memory pressure and is asking * to reclaim memory. A return value of B_TRUE indicates that the system * is under memory pressure and that the arc should adjust accordingly. */ static boolean_t arc_reclaim_needed(void) { return (arc_available_memory() < 0); } extern kmem_cache_t *zio_buf_cache[]; extern kmem_cache_t *zio_data_buf_cache[]; extern kmem_cache_t *range_seg_cache; extern kmem_cache_t *abd_chunk_cache; static __noinline void arc_kmem_reap_now(void) { size_t i; kmem_cache_t *prev_cache = NULL; kmem_cache_t *prev_data_cache = NULL; DTRACE_PROBE(arc__kmem_reap_start); #ifdef _KERNEL if (arc_meta_used >= arc_meta_limit) { /* * We are exceeding our meta-data cache limit. * Purge some DNLC entries to release holds on meta-data. */ dnlc_reduce_cache((void *)(uintptr_t)arc_reduce_dnlc_percent); } #if defined(__i386) /* * Reclaim unused memory from all kmem caches. */ kmem_reap(); #endif #endif for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { if (zio_buf_cache[i] != prev_cache) { prev_cache = zio_buf_cache[i]; kmem_cache_reap_now(zio_buf_cache[i]); } if (zio_data_buf_cache[i] != prev_data_cache) { prev_data_cache = zio_data_buf_cache[i]; kmem_cache_reap_now(zio_data_buf_cache[i]); } } kmem_cache_reap_now(abd_chunk_cache); kmem_cache_reap_now(buf_cache); kmem_cache_reap_now(hdr_full_cache); kmem_cache_reap_now(hdr_l2only_cache); kmem_cache_reap_now(range_seg_cache); #ifdef illumos if (zio_arena != NULL) { /* * Ask the vmem arena to reclaim unused memory from its * quantum caches. */ vmem_qcache_reap(zio_arena); } #endif DTRACE_PROBE(arc__kmem_reap_end); } /* * Threads can block in arc_get_data_impl() waiting for this thread to evict * enough data and signal them to proceed. When this happens, the threads in * arc_get_data_impl() are sleeping while holding the hash lock for their * particular arc header. Thus, we must be careful to never sleep on a * hash lock in this thread. This is to prevent the following deadlock: * * - Thread A sleeps on CV in arc_get_data_impl() holding hash lock "L", * waiting for the reclaim thread to signal it. * * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter, * fails, and goes to sleep forever. * * This possible deadlock is avoided by always acquiring a hash lock * using mutex_tryenter() from arc_reclaim_thread(). */ static void arc_reclaim_thread(void *dummy __unused) { hrtime_t growtime = 0; callb_cpr_t cpr; CALLB_CPR_INIT(&cpr, &arc_reclaim_lock, callb_generic_cpr, FTAG); mutex_enter(&arc_reclaim_lock); while (!arc_reclaim_thread_exit) { uint64_t evicted = 0; /* * This is necessary in order for the mdb ::arc dcmd to * show up to date information. Since the ::arc command * does not call the kstat's update function, without * this call, the command may show stale stats for the * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even * with this change, the data might be up to 1 second * out of date; but that should suffice. The arc_state_t * structures can be queried directly if more accurate * information is needed. */ if (arc_ksp != NULL) arc_ksp->ks_update(arc_ksp, KSTAT_READ); mutex_exit(&arc_reclaim_lock); /* * We call arc_adjust() before (possibly) calling * arc_kmem_reap_now(), so that we can wake up * arc_get_data_impl() sooner. */ evicted = arc_adjust(); int64_t free_memory = arc_available_memory(); if (free_memory < 0) { arc_no_grow = B_TRUE; arc_warm = B_TRUE; /* * Wait at least zfs_grow_retry (default 60) seconds * before considering growing. */ growtime = gethrtime() + SEC2NSEC(arc_grow_retry); arc_kmem_reap_now(); /* * If we are still low on memory, shrink the ARC * so that we have arc_shrink_min free space. */ free_memory = arc_available_memory(); int64_t to_free = (arc_c >> arc_shrink_shift) - free_memory; if (to_free > 0) { arc_shrink(to_free); } } else if (free_memory < arc_c >> arc_no_grow_shift) { arc_no_grow = B_TRUE; } else if (gethrtime() >= growtime) { arc_no_grow = B_FALSE; } mutex_enter(&arc_reclaim_lock); /* * If evicted is zero, we couldn't evict anything via * arc_adjust(). This could be due to hash lock * collisions, but more likely due to the majority of * arc buffers being unevictable. Therefore, even if * arc_size is above arc_c, another pass is unlikely to * be helpful and could potentially cause us to enter an * infinite loop. */ if (arc_size <= arc_c || evicted == 0) { /* * We're either no longer overflowing, or we * can't evict anything more, so we should wake * up any threads before we go to sleep. */ cv_broadcast(&arc_reclaim_waiters_cv); /* * Block until signaled, or after one second (we * might need to perform arc_kmem_reap_now() * even if we aren't being signalled) */ CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait_hires(&arc_reclaim_thread_cv, &arc_reclaim_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock); } } arc_reclaim_thread_exit = B_FALSE; cv_broadcast(&arc_reclaim_thread_cv); CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ thread_exit(); } static u_int arc_dnlc_evicts_arg; extern struct vfsops zfs_vfsops; static void arc_dnlc_evicts_thread(void *dummy __unused) { callb_cpr_t cpr; u_int percent; CALLB_CPR_INIT(&cpr, &arc_dnlc_evicts_lock, callb_generic_cpr, FTAG); mutex_enter(&arc_dnlc_evicts_lock); while (!arc_dnlc_evicts_thread_exit) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_wait(&arc_dnlc_evicts_cv, &arc_dnlc_evicts_lock); CALLB_CPR_SAFE_END(&cpr, &arc_dnlc_evicts_lock); if (arc_dnlc_evicts_arg != 0) { percent = arc_dnlc_evicts_arg; mutex_exit(&arc_dnlc_evicts_lock); #ifdef _KERNEL vnlru_free(desiredvnodes * percent / 100, &zfs_vfsops); #endif mutex_enter(&arc_dnlc_evicts_lock); /* * Clear our token only after vnlru_free() * pass is done, to avoid false queueing of * the requests. */ arc_dnlc_evicts_arg = 0; } } arc_dnlc_evicts_thread_exit = FALSE; cv_broadcast(&arc_dnlc_evicts_cv); CALLB_CPR_EXIT(&cpr); thread_exit(); } void dnlc_reduce_cache(void *arg) { u_int percent; percent = (u_int)(uintptr_t)arg; mutex_enter(&arc_dnlc_evicts_lock); if (arc_dnlc_evicts_arg == 0) { arc_dnlc_evicts_arg = percent; cv_broadcast(&arc_dnlc_evicts_cv); } mutex_exit(&arc_dnlc_evicts_lock); } /* * Adapt arc info given the number of bytes we are trying to add and * the state that we are comming from. This function is only called * when we are adding new content to the cache. */ static void arc_adapt(int bytes, arc_state_t *state) { int mult; uint64_t arc_p_min = (arc_c >> arc_p_min_shift); int64_t mrug_size = refcount_count(&arc_mru_ghost->arcs_size); int64_t mfug_size = refcount_count(&arc_mfu_ghost->arcs_size); if (state == arc_l2c_only) return; ASSERT(bytes > 0); /* * Adapt the target size of the MRU list: * - if we just hit in the MRU ghost list, then increase * the target size of the MRU list. * - if we just hit in the MFU ghost list, then increase * the target size of the MFU list by decreasing the * target size of the MRU list. */ if (state == arc_mru_ghost) { mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); } else if (state == arc_mfu_ghost) { uint64_t delta; mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); mult = MIN(mult, 10); delta = MIN(bytes * mult, arc_p); arc_p = MAX(arc_p_min, arc_p - delta); } ASSERT((int64_t)arc_p >= 0); if (arc_reclaim_needed()) { cv_signal(&arc_reclaim_thread_cv); return; } if (arc_no_grow) return; if (arc_c >= arc_c_max) return; /* * If we're within (2 * maxblocksize) bytes of the target * cache size, increment the target cache size */ if (arc_size > arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { DTRACE_PROBE1(arc__inc_adapt, int, bytes); atomic_add_64(&arc_c, (int64_t)bytes); if (arc_c > arc_c_max) arc_c = arc_c_max; else if (state == arc_anon) atomic_add_64(&arc_p, (int64_t)bytes); if (arc_p > arc_c) arc_p = arc_c; } ASSERT((int64_t)arc_p >= 0); } /* * Check if arc_size has grown past our upper threshold, determined by * zfs_arc_overflow_shift. */ static boolean_t arc_is_overflowing(void) { /* Always allow at least one block of overflow */ uint64_t overflow = MAX(SPA_MAXBLOCKSIZE, arc_c >> zfs_arc_overflow_shift); return (arc_size >= arc_c + overflow); } static abd_t * arc_get_data_abd(arc_buf_hdr_t *hdr, uint64_t size, void *tag) { arc_buf_contents_t type = arc_buf_type(hdr); arc_get_data_impl(hdr, size, tag); if (type == ARC_BUFC_METADATA) { return (abd_alloc(size, B_TRUE)); } else { ASSERT(type == ARC_BUFC_DATA); return (abd_alloc(size, B_FALSE)); } } static void * arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, void *tag) { arc_buf_contents_t type = arc_buf_type(hdr); arc_get_data_impl(hdr, size, tag); if (type == ARC_BUFC_METADATA) { return (zio_buf_alloc(size)); } else { ASSERT(type == ARC_BUFC_DATA); return (zio_data_buf_alloc(size)); } } /* * Allocate a block and return it to the caller. If we are hitting the * hard limit for the cache size, we must sleep, waiting for the eviction * thread to catch up. If we're past the target size but below the hard * limit, we'll only signal the reclaim thread and continue on. */ static void arc_get_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); arc_adapt(size, state); /* * If arc_size is currently overflowing, and has grown past our * upper limit, we must be adding data faster than the evict * thread can evict. Thus, to ensure we don't compound the * problem by adding more data and forcing arc_size to grow even * further past it's target size, we halt and wait for the * eviction thread to catch up. * * It's also possible that the reclaim thread is unable to evict * enough buffers to get arc_size below the overflow limit (e.g. * due to buffers being un-evictable, or hash lock collisions). * In this case, we want to proceed regardless if we're * overflowing; thus we don't use a while loop here. */ if (arc_is_overflowing()) { mutex_enter(&arc_reclaim_lock); /* * Now that we've acquired the lock, we may no longer be * over the overflow limit, lets check. * * We're ignoring the case of spurious wake ups. If that * were to happen, it'd let this thread consume an ARC * buffer before it should have (i.e. before we're under * the overflow limit and were signalled by the reclaim * thread). As long as that is a rare occurrence, it * shouldn't cause any harm. */ if (arc_is_overflowing()) { cv_signal(&arc_reclaim_thread_cv); cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); } mutex_exit(&arc_reclaim_lock); } VERIFY3U(hdr->b_type, ==, type); if (type == ARC_BUFC_METADATA) { arc_space_consume(size, ARC_SPACE_META); } else { arc_space_consume(size, ARC_SPACE_DATA); } /* * Update the state size. Note that ghost states have a * "ghost size" and so don't need to be updated. */ if (!GHOST_STATE(state)) { (void) refcount_add_many(&state->arcs_size, size, tag); /* * If this is reached via arc_read, the link is * protected by the hash lock. If reached via * arc_buf_alloc, the header should not be accessed by * any other thread. And, if reached via arc_read_done, * the hash lock will protect it if it's found in the * hash table; otherwise no other thread should be * trying to [add|remove]_reference it. */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); (void) refcount_add_many(&state->arcs_esize[type], size, tag); } /* * If we are growing the cache, and we are adding anonymous * data, and we have outgrown arc_p, update arc_p */ if (arc_size < arc_c && hdr->b_l1hdr.b_state == arc_anon && (refcount_count(&arc_anon->arcs_size) + refcount_count(&arc_mru->arcs_size) > arc_p)) arc_p = MIN(arc_c, arc_p + size); } ARCSTAT_BUMP(arcstat_allocated); } static void arc_free_data_abd(arc_buf_hdr_t *hdr, abd_t *abd, uint64_t size, void *tag) { arc_free_data_impl(hdr, size, tag); abd_free(abd); } static void arc_free_data_buf(arc_buf_hdr_t *hdr, void *buf, uint64_t size, void *tag) { arc_buf_contents_t type = arc_buf_type(hdr); arc_free_data_impl(hdr, size, tag); if (type == ARC_BUFC_METADATA) { zio_buf_free(buf, size); } else { ASSERT(type == ARC_BUFC_DATA); zio_data_buf_free(buf, size); } } /* * Free the arc data buffer. */ static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); /* protected by hash lock, if in the hash table */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT(state != arc_anon && state != arc_l2c_only); (void) refcount_remove_many(&state->arcs_esize[type], size, tag); } (void) refcount_remove_many(&state->arcs_size, size, tag); VERIFY3U(hdr->b_type, ==, type); if (type == ARC_BUFC_METADATA) { arc_space_return(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); arc_space_return(size, ARC_SPACE_DATA); } } /* * This routine is called whenever a buffer is accessed. * NOTE: the hash lock is dropped in this function. */ static void arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { clock_t now; ASSERT(MUTEX_HELD(hash_lock)); ASSERT(HDR_HAS_L1HDR(hdr)); if (hdr->b_l1hdr.b_state == arc_anon) { /* * This buffer is not in the cache, and does not * appear in our "ghost" list. Add the new buffer * to the MRU state. */ ASSERT0(hdr->b_l1hdr.b_arc_access); hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); arc_change_state(arc_mru, hdr, hash_lock); } else if (hdr->b_l1hdr.b_state == arc_mru) { now = ddi_get_lbolt(); /* * If this buffer is here because of a prefetch, then either: * - clear the flag if this is a "referencing" read * (any subsequent access will bump this into the MFU state). * or * - move the buffer to the head of the list if this is * another prefetch (to make it less likely to be evicted). */ if (HDR_PREFETCH(hdr)) { if (refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { /* link protected by hash lock */ ASSERT(multilist_link_active( &hdr->b_l1hdr.b_arc_node)); } else { arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); ARCSTAT_BUMP(arcstat_mru_hits); } hdr->b_l1hdr.b_arc_access = now; return; } /* * This buffer has been "accessed" only once so far, * but it is still in the cache. Move it to the MFU * state. */ if (now > hdr->b_l1hdr.b_arc_access + ARC_MINTIME) { /* * More than 125ms have passed since we * instantiated this buffer. Move it to the * most frequently used state. */ hdr->b_l1hdr.b_arc_access = now; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(arc_mfu, hdr, hash_lock); } ARCSTAT_BUMP(arcstat_mru_hits); } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { arc_state_t *new_state; /* * This buffer has been "accessed" recently, but * was evicted from the cache. Move it to the * MFU state. */ if (HDR_PREFETCH(hdr)) { new_state = arc_mru; if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); } else { new_state = arc_mfu; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); } hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); arc_change_state(new_state, hdr, hash_lock); ARCSTAT_BUMP(arcstat_mru_ghost_hits); } else if (hdr->b_l1hdr.b_state == arc_mfu) { /* * This buffer has been accessed more than once and is * still in the cache. Keep it in the MFU state. * * NOTE: an add_reference() that occurred when we did * the arc_read() will have kicked this off the list. * If it was a prefetch, we will explicitly move it to * the head of the list now. */ if ((HDR_PREFETCH(hdr)) != 0) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); /* link protected by hash_lock */ ASSERT(multilist_link_active(&hdr->b_l1hdr.b_arc_node)); } ARCSTAT_BUMP(arcstat_mfu_hits); hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { arc_state_t *new_state = arc_mfu; /* * This buffer has been accessed more than once but has * been evicted from the cache. Move it back to the * MFU state. */ if (HDR_PREFETCH(hdr)) { /* * This is a prefetch access... * move this block back to the MRU state. */ ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); new_state = arc_mru; } hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(new_state, hdr, hash_lock); ARCSTAT_BUMP(arcstat_mfu_ghost_hits); } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { /* * This buffer is on the 2nd Level ARC. */ hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(arc_mfu, hdr, hash_lock); } else { ASSERT(!"invalid arc state"); } } /* a generic arc_done_func_t which you can use */ /* ARGSUSED */ void arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) { if (zio == NULL || zio->io_error == 0) bcopy(buf->b_data, arg, arc_buf_size(buf)); arc_buf_destroy(buf, arg); } /* a generic arc_done_func_t */ void arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) { arc_buf_t **bufp = arg; if (zio && zio->io_error) { arc_buf_destroy(buf, arg); *bufp = NULL; } else { *bufp = buf; ASSERT(buf->b_data); } } static void arc_hdr_verify(arc_buf_hdr_t *hdr, blkptr_t *bp) { if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) { ASSERT3U(HDR_GET_PSIZE(hdr), ==, 0); ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); } else { if (HDR_COMPRESSION_ENABLED(hdr)) { ASSERT3U(HDR_GET_COMPRESS(hdr), ==, BP_GET_COMPRESS(bp)); } ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp)); ASSERT3U(HDR_GET_PSIZE(hdr), ==, BP_GET_PSIZE(bp)); } } static void arc_read_done(zio_t *zio) { arc_buf_hdr_t *hdr = zio->io_private; kmutex_t *hash_lock = NULL; arc_callback_t *callback_list; arc_callback_t *acb; boolean_t freeable = B_FALSE; boolean_t no_zio_error = (zio->io_error == 0); /* * The hdr was inserted into hash-table and removed from lists * prior to starting I/O. We should find this header, since * it's in the hash table, and it should be legit since it's * not possible to evict it during the I/O. The only possible * reason for it not to be found is if we were freed during the * read. */ if (HDR_IN_HASH_TABLE(hdr)) { ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); ASSERT3U(hdr->b_dva.dva_word[0], ==, BP_IDENTITY(zio->io_bp)->dva_word[0]); ASSERT3U(hdr->b_dva.dva_word[1], ==, BP_IDENTITY(zio->io_bp)->dva_word[1]); arc_buf_hdr_t *found = buf_hash_find(hdr->b_spa, zio->io_bp, &hash_lock); ASSERT((found == hdr && DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || (found == hdr && HDR_L2_READING(hdr))); ASSERT3P(hash_lock, !=, NULL); } if (no_zio_error) { /* byteswap if necessary */ if (BP_SHOULD_BYTESWAP(zio->io_bp)) { if (BP_GET_LEVEL(zio->io_bp) > 0) { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64; } else { hdr->b_l1hdr.b_byteswap = DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); } } else { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; } } arc_hdr_clear_flags(hdr, ARC_FLAG_L2_EVICTED); if (l2arc_noprefetch && HDR_PREFETCH(hdr)) arc_hdr_clear_flags(hdr, ARC_FLAG_L2CACHE); callback_list = hdr->b_l1hdr.b_acb; ASSERT3P(callback_list, !=, NULL); if (hash_lock && no_zio_error && hdr->b_l1hdr.b_state == arc_anon) { /* * Only call arc_access on anonymous buffers. This is because * if we've issued an I/O for an evicted buffer, we've already * called arc_access (to prevent any simultaneous readers from * getting confused). */ arc_access(hdr, hash_lock); } /* * If a read request has a callback (i.e. acb_done is not NULL), then we * make a buf containing the data according to the parameters which were * passed in. The implementation of arc_buf_alloc_impl() ensures that we * aren't needlessly decompressing the data multiple times. */ int callback_cnt = 0; for (acb = callback_list; acb != NULL; acb = acb->acb_next) { if (!acb->acb_done) continue; /* This is a demand read since prefetches don't use callbacks */ callback_cnt++; int error = arc_buf_alloc_impl(hdr, acb->acb_private, acb->acb_compressed, no_zio_error, &acb->acb_buf); if (no_zio_error) { zio->io_error = error; } } hdr->b_l1hdr.b_acb = NULL; arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); if (callback_cnt == 0) { ASSERT(HDR_PREFETCH(hdr)); ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); } ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || callback_list != NULL); if (no_zio_error) { arc_hdr_verify(hdr, zio->io_bp); } else { arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR); if (hdr->b_l1hdr.b_state != arc_anon) arc_change_state(arc_anon, hdr, hash_lock); if (HDR_IN_HASH_TABLE(hdr)) buf_hash_remove(hdr); freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); } /* * Broadcast before we drop the hash_lock to avoid the possibility * that the hdr (and hence the cv) might be freed before we get to * the cv_broadcast(). */ cv_broadcast(&hdr->b_l1hdr.b_cv); if (hash_lock != NULL) { mutex_exit(hash_lock); } else { /* * This block was freed while we waited for the read to * complete. It has been removed from the hash table and * moved to the anonymous state (so that it won't show up * in the cache). */ ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); } /* execute each callback and free its structure */ while ((acb = callback_list) != NULL) { if (acb->acb_done) acb->acb_done(zio, acb->acb_buf, acb->acb_private); if (acb->acb_zio_dummy != NULL) { acb->acb_zio_dummy->io_error = zio->io_error; zio_nowait(acb->acb_zio_dummy); } callback_list = acb->acb_next; kmem_free(acb, sizeof (arc_callback_t)); } if (freeable) arc_hdr_destroy(hdr); } /* * "Read" the block at the specified DVA (in bp) via the * cache. If the block is found in the cache, invoke the provided * callback immediately and return. Note that the `zio' parameter * in the callback will be NULL in this case, since no IO was * required. If the block is not in the cache pass the read request * on to the spa with a substitute callback function, so that the * requested block will be added to the cache. * * If a read request arrives for a block that has a read in-progress, * either wait for the in-progress read to complete (and return the * results); or, if this is a read with a "done" func, add a record * to the read to invoke the "done" func when the read completes, * and return; or just return. * * arc_read_done() will invoke all the requested "done" functions * for readers of this block. */ int arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, void *private, zio_priority_t priority, int zio_flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb) { arc_buf_hdr_t *hdr = NULL; kmutex_t *hash_lock = NULL; zio_t *rzio; uint64_t guid = spa_load_guid(spa); boolean_t compressed_read = (zio_flags & ZIO_FLAG_RAW) != 0; ASSERT(!BP_IS_EMBEDDED(bp) || BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); top: if (!BP_IS_EMBEDDED(bp)) { /* * Embedded BP's have no DVA and require no I/O to "read". * Create an anonymous arc buf to back it. */ hdr = buf_hash_find(guid, bp, &hash_lock); } if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_pabd != NULL) { arc_buf_t *buf = NULL; *arc_flags |= ARC_FLAG_CACHED; if (HDR_IO_IN_PROGRESS(hdr)) { if ((hdr->b_flags & ARC_FLAG_PRIO_ASYNC_READ) && priority == ZIO_PRIORITY_SYNC_READ) { /* * This sync read must wait for an * in-progress async read (e.g. a predictive * prefetch). Async reads are queued * separately at the vdev_queue layer, so * this is a form of priority inversion. * Ideally, we would "inherit" the demand * i/o's priority by moving the i/o from * the async queue to the synchronous queue, * but there is currently no mechanism to do * so. Track this so that we can evaluate * the magnitude of this potential performance * problem. * * Note that if the prefetch i/o is already * active (has been issued to the device), * the prefetch improved performance, because * we issued it sooner than we would have * without the prefetch. */ DTRACE_PROBE1(arc__sync__wait__for__async, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_sync_wait_for_async); } if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { arc_hdr_clear_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); } if (*arc_flags & ARC_FLAG_WAIT) { cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); mutex_exit(hash_lock); goto top; } ASSERT(*arc_flags & ARC_FLAG_NOWAIT); if (done) { arc_callback_t *acb = NULL; acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); acb->acb_done = done; acb->acb_private = private; acb->acb_compressed = compressed_read; if (pio != NULL) acb->acb_zio_dummy = zio_null(pio, spa, NULL, NULL, NULL, zio_flags); ASSERT3P(acb->acb_done, !=, NULL); acb->acb_next = hdr->b_l1hdr.b_acb; hdr->b_l1hdr.b_acb = acb; mutex_exit(hash_lock); return (0); } mutex_exit(hash_lock); return (0); } ASSERT(hdr->b_l1hdr.b_state == arc_mru || hdr->b_l1hdr.b_state == arc_mfu); if (done) { if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { /* * This is a demand read which does not have to * wait for i/o because we did a predictive * prefetch i/o for it, which has completed. */ DTRACE_PROBE1( arc__demand__hit__predictive__prefetch, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP( arcstat_demand_hit_predictive_prefetch); arc_hdr_clear_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); } ASSERT(!BP_IS_EMBEDDED(bp) || !BP_IS_HOLE(bp)); /* Get a buf with the desired data in it. */ VERIFY0(arc_buf_alloc_impl(hdr, private, compressed_read, B_TRUE, &buf)); } else if (*arc_flags & ARC_FLAG_PREFETCH && refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); } DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); arc_access(hdr, hash_lock); if (*arc_flags & ARC_FLAG_L2CACHE) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_hits); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, hits); if (done) done(NULL, buf, private); } else { uint64_t lsize = BP_GET_LSIZE(bp); uint64_t psize = BP_GET_PSIZE(bp); arc_callback_t *acb; vdev_t *vd = NULL; uint64_t addr = 0; boolean_t devw = B_FALSE; uint64_t size; if (hdr == NULL) { /* this block is not in the cache */ arc_buf_hdr_t *exists = NULL; arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, BP_GET_COMPRESS(bp), type); if (!BP_IS_EMBEDDED(bp)) { hdr->b_dva = *BP_IDENTITY(bp); hdr->b_birth = BP_PHYSICAL_BIRTH(bp); exists = buf_hash_insert(hdr, &hash_lock); } if (exists != NULL) { /* somebody beat us to the hash insert */ mutex_exit(hash_lock); buf_discard_identity(hdr); arc_hdr_destroy(hdr); goto top; /* restart the IO request */ } } else { /* * This block is in the ghost cache. If it was L2-only * (and thus didn't have an L1 hdr), we realloc the * header to add an L1 hdr. */ if (!HDR_HAS_L1HDR(hdr)) { hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, hdr_full_cache); } ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(GHOST_STATE(hdr->b_l1hdr.b_state)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); /* * This is a delicate dance that we play here. * This hdr is in the ghost list so we access it * to move it out of the ghost list before we * initiate the read. If it's a prefetch then * it won't have a callback so we'll remove the * reference that arc_buf_alloc_impl() created. We * do this after we've called arc_access() to * avoid hitting an assert in remove_reference(). */ arc_access(hdr, hash_lock); arc_hdr_alloc_pabd(hdr); } ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); size = arc_hdr_size(hdr); /* * If compression is enabled on the hdr, then will do * RAW I/O and will store the compressed data in the hdr's * data block. Otherwise, the hdr's data block will contain * the uncompressed data. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) { zio_flags |= ZIO_FLAG_RAW; } if (*arc_flags & ARC_FLAG_PREFETCH) arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); if (*arc_flags & ARC_FLAG_L2CACHE) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); if (BP_GET_LEVEL(bp) > 0) arc_hdr_set_flags(hdr, ARC_FLAG_INDIRECT); if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH) arc_hdr_set_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); acb->acb_done = done; acb->acb_private = private; acb->acb_compressed = compressed_read; ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); hdr->b_l1hdr.b_acb = acb; arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); if (HDR_HAS_L2HDR(hdr) && (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { devw = hdr->b_l2hdr.b_dev->l2ad_writing; addr = hdr->b_l2hdr.b_daddr; /* * Lock out device removal. */ if (vdev_is_dead(vd) || !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) vd = NULL; } if (priority == ZIO_PRIORITY_ASYNC_READ) arc_hdr_set_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); else arc_hdr_clear_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); if (hash_lock != NULL) mutex_exit(hash_lock); /* * At this point, we have a level 1 cache miss. Try again in * L2ARC if possible. */ ASSERT3U(HDR_GET_LSIZE(hdr), ==, lsize); DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, uint64_t, lsize, zbookmark_phys_t *, zb); ARCSTAT_BUMP(arcstat_misses); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, misses); #ifdef _KERNEL #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_force(curproc, RACCT_READBPS, size); racct_add_force(curproc, RACCT_READIOPS, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; #endif if (vd != NULL && l2arc_ndev != 0 && !(l2arc_norw && devw)) { /* * Read from the L2ARC if the following are true: * 1. The L2ARC vdev was previously cached. * 2. This buffer still has L2ARC metadata. * 3. This buffer isn't currently writing to the L2ARC. * 4. The L2ARC entry wasn't evicted, which may * also have invalidated the vdev. * 5. This isn't prefetch and l2arc_noprefetch is set. */ if (HDR_HAS_L2HDR(hdr) && !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { l2arc_read_callback_t *cb; abd_t *abd; uint64_t asize; DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_hits); cb = kmem_zalloc(sizeof (l2arc_read_callback_t), KM_SLEEP); cb->l2rcb_hdr = hdr; cb->l2rcb_bp = *bp; cb->l2rcb_zb = *zb; cb->l2rcb_flags = zio_flags; asize = vdev_psize_to_asize(vd, size); if (asize != size) { abd = abd_alloc_for_io(asize, HDR_ISTYPE_METADATA(hdr)); cb->l2rcb_abd = abd; } else { abd = hdr->b_l1hdr.b_pabd; } ASSERT(addr >= VDEV_LABEL_START_SIZE && addr + asize <= vd->vdev_psize - VDEV_LABEL_END_SIZE); /* * l2arc read. The SCL_L2ARC lock will be * released by l2arc_read_done(). * Issue a null zio if the underlying buffer * was squashed to zero size by compression. */ ASSERT3U(HDR_GET_COMPRESS(hdr), !=, ZIO_COMPRESS_EMPTY); rzio = zio_read_phys(pio, vd, addr, asize, abd, ZIO_CHECKSUM_OFF, l2arc_read_done, cb, priority, zio_flags | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY, B_FALSE); DTRACE_PROBE2(l2arc__read, vdev_t *, vd, zio_t *, rzio); ARCSTAT_INCR(arcstat_l2_read_bytes, size); if (*arc_flags & ARC_FLAG_NOWAIT) { zio_nowait(rzio); return (0); } ASSERT(*arc_flags & ARC_FLAG_WAIT); if (zio_wait(rzio) == 0) return (0); /* l2arc read error; goto zio_read() */ } else { DTRACE_PROBE1(l2arc__miss, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_misses); if (HDR_L2_WRITING(hdr)) ARCSTAT_BUMP(arcstat_l2_rw_clash); spa_config_exit(spa, SCL_L2ARC, vd); } } else { if (vd != NULL) spa_config_exit(spa, SCL_L2ARC, vd); if (l2arc_ndev != 0) { DTRACE_PROBE1(l2arc__miss, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_misses); } } rzio = zio_read(pio, spa, bp, hdr->b_l1hdr.b_pabd, size, arc_read_done, hdr, priority, zio_flags, zb); if (*arc_flags & ARC_FLAG_WAIT) return (zio_wait(rzio)); ASSERT(*arc_flags & ARC_FLAG_NOWAIT); zio_nowait(rzio); } return (0); } /* * Notify the arc that a block was freed, and thus will never be used again. */ void arc_freed(spa_t *spa, const blkptr_t *bp) { arc_buf_hdr_t *hdr; kmutex_t *hash_lock; uint64_t guid = spa_load_guid(spa); ASSERT(!BP_IS_EMBEDDED(bp)); hdr = buf_hash_find(guid, bp, &hash_lock); if (hdr == NULL) return; /* * We might be trying to free a block that is still doing I/O * (i.e. prefetch) or has a reference (i.e. a dedup-ed, * dmu_sync-ed block). If this block is being prefetched, then it * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr * until the I/O completes. A block may also have a reference if it is * part of a dedup-ed, dmu_synced write. The dmu_sync() function would * have written the new block to its final resting place on disk but * without the dedup flag set. This would have left the hdr in the MRU * state and discoverable. When the txg finally syncs it detects that * the block was overridden in open context and issues an override I/O. * Since this is a dedup block, the override I/O will determine if the * block is already in the DDT. If so, then it will replace the io_bp * with the bp from the DDT and allow the I/O to finish. When the I/O * reaches the done callback, dbuf_write_override_done, it will * check to see if the io_bp and io_bp_override are identical. * If they are not, then it indicates that the bp was replaced with * the bp in the DDT and the override bp is freed. This allows * us to arrive here with a reference on a block that is being * freed. So if we have an I/O in progress, or a reference to * this hdr, then we don't destroy the hdr. */ if (!HDR_HAS_L1HDR(hdr) || (!HDR_IO_IN_PROGRESS(hdr) && refcount_is_zero(&hdr->b_l1hdr.b_refcnt))) { arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); mutex_exit(hash_lock); } else { mutex_exit(hash_lock); } } /* * Release this buffer from the cache, making it an anonymous buffer. This * must be done after a read and prior to modifying the buffer contents. * If the buffer has more than one reference, we must make * a new hdr for the buffer. */ void arc_release(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; /* * It would be nice to assert that if it's DMU metadata (level > * 0 || it's the dnode file), then it must be syncing context. * But we don't know that information at this level. */ mutex_enter(&buf->b_evict_lock); ASSERT(HDR_HAS_L1HDR(hdr)); /* * We don't grab the hash lock prior to this check, because if * the buffer's header is in the arc_anon state, it won't be * linked into the hash table. */ if (hdr->b_l1hdr.b_state == arc_anon) { mutex_exit(&buf->b_evict_lock); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(!HDR_IN_HASH_TABLE(hdr)); ASSERT(!HDR_HAS_L2HDR(hdr)); ASSERT(HDR_EMPTY(hdr)); ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); hdr->b_l1hdr.b_arc_access = 0; /* * If the buf is being overridden then it may already * have a hdr that is not empty. */ buf_discard_identity(hdr); arc_buf_thaw(buf); return; } kmutex_t *hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); /* * This assignment is only valid as long as the hash_lock is * held, we must be careful not to reference state or the * b_state field after dropping the lock. */ arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3P(state, !=, arc_anon); /* this buffer is not on any list */ ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), >, 0); if (HDR_HAS_L2HDR(hdr)) { mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); /* * We have to recheck this conditional again now that * we're holding the l2ad_mtx to prevent a race with * another thread which might be concurrently calling * l2arc_evict(). In that case, l2arc_evict() might have * destroyed the header's L2 portion as we were waiting * to acquire the l2ad_mtx. */ if (HDR_HAS_L2HDR(hdr)) { l2arc_trim(hdr); arc_hdr_l2hdr_destroy(hdr); } mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); } /* * Do we have more than one buf? */ if (hdr->b_l1hdr.b_bufcnt > 1) { arc_buf_hdr_t *nhdr; uint64_t spa = hdr->b_spa; uint64_t psize = HDR_GET_PSIZE(hdr); uint64_t lsize = HDR_GET_LSIZE(hdr); enum zio_compress compress = HDR_GET_COMPRESS(hdr); arc_buf_contents_t type = arc_buf_type(hdr); VERIFY3U(hdr->b_type, ==, type); ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); (void) remove_reference(hdr, hash_lock, tag); if (arc_buf_is_shared(buf) && !ARC_BUF_COMPRESSED(buf)) { ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); ASSERT(ARC_BUF_LAST(buf)); } /* * Pull the data off of this hdr and attach it to * a new anonymous hdr. Also find the last buffer * in the hdr's buffer list. */ arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); ASSERT3P(lastbuf, !=, NULL); /* * If the current arc_buf_t and the hdr are sharing their data * buffer, then we must stop sharing that block. */ if (arc_buf_is_shared(buf)) { VERIFY(!arc_buf_is_shared(lastbuf)); /* * First, sever the block sharing relationship between * buf and the arc_buf_hdr_t. */ arc_unshare_buf(hdr, buf); /* * Now we need to recreate the hdr's b_pabd. Since we * have lastbuf handy, we try to share with it, but if * we can't then we allocate a new b_pabd and copy the * data from buf into it. */ if (arc_can_share(hdr, lastbuf)) { arc_share_buf(hdr, lastbuf); } else { arc_hdr_alloc_pabd(hdr); abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data, psize); } VERIFY3P(lastbuf->b_data, !=, NULL); } else if (HDR_SHARED_DATA(hdr)) { /* * Uncompressed shared buffers are always at the end * of the list. Compressed buffers don't have the * same requirements. This makes it hard to * simply assert that the lastbuf is shared so * we rely on the hdr's compression flags to determine * if we have a compressed, shared buffer. */ ASSERT(arc_buf_is_shared(lastbuf) || HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); ASSERT(!ARC_BUF_SHARED(buf)); } ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); ASSERT3P(state, !=, arc_l2c_only); (void) refcount_remove_many(&state->arcs_size, arc_buf_size(buf), buf); if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { ASSERT3P(state, !=, arc_l2c_only); (void) refcount_remove_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } hdr->b_l1hdr.b_bufcnt -= 1; arc_cksum_verify(buf); #ifdef illumos arc_buf_unwatch(buf); #endif mutex_exit(hash_lock); /* * Allocate a new hdr. The new hdr will contain a b_pabd * buffer which will be freed in arc_write(). */ nhdr = arc_hdr_alloc(spa, psize, lsize, compress, type); ASSERT3P(nhdr->b_l1hdr.b_buf, ==, NULL); ASSERT0(nhdr->b_l1hdr.b_bufcnt); ASSERT0(refcount_count(&nhdr->b_l1hdr.b_refcnt)); VERIFY3U(nhdr->b_type, ==, type); ASSERT(!HDR_SHARED_DATA(nhdr)); nhdr->b_l1hdr.b_buf = buf; nhdr->b_l1hdr.b_bufcnt = 1; (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); buf->b_hdr = nhdr; mutex_exit(&buf->b_evict_lock); (void) refcount_add_many(&arc_anon->arcs_size, arc_buf_size(buf), buf); } else { mutex_exit(&buf->b_evict_lock); ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); /* protected by hash lock, or hdr is on arc_anon */ ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); arc_change_state(arc_anon, hdr, hash_lock); hdr->b_l1hdr.b_arc_access = 0; mutex_exit(hash_lock); buf_discard_identity(hdr); arc_buf_thaw(buf); } } int arc_released(arc_buf_t *buf) { int released; mutex_enter(&buf->b_evict_lock); released = (buf->b_data != NULL && buf->b_hdr->b_l1hdr.b_state == arc_anon); mutex_exit(&buf->b_evict_lock); return (released); } #ifdef ZFS_DEBUG int arc_referenced(arc_buf_t *buf) { int referenced; mutex_enter(&buf->b_evict_lock); referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); mutex_exit(&buf->b_evict_lock); return (referenced); } #endif static void arc_write_ready(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; arc_buf_hdr_t *hdr = buf->b_hdr; uint64_t psize = BP_IS_HOLE(zio->io_bp) ? 0 : BP_GET_PSIZE(zio->io_bp); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); ASSERT(hdr->b_l1hdr.b_bufcnt > 0); /* * If we're reexecuting this zio because the pool suspended, then * cleanup any state that was previously set the first time the * callback was invoked. */ if (zio->io_flags & ZIO_FLAG_REEXECUTED) { arc_cksum_free(hdr); #ifdef illumos arc_buf_unwatch(buf); #endif if (hdr->b_l1hdr.b_pabd != NULL) { if (arc_buf_is_shared(buf)) { arc_unshare_buf(hdr, buf); } else { arc_hdr_free_pabd(hdr); } } } ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT(!arc_buf_is_shared(buf)); callback->awcb_ready(zio, buf, callback->awcb_private); if (HDR_IO_IN_PROGRESS(hdr)) ASSERT(zio->io_flags & ZIO_FLAG_REEXECUTED); arc_cksum_compute(buf); arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); enum zio_compress compress; if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { compress = ZIO_COMPRESS_OFF; } else { ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(zio->io_bp)); compress = BP_GET_COMPRESS(zio->io_bp); } HDR_SET_PSIZE(hdr, psize); arc_hdr_set_compress(hdr, compress); /* * Fill the hdr with data. If the hdr is compressed, the data we want * is available from the zio, otherwise we can take it from the buf. * * We might be able to share the buf's data with the hdr here. However, * doing so would cause the ARC to be full of linear ABDs if we write a * lot of shareable data. As a compromise, we check whether scattered * ABDs are allowed, and assume that if they are then the user wants * the ARC to be primarily filled with them regardless of the data being * written. Therefore, if they're allowed then we allocate one and copy * the data into it; otherwise, we share the data directly if we can. */ if (zfs_abd_scatter_enabled || !arc_can_share(hdr, buf)) { arc_hdr_alloc_pabd(hdr); /* * Ideally, we would always copy the io_abd into b_pabd, but the * user may have disabled compressed ARC, thus we must check the * hdr's compression setting rather than the io_bp's. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) { ASSERT3U(BP_GET_COMPRESS(zio->io_bp), !=, ZIO_COMPRESS_OFF); ASSERT3U(psize, >, 0); abd_copy(hdr->b_l1hdr.b_pabd, zio->io_abd, psize); } else { ASSERT3U(zio->io_orig_size, ==, arc_hdr_size(hdr)); abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data, arc_buf_size(buf)); } } else { ASSERT3P(buf->b_data, ==, abd_to_buf(zio->io_orig_abd)); ASSERT3U(zio->io_orig_size, ==, arc_buf_size(buf)); ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); arc_share_buf(hdr, buf); } arc_hdr_verify(hdr, zio->io_bp); } static void arc_write_children_ready(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; callback->awcb_children_ready(zio, buf, callback->awcb_private); } /* * The SPA calls this callback for each physical write that happens on behalf * of a logical write. See the comment in dbuf_write_physdone() for details. */ static void arc_write_physdone(zio_t *zio) { arc_write_callback_t *cb = zio->io_private; if (cb->awcb_physdone != NULL) cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); } static void arc_write_done(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); if (zio->io_error == 0) { arc_hdr_verify(hdr, zio->io_bp); if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { buf_discard_identity(hdr); } else { hdr->b_dva = *BP_IDENTITY(zio->io_bp); hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); } } else { ASSERT(HDR_EMPTY(hdr)); } /* * If the block to be written was all-zero or compressed enough to be * embedded in the BP, no write was performed so there will be no * dva/birth/checksum. The buffer must therefore remain anonymous * (and uncached). */ if (!HDR_EMPTY(hdr)) { arc_buf_hdr_t *exists; kmutex_t *hash_lock; ASSERT3U(zio->io_error, ==, 0); arc_cksum_verify(buf); exists = buf_hash_insert(hdr, &hash_lock); if (exists != NULL) { /* * This can only happen if we overwrite for * sync-to-convergence, because we remove * buffers from the hash table when we arc_free(). */ if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) panic("bad overwrite, hdr=%p exists=%p", (void *)hdr, (void *)exists); ASSERT(refcount_is_zero( &exists->b_l1hdr.b_refcnt)); arc_change_state(arc_anon, exists, hash_lock); mutex_exit(hash_lock); arc_hdr_destroy(exists); exists = buf_hash_insert(hdr, &hash_lock); ASSERT3P(exists, ==, NULL); } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { /* nopwrite */ ASSERT(zio->io_prop.zp_nopwrite); if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) panic("bad nopwrite, hdr=%p exists=%p", (void *)hdr, (void *)exists); } else { /* Dedup */ ASSERT(hdr->b_l1hdr.b_bufcnt == 1); ASSERT(hdr->b_l1hdr.b_state == arc_anon); ASSERT(BP_GET_DEDUP(zio->io_bp)); ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); } } arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); /* if it's not anon, we are doing a scrub */ if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) arc_access(hdr, hash_lock); mutex_exit(hash_lock); } else { arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); } ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); callback->awcb_done(zio, buf, callback->awcb_private); abd_put(zio->io_abd); kmem_free(callback, sizeof (arc_write_callback_t)); } zio_t * arc_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, const zio_prop_t *zp, arc_done_func_t *ready, arc_done_func_t *children_ready, arc_done_func_t *physdone, arc_done_func_t *done, void *private, zio_priority_t priority, int zio_flags, const zbookmark_phys_t *zb) { arc_buf_hdr_t *hdr = buf->b_hdr; arc_write_callback_t *callback; zio_t *zio; zio_prop_t localprop = *zp; ASSERT3P(ready, !=, NULL); ASSERT3P(done, !=, NULL); ASSERT(!HDR_IO_ERROR(hdr)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); if (l2arc) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); if (ARC_BUF_COMPRESSED(buf)) { /* * We're writing a pre-compressed buffer. Make the * compression algorithm requested by the zio_prop_t match * the pre-compressed buffer's compression algorithm. */ localprop.zp_compress = HDR_GET_COMPRESS(hdr); ASSERT3U(HDR_GET_LSIZE(hdr), !=, arc_buf_size(buf)); zio_flags |= ZIO_FLAG_RAW; } callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); callback->awcb_ready = ready; callback->awcb_children_ready = children_ready; callback->awcb_physdone = physdone; callback->awcb_done = done; callback->awcb_private = private; callback->awcb_buf = buf; /* * The hdr's b_pabd is now stale, free it now. A new data block * will be allocated when the zio pipeline calls arc_write_ready(). */ if (hdr->b_l1hdr.b_pabd != NULL) { /* * If the buf is currently sharing the data block with * the hdr then we need to break that relationship here. * The hdr will remain with a NULL data pointer and the * buf will take sole ownership of the block. */ if (arc_buf_is_shared(buf)) { arc_unshare_buf(hdr, buf); } else { arc_hdr_free_pabd(hdr); } VERIFY3P(buf->b_data, !=, NULL); arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF); } ASSERT(!arc_buf_is_shared(buf)); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); zio = zio_write(pio, spa, txg, bp, abd_get_from_buf(buf->b_data, HDR_GET_LSIZE(hdr)), HDR_GET_LSIZE(hdr), arc_buf_size(buf), &localprop, arc_write_ready, (children_ready != NULL) ? arc_write_children_ready : NULL, arc_write_physdone, arc_write_done, callback, priority, zio_flags, zb); return (zio); } static int arc_memory_throttle(uint64_t reserve, uint64_t txg) { #ifdef _KERNEL uint64_t available_memory = ptob(freemem); static uint64_t page_load = 0; static uint64_t last_txg = 0; #if defined(__i386) || !defined(UMA_MD_SMALL_ALLOC) - available_memory = - MIN(available_memory, ptob(vmem_size(heap_arena, VMEM_FREE))); + available_memory = MIN(available_memory, uma_avail()); #endif if (freemem > (uint64_t)physmem * arc_lotsfree_percent / 100) return (0); if (txg > last_txg) { last_txg = txg; page_load = 0; } /* * If we are in pageout, we know that memory is already tight, * the arc is already going to be evicting, so we just want to * continue to let page writes occur as quickly as possible. */ if (curproc == pageproc) { if (page_load > MAX(ptob(minfree), available_memory) / 4) return (SET_ERROR(ERESTART)); /* Note: reserve is inflated, so we deflate */ page_load += reserve / 8; return (0); } else if (page_load > 0 && arc_reclaim_needed()) { /* memory is low, delay before restarting */ ARCSTAT_INCR(arcstat_memory_throttle_count, 1); return (SET_ERROR(EAGAIN)); } page_load = 0; #endif return (0); } void arc_tempreserve_clear(uint64_t reserve) { atomic_add_64(&arc_tempreserve, -reserve); ASSERT((int64_t)arc_tempreserve >= 0); } int arc_tempreserve_space(uint64_t reserve, uint64_t txg) { int error; uint64_t anon_size; if (reserve > arc_c/4 && !arc_no_grow) { arc_c = MIN(arc_c_max, reserve * 4); DTRACE_PROBE1(arc__set_reserve, uint64_t, arc_c); } if (reserve > arc_c) return (SET_ERROR(ENOMEM)); /* * Don't count loaned bufs as in flight dirty data to prevent long * network delays from blocking transactions that are ready to be * assigned to a txg. */ /* assert that it has not wrapped around */ ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) - arc_loaned_bytes), 0); /* * Writes will, almost always, require additional memory allocations * in order to compress/encrypt/etc the data. We therefore need to * make sure that there is sufficient available memory for this. */ error = arc_memory_throttle(reserve, txg); if (error != 0) return (error); /* * Throttle writes when the amount of dirty data in the cache * gets too large. We try to keep the cache less than half full * of dirty blocks so that our sync times don't grow too large. * Note: if two requests come in concurrently, we might let them * both succeed, when one of them should fail. Not a huge deal. */ if (reserve + arc_tempreserve + anon_size > arc_c / 2 && anon_size > arc_c / 4) { uint64_t meta_esize = refcount_count(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); uint64_t data_esize = refcount_count(&arc_anon->arcs_esize[ARC_BUFC_DATA]); dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n", arc_tempreserve >> 10, meta_esize >> 10, data_esize >> 10, reserve >> 10, arc_c >> 10); return (SET_ERROR(ERESTART)); } atomic_add_64(&arc_tempreserve, reserve); return (0); } static void arc_kstat_update_state(arc_state_t *state, kstat_named_t *size, kstat_named_t *evict_data, kstat_named_t *evict_metadata) { size->value.ui64 = refcount_count(&state->arcs_size); evict_data->value.ui64 = refcount_count(&state->arcs_esize[ARC_BUFC_DATA]); evict_metadata->value.ui64 = refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]); } static int arc_kstat_update(kstat_t *ksp, int rw) { arc_stats_t *as = ksp->ks_data; if (rw == KSTAT_WRITE) { return (EACCES); } else { arc_kstat_update_state(arc_anon, &as->arcstat_anon_size, &as->arcstat_anon_evictable_data, &as->arcstat_anon_evictable_metadata); arc_kstat_update_state(arc_mru, &as->arcstat_mru_size, &as->arcstat_mru_evictable_data, &as->arcstat_mru_evictable_metadata); arc_kstat_update_state(arc_mru_ghost, &as->arcstat_mru_ghost_size, &as->arcstat_mru_ghost_evictable_data, &as->arcstat_mru_ghost_evictable_metadata); arc_kstat_update_state(arc_mfu, &as->arcstat_mfu_size, &as->arcstat_mfu_evictable_data, &as->arcstat_mfu_evictable_metadata); arc_kstat_update_state(arc_mfu_ghost, &as->arcstat_mfu_ghost_size, &as->arcstat_mfu_ghost_evictable_data, &as->arcstat_mfu_ghost_evictable_metadata); } return (0); } /* * This function *must* return indices evenly distributed between all * sublists of the multilist. This is needed due to how the ARC eviction * code is laid out; arc_evict_state() assumes ARC buffers are evenly * distributed between all sublists and uses this assumption when * deciding which sublist to evict from and how much to evict from it. */ unsigned int arc_state_multilist_index_func(multilist_t *ml, void *obj) { arc_buf_hdr_t *hdr = obj; /* * We rely on b_dva to generate evenly distributed index * numbers using buf_hash below. So, as an added precaution, * let's make sure we never add empty buffers to the arc lists. */ ASSERT(!HDR_EMPTY(hdr)); /* * The assumption here, is the hash value for a given * arc_buf_hdr_t will remain constant throughout it's lifetime * (i.e. it's b_spa, b_dva, and b_birth fields don't change). * Thus, we don't need to store the header's sublist index * on insertion, as this index can be recalculated on removal. * * Also, the low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. */ return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) % multilist_get_num_sublists(ml)); } #ifdef _KERNEL static eventhandler_tag arc_event_lowmem = NULL; static void arc_lowmem(void *arg __unused, int howto __unused) { mutex_enter(&arc_reclaim_lock); DTRACE_PROBE1(arc__needfree, int64_t, ((int64_t)freemem - zfs_arc_free_target) * PAGESIZE); cv_signal(&arc_reclaim_thread_cv); /* * It is unsafe to block here in arbitrary threads, because we can come * here from ARC itself and may hold ARC locks and thus risk a deadlock * with ARC reclaim thread. */ if (curproc == pageproc) (void) cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); mutex_exit(&arc_reclaim_lock); } #endif static void arc_state_init(void) { arc_anon = &ARC_anon; arc_mru = &ARC_mru; arc_mru_ghost = &ARC_mru_ghost; arc_mfu = &ARC_mfu; arc_mfu_ghost = &ARC_mfu_ghost; arc_l2c_only = &ARC_l2c_only; arc_mru->arcs_list[ARC_BUFC_METADATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_mru->arcs_list[ARC_BUFC_DATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_mru_ghost->arcs_list[ARC_BUFC_METADATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_mru_ghost->arcs_list[ARC_BUFC_DATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_mfu->arcs_list[ARC_BUFC_METADATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_mfu->arcs_list[ARC_BUFC_DATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_mfu_ghost->arcs_list[ARC_BUFC_DATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_l2c_only->arcs_list[ARC_BUFC_METADATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); arc_l2c_only->arcs_list[ARC_BUFC_DATA] = multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); refcount_create(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_anon->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mru->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_anon->arcs_size); refcount_create(&arc_mru->arcs_size); refcount_create(&arc_mru_ghost->arcs_size); refcount_create(&arc_mfu->arcs_size); refcount_create(&arc_mfu_ghost->arcs_size); refcount_create(&arc_l2c_only->arcs_size); } static void arc_state_fini(void) { refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_anon->arcs_size); refcount_destroy(&arc_mru->arcs_size); refcount_destroy(&arc_mru_ghost->arcs_size); refcount_destroy(&arc_mfu->arcs_size); refcount_destroy(&arc_mfu_ghost->arcs_size); refcount_destroy(&arc_l2c_only->arcs_size); multilist_destroy(arc_mru->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(arc_mru->arcs_list[ARC_BUFC_DATA]); multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_DATA]); multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); } uint64_t arc_max_bytes(void) { return (arc_c_max); } void arc_init(void) { int i, prefetch_tunable_set = 0; /* * allmem is "all memory that we could possibly use". */ #ifdef illumos #ifdef _KERNEL uint64_t allmem = ptob(physmem - swapfs_minfree); #else uint64_t allmem = (physmem * PAGESIZE) / 2; #endif #else uint64_t allmem = kmem_size(); #endif mutex_init(&arc_reclaim_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&arc_reclaim_thread_cv, NULL, CV_DEFAULT, NULL); cv_init(&arc_reclaim_waiters_cv, NULL, CV_DEFAULT, NULL); mutex_init(&arc_dnlc_evicts_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&arc_dnlc_evicts_cv, NULL, CV_DEFAULT, NULL); /* Convert seconds to clock ticks */ arc_min_prefetch_lifespan = 1 * hz; /* set min cache to 1/32 of all memory, or arc_abs_min, whichever is more */ arc_c_min = MAX(allmem / 32, arc_abs_min); /* set max to 5/8 of all memory, or all but 1GB, whichever is more */ if (allmem >= 1 << 30) arc_c_max = allmem - (1 << 30); else arc_c_max = arc_c_min; arc_c_max = MAX(allmem * 5 / 8, arc_c_max); /* * In userland, there's only the memory pressure that we artificially * create (see arc_available_memory()). Don't let arc_c get too * small, because it can cause transactions to be larger than * arc_c, causing arc_tempreserve_space() to fail. */ #ifndef _KERNEL arc_c_min = arc_c_max / 2; #endif #ifdef _KERNEL /* * Allow the tunables to override our calculations if they are * reasonable. */ if (zfs_arc_max > arc_abs_min && zfs_arc_max < allmem) { arc_c_max = zfs_arc_max; arc_c_min = MIN(arc_c_min, arc_c_max); } if (zfs_arc_min > arc_abs_min && zfs_arc_min <= arc_c_max) arc_c_min = zfs_arc_min; #endif arc_c = arc_c_max; arc_p = (arc_c >> 1); arc_size = 0; /* limit meta-data to 1/4 of the arc capacity */ arc_meta_limit = arc_c_max / 4; #ifdef _KERNEL /* * Metadata is stored in the kernel's heap. Don't let us * use more than half the heap for the ARC. */ +#ifdef __FreeBSD__ + arc_meta_limit = MIN(arc_meta_limit, uma_limit() / 2); +#else arc_meta_limit = MIN(arc_meta_limit, vmem_size(heap_arena, VMEM_ALLOC | VMEM_FREE) / 2); +#endif #endif /* Allow the tunable to override if it is reasonable */ if (zfs_arc_meta_limit > 0 && zfs_arc_meta_limit <= arc_c_max) arc_meta_limit = zfs_arc_meta_limit; if (arc_c_min < arc_meta_limit / 2 && zfs_arc_min == 0) arc_c_min = arc_meta_limit / 2; if (zfs_arc_meta_min > 0) { arc_meta_min = zfs_arc_meta_min; } else { arc_meta_min = arc_c_min / 2; } if (zfs_arc_grow_retry > 0) arc_grow_retry = zfs_arc_grow_retry; if (zfs_arc_shrink_shift > 0) arc_shrink_shift = zfs_arc_shrink_shift; if (zfs_arc_no_grow_shift > 0) arc_no_grow_shift = zfs_arc_no_grow_shift; /* * Ensure that arc_no_grow_shift is less than arc_shrink_shift. */ if (arc_no_grow_shift >= arc_shrink_shift) arc_no_grow_shift = arc_shrink_shift - 1; if (zfs_arc_p_min_shift > 0) arc_p_min_shift = zfs_arc_p_min_shift; /* if kmem_flags are set, lets try to use less memory */ if (kmem_debugging()) arc_c = arc_c / 2; if (arc_c < arc_c_min) arc_c = arc_c_min; zfs_arc_min = arc_c_min; zfs_arc_max = arc_c_max; arc_state_init(); buf_init(); arc_reclaim_thread_exit = B_FALSE; arc_dnlc_evicts_thread_exit = FALSE; arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (arc_ksp != NULL) { arc_ksp->ks_data = &arc_stats; arc_ksp->ks_update = arc_kstat_update; kstat_install(arc_ksp); } (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0, TS_RUN, minclsyspri); #ifdef _KERNEL arc_event_lowmem = EVENTHANDLER_REGISTER(vm_lowmem, arc_lowmem, NULL, EVENTHANDLER_PRI_FIRST); #endif (void) thread_create(NULL, 0, arc_dnlc_evicts_thread, NULL, 0, &p0, TS_RUN, minclsyspri); arc_dead = B_FALSE; arc_warm = B_FALSE; /* * Calculate maximum amount of dirty data per pool. * * If it has been set by /etc/system, take that. * Otherwise, use a percentage of physical memory defined by * zfs_dirty_data_max_percent (default 10%) with a cap at * zfs_dirty_data_max_max (default 4GB). */ if (zfs_dirty_data_max == 0) { zfs_dirty_data_max = ptob(physmem) * zfs_dirty_data_max_percent / 100; zfs_dirty_data_max = MIN(zfs_dirty_data_max, zfs_dirty_data_max_max); } #ifdef _KERNEL if (TUNABLE_INT_FETCH("vfs.zfs.prefetch_disable", &zfs_prefetch_disable)) prefetch_tunable_set = 1; #ifdef __i386__ if (prefetch_tunable_set == 0) { printf("ZFS NOTICE: Prefetch is disabled by default on i386 " "-- to enable,\n"); printf(" add \"vfs.zfs.prefetch_disable=0\" " "to /boot/loader.conf.\n"); zfs_prefetch_disable = 1; } #else if ((((uint64_t)physmem * PAGESIZE) < (1ULL << 32)) && prefetch_tunable_set == 0) { printf("ZFS NOTICE: Prefetch is disabled by default if less " "than 4GB of RAM is present;\n" " to enable, add \"vfs.zfs.prefetch_disable=0\" " "to /boot/loader.conf.\n"); zfs_prefetch_disable = 1; } #endif /* Warn about ZFS memory and address space requirements. */ if (((uint64_t)physmem * PAGESIZE) < (256 + 128 + 64) * (1 << 20)) { printf("ZFS WARNING: Recommended minimum RAM size is 512MB; " "expect unstable behavior.\n"); } if (allmem < 512 * (1 << 20)) { printf("ZFS WARNING: Recommended minimum kmem_size is 512MB; " "expect unstable behavior.\n"); printf(" Consider tuning vm.kmem_size and " "vm.kmem_size_max\n"); printf(" in /boot/loader.conf.\n"); } #endif } void arc_fini(void) { #ifdef _KERNEL if (arc_event_lowmem != NULL) EVENTHANDLER_DEREGISTER(vm_lowmem, arc_event_lowmem); #endif mutex_enter(&arc_reclaim_lock); arc_reclaim_thread_exit = B_TRUE; /* * The reclaim thread will set arc_reclaim_thread_exit back to * B_FALSE when it is finished exiting; we're waiting for that. */ while (arc_reclaim_thread_exit) { cv_signal(&arc_reclaim_thread_cv); cv_wait(&arc_reclaim_thread_cv, &arc_reclaim_lock); } mutex_exit(&arc_reclaim_lock); /* Use B_TRUE to ensure *all* buffers are evicted */ arc_flush(NULL, B_TRUE); mutex_enter(&arc_dnlc_evicts_lock); arc_dnlc_evicts_thread_exit = TRUE; /* * The user evicts thread will set arc_user_evicts_thread_exit * to FALSE when it is finished exiting; we're waiting for that. */ while (arc_dnlc_evicts_thread_exit) { cv_signal(&arc_dnlc_evicts_cv); cv_wait(&arc_dnlc_evicts_cv, &arc_dnlc_evicts_lock); } mutex_exit(&arc_dnlc_evicts_lock); arc_dead = B_TRUE; if (arc_ksp != NULL) { kstat_delete(arc_ksp); arc_ksp = NULL; } mutex_destroy(&arc_reclaim_lock); cv_destroy(&arc_reclaim_thread_cv); cv_destroy(&arc_reclaim_waiters_cv); mutex_destroy(&arc_dnlc_evicts_lock); cv_destroy(&arc_dnlc_evicts_cv); arc_state_fini(); buf_fini(); ASSERT0(arc_loaned_bytes); } /* * Level 2 ARC * * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. * It uses dedicated storage devices to hold cached data, which are populated * using large infrequent writes. The main role of this cache is to boost * the performance of random read workloads. The intended L2ARC devices * include short-stroked disks, solid state disks, and other media with * substantially faster read latency than disk. * * +-----------------------+ * | ARC | * +-----------------------+ * | ^ ^ * | | | * l2arc_feed_thread() arc_read() * | | | * | l2arc read | * V | | * +---------------+ | * | L2ARC | | * +---------------+ | * | ^ | * l2arc_write() | | * | | | * V | | * +-------+ +-------+ * | vdev | | vdev | * | cache | | cache | * +-------+ +-------+ * +=========+ .-----. * : L2ARC : |-_____-| * : devices : | Disks | * +=========+ `-_____-' * * Read requests are satisfied from the following sources, in order: * * 1) ARC * 2) vdev cache of L2ARC devices * 3) L2ARC devices * 4) vdev cache of disks * 5) disks * * Some L2ARC device types exhibit extremely slow write performance. * To accommodate for this there are some significant differences between * the L2ARC and traditional cache design: * * 1. There is no eviction path from the ARC to the L2ARC. Evictions from * the ARC behave as usual, freeing buffers and placing headers on ghost * lists. The ARC does not send buffers to the L2ARC during eviction as * this would add inflated write latencies for all ARC memory pressure. * * 2. The L2ARC attempts to cache data from the ARC before it is evicted. * It does this by periodically scanning buffers from the eviction-end of * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are * not already there. It scans until a headroom of buffers is satisfied, * which itself is a buffer for ARC eviction. If a compressible buffer is * found during scanning and selected for writing to an L2ARC device, we * temporarily boost scanning headroom during the next scan cycle to make * sure we adapt to compression effects (which might significantly reduce * the data volume we write to L2ARC). The thread that does this is * l2arc_feed_thread(), illustrated below; example sizes are included to * provide a better sense of ratio than this diagram: * * head --> tail * +---------------------+----------+ * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC * +---------------------+----------+ | o L2ARC eligible * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer * +---------------------+----------+ | * 15.9 Gbytes ^ 32 Mbytes | * headroom | * l2arc_feed_thread() * | * l2arc write hand <--[oooo]--' * | 8 Mbyte * | write max * V * +==============================+ * L2ARC dev |####|#|###|###| |####| ... | * +==============================+ * 32 Gbytes * * 3. If an ARC buffer is copied to the L2ARC but then hit instead of * evicted, then the L2ARC has cached a buffer much sooner than it probably * needed to, potentially wasting L2ARC device bandwidth and storage. It is * safe to say that this is an uncommon case, since buffers at the end of * the ARC lists have moved there due to inactivity. * * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, * then the L2ARC simply misses copying some buffers. This serves as a * pressure valve to prevent heavy read workloads from both stalling the ARC * with waits and clogging the L2ARC with writes. This also helps prevent * the potential for the L2ARC to churn if it attempts to cache content too * quickly, such as during backups of the entire pool. * * 5. After system boot and before the ARC has filled main memory, there are * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru * lists can remain mostly static. Instead of searching from tail of these * lists as pictured, the l2arc_feed_thread() will search from the list heads * for eligible buffers, greatly increasing its chance of finding them. * * The L2ARC device write speed is also boosted during this time so that * the L2ARC warms up faster. Since there have been no ARC evictions yet, * there are no L2ARC reads, and no fear of degrading read performance * through increased writes. * * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that * the vdev queue can aggregate them into larger and fewer writes. Each * device is written to in a rotor fashion, sweeping writes through * available space then repeating. * * 7. The L2ARC does not store dirty content. It never needs to flush * write buffers back to disk based storage. * * 8. If an ARC buffer is written (and dirtied) which also exists in the * L2ARC, the now stale L2ARC buffer is immediately dropped. * * The performance of the L2ARC can be tweaked by a number of tunables, which * may be necessary for different workloads: * * l2arc_write_max max write bytes per interval * l2arc_write_boost extra write bytes during device warmup * l2arc_noprefetch skip caching prefetched buffers * l2arc_headroom number of max device writes to precache * l2arc_headroom_boost when we find compressed buffers during ARC * scanning, we multiply headroom by this * percentage factor for the next scan cycle, * since more compressed buffers are likely to * be present * l2arc_feed_secs seconds between L2ARC writing * * Tunables may be removed or added as future performance improvements are * integrated, and also may become zpool properties. * * There are three key functions that control how the L2ARC warms up: * * l2arc_write_eligible() check if a buffer is eligible to cache * l2arc_write_size() calculate how much to write * l2arc_write_interval() calculate sleep delay between writes * * These three functions determine what to write, how much, and how quickly * to send writes. */ static boolean_t l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) { /* * A buffer is *not* eligible for the L2ARC if it: * 1. belongs to a different spa. * 2. is already cached on the L2ARC. * 3. has an I/O in progress (it may be an incomplete read). * 4. is flagged not eligible (zfs property). */ if (hdr->b_spa != spa_guid) { ARCSTAT_BUMP(arcstat_l2_write_spa_mismatch); return (B_FALSE); } if (HDR_HAS_L2HDR(hdr)) { ARCSTAT_BUMP(arcstat_l2_write_in_l2); return (B_FALSE); } if (HDR_IO_IN_PROGRESS(hdr)) { ARCSTAT_BUMP(arcstat_l2_write_hdr_io_in_progress); return (B_FALSE); } if (!HDR_L2CACHE(hdr)) { ARCSTAT_BUMP(arcstat_l2_write_not_cacheable); return (B_FALSE); } return (B_TRUE); } static uint64_t l2arc_write_size(void) { uint64_t size; /* * Make sure our globals have meaningful values in case the user * altered them. */ size = l2arc_write_max; if (size == 0) { cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " "be greater than zero, resetting it to the default (%d)", L2ARC_WRITE_SIZE); size = l2arc_write_max = L2ARC_WRITE_SIZE; } if (arc_warm == B_FALSE) size += l2arc_write_boost; return (size); } static clock_t l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) { clock_t interval, next, now; /* * If the ARC lists are busy, increase our write rate; if the * lists are stale, idle back. This is achieved by checking * how much we previously wrote - if it was more than half of * what we wanted, schedule the next write much sooner. */ if (l2arc_feed_again && wrote > (wanted / 2)) interval = (hz * l2arc_feed_min_ms) / 1000; else interval = hz * l2arc_feed_secs; now = ddi_get_lbolt(); next = MAX(now, MIN(now + interval, began + interval)); return (next); } /* * Cycle through L2ARC devices. This is how L2ARC load balances. * If a device is returned, this also returns holding the spa config lock. */ static l2arc_dev_t * l2arc_dev_get_next(void) { l2arc_dev_t *first, *next = NULL; /* * Lock out the removal of spas (spa_namespace_lock), then removal * of cache devices (l2arc_dev_mtx). Once a device has been selected, * both locks will be dropped and a spa config lock held instead. */ mutex_enter(&spa_namespace_lock); mutex_enter(&l2arc_dev_mtx); /* if there are no vdevs, there is nothing to do */ if (l2arc_ndev == 0) goto out; first = NULL; next = l2arc_dev_last; do { /* loop around the list looking for a non-faulted vdev */ if (next == NULL) { next = list_head(l2arc_dev_list); } else { next = list_next(l2arc_dev_list, next); if (next == NULL) next = list_head(l2arc_dev_list); } /* if we have come back to the start, bail out */ if (first == NULL) first = next; else if (next == first) break; } while (vdev_is_dead(next->l2ad_vdev)); /* if we were unable to find any usable vdevs, return NULL */ if (vdev_is_dead(next->l2ad_vdev)) next = NULL; l2arc_dev_last = next; out: mutex_exit(&l2arc_dev_mtx); /* * Grab the config lock to prevent the 'next' device from being * removed while we are writing to it. */ if (next != NULL) spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); mutex_exit(&spa_namespace_lock); return (next); } /* * Free buffers that were tagged for destruction. */ static void l2arc_do_free_on_write() { list_t *buflist; l2arc_data_free_t *df, *df_prev; mutex_enter(&l2arc_free_on_write_mtx); buflist = l2arc_free_on_write; for (df = list_tail(buflist); df; df = df_prev) { df_prev = list_prev(buflist, df); ASSERT3P(df->l2df_abd, !=, NULL); abd_free(df->l2df_abd); list_remove(buflist, df); kmem_free(df, sizeof (l2arc_data_free_t)); } mutex_exit(&l2arc_free_on_write_mtx); } /* * A write to a cache device has completed. Update all headers to allow * reads from these buffers to begin. */ static void l2arc_write_done(zio_t *zio) { l2arc_write_callback_t *cb; l2arc_dev_t *dev; list_t *buflist; arc_buf_hdr_t *head, *hdr, *hdr_prev; kmutex_t *hash_lock; int64_t bytes_dropped = 0; cb = zio->io_private; ASSERT3P(cb, !=, NULL); dev = cb->l2wcb_dev; ASSERT3P(dev, !=, NULL); head = cb->l2wcb_head; ASSERT3P(head, !=, NULL); buflist = &dev->l2ad_buflist; ASSERT3P(buflist, !=, NULL); DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, l2arc_write_callback_t *, cb); if (zio->io_error != 0) ARCSTAT_BUMP(arcstat_l2_writes_error); /* * All writes completed, or an error was hit. */ top: mutex_enter(&dev->l2ad_mtx); for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { hdr_prev = list_prev(buflist, hdr); hash_lock = HDR_LOCK(hdr); /* * We cannot use mutex_enter or else we can deadlock * with l2arc_write_buffers (due to swapping the order * the hash lock and l2ad_mtx are taken). */ if (!mutex_tryenter(hash_lock)) { /* * Missed the hash lock. We must retry so we * don't leave the ARC_FLAG_L2_WRITING bit set. */ ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); /* * We don't want to rescan the headers we've * already marked as having been written out, so * we reinsert the head node so we can pick up * where we left off. */ list_remove(buflist, head); list_insert_after(buflist, hdr, head); mutex_exit(&dev->l2ad_mtx); /* * We wait for the hash lock to become available * to try and prevent busy waiting, and increase * the chance we'll be able to acquire the lock * the next time around. */ mutex_enter(hash_lock); mutex_exit(hash_lock); goto top; } /* * We could not have been moved into the arc_l2c_only * state while in-flight due to our ARC_FLAG_L2_WRITING * bit being set. Let's just ensure that's being enforced. */ ASSERT(HDR_HAS_L1HDR(hdr)); if (zio->io_error != 0) { /* * Error - drop L2ARC entry. */ list_remove(buflist, hdr); l2arc_trim(hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); ARCSTAT_INCR(arcstat_l2_psize, -arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_l2_lsize, -HDR_GET_LSIZE(hdr)); bytes_dropped += arc_hdr_size(hdr); (void) refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); } /* * Allow ARC to begin reads and ghost list evictions to * this L2ARC entry. */ arc_hdr_clear_flags(hdr, ARC_FLAG_L2_WRITING); mutex_exit(hash_lock); } atomic_inc_64(&l2arc_writes_done); list_remove(buflist, head); ASSERT(!HDR_HAS_L1HDR(head)); kmem_cache_free(hdr_l2only_cache, head); mutex_exit(&dev->l2ad_mtx); vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); l2arc_do_free_on_write(); kmem_free(cb, sizeof (l2arc_write_callback_t)); } /* * A read to a cache device completed. Validate buffer contents before * handing over to the regular ARC routines. */ static void l2arc_read_done(zio_t *zio) { l2arc_read_callback_t *cb; arc_buf_hdr_t *hdr; kmutex_t *hash_lock; boolean_t valid_cksum; ASSERT3P(zio->io_vd, !=, NULL); ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); cb = zio->io_private; ASSERT3P(cb, !=, NULL); hdr = cb->l2rcb_hdr; ASSERT3P(hdr, !=, NULL); hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); /* * If the data was read into a temporary buffer, * move it and free the buffer. */ if (cb->l2rcb_abd != NULL) { ASSERT3U(arc_hdr_size(hdr), <, zio->io_size); if (zio->io_error == 0) { abd_copy(hdr->b_l1hdr.b_pabd, cb->l2rcb_abd, arc_hdr_size(hdr)); } /* * The following must be done regardless of whether * there was an error: * - free the temporary buffer * - point zio to the real ARC buffer * - set zio size accordingly * These are required because zio is either re-used for * an I/O of the block in the case of the error * or the zio is passed to arc_read_done() and it * needs real data. */ abd_free(cb->l2rcb_abd); zio->io_size = zio->io_orig_size = arc_hdr_size(hdr); zio->io_abd = zio->io_orig_abd = hdr->b_l1hdr.b_pabd; } ASSERT3P(zio->io_abd, !=, NULL); /* * Check this survived the L2ARC journey. */ ASSERT3P(zio->io_abd, ==, hdr->b_l1hdr.b_pabd); zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ valid_cksum = arc_cksum_is_equal(hdr, zio); if (valid_cksum && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { mutex_exit(hash_lock); zio->io_private = hdr; arc_read_done(zio); } else { mutex_exit(hash_lock); /* * Buffer didn't survive caching. Increment stats and * reissue to the original storage device. */ if (zio->io_error != 0) { ARCSTAT_BUMP(arcstat_l2_io_error); } else { zio->io_error = SET_ERROR(EIO); } if (!valid_cksum) ARCSTAT_BUMP(arcstat_l2_cksum_bad); /* * If there's no waiter, issue an async i/o to the primary * storage now. If there *is* a waiter, the caller must * issue the i/o in a context where it's OK to block. */ if (zio->io_waiter == NULL) { zio_t *pio = zio_unique_parent(zio); ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); zio_nowait(zio_read(pio, zio->io_spa, zio->io_bp, hdr->b_l1hdr.b_pabd, zio->io_size, arc_read_done, hdr, zio->io_priority, cb->l2rcb_flags, &cb->l2rcb_zb)); } } kmem_free(cb, sizeof (l2arc_read_callback_t)); } /* * This is the list priority from which the L2ARC will search for pages to * cache. This is used within loops (0..3) to cycle through lists in the * desired order. This order can have a significant effect on cache * performance. * * Currently the metadata lists are hit first, MFU then MRU, followed by * the data lists. This function returns a locked list, and also returns * the lock pointer. */ static multilist_sublist_t * l2arc_sublist_lock(int list_num) { multilist_t *ml = NULL; unsigned int idx; ASSERT(list_num >= 0 && list_num <= 3); switch (list_num) { case 0: ml = arc_mfu->arcs_list[ARC_BUFC_METADATA]; break; case 1: ml = arc_mru->arcs_list[ARC_BUFC_METADATA]; break; case 2: ml = arc_mfu->arcs_list[ARC_BUFC_DATA]; break; case 3: ml = arc_mru->arcs_list[ARC_BUFC_DATA]; break; } /* * Return a randomly-selected sublist. This is acceptable * because the caller feeds only a little bit of data for each * call (8MB). Subsequent calls will result in different * sublists being selected. */ idx = multilist_get_random_index(ml); return (multilist_sublist_lock(ml, idx)); } /* * Evict buffers from the device write hand to the distance specified in * bytes. This distance may span populated buffers, it may span nothing. * This is clearing a region on the L2ARC device ready for writing. * If the 'all' boolean is set, every buffer is evicted. */ static void l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) { list_t *buflist; arc_buf_hdr_t *hdr, *hdr_prev; kmutex_t *hash_lock; uint64_t taddr; buflist = &dev->l2ad_buflist; if (!all && dev->l2ad_first) { /* * This is the first sweep through the device. There is * nothing to evict. */ return; } if (dev->l2ad_hand >= (dev->l2ad_end - (2 * distance))) { /* * When nearing the end of the device, evict to the end * before the device write hand jumps to the start. */ taddr = dev->l2ad_end; } else { taddr = dev->l2ad_hand + distance; } DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, uint64_t, taddr, boolean_t, all); top: mutex_enter(&dev->l2ad_mtx); for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { hdr_prev = list_prev(buflist, hdr); hash_lock = HDR_LOCK(hdr); /* * We cannot use mutex_enter or else we can deadlock * with l2arc_write_buffers (due to swapping the order * the hash lock and l2ad_mtx are taken). */ if (!mutex_tryenter(hash_lock)) { /* * Missed the hash lock. Retry. */ ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); mutex_exit(&dev->l2ad_mtx); mutex_enter(hash_lock); mutex_exit(hash_lock); goto top; } /* * A header can't be on this list if it doesn't have L2 header. */ ASSERT(HDR_HAS_L2HDR(hdr)); /* Ensure this header has finished being written. */ ASSERT(!HDR_L2_WRITING(hdr)); ASSERT(!HDR_L2_WRITE_HEAD(hdr)); if (!all && (hdr->b_l2hdr.b_daddr >= taddr || hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { /* * We've evicted to the target address, * or the end of the device. */ mutex_exit(hash_lock); break; } if (!HDR_HAS_L1HDR(hdr)) { ASSERT(!HDR_L2_READING(hdr)); /* * This doesn't exist in the ARC. Destroy. * arc_hdr_destroy() will call list_remove() * and decrement arcstat_l2_lsize. */ arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); } else { ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); ARCSTAT_BUMP(arcstat_l2_evict_l1cached); /* * Invalidate issued or about to be issued * reads, since we may be about to write * over this location. */ if (HDR_L2_READING(hdr)) { ARCSTAT_BUMP(arcstat_l2_evict_reading); arc_hdr_set_flags(hdr, ARC_FLAG_L2_EVICTED); } arc_hdr_l2hdr_destroy(hdr); } mutex_exit(hash_lock); } mutex_exit(&dev->l2ad_mtx); } /* * Find and write ARC buffers to the L2ARC device. * * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid * for reading until they have completed writing. * The headroom_boost is an in-out parameter used to maintain headroom boost * state between calls to this function. * * Returns the number of bytes actually written (which may be smaller than * the delta by which the device hand has changed due to alignment). */ static uint64_t l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz) { arc_buf_hdr_t *hdr, *hdr_prev, *head; uint64_t write_asize, write_psize, write_lsize, headroom; boolean_t full; l2arc_write_callback_t *cb; zio_t *pio, *wzio; uint64_t guid = spa_load_guid(spa); int try; ASSERT3P(dev->l2ad_vdev, !=, NULL); pio = NULL; write_lsize = write_asize = write_psize = 0; full = B_FALSE; head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); arc_hdr_set_flags(head, ARC_FLAG_L2_WRITE_HEAD | ARC_FLAG_HAS_L2HDR); ARCSTAT_BUMP(arcstat_l2_write_buffer_iter); /* * Copy buffers for L2ARC writing. */ for (try = 0; try <= 3; try++) { multilist_sublist_t *mls = l2arc_sublist_lock(try); uint64_t passed_sz = 0; ARCSTAT_BUMP(arcstat_l2_write_buffer_list_iter); /* * L2ARC fast warmup. * * Until the ARC is warm and starts to evict, read from the * head of the ARC lists rather than the tail. */ if (arc_warm == B_FALSE) hdr = multilist_sublist_head(mls); else hdr = multilist_sublist_tail(mls); if (hdr == NULL) ARCSTAT_BUMP(arcstat_l2_write_buffer_list_null_iter); headroom = target_sz * l2arc_headroom; if (zfs_compressed_arc_enabled) headroom = (headroom * l2arc_headroom_boost) / 100; for (; hdr; hdr = hdr_prev) { kmutex_t *hash_lock; if (arc_warm == B_FALSE) hdr_prev = multilist_sublist_next(mls, hdr); else hdr_prev = multilist_sublist_prev(mls, hdr); ARCSTAT_INCR(arcstat_l2_write_buffer_bytes_scanned, HDR_GET_LSIZE(hdr)); hash_lock = HDR_LOCK(hdr); if (!mutex_tryenter(hash_lock)) { ARCSTAT_BUMP(arcstat_l2_write_trylock_fail); /* * Skip this buffer rather than waiting. */ continue; } passed_sz += HDR_GET_LSIZE(hdr); if (passed_sz > headroom) { /* * Searched too far. */ mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_l2_write_passed_headroom); break; } if (!l2arc_write_eligible(guid, hdr)) { mutex_exit(hash_lock); continue; } /* * We rely on the L1 portion of the header below, so * it's invalid for this header to have been evicted out * of the ghost cache, prior to being written out. The * ARC_FLAG_L2_WRITING bit ensures this won't happen. */ ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3U(HDR_GET_PSIZE(hdr), >, 0); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); ASSERT3U(arc_hdr_size(hdr), >, 0); uint64_t psize = arc_hdr_size(hdr); uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, psize); if ((write_asize + asize) > target_sz) { full = B_TRUE; mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_l2_write_full); break; } if (pio == NULL) { /* * Insert a dummy header on the buflist so * l2arc_write_done() can find where the * write buffers begin without searching. */ mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_buflist, head); mutex_exit(&dev->l2ad_mtx); cb = kmem_alloc( sizeof (l2arc_write_callback_t), KM_SLEEP); cb->l2wcb_dev = dev; cb->l2wcb_head = head; pio = zio_root(spa, l2arc_write_done, cb, ZIO_FLAG_CANFAIL); ARCSTAT_BUMP(arcstat_l2_write_pios); } hdr->b_l2hdr.b_dev = dev; hdr->b_l2hdr.b_daddr = dev->l2ad_hand; arc_hdr_set_flags(hdr, ARC_FLAG_L2_WRITING | ARC_FLAG_HAS_L2HDR); mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_buflist, hdr); mutex_exit(&dev->l2ad_mtx); (void) refcount_add_many(&dev->l2ad_alloc, psize, hdr); /* * Normally the L2ARC can use the hdr's data, but if * we're sharing data between the hdr and one of its * bufs, L2ARC needs its own copy of the data so that * the ZIO below can't race with the buf consumer. * Another case where we need to create a copy of the * data is when the buffer size is not device-aligned * and we need to pad the block to make it such. * That also keeps the clock hand suitably aligned. * * To ensure that the copy will be available for the * lifetime of the ZIO and be cleaned up afterwards, we * add it to the l2arc_free_on_write queue. */ abd_t *to_write; if (!HDR_SHARED_DATA(hdr) && psize == asize) { to_write = hdr->b_l1hdr.b_pabd; } else { to_write = abd_alloc_for_io(asize, HDR_ISTYPE_METADATA(hdr)); abd_copy(to_write, hdr->b_l1hdr.b_pabd, psize); if (asize != psize) { abd_zero_off(to_write, psize, asize - psize); } l2arc_free_abd_on_write(to_write, asize, arc_buf_type(hdr)); } wzio = zio_write_phys(pio, dev->l2ad_vdev, hdr->b_l2hdr.b_daddr, asize, to_write, ZIO_CHECKSUM_OFF, NULL, hdr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE); write_lsize += HDR_GET_LSIZE(hdr); DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, zio_t *, wzio); write_psize += psize; write_asize += asize; dev->l2ad_hand += asize; mutex_exit(hash_lock); (void) zio_nowait(wzio); } multilist_sublist_unlock(mls); if (full == B_TRUE) break; } /* No buffers selected for writing? */ if (pio == NULL) { ASSERT0(write_lsize); ASSERT(!HDR_HAS_L1HDR(head)); kmem_cache_free(hdr_l2only_cache, head); return (0); } ASSERT3U(write_psize, <=, target_sz); ARCSTAT_BUMP(arcstat_l2_writes_sent); ARCSTAT_INCR(arcstat_l2_write_bytes, write_psize); ARCSTAT_INCR(arcstat_l2_lsize, write_lsize); ARCSTAT_INCR(arcstat_l2_psize, write_psize); vdev_space_update(dev->l2ad_vdev, write_psize, 0, 0); /* * Bump device hand to the device start if it is approaching the end. * l2arc_evict() will already have evicted ahead for this case. */ if (dev->l2ad_hand >= (dev->l2ad_end - target_sz)) { dev->l2ad_hand = dev->l2ad_start; dev->l2ad_first = B_FALSE; } dev->l2ad_writing = B_TRUE; (void) zio_wait(pio); dev->l2ad_writing = B_FALSE; return (write_asize); } /* * This thread feeds the L2ARC at regular intervals. This is the beating * heart of the L2ARC. */ static void l2arc_feed_thread(void *dummy __unused) { callb_cpr_t cpr; l2arc_dev_t *dev; spa_t *spa; uint64_t size, wrote; clock_t begin, next = ddi_get_lbolt(); CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); mutex_enter(&l2arc_feed_thr_lock); while (l2arc_thread_exit == 0) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock, next - ddi_get_lbolt()); CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); next = ddi_get_lbolt() + hz; /* * Quick check for L2ARC devices. */ mutex_enter(&l2arc_dev_mtx); if (l2arc_ndev == 0) { mutex_exit(&l2arc_dev_mtx); continue; } mutex_exit(&l2arc_dev_mtx); begin = ddi_get_lbolt(); /* * This selects the next l2arc device to write to, and in * doing so the next spa to feed from: dev->l2ad_spa. This * will return NULL if there are now no l2arc devices or if * they are all faulted. * * If a device is returned, its spa's config lock is also * held to prevent device removal. l2arc_dev_get_next() * will grab and release l2arc_dev_mtx. */ if ((dev = l2arc_dev_get_next()) == NULL) continue; spa = dev->l2ad_spa; ASSERT3P(spa, !=, NULL); /* * If the pool is read-only then force the feed thread to * sleep a little longer. */ if (!spa_writeable(spa)) { next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; spa_config_exit(spa, SCL_L2ARC, dev); continue; } /* * Avoid contributing to memory pressure. */ if (arc_reclaim_needed()) { ARCSTAT_BUMP(arcstat_l2_abort_lowmem); spa_config_exit(spa, SCL_L2ARC, dev); continue; } ARCSTAT_BUMP(arcstat_l2_feeds); size = l2arc_write_size(); /* * Evict L2ARC buffers that will be overwritten. */ l2arc_evict(dev, size, B_FALSE); /* * Write ARC buffers. */ wrote = l2arc_write_buffers(spa, dev, size); /* * Calculate interval between writes. */ next = l2arc_write_interval(begin, size, wrote); spa_config_exit(spa, SCL_L2ARC, dev); } l2arc_thread_exit = 0; cv_broadcast(&l2arc_feed_thr_cv); CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ thread_exit(); } boolean_t l2arc_vdev_present(vdev_t *vd) { l2arc_dev_t *dev; mutex_enter(&l2arc_dev_mtx); for (dev = list_head(l2arc_dev_list); dev != NULL; dev = list_next(l2arc_dev_list, dev)) { if (dev->l2ad_vdev == vd) break; } mutex_exit(&l2arc_dev_mtx); return (dev != NULL); } /* * Add a vdev for use by the L2ARC. By this point the spa has already * validated the vdev and opened it. */ void l2arc_add_vdev(spa_t *spa, vdev_t *vd) { l2arc_dev_t *adddev; ASSERT(!l2arc_vdev_present(vd)); vdev_ashift_optimize(vd); /* * Create a new l2arc device entry. */ adddev = kmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); adddev->l2ad_spa = spa; adddev->l2ad_vdev = vd; adddev->l2ad_start = VDEV_LABEL_START_SIZE; adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); adddev->l2ad_hand = adddev->l2ad_start; adddev->l2ad_first = B_TRUE; adddev->l2ad_writing = B_FALSE; mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); /* * This is a list of all ARC buffers that are still valid on the * device. */ list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); refcount_create(&adddev->l2ad_alloc); /* * Add device to global list */ mutex_enter(&l2arc_dev_mtx); list_insert_head(l2arc_dev_list, adddev); atomic_inc_64(&l2arc_ndev); mutex_exit(&l2arc_dev_mtx); } /* * Remove a vdev from the L2ARC. */ void l2arc_remove_vdev(vdev_t *vd) { l2arc_dev_t *dev, *nextdev, *remdev = NULL; /* * Find the device by vdev */ mutex_enter(&l2arc_dev_mtx); for (dev = list_head(l2arc_dev_list); dev; dev = nextdev) { nextdev = list_next(l2arc_dev_list, dev); if (vd == dev->l2ad_vdev) { remdev = dev; break; } } ASSERT3P(remdev, !=, NULL); /* * Remove device from global list */ list_remove(l2arc_dev_list, remdev); l2arc_dev_last = NULL; /* may have been invalidated */ atomic_dec_64(&l2arc_ndev); mutex_exit(&l2arc_dev_mtx); /* * Clear all buflists and ARC references. L2ARC device flush. */ l2arc_evict(remdev, 0, B_TRUE); list_destroy(&remdev->l2ad_buflist); mutex_destroy(&remdev->l2ad_mtx); refcount_destroy(&remdev->l2ad_alloc); kmem_free(remdev, sizeof (l2arc_dev_t)); } void l2arc_init(void) { l2arc_thread_exit = 0; l2arc_ndev = 0; l2arc_writes_sent = 0; l2arc_writes_done = 0; mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); l2arc_dev_list = &L2ARC_dev_list; l2arc_free_on_write = &L2ARC_free_on_write; list_create(l2arc_dev_list, sizeof (l2arc_dev_t), offsetof(l2arc_dev_t, l2ad_node)); list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), offsetof(l2arc_data_free_t, l2df_list_node)); } void l2arc_fini(void) { /* * This is called from dmu_fini(), which is called from spa_fini(); * Because of this, we can assume that all l2arc devices have * already been removed when the pools themselves were removed. */ l2arc_do_free_on_write(); mutex_destroy(&l2arc_feed_thr_lock); cv_destroy(&l2arc_feed_thr_cv); mutex_destroy(&l2arc_dev_mtx); mutex_destroy(&l2arc_free_on_write_mtx); list_destroy(l2arc_dev_list); list_destroy(l2arc_free_on_write); } void l2arc_start(void) { if (!(spa_mode_global & FWRITE)) return; (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, TS_RUN, minclsyspri); } void l2arc_stop(void) { if (!(spa_mode_global & FWRITE)) return; mutex_enter(&l2arc_feed_thr_lock); cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ l2arc_thread_exit = 1; while (l2arc_thread_exit != 0) cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); mutex_exit(&l2arc_feed_thr_lock); } Index: head/sys/vm/uma.h =================================================================== --- head/sys/vm/uma.h (revision 327484) +++ head/sys/vm/uma.h (revision 327485) @@ -1,701 +1,709 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002, 2003, 2004, 2005 Jeffrey Roberson * Copyright (c) 2004, 2005 Bosko Milekic * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ * */ /* * uma.h - External definitions for the Universal Memory Allocator * */ #ifndef _VM_UMA_H_ #define _VM_UMA_H_ #include /* For NULL */ #include /* For M_* */ /* User visible parameters */ #define UMA_SMALLEST_UNIT (PAGE_SIZE / 256) /* Smallest item allocated */ /* Types and type defs */ struct uma_zone; /* Opaque type used as a handle to the zone */ typedef struct uma_zone * uma_zone_t; void zone_drain(uma_zone_t); /* * Item constructor * * Arguments: * item A pointer to the memory which has been allocated. * arg The arg field passed to uma_zalloc_arg * size The size of the allocated item * flags See zalloc flags * * Returns: * 0 on success * errno on failure * * Discussion: * The constructor is called just before the memory is returned * to the user. It may block if necessary. */ typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags); /* * Item destructor * * Arguments: * item A pointer to the memory which has been allocated. * size The size of the item being destructed. * arg Argument passed through uma_zfree_arg * * Returns: * Nothing * * Discussion: * The destructor may perform operations that differ from those performed * by the initializer, but it must leave the object in the same state. * This IS type stable storage. This is called after EVERY zfree call. */ typedef void (*uma_dtor)(void *mem, int size, void *arg); /* * Item initializer * * Arguments: * item A pointer to the memory which has been allocated. * size The size of the item being initialized. * flags See zalloc flags * * Returns: * 0 on success * errno on failure * * Discussion: * The initializer is called when the memory is cached in the uma zone. * The initializer and the destructor should leave the object in the same * state. */ typedef int (*uma_init)(void *mem, int size, int flags); /* * Item discard function * * Arguments: * item A pointer to memory which has been 'freed' but has not left the * zone's cache. * size The size of the item being discarded. * * Returns: * Nothing * * Discussion: * This routine is called when memory leaves a zone and is returned to the * system for other uses. It is the counter-part to the init function. */ typedef void (*uma_fini)(void *mem, int size); /* * Import new memory into a cache zone. */ typedef int (*uma_import)(void *arg, void **store, int count, int flags); /* * Free memory from a cache zone. */ typedef void (*uma_release)(void *arg, void **store, int count); /* * What's the difference between initializing and constructing? * * The item is initialized when it is cached, and this is the state that the * object should be in when returned to the allocator. The purpose of this is * to remove some code which would otherwise be called on each allocation by * utilizing a known, stable state. This differs from the constructor which * will be called on EVERY allocation. * * For example, in the initializer you may want to initialize embedded locks, * NULL list pointers, set up initial states, magic numbers, etc. This way if * the object is held in the allocator and re-used it won't be necessary to * re-initialize it. * * The constructor may be used to lock a data structure, link it on to lists, * bump reference counts or total counts of outstanding structures, etc. * */ /* Function proto types */ /* * Create a new uma zone * * Arguments: * name The text name of the zone for debugging and stats. This memory * should not be freed until the zone has been deallocated. * size The size of the object that is being created. * ctor The constructor that is called when the object is allocated. * dtor The destructor that is called when the object is freed. * init An initializer that sets up the initial state of the memory. * fini A discard function that undoes initialization done by init. * ctor/dtor/init/fini may all be null, see notes above. * align A bitmask that corresponds to the requested alignment * eg 4 would be 0x3 * flags A set of parameters that control the behavior of the zone. * * Returns: * A pointer to a structure which is intended to be opaque to users of * the interface. The value may be null if the wait flag is not set. */ uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor, uma_dtor dtor, uma_init uminit, uma_fini fini, int align, uint32_t flags); /* * Create a secondary uma zone * * Arguments: * name The text name of the zone for debugging and stats. This memory * should not be freed until the zone has been deallocated. * ctor The constructor that is called when the object is allocated. * dtor The destructor that is called when the object is freed. * zinit An initializer that sets up the initial state of the memory * as the object passes from the Keg's slab to the Zone's cache. * zfini A discard function that undoes initialization done by init * as the object passes from the Zone's cache to the Keg's slab. * * ctor/dtor/zinit/zfini may all be null, see notes above. * Note that the zinit and zfini specified here are NOT * exactly the same as the init/fini specified to uma_zcreate() * when creating a master zone. These zinit/zfini are called * on the TRANSITION from keg to zone (and vice-versa). Once * these are set, the primary zone may alter its init/fini * (which are called when the object passes from VM to keg) * using uma_zone_set_init/fini()) as well as its own * zinit/zfini (unset by default for master zone) with * uma_zone_set_zinit/zfini() (note subtle 'z' prefix). * * master A reference to this zone's Master Zone (Primary Zone), * which contains the backing Keg for the Secondary Zone * being added. * * Returns: * A pointer to a structure which is intended to be opaque to users of * the interface. The value may be null if the wait flag is not set. */ uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_zone_t master); /* * Add a second master to a secondary zone. This provides multiple data * backends for objects with the same size. Both masters must have * compatible allocation flags. Presently, UMA_ZONE_MALLOC type zones are * the only supported. * * Returns: * Error on failure, 0 on success. */ int uma_zsecond_add(uma_zone_t zone, uma_zone_t master); /* * Create cache-only zones. * * This allows uma's per-cpu cache facilities to handle arbitrary * pointers. Consumers must specify the import and release functions to * fill and destroy caches. UMA does not allocate any memory for these * zones. The 'arg' parameter is passed to import/release and is caller * specific. */ uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_import zimport, uma_release zrelease, void *arg, int flags); /* * Definitions for uma_zcreate flags * * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to * overlap when adding new features. 0xff000000 is in use by uma_int.h. */ #define UMA_ZONE_PAGEABLE 0x0001 /* Return items not fully backed by physical memory XXX Not yet */ #define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */ #define UMA_ZONE_STATIC 0x0004 /* Statically sized zone */ #define UMA_ZONE_OFFPAGE 0x0008 /* Force the slab structure allocation off of the real memory */ #define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */ #define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */ #define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */ #define UMA_ZONE_VM 0x0080 /* * Used for internal vm datastructures * only. */ #define UMA_ZONE_HASH 0x0100 /* * Use a hash table instead of caching * information in the vm_page. */ #define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */ /* 0x0400 Unused */ #define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets */ #define UMA_ZONE_CACHESPREAD 0x1000 /* * Spread memory start locations across * all possible cache lines. May * require many virtually contiguous * backend pages and can fail early. */ #define UMA_ZONE_VTOSLAB 0x2000 /* Zone uses vtoslab for lookup. */ #define UMA_ZONE_NODUMP 0x4000 /* * Zone's pages will not be included in * mini-dumps. */ #define UMA_ZONE_PCPU 0x8000 /* * Allocates mp_maxid + 1 slabs sized to * sizeof(struct pcpu). */ /* * These flags are shared between the keg and zone. In zones wishing to add * new kegs these flags must be compatible. Some are determined based on * physical parameters of the request and may not be provided by the consumer. */ #define UMA_ZONE_INHERIT \ (UMA_ZONE_OFFPAGE | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \ UMA_ZONE_HASH | UMA_ZONE_VTOSLAB | UMA_ZONE_PCPU) /* Definitions for align */ #define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */ #define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */ #define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */ #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */ #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */ #define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */ #define UMA_ALIGNOF(type) (_Alignof(type) - 1) /* Alignment fit for 'type' */ /* * Destroys an empty uma zone. If the zone is not empty uma complains loudly. * * Arguments: * zone The zone we want to destroy. * */ void uma_zdestroy(uma_zone_t zone); /* * Allocates an item out of a zone * * Arguments: * zone The zone we are allocating from * arg This data is passed to the ctor function * flags See sys/malloc.h for available flags. * * Returns: * A non-null pointer to an initialized element from the zone is * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer * may be returned if the zone is empty or the ctor failed. */ void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags); /* * Allocates an item out of a zone without supplying an argument * * This is just a wrapper for uma_zalloc_arg for convenience. * */ static __inline void *uma_zalloc(uma_zone_t zone, int flags); static __inline void * uma_zalloc(uma_zone_t zone, int flags) { return uma_zalloc_arg(zone, NULL, flags); } /* * Frees an item back into the specified zone. * * Arguments: * zone The zone the item was originally allocated out of. * item The memory to be freed. * arg Argument passed to the destructor * * Returns: * Nothing. */ void uma_zfree_arg(uma_zone_t zone, void *item, void *arg); /* * Frees an item back to a zone without supplying an argument * * This is just a wrapper for uma_zfree_arg for convenience. * */ static __inline void uma_zfree(uma_zone_t zone, void *item); static __inline void uma_zfree(uma_zone_t zone, void *item) { uma_zfree_arg(zone, item, NULL); } /* * Wait until the specified zone can allocate an item. */ void uma_zwait(uma_zone_t zone); /* * XXX The rest of the prototypes in this header are h0h0 magic for the VM. * If you think you need to use it for a normal zone you're probably incorrect. */ /* * Backend page supplier routines * * Arguments: * zone The zone that is requesting pages. * size The number of bytes being requested. * pflag Flags for these memory pages, see below. * wait Indicates our willingness to block. * * Returns: * A pointer to the allocated memory or NULL on failure. */ typedef void *(*uma_alloc)(uma_zone_t zone, vm_size_t size, uint8_t *pflag, int wait); /* * Backend page free routines * * Arguments: * item A pointer to the previously allocated pages. * size The original size of the allocation. * pflag The flags for the slab. See UMA_SLAB_* below. * * Returns: * None */ typedef void (*uma_free)(void *item, vm_size_t size, uint8_t pflag); /* * Sets up the uma allocator. (Called by vm_mem_init) * * Arguments: * bootmem A pointer to memory used to bootstrap the system. * * Returns: * Nothing * * Discussion: * This memory is used for zones which allocate things before the * backend page supplier can give us pages. It should be * UMA_SLAB_SIZE * boot_pages bytes. (see uma_int.h) * */ void uma_startup(void *bootmem, int boot_pages); /* * Finishes starting up the allocator. This should * be called when kva is ready for normal allocs. * * Arguments: * None * * Returns: * Nothing * * Discussion: * uma_startup2 is called by kmeminit() to enable us of uma for malloc. */ void uma_startup2(void); /* * Reclaims unused memory for all zones * * Arguments: * None * Returns: * None * * This should only be called by the page out daemon. */ void uma_reclaim(void); /* * Sets the alignment mask to be used for all zones requesting cache * alignment. Should be called by MD boot code prior to starting VM/UMA. * * Arguments: * align The alignment mask * * Returns: * Nothing */ void uma_set_align(int align); /* * Set a reserved number of items to hold for M_USE_RESERVE allocations. All * other requests must allocate new backing pages. */ void uma_zone_reserve(uma_zone_t zone, int nitems); /* * Reserves the maximum KVA space required by the zone and configures the zone * to use a VM_ALLOC_NOOBJ-based backend allocator. * * Arguments: * zone The zone to update. * nitems The upper limit on the number of items that can be allocated. * * Returns: * 0 if KVA space can not be allocated * 1 if successful * * Discussion: * When the machine supports a direct map and the zone's items are smaller * than a page, the zone will use the direct map instead of allocating KVA * space. */ int uma_zone_reserve_kva(uma_zone_t zone, int nitems); /* * Sets a high limit on the number of items allowed in a zone * * Arguments: * zone The zone to limit * nitems The requested upper limit on the number of items allowed * * Returns: * int The effective value of nitems after rounding up based on page size */ int uma_zone_set_max(uma_zone_t zone, int nitems); /* * Obtains the effective limit on the number of items in a zone * * Arguments: * zone The zone to obtain the effective limit from * * Return: * 0 No limit * int The effective limit of the zone */ int uma_zone_get_max(uma_zone_t zone); /* * Sets a warning to be printed when limit is reached * * Arguments: * zone The zone we will warn about * warning Warning content * * Returns: * Nothing */ void uma_zone_set_warning(uma_zone_t zone, const char *warning); /* * Sets a function to run when limit is reached * * Arguments: * zone The zone to which this applies * fx The function ro run * * Returns: * Nothing */ typedef void (*uma_maxaction_t)(uma_zone_t, int); void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t); /* * Obtains the approximate current number of items allocated from a zone * * Arguments: * zone The zone to obtain the current allocation count from * * Return: * int The approximate current number of items allocated from the zone */ int uma_zone_get_cur(uma_zone_t zone); /* * The following two routines (uma_zone_set_init/fini) * are used to set the backend init/fini pair which acts on an * object as it becomes allocated and is placed in a slab within * the specified zone's backing keg. These should probably not * be changed once allocations have already begun, but only be set * immediately upon zone creation. */ void uma_zone_set_init(uma_zone_t zone, uma_init uminit); void uma_zone_set_fini(uma_zone_t zone, uma_fini fini); /* * The following two routines (uma_zone_set_zinit/zfini) are * used to set the zinit/zfini pair which acts on an object as * it passes from the backing Keg's slab cache to the * specified Zone's bucket cache. These should probably not * be changed once allocations have already begun, but only be set * immediately upon zone creation. */ void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit); void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini); /* * Replaces the standard backend allocator for this zone. * * Arguments: * zone The zone whose backend allocator is being changed. * allocf A pointer to the allocation function * * Returns: * Nothing * * Discussion: * This could be used to implement pageable allocation, or perhaps * even DMA allocators if used in conjunction with the OFFPAGE * zone flag. */ void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf); /* * Used for freeing memory provided by the allocf above * * Arguments: * zone The zone that intends to use this free routine. * freef The page freeing routine. * * Returns: * Nothing */ void uma_zone_set_freef(uma_zone_t zone, uma_free freef); /* * These flags are setable in the allocf and visible in the freef. */ #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */ #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kernel_map */ #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */ #define UMA_SLAB_OFFP 0x10 /* Slab is managed separately */ #define UMA_SLAB_MALLOC 0x20 /* Slab is a large malloc slab */ /* 0x02, 0x40 and 0x80 are available */ /* * Used to pre-fill a zone with some number of items * * Arguments: * zone The zone to fill * itemcnt The number of items to reserve * * Returns: * Nothing * * NOTE: This is blocking and should only be done at startup */ void uma_prealloc(uma_zone_t zone, int itemcnt); /* * Used to determine if a fixed-size zone is exhausted. * * Arguments: * zone The zone to check * * Returns: * Non-zero if zone is exhausted. */ int uma_zone_exhausted(uma_zone_t zone); int uma_zone_exhausted_nolock(uma_zone_t zone); /* * Common UMA_ZONE_PCPU zones. */ extern uma_zone_t pcpu_zone_64; extern uma_zone_t pcpu_zone_ptr; /* * Exported statistics structures to be used by user space monitoring tools. * Statistics stream consists of a uma_stream_header, followed by a series of * alternative uma_type_header and uma_type_stat structures. */ #define UMA_STREAM_VERSION 0x00000001 struct uma_stream_header { uint32_t ush_version; /* Stream format version. */ uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */ uint32_t ush_count; /* Number of records. */ uint32_t _ush_pad; /* Pad/reserved field. */ }; #define UTH_MAX_NAME 32 #define UTH_ZONE_SECONDARY 0x00000001 struct uma_type_header { /* * Static per-zone data, some extracted from the supporting keg. */ char uth_name[UTH_MAX_NAME]; uint32_t uth_align; /* Keg: alignment. */ uint32_t uth_size; /* Keg: requested size of item. */ uint32_t uth_rsize; /* Keg: real size of item. */ uint32_t uth_maxpages; /* Keg: maximum number of pages. */ uint32_t uth_limit; /* Keg: max items to allocate. */ /* * Current dynamic zone/keg-derived statistics. */ uint32_t uth_pages; /* Keg: pages allocated. */ uint32_t uth_keg_free; /* Keg: items free. */ uint32_t uth_zone_free; /* Zone: items free. */ uint32_t uth_bucketsize; /* Zone: desired bucket size. */ uint32_t uth_zone_flags; /* Zone: flags. */ uint64_t uth_allocs; /* Zone: number of allocations. */ uint64_t uth_frees; /* Zone: number of frees. */ uint64_t uth_fails; /* Zone: number of alloc failures. */ uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */ uint64_t _uth_reserved1[2]; /* Reserved. */ }; struct uma_percpu_stat { uint64_t ups_allocs; /* Cache: number of allocations. */ uint64_t ups_frees; /* Cache: number of frees. */ uint64_t ups_cache_free; /* Cache: free items in cache. */ uint64_t _ups_reserved[5]; /* Reserved. */ }; void uma_reclaim_wakeup(void); void uma_reclaim_worker(void *); +unsigned long uma_limit(void); + +/* Return the amount of memory managed by UMA. */ +unsigned long uma_size(void); + +/* Return the amount of memory remaining. May be negative. */ +long uma_avail(void); + #endif /* _VM_UMA_H_ */ Index: head/sys/vm/uma_core.c =================================================================== --- head/sys/vm/uma_core.c (revision 327484) +++ head/sys/vm/uma_core.c (revision 327485) @@ -1,3679 +1,3687 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002-2005, 2009, 2013 Jeffrey Roberson * Copyright (c) 2004, 2005 Bosko Milekic * Copyright (c) 2004-2006 Robert N. M. Watson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* * uma_core.c Implementation of the Universal Memory allocator * * This allocator is intended to replace the multitude of similar object caches * in the standard FreeBSD kernel. The intent is to be flexible as well as * efficient. A primary design goal is to return unused memory to the rest of * the system. This will make the system as a whole more flexible due to the * ability to move memory to subsystems which most need it instead of leaving * pools of reserved memory unused. * * The basic ideas stem from similar slab/zone based allocators whose algorithms * are well known. * */ /* * TODO: * - Improve memory usage for large allocations * - Investigate cache size adjustments */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_param.h" #include "opt_vm.h" #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DEBUG_MEMGUARD #include #endif /* * This is the zone and keg from which all zones are spawned. The idea is that * even the zone & keg heads are allocated from the allocator, so we use the * bss section to bootstrap us. */ static struct uma_keg masterkeg; static struct uma_zone masterzone_k; static struct uma_zone masterzone_z; static uma_zone_t kegs = &masterzone_k; static uma_zone_t zones = &masterzone_z; /* This is the zone from which all of uma_slab_t's are allocated. */ static uma_zone_t slabzone; /* * The initial hash tables come out of this zone so they can be allocated * prior to malloc coming up. */ static uma_zone_t hashzone; /* The boot-time adjusted value for cache line alignment. */ int uma_align_cache = 64 - 1; static MALLOC_DEFINE(M_UMAHASH, "UMAHash", "UMA Hash Buckets"); /* * Are we allowed to allocate buckets? */ static int bucketdisable = 1; /* Linked list of all kegs in the system */ static LIST_HEAD(,uma_keg) uma_kegs = LIST_HEAD_INITIALIZER(uma_kegs); /* Linked list of all cache-only zones in the system */ static LIST_HEAD(,uma_zone) uma_cachezones = LIST_HEAD_INITIALIZER(uma_cachezones); /* This RW lock protects the keg list */ static struct rwlock_padalign __exclusive_cache_line uma_rwlock; /* * Pointer and counter to pool of pages, that is preallocated at * startup to bootstrap UMA. Early zones continue to use the pool * until it is depleted, so allocations may happen after boot, thus * we need a mutex to protect it. */ static char *bootmem; static int boot_pages; static struct mtx uma_boot_pages_mtx; static struct sx uma_drain_lock; /* kmem soft limit. */ -static unsigned long uma_kmem_limit; +static unsigned long uma_kmem_limit = LONG_MAX; static volatile unsigned long uma_kmem_total; /* Is the VM done starting up? */ static int booted = 0; #define UMA_STARTUP 1 #define UMA_STARTUP2 2 /* * This is the handle used to schedule events that need to happen * outside of the allocation fast path. */ static struct callout uma_callout; #define UMA_TIMEOUT 20 /* Seconds for callout interval. */ /* * This structure is passed as the zone ctor arg so that I don't have to create * a special allocation function just for zones. */ struct uma_zctor_args { const char *name; size_t size; uma_ctor ctor; uma_dtor dtor; uma_init uminit; uma_fini fini; uma_import import; uma_release release; void *arg; uma_keg_t keg; int align; uint32_t flags; }; struct uma_kctor_args { uma_zone_t zone; size_t size; uma_init uminit; uma_fini fini; int align; uint32_t flags; }; struct uma_bucket_zone { uma_zone_t ubz_zone; char *ubz_name; int ubz_entries; /* Number of items it can hold. */ int ubz_maxsize; /* Maximum allocation size per-item. */ }; /* * Compute the actual number of bucket entries to pack them in power * of two sizes for more efficient space utilization. */ #define BUCKET_SIZE(n) \ (((sizeof(void *) * (n)) - sizeof(struct uma_bucket)) / sizeof(void *)) #define BUCKET_MAX BUCKET_SIZE(256) struct uma_bucket_zone bucket_zones[] = { { NULL, "4 Bucket", BUCKET_SIZE(4), 4096 }, { NULL, "6 Bucket", BUCKET_SIZE(6), 3072 }, { NULL, "8 Bucket", BUCKET_SIZE(8), 2048 }, { NULL, "12 Bucket", BUCKET_SIZE(12), 1536 }, { NULL, "16 Bucket", BUCKET_SIZE(16), 1024 }, { NULL, "32 Bucket", BUCKET_SIZE(32), 512 }, { NULL, "64 Bucket", BUCKET_SIZE(64), 256 }, { NULL, "128 Bucket", BUCKET_SIZE(128), 128 }, { NULL, "256 Bucket", BUCKET_SIZE(256), 64 }, { NULL, NULL, 0} }; /* * Flags and enumerations to be passed to internal functions. */ enum zfreeskip { SKIP_NONE = 0, SKIP_DTOR, SKIP_FINI }; /* Prototypes.. */ static void *noobj_alloc(uma_zone_t, vm_size_t, uint8_t *, int); static void *page_alloc(uma_zone_t, vm_size_t, uint8_t *, int); static void *startup_alloc(uma_zone_t, vm_size_t, uint8_t *, int); static void page_free(void *, vm_size_t, uint8_t); static uma_slab_t keg_alloc_slab(uma_keg_t, uma_zone_t, int); static void cache_drain(uma_zone_t); static void bucket_drain(uma_zone_t, uma_bucket_t); static void bucket_cache_drain(uma_zone_t zone); static int keg_ctor(void *, int, void *, int); static void keg_dtor(void *, int, void *); static int zone_ctor(void *, int, void *, int); static void zone_dtor(void *, int, void *); static int zero_init(void *, int, int); static void keg_small_init(uma_keg_t keg); static void keg_large_init(uma_keg_t keg); static void zone_foreach(void (*zfunc)(uma_zone_t)); static void zone_timeout(uma_zone_t zone); static int hash_alloc(struct uma_hash *); static int hash_expand(struct uma_hash *, struct uma_hash *); static void hash_free(struct uma_hash *hash); static void uma_timeout(void *); static void uma_startup3(void); static void *zone_alloc_item(uma_zone_t, void *, int); static void zone_free_item(uma_zone_t, void *, void *, enum zfreeskip); static void bucket_enable(void); static void bucket_init(void); static uma_bucket_t bucket_alloc(uma_zone_t zone, void *, int); static void bucket_free(uma_zone_t zone, uma_bucket_t, void *); static void bucket_zone_drain(void); static uma_bucket_t zone_alloc_bucket(uma_zone_t zone, void *, int flags); static uma_slab_t zone_fetch_slab(uma_zone_t zone, uma_keg_t last, int flags); static uma_slab_t zone_fetch_slab_multi(uma_zone_t zone, uma_keg_t last, int flags); static void *slab_alloc_item(uma_keg_t keg, uma_slab_t slab); static void slab_free_item(uma_keg_t keg, uma_slab_t slab, void *item); static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini, int align, uint32_t flags); static int zone_import(uma_zone_t zone, void **bucket, int max, int flags); static void zone_release(uma_zone_t zone, void **bucket, int cnt); static void uma_zero_item(void *item, uma_zone_t zone); void uma_print_zone(uma_zone_t); void uma_print_stats(void); static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS); static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS); #ifdef INVARIANTS static void uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item); static void uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item); #endif SYSINIT(uma_startup3, SI_SUB_VM_CONF, SI_ORDER_SECOND, uma_startup3, NULL); SYSCTL_PROC(_vm, OID_AUTO, zone_count, CTLFLAG_RD|CTLTYPE_INT, 0, 0, sysctl_vm_zone_count, "I", "Number of UMA zones"); SYSCTL_PROC(_vm, OID_AUTO, zone_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 0, 0, sysctl_vm_zone_stats, "s,struct uma_type_header", "Zone Stats"); static int zone_warnings = 1; SYSCTL_INT(_vm, OID_AUTO, zone_warnings, CTLFLAG_RWTUN, &zone_warnings, 0, "Warn when UMA zones becomes full"); /* Adjust bytes under management by UMA. */ static inline void uma_total_dec(unsigned long size) { atomic_subtract_long(&uma_kmem_total, size); } static inline void uma_total_inc(unsigned long size) { if (atomic_fetchadd_long(&uma_kmem_total, size) > uma_kmem_limit) uma_reclaim_wakeup(); } /* * This routine checks to see whether or not it's safe to enable buckets. */ static void bucket_enable(void) { bucketdisable = vm_page_count_min(); } /* * Initialize bucket_zones, the array of zones of buckets of various sizes. * * For each zone, calculate the memory required for each bucket, consisting * of the header and an array of pointers. */ static void bucket_init(void) { struct uma_bucket_zone *ubz; int size; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) { size = roundup(sizeof(struct uma_bucket), sizeof(void *)); size += sizeof(void *) * ubz->ubz_entries; ubz->ubz_zone = uma_zcreate(ubz->ubz_name, size, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_MTXCLASS | UMA_ZFLAG_BUCKET); } } /* * Given a desired number of entries for a bucket, return the zone from which * to allocate the bucket. */ static struct uma_bucket_zone * bucket_zone_lookup(int entries) { struct uma_bucket_zone *ubz; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) if (ubz->ubz_entries >= entries) return (ubz); ubz--; return (ubz); } static int bucket_select(int size) { struct uma_bucket_zone *ubz; ubz = &bucket_zones[0]; if (size > ubz->ubz_maxsize) return MAX((ubz->ubz_maxsize * ubz->ubz_entries) / size, 1); for (; ubz->ubz_entries != 0; ubz++) if (ubz->ubz_maxsize < size) break; ubz--; return (ubz->ubz_entries); } static uma_bucket_t bucket_alloc(uma_zone_t zone, void *udata, int flags) { struct uma_bucket_zone *ubz; uma_bucket_t bucket; /* * This is to stop us from allocating per cpu buckets while we're * running out of vm.boot_pages. Otherwise, we would exhaust the * boot pages. This also prevents us from allocating buckets in * low memory situations. */ if (bucketdisable) return (NULL); /* * To limit bucket recursion we store the original zone flags * in a cookie passed via zalloc_arg/zfree_arg. This allows the * NOVM flag to persist even through deep recursions. We also * store ZFLAG_BUCKET once we have recursed attempting to allocate * a bucket for a bucket zone so we do not allow infinite bucket * recursion. This cookie will even persist to frees of unused * buckets via the allocation path or bucket allocations in the * free path. */ if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0) udata = (void *)(uintptr_t)zone->uz_flags; else { if ((uintptr_t)udata & UMA_ZFLAG_BUCKET) return (NULL); udata = (void *)((uintptr_t)udata | UMA_ZFLAG_BUCKET); } if ((uintptr_t)udata & UMA_ZFLAG_CACHEONLY) flags |= M_NOVM; ubz = bucket_zone_lookup(zone->uz_count); if (ubz->ubz_zone == zone && (ubz + 1)->ubz_entries != 0) ubz++; bucket = uma_zalloc_arg(ubz->ubz_zone, udata, flags); if (bucket) { #ifdef INVARIANTS bzero(bucket->ub_bucket, sizeof(void *) * ubz->ubz_entries); #endif bucket->ub_cnt = 0; bucket->ub_entries = ubz->ubz_entries; } return (bucket); } static void bucket_free(uma_zone_t zone, uma_bucket_t bucket, void *udata) { struct uma_bucket_zone *ubz; KASSERT(bucket->ub_cnt == 0, ("bucket_free: Freeing a non free bucket.")); if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0) udata = (void *)(uintptr_t)zone->uz_flags; ubz = bucket_zone_lookup(bucket->ub_entries); uma_zfree_arg(ubz->ubz_zone, bucket, udata); } static void bucket_zone_drain(void) { struct uma_bucket_zone *ubz; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) zone_drain(ubz->ubz_zone); } static void zone_log_warning(uma_zone_t zone) { static const struct timeval warninterval = { 300, 0 }; if (!zone_warnings || zone->uz_warning == NULL) return; if (ratecheck(&zone->uz_ratecheck, &warninterval)) printf("[zone: %s] %s\n", zone->uz_name, zone->uz_warning); } static inline void zone_maxaction(uma_zone_t zone) { if (zone->uz_maxaction.ta_func != NULL) taskqueue_enqueue(taskqueue_thread, &zone->uz_maxaction); } static void zone_foreach_keg(uma_zone_t zone, void (*kegfn)(uma_keg_t)) { uma_klink_t klink; LIST_FOREACH(klink, &zone->uz_kegs, kl_link) kegfn(klink->kl_keg); } /* * Routine called by timeout which is used to fire off some time interval * based calculations. (stats, hash size, etc.) * * Arguments: * arg Unused * * Returns: * Nothing */ static void uma_timeout(void *unused) { bucket_enable(); zone_foreach(zone_timeout); /* Reschedule this event */ callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); } /* * Routine to perform timeout driven calculations. This expands the * hashes and does per cpu statistics aggregation. * * Returns nothing. */ static void keg_timeout(uma_keg_t keg) { KEG_LOCK(keg); /* * Expand the keg hash table. * * This is done if the number of slabs is larger than the hash size. * What I'm trying to do here is completely reduce collisions. This * may be a little aggressive. Should I allow for two collisions max? */ if (keg->uk_flags & UMA_ZONE_HASH && keg->uk_pages / keg->uk_ppera >= keg->uk_hash.uh_hashsize) { struct uma_hash newhash; struct uma_hash oldhash; int ret; /* * This is so involved because allocating and freeing * while the keg lock is held will lead to deadlock. * I have to do everything in stages and check for * races. */ newhash = keg->uk_hash; KEG_UNLOCK(keg); ret = hash_alloc(&newhash); KEG_LOCK(keg); if (ret) { if (hash_expand(&keg->uk_hash, &newhash)) { oldhash = keg->uk_hash; keg->uk_hash = newhash; } else oldhash = newhash; KEG_UNLOCK(keg); hash_free(&oldhash); return; } } KEG_UNLOCK(keg); } static void zone_timeout(uma_zone_t zone) { zone_foreach_keg(zone, &keg_timeout); } /* * Allocate and zero fill the next sized hash table from the appropriate * backing store. * * Arguments: * hash A new hash structure with the old hash size in uh_hashsize * * Returns: * 1 on success and 0 on failure. */ static int hash_alloc(struct uma_hash *hash) { int oldsize; int alloc; oldsize = hash->uh_hashsize; /* We're just going to go to a power of two greater */ if (oldsize) { hash->uh_hashsize = oldsize * 2; alloc = sizeof(hash->uh_slab_hash[0]) * hash->uh_hashsize; hash->uh_slab_hash = (struct slabhead *)malloc(alloc, M_UMAHASH, M_NOWAIT); } else { alloc = sizeof(hash->uh_slab_hash[0]) * UMA_HASH_SIZE_INIT; hash->uh_slab_hash = zone_alloc_item(hashzone, NULL, M_WAITOK); hash->uh_hashsize = UMA_HASH_SIZE_INIT; } if (hash->uh_slab_hash) { bzero(hash->uh_slab_hash, alloc); hash->uh_hashmask = hash->uh_hashsize - 1; return (1); } return (0); } /* * Expands the hash table for HASH zones. This is done from zone_timeout * to reduce collisions. This must not be done in the regular allocation * path, otherwise, we can recurse on the vm while allocating pages. * * Arguments: * oldhash The hash you want to expand * newhash The hash structure for the new table * * Returns: * Nothing * * Discussion: */ static int hash_expand(struct uma_hash *oldhash, struct uma_hash *newhash) { uma_slab_t slab; int hval; int i; if (!newhash->uh_slab_hash) return (0); if (oldhash->uh_hashsize >= newhash->uh_hashsize) return (0); /* * I need to investigate hash algorithms for resizing without a * full rehash. */ for (i = 0; i < oldhash->uh_hashsize; i++) while (!SLIST_EMPTY(&oldhash->uh_slab_hash[i])) { slab = SLIST_FIRST(&oldhash->uh_slab_hash[i]); SLIST_REMOVE_HEAD(&oldhash->uh_slab_hash[i], us_hlink); hval = UMA_HASH(newhash, slab->us_data); SLIST_INSERT_HEAD(&newhash->uh_slab_hash[hval], slab, us_hlink); } return (1); } /* * Free the hash bucket to the appropriate backing store. * * Arguments: * slab_hash The hash bucket we're freeing * hashsize The number of entries in that hash bucket * * Returns: * Nothing */ static void hash_free(struct uma_hash *hash) { if (hash->uh_slab_hash == NULL) return; if (hash->uh_hashsize == UMA_HASH_SIZE_INIT) zone_free_item(hashzone, hash->uh_slab_hash, NULL, SKIP_NONE); else free(hash->uh_slab_hash, M_UMAHASH); } /* * Frees all outstanding items in a bucket * * Arguments: * zone The zone to free to, must be unlocked. * bucket The free/alloc bucket with items, cpu queue must be locked. * * Returns: * Nothing */ static void bucket_drain(uma_zone_t zone, uma_bucket_t bucket) { int i; if (bucket == NULL) return; if (zone->uz_fini) for (i = 0; i < bucket->ub_cnt; i++) zone->uz_fini(bucket->ub_bucket[i], zone->uz_size); zone->uz_release(zone->uz_arg, bucket->ub_bucket, bucket->ub_cnt); bucket->ub_cnt = 0; } /* * Drains the per cpu caches for a zone. * * NOTE: This may only be called while the zone is being turn down, and not * during normal operation. This is necessary in order that we do not have * to migrate CPUs to drain the per-CPU caches. * * Arguments: * zone The zone to drain, must be unlocked. * * Returns: * Nothing */ static void cache_drain(uma_zone_t zone) { uma_cache_t cache; int cpu; /* * XXX: It is safe to not lock the per-CPU caches, because we're * tearing down the zone anyway. I.e., there will be no further use * of the caches at this point. * * XXX: It would good to be able to assert that the zone is being * torn down to prevent improper use of cache_drain(). * * XXX: We lock the zone before passing into bucket_cache_drain() as * it is used elsewhere. Should the tear-down path be made special * there in some form? */ CPU_FOREACH(cpu) { cache = &zone->uz_cpu[cpu]; bucket_drain(zone, cache->uc_allocbucket); bucket_drain(zone, cache->uc_freebucket); if (cache->uc_allocbucket != NULL) bucket_free(zone, cache->uc_allocbucket, NULL); if (cache->uc_freebucket != NULL) bucket_free(zone, cache->uc_freebucket, NULL); cache->uc_allocbucket = cache->uc_freebucket = NULL; } ZONE_LOCK(zone); bucket_cache_drain(zone); ZONE_UNLOCK(zone); } static void cache_shrink(uma_zone_t zone) { if (zone->uz_flags & UMA_ZFLAG_INTERNAL) return; ZONE_LOCK(zone); zone->uz_count = (zone->uz_count_min + zone->uz_count) / 2; ZONE_UNLOCK(zone); } static void cache_drain_safe_cpu(uma_zone_t zone) { uma_cache_t cache; uma_bucket_t b1, b2; if (zone->uz_flags & UMA_ZFLAG_INTERNAL) return; b1 = b2 = NULL; ZONE_LOCK(zone); critical_enter(); cache = &zone->uz_cpu[curcpu]; if (cache->uc_allocbucket) { if (cache->uc_allocbucket->ub_cnt != 0) LIST_INSERT_HEAD(&zone->uz_buckets, cache->uc_allocbucket, ub_link); else b1 = cache->uc_allocbucket; cache->uc_allocbucket = NULL; } if (cache->uc_freebucket) { if (cache->uc_freebucket->ub_cnt != 0) LIST_INSERT_HEAD(&zone->uz_buckets, cache->uc_freebucket, ub_link); else b2 = cache->uc_freebucket; cache->uc_freebucket = NULL; } critical_exit(); ZONE_UNLOCK(zone); if (b1) bucket_free(zone, b1, NULL); if (b2) bucket_free(zone, b2, NULL); } /* * Safely drain per-CPU caches of a zone(s) to alloc bucket. * This is an expensive call because it needs to bind to all CPUs * one by one and enter a critical section on each of them in order * to safely access their cache buckets. * Zone lock must not be held on call this function. */ static void cache_drain_safe(uma_zone_t zone) { int cpu; /* * Polite bucket sizes shrinking was not enouth, shrink aggressively. */ if (zone) cache_shrink(zone); else zone_foreach(cache_shrink); CPU_FOREACH(cpu) { thread_lock(curthread); sched_bind(curthread, cpu); thread_unlock(curthread); if (zone) cache_drain_safe_cpu(zone); else zone_foreach(cache_drain_safe_cpu); } thread_lock(curthread); sched_unbind(curthread); thread_unlock(curthread); } /* * Drain the cached buckets from a zone. Expects a locked zone on entry. */ static void bucket_cache_drain(uma_zone_t zone) { uma_bucket_t bucket; /* * Drain the bucket queues and free the buckets, we just keep two per * cpu (alloc/free). */ while ((bucket = LIST_FIRST(&zone->uz_buckets)) != NULL) { LIST_REMOVE(bucket, ub_link); ZONE_UNLOCK(zone); bucket_drain(zone, bucket); bucket_free(zone, bucket, NULL); ZONE_LOCK(zone); } /* * Shrink further bucket sizes. Price of single zone lock collision * is probably lower then price of global cache drain. */ if (zone->uz_count > zone->uz_count_min) zone->uz_count--; } static void keg_free_slab(uma_keg_t keg, uma_slab_t slab, int start) { uint8_t *mem; int i; uint8_t flags; CTR4(KTR_UMA, "keg_free_slab keg %s(%p) slab %p, returning %d bytes", keg->uk_name, keg, slab, PAGE_SIZE * keg->uk_ppera); mem = slab->us_data; flags = slab->us_flags; i = start; if (keg->uk_fini != NULL) { for (i--; i > -1; i--) keg->uk_fini(slab->us_data + (keg->uk_rsize * i), keg->uk_size); } if (keg->uk_flags & UMA_ZONE_OFFPAGE) zone_free_item(keg->uk_slabzone, slab, NULL, SKIP_NONE); keg->uk_freef(mem, PAGE_SIZE * keg->uk_ppera, flags); uma_total_dec(PAGE_SIZE * keg->uk_ppera); } /* * Frees pages from a keg back to the system. This is done on demand from * the pageout daemon. * * Returns nothing. */ static void keg_drain(uma_keg_t keg) { struct slabhead freeslabs = { 0 }; uma_slab_t slab, tmp; /* * We don't want to take pages from statically allocated kegs at this * time */ if (keg->uk_flags & UMA_ZONE_NOFREE || keg->uk_freef == NULL) return; CTR3(KTR_UMA, "keg_drain %s(%p) free items: %u", keg->uk_name, keg, keg->uk_free); KEG_LOCK(keg); if (keg->uk_free == 0) goto finished; LIST_FOREACH_SAFE(slab, &keg->uk_free_slab, us_link, tmp) { /* We have nowhere to free these to. */ if (slab->us_flags & UMA_SLAB_BOOT) continue; LIST_REMOVE(slab, us_link); keg->uk_pages -= keg->uk_ppera; keg->uk_free -= keg->uk_ipers; if (keg->uk_flags & UMA_ZONE_HASH) UMA_HASH_REMOVE(&keg->uk_hash, slab, slab->us_data); SLIST_INSERT_HEAD(&freeslabs, slab, us_hlink); } finished: KEG_UNLOCK(keg); while ((slab = SLIST_FIRST(&freeslabs)) != NULL) { SLIST_REMOVE(&freeslabs, slab, uma_slab, us_hlink); keg_free_slab(keg, slab, keg->uk_ipers); } } static void zone_drain_wait(uma_zone_t zone, int waitok) { /* * Set draining to interlock with zone_dtor() so we can release our * locks as we go. Only dtor() should do a WAITOK call since it * is the only call that knows the structure will still be available * when it wakes up. */ ZONE_LOCK(zone); while (zone->uz_flags & UMA_ZFLAG_DRAINING) { if (waitok == M_NOWAIT) goto out; msleep(zone, zone->uz_lockptr, PVM, "zonedrain", 1); } zone->uz_flags |= UMA_ZFLAG_DRAINING; bucket_cache_drain(zone); ZONE_UNLOCK(zone); /* * The DRAINING flag protects us from being freed while * we're running. Normally the uma_rwlock would protect us but we * must be able to release and acquire the right lock for each keg. */ zone_foreach_keg(zone, &keg_drain); ZONE_LOCK(zone); zone->uz_flags &= ~UMA_ZFLAG_DRAINING; wakeup(zone); out: ZONE_UNLOCK(zone); } void zone_drain(uma_zone_t zone) { zone_drain_wait(zone, M_NOWAIT); } /* * Allocate a new slab for a keg. This does not insert the slab onto a list. * * Arguments: * wait Shall we wait? * * Returns: * The slab that was allocated or NULL if there is no memory and the * caller specified M_NOWAIT. */ static uma_slab_t keg_alloc_slab(uma_keg_t keg, uma_zone_t zone, int wait) { uma_alloc allocf; uma_slab_t slab; unsigned long size; uint8_t *mem; uint8_t flags; int i; mtx_assert(&keg->uk_lock, MA_OWNED); slab = NULL; mem = NULL; allocf = keg->uk_allocf; KEG_UNLOCK(keg); size = keg->uk_ppera * PAGE_SIZE; if (keg->uk_flags & UMA_ZONE_OFFPAGE) { slab = zone_alloc_item(keg->uk_slabzone, NULL, wait); if (slab == NULL) goto out; } /* * This reproduces the old vm_zone behavior of zero filling pages the * first time they are added to a zone. * * Malloced items are zeroed in uma_zalloc. */ if ((keg->uk_flags & UMA_ZONE_MALLOC) == 0) wait |= M_ZERO; else wait &= ~M_ZERO; if (keg->uk_flags & UMA_ZONE_NODUMP) wait |= M_NODUMP; /* zone is passed for legacy reasons. */ mem = allocf(zone, size, &flags, wait); if (mem == NULL) { if (keg->uk_flags & UMA_ZONE_OFFPAGE) zone_free_item(keg->uk_slabzone, slab, NULL, SKIP_NONE); slab = NULL; goto out; } uma_total_inc(size); /* Point the slab into the allocated memory */ if (!(keg->uk_flags & UMA_ZONE_OFFPAGE)) slab = (uma_slab_t )(mem + keg->uk_pgoff); if (keg->uk_flags & UMA_ZONE_VTOSLAB) for (i = 0; i < keg->uk_ppera; i++) vsetslab((vm_offset_t)mem + (i * PAGE_SIZE), slab); slab->us_keg = keg; slab->us_data = mem; slab->us_freecount = keg->uk_ipers; slab->us_flags = flags; BIT_FILL(SLAB_SETSIZE, &slab->us_free); #ifdef INVARIANTS BIT_ZERO(SLAB_SETSIZE, &slab->us_debugfree); #endif if (keg->uk_init != NULL) { for (i = 0; i < keg->uk_ipers; i++) if (keg->uk_init(slab->us_data + (keg->uk_rsize * i), keg->uk_size, wait) != 0) break; if (i != keg->uk_ipers) { keg_free_slab(keg, slab, i); slab = NULL; goto out; } } out: KEG_LOCK(keg); CTR3(KTR_UMA, "keg_alloc_slab: allocated slab %p for %s(%p)", slab, keg->uk_name, keg); if (slab != NULL) { if (keg->uk_flags & UMA_ZONE_HASH) UMA_HASH_INSERT(&keg->uk_hash, slab, mem); keg->uk_pages += keg->uk_ppera; keg->uk_free += keg->uk_ipers; } return (slab); } /* * This function is intended to be used early on in place of page_alloc() so * that we may use the boot time page cache to satisfy allocations before * the VM is ready. */ static void * startup_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *pflag, int wait) { uma_keg_t keg; void *mem; int pages; keg = zone_first_keg(zone); pages = howmany(bytes, PAGE_SIZE); KASSERT(pages > 0, ("startup_alloc can't reserve 0 pages\n")); /* * Check our small startup cache to see if it has pages remaining. */ mtx_lock(&uma_boot_pages_mtx); if (pages <= boot_pages) { mem = bootmem; boot_pages -= pages; bootmem += pages * PAGE_SIZE; mtx_unlock(&uma_boot_pages_mtx); *pflag = UMA_SLAB_BOOT; return (mem); } mtx_unlock(&uma_boot_pages_mtx); if (booted < UMA_STARTUP2) panic("UMA: Increase vm.boot_pages"); /* * Now that we've booted reset these users to their real allocator. */ #ifdef UMA_MD_SMALL_ALLOC keg->uk_allocf = (keg->uk_ppera > 1) ? page_alloc : uma_small_alloc; #else keg->uk_allocf = page_alloc; #endif return keg->uk_allocf(zone, bytes, pflag, wait); } /* * Allocates a number of pages from the system * * Arguments: * bytes The number of bytes requested * wait Shall we wait? * * Returns: * A pointer to the alloced memory or possibly * NULL if M_NOWAIT is set. */ static void * page_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *pflag, int wait) { void *p; /* Returned page */ *pflag = UMA_SLAB_KERNEL; p = (void *) kmem_malloc(kernel_arena, bytes, wait); return (p); } /* * Allocates a number of pages from within an object * * Arguments: * bytes The number of bytes requested * wait Shall we wait? * * Returns: * A pointer to the alloced memory or possibly * NULL if M_NOWAIT is set. */ static void * noobj_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *flags, int wait) { TAILQ_HEAD(, vm_page) alloctail; u_long npages; vm_offset_t retkva, zkva; vm_page_t p, p_next; uma_keg_t keg; TAILQ_INIT(&alloctail); keg = zone_first_keg(zone); npages = howmany(bytes, PAGE_SIZE); while (npages > 0) { p = vm_page_alloc(NULL, 0, VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED | VM_ALLOC_NOOBJ | ((wait & M_WAITOK) != 0 ? VM_ALLOC_WAITOK : VM_ALLOC_NOWAIT)); if (p != NULL) { /* * Since the page does not belong to an object, its * listq is unused. */ TAILQ_INSERT_TAIL(&alloctail, p, listq); npages--; continue; } /* * Page allocation failed, free intermediate pages and * exit. */ TAILQ_FOREACH_SAFE(p, &alloctail, listq, p_next) { vm_page_unwire(p, PQ_NONE); vm_page_free(p); } return (NULL); } *flags = UMA_SLAB_PRIV; zkva = keg->uk_kva + atomic_fetchadd_long(&keg->uk_offset, round_page(bytes)); retkva = zkva; TAILQ_FOREACH(p, &alloctail, listq) { pmap_qenter(zkva, &p, 1); zkva += PAGE_SIZE; } return ((void *)retkva); } /* * Frees a number of pages to the system * * Arguments: * mem A pointer to the memory to be freed * size The size of the memory being freed * flags The original p->us_flags field * * Returns: * Nothing */ static void page_free(void *mem, vm_size_t size, uint8_t flags) { struct vmem *vmem; if (flags & UMA_SLAB_KERNEL) vmem = kernel_arena; else panic("UMA: page_free used with invalid flags %x", flags); kmem_free(vmem, (vm_offset_t)mem, size); } /* * Zero fill initializer * * Arguments/Returns follow uma_init specifications */ static int zero_init(void *mem, int size, int flags) { bzero(mem, size); return (0); } /* * Finish creating a small uma keg. This calculates ipers, and the keg size. * * Arguments * keg The zone we should initialize * * Returns * Nothing */ static void keg_small_init(uma_keg_t keg) { u_int rsize; u_int memused; u_int wastedspace; u_int shsize; u_int slabsize; if (keg->uk_flags & UMA_ZONE_PCPU) { u_int ncpus = (mp_maxid + 1) ? (mp_maxid + 1) : MAXCPU; slabsize = sizeof(struct pcpu); keg->uk_ppera = howmany(ncpus * sizeof(struct pcpu), PAGE_SIZE); } else { slabsize = UMA_SLAB_SIZE; keg->uk_ppera = 1; } /* * Calculate the size of each allocation (rsize) according to * alignment. If the requested size is smaller than we have * allocation bits for we round it up. */ rsize = keg->uk_size; if (rsize < slabsize / SLAB_SETSIZE) rsize = slabsize / SLAB_SETSIZE; if (rsize & keg->uk_align) rsize = (rsize & ~keg->uk_align) + (keg->uk_align + 1); keg->uk_rsize = rsize; KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0 || keg->uk_rsize < sizeof(struct pcpu), ("%s: size %u too large", __func__, keg->uk_rsize)); if (keg->uk_flags & UMA_ZONE_OFFPAGE) shsize = 0; else shsize = sizeof(struct uma_slab); keg->uk_ipers = (slabsize - shsize) / rsize; KASSERT(keg->uk_ipers > 0 && keg->uk_ipers <= SLAB_SETSIZE, ("%s: keg->uk_ipers %u", __func__, keg->uk_ipers)); memused = keg->uk_ipers * rsize + shsize; wastedspace = slabsize - memused; /* * We can't do OFFPAGE if we're internal or if we've been * asked to not go to the VM for buckets. If we do this we * may end up going to the VM for slabs which we do not * want to do if we're UMA_ZFLAG_CACHEONLY as a result * of UMA_ZONE_VM, which clearly forbids it. */ if ((keg->uk_flags & UMA_ZFLAG_INTERNAL) || (keg->uk_flags & UMA_ZFLAG_CACHEONLY)) return; /* * See if using an OFFPAGE slab will limit our waste. Only do * this if it permits more items per-slab. * * XXX We could try growing slabsize to limit max waste as well. * Historically this was not done because the VM could not * efficiently handle contiguous allocations. */ if ((wastedspace >= slabsize / UMA_MAX_WASTE) && (keg->uk_ipers < (slabsize / keg->uk_rsize))) { keg->uk_ipers = slabsize / keg->uk_rsize; KASSERT(keg->uk_ipers > 0 && keg->uk_ipers <= SLAB_SETSIZE, ("%s: keg->uk_ipers %u", __func__, keg->uk_ipers)); CTR6(KTR_UMA, "UMA decided we need offpage slab headers for " "keg: %s(%p), calculated wastedspace = %d, " "maximum wasted space allowed = %d, " "calculated ipers = %d, " "new wasted space = %d\n", keg->uk_name, keg, wastedspace, slabsize / UMA_MAX_WASTE, keg->uk_ipers, slabsize - keg->uk_ipers * keg->uk_rsize); keg->uk_flags |= UMA_ZONE_OFFPAGE; } if ((keg->uk_flags & UMA_ZONE_OFFPAGE) && (keg->uk_flags & UMA_ZONE_VTOSLAB) == 0) keg->uk_flags |= UMA_ZONE_HASH; } /* * Finish creating a large (> UMA_SLAB_SIZE) uma kegs. Just give in and do * OFFPAGE for now. When I can allow for more dynamic slab sizes this will be * more complicated. * * Arguments * keg The keg we should initialize * * Returns * Nothing */ static void keg_large_init(uma_keg_t keg) { u_int shsize; KASSERT(keg != NULL, ("Keg is null in keg_large_init")); KASSERT((keg->uk_flags & UMA_ZFLAG_CACHEONLY) == 0, ("keg_large_init: Cannot large-init a UMA_ZFLAG_CACHEONLY keg")); KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0, ("%s: Cannot large-init a UMA_ZONE_PCPU keg", __func__)); keg->uk_ppera = howmany(keg->uk_size, PAGE_SIZE); keg->uk_ipers = 1; keg->uk_rsize = keg->uk_size; /* Check whether we have enough space to not do OFFPAGE. */ if ((keg->uk_flags & UMA_ZONE_OFFPAGE) == 0) { shsize = sizeof(struct uma_slab); if (shsize & UMA_ALIGN_PTR) shsize = (shsize & ~UMA_ALIGN_PTR) + (UMA_ALIGN_PTR + 1); if (PAGE_SIZE * keg->uk_ppera - keg->uk_rsize < shsize) { /* * We can't do OFFPAGE if we're internal, in which case * we need an extra page per allocation to contain the * slab header. */ if ((keg->uk_flags & UMA_ZFLAG_INTERNAL) == 0) keg->uk_flags |= UMA_ZONE_OFFPAGE; else keg->uk_ppera++; } } if ((keg->uk_flags & UMA_ZONE_OFFPAGE) && (keg->uk_flags & UMA_ZONE_VTOSLAB) == 0) keg->uk_flags |= UMA_ZONE_HASH; } static void keg_cachespread_init(uma_keg_t keg) { int alignsize; int trailer; int pages; int rsize; KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0, ("%s: Cannot cachespread-init a UMA_ZONE_PCPU keg", __func__)); alignsize = keg->uk_align + 1; rsize = keg->uk_size; /* * We want one item to start on every align boundary in a page. To * do this we will span pages. We will also extend the item by the * size of align if it is an even multiple of align. Otherwise, it * would fall on the same boundary every time. */ if (rsize & keg->uk_align) rsize = (rsize & ~keg->uk_align) + alignsize; if ((rsize & alignsize) == 0) rsize += alignsize; trailer = rsize - keg->uk_size; pages = (rsize * (PAGE_SIZE / alignsize)) / PAGE_SIZE; pages = MIN(pages, (128 * 1024) / PAGE_SIZE); keg->uk_rsize = rsize; keg->uk_ppera = pages; keg->uk_ipers = ((pages * PAGE_SIZE) + trailer) / rsize; keg->uk_flags |= UMA_ZONE_OFFPAGE | UMA_ZONE_VTOSLAB; KASSERT(keg->uk_ipers <= SLAB_SETSIZE, ("%s: keg->uk_ipers too high(%d) increase max_ipers", __func__, keg->uk_ipers)); } /* * Keg header ctor. This initializes all fields, locks, etc. And inserts * the keg onto the global keg list. * * Arguments/Returns follow uma_ctor specifications * udata Actually uma_kctor_args */ static int keg_ctor(void *mem, int size, void *udata, int flags) { struct uma_kctor_args *arg = udata; uma_keg_t keg = mem; uma_zone_t zone; bzero(keg, size); keg->uk_size = arg->size; keg->uk_init = arg->uminit; keg->uk_fini = arg->fini; keg->uk_align = arg->align; keg->uk_free = 0; keg->uk_reserve = 0; keg->uk_pages = 0; keg->uk_flags = arg->flags; keg->uk_slabzone = NULL; /* * The master zone is passed to us at keg-creation time. */ zone = arg->zone; keg->uk_name = zone->uz_name; if (arg->flags & UMA_ZONE_VM) keg->uk_flags |= UMA_ZFLAG_CACHEONLY; if (arg->flags & UMA_ZONE_ZINIT) keg->uk_init = zero_init; if (arg->flags & UMA_ZONE_MALLOC) keg->uk_flags |= UMA_ZONE_VTOSLAB; if (arg->flags & UMA_ZONE_PCPU) #ifdef SMP keg->uk_flags |= UMA_ZONE_OFFPAGE; #else keg->uk_flags &= ~UMA_ZONE_PCPU; #endif if (keg->uk_flags & UMA_ZONE_CACHESPREAD) { keg_cachespread_init(keg); } else { if (keg->uk_size > (UMA_SLAB_SIZE - sizeof(struct uma_slab))) keg_large_init(keg); else keg_small_init(keg); } if (keg->uk_flags & UMA_ZONE_OFFPAGE) keg->uk_slabzone = slabzone; /* * If we haven't booted yet we need allocations to go through the * startup cache until the vm is ready. */ if (booted < UMA_STARTUP2) keg->uk_allocf = startup_alloc; #ifdef UMA_MD_SMALL_ALLOC else if (keg->uk_ppera == 1) keg->uk_allocf = uma_small_alloc; #endif else keg->uk_allocf = page_alloc; #ifdef UMA_MD_SMALL_ALLOC if (keg->uk_ppera == 1) keg->uk_freef = uma_small_free; else #endif keg->uk_freef = page_free; /* * Initialize keg's lock */ KEG_LOCK_INIT(keg, (arg->flags & UMA_ZONE_MTXCLASS)); /* * If we're putting the slab header in the actual page we need to * figure out where in each page it goes. This calculates a right * justified offset into the memory on an ALIGN_PTR boundary. */ if (!(keg->uk_flags & UMA_ZONE_OFFPAGE)) { u_int totsize; /* Size of the slab struct and free list */ totsize = sizeof(struct uma_slab); if (totsize & UMA_ALIGN_PTR) totsize = (totsize & ~UMA_ALIGN_PTR) + (UMA_ALIGN_PTR + 1); keg->uk_pgoff = (PAGE_SIZE * keg->uk_ppera) - totsize; /* * The only way the following is possible is if with our * UMA_ALIGN_PTR adjustments we are now bigger than * UMA_SLAB_SIZE. I haven't checked whether this is * mathematically possible for all cases, so we make * sure here anyway. */ totsize = keg->uk_pgoff + sizeof(struct uma_slab); if (totsize > PAGE_SIZE * keg->uk_ppera) { printf("zone %s ipers %d rsize %d size %d\n", zone->uz_name, keg->uk_ipers, keg->uk_rsize, keg->uk_size); panic("UMA slab won't fit."); } } if (keg->uk_flags & UMA_ZONE_HASH) hash_alloc(&keg->uk_hash); CTR5(KTR_UMA, "keg_ctor %p zone %s(%p) out %d free %d\n", keg, zone->uz_name, zone, (keg->uk_pages / keg->uk_ppera) * keg->uk_ipers - keg->uk_free, keg->uk_free); LIST_INSERT_HEAD(&keg->uk_zones, zone, uz_link); rw_wlock(&uma_rwlock); LIST_INSERT_HEAD(&uma_kegs, keg, uk_link); rw_wunlock(&uma_rwlock); return (0); } /* * Zone header ctor. This initializes all fields, locks, etc. * * Arguments/Returns follow uma_ctor specifications * udata Actually uma_zctor_args */ static int zone_ctor(void *mem, int size, void *udata, int flags) { struct uma_zctor_args *arg = udata; uma_zone_t zone = mem; uma_zone_t z; uma_keg_t keg; bzero(zone, size); zone->uz_name = arg->name; zone->uz_ctor = arg->ctor; zone->uz_dtor = arg->dtor; zone->uz_slab = zone_fetch_slab; zone->uz_init = NULL; zone->uz_fini = NULL; zone->uz_allocs = 0; zone->uz_frees = 0; zone->uz_fails = 0; zone->uz_sleeps = 0; zone->uz_count = 0; zone->uz_count_min = 0; zone->uz_flags = 0; zone->uz_warning = NULL; timevalclear(&zone->uz_ratecheck); keg = arg->keg; ZONE_LOCK_INIT(zone, (arg->flags & UMA_ZONE_MTXCLASS)); /* * This is a pure cache zone, no kegs. */ if (arg->import) { if (arg->flags & UMA_ZONE_VM) arg->flags |= UMA_ZFLAG_CACHEONLY; zone->uz_flags = arg->flags; zone->uz_size = arg->size; zone->uz_import = arg->import; zone->uz_release = arg->release; zone->uz_arg = arg->arg; zone->uz_lockptr = &zone->uz_lock; rw_wlock(&uma_rwlock); LIST_INSERT_HEAD(&uma_cachezones, zone, uz_link); rw_wunlock(&uma_rwlock); goto out; } /* * Use the regular zone/keg/slab allocator. */ zone->uz_import = (uma_import)zone_import; zone->uz_release = (uma_release)zone_release; zone->uz_arg = zone; if (arg->flags & UMA_ZONE_SECONDARY) { KASSERT(arg->keg != NULL, ("Secondary zone on zero'd keg")); zone->uz_init = arg->uminit; zone->uz_fini = arg->fini; zone->uz_lockptr = &keg->uk_lock; zone->uz_flags |= UMA_ZONE_SECONDARY; rw_wlock(&uma_rwlock); ZONE_LOCK(zone); LIST_FOREACH(z, &keg->uk_zones, uz_link) { if (LIST_NEXT(z, uz_link) == NULL) { LIST_INSERT_AFTER(z, zone, uz_link); break; } } ZONE_UNLOCK(zone); rw_wunlock(&uma_rwlock); } else if (keg == NULL) { if ((keg = uma_kcreate(zone, arg->size, arg->uminit, arg->fini, arg->align, arg->flags)) == NULL) return (ENOMEM); } else { struct uma_kctor_args karg; int error; /* We should only be here from uma_startup() */ karg.size = arg->size; karg.uminit = arg->uminit; karg.fini = arg->fini; karg.align = arg->align; karg.flags = arg->flags; karg.zone = zone; error = keg_ctor(arg->keg, sizeof(struct uma_keg), &karg, flags); if (error) return (error); } /* * Link in the first keg. */ zone->uz_klink.kl_keg = keg; LIST_INSERT_HEAD(&zone->uz_kegs, &zone->uz_klink, kl_link); zone->uz_lockptr = &keg->uk_lock; zone->uz_size = keg->uk_size; zone->uz_flags |= (keg->uk_flags & (UMA_ZONE_INHERIT | UMA_ZFLAG_INHERIT)); /* * Some internal zones don't have room allocated for the per cpu * caches. If we're internal, bail out here. */ if (keg->uk_flags & UMA_ZFLAG_INTERNAL) { KASSERT((zone->uz_flags & UMA_ZONE_SECONDARY) == 0, ("Secondary zone requested UMA_ZFLAG_INTERNAL")); return (0); } out: if ((arg->flags & UMA_ZONE_MAXBUCKET) == 0) zone->uz_count = bucket_select(zone->uz_size); else zone->uz_count = BUCKET_MAX; zone->uz_count_min = zone->uz_count; return (0); } /* * Keg header dtor. This frees all data, destroys locks, frees the hash * table and removes the keg from the global list. * * Arguments/Returns follow uma_dtor specifications * udata unused */ static void keg_dtor(void *arg, int size, void *udata) { uma_keg_t keg; keg = (uma_keg_t)arg; KEG_LOCK(keg); if (keg->uk_free != 0) { printf("Freed UMA keg (%s) was not empty (%d items). " " Lost %d pages of memory.\n", keg->uk_name ? keg->uk_name : "", keg->uk_free, keg->uk_pages); } KEG_UNLOCK(keg); hash_free(&keg->uk_hash); KEG_LOCK_FINI(keg); } /* * Zone header dtor. * * Arguments/Returns follow uma_dtor specifications * udata unused */ static void zone_dtor(void *arg, int size, void *udata) { uma_klink_t klink; uma_zone_t zone; uma_keg_t keg; zone = (uma_zone_t)arg; keg = zone_first_keg(zone); if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) cache_drain(zone); rw_wlock(&uma_rwlock); LIST_REMOVE(zone, uz_link); rw_wunlock(&uma_rwlock); /* * XXX there are some races here where * the zone can be drained but zone lock * released and then refilled before we * remove it... we dont care for now */ zone_drain_wait(zone, M_WAITOK); /* * Unlink all of our kegs. */ while ((klink = LIST_FIRST(&zone->uz_kegs)) != NULL) { klink->kl_keg = NULL; LIST_REMOVE(klink, kl_link); if (klink == &zone->uz_klink) continue; free(klink, M_TEMP); } /* * We only destroy kegs from non secondary zones. */ if (keg != NULL && (zone->uz_flags & UMA_ZONE_SECONDARY) == 0) { rw_wlock(&uma_rwlock); LIST_REMOVE(keg, uk_link); rw_wunlock(&uma_rwlock); zone_free_item(kegs, keg, NULL, SKIP_NONE); } ZONE_LOCK_FINI(zone); } /* * Traverses every zone in the system and calls a callback * * Arguments: * zfunc A pointer to a function which accepts a zone * as an argument. * * Returns: * Nothing */ static void zone_foreach(void (*zfunc)(uma_zone_t)) { uma_keg_t keg; uma_zone_t zone; rw_rlock(&uma_rwlock); LIST_FOREACH(keg, &uma_kegs, uk_link) { LIST_FOREACH(zone, &keg->uk_zones, uz_link) zfunc(zone); } rw_runlock(&uma_rwlock); } /* Public functions */ /* See uma.h */ void uma_startup(void *mem, int npages) { struct uma_zctor_args args; rw_init(&uma_rwlock, "UMA lock"); /* "manually" create the initial zone */ memset(&args, 0, sizeof(args)); args.name = "UMA Kegs"; args.size = sizeof(struct uma_keg); args.ctor = keg_ctor; args.dtor = keg_dtor; args.uminit = zero_init; args.fini = NULL; args.keg = &masterkeg; args.align = 32 - 1; args.flags = UMA_ZFLAG_INTERNAL; /* The initial zone has no Per cpu queues so it's smaller */ zone_ctor(kegs, sizeof(struct uma_zone), &args, M_WAITOK); mtx_init(&uma_boot_pages_mtx, "UMA boot pages", NULL, MTX_DEF); bootmem = mem; boot_pages = npages; args.name = "UMA Zones"; args.size = sizeof(struct uma_zone) + (sizeof(struct uma_cache) * (mp_maxid + 1)); args.ctor = zone_ctor; args.dtor = zone_dtor; args.uminit = zero_init; args.fini = NULL; args.keg = NULL; args.align = 32 - 1; args.flags = UMA_ZFLAG_INTERNAL; /* The initial zone has no Per cpu queues so it's smaller */ zone_ctor(zones, sizeof(struct uma_zone), &args, M_WAITOK); /* Now make a zone for slab headers */ slabzone = uma_zcreate("UMA Slabs", sizeof(struct uma_slab), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); hashzone = uma_zcreate("UMA Hash", sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); bucket_init(); booted = UMA_STARTUP; } /* see uma.h */ void uma_startup2(void) { booted = UMA_STARTUP2; bucket_enable(); sx_init(&uma_drain_lock, "umadrain"); } /* * Initialize our callout handle * */ static void uma_startup3(void) { callout_init(&uma_callout, 1); callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); } static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini, int align, uint32_t flags) { struct uma_kctor_args args; args.size = size; args.uminit = uminit; args.fini = fini; args.align = (align == UMA_ALIGN_CACHE) ? uma_align_cache : align; args.flags = flags; args.zone = zone; return (zone_alloc_item(kegs, &args, M_WAITOK)); } /* See uma.h */ void uma_set_align(int align) { if (align != UMA_ALIGN_CACHE) uma_align_cache = align; } /* See uma.h */ uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor, uma_dtor dtor, uma_init uminit, uma_fini fini, int align, uint32_t flags) { struct uma_zctor_args args; uma_zone_t res; bool locked; KASSERT(powerof2(align + 1), ("invalid zone alignment %d for \"%s\"", align, name)); /* This stuff is essential for the zone ctor */ memset(&args, 0, sizeof(args)); args.name = name; args.size = size; args.ctor = ctor; args.dtor = dtor; args.uminit = uminit; args.fini = fini; #ifdef INVARIANTS /* * If a zone is being created with an empty constructor and * destructor, pass UMA constructor/destructor which checks for * memory use after free. */ if ((!(flags & (UMA_ZONE_ZINIT | UMA_ZONE_NOFREE))) && ctor == NULL && dtor == NULL && uminit == NULL && fini == NULL) { args.ctor = trash_ctor; args.dtor = trash_dtor; args.uminit = trash_init; args.fini = trash_fini; } #endif args.align = align; args.flags = flags; args.keg = NULL; if (booted < UMA_STARTUP2) { locked = false; } else { sx_slock(&uma_drain_lock); locked = true; } res = zone_alloc_item(zones, &args, M_WAITOK); if (locked) sx_sunlock(&uma_drain_lock); return (res); } /* See uma.h */ uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_zone_t master) { struct uma_zctor_args args; uma_keg_t keg; uma_zone_t res; bool locked; keg = zone_first_keg(master); memset(&args, 0, sizeof(args)); args.name = name; args.size = keg->uk_size; args.ctor = ctor; args.dtor = dtor; args.uminit = zinit; args.fini = zfini; args.align = keg->uk_align; args.flags = keg->uk_flags | UMA_ZONE_SECONDARY; args.keg = keg; if (booted < UMA_STARTUP2) { locked = false; } else { sx_slock(&uma_drain_lock); locked = true; } /* XXX Attaches only one keg of potentially many. */ res = zone_alloc_item(zones, &args, M_WAITOK); if (locked) sx_sunlock(&uma_drain_lock); return (res); } /* See uma.h */ uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_import zimport, uma_release zrelease, void *arg, int flags) { struct uma_zctor_args args; memset(&args, 0, sizeof(args)); args.name = name; args.size = size; args.ctor = ctor; args.dtor = dtor; args.uminit = zinit; args.fini = zfini; args.import = zimport; args.release = zrelease; args.arg = arg; args.align = 0; args.flags = flags; return (zone_alloc_item(zones, &args, M_WAITOK)); } static void zone_lock_pair(uma_zone_t a, uma_zone_t b) { if (a < b) { ZONE_LOCK(a); mtx_lock_flags(b->uz_lockptr, MTX_DUPOK); } else { ZONE_LOCK(b); mtx_lock_flags(a->uz_lockptr, MTX_DUPOK); } } static void zone_unlock_pair(uma_zone_t a, uma_zone_t b) { ZONE_UNLOCK(a); ZONE_UNLOCK(b); } int uma_zsecond_add(uma_zone_t zone, uma_zone_t master) { uma_klink_t klink; uma_klink_t kl; int error; error = 0; klink = malloc(sizeof(*klink), M_TEMP, M_WAITOK | M_ZERO); zone_lock_pair(zone, master); /* * zone must use vtoslab() to resolve objects and must already be * a secondary. */ if ((zone->uz_flags & (UMA_ZONE_VTOSLAB | UMA_ZONE_SECONDARY)) != (UMA_ZONE_VTOSLAB | UMA_ZONE_SECONDARY)) { error = EINVAL; goto out; } /* * The new master must also use vtoslab(). */ if ((zone->uz_flags & UMA_ZONE_VTOSLAB) != UMA_ZONE_VTOSLAB) { error = EINVAL; goto out; } /* * The underlying object must be the same size. rsize * may be different. */ if (master->uz_size != zone->uz_size) { error = E2BIG; goto out; } /* * Put it at the end of the list. */ klink->kl_keg = zone_first_keg(master); LIST_FOREACH(kl, &zone->uz_kegs, kl_link) { if (LIST_NEXT(kl, kl_link) == NULL) { LIST_INSERT_AFTER(kl, klink, kl_link); break; } } klink = NULL; zone->uz_flags |= UMA_ZFLAG_MULTI; zone->uz_slab = zone_fetch_slab_multi; out: zone_unlock_pair(zone, master); if (klink != NULL) free(klink, M_TEMP); return (error); } /* See uma.h */ void uma_zdestroy(uma_zone_t zone) { sx_slock(&uma_drain_lock); zone_free_item(zones, zone, NULL, SKIP_NONE); sx_sunlock(&uma_drain_lock); } void uma_zwait(uma_zone_t zone) { void *item; item = uma_zalloc_arg(zone, NULL, M_WAITOK); uma_zfree(zone, item); } /* See uma.h */ void * uma_zalloc_arg(uma_zone_t zone, void *udata, int flags) { void *item; uma_cache_t cache; uma_bucket_t bucket; int lockfail; int cpu; /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), 1, RANDOM_UMA); /* This is the fast path allocation */ CTR4(KTR_UMA, "uma_zalloc_arg thread %x zone %s(%p) flags %d", curthread, zone->uz_name, zone, flags); if (flags & M_WAITOK) { WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "uma_zalloc_arg: zone \"%s\"", zone->uz_name); } KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zalloc_arg: called with spinlock or critical section held")); #ifdef DEBUG_MEMGUARD if (memguard_cmp_zone(zone)) { item = memguard_alloc(zone->uz_size, flags); if (item != NULL) { if (zone->uz_init != NULL && zone->uz_init(item, zone->uz_size, flags) != 0) return (NULL); if (zone->uz_ctor != NULL && zone->uz_ctor(item, zone->uz_size, udata, flags) != 0) { zone->uz_fini(item, zone->uz_size); return (NULL); } return (item); } /* This is unfortunate but should not be fatal. */ } #endif /* * If possible, allocate from the per-CPU cache. There are two * requirements for safe access to the per-CPU cache: (1) the thread * accessing the cache must not be preempted or yield during access, * and (2) the thread must not migrate CPUs without switching which * cache it accesses. We rely on a critical section to prevent * preemption and migration. We release the critical section in * order to acquire the zone mutex if we are unable to allocate from * the current cache; when we re-acquire the critical section, we * must detect and handle migration if it has occurred. */ critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; zalloc_start: bucket = cache->uc_allocbucket; if (bucket != NULL && bucket->ub_cnt > 0) { bucket->ub_cnt--; item = bucket->ub_bucket[bucket->ub_cnt]; #ifdef INVARIANTS bucket->ub_bucket[bucket->ub_cnt] = NULL; #endif KASSERT(item != NULL, ("uma_zalloc: Bucket pointer mangled.")); cache->uc_allocs++; critical_exit(); if (zone->uz_ctor != NULL && zone->uz_ctor(item, zone->uz_size, udata, flags) != 0) { atomic_add_long(&zone->uz_fails, 1); zone_free_item(zone, item, udata, SKIP_DTOR); return (NULL); } #ifdef INVARIANTS uma_dbg_alloc(zone, NULL, item); #endif if (flags & M_ZERO) uma_zero_item(item, zone); return (item); } /* * We have run out of items in our alloc bucket. * See if we can switch with our free bucket. */ bucket = cache->uc_freebucket; if (bucket != NULL && bucket->ub_cnt > 0) { CTR2(KTR_UMA, "uma_zalloc: zone %s(%p) swapping empty with alloc", zone->uz_name, zone); cache->uc_freebucket = cache->uc_allocbucket; cache->uc_allocbucket = bucket; goto zalloc_start; } /* * Discard any empty allocation bucket while we hold no locks. */ bucket = cache->uc_allocbucket; cache->uc_allocbucket = NULL; critical_exit(); if (bucket != NULL) bucket_free(zone, bucket, udata); /* Short-circuit for zones without buckets and low memory. */ if (zone->uz_count == 0 || bucketdisable) goto zalloc_item; /* * Attempt to retrieve the item from the per-CPU cache has failed, so * we must go back to the zone. This requires the zone lock, so we * must drop the critical section, then re-acquire it when we go back * to the cache. Since the critical section is released, we may be * preempted or migrate. As such, make sure not to maintain any * thread-local state specific to the cache from prior to releasing * the critical section. */ lockfail = 0; if (ZONE_TRYLOCK(zone) == 0) { /* Record contention to size the buckets. */ ZONE_LOCK(zone); lockfail = 1; } critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; /* * Since we have locked the zone we may as well send back our stats. */ atomic_add_long(&zone->uz_allocs, cache->uc_allocs); atomic_add_long(&zone->uz_frees, cache->uc_frees); cache->uc_allocs = 0; cache->uc_frees = 0; /* See if we lost the race to fill the cache. */ if (cache->uc_allocbucket != NULL) { ZONE_UNLOCK(zone); goto zalloc_start; } /* * Check the zone's cache of buckets. */ if ((bucket = LIST_FIRST(&zone->uz_buckets)) != NULL) { KASSERT(bucket->ub_cnt != 0, ("uma_zalloc_arg: Returning an empty bucket.")); LIST_REMOVE(bucket, ub_link); cache->uc_allocbucket = bucket; ZONE_UNLOCK(zone); goto zalloc_start; } /* We are no longer associated with this CPU. */ critical_exit(); /* * We bump the uz count when the cache size is insufficient to * handle the working set. */ if (lockfail && zone->uz_count < BUCKET_MAX) zone->uz_count++; ZONE_UNLOCK(zone); /* * Now lets just fill a bucket and put it on the free list. If that * works we'll restart the allocation from the beginning and it * will use the just filled bucket. */ bucket = zone_alloc_bucket(zone, udata, flags); CTR3(KTR_UMA, "uma_zalloc: zone %s(%p) bucket zone returned %p", zone->uz_name, zone, bucket); if (bucket != NULL) { ZONE_LOCK(zone); critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; /* * See if we lost the race or were migrated. Cache the * initialized bucket to make this less likely or claim * the memory directly. */ if (cache->uc_allocbucket == NULL) cache->uc_allocbucket = bucket; else LIST_INSERT_HEAD(&zone->uz_buckets, bucket, ub_link); ZONE_UNLOCK(zone); goto zalloc_start; } /* * We may not be able to get a bucket so return an actual item. */ zalloc_item: item = zone_alloc_item(zone, udata, flags); return (item); } static uma_slab_t keg_fetch_slab(uma_keg_t keg, uma_zone_t zone, int flags) { uma_slab_t slab; int reserve; mtx_assert(&keg->uk_lock, MA_OWNED); slab = NULL; reserve = 0; if ((flags & M_USE_RESERVE) == 0) reserve = keg->uk_reserve; for (;;) { /* * Find a slab with some space. Prefer slabs that are partially * used over those that are totally full. This helps to reduce * fragmentation. */ if (keg->uk_free > reserve) { if (!LIST_EMPTY(&keg->uk_part_slab)) { slab = LIST_FIRST(&keg->uk_part_slab); } else { slab = LIST_FIRST(&keg->uk_free_slab); LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_part_slab, slab, us_link); } MPASS(slab->us_keg == keg); return (slab); } /* * M_NOVM means don't ask at all! */ if (flags & M_NOVM) break; if (keg->uk_maxpages && keg->uk_pages >= keg->uk_maxpages) { keg->uk_flags |= UMA_ZFLAG_FULL; /* * If this is not a multi-zone, set the FULL bit. * Otherwise slab_multi() takes care of it. */ if ((zone->uz_flags & UMA_ZFLAG_MULTI) == 0) { zone->uz_flags |= UMA_ZFLAG_FULL; zone_log_warning(zone); zone_maxaction(zone); } if (flags & M_NOWAIT) break; zone->uz_sleeps++; msleep(keg, &keg->uk_lock, PVM, "keglimit", 0); continue; } slab = keg_alloc_slab(keg, zone, flags); /* * If we got a slab here it's safe to mark it partially used * and return. We assume that the caller is going to remove * at least one item. */ if (slab) { MPASS(slab->us_keg == keg); LIST_INSERT_HEAD(&keg->uk_part_slab, slab, us_link); return (slab); } /* * We might not have been able to get a slab but another cpu * could have while we were unlocked. Check again before we * fail. */ flags |= M_NOVM; } return (slab); } static uma_slab_t zone_fetch_slab(uma_zone_t zone, uma_keg_t keg, int flags) { uma_slab_t slab; if (keg == NULL) { keg = zone_first_keg(zone); KEG_LOCK(keg); } for (;;) { slab = keg_fetch_slab(keg, zone, flags); if (slab) return (slab); if (flags & (M_NOWAIT | M_NOVM)) break; } KEG_UNLOCK(keg); return (NULL); } /* * uma_zone_fetch_slab_multi: Fetches a slab from one available keg. Returns * with the keg locked. On NULL no lock is held. * * The last pointer is used to seed the search. It is not required. */ static uma_slab_t zone_fetch_slab_multi(uma_zone_t zone, uma_keg_t last, int rflags) { uma_klink_t klink; uma_slab_t slab; uma_keg_t keg; int flags; int empty; int full; /* * Don't wait on the first pass. This will skip limit tests * as well. We don't want to block if we can find a provider * without blocking. */ flags = (rflags & ~M_WAITOK) | M_NOWAIT; /* * Use the last slab allocated as a hint for where to start * the search. */ if (last != NULL) { slab = keg_fetch_slab(last, zone, flags); if (slab) return (slab); KEG_UNLOCK(last); } /* * Loop until we have a slab incase of transient failures * while M_WAITOK is specified. I'm not sure this is 100% * required but we've done it for so long now. */ for (;;) { empty = 0; full = 0; /* * Search the available kegs for slabs. Be careful to hold the * correct lock while calling into the keg layer. */ LIST_FOREACH(klink, &zone->uz_kegs, kl_link) { keg = klink->kl_keg; KEG_LOCK(keg); if ((keg->uk_flags & UMA_ZFLAG_FULL) == 0) { slab = keg_fetch_slab(keg, zone, flags); if (slab) return (slab); } if (keg->uk_flags & UMA_ZFLAG_FULL) full++; else empty++; KEG_UNLOCK(keg); } if (rflags & (M_NOWAIT | M_NOVM)) break; flags = rflags; /* * All kegs are full. XXX We can't atomically check all kegs * and sleep so just sleep for a short period and retry. */ if (full && !empty) { ZONE_LOCK(zone); zone->uz_flags |= UMA_ZFLAG_FULL; zone->uz_sleeps++; zone_log_warning(zone); zone_maxaction(zone); msleep(zone, zone->uz_lockptr, PVM, "zonelimit", hz/100); zone->uz_flags &= ~UMA_ZFLAG_FULL; ZONE_UNLOCK(zone); continue; } } return (NULL); } static void * slab_alloc_item(uma_keg_t keg, uma_slab_t slab) { void *item; uint8_t freei; MPASS(keg == slab->us_keg); mtx_assert(&keg->uk_lock, MA_OWNED); freei = BIT_FFS(SLAB_SETSIZE, &slab->us_free) - 1; BIT_CLR(SLAB_SETSIZE, freei, &slab->us_free); item = slab->us_data + (keg->uk_rsize * freei); slab->us_freecount--; keg->uk_free--; /* Move this slab to the full list */ if (slab->us_freecount == 0) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_full_slab, slab, us_link); } return (item); } static int zone_import(uma_zone_t zone, void **bucket, int max, int flags) { uma_slab_t slab; uma_keg_t keg; int i; slab = NULL; keg = NULL; /* Try to keep the buckets totally full */ for (i = 0; i < max; ) { if ((slab = zone->uz_slab(zone, keg, flags)) == NULL) break; keg = slab->us_keg; while (slab->us_freecount && i < max) { bucket[i++] = slab_alloc_item(keg, slab); if (keg->uk_free <= keg->uk_reserve) break; } /* Don't grab more than one slab at a time. */ flags &= ~M_WAITOK; flags |= M_NOWAIT; } if (slab != NULL) KEG_UNLOCK(keg); return i; } static uma_bucket_t zone_alloc_bucket(uma_zone_t zone, void *udata, int flags) { uma_bucket_t bucket; int max; /* Don't wait for buckets, preserve caller's NOVM setting. */ bucket = bucket_alloc(zone, udata, M_NOWAIT | (flags & M_NOVM)); if (bucket == NULL) return (NULL); max = MIN(bucket->ub_entries, zone->uz_count); bucket->ub_cnt = zone->uz_import(zone->uz_arg, bucket->ub_bucket, max, flags); /* * Initialize the memory if necessary. */ if (bucket->ub_cnt != 0 && zone->uz_init != NULL) { int i; for (i = 0; i < bucket->ub_cnt; i++) if (zone->uz_init(bucket->ub_bucket[i], zone->uz_size, flags) != 0) break; /* * If we couldn't initialize the whole bucket, put the * rest back onto the freelist. */ if (i != bucket->ub_cnt) { zone->uz_release(zone->uz_arg, &bucket->ub_bucket[i], bucket->ub_cnt - i); #ifdef INVARIANTS bzero(&bucket->ub_bucket[i], sizeof(void *) * (bucket->ub_cnt - i)); #endif bucket->ub_cnt = i; } } if (bucket->ub_cnt == 0) { bucket_free(zone, bucket, udata); atomic_add_long(&zone->uz_fails, 1); return (NULL); } return (bucket); } /* * Allocates a single item from a zone. * * Arguments * zone The zone to alloc for. * udata The data to be passed to the constructor. * flags M_WAITOK, M_NOWAIT, M_ZERO. * * Returns * NULL if there is no memory and M_NOWAIT is set * An item if successful */ static void * zone_alloc_item(uma_zone_t zone, void *udata, int flags) { void *item; item = NULL; if (zone->uz_import(zone->uz_arg, &item, 1, flags) != 1) goto fail; atomic_add_long(&zone->uz_allocs, 1); /* * We have to call both the zone's init (not the keg's init) * and the zone's ctor. This is because the item is going from * a keg slab directly to the user, and the user is expecting it * to be both zone-init'd as well as zone-ctor'd. */ if (zone->uz_init != NULL) { if (zone->uz_init(item, zone->uz_size, flags) != 0) { zone_free_item(zone, item, udata, SKIP_FINI); goto fail; } } if (zone->uz_ctor != NULL) { if (zone->uz_ctor(item, zone->uz_size, udata, flags) != 0) { zone_free_item(zone, item, udata, SKIP_DTOR); goto fail; } } #ifdef INVARIANTS uma_dbg_alloc(zone, NULL, item); #endif if (flags & M_ZERO) uma_zero_item(item, zone); CTR3(KTR_UMA, "zone_alloc_item item %p from %s(%p)", item, zone->uz_name, zone); return (item); fail: CTR2(KTR_UMA, "zone_alloc_item failed from %s(%p)", zone->uz_name, zone); atomic_add_long(&zone->uz_fails, 1); return (NULL); } /* See uma.h */ void uma_zfree_arg(uma_zone_t zone, void *item, void *udata) { uma_cache_t cache; uma_bucket_t bucket; int lockfail; int cpu; /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), 1, RANDOM_UMA); CTR2(KTR_UMA, "uma_zfree_arg thread %x zone %s", curthread, zone->uz_name); KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zfree_arg: called with spinlock or critical section held")); /* uma_zfree(..., NULL) does nothing, to match free(9). */ if (item == NULL) return; #ifdef DEBUG_MEMGUARD if (is_memguard_addr(item)) { if (zone->uz_dtor != NULL) zone->uz_dtor(item, zone->uz_size, udata); if (zone->uz_fini != NULL) zone->uz_fini(item, zone->uz_size); memguard_free(item); return; } #endif #ifdef INVARIANTS if (zone->uz_flags & UMA_ZONE_MALLOC) uma_dbg_free(zone, udata, item); else uma_dbg_free(zone, NULL, item); #endif if (zone->uz_dtor != NULL) zone->uz_dtor(item, zone->uz_size, udata); /* * The race here is acceptable. If we miss it we'll just have to wait * a little longer for the limits to be reset. */ if (zone->uz_flags & UMA_ZFLAG_FULL) goto zfree_item; /* * If possible, free to the per-CPU cache. There are two * requirements for safe access to the per-CPU cache: (1) the thread * accessing the cache must not be preempted or yield during access, * and (2) the thread must not migrate CPUs without switching which * cache it accesses. We rely on a critical section to prevent * preemption and migration. We release the critical section in * order to acquire the zone mutex if we are unable to free to the * current cache; when we re-acquire the critical section, we must * detect and handle migration if it has occurred. */ zfree_restart: critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; zfree_start: /* * Try to free into the allocbucket first to give LIFO ordering * for cache-hot datastructures. Spill over into the freebucket * if necessary. Alloc will swap them if one runs dry. */ bucket = cache->uc_allocbucket; if (bucket == NULL || bucket->ub_cnt >= bucket->ub_entries) bucket = cache->uc_freebucket; if (bucket != NULL && bucket->ub_cnt < bucket->ub_entries) { KASSERT(bucket->ub_bucket[bucket->ub_cnt] == NULL, ("uma_zfree: Freeing to non free bucket index.")); bucket->ub_bucket[bucket->ub_cnt] = item; bucket->ub_cnt++; cache->uc_frees++; critical_exit(); return; } /* * We must go back the zone, which requires acquiring the zone lock, * which in turn means we must release and re-acquire the critical * section. Since the critical section is released, we may be * preempted or migrate. As such, make sure not to maintain any * thread-local state specific to the cache from prior to releasing * the critical section. */ critical_exit(); if (zone->uz_count == 0 || bucketdisable) goto zfree_item; lockfail = 0; if (ZONE_TRYLOCK(zone) == 0) { /* Record contention to size the buckets. */ ZONE_LOCK(zone); lockfail = 1; } critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; /* * Since we have locked the zone we may as well send back our stats. */ atomic_add_long(&zone->uz_allocs, cache->uc_allocs); atomic_add_long(&zone->uz_frees, cache->uc_frees); cache->uc_allocs = 0; cache->uc_frees = 0; bucket = cache->uc_freebucket; if (bucket != NULL && bucket->ub_cnt < bucket->ub_entries) { ZONE_UNLOCK(zone); goto zfree_start; } cache->uc_freebucket = NULL; /* We are no longer associated with this CPU. */ critical_exit(); /* Can we throw this on the zone full list? */ if (bucket != NULL) { CTR3(KTR_UMA, "uma_zfree: zone %s(%p) putting bucket %p on free list", zone->uz_name, zone, bucket); /* ub_cnt is pointing to the last free item */ KASSERT(bucket->ub_cnt != 0, ("uma_zfree: Attempting to insert an empty bucket onto the full list.\n")); LIST_INSERT_HEAD(&zone->uz_buckets, bucket, ub_link); } /* * We bump the uz count when the cache size is insufficient to * handle the working set. */ if (lockfail && zone->uz_count < BUCKET_MAX) zone->uz_count++; ZONE_UNLOCK(zone); bucket = bucket_alloc(zone, udata, M_NOWAIT); CTR3(KTR_UMA, "uma_zfree: zone %s(%p) allocated bucket %p", zone->uz_name, zone, bucket); if (bucket) { critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; if (cache->uc_freebucket == NULL) { cache->uc_freebucket = bucket; goto zfree_start; } /* * We lost the race, start over. We have to drop our * critical section to free the bucket. */ critical_exit(); bucket_free(zone, bucket, udata); goto zfree_restart; } /* * If nothing else caught this, we'll just do an internal free. */ zfree_item: zone_free_item(zone, item, udata, SKIP_DTOR); return; } static void slab_free_item(uma_keg_t keg, uma_slab_t slab, void *item) { uint8_t freei; mtx_assert(&keg->uk_lock, MA_OWNED); MPASS(keg == slab->us_keg); /* Do we need to remove from any lists? */ if (slab->us_freecount+1 == keg->uk_ipers) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_free_slab, slab, us_link); } else if (slab->us_freecount == 0) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_part_slab, slab, us_link); } /* Slab management. */ freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; BIT_SET(SLAB_SETSIZE, freei, &slab->us_free); slab->us_freecount++; /* Keg statistics. */ keg->uk_free++; } static void zone_release(uma_zone_t zone, void **bucket, int cnt) { void *item; uma_slab_t slab; uma_keg_t keg; uint8_t *mem; int clearfull; int i; clearfull = 0; keg = zone_first_keg(zone); KEG_LOCK(keg); for (i = 0; i < cnt; i++) { item = bucket[i]; if (!(zone->uz_flags & UMA_ZONE_VTOSLAB)) { mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK)); if (zone->uz_flags & UMA_ZONE_HASH) { slab = hash_sfind(&keg->uk_hash, mem); } else { mem += keg->uk_pgoff; slab = (uma_slab_t)mem; } } else { slab = vtoslab((vm_offset_t)item); if (slab->us_keg != keg) { KEG_UNLOCK(keg); keg = slab->us_keg; KEG_LOCK(keg); } } slab_free_item(keg, slab, item); if (keg->uk_flags & UMA_ZFLAG_FULL) { if (keg->uk_pages < keg->uk_maxpages) { keg->uk_flags &= ~UMA_ZFLAG_FULL; clearfull = 1; } /* * We can handle one more allocation. Since we're * clearing ZFLAG_FULL, wake up all procs blocked * on pages. This should be uncommon, so keeping this * simple for now (rather than adding count of blocked * threads etc). */ wakeup(keg); } } KEG_UNLOCK(keg); if (clearfull) { ZONE_LOCK(zone); zone->uz_flags &= ~UMA_ZFLAG_FULL; wakeup(zone); ZONE_UNLOCK(zone); } } /* * Frees a single item to any zone. * * Arguments: * zone The zone to free to * item The item we're freeing * udata User supplied data for the dtor * skip Skip dtors and finis */ static void zone_free_item(uma_zone_t zone, void *item, void *udata, enum zfreeskip skip) { #ifdef INVARIANTS if (skip == SKIP_NONE) { if (zone->uz_flags & UMA_ZONE_MALLOC) uma_dbg_free(zone, udata, item); else uma_dbg_free(zone, NULL, item); } #endif if (skip < SKIP_DTOR && zone->uz_dtor) zone->uz_dtor(item, zone->uz_size, udata); if (skip < SKIP_FINI && zone->uz_fini) zone->uz_fini(item, zone->uz_size); atomic_add_long(&zone->uz_frees, 1); zone->uz_release(zone->uz_arg, &item, 1); } /* See uma.h */ int uma_zone_set_max(uma_zone_t zone, int nitems) { uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return (0); KEG_LOCK(keg); keg->uk_maxpages = (nitems / keg->uk_ipers) * keg->uk_ppera; if (keg->uk_maxpages * keg->uk_ipers < nitems) keg->uk_maxpages += keg->uk_ppera; nitems = (keg->uk_maxpages / keg->uk_ppera) * keg->uk_ipers; KEG_UNLOCK(keg); return (nitems); } /* See uma.h */ int uma_zone_get_max(uma_zone_t zone) { int nitems; uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return (0); KEG_LOCK(keg); nitems = (keg->uk_maxpages / keg->uk_ppera) * keg->uk_ipers; KEG_UNLOCK(keg); return (nitems); } /* See uma.h */ void uma_zone_set_warning(uma_zone_t zone, const char *warning) { ZONE_LOCK(zone); zone->uz_warning = warning; ZONE_UNLOCK(zone); } /* See uma.h */ void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t maxaction) { ZONE_LOCK(zone); TASK_INIT(&zone->uz_maxaction, 0, (task_fn_t *)maxaction, zone); ZONE_UNLOCK(zone); } /* See uma.h */ int uma_zone_get_cur(uma_zone_t zone) { int64_t nitems; u_int i; ZONE_LOCK(zone); nitems = zone->uz_allocs - zone->uz_frees; CPU_FOREACH(i) { /* * See the comment in sysctl_vm_zone_stats() regarding the * safety of accessing the per-cpu caches. With the zone lock * held, it is safe, but can potentially result in stale data. */ nitems += zone->uz_cpu[i].uc_allocs - zone->uz_cpu[i].uc_frees; } ZONE_UNLOCK(zone); return (nitems < 0 ? 0 : nitems); } /* See uma.h */ void uma_zone_set_init(uma_zone_t zone, uma_init uminit) { uma_keg_t keg; keg = zone_first_keg(zone); KASSERT(keg != NULL, ("uma_zone_set_init: Invalid zone type")); KEG_LOCK(keg); KASSERT(keg->uk_pages == 0, ("uma_zone_set_init on non-empty keg")); keg->uk_init = uminit; KEG_UNLOCK(keg); } /* See uma.h */ void uma_zone_set_fini(uma_zone_t zone, uma_fini fini) { uma_keg_t keg; keg = zone_first_keg(zone); KASSERT(keg != NULL, ("uma_zone_set_fini: Invalid zone type")); KEG_LOCK(keg); KASSERT(keg->uk_pages == 0, ("uma_zone_set_fini on non-empty keg")); keg->uk_fini = fini; KEG_UNLOCK(keg); } /* See uma.h */ void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit) { ZONE_LOCK(zone); KASSERT(zone_first_keg(zone)->uk_pages == 0, ("uma_zone_set_zinit on non-empty keg")); zone->uz_init = zinit; ZONE_UNLOCK(zone); } /* See uma.h */ void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini) { ZONE_LOCK(zone); KASSERT(zone_first_keg(zone)->uk_pages == 0, ("uma_zone_set_zfini on non-empty keg")); zone->uz_fini = zfini; ZONE_UNLOCK(zone); } /* See uma.h */ /* XXX uk_freef is not actually used with the zone locked */ void uma_zone_set_freef(uma_zone_t zone, uma_free freef) { uma_keg_t keg; keg = zone_first_keg(zone); KASSERT(keg != NULL, ("uma_zone_set_freef: Invalid zone type")); KEG_LOCK(keg); keg->uk_freef = freef; KEG_UNLOCK(keg); } /* See uma.h */ /* XXX uk_allocf is not actually used with the zone locked */ void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf) { uma_keg_t keg; keg = zone_first_keg(zone); KEG_LOCK(keg); keg->uk_allocf = allocf; KEG_UNLOCK(keg); } /* See uma.h */ void uma_zone_reserve(uma_zone_t zone, int items) { uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return; KEG_LOCK(keg); keg->uk_reserve = items; KEG_UNLOCK(keg); return; } /* See uma.h */ int uma_zone_reserve_kva(uma_zone_t zone, int count) { uma_keg_t keg; vm_offset_t kva; u_int pages; keg = zone_first_keg(zone); if (keg == NULL) return (0); pages = count / keg->uk_ipers; if (pages * keg->uk_ipers < count) pages++; pages *= keg->uk_ppera; #ifdef UMA_MD_SMALL_ALLOC if (keg->uk_ppera > 1) { #else if (1) { #endif kva = kva_alloc((vm_size_t)pages * PAGE_SIZE); if (kva == 0) return (0); } else kva = 0; KEG_LOCK(keg); keg->uk_kva = kva; keg->uk_offset = 0; keg->uk_maxpages = pages; #ifdef UMA_MD_SMALL_ALLOC keg->uk_allocf = (keg->uk_ppera > 1) ? noobj_alloc : uma_small_alloc; #else keg->uk_allocf = noobj_alloc; #endif keg->uk_flags |= UMA_ZONE_NOFREE; KEG_UNLOCK(keg); return (1); } /* See uma.h */ void uma_prealloc(uma_zone_t zone, int items) { int slabs; uma_slab_t slab; uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return; KEG_LOCK(keg); slabs = items / keg->uk_ipers; if (slabs * keg->uk_ipers < items) slabs++; while (slabs > 0) { slab = keg_alloc_slab(keg, zone, M_WAITOK); if (slab == NULL) break; MPASS(slab->us_keg == keg); LIST_INSERT_HEAD(&keg->uk_free_slab, slab, us_link); slabs--; } KEG_UNLOCK(keg); } /* See uma.h */ static void uma_reclaim_locked(bool kmem_danger) { CTR0(KTR_UMA, "UMA: vm asked us to release pages!"); sx_assert(&uma_drain_lock, SA_XLOCKED); bucket_enable(); zone_foreach(zone_drain); if (vm_page_count_min() || kmem_danger) { cache_drain_safe(NULL); zone_foreach(zone_drain); } /* * Some slabs may have been freed but this zone will be visited early * we visit again so that we can free pages that are empty once other * zones are drained. We have to do the same for buckets. */ zone_drain(slabzone); bucket_zone_drain(); } void uma_reclaim(void) { sx_xlock(&uma_drain_lock); uma_reclaim_locked(false); sx_xunlock(&uma_drain_lock); } static volatile int uma_reclaim_needed; void uma_reclaim_wakeup(void) { if (atomic_fetchadd_int(&uma_reclaim_needed, 1) == 0) wakeup(uma_reclaim); } void uma_reclaim_worker(void *arg __unused) { for (;;) { sx_xlock(&uma_drain_lock); while (atomic_load_int(&uma_reclaim_needed) == 0) sx_sleep(uma_reclaim, &uma_drain_lock, PVM, "umarcl", hz); sx_xunlock(&uma_drain_lock); EVENTHANDLER_INVOKE(vm_lowmem, VM_LOW_KMEM); sx_xlock(&uma_drain_lock); uma_reclaim_locked(true); atomic_store_int(&uma_reclaim_needed, 0); sx_xunlock(&uma_drain_lock); /* Don't fire more than once per-second. */ pause("umarclslp", hz); } } /* See uma.h */ int uma_zone_exhausted(uma_zone_t zone) { int full; ZONE_LOCK(zone); full = (zone->uz_flags & UMA_ZFLAG_FULL); ZONE_UNLOCK(zone); return (full); } int uma_zone_exhausted_nolock(uma_zone_t zone) { return (zone->uz_flags & UMA_ZFLAG_FULL); } void * uma_large_malloc(vm_size_t size, int wait) { void *mem; uma_slab_t slab; uint8_t flags; slab = zone_alloc_item(slabzone, NULL, wait); if (slab == NULL) return (NULL); mem = page_alloc(NULL, size, &flags, wait); if (mem) { vsetslab((vm_offset_t)mem, slab); slab->us_data = mem; slab->us_flags = flags | UMA_SLAB_MALLOC; slab->us_size = size; uma_total_inc(size); } else { zone_free_item(slabzone, slab, NULL, SKIP_NONE); } return (mem); } void uma_large_free(uma_slab_t slab) { page_free(slab->us_data, slab->us_size, slab->us_flags); uma_total_dec(slab->us_size); zone_free_item(slabzone, slab, NULL, SKIP_NONE); } static void uma_zero_item(void *item, uma_zone_t zone) { int i; if (zone->uz_flags & UMA_ZONE_PCPU) { CPU_FOREACH(i) bzero(zpcpu_get_cpu(item, i), zone->uz_size); } else bzero(item, zone->uz_size); } unsigned long uma_limit(void) { return (uma_kmem_limit); } void uma_set_limit(unsigned long limit) { uma_kmem_limit = limit; } unsigned long uma_size(void) { - return uma_kmem_total; + return (uma_kmem_total); +} + +long +uma_avail(void) +{ + + return (uma_kmem_limit - uma_kmem_total); } void uma_print_stats(void) { zone_foreach(uma_print_zone); } static void slab_print(uma_slab_t slab) { printf("slab: keg %p, data %p, freecount %d\n", slab->us_keg, slab->us_data, slab->us_freecount); } static void cache_print(uma_cache_t cache) { printf("alloc: %p(%d), free: %p(%d)\n", cache->uc_allocbucket, cache->uc_allocbucket?cache->uc_allocbucket->ub_cnt:0, cache->uc_freebucket, cache->uc_freebucket?cache->uc_freebucket->ub_cnt:0); } static void uma_print_keg(uma_keg_t keg) { uma_slab_t slab; printf("keg: %s(%p) size %d(%d) flags %#x ipers %d ppera %d " "out %d free %d limit %d\n", keg->uk_name, keg, keg->uk_size, keg->uk_rsize, keg->uk_flags, keg->uk_ipers, keg->uk_ppera, (keg->uk_pages / keg->uk_ppera) * keg->uk_ipers - keg->uk_free, keg->uk_free, (keg->uk_maxpages / keg->uk_ppera) * keg->uk_ipers); printf("Part slabs:\n"); LIST_FOREACH(slab, &keg->uk_part_slab, us_link) slab_print(slab); printf("Free slabs:\n"); LIST_FOREACH(slab, &keg->uk_free_slab, us_link) slab_print(slab); printf("Full slabs:\n"); LIST_FOREACH(slab, &keg->uk_full_slab, us_link) slab_print(slab); } void uma_print_zone(uma_zone_t zone) { uma_cache_t cache; uma_klink_t kl; int i; printf("zone: %s(%p) size %d flags %#x\n", zone->uz_name, zone, zone->uz_size, zone->uz_flags); LIST_FOREACH(kl, &zone->uz_kegs, kl_link) uma_print_keg(kl->kl_keg); CPU_FOREACH(i) { cache = &zone->uz_cpu[i]; printf("CPU %d Cache:\n", i); cache_print(cache); } } #ifdef DDB /* * Generate statistics across both the zone and its per-cpu cache's. Return * desired statistics if the pointer is non-NULL for that statistic. * * Note: does not update the zone statistics, as it can't safely clear the * per-CPU cache statistic. * * XXXRW: Following the uc_allocbucket and uc_freebucket pointers here isn't * safe from off-CPU; we should modify the caches to track this information * directly so that we don't have to. */ static void uma_zone_sumstat(uma_zone_t z, int *cachefreep, uint64_t *allocsp, uint64_t *freesp, uint64_t *sleepsp) { uma_cache_t cache; uint64_t allocs, frees, sleeps; int cachefree, cpu; allocs = frees = sleeps = 0; cachefree = 0; CPU_FOREACH(cpu) { cache = &z->uz_cpu[cpu]; if (cache->uc_allocbucket != NULL) cachefree += cache->uc_allocbucket->ub_cnt; if (cache->uc_freebucket != NULL) cachefree += cache->uc_freebucket->ub_cnt; allocs += cache->uc_allocs; frees += cache->uc_frees; } allocs += z->uz_allocs; frees += z->uz_frees; sleeps += z->uz_sleeps; if (cachefreep != NULL) *cachefreep = cachefree; if (allocsp != NULL) *allocsp = allocs; if (freesp != NULL) *freesp = frees; if (sleepsp != NULL) *sleepsp = sleeps; } #endif /* DDB */ static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS) { uma_keg_t kz; uma_zone_t z; int count; count = 0; rw_rlock(&uma_rwlock); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) count++; } rw_runlock(&uma_rwlock); return (sysctl_handle_int(oidp, &count, 0, req)); } static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS) { struct uma_stream_header ush; struct uma_type_header uth; struct uma_percpu_stat ups; uma_bucket_t bucket; struct sbuf sbuf; uma_cache_t cache; uma_klink_t kl; uma_keg_t kz; uma_zone_t z; uma_keg_t k; int count, error, i; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); count = 0; rw_rlock(&uma_rwlock); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) count++; } /* * Insert stream header. */ bzero(&ush, sizeof(ush)); ush.ush_version = UMA_STREAM_VERSION; ush.ush_maxcpus = (mp_maxid + 1); ush.ush_count = count; (void)sbuf_bcat(&sbuf, &ush, sizeof(ush)); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) { bzero(&uth, sizeof(uth)); ZONE_LOCK(z); strlcpy(uth.uth_name, z->uz_name, UTH_MAX_NAME); uth.uth_align = kz->uk_align; uth.uth_size = kz->uk_size; uth.uth_rsize = kz->uk_rsize; LIST_FOREACH(kl, &z->uz_kegs, kl_link) { k = kl->kl_keg; uth.uth_maxpages += k->uk_maxpages; uth.uth_pages += k->uk_pages; uth.uth_keg_free += k->uk_free; uth.uth_limit = (k->uk_maxpages / k->uk_ppera) * k->uk_ipers; } /* * A zone is secondary is it is not the first entry * on the keg's zone list. */ if ((z->uz_flags & UMA_ZONE_SECONDARY) && (LIST_FIRST(&kz->uk_zones) != z)) uth.uth_zone_flags = UTH_ZONE_SECONDARY; LIST_FOREACH(bucket, &z->uz_buckets, ub_link) uth.uth_zone_free += bucket->ub_cnt; uth.uth_allocs = z->uz_allocs; uth.uth_frees = z->uz_frees; uth.uth_fails = z->uz_fails; uth.uth_sleeps = z->uz_sleeps; (void)sbuf_bcat(&sbuf, &uth, sizeof(uth)); /* * While it is not normally safe to access the cache * bucket pointers while not on the CPU that owns the * cache, we only allow the pointers to be exchanged * without the zone lock held, not invalidated, so * accept the possible race associated with bucket * exchange during monitoring. */ for (i = 0; i < (mp_maxid + 1); i++) { bzero(&ups, sizeof(ups)); if (kz->uk_flags & UMA_ZFLAG_INTERNAL) goto skip; if (CPU_ABSENT(i)) goto skip; cache = &z->uz_cpu[i]; if (cache->uc_allocbucket != NULL) ups.ups_cache_free += cache->uc_allocbucket->ub_cnt; if (cache->uc_freebucket != NULL) ups.ups_cache_free += cache->uc_freebucket->ub_cnt; ups.ups_allocs = cache->uc_allocs; ups.ups_frees = cache->uc_frees; skip: (void)sbuf_bcat(&sbuf, &ups, sizeof(ups)); } ZONE_UNLOCK(z); } } rw_runlock(&uma_rwlock); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } int sysctl_handle_uma_zone_max(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = *(uma_zone_t *)arg1; int error, max; max = uma_zone_get_max(zone); error = sysctl_handle_int(oidp, &max, 0, req); if (error || !req->newptr) return (error); uma_zone_set_max(zone, max); return (0); } int sysctl_handle_uma_zone_cur(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = *(uma_zone_t *)arg1; int cur; cur = uma_zone_get_cur(zone); return (sysctl_handle_int(oidp, &cur, 0, req)); } #ifdef INVARIANTS static uma_slab_t uma_dbg_getslab(uma_zone_t zone, void *item) { uma_slab_t slab; uma_keg_t keg; uint8_t *mem; mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK)); if (zone->uz_flags & UMA_ZONE_VTOSLAB) { slab = vtoslab((vm_offset_t)mem); } else { /* * It is safe to return the slab here even though the * zone is unlocked because the item's allocation state * essentially holds a reference. */ ZONE_LOCK(zone); keg = LIST_FIRST(&zone->uz_kegs)->kl_keg; if (keg->uk_flags & UMA_ZONE_HASH) slab = hash_sfind(&keg->uk_hash, mem); else slab = (uma_slab_t)(mem + keg->uk_pgoff); ZONE_UNLOCK(zone); } return (slab); } /* * Set up the slab's freei data such that uma_dbg_free can function. * */ static void uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item) { uma_keg_t keg; int freei; if (zone_first_keg(zone) == NULL) return; if (slab == NULL) { slab = uma_dbg_getslab(zone, item); if (slab == NULL) panic("uma: item %p did not belong to zone %s\n", item, zone->uz_name); } keg = slab->us_keg; freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; if (BIT_ISSET(SLAB_SETSIZE, freei, &slab->us_debugfree)) panic("Duplicate alloc of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); BIT_SET_ATOMIC(SLAB_SETSIZE, freei, &slab->us_debugfree); return; } /* * Verifies freed addresses. Checks for alignment, valid slab membership * and duplicate frees. * */ static void uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item) { uma_keg_t keg; int freei; if (zone_first_keg(zone) == NULL) return; if (slab == NULL) { slab = uma_dbg_getslab(zone, item); if (slab == NULL) panic("uma: Freed item %p did not belong to zone %s\n", item, zone->uz_name); } keg = slab->us_keg; freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; if (freei >= keg->uk_ipers) panic("Invalid free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); if (((freei * keg->uk_rsize) + slab->us_data) != item) panic("Unaligned free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); if (!BIT_ISSET(SLAB_SETSIZE, freei, &slab->us_debugfree)) panic("Duplicate free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); BIT_CLR_ATOMIC(SLAB_SETSIZE, freei, &slab->us_debugfree); } #endif /* INVARIANTS */ #ifdef DDB DB_SHOW_COMMAND(uma, db_show_uma) { uint64_t allocs, frees, sleeps; uma_bucket_t bucket; uma_keg_t kz; uma_zone_t z; int cachefree; db_printf("%18s %8s %8s %8s %12s %8s %8s\n", "Zone", "Size", "Used", "Free", "Requests", "Sleeps", "Bucket"); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) { if (kz->uk_flags & UMA_ZFLAG_INTERNAL) { allocs = z->uz_allocs; frees = z->uz_frees; sleeps = z->uz_sleeps; cachefree = 0; } else uma_zone_sumstat(z, &cachefree, &allocs, &frees, &sleeps); if (!((z->uz_flags & UMA_ZONE_SECONDARY) && (LIST_FIRST(&kz->uk_zones) != z))) cachefree += kz->uk_free; LIST_FOREACH(bucket, &z->uz_buckets, ub_link) cachefree += bucket->ub_cnt; db_printf("%18s %8ju %8jd %8d %12ju %8ju %8u\n", z->uz_name, (uintmax_t)kz->uk_size, (intmax_t)(allocs - frees), cachefree, (uintmax_t)allocs, sleeps, z->uz_count); if (db_pager_quit) return; } } } DB_SHOW_COMMAND(umacache, db_show_umacache) { uint64_t allocs, frees; uma_bucket_t bucket; uma_zone_t z; int cachefree; db_printf("%18s %8s %8s %8s %12s %8s\n", "Zone", "Size", "Used", "Free", "Requests", "Bucket"); LIST_FOREACH(z, &uma_cachezones, uz_link) { uma_zone_sumstat(z, &cachefree, &allocs, &frees, NULL); LIST_FOREACH(bucket, &z->uz_buckets, ub_link) cachefree += bucket->ub_cnt; db_printf("%18s %8ju %8jd %8d %12ju %8u\n", z->uz_name, (uintmax_t)z->uz_size, (intmax_t)(allocs - frees), cachefree, (uintmax_t)allocs, z->uz_count); if (db_pager_quit) return; } } #endif /* DDB */ Index: head/sys/vm/uma_int.h =================================================================== --- head/sys/vm/uma_int.h (revision 327484) +++ head/sys/vm/uma_int.h (revision 327485) @@ -1,437 +1,433 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002-2005, 2009, 2013 Jeffrey Roberson * Copyright (c) 2004, 2005 Bosko Milekic * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ * */ #include #include /* * This file includes definitions, structures, prototypes, and inlines that * should not be used outside of the actual implementation of UMA. */ /* * Here's a quick description of the relationship between the objects: * * Kegs contain lists of slabs which are stored in either the full bin, empty * bin, or partially allocated bin, to reduce fragmentation. They also contain * the user supplied value for size, which is adjusted for alignment purposes * and rsize is the result of that. The Keg also stores information for * managing a hash of page addresses that maps pages to uma_slab_t structures * for pages that don't have embedded uma_slab_t's. * * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may * be allocated off the page from a special slab zone. The free list within a * slab is managed with a bitmask. For item sizes that would yield more than * 10% memory waste we potentially allocate a separate uma_slab_t if this will * improve the number of items per slab that will fit. * * The only really gross cases, with regards to memory waste, are for those * items that are just over half the page size. You can get nearly 50% waste, * so you fall back to the memory footprint of the power of two allocator. I * have looked at memory allocation sizes on many of the machines available to * me, and there does not seem to be an abundance of allocations at this range * so at this time it may not make sense to optimize for it. This can, of * course, be solved with dynamic slab sizes. * * Kegs may serve multiple Zones but by far most of the time they only serve * one. When a Zone is created, a Keg is allocated and setup for it. While * the backing Keg stores slabs, the Zone caches Buckets of items allocated * from the slabs. Each Zone is equipped with an init/fini and ctor/dtor * pair, as well as with its own set of small per-CPU caches, layered above * the Zone's general Bucket cache. * * The PCPU caches are protected by critical sections, and may be accessed * safely only from their associated CPU, while the Zones backed by the same * Keg all share a common Keg lock (to coalesce contention on the backing * slabs). The backing Keg typically only serves one Zone but in the case of * multiple Zones, one of the Zones is considered the Master Zone and all * Zone-related stats from the Keg are done in the Master Zone. For an * example of a Multi-Zone setup, refer to the Mbuf allocation code. */ /* * This is the representation for normal (Non OFFPAGE slab) * * i == item * s == slab pointer * * <---------------- Page (UMA_SLAB_SIZE) ------------------> * ___________________________________________________________ * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ | * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header|| * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________|| * |___________________________________________________________| * * * This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE. * * ___________________________________________________________ * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ | * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| | * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| | * |___________________________________________________________| * ___________ ^ * |slab header| | * |___________|---* * */ #ifndef VM_UMA_INT_H #define VM_UMA_INT_H #define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */ #define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */ #define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */ #define UMA_BOOT_PAGES 64 /* Pages allocated for startup */ #define UMA_BOOT_PAGES_ZONES 32 /* Multiplier for pages to reserve */ /* if uma_zone > PAGE_SIZE */ /* Max waste percentage before going to off page slab management */ #define UMA_MAX_WASTE 10 /* * I doubt there will be many cases where this is exceeded. This is the initial * size of the hash table for uma_slabs that are managed off page. This hash * does expand by powers of two. Currently it doesn't get smaller. */ #define UMA_HASH_SIZE_INIT 32 /* * I should investigate other hashing algorithms. This should yield a low * number of collisions if the pages are relatively contiguous. */ #define UMA_HASH(h, s) ((((uintptr_t)s) >> UMA_SLAB_SHIFT) & (h)->uh_hashmask) #define UMA_HASH_INSERT(h, s, mem) \ SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \ (mem))], (s), us_hlink) #define UMA_HASH_REMOVE(h, s, mem) \ SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \ (mem))], (s), uma_slab, us_hlink) /* Hash table for freed address -> slab translation */ SLIST_HEAD(slabhead, uma_slab); struct uma_hash { struct slabhead *uh_slab_hash; /* Hash table for slabs */ int uh_hashsize; /* Current size of the hash table */ int uh_hashmask; /* Mask used during hashing */ }; /* * align field or structure to cache line */ #if defined(__amd64__) #define UMA_ALIGN __aligned(CACHE_LINE_SIZE) #else #define UMA_ALIGN #endif /* * Structures for per cpu queues. */ struct uma_bucket { LIST_ENTRY(uma_bucket) ub_link; /* Link into the zone */ int16_t ub_cnt; /* Count of free items. */ int16_t ub_entries; /* Max items. */ void *ub_bucket[]; /* actual allocation storage */ }; typedef struct uma_bucket * uma_bucket_t; struct uma_cache { uma_bucket_t uc_freebucket; /* Bucket we're freeing to */ uma_bucket_t uc_allocbucket; /* Bucket to allocate from */ uint64_t uc_allocs; /* Count of allocations */ uint64_t uc_frees; /* Count of frees */ } UMA_ALIGN; typedef struct uma_cache * uma_cache_t; /* * Keg management structure * * TODO: Optimize for cache line size * */ struct uma_keg { struct mtx_padalign uk_lock; /* Lock for the keg */ struct uma_hash uk_hash; LIST_HEAD(,uma_zone) uk_zones; /* Keg's zones */ LIST_HEAD(,uma_slab) uk_part_slab; /* partially allocated slabs */ LIST_HEAD(,uma_slab) uk_free_slab; /* empty slab list */ LIST_HEAD(,uma_slab) uk_full_slab; /* full slabs */ uint32_t uk_align; /* Alignment mask */ uint32_t uk_pages; /* Total page count */ uint32_t uk_free; /* Count of items free in slabs */ uint32_t uk_reserve; /* Number of reserved items. */ uint32_t uk_size; /* Requested size of each item */ uint32_t uk_rsize; /* Real size of each item */ uint32_t uk_maxpages; /* Maximum number of pages to alloc */ uma_init uk_init; /* Keg's init routine */ uma_fini uk_fini; /* Keg's fini routine */ uma_alloc uk_allocf; /* Allocation function */ uma_free uk_freef; /* Free routine */ u_long uk_offset; /* Next free offset from base KVA */ vm_offset_t uk_kva; /* Zone base KVA */ uma_zone_t uk_slabzone; /* Slab zone backing us, if OFFPAGE */ uint32_t uk_pgoff; /* Offset to uma_slab struct */ uint16_t uk_ppera; /* pages per allocation from backend */ uint16_t uk_ipers; /* Items per slab */ uint32_t uk_flags; /* Internal flags */ /* Least used fields go to the last cache line. */ const char *uk_name; /* Name of creating zone. */ LIST_ENTRY(uma_keg) uk_link; /* List of all kegs */ }; typedef struct uma_keg * uma_keg_t; /* * Free bits per-slab. */ #define SLAB_SETSIZE (PAGE_SIZE / UMA_SMALLEST_UNIT) BITSET_DEFINE(slabbits, SLAB_SETSIZE); /* * The slab structure manages a single contiguous allocation from backing * store and subdivides it into individually allocatable items. */ struct uma_slab { uma_keg_t us_keg; /* Keg we live in */ union { LIST_ENTRY(uma_slab) _us_link; /* slabs in zone */ unsigned long _us_size; /* Size of allocation */ } us_type; SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */ uint8_t *us_data; /* First item */ struct slabbits us_free; /* Free bitmask. */ #ifdef INVARIANTS struct slabbits us_debugfree; /* Debug bitmask. */ #endif uint16_t us_freecount; /* How many are free? */ uint8_t us_flags; /* Page flags see uma.h */ uint8_t us_pad; /* Pad to 32bits, unused. */ }; #define us_link us_type._us_link #define us_size us_type._us_size typedef struct uma_slab * uma_slab_t; typedef uma_slab_t (*uma_slaballoc)(uma_zone_t, uma_keg_t, int); struct uma_klink { LIST_ENTRY(uma_klink) kl_link; uma_keg_t kl_keg; }; typedef struct uma_klink *uma_klink_t; /* * Zone management structure * * TODO: Optimize for cache line size * */ struct uma_zone { struct mtx_padalign uz_lock; /* Lock for the zone */ struct mtx_padalign *uz_lockptr; const char *uz_name; /* Text name of the zone */ LIST_ENTRY(uma_zone) uz_link; /* List of all zones in keg */ LIST_HEAD(,uma_bucket) uz_buckets; /* full buckets */ LIST_HEAD(,uma_klink) uz_kegs; /* List of kegs. */ struct uma_klink uz_klink; /* klink for first keg. */ uma_slaballoc uz_slab; /* Allocate a slab from the backend. */ uma_ctor uz_ctor; /* Constructor for each allocation */ uma_dtor uz_dtor; /* Destructor */ uma_init uz_init; /* Initializer for each item */ uma_fini uz_fini; /* Finalizer for each item. */ uma_import uz_import; /* Import new memory to cache. */ uma_release uz_release; /* Release memory from cache. */ void *uz_arg; /* Import/release argument. */ uint32_t uz_flags; /* Flags inherited from kegs */ uint32_t uz_size; /* Size inherited from kegs */ volatile u_long uz_allocs UMA_ALIGN; /* Total number of allocations */ volatile u_long uz_fails; /* Total number of alloc failures */ volatile u_long uz_frees; /* Total number of frees */ uint64_t uz_sleeps; /* Total number of alloc sleeps */ uint16_t uz_count; /* Amount of items in full bucket */ uint16_t uz_count_min; /* Minimal amount of items there */ /* The next two fields are used to print a rate-limited warnings. */ const char *uz_warning; /* Warning to print on failure */ struct timeval uz_ratecheck; /* Warnings rate-limiting */ struct task uz_maxaction; /* Task to run when at limit */ /* * This HAS to be the last item because we adjust the zone size * based on NCPU and then allocate the space for the zones. */ struct uma_cache uz_cpu[1]; /* Per cpu caches */ }; /* * These flags must not overlap with the UMA_ZONE flags specified in uma.h. */ #define UMA_ZFLAG_MULTI 0x04000000 /* Multiple kegs in the zone. */ #define UMA_ZFLAG_DRAINING 0x08000000 /* Running zone_drain. */ #define UMA_ZFLAG_BUCKET 0x10000000 /* Bucket zone. */ #define UMA_ZFLAG_INTERNAL 0x20000000 /* No offpage no PCPU. */ #define UMA_ZFLAG_FULL 0x40000000 /* Reached uz_maxpages */ #define UMA_ZFLAG_CACHEONLY 0x80000000 /* Don't ask VM for buckets. */ #define UMA_ZFLAG_INHERIT \ (UMA_ZFLAG_INTERNAL | UMA_ZFLAG_CACHEONLY | UMA_ZFLAG_BUCKET) static inline uma_keg_t zone_first_keg(uma_zone_t zone) { uma_klink_t klink; klink = LIST_FIRST(&zone->uz_kegs); return (klink != NULL) ? klink->kl_keg : NULL; } #undef UMA_ALIGN #ifdef _KERNEL /* Internal prototypes */ static __inline uma_slab_t hash_sfind(struct uma_hash *hash, uint8_t *data); void *uma_large_malloc(vm_size_t size, int wait); void uma_large_free(uma_slab_t slab); /* Lock Macros */ #define KEG_LOCK_INIT(k, lc) \ do { \ if ((lc)) \ mtx_init(&(k)->uk_lock, (k)->uk_name, \ (k)->uk_name, MTX_DEF | MTX_DUPOK); \ else \ mtx_init(&(k)->uk_lock, (k)->uk_name, \ "UMA zone", MTX_DEF | MTX_DUPOK); \ } while (0) #define KEG_LOCK_FINI(k) mtx_destroy(&(k)->uk_lock) #define KEG_LOCK(k) mtx_lock(&(k)->uk_lock) #define KEG_UNLOCK(k) mtx_unlock(&(k)->uk_lock) #define ZONE_LOCK_INIT(z, lc) \ do { \ if ((lc)) \ mtx_init(&(z)->uz_lock, (z)->uz_name, \ (z)->uz_name, MTX_DEF | MTX_DUPOK); \ else \ mtx_init(&(z)->uz_lock, (z)->uz_name, \ "UMA zone", MTX_DEF | MTX_DUPOK); \ } while (0) #define ZONE_LOCK(z) mtx_lock((z)->uz_lockptr) #define ZONE_TRYLOCK(z) mtx_trylock((z)->uz_lockptr) #define ZONE_UNLOCK(z) mtx_unlock((z)->uz_lockptr) #define ZONE_LOCK_FINI(z) mtx_destroy(&(z)->uz_lock) /* * Find a slab within a hash table. This is used for OFFPAGE zones to lookup * the slab structure. * * Arguments: * hash The hash table to search. * data The base page of the item. * * Returns: * A pointer to a slab if successful, else NULL. */ static __inline uma_slab_t hash_sfind(struct uma_hash *hash, uint8_t *data) { uma_slab_t slab; int hval; hval = UMA_HASH(hash, data); SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) { if ((uint8_t *)slab->us_data == data) return (slab); } return (NULL); } static __inline uma_slab_t vtoslab(vm_offset_t va) { vm_page_t p; p = PHYS_TO_VM_PAGE(pmap_kextract(va)); return ((uma_slab_t)p->plinks.s.pv); } static __inline void vsetslab(vm_offset_t va, uma_slab_t slab) { vm_page_t p; p = PHYS_TO_VM_PAGE(pmap_kextract(va)); p->plinks.s.pv = slab; } /* * The following two functions may be defined by architecture specific code * if they can provide more efficient allocation functions. This is useful * for using direct mapped addresses. */ void *uma_small_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *pflag, int wait); void uma_small_free(void *mem, vm_size_t size, uint8_t flags); /* Set a global soft limit on UMA managed memory. */ void uma_set_limit(unsigned long limit); -unsigned long uma_limit(void); - -/* Return the amount of memory managed by UMA. */ -unsigned long uma_size(void); #endif /* _KERNEL */ #endif /* VM_UMA_INT_H */