*/
/*
* 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) 2018, Joyent, Inc.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
* Copyright (c) 2014 by Saso Kiselkov. All rights reserved.
- * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2017 Nexenta Systems, Inc. All rights reserved.
*/
/*
* 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.
+ *
+ * The L1ARC has a slightly different system for storing encrypted data.
+ * Raw (encrypted + possibly compressed) data has a few subtle differences from
+ * data that is just compressed. The biggest difference is that it is not
+ * possible to decrypt encrypted data (or visa versa) if the keys aren't loaded.
+ * The other difference is that encryption cannot be treated as a suggestion.
+ * If a caller would prefer compressed data, but they actually wind up with
+ * uncompressed data the worst thing that could happen is there might be a
+ * performance hit. If the caller requests encrypted data, however, we must be
+ * sure they actually get it or else secret information could be leaked. Raw
+ * data is stored in hdr->b_crypt_hdr.b_rabd. An encrypted header, therefore,
+ * may have both an encrypted version and a decrypted version of its data at
+ * once. When a caller needs a raw arc_buf_t, it is allocated and the data is
+ * copied out of this header. To avoid complications with b_pabd, raw buffers
+ * cannot be shared.
*/
#include <sys/spa.h>
#include <sys/zio_checksum.h>
#include <sys/multilist.h>
#include <sys/abd.h>
+#include <sys/zil.h>
+#include <sys/fm/fs/zfs.h>
#ifdef _KERNEL
+#include <sys/shrinker.h>
#include <sys/vmsystm.h>
-#include <vm/anon.h>
-#include <sys/fs/swapnode.h>
#include <sys/zpl.h>
-#include <linux/mm_compat.h>
+#include <linux/page_compat.h>
#endif
#include <sys/callb.h>
#include <sys/kstat.h>
-#include <sys/dmu_tx.h>
+#include <sys/zthr.h>
#include <zfs_fletcher.h>
#include <sys/arc_impl.h>
#include <sys/trace_arc.h>
+#include <sys/aggsum.h>
+#include <sys/cityhash.h>
#ifndef _KERNEL
/* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */
boolean_t arc_watch = B_FALSE;
#endif
-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;
+/*
+ * This thread's job is to keep enough free memory in the system, by
+ * calling arc_kmem_reap_soon() plus arc_reduce_target_size(), which improves
+ * arc_available_memory().
+ */
+static zthr_t *arc_reap_zthr;
+
+/*
+ * This thread's job is to keep arc_size under arc_c, by calling
+ * arc_adjust(), which improves arc_is_overflowing().
+ */
+static zthr_t *arc_adjust_zthr;
+
+static kmutex_t arc_adjust_lock;
+static kcondvar_t arc_adjust_waiters_cv;
+static boolean_t arc_adjust_needed = B_FALSE;
/*
* The number of headers to evict in arc_evict_state_impl() before
int zfs_arc_evict_batch_limit = 10;
/* number of seconds before growing cache again */
-static int arc_grow_retry = 5;
+static int arc_grow_retry = 5;
+
+/*
+ * Minimum time between calls to arc_kmem_reap_soon().
+ */
+int arc_kmem_cache_reap_retry_ms = 1000;
/* shift of arc_c for calculating overflow limit in arc_get_data_impl */
-int zfs_arc_overflow_shift = 8;
+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;
+int arc_p_min_shift = 4;
/* log2(fraction of arc to reclaim) */
-static int arc_shrink_shift = 7;
+static int arc_shrink_shift = 7;
+
+/* percent of pagecache to reclaim arc to */
+#ifdef _KERNEL
+static uint_t zfs_arc_pc_percent = 0;
+#endif
/*
* log2(fraction of ARC which must be free to allow growing).
* minimum lifespan of a prefetch block in clock ticks
* (initialized in arc_init())
*/
-static int arc_min_prefetch_lifespan;
+static int arc_min_prefetch_ms;
+static int arc_min_prescient_prefetch_ms;
/*
* If this percent of memory is free, don't throttle.
*/
int arc_lotsfree_percent = 10;
-static int arc_dead;
+/*
+ * hdr_recl() uses this to determine if the arc is up and running.
+ */
+static boolean_t arc_initialized;
/*
* The arc has filled available memory and has now warmed up.
int zfs_arc_p_min_shift = 0;
int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */
+/*
+ * ARC dirty data constraints for arc_tempreserve_space() throttle.
+ */
+unsigned long zfs_arc_dirty_limit_percent = 50; /* total dirty data limit */
+unsigned long zfs_arc_anon_limit_percent = 25; /* anon block dirty limit */
+unsigned long zfs_arc_pool_dirty_percent = 20; /* each pool's anon allowance */
+
+/*
+ * Enable or disable compressed arc buffers.
+ */
int zfs_compressed_arc_enabled = B_TRUE;
/*
* These tunables are Linux specific
*/
unsigned long zfs_arc_sys_free = 0;
-int zfs_arc_min_prefetch_lifespan = 0;
-int zfs_arc_p_aggressive_disable = 1;
+int zfs_arc_min_prefetch_ms = 0;
+int zfs_arc_min_prescient_prefetch_ms = 0;
int zfs_arc_p_dampener_disable = 1;
int zfs_arc_meta_prune = 10000;
int zfs_arc_meta_strategy = ARC_STRATEGY_META_BALANCED;
* by multiple buffers.
*/
kstat_named_t arcstat_mutex_miss;
+ /*
+ * Number of buffers skipped when updating the access state due to the
+ * header having already been released after acquiring the hash lock.
+ */
+ kstat_named_t arcstat_access_skip;
/*
* Number of buffers skipped because they have I/O in progress, are
- * indrect prefetch buffers that have not lived long enough, or are
+ * indirect 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;
kstat_named_t arcstat_c;
kstat_named_t arcstat_c_min;
kstat_named_t arcstat_c_max;
+ /* Not updated directly; only synced in arc_kstat_update. */
kstat_named_t arcstat_size;
/*
* Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd.
* (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).
+ * Not updated directly; only synced in arc_kstat_update.
*/
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).
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_data_size;
/*
* 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).
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_metadata_size;
/*
* Number of bytes consumed by dmu_buf_impl_t objects.
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_dbuf_size;
/*
* Number of bytes consumed by dnode_t objects.
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_dnode_size;
/*
* Number of bytes consumed by bonus buffers.
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_bonus_size;
/*
* 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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_anon_size;
/*
* 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).
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_anon_evictable_data;
/*
* 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).
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_anon_evictable_metadata;
/*
* 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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_size;
/*
* 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).
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_evictable_data;
/*
* 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).
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_evictable_metadata;
/*
* 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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_ghost_evictable_metadata;
/*
* 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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
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.
+ * Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mfu_ghost_evictable_metadata;
kstat_named_t arcstat_l2_hits;
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_size;
- kstat_named_t arcstat_l2_asize;
+ kstat_named_t arcstat_l2_lsize;
+ kstat_named_t arcstat_l2_psize;
+ /* Not updated directly; only synced in arc_kstat_update. */
kstat_named_t arcstat_l2_hdr_size;
kstat_named_t arcstat_memory_throttle_count;
kstat_named_t arcstat_memory_direct_count;
kstat_named_t arcstat_memory_indirect_count;
+ kstat_named_t arcstat_memory_all_bytes;
+ kstat_named_t arcstat_memory_free_bytes;
+ kstat_named_t arcstat_memory_available_bytes;
kstat_named_t arcstat_no_grow;
kstat_named_t arcstat_tempreserve;
kstat_named_t arcstat_loaned_bytes;
kstat_named_t arcstat_prune;
+ /* Not updated directly; only synced in arc_kstat_update. */
kstat_named_t arcstat_meta_used;
kstat_named_t arcstat_meta_limit;
kstat_named_t arcstat_dnode_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_async_upgrade_sync;
kstat_named_t arcstat_demand_hit_predictive_prefetch;
+ kstat_named_t arcstat_demand_hit_prescient_prefetch;
kstat_named_t arcstat_need_free;
kstat_named_t arcstat_sys_free;
+ kstat_named_t arcstat_raw_size;
} arc_stats_t;
static arc_stats_t arc_stats = {
{ "mfu_ghost_hits", KSTAT_DATA_UINT64 },
{ "deleted", KSTAT_DATA_UINT64 },
{ "mutex_miss", KSTAT_DATA_UINT64 },
+ { "access_skip", KSTAT_DATA_UINT64 },
{ "evict_skip", KSTAT_DATA_UINT64 },
{ "evict_not_enough", KSTAT_DATA_UINT64 },
{ "evict_l2_cached", KSTAT_DATA_UINT64 },
{ "memory_throttle_count", KSTAT_DATA_UINT64 },
{ "memory_direct_count", KSTAT_DATA_UINT64 },
{ "memory_indirect_count", KSTAT_DATA_UINT64 },
+ { "memory_all_bytes", KSTAT_DATA_UINT64 },
+ { "memory_free_bytes", KSTAT_DATA_UINT64 },
+ { "memory_available_bytes", KSTAT_DATA_INT64 },
{ "arc_no_grow", KSTAT_DATA_UINT64 },
{ "arc_tempreserve", KSTAT_DATA_UINT64 },
{ "arc_loaned_bytes", KSTAT_DATA_UINT64 },
{ "arc_dnode_limit", KSTAT_DATA_UINT64 },
{ "arc_meta_max", KSTAT_DATA_UINT64 },
{ "arc_meta_min", KSTAT_DATA_UINT64 },
- { "sync_wait_for_async", KSTAT_DATA_UINT64 },
+ { "async_upgrade_sync", KSTAT_DATA_UINT64 },
{ "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64 },
+ { "demand_hit_prescient_prefetch", KSTAT_DATA_UINT64 },
{ "arc_need_free", KSTAT_DATA_UINT64 },
- { "arc_sys_free", KSTAT_DATA_UINT64 }
+ { "arc_sys_free", KSTAT_DATA_UINT64 },
+ { "arc_raw_size", KSTAT_DATA_UINT64 }
};
#define ARCSTAT(stat) (arc_stats.stat.value.ui64)
* 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_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */
#define arc_dnode_limit ARCSTAT(arcstat_dnode_limit) /* max size for dnodes */
#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 */
-#define arc_dbuf_size ARCSTAT(arcstat_dbuf_size) /* dbuf metadata */
-#define arc_dnode_size ARCSTAT(arcstat_dnode_size) /* dnode metadata */
-#define arc_bonus_size ARCSTAT(arcstat_bonus_size) /* bonus buffer metadata */
#define arc_need_free ARCSTAT(arcstat_need_free) /* bytes to be freed */
#define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */
+/* size of all b_rabd's in entire arc */
+#define arc_raw_size ARCSTAT(arcstat_raw_size)
/* compressed size of entire arc */
#define arc_compressed_size ARCSTAT(arcstat_compressed_size)
/* uncompressed size of entire arc */
/* number of bytes in the arc from arc_buf_t's */
#define arc_overhead_size ARCSTAT(arcstat_overhead_size)
+/*
+ * There are also some ARC variables that we want to export, but that are
+ * updated so often that having the canonical representation be the statistic
+ * variable causes a performance bottleneck. We want to use aggsum_t's for these
+ * instead, but still be able to export the kstat in the same way as before.
+ * The solution is to always use the aggsum version, except in the kstat update
+ * callback.
+ */
+aggsum_t arc_size;
+aggsum_t arc_meta_used;
+aggsum_t astat_data_size;
+aggsum_t astat_metadata_size;
+aggsum_t astat_dbuf_size;
+aggsum_t astat_dnode_size;
+aggsum_t astat_bonus_size;
+aggsum_t astat_hdr_size;
+aggsum_t astat_l2_hdr_size;
+
+static hrtime_t arc_growtime;
static list_t arc_prune_list;
static kmutex_t arc_prune_mtx;
static taskq_t *arc_prune_taskq;
#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_PRESCIENT_PREFETCH(hdr) \
+ ((hdr)->b_flags & ARC_FLAG_PRESCIENT_PREFETCH)
#define HDR_COMPRESSION_ENABLED(hdr) \
((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC)
#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_PROTECTED(hdr) ((hdr)->b_flags & ARC_FLAG_PROTECTED)
+#define HDR_NOAUTH(hdr) ((hdr)->b_flags & ARC_FLAG_NOAUTH)
#define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA)
#define 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)
+#define HDR_HAS_RABD(hdr) \
+ (HDR_HAS_L1HDR(hdr) && HDR_PROTECTED(hdr) && \
+ (hdr)->b_crypt_hdr.b_rabd != NULL)
+#define HDR_ENCRYPTED(hdr) \
+ (HDR_PROTECTED(hdr) && DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot))
+#define HDR_AUTHENTICATED(hdr) \
+ (HDR_PROTECTED(hdr) && !DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot))
/* For storing compression mode in b_flags */
#define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1)
#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)
+#define ARC_BUF_ENCRYPTED(buf) ((buf)->b_flags & ARC_BUF_FLAG_ENCRYPTED)
/*
* Other sizes
*/
-#define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t))
+#define HDR_FULL_CRYPT_SIZE ((int64_t)sizeof (arc_buf_hdr_t))
+#define HDR_FULL_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_crypt_hdr))
#define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr))
/*
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_data_free {
list_node_t l2df_list_node;
} l2arc_data_free_t;
+typedef enum arc_fill_flags {
+ ARC_FILL_LOCKED = 1 << 0, /* hdr lock is held */
+ ARC_FILL_COMPRESSED = 1 << 1, /* fill with compressed data */
+ ARC_FILL_ENCRYPTED = 1 << 2, /* fill with encrypted data */
+ ARC_FILL_NOAUTH = 1 << 3, /* don't attempt to authenticate */
+ ARC_FILL_IN_PLACE = 1 << 4 /* fill in place (special case) */
+} arc_fill_flags_t;
+
static kmutex_t l2arc_feed_thr_lock;
static kcondvar_t l2arc_feed_thr_cv;
static uint8_t l2arc_thread_exit;
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_hdr_free_abd(arc_buf_hdr_t *, boolean_t);
+static void arc_hdr_alloc_abd(arc_buf_hdr_t *, boolean_t);
static void arc_access(arc_buf_hdr_t *, kmutex_t *);
static boolean_t arc_is_overflowing(void);
static void arc_buf_watch(arc_buf_t *);
static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *);
static void l2arc_read_done(zio_t *);
+
+/*
+ * We use Cityhash for this. It's fast, and has good hash properties without
+ * requiring any large static buffers.
+ */
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);
+ return (cityhash4(spa, dva->dva_word[0], dva->dva_word[1], birth));
}
#define HDR_EMPTY(hdr) \
/*
* Global data structures and functions for the buf kmem cache.
*/
+
static kmem_cache_t *hdr_full_cache;
+static kmem_cache_t *hdr_full_crypt_cache;
static kmem_cache_t *hdr_l2only_cache;
static kmem_cache_t *buf_cache;
{
int i;
-#if defined(_KERNEL) && defined(HAVE_SPL)
+#if defined(_KERNEL)
/*
* Large allocations which do not require contiguous pages
* should be using vmem_free() in the linux kernel\
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_full_crypt_cache);
kmem_cache_destroy(hdr_l2only_cache);
kmem_cache_destroy(buf_cache);
}
arc_buf_hdr_t *hdr = vbuf;
bzero(hdr, HDR_FULL_SIZE);
+ hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL);
- refcount_create(&hdr->b_l1hdr.b_refcnt);
+ zfs_refcount_create(&hdr->b_l1hdr.b_refcnt);
mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL);
list_link_init(&hdr->b_l1hdr.b_arc_node);
list_link_init(&hdr->b_l2hdr.b_l2node);
return (0);
}
+/* ARGSUSED */
+static int
+hdr_full_crypt_cons(void *vbuf, void *unused, int kmflag)
+{
+ arc_buf_hdr_t *hdr = vbuf;
+
+ hdr_full_cons(vbuf, unused, kmflag);
+ bzero(&hdr->b_crypt_hdr, sizeof (hdr->b_crypt_hdr));
+ arc_space_consume(sizeof (hdr->b_crypt_hdr), ARC_SPACE_HDRS);
+
+ return (0);
+}
+
/* ARGSUSED */
static int
hdr_l2only_cons(void *vbuf, void *unused, int kmflag)
ASSERT(HDR_EMPTY(hdr));
cv_destroy(&hdr->b_l1hdr.b_cv);
- refcount_destroy(&hdr->b_l1hdr.b_refcnt);
+ zfs_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_full_crypt_dest(void *vbuf, void *unused)
+{
+ arc_buf_hdr_t *hdr = vbuf;
+
+ hdr_full_dest(vbuf, unused);
+ arc_space_return(sizeof (hdr->b_crypt_hdr), ARC_SPACE_HDRS);
+}
+
/* ARGSUSED */
static void
hdr_l2only_dest(void *vbuf, void *unused)
* 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);
+ if (arc_initialized)
+ zthr_wakeup(arc_reap_zthr);
}
static void
hsize <<= 1;
retry:
buf_hash_table.ht_mask = hsize - 1;
-#if defined(_KERNEL) && defined(HAVE_SPL)
+#if defined(_KERNEL)
/*
* Large allocations which do not require contiguous pages
* should be using vmem_alloc() in the linux kernel
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_full_crypt_cache = kmem_cache_create("arc_buf_hdr_t_full_crypt",
+ HDR_FULL_CRYPT_SIZE, 0, hdr_full_crypt_cons, hdr_full_crypt_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);
return (HDR_GET_LSIZE(buf->b_hdr));
}
+/*
+ * This function will return B_TRUE if the buffer is encrypted in memory.
+ * This buffer can be decrypted by calling arc_untransform().
+ */
+boolean_t
+arc_is_encrypted(arc_buf_t *buf)
+{
+ return (ARC_BUF_ENCRYPTED(buf) != 0);
+}
+
+/*
+ * Returns B_TRUE if the buffer represents data that has not had its MAC
+ * verified yet.
+ */
+boolean_t
+arc_is_unauthenticated(arc_buf_t *buf)
+{
+ return (HDR_NOAUTH(buf->b_hdr) != 0);
+}
+
+void
+arc_get_raw_params(arc_buf_t *buf, boolean_t *byteorder, uint8_t *salt,
+ uint8_t *iv, uint8_t *mac)
+{
+ arc_buf_hdr_t *hdr = buf->b_hdr;
+
+ ASSERT(HDR_PROTECTED(hdr));
+
+ bcopy(hdr->b_crypt_hdr.b_salt, salt, ZIO_DATA_SALT_LEN);
+ bcopy(hdr->b_crypt_hdr.b_iv, iv, ZIO_DATA_IV_LEN);
+ bcopy(hdr->b_crypt_hdr.b_mac, mac, ZIO_DATA_MAC_LEN);
+ *byteorder = (hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS) ?
+ ZFS_HOST_BYTEORDER : !ZFS_HOST_BYTEORDER;
+}
+
+/*
+ * Indicates how this buffer is compressed in memory. If it is not compressed
+ * the value will be ZIO_COMPRESS_OFF. It can be made normally readable with
+ * arc_untransform() as long as it is also unencrypted.
+ */
enum zio_compress
arc_get_compression(arc_buf_t *buf)
{
HDR_GET_COMPRESS(buf->b_hdr) : ZIO_COMPRESS_OFF);
}
+/*
+ * Return the compression algorithm used to store this data in the ARC. If ARC
+ * compression is enabled or this is an encrypted block, this will be the same
+ * as what's used to store it on-disk. Otherwise, this will be ZIO_COMPRESS_OFF.
+ */
+static inline enum zio_compress
+arc_hdr_get_compress(arc_buf_hdr_t *hdr)
+{
+ return (HDR_COMPRESSION_ENABLED(hdr) ?
+ HDR_GET_COMPRESS(hdr) : ZIO_COMPRESS_OFF);
+}
+
static inline boolean_t
arc_buf_is_shared(arc_buf_t *buf)
{
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));
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
}
+/*
+ * Return true iff at least one of the bufs on hdr is not compressed.
+ * Encrypted buffers count as compressed.
+ */
+static boolean_t
+arc_hdr_has_uncompressed_buf(arc_buf_hdr_t *hdr)
+{
+ ASSERT(hdr->b_l1hdr.b_state == arc_anon ||
+ MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(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,
if (!(zfs_flags & ZFS_DEBUG_MODIFY))
return;
- if (ARC_BUF_COMPRESSED(buf)) {
+ if (ARC_BUF_COMPRESSED(buf))
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;
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
}
+/*
+ * This function makes the assumption that data stored in the L2ARC
+ * will be transformed exactly as it is in the main pool. Because of
+ * this we can verify the checksum against the reading process's bp.
+ */
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) {
- uint64_t lsize;
- uint64_t csize;
- void *cbuf;
- ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF);
-
- cbuf = zio_buf_alloc(HDR_GET_PSIZE(hdr));
- lsize = HDR_GET_LSIZE(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
* 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,
+ return (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);
}
/*
ASSERT(HDR_HAS_L1HDR(hdr));
mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock);
- if (hdr->b_l1hdr.b_freeze_cksum != NULL) {
- mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
- return;
- } else if (ARC_BUF_COMPRESSED(buf)) {
+ if (hdr->b_l1hdr.b_freeze_cksum != NULL || ARC_BUF_COMPRESSED(buf)) {
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
return;
}
+ ASSERT(!ARC_BUF_ENCRYPTED(buf));
ASSERT(!ARC_BUF_COMPRESSED(buf));
hdr->b_l1hdr.b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t),
KM_SLEEP);
{
#ifndef _KERNEL
if (arc_watch) {
- ASSERT0(mprotect(buf->b_data, HDR_GET_LSIZE(buf->b_hdr),
+ ASSERT0(mprotect(buf->b_data, arc_buf_size(buf),
PROT_READ | PROT_WRITE));
}
#endif
arc_cksum_verify(buf);
/*
- * Compressed buffers do not manipulate the b_freeze_cksum or
- * allocate b_thawed.
+ * Compressed buffers do not manipulate the b_freeze_cksum.
*/
- if (ARC_BUF_COMPRESSED(buf)) {
+ if (ARC_BUF_COMPRESSED(buf))
return;
- }
ASSERT(HDR_HAS_L1HDR(hdr));
arc_cksum_free(hdr);
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)) {
+ if (ARC_BUF_COMPRESSED(buf))
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);
+ ASSERT(HDR_HAS_L1HDR(buf->b_hdr));
arc_cksum_compute(buf);
- mutex_exit(hash_lock);
}
/*
*/
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));
}
+
+ HDR_SET_COMPRESS(hdr, cmp);
+ ASSERT3U(HDR_GET_COMPRESS(hdr), ==, cmp);
}
/*
arc_buf_try_copy_decompressed_data(arc_buf_t *buf)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
- arc_buf_t *from;
boolean_t copied = B_FALSE;
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT3P(buf->b_data, !=, NULL);
ASSERT(!ARC_BUF_COMPRESSED(buf));
- for (from = hdr->b_l1hdr.b_buf; from != NULL;
+ 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) {
return (copied);
}
+/*
+ * 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 (arc_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);
+}
+
+static int
+arc_hdr_authenticate(arc_buf_hdr_t *hdr, spa_t *spa, uint64_t dsobj)
+{
+ int ret;
+ uint64_t csize;
+ uint64_t lsize = HDR_GET_LSIZE(hdr);
+ uint64_t psize = HDR_GET_PSIZE(hdr);
+ void *tmpbuf = NULL;
+ abd_t *abd = hdr->b_l1hdr.b_pabd;
+
+ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr));
+ ASSERT(HDR_AUTHENTICATED(hdr));
+ ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+
+ /*
+ * The MAC is calculated on the compressed data that is stored on disk.
+ * However, if compressed arc is disabled we will only have the
+ * decompressed data available to us now. Compress it into a temporary
+ * abd so we can verify the MAC. The performance overhead of this will
+ * be relatively low, since most objects in an encrypted objset will
+ * be encrypted (instead of authenticated) anyway.
+ */
+ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
+ !HDR_COMPRESSION_ENABLED(hdr)) {
+ tmpbuf = zio_buf_alloc(lsize);
+ abd = abd_get_from_buf(tmpbuf, lsize);
+ abd_take_ownership_of_buf(abd, B_TRUE);
+
+ csize = zio_compress_data(HDR_GET_COMPRESS(hdr),
+ hdr->b_l1hdr.b_pabd, tmpbuf, lsize);
+ ASSERT3U(csize, <=, psize);
+ abd_zero_off(abd, csize, psize - csize);
+ }
+
+ /*
+ * Authentication is best effort. We authenticate whenever the key is
+ * available. If we succeed we clear ARC_FLAG_NOAUTH.
+ */
+ if (hdr->b_crypt_hdr.b_ot == DMU_OT_OBJSET) {
+ ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF);
+ ASSERT3U(lsize, ==, psize);
+ ret = spa_do_crypt_objset_mac_abd(B_FALSE, spa, dsobj, abd,
+ psize, hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS);
+ } else {
+ ret = spa_do_crypt_mac_abd(B_FALSE, spa, dsobj, abd, psize,
+ hdr->b_crypt_hdr.b_mac);
+ }
+
+ if (ret == 0)
+ arc_hdr_clear_flags(hdr, ARC_FLAG_NOAUTH);
+ else if (ret != ENOENT)
+ goto error;
+
+ if (tmpbuf != NULL)
+ abd_free(abd);
+
+ return (0);
+
+error:
+ if (tmpbuf != NULL)
+ abd_free(abd);
+
+ return (ret);
+}
+
+/*
+ * This function will take a header that only has raw encrypted data in
+ * b_crypt_hdr.b_rabd and decrypt it into a new buffer which is stored in
+ * b_l1hdr.b_pabd. If designated in the header flags, this function will
+ * also decompress the data.
+ */
+static int
+arc_hdr_decrypt(arc_buf_hdr_t *hdr, spa_t *spa, const zbookmark_phys_t *zb)
+{
+ int ret;
+ abd_t *cabd = NULL;
+ void *tmp = NULL;
+ boolean_t no_crypt = B_FALSE;
+ boolean_t bswap = (hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS);
+
+ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr));
+ ASSERT(HDR_ENCRYPTED(hdr));
+
+ arc_hdr_alloc_abd(hdr, B_FALSE);
+
+ ret = spa_do_crypt_abd(B_FALSE, spa, zb, hdr->b_crypt_hdr.b_ot,
+ B_FALSE, bswap, hdr->b_crypt_hdr.b_salt, hdr->b_crypt_hdr.b_iv,
+ hdr->b_crypt_hdr.b_mac, HDR_GET_PSIZE(hdr), hdr->b_l1hdr.b_pabd,
+ hdr->b_crypt_hdr.b_rabd, &no_crypt);
+ if (ret != 0)
+ goto error;
+
+ if (no_crypt) {
+ abd_copy(hdr->b_l1hdr.b_pabd, hdr->b_crypt_hdr.b_rabd,
+ HDR_GET_PSIZE(hdr));
+ }
+
+ /*
+ * If this header has disabled arc compression but the b_pabd is
+ * compressed after decrypting it, we need to decompress the newly
+ * decrypted data.
+ */
+ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
+ !HDR_COMPRESSION_ENABLED(hdr)) {
+ /*
+ * We want to make sure that we are correctly honoring the
+ * zfs_abd_scatter_enabled setting, so we allocate an abd here
+ * and then loan a buffer from it, rather than allocating a
+ * linear buffer and wrapping it in an abd later.
+ */
+ cabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr);
+ tmp = abd_borrow_buf(cabd, arc_hdr_size(hdr));
+
+ ret = zio_decompress_data(HDR_GET_COMPRESS(hdr),
+ hdr->b_l1hdr.b_pabd, tmp, HDR_GET_PSIZE(hdr),
+ HDR_GET_LSIZE(hdr));
+ if (ret != 0) {
+ abd_return_buf(cabd, tmp, arc_hdr_size(hdr));
+ goto error;
+ }
+
+ abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr));
+ arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd,
+ arc_hdr_size(hdr), hdr);
+ hdr->b_l1hdr.b_pabd = cabd;
+ }
+
+ return (0);
+
+error:
+ arc_hdr_free_abd(hdr, B_FALSE);
+ if (cabd != NULL)
+ arc_free_data_buf(hdr, cabd, arc_hdr_size(hdr), hdr);
+
+ return (ret);
+}
+
+/*
+ * This function is called during arc_buf_fill() to prepare the header's
+ * abd plaintext pointer for use. This involves authenticated protected
+ * data and decrypting encrypted data into the plaintext abd.
+ */
+static int
+arc_fill_hdr_crypt(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, spa_t *spa,
+ const zbookmark_phys_t *zb, boolean_t noauth)
+{
+ int ret;
+
+ ASSERT(HDR_PROTECTED(hdr));
+
+ if (hash_lock != NULL)
+ mutex_enter(hash_lock);
+
+ if (HDR_NOAUTH(hdr) && !noauth) {
+ /*
+ * The caller requested authenticated data but our data has
+ * not been authenticated yet. Verify the MAC now if we can.
+ */
+ ret = arc_hdr_authenticate(hdr, spa, zb->zb_objset);
+ if (ret != 0)
+ goto error;
+ } else if (HDR_HAS_RABD(hdr) && hdr->b_l1hdr.b_pabd == NULL) {
+ /*
+ * If we only have the encrypted version of the data, but the
+ * unencrypted version was requested we take this opportunity
+ * to store the decrypted version in the header for future use.
+ */
+ ret = arc_hdr_decrypt(hdr, spa, zb);
+ if (ret != 0)
+ goto error;
+ }
+
+ ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+
+ if (hash_lock != NULL)
+ mutex_exit(hash_lock);
+
+ return (0);
+
+error:
+ if (hash_lock != NULL)
+ mutex_exit(hash_lock);
+
+ return (ret);
+}
+
+/*
+ * This function is used by the dbuf code to decrypt bonus buffers in place.
+ * The dbuf code itself doesn't have any locking for decrypting a shared dnode
+ * block, so we use the hash lock here to protect against concurrent calls to
+ * arc_buf_fill().
+ */
+static void
+arc_buf_untransform_in_place(arc_buf_t *buf, kmutex_t *hash_lock)
+{
+ arc_buf_hdr_t *hdr = buf->b_hdr;
+
+ ASSERT(HDR_ENCRYPTED(hdr));
+ ASSERT3U(hdr->b_crypt_hdr.b_ot, ==, DMU_OT_DNODE);
+ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr));
+ ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+
+ zio_crypt_copy_dnode_bonus(hdr->b_l1hdr.b_pabd, buf->b_data,
+ arc_buf_size(buf));
+ buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED;
+ buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED;
+ hdr->b_crypt_hdr.b_ebufcnt -= 1;
+}
+
/*
* 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 correct-sized data buffer.
*/
static int
-arc_buf_fill(arc_buf_t *buf, boolean_t compressed)
+arc_buf_fill(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb,
+ arc_fill_flags_t flags)
{
+ int error = 0;
arc_buf_hdr_t *hdr = buf->b_hdr;
- boolean_t hdr_compressed = (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF);
+ boolean_t hdr_compressed =
+ (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF);
+ boolean_t compressed = (flags & ARC_FILL_COMPRESSED) != 0;
+ boolean_t encrypted = (flags & ARC_FILL_ENCRYPTED) != 0;
dmu_object_byteswap_t bswap = hdr->b_l1hdr.b_byteswap;
+ kmutex_t *hash_lock = (flags & ARC_FILL_LOCKED) ? NULL : HDR_LOCK(hdr);
ASSERT3P(buf->b_data, !=, NULL);
- IMPLY(compressed, hdr_compressed);
+ IMPLY(compressed, hdr_compressed || ARC_BUF_ENCRYPTED(buf));
IMPLY(compressed, ARC_BUF_COMPRESSED(buf));
+ IMPLY(encrypted, HDR_ENCRYPTED(hdr));
+ IMPLY(encrypted, ARC_BUF_ENCRYPTED(buf));
+ IMPLY(encrypted, ARC_BUF_COMPRESSED(buf));
+ IMPLY(encrypted, !ARC_BUF_SHARED(buf));
+
+ /*
+ * If the caller wanted encrypted data we just need to copy it from
+ * b_rabd and potentially byteswap it. We won't be able to do any
+ * further transforms on it.
+ */
+ if (encrypted) {
+ ASSERT(HDR_HAS_RABD(hdr));
+ abd_copy_to_buf(buf->b_data, hdr->b_crypt_hdr.b_rabd,
+ HDR_GET_PSIZE(hdr));
+ goto byteswap;
+ }
+
+ /*
+ * Adjust encrypted and authenticated headers to accomodate
+ * the request if needed. Dnode blocks (ARC_FILL_IN_PLACE) are
+ * allowed to fail decryption due to keys not being loaded
+ * without being marked as an IO error.
+ */
+ if (HDR_PROTECTED(hdr)) {
+ error = arc_fill_hdr_crypt(hdr, hash_lock, spa,
+ zb, !!(flags & ARC_FILL_NOAUTH));
+ if (error == EACCES && (flags & ARC_FILL_IN_PLACE) != 0) {
+ return (error);
+ } else if (error != 0) {
+ if (hash_lock != NULL)
+ mutex_enter(hash_lock);
+ arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR);
+ if (hash_lock != NULL)
+ mutex_exit(hash_lock);
+ return (error);
+ }
+ }
+
+ /*
+ * There is a special case here for dnode blocks which are
+ * decrypting their bonus buffers. These blocks may request to
+ * be decrypted in-place. This is necessary because there may
+ * be many dnodes pointing into this buffer and there is
+ * currently no method to synchronize replacing the backing
+ * b_data buffer and updating all of the pointers. Here we use
+ * the hash lock to ensure there are no races. If the need
+ * arises for other types to be decrypted in-place, they must
+ * add handling here as well.
+ */
+ if ((flags & ARC_FILL_IN_PLACE) != 0) {
+ ASSERT(!hdr_compressed);
+ ASSERT(!compressed);
+ ASSERT(!encrypted);
+
+ if (HDR_ENCRYPTED(hdr) && ARC_BUF_ENCRYPTED(buf)) {
+ ASSERT3U(hdr->b_crypt_hdr.b_ot, ==, DMU_OT_DNODE);
+
+ if (hash_lock != NULL)
+ mutex_enter(hash_lock);
+ arc_buf_untransform_in_place(buf, hash_lock);
+ if (hash_lock != NULL)
+ mutex_exit(hash_lock);
+
+ /* Compute the hdr's checksum if necessary */
+ arc_cksum_compute(buf);
+ }
+
+ return (0);
+ }
if (hdr_compressed == compressed) {
if (!arc_buf_is_shared(buf)) {
ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, !=, NULL);
return (0);
} else {
- int error = zio_decompress_data(HDR_GET_COMPRESS(hdr),
+ error = zio_decompress_data(HDR_GET_COMPRESS(hdr),
hdr->b_l1hdr.b_pabd, buf->b_data,
HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr));
if (error != 0) {
zfs_dbgmsg(
"hdr %p, compress %d, psize %d, lsize %d",
- hdr, HDR_GET_COMPRESS(hdr),
+ hdr, arc_hdr_get_compress(hdr),
HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr));
+ if (hash_lock != NULL)
+ mutex_enter(hash_lock);
+ arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR);
+ if (hash_lock != NULL)
+ mutex_exit(hash_lock);
return (SET_ERROR(EIO));
}
}
}
+byteswap:
/* Byteswap the buf's data if necessary */
if (bswap != DMU_BSWAP_NUMFUNCS) {
ASSERT(!HDR_SHARED_DATA(hdr));
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.
+ * If this function is being called to decrypt an encrypted buffer or verify an
+ * authenticated one, the key must be loaded and a mapping must be made
+ * available in the keystore via spa_keystore_create_mapping() or one of its
+ * callers.
*/
-static uint64_t
-arc_hdr_size(arc_buf_hdr_t *hdr)
+int
+arc_untransform(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb,
+ boolean_t in_place)
{
- uint64_t size;
+ int ret;
+ arc_fill_flags_t flags = 0;
- 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);
+ if (in_place)
+ flags |= ARC_FILL_IN_PLACE;
+
+ ret = arc_buf_fill(buf, spa, zb, flags);
+ if (ret == ECKSUM) {
+ /*
+ * Convert authentication and decryption errors to EIO
+ * (and generate an ereport) before leaving the ARC.
+ */
+ ret = SET_ERROR(EIO);
+ spa_log_error(spa, zb);
+ zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION,
+ spa, NULL, zb, NULL, 0, 0);
}
- return (size);
+
+ return (ret);
}
/*
arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state)
{
arc_buf_contents_t type = arc_buf_type(hdr);
- arc_buf_t *buf;
ASSERT(HDR_HAS_L1HDR(hdr));
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],
+ ASSERT(!HDR_HAS_RABD(hdr));
+ (void) zfs_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],
+ (void) zfs_refcount_add_many(&state->arcs_esize[type],
arc_hdr_size(hdr), hdr);
}
- for (buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) {
+ if (HDR_HAS_RABD(hdr)) {
+ (void) zfs_refcount_add_many(&state->arcs_esize[type],
+ HDR_GET_PSIZE(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],
+ (void) zfs_refcount_add_many(&state->arcs_esize[type],
arc_buf_size(buf), buf);
}
}
arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state)
{
arc_buf_contents_t type = arc_buf_type(hdr);
- arc_buf_t *buf;
ASSERT(HDR_HAS_L1HDR(hdr));
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],
+ ASSERT(!HDR_HAS_RABD(hdr));
+ (void) zfs_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],
+ (void) zfs_refcount_remove_many(&state->arcs_esize[type],
arc_hdr_size(hdr), hdr);
}
- for (buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) {
+ if (HDR_HAS_RABD(hdr)) {
+ (void) zfs_refcount_remove_many(&state->arcs_esize[type],
+ HDR_GET_PSIZE(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],
+ (void) zfs_refcount_remove_many(&state->arcs_esize[type],
arc_buf_size(buf), buf);
}
}
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));
+ ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
}
state = hdr->b_l1hdr.b_state;
- if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) &&
+ if ((zfs_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)],
+ multilist_remove(state->arcs_list[arc_buf_type(hdr)],
hdr);
arc_evictable_space_decrement(hdr, 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) &&
+ if (((cnt = zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) &&
(state != arc_anon)) {
- multilist_insert(&state->arcs_list[arc_buf_type(hdr)], hdr);
+ multilist_insert(state->arcs_list[arc_buf_type(hdr)], hdr);
ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0);
arc_evictable_space_increment(hdr, state);
}
abi->abi_mru_ghost_hits = l1hdr->b_mru_ghost_hits;
abi->abi_mfu_hits = l1hdr->b_mfu_hits;
abi->abi_mfu_ghost_hits = l1hdr->b_mfu_ghost_hits;
- abi->abi_holds = refcount_count(&l1hdr->b_refcnt);
+ abi->abi_holds = zfs_refcount_count(&l1hdr->b_refcnt);
}
if (l2hdr) {
*/
if (HDR_HAS_L1HDR(hdr)) {
old_state = hdr->b_l1hdr.b_state;
- refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt);
+ refcnt = zfs_refcount_count(&hdr->b_l1hdr.b_refcnt);
bufcnt = hdr->b_l1hdr.b_bufcnt;
- update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pabd != NULL);
+ update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pabd != NULL ||
+ HDR_HAS_RABD(hdr));
} else {
old_state = arc_l2c_only;
refcnt = 0;
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);
+ multilist_remove(old_state->arcs_list[buftype], hdr);
if (GHOST_STATE(old_state)) {
ASSERT0(bufcnt);
* beforehand.
*/
ASSERT(HDR_HAS_L1HDR(hdr));
- multilist_insert(&new_state->arcs_list[buftype], hdr);
+ multilist_insert(new_state->arcs_list[buftype], hdr);
if (GHOST_STATE(new_state)) {
ASSERT0(bufcnt);
* the reference. As a result, we use the arc
* header pointer for the reference.
*/
- (void) refcount_add_many(&new_state->arcs_size,
+ (void) zfs_refcount_add_many(&new_state->arcs_size,
HDR_GET_LSIZE(hdr), hdr);
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
+ ASSERT(!HDR_HAS_RABD(hdr));
} else {
- arc_buf_t *buf;
uint32_t buffers = 0;
/*
* thus we must remove each of these references one
* at a time.
*/
- for (buf = hdr->b_l1hdr.b_buf; buf != NULL;
+ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL;
buf = buf->b_next) {
ASSERT3U(bufcnt, !=, 0);
buffers++;
if (arc_buf_is_shared(buf))
continue;
- (void) refcount_add_many(&new_state->arcs_size,
+ (void) zfs_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,
+ (void) zfs_refcount_add_many(
+ &new_state->arcs_size,
arc_hdr_size(hdr), hdr);
- } else {
- ASSERT(GHOST_STATE(old_state));
+ }
+
+ if (HDR_HAS_RABD(hdr)) {
+ (void) zfs_refcount_add_many(
+ &new_state->arcs_size,
+ HDR_GET_PSIZE(hdr), hdr);
}
}
}
if (GHOST_STATE(old_state)) {
ASSERT0(bufcnt);
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
+ ASSERT(!HDR_HAS_RABD(hdr));
/*
* When moving a header off of a ghost state,
* header on the ghost state.
*/
- (void) refcount_remove_many(&old_state->arcs_size,
+ (void) zfs_refcount_remove_many(&old_state->arcs_size,
HDR_GET_LSIZE(hdr), hdr);
} else {
- arc_buf_t *buf;
uint32_t buffers = 0;
/*
* thus we must remove each of these references one
* at a time.
*/
- for (buf = hdr->b_l1hdr.b_buf; buf != NULL;
+ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL;
buf = buf->b_next) {
ASSERT3U(bufcnt, !=, 0);
buffers++;
if (arc_buf_is_shared(buf))
continue;
- (void) refcount_remove_many(
+ (void) zfs_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);
+ ASSERT(hdr->b_l1hdr.b_pabd != NULL ||
+ HDR_HAS_RABD(hdr));
+
+ if (hdr->b_l1hdr.b_pabd != NULL) {
+ (void) zfs_refcount_remove_many(
+ &old_state->arcs_size, arc_hdr_size(hdr),
+ hdr);
+ }
+
+ if (HDR_HAS_RABD(hdr)) {
+ (void) zfs_refcount_remove_many(
+ &old_state->arcs_size, HDR_GET_PSIZE(hdr),
+ hdr);
+ }
}
}
* 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]));
+ ASSERT(multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_DATA]) &&
+ multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_METADATA]));
}
void
default:
break;
case ARC_SPACE_DATA:
- ARCSTAT_INCR(arcstat_data_size, space);
+ aggsum_add(&astat_data_size, space);
break;
case ARC_SPACE_META:
- ARCSTAT_INCR(arcstat_metadata_size, space);
+ aggsum_add(&astat_metadata_size, space);
break;
case ARC_SPACE_BONUS:
- ARCSTAT_INCR(arcstat_bonus_size, space);
+ aggsum_add(&astat_bonus_size, space);
break;
case ARC_SPACE_DNODE:
- ARCSTAT_INCR(arcstat_dnode_size, space);
+ aggsum_add(&astat_dnode_size, space);
break;
case ARC_SPACE_DBUF:
- ARCSTAT_INCR(arcstat_dbuf_size, space);
+ aggsum_add(&astat_dbuf_size, space);
break;
case ARC_SPACE_HDRS:
- ARCSTAT_INCR(arcstat_hdr_size, space);
+ aggsum_add(&astat_hdr_size, space);
break;
case ARC_SPACE_L2HDRS:
- ARCSTAT_INCR(arcstat_l2_hdr_size, space);
+ aggsum_add(&astat_l2_hdr_size, space);
break;
}
if (type != ARC_SPACE_DATA)
- ARCSTAT_INCR(arcstat_meta_used, space);
+ aggsum_add(&arc_meta_used, space);
- atomic_add_64(&arc_size, space);
+ aggsum_add(&arc_size, space);
}
void
default:
break;
case ARC_SPACE_DATA:
- ARCSTAT_INCR(arcstat_data_size, -space);
+ aggsum_add(&astat_data_size, -space);
break;
case ARC_SPACE_META:
- ARCSTAT_INCR(arcstat_metadata_size, -space);
+ aggsum_add(&astat_metadata_size, -space);
break;
case ARC_SPACE_BONUS:
- ARCSTAT_INCR(arcstat_bonus_size, -space);
+ aggsum_add(&astat_bonus_size, -space);
break;
case ARC_SPACE_DNODE:
- ARCSTAT_INCR(arcstat_dnode_size, -space);
+ aggsum_add(&astat_dnode_size, -space);
break;
case ARC_SPACE_DBUF:
- ARCSTAT_INCR(arcstat_dbuf_size, -space);
+ aggsum_add(&astat_dbuf_size, -space);
break;
case ARC_SPACE_HDRS:
- ARCSTAT_INCR(arcstat_hdr_size, -space);
+ aggsum_add(&astat_hdr_size, -space);
break;
case ARC_SPACE_L2HDRS:
- ARCSTAT_INCR(arcstat_l2_hdr_size, -space);
+ aggsum_add(&astat_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(aggsum_compare(&arc_meta_used, space) >= 0);
+ /*
+ * We use the upper bound here rather than the precise value
+ * because the arc_meta_max value doesn't need to be
+ * precise. It's only consumed by humans via arcstats.
+ */
+ if (arc_meta_max < aggsum_upper_bound(&arc_meta_used))
+ arc_meta_max = aggsum_upper_bound(&arc_meta_used);
+ aggsum_add(&arc_meta_used, -space);
}
- ASSERT(arc_size >= space);
- atomic_add_64(&arc_size, -space);
+ ASSERT(aggsum_compare(&arc_size, space) >= 0);
+ aggsum_add(&arc_size, -space);
}
/*
static boolean_t
arc_can_share(arc_buf_hdr_t *hdr, arc_buf_t *buf)
{
- boolean_t hdr_compressed, buf_compressed;
/*
* 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
+ * 1. the buffer is not encrypted
+ * 2. the hdr's compression matches the buf's compression
+ * 3. the hdr doesn't need to be byteswapped
+ * 4. the hdr isn't already being shared
+ * 5. the buf is either compressed or it is the last buf in the hdr list
*
- * Criterion #4 maintains the invariant that shared uncompressed
+ * Criterion #5 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
* sharing if the new buf isn't the first to be added.
*/
ASSERT3P(buf->b_hdr, ==, hdr);
- hdr_compressed = HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF;
- buf_compressed = ARC_BUF_COMPRESSED(buf) != 0;
- return (buf_compressed == hdr_compressed &&
+ boolean_t hdr_compressed =
+ arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF;
+ boolean_t buf_compressed = ARC_BUF_COMPRESSED(buf) != 0;
+ return (!ARC_BUF_ENCRYPTED(buf) &&
+ buf_compressed == hdr_compressed &&
hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS &&
!HDR_SHARED_DATA(hdr) &&
(ARC_BUF_LAST(buf) || ARC_BUF_COMPRESSED(buf)));
* 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,
+arc_buf_alloc_impl(arc_buf_hdr_t *hdr, spa_t *spa, const zbookmark_phys_t *zb,
+ void *tag, boolean_t encrypted, boolean_t compressed, boolean_t noauth,
boolean_t fill, arc_buf_t **ret)
{
arc_buf_t *buf;
- boolean_t can_share;
+ arc_fill_flags_t flags = ARC_FILL_LOCKED;
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT3U(HDR_GET_LSIZE(hdr), >, 0);
hdr->b_type == ARC_BUFC_METADATA);
ASSERT3P(ret, !=, NULL);
ASSERT3P(*ret, ==, NULL);
+ IMPLY(encrypted, compressed);
hdr->b_l1hdr.b_mru_hits = 0;
hdr->b_l1hdr.b_mru_ghost_hits = 0;
/*
* Only honor requests for compressed bufs if the hdr is actually
- * compressed.
+ * compressed. This must be overriden if the buffer is encrypted since
+ * encrypted buffers cannot be decompressed.
*/
- if (compressed && HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF)
+ if (encrypted) {
+ buf->b_flags |= ARC_BUF_FLAG_COMPRESSED;
+ buf->b_flags |= ARC_BUF_FLAG_ENCRYPTED;
+ flags |= ARC_FILL_COMPRESSED | ARC_FILL_ENCRYPTED;
+ } else if (compressed &&
+ arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF) {
buf->b_flags |= ARC_BUF_FLAG_COMPRESSED;
+ flags |= ARC_FILL_COMPRESSED;
+ }
- /*
- * Although the ARC should handle it correctly, levels above the ARC
- * should prevent us from having multiple compressed bufs off the same
- * hdr. To ensure we notice it if this behavior changes, we assert this
- * here the best we can.
- */
- IMPLY(ARC_BUF_COMPRESSED(buf), !HDR_SHARED_DATA(hdr));
+ if (noauth) {
+ ASSERT0(encrypted);
+ flags |= ARC_FILL_NOAUTH;
+ }
/*
* If the hdr's data can be shared then we share the data buffer and
* Second, the hdr's ABD must be linear so that the buf's user doesn't
* need to be ABD-aware.
*/
- can_share = arc_can_share(hdr, buf) && !HDR_L2_WRITING(hdr) &&
- abd_is_linear(hdr->b_l1hdr.b_pabd);
+ boolean_t can_share = arc_can_share(hdr, buf) && !HDR_L2_WRITING(hdr) &&
+ hdr->b_l1hdr.b_pabd != NULL && abd_is_linear(hdr->b_l1hdr.b_pabd);
/* Set up b_data and sharing */
if (can_share) {
hdr->b_l1hdr.b_buf = buf;
hdr->b_l1hdr.b_bufcnt += 1;
+ if (encrypted)
+ hdr->b_crypt_hdr.b_ebufcnt += 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));
+ ASSERT3P(zb, !=, NULL);
+ return (arc_buf_fill(buf, spa, zb, flags));
}
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
arc_buf_t *buf = arc_alloc_buf(spa, arc_onloan_tag,
is_metadata ? ARC_BUFC_METADATA : ARC_BUFC_DATA, size);
- atomic_add_64(&arc_loaned_bytes, size);
+ arc_loaned_bytes_update(arc_buf_size(buf));
+
return (buf);
}
arc_buf_t *buf = arc_alloc_compressed_buf(spa, arc_onloan_tag,
psize, lsize, compression_type);
+ arc_loaned_bytes_update(arc_buf_size(buf));
+
+ return (buf);
+}
+
+arc_buf_t *
+arc_loan_raw_buf(spa_t *spa, uint64_t dsobj, boolean_t byteorder,
+ const uint8_t *salt, const uint8_t *iv, const uint8_t *mac,
+ dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
+ enum zio_compress compression_type)
+{
+ arc_buf_t *buf = arc_alloc_raw_buf(spa, arc_onloan_tag, dsobj,
+ byteorder, salt, iv, mac, ot, psize, lsize, compression_type);
+
atomic_add_64(&arc_loaned_bytes, psize);
return (buf);
}
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);
+ (void) zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, tag);
+ (void) zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag);
- atomic_add_64(&arc_loaned_bytes, -arc_buf_size(buf));
+ arc_loaned_bytes_update(-arc_buf_size(buf));
}
/* Detach an arc_buf from a dbuf (tag) */
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);
+ (void) zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag);
+ (void) zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, tag);
- atomic_add_64(&arc_loaned_bytes, -arc_buf_size(buf));
+ arc_loaned_bytes_update(arc_buf_size(buf));
}
static void
}
static void
-arc_hdr_free_on_write(arc_buf_hdr_t *hdr)
+arc_hdr_free_on_write(arc_buf_hdr_t *hdr, boolean_t free_rdata)
{
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);
+ uint64_t size = (free_rdata) ? HDR_GET_PSIZE(hdr) : 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(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT(state != arc_anon && state != arc_l2c_only);
- (void) refcount_remove_many(&state->arcs_esize[type],
+ (void) zfs_refcount_remove_many(&state->arcs_esize[type],
size, hdr);
}
- (void) refcount_remove_many(&state->arcs_size, size, hdr);
+ (void) zfs_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);
+ if (free_rdata) {
+ l2arc_free_abd_on_write(hdr->b_crypt_hdr.b_rabd, size, type);
+ } else {
+ l2arc_free_abd_on_write(hdr->b_l1hdr.b_pabd, size, type);
+ }
}
/*
{
ASSERT(arc_can_share(hdr, buf));
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
+ ASSERT(!ARC_BUF_ENCRYPTED(buf));
ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr));
/*
* refcount ownership to the hdr since it always owns
* the refcount whenever an arc_buf_t is shared.
*/
- refcount_transfer_ownership(&hdr->b_l1hdr.b_state->arcs_size, buf, hdr);
+ zfs_refcount_transfer_ownership_many(&hdr->b_l1hdr.b_state->arcs_size,
+ arc_hdr_size(hdr), 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));
* We are no longer sharing this buffer so we need
* to transfer its ownership to the rightful owner.
*/
- refcount_transfer_ownership(&hdr->b_l1hdr.b_state->arcs_size, hdr, buf);
+ zfs_refcount_transfer_ownership_many(&hdr->b_l1hdr.b_state->arcs_size,
+ arc_hdr_size(hdr), 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);
static arc_buf_t *
arc_buf_remove(arc_buf_hdr_t *hdr, arc_buf_t *buf)
{
- arc_buf_t **bufp = &hdr->b_l1hdr.b_buf;
- arc_buf_t *lastbuf = NULL;
-
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.
static void
arc_buf_destroy_impl(arc_buf_t *buf)
{
- arc_buf_t *lastbuf;
arc_buf_hdr_t *hdr = buf->b_hdr;
/*
ASSERT(hdr->b_l1hdr.b_bufcnt > 0);
hdr->b_l1hdr.b_bufcnt -= 1;
+
+ if (ARC_BUF_ENCRYPTED(buf)) {
+ hdr->b_crypt_hdr.b_ebufcnt -= 1;
+
+ /*
+ * If we have no more encrypted buffers and we've
+ * already gotten a copy of the decrypted data we can
+ * free b_rabd to save some space.
+ */
+ if (hdr->b_crypt_hdr.b_ebufcnt == 0 &&
+ HDR_HAS_RABD(hdr) && hdr->b_l1hdr.b_pabd != NULL &&
+ !HDR_IO_IN_PROGRESS(hdr)) {
+ arc_hdr_free_abd(hdr, B_TRUE);
+ }
+ }
}
- lastbuf = arc_buf_remove(hdr, buf);
+ arc_buf_t *lastbuf = arc_buf_remove(hdr, buf);
if (ARC_BUF_SHARED(buf) && !ARC_BUF_COMPRESSED(buf)) {
/*
* 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.
+ * wasted temporarily. We must also be careful not to share
+ * encrypted buffers, since they cannot be shared.
*/
- if (lastbuf != NULL) {
+ if (lastbuf != NULL && !ARC_BUF_ENCRYPTED(lastbuf)) {
/* 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);
+ arc_hdr_free_abd(hdr, B_FALSE);
/*
* We must setup a new shared block between the
*/
ASSERT3P(lastbuf, !=, NULL);
ASSERT(arc_buf_is_shared(lastbuf) ||
- HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF);
+ arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF);
}
- if (hdr->b_l1hdr.b_bufcnt == 0)
+ /*
+ * 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;
}
static void
-arc_hdr_alloc_pabd(arc_buf_hdr_t *hdr)
+arc_hdr_alloc_abd(arc_buf_hdr_t *hdr, boolean_t alloc_rdata)
{
+ uint64_t size;
+
ASSERT3U(HDR_GET_LSIZE(hdr), >, 0);
ASSERT(HDR_HAS_L1HDR(hdr));
- ASSERT(!HDR_SHARED_DATA(hdr));
+ ASSERT(!HDR_SHARED_DATA(hdr) || alloc_rdata);
+ IMPLY(alloc_rdata, HDR_PROTECTED(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);
+ if (alloc_rdata) {
+ size = HDR_GET_PSIZE(hdr);
+ ASSERT3P(hdr->b_crypt_hdr.b_rabd, ==, NULL);
+ hdr->b_crypt_hdr.b_rabd = arc_get_data_abd(hdr, size, hdr);
+ ASSERT3P(hdr->b_crypt_hdr.b_rabd, !=, NULL);
+ ARCSTAT_INCR(arcstat_raw_size, size);
+ } else {
+ size = arc_hdr_size(hdr);
+ ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
+ hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, size, hdr);
+ ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+ }
- ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr));
+ ARCSTAT_INCR(arcstat_compressed_size, size);
ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr));
}
static void
-arc_hdr_free_pabd(arc_buf_hdr_t *hdr)
+arc_hdr_free_abd(arc_buf_hdr_t *hdr, boolean_t free_rdata)
{
+ uint64_t size = (free_rdata) ? HDR_GET_PSIZE(hdr) : arc_hdr_size(hdr);
+
ASSERT(HDR_HAS_L1HDR(hdr));
- ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+ ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr));
+ IMPLY(free_rdata, HDR_HAS_RABD(hdr));
/*
* If the hdr is currently being written to the l2arc then
* writing it to the l2arc device.
*/
if (HDR_L2_WRITING(hdr)) {
- arc_hdr_free_on_write(hdr);
+ arc_hdr_free_on_write(hdr, free_rdata);
ARCSTAT_BUMP(arcstat_l2_free_on_write);
+ } else if (free_rdata) {
+ arc_free_data_abd(hdr, hdr->b_crypt_hdr.b_rabd, size, hdr);
} else {
- arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd,
- arc_hdr_size(hdr), hdr);
+ arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, size, hdr);
}
- hdr->b_l1hdr.b_pabd = NULL;
- hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
- ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr));
+ if (free_rdata) {
+ hdr->b_crypt_hdr.b_rabd = NULL;
+ ARCSTAT_INCR(arcstat_raw_size, -size);
+ } else {
+ hdr->b_l1hdr.b_pabd = NULL;
+ }
+
+ if (hdr->b_l1hdr.b_pabd == NULL && !HDR_HAS_RABD(hdr))
+ hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
+
+ ARCSTAT_INCR(arcstat_compressed_size, -size);
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)
+ boolean_t protected, enum zio_compress compression_type,
+ arc_buf_contents_t type, boolean_t alloc_rdata)
{
arc_buf_hdr_t *hdr;
VERIFY(type == ARC_BUFC_DATA || type == ARC_BUFC_METADATA);
+ if (protected) {
+ hdr = kmem_cache_alloc(hdr_full_crypt_cache, KM_PUSHPAGE);
+ } else {
+ hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE);
+ }
- hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE);
ASSERT(HDR_EMPTY(hdr));
ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL);
HDR_SET_PSIZE(hdr, psize);
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);
+ if (protected)
+ arc_hdr_set_flags(hdr, ARC_FLAG_PROTECTED);
hdr->b_l1hdr.b_state = arc_anon;
hdr->b_l1hdr.b_arc_access = 0;
* 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));
+ arc_hdr_alloc_abd(hdr, alloc_rdata);
+ ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
return (hdr);
}
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(HDR_HAS_L2HDR(hdr));
ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) ||
(old == hdr_l2only_cache && new == hdr_full_cache));
+ /*
+ * if the caller wanted a new full header and the header is to be
+ * encrypted we will actually allocate the header from the full crypt
+ * cache instead. The same applies to freeing from the old cache.
+ */
+ if (HDR_PROTECTED(hdr) && new == hdr_full_cache)
+ new = hdr_full_crypt_cache;
+ if (HDR_PROTECTED(hdr) && old == hdr_full_cache)
+ old = hdr_full_crypt_cache;
+
nhdr = kmem_cache_alloc(new, KM_PUSHPAGE);
ASSERT(MUTEX_HELD(HDR_LOCK(hdr)));
bcopy(hdr, nhdr, HDR_L2ONLY_SIZE);
- if (new == hdr_full_cache) {
+ if (new == hdr_full_cache || new == hdr_full_crypt_cache) {
arc_hdr_set_flags(nhdr, ARC_FLAG_HAS_L1HDR);
/*
* arc_access and arc_change_state need to be aware that a
/* Verify previous threads set to NULL before freeing */
ASSERT3P(nhdr->b_l1hdr.b_pabd, ==, NULL);
+ ASSERT(!HDR_HAS_RABD(hdr));
} else {
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
ASSERT0(hdr->b_l1hdr.b_bufcnt);
*/
VERIFY(!HDR_L2_WRITING(hdr));
VERIFY3P(hdr->b_l1hdr.b_pabd, ==, NULL);
+ ASSERT(!HDR_HAS_RABD(hdr));
arc_hdr_clear_flags(nhdr, ARC_FLAG_HAS_L1HDR);
}
* 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);
+ (void) zfs_refcount_remove_many(&dev->l2ad_alloc,
+ arc_hdr_size(hdr), hdr);
+ (void) zfs_refcount_add_many(&dev->l2ad_alloc,
+ arc_hdr_size(nhdr), nhdr);
buf_discard_identity(hdr);
kmem_cache_free(old, hdr);
return (nhdr);
}
+/*
+ * This function allows an L1 header to be reallocated as a crypt
+ * header and vice versa. If we are going to a crypt header, the
+ * new fields will be zeroed out.
+ */
+static arc_buf_hdr_t *
+arc_hdr_realloc_crypt(arc_buf_hdr_t *hdr, boolean_t need_crypt)
+{
+ arc_buf_hdr_t *nhdr;
+ arc_buf_t *buf;
+ kmem_cache_t *ncache, *ocache;
+ unsigned nsize, osize;
+
+ /*
+ * This function requires that hdr is in the arc_anon state.
+ * Therefore it won't have any L2ARC data for us to worry
+ * about copying.
+ */
+ ASSERT(HDR_HAS_L1HDR(hdr));
+ ASSERT(!HDR_HAS_L2HDR(hdr));
+ ASSERT3U(!!HDR_PROTECTED(hdr), !=, need_crypt);
+ ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
+ ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node));
+ ASSERT(!list_link_active(&hdr->b_l2hdr.b_l2node));
+ ASSERT3P(hdr->b_hash_next, ==, NULL);
+
+ if (need_crypt) {
+ ncache = hdr_full_crypt_cache;
+ nsize = sizeof (hdr->b_crypt_hdr);
+ ocache = hdr_full_cache;
+ osize = HDR_FULL_SIZE;
+ } else {
+ ncache = hdr_full_cache;
+ nsize = HDR_FULL_SIZE;
+ ocache = hdr_full_crypt_cache;
+ osize = sizeof (hdr->b_crypt_hdr);
+ }
+
+ nhdr = kmem_cache_alloc(ncache, KM_PUSHPAGE);
+
+ /*
+ * Copy all members that aren't locks or condvars to the new header.
+ * No lists are pointing to us (as we asserted above), so we don't
+ * need to worry about the list nodes.
+ */
+ nhdr->b_dva = hdr->b_dva;
+ nhdr->b_birth = hdr->b_birth;
+ nhdr->b_type = hdr->b_type;
+ nhdr->b_flags = hdr->b_flags;
+ nhdr->b_psize = hdr->b_psize;
+ nhdr->b_lsize = hdr->b_lsize;
+ nhdr->b_spa = hdr->b_spa;
+ nhdr->b_l1hdr.b_freeze_cksum = hdr->b_l1hdr.b_freeze_cksum;
+ nhdr->b_l1hdr.b_bufcnt = hdr->b_l1hdr.b_bufcnt;
+ nhdr->b_l1hdr.b_byteswap = hdr->b_l1hdr.b_byteswap;
+ nhdr->b_l1hdr.b_state = hdr->b_l1hdr.b_state;
+ nhdr->b_l1hdr.b_arc_access = hdr->b_l1hdr.b_arc_access;
+ nhdr->b_l1hdr.b_mru_hits = hdr->b_l1hdr.b_mru_hits;
+ nhdr->b_l1hdr.b_mru_ghost_hits = hdr->b_l1hdr.b_mru_ghost_hits;
+ nhdr->b_l1hdr.b_mfu_hits = hdr->b_l1hdr.b_mfu_hits;
+ nhdr->b_l1hdr.b_mfu_ghost_hits = hdr->b_l1hdr.b_mfu_ghost_hits;
+ nhdr->b_l1hdr.b_l2_hits = hdr->b_l1hdr.b_l2_hits;
+ nhdr->b_l1hdr.b_acb = hdr->b_l1hdr.b_acb;
+ nhdr->b_l1hdr.b_pabd = hdr->b_l1hdr.b_pabd;
+
+ /*
+ * This zfs_refcount_add() exists only to ensure that the individual
+ * arc buffers always point to a header that is referenced, avoiding
+ * a small race condition that could trigger ASSERTs.
+ */
+ (void) zfs_refcount_add(&nhdr->b_l1hdr.b_refcnt, FTAG);
+ nhdr->b_l1hdr.b_buf = hdr->b_l1hdr.b_buf;
+ for (buf = nhdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) {
+ mutex_enter(&buf->b_evict_lock);
+ buf->b_hdr = nhdr;
+ mutex_exit(&buf->b_evict_lock);
+ }
+
+ zfs_refcount_transfer(&nhdr->b_l1hdr.b_refcnt, &hdr->b_l1hdr.b_refcnt);
+ (void) zfs_refcount_remove(&nhdr->b_l1hdr.b_refcnt, FTAG);
+ ASSERT0(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt));
+
+ if (need_crypt) {
+ arc_hdr_set_flags(nhdr, ARC_FLAG_PROTECTED);
+ } else {
+ arc_hdr_clear_flags(nhdr, ARC_FLAG_PROTECTED);
+ }
+
+ /* unset all members of the original hdr */
+ bzero(&hdr->b_dva, sizeof (dva_t));
+ hdr->b_birth = 0;
+ hdr->b_type = ARC_BUFC_INVALID;
+ hdr->b_flags = 0;
+ hdr->b_psize = 0;
+ hdr->b_lsize = 0;
+ hdr->b_spa = 0;
+ hdr->b_l1hdr.b_freeze_cksum = NULL;
+ hdr->b_l1hdr.b_buf = NULL;
+ hdr->b_l1hdr.b_bufcnt = 0;
+ hdr->b_l1hdr.b_byteswap = 0;
+ hdr->b_l1hdr.b_state = NULL;
+ hdr->b_l1hdr.b_arc_access = 0;
+ hdr->b_l1hdr.b_mru_hits = 0;
+ hdr->b_l1hdr.b_mru_ghost_hits = 0;
+ hdr->b_l1hdr.b_mfu_hits = 0;
+ hdr->b_l1hdr.b_mfu_ghost_hits = 0;
+ hdr->b_l1hdr.b_l2_hits = 0;
+ hdr->b_l1hdr.b_acb = NULL;
+ hdr->b_l1hdr.b_pabd = NULL;
+
+ if (ocache == hdr_full_crypt_cache) {
+ ASSERT(!HDR_HAS_RABD(hdr));
+ hdr->b_crypt_hdr.b_ot = DMU_OT_NONE;
+ hdr->b_crypt_hdr.b_ebufcnt = 0;
+ hdr->b_crypt_hdr.b_dsobj = 0;
+ bzero(hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN);
+ bzero(hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN);
+ bzero(hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN);
+ }
+
+ buf_discard_identity(hdr);
+ kmem_cache_free(ocache, hdr);
+
+ return (nhdr);
+}
+
+/*
+ * This function is used by the send / receive code to convert a newly
+ * allocated arc_buf_t to one that is suitable for a raw encrypted write. It
+ * is also used to allow the root objset block to be uupdated without altering
+ * its embedded MACs. Both block types will always be uncompressed so we do not
+ * have to worry about compression type or psize.
+ */
+void
+arc_convert_to_raw(arc_buf_t *buf, uint64_t dsobj, boolean_t byteorder,
+ dmu_object_type_t ot, const uint8_t *salt, const uint8_t *iv,
+ const uint8_t *mac)
+{
+ arc_buf_hdr_t *hdr = buf->b_hdr;
+
+ ASSERT(ot == DMU_OT_DNODE || ot == DMU_OT_OBJSET);
+ ASSERT(HDR_HAS_L1HDR(hdr));
+ ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
+
+ buf->b_flags |= (ARC_BUF_FLAG_COMPRESSED | ARC_BUF_FLAG_ENCRYPTED);
+ if (!HDR_PROTECTED(hdr))
+ hdr = arc_hdr_realloc_crypt(hdr, B_TRUE);
+ hdr->b_crypt_hdr.b_dsobj = dsobj;
+ hdr->b_crypt_hdr.b_ot = ot;
+ hdr->b_l1hdr.b_byteswap = (byteorder == ZFS_HOST_BYTEORDER) ?
+ DMU_BSWAP_NUMFUNCS : DMU_OT_BYTESWAP(ot);
+ if (!arc_hdr_has_uncompressed_buf(hdr))
+ arc_cksum_free(hdr);
+
+ if (salt != NULL)
+ bcopy(salt, hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN);
+ if (iv != NULL)
+ bcopy(iv, hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN);
+ if (mac != NULL)
+ bcopy(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN);
+}
+
/*
* 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_t *buf;
arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), size, size,
- ZIO_COMPRESS_OFF, type);
+ B_FALSE, ZIO_COMPRESS_OFF, type, B_FALSE);
ASSERT(!MUTEX_HELD(HDR_LOCK(hdr)));
- buf = NULL;
- VERIFY0(arc_buf_alloc_impl(hdr, tag, B_FALSE, B_FALSE, &buf));
+ arc_buf_t *buf = NULL;
+ VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_FALSE, B_FALSE,
+ 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);
+ ASSERT3U(compression_type, >, ZIO_COMPRESS_OFF);
+ ASSERT3U(compression_type, <, ZIO_COMPRESS_FUNCTIONS);
+
+ arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize,
+ B_FALSE, compression_type, ARC_BUFC_DATA, B_FALSE);
+ ASSERT(!MUTEX_HELD(HDR_LOCK(hdr)));
+
+ arc_buf_t *buf = NULL;
+ VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_FALSE,
+ B_TRUE, B_FALSE, 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_untransform() 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_abd(hdr, B_FALSE);
+ arc_share_buf(hdr, 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,
+arc_alloc_raw_buf(spa_t *spa, void *tag, uint64_t dsobj, boolean_t byteorder,
+ const uint8_t *salt, const uint8_t *iv, const uint8_t *mac,
+ dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
enum zio_compress compression_type)
{
arc_buf_hdr_t *hdr;
arc_buf_t *buf;
+ arc_buf_contents_t type = DMU_OT_IS_METADATA(ot) ?
+ ARC_BUFC_METADATA : ARC_BUFC_DATA;
+
ASSERT3U(lsize, >, 0);
ASSERT3U(lsize, >=, psize);
- ASSERT(compression_type > ZIO_COMPRESS_OFF);
- ASSERT(compression_type < ZIO_COMPRESS_FUNCTIONS);
+ ASSERT3U(compression_type, >=, ZIO_COMPRESS_OFF);
+ ASSERT3U(compression_type, <, ZIO_COMPRESS_FUNCTIONS);
- hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize,
- compression_type, ARC_BUFC_DATA);
+ hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, B_TRUE,
+ compression_type, type, B_TRUE);
ASSERT(!MUTEX_HELD(HDR_LOCK(hdr)));
+ hdr->b_crypt_hdr.b_dsobj = dsobj;
+ hdr->b_crypt_hdr.b_ot = ot;
+ hdr->b_l1hdr.b_byteswap = (byteorder == ZFS_HOST_BYTEORDER) ?
+ DMU_BSWAP_NUMFUNCS : DMU_OT_BYTESWAP(ot);
+ bcopy(salt, hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN);
+ bcopy(iv, hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN);
+ bcopy(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN);
+
+ /*
+ * This buffer will be considered encrypted even if the ot is not an
+ * encrypted type. It will become authenticated instead in
+ * arc_write_ready().
+ */
buf = NULL;
- VERIFY0(arc_buf_alloc_impl(hdr, tag, B_TRUE, B_FALSE, &buf));
+ VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_TRUE, B_TRUE,
+ B_FALSE, 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);
}
{
l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr;
l2arc_dev_t *dev = l2hdr->b_dev;
- uint64_t asize = arc_hdr_size(hdr);
+ uint64_t psize = HDR_GET_PSIZE(hdr);
+ uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, psize);
ASSERT(MUTEX_HELD(&dev->l2ad_mtx));
ASSERT(HDR_HAS_L2HDR(hdr));
list_remove(&dev->l2ad_buflist, hdr);
- ARCSTAT_INCR(arcstat_l2_asize, -asize);
- ARCSTAT_INCR(arcstat_l2_size, -HDR_GET_LSIZE(hdr));
+ ARCSTAT_INCR(arcstat_l2_psize, -psize);
+ ARCSTAT_INCR(arcstat_l2_lsize, -HDR_GET_LSIZE(hdr));
vdev_space_update(dev->l2ad_vdev, -asize, 0, 0);
- (void) refcount_remove_many(&dev->l2ad_alloc, asize, hdr);
+ (void) zfs_refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr),
+ hdr);
arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR);
}
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));
+ ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
}
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
while (hdr->b_l1hdr.b_buf != NULL)
arc_buf_destroy_impl(hdr->b_l1hdr.b_buf);
- if (hdr->b_l1hdr.b_pabd != NULL)
- arc_hdr_free_pabd(hdr);
+ if (hdr->b_l1hdr.b_pabd != NULL) {
+ arc_hdr_free_abd(hdr, B_FALSE);
+ }
+
+ if (HDR_HAS_RABD(hdr))
+ arc_hdr_free_abd(hdr, B_TRUE);
}
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);
+
+ if (!HDR_PROTECTED(hdr)) {
+ kmem_cache_free(hdr_full_cache, hdr);
+ } else {
+ kmem_cache_free(hdr_full_crypt_cache, hdr);
+ }
} else {
kmem_cache_free(hdr_l2only_cache, hdr);
}
{
arc_state_t *evicted_state, *state;
int64_t bytes_evicted = 0;
+ int min_lifetime = HDR_PRESCIENT_PREFETCH(hdr) ?
+ arc_min_prescient_prefetch_ms : arc_min_prefetch_ms;
ASSERT(MUTEX_HELD(hash_lock));
ASSERT(HDR_HAS_L1HDR(hdr));
if (HDR_HAS_L2HDR(hdr)) {
ASSERT(hdr->b_l1hdr.b_pabd == NULL);
+ ASSERT(!HDR_HAS_RABD(hdr));
/*
* This buffer is cached on the 2nd Level ARC;
* don't destroy the header.
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)) {
+ MSEC_TO_TICK(min_lifetime))) {
ARCSTAT_BUMP(arcstat_evict_skip);
return (bytes_evicted);
}
- ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt));
+ ASSERT0(zfs_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)) {
* This ensures that the accounting is updated correctly
* in arc_free_data_impl().
*/
- arc_hdr_free_pabd(hdr);
+ if (hdr->b_l1hdr.b_pabd != NULL)
+ arc_hdr_free_abd(hdr, B_FALSE);
+
+ if (HDR_HAS_RABD(hdr))
+ arc_hdr_free_abd(hdr, B_TRUE);
arc_change_state(evicted_state, hdr, hash_lock);
ASSERT(HDR_IN_HASH_TABLE(hdr));
* 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.
+ * using cv_broadcast here, they will be woken
+ * up via cv_broadcast in arc_adjust_cb() just
+ * before arc_adjust_zthr sleeps.
*/
- mutex_enter(&arc_reclaim_lock);
+ mutex_enter(&arc_adjust_lock);
if (!arc_is_overflowing())
- cv_signal(&arc_reclaim_waiters_cv);
- mutex_exit(&arc_reclaim_lock);
+ cv_signal(&arc_adjust_waiters_cv);
+ mutex_exit(&arc_adjust_lock);
} else {
ARCSTAT_BUMP(arcstat_mutex_miss);
}
arc_buf_contents_t type)
{
uint64_t total_evicted = 0;
- multilist_t *ml = &state->arcs_list[type];
+ multilist_t *ml = state->arcs_list[type];
int num_sublists;
arc_buf_hdr_t **markers;
- int i;
IMPLY(bytes < 0, bytes == ARC_EVICT_ALL);
* than starting from the tail each time.
*/
markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP);
- for (i = 0; i < num_sublists; i++) {
+ for (int i = 0; i < num_sublists; i++) {
multilist_sublist_t *mls;
markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP);
* Request that 10% of the LRUs be scanned by the superblock
* shrinker.
*/
- if (type == ARC_BUFC_DATA && arc_dnode_size > arc_dnode_limit)
- arc_prune_async((arc_dnode_size - arc_dnode_limit) /
- sizeof (dnode_t) / zfs_arc_dnode_reduce_percent);
+ if (type == ARC_BUFC_DATA && aggsum_compare(&astat_dnode_size,
+ arc_dnode_limit) > 0) {
+ arc_prune_async((aggsum_upper_bound(&astat_dnode_size) -
+ arc_dnode_limit) / sizeof (dnode_t) /
+ zfs_arc_dnode_reduce_percent);
+ }
/*
* Start eviction using a randomly selected sublist,
* (e.g. index 0) would cause evictions to favor certain
* sublists over others.
*/
- for (i = 0; i < num_sublists; i++) {
+ for (int i = 0; i < num_sublists; i++) {
uint64_t bytes_remaining;
uint64_t bytes_evicted;
}
}
- for (i = 0; i < num_sublists; i++) {
+ 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);
{
uint64_t evicted = 0;
- while (refcount_count(&state->arcs_esize[type]) != 0) {
+ while (zfs_refcount_count(&state->arcs_esize[type]) != 0) {
evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type);
if (!retry)
if (func != NULL)
func(ap->p_adjust, ap->p_private);
- refcount_remove(&ap->p_refcnt, func);
+ zfs_refcount_remove(&ap->p_refcnt, func);
}
/*
for (ap = list_head(&arc_prune_list); ap != NULL;
ap = list_next(&arc_prune_list, ap)) {
- if (refcount_count(&ap->p_refcnt) >= 2)
+ if (zfs_refcount_count(&ap->p_refcnt) >= 2)
continue;
- refcount_add(&ap->p_refcnt, ap->p_pfunc);
+ zfs_refcount_add(&ap->p_refcnt, ap->p_pfunc);
ap->p_adjust = adjust;
if (taskq_dispatch(arc_prune_taskq, arc_prune_task,
ap, TQ_SLEEP) == TASKQID_INVALID) {
- refcount_remove(&ap->p_refcnt, ap->p_pfunc);
+ zfs_refcount_remove(&ap->p_refcnt, ap->p_pfunc);
continue;
}
ARCSTAT_BUMP(arcstat_prune);
{
int64_t delta;
- if (bytes > 0 && refcount_count(&state->arcs_esize[type]) > 0) {
- delta = MIN(refcount_count(&state->arcs_esize[type]), bytes);
+ if (bytes > 0 && zfs_refcount_count(&state->arcs_esize[type]) > 0) {
+ delta = MIN(zfs_refcount_count(&state->arcs_esize[type]),
+ bytes);
return (arc_evict_state(state, spa, delta, type));
}
*
* Therefore, this function has been updated to make alternating passes
* over the ARC releasing data buffers and then newly unheld meta data
- * buffers. This ensures forward progress is maintained and arc_meta_used
+ * buffers. This ensures forward progress is maintained and meta_used
* will decrease. Normally this is sufficient, but if required the ARC
* will call the registered prune callbacks causing dentry and inodes to
* be dropped from the VFS cache. This will make dnode meta data buffers
* available for reclaim.
*/
static uint64_t
-arc_adjust_meta_balanced(void)
+arc_adjust_meta_balanced(uint64_t meta_used)
{
int64_t delta, prune = 0, adjustmnt;
uint64_t total_evicted = 0;
* metadata from the MFU. I think we probably need to implement a
* "metadata arc_p" value to do this properly.
*/
- adjustmnt = arc_meta_used - arc_meta_limit;
+ adjustmnt = meta_used - arc_meta_limit;
- if (adjustmnt > 0 && refcount_count(&arc_mru->arcs_esize[type]) > 0) {
- delta = MIN(refcount_count(&arc_mru->arcs_esize[type]),
+ if (adjustmnt > 0 &&
+ zfs_refcount_count(&arc_mru->arcs_esize[type]) > 0) {
+ delta = MIN(zfs_refcount_count(&arc_mru->arcs_esize[type]),
adjustmnt);
total_evicted += arc_adjust_impl(arc_mru, 0, delta, type);
adjustmnt -= delta;
* simply decrement the amount of data evicted from the MRU.
*/
- if (adjustmnt > 0 && refcount_count(&arc_mfu->arcs_esize[type]) > 0) {
- delta = MIN(refcount_count(&arc_mfu->arcs_esize[type]),
+ if (adjustmnt > 0 &&
+ zfs_refcount_count(&arc_mfu->arcs_esize[type]) > 0) {
+ delta = MIN(zfs_refcount_count(&arc_mfu->arcs_esize[type]),
adjustmnt);
total_evicted += arc_adjust_impl(arc_mfu, 0, delta, type);
}
- adjustmnt = arc_meta_used - arc_meta_limit;
+ adjustmnt = meta_used - arc_meta_limit;
if (adjustmnt > 0 &&
- refcount_count(&arc_mru_ghost->arcs_esize[type]) > 0) {
+ zfs_refcount_count(&arc_mru_ghost->arcs_esize[type]) > 0) {
delta = MIN(adjustmnt,
- refcount_count(&arc_mru_ghost->arcs_esize[type]));
+ zfs_refcount_count(&arc_mru_ghost->arcs_esize[type]));
total_evicted += arc_adjust_impl(arc_mru_ghost, 0, delta, type);
adjustmnt -= delta;
}
if (adjustmnt > 0 &&
- refcount_count(&arc_mfu_ghost->arcs_esize[type]) > 0) {
+ zfs_refcount_count(&arc_mfu_ghost->arcs_esize[type]) > 0) {
delta = MIN(adjustmnt,
- refcount_count(&arc_mfu_ghost->arcs_esize[type]));
+ zfs_refcount_count(&arc_mfu_ghost->arcs_esize[type]));
total_evicted += arc_adjust_impl(arc_mfu_ghost, 0, delta, type);
}
* meta buffers. Requests to the upper layers will be made with
* increasingly large scan sizes until the ARC is below the limit.
*/
- if (arc_meta_used > arc_meta_limit) {
+ if (meta_used > arc_meta_limit) {
if (type == ARC_BUFC_DATA) {
type = ARC_BUFC_METADATA;
} else {
* capped by the arc_meta_limit tunable.
*/
static uint64_t
-arc_adjust_meta_only(void)
+arc_adjust_meta_only(uint64_t meta_used)
{
uint64_t total_evicted = 0;
int64_t target;
* 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));
+ target = MIN((int64_t)(meta_used - arc_meta_limit),
+ (int64_t)(zfs_refcount_count(&arc_anon->arcs_size) +
+ zfs_refcount_count(&arc_mru->arcs_size) - arc_p));
total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA);
* 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)));
+ target = MIN((int64_t)(meta_used - arc_meta_limit),
+ (int64_t)(zfs_refcount_count(&arc_mfu->arcs_size) -
+ (arc_c - arc_p)));
total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA);
}
static uint64_t
-arc_adjust_meta(void)
+arc_adjust_meta(uint64_t meta_used)
{
if (zfs_arc_meta_strategy == ARC_STRATEGY_META_ONLY)
- return (arc_adjust_meta_only());
+ return (arc_adjust_meta_only(meta_used));
else
- return (arc_adjust_meta_balanced());
+ return (arc_adjust_meta_balanced(meta_used));
}
/*
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];
+ 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;
uint64_t total_evicted = 0;
uint64_t bytes;
int64_t target;
+ uint64_t asize = aggsum_value(&arc_size);
+ uint64_t ameta = aggsum_value(&arc_meta_used);
/*
* If we're over arc_meta_limit, we want to correct that before
* potentially evicting data buffers below.
*/
- total_evicted += arc_adjust_meta();
+ total_evicted += arc_adjust_meta(ameta);
/*
* Adjust MRU size
* 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));
+ target = MIN((int64_t)(asize - arc_c),
+ (int64_t)(zfs_refcount_count(&arc_anon->arcs_size) +
+ zfs_refcount_count(&arc_mru->arcs_size) + ameta - arc_p));
/*
* If we're below arc_meta_min, always prefer to evict data.
* type, spill over into the next type.
*/
if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA &&
- arc_meta_used > arc_meta_min) {
+ ameta > arc_meta_min) {
bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA);
total_evicted += bytes;
arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA);
}
+ /*
+ * Re-sum ARC stats after the first round of evictions.
+ */
+ asize = aggsum_value(&arc_size);
+ ameta = aggsum_value(&arc_meta_used);
+
+
/*
* Adjust MFU size
*
* 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;
+ target = asize - arc_c;
if (arc_adjust_type(arc_mfu) == ARC_BUFC_METADATA &&
- arc_meta_used > arc_meta_min) {
+ ameta > arc_meta_min) {
bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA);
total_evicted += bytes;
* 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;
+ target = zfs_refcount_count(&arc_mru->arcs_size) +
+ zfs_refcount_count(&arc_mru_ghost->arcs_size) - arc_c;
bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA);
total_evicted += bytes;
* 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;
+ target = zfs_refcount_count(&arc_mru_ghost->arcs_size) +
+ zfs_refcount_count(&arc_mfu_ghost->arcs_size) - arc_c;
bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA);
total_evicted += bytes;
(void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry);
}
-void
-arc_shrink(int64_t to_free)
+static void
+arc_reduce_target_size(int64_t to_free)
{
+ uint64_t asize = aggsum_value(&arc_size);
uint64_t c = arc_c;
if (c > to_free && c - to_free > arc_c_min) {
arc_c = c - to_free;
atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift));
- if (arc_c > arc_size)
- arc_c = MAX(arc_size, arc_c_min);
+ if (asize < arc_c)
+ arc_c = MAX(asize, arc_c_min);
if (arc_p > arc_c)
arc_p = (arc_c >> 1);
ASSERT(arc_c >= arc_c_min);
arc_c = arc_c_min;
}
- if (arc_size > arc_c)
- (void) arc_adjust();
+ if (asize > arc_c) {
+ /* See comment in arc_adjust_cb_check() on why lock+flag */
+ mutex_enter(&arc_adjust_lock);
+ arc_adjust_needed = B_TRUE;
+ mutex_exit(&arc_adjust_lock);
+ zthr_wakeup(arc_adjust_zthr);
+ }
}
-
/*
* Return maximum amount of memory that we could possibly use. Reduced
* to half of all memory in user space which is primarily used for testing.
arc_all_memory(void)
{
#ifdef _KERNEL
- return (MIN(ptob(physmem),
- vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC)));
+#ifdef CONFIG_HIGHMEM
+ return (ptob(zfs_totalram_pages - totalhigh_pages));
+#else
+ return (ptob(zfs_totalram_pages));
+#endif /* CONFIG_HIGHMEM */
#else
return (ptob(physmem) / 2);
-#endif
+#endif /* _KERNEL */
+}
+
+/*
+ * Return the amount of memory that is considered free. In user space
+ * which is primarily used for testing we pretend that free memory ranges
+ * from 0-20% of all memory.
+ */
+static uint64_t
+arc_free_memory(void)
+{
+#ifdef _KERNEL
+#ifdef CONFIG_HIGHMEM
+ struct sysinfo si;
+ si_meminfo(&si);
+ return (ptob(si.freeram - si.freehigh));
+#else
+ return (ptob(nr_free_pages() +
+ nr_inactive_file_pages() +
+ nr_inactive_anon_pages() +
+ nr_slab_reclaimable_pages()));
+
+#endif /* CONFIG_HIGHMEM */
+#else
+ return (spa_get_random(arc_all_memory() * 20 / 100));
+#endif /* _KERNEL */
}
typedef enum free_memory_reason_t {
int64_t lowest = INT64_MAX;
free_memory_reason_t r = FMR_UNKNOWN;
#ifdef _KERNEL
- uint64_t available_memory = ptob(freemem);
int64_t n;
#ifdef __linux__
+#ifdef freemem
+#undef freemem
+#endif
pgcnt_t needfree = btop(arc_need_free);
pgcnt_t lotsfree = btop(arc_sys_free);
pgcnt_t desfree = 0;
-#endif
-
-#if defined(__i386)
- available_memory =
- MIN(available_memory, vmem_size(heap_arena, VMEM_FREE));
+ pgcnt_t freemem = btop(arc_free_memory());
#endif
if (needfree > 0) {
* 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 * (btop(available_memory) - lotsfree - needfree - desfree);
+ n = PAGESIZE * (freemem - lotsfree - needfree - desfree);
if (n < lowest) {
lowest = n;
r = FMR_LOTSFREE;
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
}
#endif
-#if defined(__i386)
+#if defined(_ILP32)
/*
- * If we're on an i386 platform, it's possible that we'll exhaust the
+ * If we're on a 32-bit 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
}
static void
-arc_kmem_reap_now(void)
+arc_kmem_reap_soon(void)
{
size_t i;
kmem_cache_t *prev_cache = NULL;
extern kmem_cache_t *zio_data_buf_cache[];
extern kmem_cache_t *range_seg_cache;
- if ((arc_meta_used >= arc_meta_limit) && zfs_arc_meta_prune) {
+#ifdef _KERNEL
+ if ((aggsum_compare(&arc_meta_used, arc_meta_limit) >= 0) &&
+ zfs_arc_meta_prune) {
/*
* We are exceeding our meta-data cache limit.
* Prune some entries to release holds on meta-data.
*/
arc_prune_async(zfs_arc_meta_prune);
}
+#if defined(_ILP32)
+ /*
+ * Reclaim unused memory from all kmem caches.
+ */
+ kmem_reap();
+#endif
+#endif
for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) {
-#ifdef _ILP32
+#if defined(_ILP32)
/* reach upper limit of cache size on 32-bit */
if (zio_buf_cache[i] == NULL)
break;
}
}
+/* ARGSUSED */
+static boolean_t
+arc_adjust_cb_check(void *arg, zthr_t *zthr)
+{
+ /*
+ * This is necessary in order to keep the kstat information
+ * up to date for tools that display kstat data such as the
+ * mdb ::arc dcmd and the Linux crash utility. These tools
+ * typically do not call kstat's update function, but simply
+ * dump out stats from the most recent update. Without
+ * this call, these commands 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(the arc_adjust_zthr has a maximum sleep
+ * time of 1 second); 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);
+
+ /*
+ * We have to rely on arc_get_data_impl() to tell us when to adjust,
+ * rather than checking if we are overflowing here, so that we are
+ * sure to not leave arc_get_data_impl() waiting on
+ * arc_adjust_waiters_cv. If we have become "not overflowing" since
+ * arc_get_data_impl() checked, we need to wake it up. We could
+ * broadcast the CV here, but arc_get_data_impl() may have not yet
+ * gone to sleep. We would need to use a mutex to ensure that this
+ * function doesn't broadcast until arc_get_data_impl() has gone to
+ * sleep (e.g. the arc_adjust_lock). However, the lock ordering of
+ * such a lock would necessarily be incorrect with respect to the
+ * zthr_lock, which is held before this function is called, and is
+ * held by arc_get_data_impl() when it calls zthr_wakeup().
+ */
+ return (arc_adjust_needed);
+}
+
/*
- * 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().
+ * Keep arc_size under arc_c by running arc_adjust which evicts data
+ * from the ARC.
*/
+/* ARGSUSED */
static void
-arc_reclaim_thread(void)
+arc_adjust_cb(void *arg, zthr_t *zthr)
{
- fstrans_cookie_t cookie = spl_fstrans_mark();
- 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) {
- int64_t to_free;
- uint64_t evicted = 0;
+ uint64_t evicted = 0;
+ fstrans_cookie_t cookie = spl_fstrans_mark();
- arc_tuning_update();
+ /* Evict from cache */
+ evicted = arc_adjust();
+ /*
+ * 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. Additionally, zthr_iscancelled() is
+ * checked here so that if the arc is shutting down, the
+ * broadcast will wake any remaining arc adjust waiters.
+ */
+ mutex_enter(&arc_adjust_lock);
+ arc_adjust_needed = !zthr_iscancelled(arc_adjust_zthr) &&
+ evicted > 0 && aggsum_compare(&arc_size, arc_c) > 0;
+ if (!arc_adjust_needed) {
/*
- * 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.
+ * We're either no longer overflowing, or we
+ * can't evict anything more, so we should wake
+ * arc_get_data_impl() sooner.
*/
-#ifndef __linux__
- if (arc_ksp != NULL)
- arc_ksp->ks_update(arc_ksp, KSTAT_READ);
-#endif
- mutex_exit(&arc_reclaim_lock);
+ cv_broadcast(&arc_adjust_waiters_cv);
+ arc_need_free = 0;
+ }
+ mutex_exit(&arc_adjust_lock);
+ spl_fstrans_unmark(cookie);
+}
+
+/* ARGSUSED */
+static boolean_t
+arc_reap_cb_check(void *arg, zthr_t *zthr)
+{
+ int64_t free_memory = arc_available_memory();
+
+ /*
+ * If a kmem reap is already active, don't schedule more. We must
+ * check for this because kmem_cache_reap_soon() won't actually
+ * block on the cache being reaped (this is to prevent callers from
+ * becoming implicitly blocked by a system-wide kmem reap -- which,
+ * on a system with many, many full magazines, can take minutes).
+ */
+ if (!kmem_cache_reap_active() && free_memory < 0) {
+ arc_no_grow = B_TRUE;
+ arc_warm = B_TRUE;
/*
- * We call arc_adjust() before (possibly) calling
- * arc_kmem_reap_now(), so that we can wake up
- * arc_get_data_buf() sooner.
+ * Wait at least zfs_grow_retry (default 5) seconds
+ * before considering growing.
*/
- evicted = arc_adjust();
+ arc_growtime = gethrtime() + SEC2NSEC(arc_grow_retry);
+ return (B_TRUE);
+ } else if (free_memory < arc_c >> arc_no_grow_shift) {
+ arc_no_grow = B_TRUE;
+ } else if (gethrtime() >= arc_growtime) {
+ arc_no_grow = B_FALSE;
+ }
- int64_t free_memory = arc_available_memory();
- if (free_memory < 0) {
+ return (B_FALSE);
+}
- arc_no_grow = B_TRUE;
- arc_warm = B_TRUE;
+/*
+ * Keep enough free memory in the system by reaping the ARC's kmem
+ * caches. To cause more slabs to be reapable, we may reduce the
+ * target size of the cache (arc_c), causing the arc_adjust_cb()
+ * to free more buffers.
+ */
+/* ARGSUSED */
+static void
+arc_reap_cb(void *arg, zthr_t *zthr)
+{
+ int64_t free_memory;
+ fstrans_cookie_t cookie = spl_fstrans_mark();
- /*
- * Wait at least zfs_grow_retry (default 5) seconds
- * before considering growing.
- */
- growtime = gethrtime() + SEC2NSEC(arc_grow_retry);
+ /*
+ * Kick off asynchronous kmem_reap()'s of all our caches.
+ */
+ arc_kmem_reap_soon();
- arc_kmem_reap_now();
+ /*
+ * Wait at least arc_kmem_cache_reap_retry_ms between
+ * arc_kmem_reap_soon() calls. Without this check it is possible to
+ * end up in a situation where we spend lots of time reaping
+ * caches, while we're near arc_c_min. Waiting here also gives the
+ * subsequent free memory check a chance of finding that the
+ * asynchronous reap has already freed enough memory, and we don't
+ * need to call arc_reduce_target_size().
+ */
+ delay((hz * arc_kmem_cache_reap_retry_ms + 999) / 1000);
- /*
- * If we are still low on memory, shrink the ARC
- * so that we have arc_shrink_min free space.
- */
- free_memory = arc_available_memory();
+ /*
+ * Reduce the target size as needed to maintain the amount of free
+ * memory in the system at a fraction of the arc_size (1/128th by
+ * default). If oversubscribed (free_memory < 0) then reduce the
+ * target arc_size by the deficit amount plus the fractional
+ * amount. If free memory is positive but less then the fractional
+ * amount, reduce by what is needed to hit the fractional amount.
+ */
+ free_memory = arc_available_memory();
- to_free = (arc_c >> arc_shrink_shift) - free_memory;
- if (to_free > 0) {
+ int64_t to_free =
+ (arc_c >> arc_shrink_shift) - free_memory;
+ if (to_free > 0) {
#ifdef _KERNEL
- to_free = MAX(to_free, arc_need_free);
+ to_free = MAX(to_free, arc_need_free);
#endif
- 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 and clear
- * arc_need_free since nothing more can be done.
- */
- cv_broadcast(&arc_reclaim_waiters_cv);
- arc_need_free = 0;
-
- /*
- * 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_sig_hires(&arc_reclaim_thread_cv,
- &arc_reclaim_lock, SEC2NSEC(1), MSEC2NSEC(1), 0);
- CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock);
- }
+ arc_reduce_target_size(to_free);
}
-
- arc_reclaim_thread_exit = B_FALSE;
- cv_broadcast(&arc_reclaim_thread_cv);
- CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */
spl_fstrans_unmark(cookie);
- thread_exit();
}
#ifdef _KERNEL
static uint64_t
arc_evictable_memory(void)
{
+ int64_t asize = aggsum_value(&arc_size);
uint64_t arc_clean =
- refcount_count(&arc_mru->arcs_esize[ARC_BUFC_DATA]) +
- refcount_count(&arc_mru->arcs_esize[ARC_BUFC_METADATA]) +
- refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_DATA]) +
- refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]);
- uint64_t ghost_clean =
- refcount_count(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]) +
- refcount_count(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]) +
- refcount_count(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]) +
- refcount_count(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]);
- uint64_t arc_dirty = MAX((int64_t)arc_size - (int64_t)arc_clean, 0);
+ zfs_refcount_count(&arc_mru->arcs_esize[ARC_BUFC_DATA]) +
+ zfs_refcount_count(&arc_mru->arcs_esize[ARC_BUFC_METADATA]) +
+ zfs_refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_DATA]) +
+ zfs_refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]);
+ uint64_t arc_dirty = MAX((int64_t)asize - (int64_t)arc_clean, 0);
+
+ /*
+ * Scale reported evictable memory in proportion to page cache, cap
+ * at specified min/max.
+ */
+ uint64_t min = (ptob(nr_file_pages()) / 100) * zfs_arc_pc_percent;
+ min = MAX(arc_c_min, MIN(arc_c_max, min));
- if (arc_dirty >= arc_c_min)
- return (ghost_clean + arc_clean);
+ if (arc_dirty >= min)
+ return (arc_clean);
- return (ghost_clean + MAX((int64_t)arc_size - (int64_t)arc_c_min, 0));
+ return (MAX((int64_t)asize - (int64_t)min, 0));
}
/*
return (SHRINK_STOP);
/* Reclaim in progress */
- if (mutex_tryenter(&arc_reclaim_lock) == 0)
- return (SHRINK_STOP);
+ if (mutex_tryenter(&arc_adjust_lock) == 0) {
+ ARCSTAT_INCR(arcstat_need_free, ptob(sc->nr_to_scan));
+ return (0);
+ }
- mutex_exit(&arc_reclaim_lock);
+ mutex_exit(&arc_adjust_lock);
/*
* Evict the requested number of pages by shrinking arc_c the
- * requested amount. If there is nothing left to evict just
- * reap whatever we can from the various arc slabs.
+ * requested amount.
*/
if (pages > 0) {
- arc_shrink(ptob(sc->nr_to_scan));
- arc_kmem_reap_now();
+ arc_reduce_target_size(ptob(sc->nr_to_scan));
+ if (current_is_kswapd())
+ arc_kmem_reap_soon();
#ifdef HAVE_SPLIT_SHRINKER_CALLBACK
- pages = MAX(pages - btop(arc_evictable_memory()), 0);
+ pages = MAX((int64_t)pages -
+ (int64_t)btop(arc_evictable_memory()), 0);
#else
pages = btop(arc_evictable_memory());
#endif
- } else {
- arc_kmem_reap_now();
+ /*
+ * We've shrunk what we can, wake up threads.
+ */
+ cv_broadcast(&arc_adjust_waiters_cv);
+ } else
pages = SHRINK_STOP;
- }
-
- /*
- * We've reaped what we can, wake up threads.
- */
- cv_broadcast(&arc_reclaim_waiters_cv);
/*
* When direct reclaim is observed it usually indicates a rapid
ARCSTAT_BUMP(arcstat_memory_indirect_count);
} else {
arc_no_grow = B_TRUE;
- arc_need_free = ptob(sc->nr_to_scan);
+ arc_kmem_reap_soon();
ARCSTAT_BUMP(arcstat_memory_direct_count);
}
{
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);
+ int64_t mrug_size = zfs_refcount_count(&arc_mru_ghost->arcs_size);
+ int64_t mfug_size = zfs_refcount_count(&arc_mfu_ghost->arcs_size);
if (state == arc_l2c_only)
return;
}
ASSERT((int64_t)arc_p >= 0);
+ /*
+ * Wake reap thread if we do not have any available memory
+ */
if (arc_reclaim_needed()) {
- cv_signal(&arc_reclaim_thread_cv);
+ zthr_wakeup(arc_reap_zthr);
return;
}
* cache size, increment the target cache size
*/
ASSERT3U(arc_c, >=, 2ULL << SPA_MAXBLOCKSHIFT);
- if (arc_size >= arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) {
+ if (aggsum_compare(&arc_size, arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) >=
+ 0) {
atomic_add_64(&arc_c, (int64_t)bytes);
if (arc_c > arc_c_max)
arc_c = arc_c_max;
uint64_t overflow = MAX(SPA_MAXBLOCKSIZE,
arc_c >> zfs_arc_overflow_shift);
- return (arc_size >= arc_c + overflow);
+ /*
+ * We just compare the lower bound here for performance reasons. Our
+ * primary goals are to make sure that the arc never grows without
+ * bound, and that it can reach its maximum size. This check
+ * accomplishes both goals. The maximum amount we could run over by is
+ * 2 * aggsum_borrow_multiplier * NUM_CPUS * the average size of a block
+ * in the ARC. In practice, that's in the tens of MB, which is low
+ * enough to be safe.
+ */
+ return (aggsum_lower_bound(&arc_size) >= arc_c + overflow);
}
static abd_t *
* overflowing; thus we don't use a while loop here.
*/
if (arc_is_overflowing()) {
- mutex_enter(&arc_reclaim_lock);
+ mutex_enter(&arc_adjust_lock);
/*
* Now that we've acquired the lock, we may no longer be
* shouldn't cause any harm.
*/
if (arc_is_overflowing()) {
- cv_signal(&arc_reclaim_thread_cv);
- cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock);
+ arc_adjust_needed = B_TRUE;
+ zthr_wakeup(arc_adjust_zthr);
+ (void) cv_wait(&arc_adjust_waiters_cv,
+ &arc_adjust_lock);
}
-
- mutex_exit(&arc_reclaim_lock);
+ mutex_exit(&arc_adjust_lock);
}
VERIFY3U(hdr->b_type, ==, type);
*/
if (!GHOST_STATE(state)) {
- (void) refcount_add_many(&state->arcs_size, size, tag);
+ (void) zfs_refcount_add_many(&state->arcs_size, size, tag);
/*
* If this is reached via arc_read, the link is
* 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],
+ ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
+ (void) zfs_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))
+ if (aggsum_compare(&arc_size, arc_c) < 0 &&
+ hdr->b_l1hdr.b_state == arc_anon &&
+ (zfs_refcount_count(&arc_anon->arcs_size) +
+ zfs_refcount_count(&arc_mru->arcs_size) > arc_p))
arc_p = MIN(arc_c, arc_p + size);
}
}
/* 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(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT(state != arc_anon && state != arc_l2c_only);
- (void) refcount_remove_many(&state->arcs_esize[type],
+ (void) zfs_refcount_remove_many(&state->arcs_esize[type],
size, tag);
}
- (void) refcount_remove_many(&state->arcs_size, size, tag);
+ (void) zfs_refcount_remove_many(&state->arcs_size, size, tag);
VERIFY3U(hdr->b_type, ==, type);
if (type == ARC_BUFC_METADATA) {
* - 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) {
+ if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) {
+ if (zfs_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);
+ arc_hdr_clear_flags(hdr,
+ ARC_FLAG_PREFETCH |
+ ARC_FLAG_PRESCIENT_PREFETCH);
atomic_inc_32(&hdr->b_l1hdr.b_mru_hits);
ARCSTAT_BUMP(arcstat_mru_hits);
}
* MFU state.
*/
- if (HDR_PREFETCH(hdr)) {
+ if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) {
new_state = arc_mru;
- if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0)
- arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH);
+ if (zfs_refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) {
+ arc_hdr_clear_flags(hdr,
+ ARC_FLAG_PREFETCH |
+ ARC_FLAG_PRESCIENT_PREFETCH);
+ }
DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr);
} else {
new_state = arc_mfu;
* 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));
- }
+
atomic_inc_32(&hdr->b_l1hdr.b_mfu_hits);
ARCSTAT_BUMP(arcstat_mfu_hits);
hdr->b_l1hdr.b_arc_access = ddi_get_lbolt();
* MFU state.
*/
- if (HDR_PREFETCH(hdr)) {
+ if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_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;
}
}
}
-/* a generic arc_done_func_t which you can use */
+/*
+ * This routine is called by dbuf_hold() to update the arc_access() state
+ * which otherwise would be skipped for entries in the dbuf cache.
+ */
+void
+arc_buf_access(arc_buf_t *buf)
+{
+ mutex_enter(&buf->b_evict_lock);
+ arc_buf_hdr_t *hdr = buf->b_hdr;
+
+ /*
+ * Avoid taking the hash_lock when possible as an optimization.
+ * The header must be checked again under the hash_lock in order
+ * to handle the case where it is concurrently being released.
+ */
+ if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) {
+ mutex_exit(&buf->b_evict_lock);
+ return;
+ }
+
+ kmutex_t *hash_lock = HDR_LOCK(hdr);
+ mutex_enter(hash_lock);
+
+ if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) {
+ mutex_exit(hash_lock);
+ mutex_exit(&buf->b_evict_lock);
+ ARCSTAT_BUMP(arcstat_access_skip);
+ return;
+ }
+
+ mutex_exit(&buf->b_evict_lock);
+
+ ASSERT(hdr->b_l1hdr.b_state == arc_mru ||
+ hdr->b_l1hdr.b_state == arc_mfu);
+
+ DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr);
+ arc_access(hdr, hash_lock);
+ mutex_exit(hash_lock);
+
+ ARCSTAT_BUMP(arcstat_hits);
+ ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr) && !HDR_PRESCIENT_PREFETCH(hdr),
+ demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, hits);
+}
+
+/* a generic arc_read_done_func_t which you can use */
/* ARGSUSED */
void
-arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg)
+arc_bcopy_func(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp,
+ arc_buf_t *buf, void *arg)
{
- if (zio == NULL || zio->io_error == 0)
- bcopy(buf->b_data, arg, arc_buf_size(buf));
+ if (buf == NULL)
+ return;
+
+ bcopy(buf->b_data, arg, arc_buf_size(buf));
arc_buf_destroy(buf, arg);
}
-/* a generic arc_done_func_t */
+/* a generic arc_read_done_func_t */
+/* ARGSUSED */
void
-arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg)
+arc_getbuf_func(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp,
+ arc_buf_t *buf, void *arg)
{
arc_buf_t **bufp = arg;
- if (zio && zio->io_error) {
- arc_buf_destroy(buf, arg);
+
+ if (buf == NULL) {
+ ASSERT(zio == NULL || zio->io_error != 0);
*bufp = NULL;
} else {
+ ASSERT(zio == NULL || zio->io_error == 0);
*bufp = buf;
- ASSERT(buf->b_data);
+ ASSERT(buf->b_data != NULL);
}
}
{
if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) {
ASSERT3U(HDR_GET_PSIZE(hdr), ==, 0);
- ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF);
+ ASSERT3U(arc_hdr_get_compress(hdr), ==, ZIO_COMPRESS_OFF);
} else {
if (HDR_COMPRESSION_ENABLED(hdr)) {
- ASSERT3U(HDR_GET_COMPRESS(hdr), ==,
+ ASSERT3U(arc_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));
+ ASSERT3U(!!HDR_PROTECTED(hdr), ==, BP_IS_PROTECTED(bp));
}
}
static void
arc_read_done(zio_t *zio)
{
+ blkptr_t *bp = zio->io_bp;
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);
- int callback_cnt = 0;
+
/*
* The hdr was inserted into hash-table and removed from lists
* prior to starting I/O. We should find this header, since
ASSERT3P(hash_lock, !=, NULL);
}
- if (no_zio_error) {
+ if (BP_IS_PROTECTED(bp)) {
+ hdr->b_crypt_hdr.b_ot = BP_GET_TYPE(bp);
+ hdr->b_crypt_hdr.b_dsobj = zio->io_bookmark.zb_objset;
+ zio_crypt_decode_params_bp(bp, hdr->b_crypt_hdr.b_salt,
+ hdr->b_crypt_hdr.b_iv);
+
+ if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) {
+ void *tmpbuf;
+
+ tmpbuf = abd_borrow_buf_copy(zio->io_abd,
+ sizeof (zil_chain_t));
+ zio_crypt_decode_mac_zil(tmpbuf,
+ hdr->b_crypt_hdr.b_mac);
+ abd_return_buf(zio->io_abd, tmpbuf,
+ sizeof (zil_chain_t));
+ } else {
+ zio_crypt_decode_mac_bp(bp, hdr->b_crypt_hdr.b_mac);
+ }
+ }
+
+ if (zio->io_error == 0) {
/* byteswap if necessary */
if (BP_SHOULD_BYTESWAP(zio->io_bp)) {
if (BP_GET_LEVEL(zio->io_bp) > 0) {
callback_list = hdr->b_l1hdr.b_acb;
ASSERT3P(callback_list, !=, NULL);
- if (hash_lock && no_zio_error && hdr->b_l1hdr.b_state == arc_anon) {
+ if (hash_lock && zio->io_error == 0 &&
+ 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
* 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) {
- int error;
if (!acb->acb_done)
continue;
- /* This is a demand read since prefetches don't use callbacks */
-
callback_cnt++;
- error = arc_buf_alloc_impl(hdr, acb->acb_private,
- acb->acb_compressed, no_zio_error, &acb->acb_buf);
- if (no_zio_error) {
+ if (zio->io_error != 0)
+ continue;
+
+ int error = arc_buf_alloc_impl(hdr, zio->io_spa,
+ &acb->acb_zb, acb->acb_private, acb->acb_encrypted,
+ acb->acb_compressed, acb->acb_noauth, B_TRUE,
+ &acb->acb_buf);
+
+ /*
+ * Assert non-speculative zios didn't fail because an
+ * encryption key wasn't loaded
+ */
+ ASSERT((zio->io_flags & ZIO_FLAG_SPECULATIVE) ||
+ error != EACCES);
+
+ /*
+ * If we failed to decrypt, report an error now (as the zio
+ * layer would have done if it had done the transforms).
+ */
+ if (error == ECKSUM) {
+ ASSERT(BP_IS_PROTECTED(bp));
+ error = SET_ERROR(EIO);
+ if ((zio->io_flags & ZIO_FLAG_SPECULATIVE) == 0) {
+ spa_log_error(zio->io_spa, &acb->acb_zb);
+ zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION,
+ zio->io_spa, NULL, &acb->acb_zb, zio, 0, 0);
+ }
+ }
+
+ if (error != 0) {
+ /*
+ * Decompression or decryption failed. Set
+ * io_error so that when we call acb_done
+ * (below), we will indicate that the read
+ * failed. Note that in the unusual case
+ * where one callback is compressed and another
+ * uncompressed, we will mark all of them
+ * as failed, even though the uncompressed
+ * one can't actually fail. In this case,
+ * the hdr will not be anonymous, because
+ * if there are multiple callbacks, it's
+ * because multiple threads found the same
+ * arc buf in the hash table.
+ */
zio->io_error = error;
}
}
+
+ /*
+ * If there are multiple callbacks, we must have the hash lock,
+ * because the only way for multiple threads to find this hdr is
+ * in the hash table. This ensures that if there are multiple
+ * callbacks, the hdr is not anonymous. If it were anonymous,
+ * we couldn't use arc_buf_destroy() in the error case below.
+ */
+ ASSERT(callback_cnt < 2 || hash_lock != NULL);
+
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);
- }
+ if (callback_cnt == 0)
+ ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr));
- ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) ||
+ ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt) ||
callback_list != NULL);
- if (no_zio_error) {
+ if (zio->io_error == 0) {
arc_hdr_verify(hdr, zio->io_bp);
} else {
arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR);
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);
+ freeable = zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt);
}
/*
* in the cache).
*/
ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
- freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt);
+ freeable = zfs_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_done != NULL) {
+ if (zio->io_error != 0 && acb->acb_buf != NULL) {
+ /*
+ * If arc_buf_alloc_impl() fails during
+ * decompression, the buf will still be
+ * allocated, and needs to be freed here.
+ */
+ arc_buf_destroy(acb->acb_buf,
+ acb->acb_private);
+ acb->acb_buf = NULL;
+ }
+ acb->acb_done(zio, &zio->io_bookmark, zio->io_bp,
+ acb->acb_buf, acb->acb_private);
+ }
if (acb->acb_zio_dummy != NULL) {
acb->acb_zio_dummy->io_error = zio->io_error;
* 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_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
+ arc_read_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;
+ boolean_t compressed_read = (zio_flags & ZIO_FLAG_RAW_COMPRESS) != 0;
+ boolean_t encrypted_read = BP_IS_ENCRYPTED(bp) &&
+ (zio_flags & ZIO_FLAG_RAW_ENCRYPT) != 0;
+ boolean_t noauth_read = BP_IS_AUTHENTICATED(bp) &&
+ (zio_flags & ZIO_FLAG_RAW_ENCRYPT) != 0;
int rc = 0;
ASSERT(!BP_IS_EMBEDDED(bp) ||
hdr = buf_hash_find(guid, bp, &hash_lock);
}
- if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_pabd != NULL) {
+ /*
+ * Determine if we have an L1 cache hit or a cache miss. For simplicity
+ * we maintain encrypted data seperately from compressed / uncompressed
+ * data. If the user is requesting raw encrypted data and we don't have
+ * that in the header we will read from disk to guarantee that we can
+ * get it even if the encryption keys aren't loaded.
+ */
+ if (hdr != NULL && HDR_HAS_L1HDR(hdr) && (HDR_HAS_RABD(hdr) ||
+ (hdr->b_l1hdr.b_pabd != NULL && !encrypted_read))) {
arc_buf_t *buf = NULL;
*arc_flags |= ARC_FLAG_CACHED;
if (HDR_IO_IN_PROGRESS(hdr)) {
+ zio_t *head_zio = hdr->b_l1hdr.b_acb->acb_zio_head;
+ ASSERT3P(head_zio, !=, NULL);
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.
+ * This is a sync read that needs to wait for
+ * an in-flight async read. Request that the
+ * zio have its priority upgraded.
*/
- DTRACE_PROBE1(arc__sync__wait__for__async,
+ zio_change_priority(head_zio, priority);
+ DTRACE_PROBE1(arc__async__upgrade__sync,
arc_buf_hdr_t *, hdr);
- ARCSTAT_BUMP(arcstat_sync_wait_for_async);
+ ARCSTAT_BUMP(arcstat_async_upgrade_sync);
}
if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) {
arc_hdr_clear_flags(hdr,
KM_SLEEP);
acb->acb_done = done;
acb->acb_private = private;
+ acb->acb_compressed = compressed_read;
+ acb->acb_encrypted = encrypted_read;
+ acb->acb_noauth = noauth_read;
+ acb->acb_zb = *zb;
if (pio != NULL)
acb->acb_zio_dummy = zio_null(pio,
spa, NULL, NULL, NULL, zio_flags);
ASSERT3P(acb->acb_done, !=, NULL);
+ acb->acb_zio_head = head_zio;
acb->acb_next = hdr->b_l1hdr.b_acb;
hdr->b_l1hdr.b_acb = acb;
mutex_exit(hash_lock);
arc_hdr_clear_flags(hdr,
ARC_FLAG_PREDICTIVE_PREFETCH);
}
+
+ if (hdr->b_flags & ARC_FLAG_PRESCIENT_PREFETCH) {
+ ARCSTAT_BUMP(
+ arcstat_demand_hit_prescient_prefetch);
+ arc_hdr_clear_flags(hdr,
+ ARC_FLAG_PRESCIENT_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));
+ rc = arc_buf_alloc_impl(hdr, spa, zb, private,
+ encrypted_read, compressed_read, noauth_read,
+ B_TRUE, &buf);
+ if (rc == ECKSUM) {
+ /*
+ * Convert authentication and decryption errors
+ * to EIO (and generate an ereport if needed)
+ * before leaving the ARC.
+ */
+ rc = SET_ERROR(EIO);
+ if ((zio_flags & ZIO_FLAG_SPECULATIVE) == 0) {
+ spa_log_error(spa, zb);
+ zfs_ereport_post(
+ FM_EREPORT_ZFS_AUTHENTICATION,
+ spa, NULL, zb, NULL, 0, 0);
+ }
+ }
+ if (rc != 0) {
+ (void) remove_reference(hdr, hash_lock,
+ private);
+ arc_buf_destroy_impl(buf);
+ buf = NULL;
+ }
+
+ /* assert any errors weren't due to unloaded keys */
+ ASSERT((zio_flags & ZIO_FLAG_SPECULATIVE) ||
+ rc != EACCES);
} else if (*arc_flags & ARC_FLAG_PREFETCH &&
- refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) {
+ zfs_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_PRESCIENT_PREFETCH)
+ arc_hdr_set_flags(hdr, ARC_FLAG_PRESCIENT_PREFETCH);
if (*arc_flags & ARC_FLAG_L2CACHE)
arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE);
mutex_exit(hash_lock);
data, metadata, hits);
if (done)
- done(NULL, buf, private);
+ done(NULL, zb, bp, buf, private);
} else {
uint64_t lsize = BP_GET_LSIZE(bp);
uint64_t psize = BP_GET_PSIZE(bp);
uint64_t addr = 0;
boolean_t devw = B_FALSE;
uint64_t size;
+ abd_t *hdr_abd;
/*
* Gracefully handle a damaged logical block size as a
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);
+ BP_IS_PROTECTED(bp), BP_GET_COMPRESS(bp), type,
+ encrypted_read);
if (!BP_IS_EMBEDDED(bp)) {
hdr->b_dva = *BP_IDENTITY(bp);
}
} 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.
+ * This block is in the ghost cache or encrypted data
+ * was requested and we didn't have it. 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);
+ if (GHOST_STATE(hdr->b_l1hdr.b_state)) {
+ ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
+ ASSERT(!HDR_HAS_RABD(hdr));
+ ASSERT(!HDR_IO_IN_PROGRESS(hdr));
+ ASSERT0(zfs_refcount_count(
+ &hdr->b_l1hdr.b_refcnt));
+ ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
+ ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL);
+ } else if (HDR_IO_IN_PROGRESS(hdr)) {
+ /*
+ * If this header already had an IO in progress
+ * and we are performing another IO to fetch
+ * encrypted data we must wait until the first
+ * IO completes so as not to confuse
+ * arc_read_done(). This should be very rare
+ * and so the performance impact shouldn't
+ * matter.
+ */
+ cv_wait(&hdr->b_l1hdr.b_cv, hash_lock);
+ mutex_exit(hash_lock);
+ goto top;
+ }
/*
* 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
+ * This hdr might be 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
* avoid hitting an assert in remove_reference().
*/
arc_access(hdr, hash_lock);
- arc_hdr_alloc_pabd(hdr);
+ arc_hdr_alloc_abd(hdr, encrypted_read);
}
- 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) {
+ if (encrypted_read) {
+ ASSERT(HDR_HAS_RABD(hdr));
+ size = HDR_GET_PSIZE(hdr);
+ hdr_abd = hdr->b_crypt_hdr.b_rabd;
zio_flags |= ZIO_FLAG_RAW;
+ } else {
+ ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+ size = arc_hdr_size(hdr);
+ hdr_abd = hdr->b_l1hdr.b_pabd;
+
+ if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF) {
+ zio_flags |= ZIO_FLAG_RAW_COMPRESS;
+ }
+
+ /*
+ * For authenticated bp's, we do not ask the ZIO layer
+ * to authenticate them since this will cause the entire
+ * IO to fail if the key isn't loaded. Instead, we
+ * defer authentication until arc_buf_fill(), which will
+ * verify the data when the key is available.
+ */
+ if (BP_IS_AUTHENTICATED(bp))
+ zio_flags |= ZIO_FLAG_RAW_ENCRYPT;
}
- if (*arc_flags & ARC_FLAG_PREFETCH)
+ if (*arc_flags & ARC_FLAG_PREFETCH &&
+ zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt))
arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH);
+ if (*arc_flags & ARC_FLAG_PRESCIENT_PREFETCH)
+ arc_hdr_set_flags(hdr, ARC_FLAG_PRESCIENT_PREFETCH);
if (*arc_flags & ARC_FLAG_L2CACHE)
arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE);
+ if (BP_IS_AUTHENTICATED(bp))
+ arc_hdr_set_flags(hdr, ARC_FLAG_NOAUTH);
if (BP_GET_LEVEL(bp) > 0)
arc_hdr_set_flags(hdr, ARC_FLAG_INDIRECT);
if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH)
acb->acb_done = done;
acb->acb_private = private;
acb->acb_compressed = compressed_read;
+ acb->acb_encrypted = encrypted_read;
+ acb->acb_noauth = noauth_read;
+ acb->acb_zb = *zb;
ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL);
hdr->b_l1hdr.b_acb = acb;
devw = hdr->b_l2hdr.b_dev->l2ad_writing;
addr = hdr->b_l2hdr.b_daddr;
/*
- * Lock out device removal.
+ * Lock out L2ARC device removal.
*/
if (vdev_is_dead(vd) ||
!spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER))
vd = NULL;
}
- if (priority == ZIO_PRIORITY_ASYNC_READ)
+ /*
+ * We count both async reads and scrub IOs as asynchronous so
+ * that both can be upgraded in the event of a cache hit while
+ * the read IO is still in-flight.
+ */
+ if (priority == ZIO_PRIORITY_ASYNC_READ ||
+ priority == ZIO_PRIORITY_SCRUB)
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.
!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->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_abd;
+ }
+
ASSERT(addr >= VDEV_LABEL_START_SIZE &&
- addr + lsize < vd->vdev_psize -
+ addr + asize <= vd->vdev_psize -
VDEV_LABEL_END_SIZE);
/*
* Issue a null zio if the underlying buffer
* was squashed to zero size by compression.
*/
- ASSERT3U(HDR_GET_COMPRESS(hdr), !=,
+ ASSERT3U(arc_hdr_get_compress(hdr), !=,
ZIO_COMPRESS_EMPTY);
rzio = zio_read_phys(pio, vd, addr,
- size, hdr->b_l1hdr.b_pabd,
+ 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);
+ acb->acb_zio_head = rzio;
+
+ if (hash_lock != NULL)
+ mutex_exit(hash_lock);
DTRACE_PROBE2(l2arc__read, vdev_t *, vd,
zio_t *, rzio);
- ARCSTAT_INCR(arcstat_l2_read_bytes, size);
+ ARCSTAT_INCR(arcstat_l2_read_bytes,
+ HDR_GET_PSIZE(hdr));
if (*arc_flags & ARC_FLAG_NOWAIT) {
zio_nowait(rzio);
goto out;
/* l2arc read error; goto zio_read() */
+ if (hash_lock != NULL)
+ mutex_enter(hash_lock);
} else {
DTRACE_PROBE1(l2arc__miss,
arc_buf_hdr_t *, hdr);
}
}
- rzio = zio_read(pio, spa, bp, hdr->b_l1hdr.b_pabd, size,
+ rzio = zio_read(pio, spa, bp, hdr_abd, size,
arc_read_done, hdr, priority, zio_flags, zb);
+ acb->acb_zio_head = rzio;
+
+ if (hash_lock != NULL)
+ mutex_exit(hash_lock);
if (*arc_flags & ARC_FLAG_WAIT) {
rc = zio_wait(rzio);
}
out:
- spa_read_history_add(spa, zb, *arc_flags);
+ /* embedded bps don't actually go to disk */
+ if (!BP_IS_EMBEDDED(bp))
+ spa_read_history_add(spa, zb, *arc_flags);
return (rc);
}
p->p_pfunc = func;
p->p_private = private;
list_link_init(&p->p_node);
- refcount_create(&p->p_refcnt);
+ zfs_refcount_create(&p->p_refcnt);
mutex_enter(&arc_prune_mtx);
- refcount_add(&p->p_refcnt, &arc_prune_list);
+ zfs_refcount_add(&p->p_refcnt, &arc_prune_list);
list_insert_head(&arc_prune_list, p);
mutex_exit(&arc_prune_mtx);
boolean_t wait = B_FALSE;
mutex_enter(&arc_prune_mtx);
list_remove(&arc_prune_list, p);
- if (refcount_remove(&p->p_refcnt, &arc_prune_list) > 0)
+ if (zfs_refcount_remove(&p->p_refcnt, &arc_prune_list) > 0)
wait = B_TRUE;
mutex_exit(&arc_prune_mtx);
/* wait for arc_prune_task to finish */
if (wait)
taskq_wait_outstanding(arc_prune_taskq, 0);
- ASSERT0(refcount_count(&p->p_refcnt));
- refcount_destroy(&p->p_refcnt);
+ ASSERT0(zfs_refcount_count(&p->p_refcnt));
+ zfs_refcount_destroy(&p->p_refcnt);
kmem_free(p, sizeof (*p));
}
* 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))) {
+ zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt))) {
arc_change_state(arc_anon, hdr, hash_lock);
arc_hdr_destroy(hdr);
mutex_exit(hash_lock);
void
arc_release(arc_buf_t *buf, void *tag)
{
- kmutex_t *hash_lock;
- arc_state_t *state;
arc_buf_hdr_t *hdr = buf->b_hdr;
/*
ASSERT(HDR_EMPTY(hdr));
ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1);
- ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1);
+ ASSERT3S(zfs_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;
return;
}
- hash_lock = HDR_LOCK(hdr);
+ kmutex_t *hash_lock = HDR_LOCK(hdr);
mutex_enter(hash_lock);
/*
* held, we must be careful not to reference state or the
* b_state field after dropping the lock.
*/
- state = hdr->b_l1hdr.b_state;
+ 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);
+ ASSERT3S(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt), >, 0);
if (HDR_HAS_L2HDR(hdr)) {
mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx);
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);
+ boolean_t protected = HDR_PROTECTED(hdr);
+ enum zio_compress compress = arc_hdr_get_compress(hdr);
arc_buf_contents_t type = arc_buf_type(hdr);
- arc_buf_t *lastbuf = NULL;
VERIFY3U(hdr->b_type, ==, type);
ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL);
* a new anonymous hdr. Also find the last buffer
* in the hdr's buffer list.
*/
- lastbuf = arc_buf_remove(hdr, buf);
+ arc_buf_t *lastbuf = arc_buf_remove(hdr, buf);
ASSERT3P(lastbuf, !=, NULL);
/*
/*
* First, sever the block sharing relationship between
- * buf and the arc_buf_hdr_t. Then, setup a new
- * block sharing relationship with the last buffer
- * on the arc_buf_t list.
+ * buf and the arc_buf_hdr_t.
*/
arc_unshare_buf(hdr, buf);
if (arc_can_share(hdr, lastbuf)) {
arc_share_buf(hdr, lastbuf);
} else {
- arc_hdr_alloc_pabd(hdr);
+ arc_hdr_alloc_abd(hdr, B_FALSE);
abd_copy_from_buf(hdr->b_l1hdr.b_pabd,
buf->b_data, psize);
}
* if we have a compressed, shared buffer.
*/
ASSERT(arc_buf_is_shared(lastbuf) ||
- HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF);
+ arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF);
ASSERT(!ARC_BUF_SHARED(buf));
}
- ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+
+ ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr));
ASSERT3P(state, !=, arc_l2c_only);
- (void) refcount_remove_many(&state->arcs_size,
+ (void) zfs_refcount_remove_many(&state->arcs_size,
arc_buf_size(buf), buf);
- if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) {
+ if (zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) {
ASSERT3P(state, !=, arc_l2c_only);
- (void) refcount_remove_many(&state->arcs_esize[type],
+ (void) zfs_refcount_remove_many(
+ &state->arcs_esize[type],
arc_buf_size(buf), buf);
}
hdr->b_l1hdr.b_bufcnt -= 1;
+ if (ARC_BUF_ENCRYPTED(buf))
+ hdr->b_crypt_hdr.b_ebufcnt -= 1;
+
arc_cksum_verify(buf);
arc_buf_unwatch(buf);
+ /* if this is the last uncompressed buf free the checksum */
+ if (!arc_hdr_has_uncompressed_buf(hdr))
+ arc_cksum_free(hdr);
+
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);
+ nhdr = arc_hdr_alloc(spa, psize, lsize, protected,
+ compress, type, HDR_HAS_RABD(hdr));
ASSERT3P(nhdr->b_l1hdr.b_buf, ==, NULL);
ASSERT0(nhdr->b_l1hdr.b_bufcnt);
- ASSERT0(refcount_count(&nhdr->b_l1hdr.b_refcnt));
+ ASSERT0(zfs_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;
+ if (ARC_BUF_ENCRYPTED(buf))
+ nhdr->b_crypt_hdr.b_ebufcnt = 1;
nhdr->b_l1hdr.b_mru_hits = 0;
nhdr->b_l1hdr.b_mru_ghost_hits = 0;
nhdr->b_l1hdr.b_mfu_hits = 0;
nhdr->b_l1hdr.b_mfu_ghost_hits = 0;
nhdr->b_l1hdr.b_l2_hits = 0;
- (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag);
+ (void) zfs_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,
- HDR_GET_LSIZE(nhdr), buf);
+ (void) zfs_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);
+ ASSERT(zfs_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));
hdr->b_l1hdr.b_l2_hits = 0;
arc_change_state(arc_anon, hdr, hash_lock);
hdr->b_l1hdr.b_arc_access = 0;
- mutex_exit(hash_lock);
+ mutex_exit(hash_lock);
buf_discard_identity(hdr);
arc_buf_thaw(buf);
}
int referenced;
mutex_enter(&buf->b_evict_lock);
- referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt));
+ referenced = (zfs_refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt));
mutex_exit(&buf->b_evict_lock);
return (referenced);
}
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);
- enum zio_compress compress;
+ blkptr_t *bp = zio->io_bp;
+ uint64_t psize = BP_IS_HOLE(bp) ? 0 : BP_GET_PSIZE(bp);
fstrans_cookie_t cookie = spl_fstrans_mark();
ASSERT(HDR_HAS_L1HDR(hdr));
- ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt));
+ ASSERT(!zfs_refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt));
ASSERT(hdr->b_l1hdr.b_bufcnt > 0);
/*
if (arc_buf_is_shared(buf)) {
arc_unshare_buf(hdr, buf);
} else {
- arc_hdr_free_pabd(hdr);
+ arc_hdr_free_abd(hdr, B_FALSE);
}
}
+
+ if (HDR_HAS_RABD(hdr))
+ arc_hdr_free_abd(hdr, B_TRUE);
}
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
+ ASSERT(!HDR_HAS_RABD(hdr));
ASSERT(!HDR_SHARED_DATA(hdr));
ASSERT(!arc_buf_is_shared(buf));
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);
- if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) {
+ if (BP_IS_PROTECTED(bp) != !!HDR_PROTECTED(hdr))
+ hdr = arc_hdr_realloc_crypt(hdr, BP_IS_PROTECTED(bp));
+
+ if (BP_IS_PROTECTED(bp)) {
+ /* ZIL blocks are written through zio_rewrite */
+ ASSERT3U(BP_GET_TYPE(bp), !=, DMU_OT_INTENT_LOG);
+ ASSERT(HDR_PROTECTED(hdr));
+
+ if (BP_SHOULD_BYTESWAP(bp)) {
+ if (BP_GET_LEVEL(bp) > 0) {
+ hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64;
+ } else {
+ hdr->b_l1hdr.b_byteswap =
+ DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
+ }
+ } else {
+ hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
+ }
+
+ hdr->b_crypt_hdr.b_ot = BP_GET_TYPE(bp);
+ hdr->b_crypt_hdr.b_dsobj = zio->io_bookmark.zb_objset;
+ zio_crypt_decode_params_bp(bp, hdr->b_crypt_hdr.b_salt,
+ hdr->b_crypt_hdr.b_iv);
+ zio_crypt_decode_mac_bp(bp, hdr->b_crypt_hdr.b_mac);
+ }
+
+ /*
+ * If this block was written for raw encryption but the zio layer
+ * ended up only authenticating it, adjust the buffer flags now.
+ */
+ if (BP_IS_AUTHENTICATED(bp) && ARC_BUF_ENCRYPTED(buf)) {
+ arc_hdr_set_flags(hdr, ARC_FLAG_NOAUTH);
+ buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED;
+ if (BP_GET_COMPRESS(bp) == ZIO_COMPRESS_OFF)
+ buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED;
+ } else if (BP_IS_HOLE(bp) && ARC_BUF_ENCRYPTED(buf)) {
+ buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED;
+ buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED;
+ }
+
+ /* this must be done after the buffer flags are adjusted */
+ arc_cksum_compute(buf);
+
+ enum zio_compress compress;
+ if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) {
compress = ZIO_COMPRESS_OFF;
} else {
- ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(zio->io_bp));
- compress = BP_GET_COMPRESS(zio->io_bp);
+ ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp));
+ compress = BP_GET_COMPRESS(bp);
}
HDR_SET_PSIZE(hdr, psize);
arc_hdr_set_compress(hdr, compress);
+ if (zio->io_error != 0 || psize == 0)
+ goto out;
+
/*
- * 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.
+ * Fill the hdr with data. If the buffer is encrypted we have no choice
+ * but to copy the data into b_radb. 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
* 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);
-
+ if (ARC_BUF_ENCRYPTED(buf)) {
+ ASSERT3U(psize, >, 0);
+ ASSERT(ARC_BUF_COMPRESSED(buf));
+ arc_hdr_alloc_abd(hdr, B_TRUE);
+ abd_copy(hdr->b_crypt_hdr.b_rabd, zio->io_abd, psize);
+ } else if (zfs_abd_scatter_enabled || !arc_can_share(hdr, buf)) {
/*
* 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);
+ if (BP_IS_ENCRYPTED(bp)) {
ASSERT3U(psize, >, 0);
-
+ arc_hdr_alloc_abd(hdr, B_TRUE);
+ abd_copy(hdr->b_crypt_hdr.b_rabd, zio->io_abd, psize);
+ } else if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF &&
+ !ARC_BUF_COMPRESSED(buf)) {
+ ASSERT3U(psize, >, 0);
+ arc_hdr_alloc_abd(hdr, B_FALSE);
abd_copy(hdr->b_l1hdr.b_pabd, zio->io_abd, psize);
} else {
ASSERT3U(zio->io_orig_size, ==, arc_hdr_size(hdr));
-
+ arc_hdr_alloc_abd(hdr, B_FALSE);
abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data,
arc_buf_size(buf));
}
arc_share_buf(hdr, buf);
}
- arc_hdr_verify(hdr, zio->io_bp);
+out:
+ arc_hdr_verify(hdr, bp);
spl_fstrans_unmark(cookie);
}
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(
+ ASSERT(zfs_refcount_is_zero(
&exists->b_l1hdr.b_refcnt));
arc_change_state(arc_anon, exists, hash_lock);
mutex_exit(hash_lock);
arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS);
}
- ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
+ ASSERT(!zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
callback->awcb_done(zio, buf, callback->awcb_private);
abd_put(zio->io_abd);
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,
+ const zio_prop_t *zp, arc_write_done_func_t *ready,
+ arc_write_done_func_t *children_ready, arc_write_done_func_t *physdone,
+ arc_write_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);
ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0);
if (l2arc)
arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE);
- if (ARC_BUF_COMPRESSED(buf)) {
- ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_OFF);
+
+ if (ARC_BUF_ENCRYPTED(buf)) {
+ ASSERT(ARC_BUF_COMPRESSED(buf));
+ localprop.zp_encrypt = B_TRUE;
+ localprop.zp_compress = HDR_GET_COMPRESS(hdr);
+ localprop.zp_byteorder =
+ (hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS) ?
+ ZFS_HOST_BYTEORDER : !ZFS_HOST_BYTEORDER;
+ bcopy(hdr->b_crypt_hdr.b_salt, localprop.zp_salt,
+ ZIO_DATA_SALT_LEN);
+ bcopy(hdr->b_crypt_hdr.b_iv, localprop.zp_iv,
+ ZIO_DATA_IV_LEN);
+ bcopy(hdr->b_crypt_hdr.b_mac, localprop.zp_mac,
+ ZIO_DATA_MAC_LEN);
+ if (DMU_OT_IS_ENCRYPTED(localprop.zp_type)) {
+ localprop.zp_nopwrite = B_FALSE;
+ localprop.zp_copies =
+ MIN(localprop.zp_copies, SPA_DVAS_PER_BP - 1);
+ }
zio_flags |= ZIO_FLAG_RAW;
+ } else if (ARC_BUF_COMPRESSED(buf)) {
+ ASSERT3U(HDR_GET_LSIZE(hdr), !=, arc_buf_size(buf));
+ localprop.zp_compress = HDR_GET_COMPRESS(hdr);
+ zio_flags |= ZIO_FLAG_RAW_COMPRESS;
}
callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP);
callback->awcb_ready = ready;
if (arc_buf_is_shared(buf)) {
arc_unshare_buf(hdr, buf);
} else {
- arc_hdr_free_pabd(hdr);
+ arc_hdr_free_abd(hdr, B_FALSE);
}
VERIFY3P(buf->b_data, !=, NULL);
- arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF);
}
+
+ if (HDR_HAS_RABD(hdr))
+ arc_hdr_free_abd(hdr, B_TRUE);
+
+ if (!(zio_flags & ZIO_FLAG_RAW))
+ 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), zp,
- arc_write_ready,
+ 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);
}
static int
-arc_memory_throttle(uint64_t reserve, uint64_t txg)
+arc_memory_throttle(spa_t *spa, 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;
-#ifdef __linux__
- pgcnt_t minfree = btop(arc_sys_free / 4);
-#endif
+ uint64_t available_memory = arc_free_memory();
-#if defined(__i386)
+#if defined(_ILP32)
available_memory =
MIN(available_memory, vmem_size(heap_arena, VMEM_FREE));
#endif
if (available_memory > arc_all_memory() * arc_lotsfree_percent / 100)
return (0);
- if (txg > last_txg) {
- last_txg = txg;
- page_load = 0;
+ if (txg > spa->spa_lowmem_last_txg) {
+ spa->spa_lowmem_last_txg = txg;
+ spa->spa_lowmem_page_load = 0;
}
/*
* If we are in pageout, we know that memory is already tight,
* continue to let page writes occur as quickly as possible.
*/
if (current_is_kswapd()) {
- if (page_load > MAX(ptob(minfree), available_memory) / 4) {
+ if (spa->spa_lowmem_page_load >
+ MAX(arc_sys_free / 4, available_memory) / 4) {
DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim);
return (SET_ERROR(ERESTART));
}
/* Note: reserve is inflated, so we deflate */
- page_load += reserve / 8;
+ atomic_add_64(&spa->spa_lowmem_page_load, reserve / 8);
return (0);
- } else if (page_load > 0 && arc_reclaim_needed()) {
+ } else if (spa->spa_lowmem_page_load > 0 && arc_reclaim_needed()) {
/* memory is low, delay before restarting */
ARCSTAT_INCR(arcstat_memory_throttle_count, 1);
DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim);
return (SET_ERROR(EAGAIN));
}
- page_load = 0;
-#endif
+ spa->spa_lowmem_page_load = 0;
+#endif /* _KERNEL */
return (0);
}
}
int
-arc_tempreserve_space(uint64_t reserve, uint64_t txg)
+arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg)
{
int error;
uint64_t anon_size;
* network delays from blocking transactions that are ready to be
* assigned to a txg.
*/
- anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) -
+
+ /* assert that it has not wrapped around */
+ ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0);
+
+ anon_size = MAX((int64_t)(zfs_refcount_count(&arc_anon->arcs_size) -
arc_loaned_bytes), 0);
/*
* 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);
+ error = arc_memory_throttle(spa, 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.
+ *
+ * In the case of one pool being built on another pool, we want
+ * to make sure we don't end up throttling the lower (backing)
+ * pool when the upper pool is the majority contributor to dirty
+ * data. To insure we make forward progress during throttling, we
+ * also check the current pool's net dirty data and only throttle
+ * if it exceeds zfs_arc_pool_dirty_percent of the anonymous dirty
+ * data in the cache.
+ *
* Note: if two requests come in concurrently, we might let them
* both succeed, when one of them should fail. Not a huge deal.
*/
+ uint64_t total_dirty = reserve + arc_tempreserve + anon_size;
+ uint64_t spa_dirty_anon = spa_dirty_data(spa);
- 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]);
+ if (total_dirty > arc_c * zfs_arc_dirty_limit_percent / 100 &&
+ anon_size > arc_c * zfs_arc_anon_limit_percent / 100 &&
+ spa_dirty_anon > anon_size * zfs_arc_pool_dirty_percent / 100) {
+#ifdef ZFS_DEBUG
+ uint64_t meta_esize = zfs_refcount_count(
+ &arc_anon->arcs_esize[ARC_BUFC_METADATA]);
uint64_t data_esize =
- refcount_count(&arc_anon->arcs_esize[ARC_BUFC_DATA]);
+ zfs_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);
+#endif
DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle);
return (SET_ERROR(ERESTART));
}
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);
+ size->value.ui64 = zfs_refcount_count(&state->arcs_size);
evict_data->value.ui64 =
- refcount_count(&state->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_count(&state->arcs_esize[ARC_BUFC_DATA]);
evict_metadata->value.ui64 =
- refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]);
}
static int
arc_stats_t *as = ksp->ks_data;
if (rw == KSTAT_WRITE) {
- return (EACCES);
+ return (SET_ERROR(EACCES));
} else {
arc_kstat_update_state(arc_anon,
&as->arcstat_anon_size,
&as->arcstat_mfu_ghost_size,
&as->arcstat_mfu_ghost_evictable_data,
&as->arcstat_mfu_ghost_evictable_metadata);
+
+ ARCSTAT(arcstat_size) = aggsum_value(&arc_size);
+ ARCSTAT(arcstat_meta_used) = aggsum_value(&arc_meta_used);
+ ARCSTAT(arcstat_data_size) = aggsum_value(&astat_data_size);
+ ARCSTAT(arcstat_metadata_size) =
+ aggsum_value(&astat_metadata_size);
+ ARCSTAT(arcstat_hdr_size) = aggsum_value(&astat_hdr_size);
+ ARCSTAT(arcstat_l2_hdr_size) = aggsum_value(&astat_l2_hdr_size);
+ ARCSTAT(arcstat_dbuf_size) = aggsum_value(&astat_dbuf_size);
+ ARCSTAT(arcstat_dnode_size) = aggsum_value(&astat_dnode_size);
+ ARCSTAT(arcstat_bonus_size) = aggsum_value(&astat_bonus_size);
+
+ as->arcstat_memory_all_bytes.value.ui64 =
+ arc_all_memory();
+ as->arcstat_memory_free_bytes.value.ui64 =
+ arc_free_memory();
+ as->arcstat_memory_available_bytes.value.i64 =
+ arc_available_memory();
}
return (0);
static void
arc_tuning_update(void)
{
- uint64_t percent, allmem = arc_all_memory();
+ uint64_t allmem = arc_all_memory();
+ unsigned long limit;
/* Valid range: 64M - <all physical memory> */
if ((zfs_arc_max) && (zfs_arc_max != arc_c_max) &&
- (zfs_arc_max > 64 << 20) && (zfs_arc_max < allmem) &&
+ (zfs_arc_max >= 64 << 20) && (zfs_arc_max < allmem) &&
(zfs_arc_max > arc_c_min)) {
arc_c_max = zfs_arc_max;
arc_c = arc_c_max;
arc_p = (arc_c >> 1);
- /* Valid range of arc_meta_limit: arc_meta_min - arc_c_max */
- percent = MIN(zfs_arc_meta_limit_percent, 100);
- arc_meta_limit = MAX(arc_meta_min, (percent * arc_c_max) / 100);
- percent = MIN(zfs_arc_dnode_limit_percent, 100);
- arc_dnode_limit = (percent * arc_meta_limit) / 100;
+ if (arc_meta_limit > arc_c_max)
+ arc_meta_limit = arc_c_max;
+ if (arc_dnode_limit > arc_meta_limit)
+ arc_dnode_limit = arc_meta_limit;
}
/* Valid range: 32M - <arc_c_max> */
(zfs_arc_meta_min >= 1ULL << SPA_MAXBLOCKSHIFT) &&
(zfs_arc_meta_min <= arc_c_max)) {
arc_meta_min = zfs_arc_meta_min;
- arc_meta_limit = MAX(arc_meta_limit, arc_meta_min);
- arc_dnode_limit = arc_meta_limit / 10;
+ if (arc_meta_limit < arc_meta_min)
+ arc_meta_limit = arc_meta_min;
+ if (arc_dnode_limit < arc_meta_min)
+ arc_dnode_limit = arc_meta_min;
}
/* Valid range: <arc_meta_min> - <arc_c_max> */
- if ((zfs_arc_meta_limit) && (zfs_arc_meta_limit != arc_meta_limit) &&
- (zfs_arc_meta_limit >= zfs_arc_meta_min) &&
- (zfs_arc_meta_limit <= arc_c_max))
- arc_meta_limit = zfs_arc_meta_limit;
-
- /* Valid range: <arc_meta_min> - <arc_c_max> */
- if ((zfs_arc_dnode_limit) && (zfs_arc_dnode_limit != arc_dnode_limit) &&
- (zfs_arc_dnode_limit >= zfs_arc_meta_min) &&
- (zfs_arc_dnode_limit <= arc_c_max))
- arc_dnode_limit = zfs_arc_dnode_limit;
+ limit = zfs_arc_meta_limit ? zfs_arc_meta_limit :
+ MIN(zfs_arc_meta_limit_percent, 100) * arc_c_max / 100;
+ if ((limit != arc_meta_limit) &&
+ (limit >= arc_meta_min) &&
+ (limit <= arc_c_max))
+ arc_meta_limit = limit;
+
+ /* Valid range: <arc_meta_min> - <arc_meta_limit> */
+ limit = zfs_arc_dnode_limit ? zfs_arc_dnode_limit :
+ MIN(zfs_arc_dnode_limit_percent, 100) * arc_meta_limit / 100;
+ if ((limit != arc_dnode_limit) &&
+ (limit >= arc_meta_min) &&
+ (limit <= arc_meta_limit))
+ arc_dnode_limit = limit;
/* Valid range: 1 - N */
if (zfs_arc_grow_retry)
if (zfs_arc_p_min_shift)
arc_p_min_shift = zfs_arc_p_min_shift;
- /* Valid range: 1 - N ticks */
- if (zfs_arc_min_prefetch_lifespan)
- arc_min_prefetch_lifespan = zfs_arc_min_prefetch_lifespan;
+ /* Valid range: 1 - N ms */
+ if (zfs_arc_min_prefetch_ms)
+ arc_min_prefetch_ms = zfs_arc_min_prefetch_ms;
+
+ /* Valid range: 1 - N ms */
+ if (zfs_arc_min_prescient_prefetch_ms) {
+ arc_min_prescient_prefetch_ms =
+ zfs_arc_min_prescient_prefetch_ms;
+ }
/* Valid range: 0 - 100 */
if ((zfs_arc_lotsfree_percent >= 0) &&
arc_mfu_ghost = &ARC_mfu_ghost;
arc_l2c_only = &ARC_l2c_only;
- multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA],
- sizeof (arc_buf_hdr_t),
+ 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);
- multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA],
- sizeof (arc_buf_hdr_t),
+ 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);
+ zfs_refcount_create(&arc_anon->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_create(&arc_anon->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_create(&arc_mru->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_create(&arc_mru->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]);
+
+ zfs_refcount_create(&arc_anon->arcs_size);
+ zfs_refcount_create(&arc_mru->arcs_size);
+ zfs_refcount_create(&arc_mru_ghost->arcs_size);
+ zfs_refcount_create(&arc_mfu->arcs_size);
+ zfs_refcount_create(&arc_mfu_ghost->arcs_size);
+ zfs_refcount_create(&arc_l2c_only->arcs_size);
+
+ aggsum_init(&arc_meta_used, 0);
+ aggsum_init(&arc_size, 0);
+ aggsum_init(&astat_data_size, 0);
+ aggsum_init(&astat_metadata_size, 0);
+ aggsum_init(&astat_hdr_size, 0);
+ aggsum_init(&astat_l2_hdr_size, 0);
+ aggsum_init(&astat_bonus_size, 0);
+ aggsum_init(&astat_dnode_size, 0);
+ aggsum_init(&astat_dbuf_size, 0);
arc_anon->arcs_state = ARC_STATE_ANON;
arc_mru->arcs_state = ARC_STATE_MRU;
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]);
- multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]);
- multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]);
+ zfs_refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]);
+ zfs_refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]);
+ zfs_refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]);
+
+ zfs_refcount_destroy(&arc_anon->arcs_size);
+ zfs_refcount_destroy(&arc_mru->arcs_size);
+ zfs_refcount_destroy(&arc_mru_ghost->arcs_size);
+ zfs_refcount_destroy(&arc_mfu->arcs_size);
+ zfs_refcount_destroy(&arc_mfu_ghost->arcs_size);
+ zfs_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]);
+ multilist_destroy(arc_l2c_only->arcs_list[ARC_BUFC_METADATA]);
+ multilist_destroy(arc_l2c_only->arcs_list[ARC_BUFC_DATA]);
+
+ aggsum_fini(&arc_meta_used);
+ aggsum_fini(&arc_size);
+ aggsum_fini(&astat_data_size);
+ aggsum_fini(&astat_metadata_size);
+ aggsum_fini(&astat_hdr_size);
+ aggsum_fini(&astat_l2_hdr_size);
+ aggsum_fini(&astat_bonus_size);
+ aggsum_fini(&astat_dnode_size);
+ aggsum_fini(&astat_dbuf_size);
}
uint64_t
-arc_max_bytes(void)
+arc_target_bytes(void)
{
- return (arc_c_max);
+ return (arc_c);
}
void
arc_init(void)
{
uint64_t percent, allmem = arc_all_memory();
+ mutex_init(&arc_adjust_lock, NULL, MUTEX_DEFAULT, NULL);
+ cv_init(&arc_adjust_waiters_cv, NULL, CV_DEFAULT, NULL);
- 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);
-
- /* Convert seconds to clock ticks */
- arc_min_prefetch_lifespan = 1 * hz;
+ arc_min_prefetch_ms = 1000;
+ arc_min_prescient_prefetch_ms = 6000;
#ifdef _KERNEL
/*
/* Set max to 1/2 of all memory */
arc_c_max = allmem / 2;
+#ifdef _KERNEL
+ /* Set min cache to 1/32 of all memory, or 32MB, whichever is more */
+ arc_c_min = MAX(allmem / 32, 2ULL << SPA_MAXBLOCKSHIFT);
+#else
/*
* 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 = MAX(arc_c_max / 2, 2ULL << SPA_MAXBLOCKSHIFT);
-#else
- arc_c_min = 2ULL << SPA_MAXBLOCKSHIFT;
#endif
arc_c = arc_c_max;
arc_p = (arc_c >> 1);
- arc_size = 0;
/* Set min to 1/2 of arc_c_min */
arc_meta_min = 1ULL << SPA_MAXBLOCKSHIFT;
arc_c = arc_c_min;
arc_state_init();
+
+ /*
+ * The arc must be "uninitialized", so that hdr_recl() (which is
+ * registered by buf_init()) will not access arc_reap_zthr before
+ * it is created.
+ */
+ ASSERT(!arc_initialized);
buf_init();
list_create(&arc_prune_list, sizeof (arc_prune_t),
arc_prune_taskq = taskq_create("arc_prune", max_ncpus, defclsyspri,
max_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC);
- arc_reclaim_thread_exit = B_FALSE;
-
arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED,
sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
kstat_install(arc_ksp);
}
- (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0,
- TS_RUN, defclsyspri);
+ arc_adjust_zthr = zthr_create(arc_adjust_cb_check,
+ arc_adjust_cb, NULL);
+ arc_reap_zthr = zthr_create_timer(arc_reap_cb_check,
+ arc_reap_cb, NULL, SEC2NSEC(1));
- arc_dead = B_FALSE;
+ arc_initialized = B_TRUE;
arc_warm = B_FALSE;
/*
* If it has been set by a module parameter, 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 25% of physical memory).
+ * zfs_dirty_data_max_max (default 4G or 25% of physical memory).
*/
if (zfs_dirty_data_max_max == 0)
- zfs_dirty_data_max_max = allmem *
- zfs_dirty_data_max_max_percent / 100;
+ zfs_dirty_data_max_max = MIN(4ULL * 1024 * 1024 * 1024,
+ allmem * zfs_dirty_data_max_max_percent / 100);
if (zfs_dirty_data_max == 0) {
zfs_dirty_data_max = allmem *
spl_unregister_shrinker(&arc_shrinker);
#endif /* _KERNEL */
- 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);
- arc_dead = B_TRUE;
+ arc_initialized = B_FALSE;
if (arc_ksp != NULL) {
kstat_delete(arc_ksp);
mutex_enter(&arc_prune_mtx);
while ((p = list_head(&arc_prune_list)) != NULL) {
list_remove(&arc_prune_list, p);
- refcount_remove(&p->p_refcnt, &arc_prune_list);
- refcount_destroy(&p->p_refcnt);
+ zfs_refcount_remove(&p->p_refcnt, &arc_prune_list);
+ zfs_refcount_destroy(&p->p_refcnt);
kmem_free(p, sizeof (*p));
}
mutex_exit(&arc_prune_mtx);
list_destroy(&arc_prune_list);
mutex_destroy(&arc_prune_mtx);
- mutex_destroy(&arc_reclaim_lock);
- cv_destroy(&arc_reclaim_thread_cv);
- cv_destroy(&arc_reclaim_waiters_cv);
+ (void) zthr_cancel(arc_adjust_zthr);
+ zthr_destroy(arc_adjust_zthr);
- arc_state_fini();
+ (void) zthr_cancel(arc_reap_zthr);
+ zthr_destroy(arc_reap_zthr);
+
+ mutex_destroy(&arc_adjust_lock);
+ cv_destroy(&arc_adjust_waiters_cv);
+
+ /*
+ * buf_fini() must proceed arc_state_fini() because buf_fin() may
+ * trigger the release of kmem magazines, which can callback to
+ * arc_space_return() which accesses aggsums freed in act_state_fini().
+ */
buf_fini();
+ arc_state_fini();
ASSERT0(arc_loaned_bytes);
}
list_remove(buflist, hdr);
arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR);
- ARCSTAT_INCR(arcstat_l2_asize, -arc_hdr_size(hdr));
- ARCSTAT_INCR(arcstat_l2_size, -HDR_GET_LSIZE(hdr));
+ uint64_t psize = HDR_GET_PSIZE(hdr);
+ ARCSTAT_INCR(arcstat_l2_psize, -psize);
+ ARCSTAT_INCR(arcstat_l2_lsize, -HDR_GET_LSIZE(hdr));
- bytes_dropped += arc_hdr_size(hdr);
- (void) refcount_remove_many(&dev->l2ad_alloc,
+ bytes_dropped +=
+ vdev_psize_to_asize(dev->l2ad_vdev, psize);
+ (void) zfs_refcount_remove_many(&dev->l2ad_alloc,
arc_hdr_size(hdr), hdr);
}
kmem_free(cb, sizeof (l2arc_write_callback_t));
}
+static int
+l2arc_untransform(zio_t *zio, l2arc_read_callback_t *cb)
+{
+ int ret;
+ spa_t *spa = zio->io_spa;
+ arc_buf_hdr_t *hdr = cb->l2rcb_hdr;
+ blkptr_t *bp = zio->io_bp;
+ uint8_t salt[ZIO_DATA_SALT_LEN];
+ uint8_t iv[ZIO_DATA_IV_LEN];
+ uint8_t mac[ZIO_DATA_MAC_LEN];
+ boolean_t no_crypt = B_FALSE;
+
+ /*
+ * ZIL data is never be written to the L2ARC, so we don't need
+ * special handling for its unique MAC storage.
+ */
+ ASSERT3U(BP_GET_TYPE(bp), !=, DMU_OT_INTENT_LOG);
+ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)));
+ ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+
+ /*
+ * If the data was encrypted, decrypt it now. Note that
+ * we must check the bp here and not the hdr, since the
+ * hdr does not have its encryption parameters updated
+ * until arc_read_done().
+ */
+ if (BP_IS_ENCRYPTED(bp)) {
+ abd_t *eabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr);
+
+ zio_crypt_decode_params_bp(bp, salt, iv);
+ zio_crypt_decode_mac_bp(bp, mac);
+
+ ret = spa_do_crypt_abd(B_FALSE, spa, &cb->l2rcb_zb,
+ BP_GET_TYPE(bp), BP_GET_DEDUP(bp), BP_SHOULD_BYTESWAP(bp),
+ salt, iv, mac, HDR_GET_PSIZE(hdr), eabd,
+ hdr->b_l1hdr.b_pabd, &no_crypt);
+ if (ret != 0) {
+ arc_free_data_abd(hdr, eabd, arc_hdr_size(hdr), hdr);
+ goto error;
+ }
+
+ /*
+ * If we actually performed decryption, replace b_pabd
+ * with the decrypted data. Otherwise we can just throw
+ * our decryption buffer away.
+ */
+ if (!no_crypt) {
+ arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd,
+ arc_hdr_size(hdr), hdr);
+ hdr->b_l1hdr.b_pabd = eabd;
+ zio->io_abd = eabd;
+ } else {
+ arc_free_data_abd(hdr, eabd, arc_hdr_size(hdr), hdr);
+ }
+ }
+
+ /*
+ * If the L2ARC block was compressed, but ARC compression
+ * is disabled we decompress the data into a new buffer and
+ * replace the existing data.
+ */
+ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
+ !HDR_COMPRESSION_ENABLED(hdr)) {
+ abd_t *cabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr);
+ void *tmp = abd_borrow_buf(cabd, arc_hdr_size(hdr));
+
+ ret = zio_decompress_data(HDR_GET_COMPRESS(hdr),
+ hdr->b_l1hdr.b_pabd, tmp, HDR_GET_PSIZE(hdr),
+ HDR_GET_LSIZE(hdr));
+ if (ret != 0) {
+ abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr));
+ arc_free_data_abd(hdr, cabd, arc_hdr_size(hdr), hdr);
+ goto error;
+ }
+
+ abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr));
+ arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd,
+ arc_hdr_size(hdr), hdr);
+ hdr->b_l1hdr.b_pabd = cabd;
+ zio->io_abd = cabd;
+ zio->io_size = HDR_GET_LSIZE(hdr);
+ }
+
+ return (0);
+
+error:
+ return (ret);
+}
+
+
/*
* 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;
+ int tfm_error = 0;
+ l2arc_read_callback_t *cb = zio->io_private;
arc_buf_hdr_t *hdr;
kmutex_t *hash_lock;
boolean_t valid_cksum;
+ boolean_t using_rdata = (BP_IS_ENCRYPTED(&cb->l2rcb_bp) &&
+ (cb->l2rcb_flags & ZIO_FLAG_RAW_ENCRYPT));
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);
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) {
+ if (using_rdata) {
+ abd_copy(hdr->b_crypt_hdr.b_rabd,
+ cb->l2rcb_abd, arc_hdr_size(hdr));
+ } else {
+ 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);
+
+ if (using_rdata) {
+ ASSERT(HDR_HAS_RABD(hdr));
+ zio->io_abd = zio->io_orig_abd =
+ hdr->b_crypt_hdr.b_rabd;
+ } else {
+ ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
+ 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);
+ ASSERT(zio->io_abd == hdr->b_l1hdr.b_pabd ||
+ (HDR_HAS_RABD(hdr) && zio->io_abd == hdr->b_crypt_hdr.b_rabd));
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)) {
+
+ /*
+ * b_rabd will always match the data as it exists on disk if it is
+ * being used. Therefore if we are reading into b_rabd we do not
+ * attempt to untransform the data.
+ */
+ if (valid_cksum && !using_rdata)
+ tfm_error = l2arc_untransform(zio, cb);
+
+ if (valid_cksum && tfm_error == 0 && zio->io_error == 0 &&
+ !HDR_L2_EVICTED(hdr)) {
mutex_exit(hash_lock);
zio->io_private = hdr;
arc_read_done(zio);
} else {
zio->io_error = SET_ERROR(EIO);
}
- if (!valid_cksum)
+ if (!valid_cksum || tfm_error != 0)
ARCSTAT_BUMP(arcstat_l2_cksum_bad);
/*
*/
if (zio->io_waiter == NULL) {
zio_t *pio = zio_unique_parent(zio);
+ void *abd = (using_rdata) ?
+ hdr->b_crypt_hdr.b_rabd : hdr->b_l1hdr.b_pabd;
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,
+ abd, zio->io_size, arc_read_done,
hdr, zio->io_priority, cb->l2rcb_flags,
&cb->l2rcb_zb));
}
switch (list_num) {
case 0:
- ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA];
+ ml = arc_mfu->arcs_list[ARC_BUFC_METADATA];
break;
case 1:
- ml = &arc_mru->arcs_list[ARC_BUFC_METADATA];
+ ml = arc_mru->arcs_list[ARC_BUFC_METADATA];
break;
case 2:
- ml = &arc_mfu->arcs_list[ARC_BUFC_DATA];
+ ml = arc_mfu->arcs_list[ARC_BUFC_DATA];
break;
case 3:
- ml = &arc_mru->arcs_list[ARC_BUFC_DATA];
+ ml = arc_mru->arcs_list[ARC_BUFC_DATA];
break;
default:
return (NULL);
goto top;
}
- if (HDR_L2_WRITE_HEAD(hdr)) {
- /*
- * We hit a write head node. Leave it for
- * l2arc_write_done().
- */
- list_remove(buflist, hdr);
- mutex_exit(hash_lock);
- continue;
- }
+ /*
+ * 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_HAS_L2HDR(hdr) &&
- (hdr->b_l2hdr.b_daddr > taddr ||
+ if (!all && (hdr->b_l2hdr.b_daddr >= taddr ||
hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) {
/*
* We've evicted to the target address,
break;
}
- ASSERT(HDR_HAS_L2HDR(hdr));
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_size.
+ * and decrement arcstat_l2_lsize.
*/
arc_change_state(arc_anon, hdr, hash_lock);
arc_hdr_destroy(hdr);
arc_hdr_set_flags(hdr, ARC_FLAG_L2_EVICTED);
}
- /* Ensure this header has finished being written */
- ASSERT(!HDR_L2_WRITING(hdr));
-
arc_hdr_l2hdr_destroy(hdr);
}
mutex_exit(hash_lock);
mutex_exit(&dev->l2ad_mtx);
}
+/*
+ * Handle any abd transforms that might be required for writing to the L2ARC.
+ * If successful, this function will always return an abd with the data
+ * transformed as it is on disk in a new abd of asize bytes.
+ */
+static int
+l2arc_apply_transforms(spa_t *spa, arc_buf_hdr_t *hdr, uint64_t asize,
+ abd_t **abd_out)
+{
+ int ret;
+ void *tmp = NULL;
+ abd_t *cabd = NULL, *eabd = NULL, *to_write = hdr->b_l1hdr.b_pabd;
+ enum zio_compress compress = HDR_GET_COMPRESS(hdr);
+ uint64_t psize = HDR_GET_PSIZE(hdr);
+ uint64_t size = arc_hdr_size(hdr);
+ boolean_t ismd = HDR_ISTYPE_METADATA(hdr);
+ boolean_t bswap = (hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS);
+ dsl_crypto_key_t *dck = NULL;
+ uint8_t mac[ZIO_DATA_MAC_LEN] = { 0 };
+ boolean_t no_crypt = B_FALSE;
+
+ ASSERT((HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
+ !HDR_COMPRESSION_ENABLED(hdr)) ||
+ HDR_ENCRYPTED(hdr) || HDR_SHARED_DATA(hdr) || psize != asize);
+ ASSERT3U(psize, <=, asize);
+
+ /*
+ * If this data simply needs its own buffer, we simply allocate it
+ * and copy the data. This may be done to elimiate a depedency on a
+ * shared buffer or to reallocate the buffer to match asize.
+ */
+ if (HDR_HAS_RABD(hdr) && asize != psize) {
+ ASSERT3U(asize, >=, psize);
+ to_write = abd_alloc_for_io(asize, ismd);
+ abd_copy(to_write, hdr->b_crypt_hdr.b_rabd, psize);
+ if (psize != asize)
+ abd_zero_off(to_write, psize, asize - psize);
+ goto out;
+ }
+
+ if ((compress == ZIO_COMPRESS_OFF || HDR_COMPRESSION_ENABLED(hdr)) &&
+ !HDR_ENCRYPTED(hdr)) {
+ ASSERT3U(size, ==, psize);
+ to_write = abd_alloc_for_io(asize, ismd);
+ abd_copy(to_write, hdr->b_l1hdr.b_pabd, size);
+ if (size != asize)
+ abd_zero_off(to_write, size, asize - size);
+ goto out;
+ }
+
+ if (compress != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) {
+ cabd = abd_alloc_for_io(asize, ismd);
+ tmp = abd_borrow_buf(cabd, asize);
+
+ psize = zio_compress_data(compress, to_write, tmp, size);
+ ASSERT3U(psize, <=, HDR_GET_PSIZE(hdr));
+ if (psize < asize)
+ bzero((char *)tmp + psize, asize - psize);
+ psize = HDR_GET_PSIZE(hdr);
+ abd_return_buf_copy(cabd, tmp, asize);
+ to_write = cabd;
+ }
+
+ if (HDR_ENCRYPTED(hdr)) {
+ eabd = abd_alloc_for_io(asize, ismd);
+
+ /*
+ * If the dataset was disowned before the buffer
+ * made it to this point, the key to re-encrypt
+ * it won't be available. In this case we simply
+ * won't write the buffer to the L2ARC.
+ */
+ ret = spa_keystore_lookup_key(spa, hdr->b_crypt_hdr.b_dsobj,
+ FTAG, &dck);
+ if (ret != 0)
+ goto error;
+
+ ret = zio_do_crypt_abd(B_TRUE, &dck->dck_key,
+ hdr->b_crypt_hdr.b_ot, bswap, hdr->b_crypt_hdr.b_salt,
+ hdr->b_crypt_hdr.b_iv, mac, psize, to_write, eabd,
+ &no_crypt);
+ if (ret != 0)
+ goto error;
+
+ if (no_crypt)
+ abd_copy(eabd, to_write, psize);
+
+ if (psize != asize)
+ abd_zero_off(eabd, psize, asize - psize);
+
+ /* assert that the MAC we got here matches the one we saved */
+ ASSERT0(bcmp(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN));
+ spa_keystore_dsl_key_rele(spa, dck, FTAG);
+
+ if (to_write == cabd)
+ abd_free(cabd);
+
+ to_write = eabd;
+ }
+
+out:
+ ASSERT3P(to_write, !=, hdr->b_l1hdr.b_pabd);
+ *abd_out = to_write;
+ return (0);
+
+error:
+ if (dck != NULL)
+ spa_keystore_dsl_key_rele(spa, dck, FTAG);
+ if (cabd != NULL)
+ abd_free(cabd);
+ if (eabd != NULL)
+ abd_free(eabd);
+
+ *abd_out = NULL;
+ return (ret);
+}
+
/*
* Find and write ARC buffers to the L2ARC device.
*
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_sz, headroom;
+ 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_sz = write_asize = write_psize = 0;
+ 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);
/*
* Copy buffers for L2ARC writing.
*/
- for (try = 0; try < L2ARC_FEED_TYPES; try++) {
+ for (int try = 0; try < L2ARC_FEED_TYPES; try++) {
multilist_sublist_t *mls = l2arc_sublist_lock(try);
uint64_t passed_sz = 0;
for (; hdr; hdr = hdr_prev) {
kmutex_t *hash_lock;
- uint64_t asize, size;
- abd_t *to_write;
+ abd_t *to_write = NULL;
if (arc_warm == B_FALSE)
hdr_prev = multilist_sublist_next(mls, hdr);
continue;
}
- if ((write_asize + HDR_GET_LSIZE(hdr)) > target_sz) {
+ /*
+ * 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);
+ ASSERT3U(arc_hdr_size(hdr), >, 0);
+ ASSERT(hdr->b_l1hdr.b_pabd != NULL ||
+ HDR_HAS_RABD(hdr));
+ uint64_t psize = HDR_GET_PSIZE(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);
break;
}
+ /*
+ * 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.
+ */
+ arc_hdr_set_flags(hdr, ARC_FLAG_L2_WRITING);
+ ASSERT(HDR_HAS_L1HDR(hdr));
+
+ ASSERT3U(HDR_GET_PSIZE(hdr), >, 0);
+ ASSERT(hdr->b_l1hdr.b_pabd != NULL ||
+ HDR_HAS_RABD(hdr));
+ ASSERT3U(arc_hdr_size(hdr), >, 0);
+
+ /*
+ * If this header has b_rabd, we can use this since it
+ * must always match the data exactly as it exists on
+ * disk. Otherwise, the L2ARC can normally 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. To ensure that this 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. If we need to apply any transforms to the
+ * data (compression, encryption) we will also need the
+ * extra buffer.
+ */
+ if (HDR_HAS_RABD(hdr) && psize == asize) {
+ to_write = hdr->b_crypt_hdr.b_rabd;
+ } else if ((HDR_COMPRESSION_ENABLED(hdr) ||
+ HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_OFF) &&
+ !HDR_ENCRYPTED(hdr) && !HDR_SHARED_DATA(hdr) &&
+ psize == asize) {
+ to_write = hdr->b_l1hdr.b_pabd;
+ } else {
+ int ret;
+ arc_buf_contents_t type = arc_buf_type(hdr);
+
+ ret = l2arc_apply_transforms(spa, hdr, asize,
+ &to_write);
+ if (ret != 0) {
+ arc_hdr_clear_flags(hdr,
+ ARC_FLAG_L2_WRITING);
+ mutex_exit(hash_lock);
+ continue;
+ }
+
+ l2arc_free_abd_on_write(to_write, asize, type);
+ }
+
if (pio == NULL) {
/*
* Insert a dummy header on the buflist so
hdr->b_l2hdr.b_hits = 0;
hdr->b_l2hdr.b_daddr = dev->l2ad_hand;
- arc_hdr_set_flags(hdr,
- ARC_FLAG_L2_WRITING | ARC_FLAG_HAS_L2HDR);
+ arc_hdr_set_flags(hdr, ARC_FLAG_HAS_L2HDR);
mutex_enter(&dev->l2ad_mtx);
list_insert_head(&dev->l2ad_buflist, hdr);
mutex_exit(&dev->l2ad_mtx);
- /*
- * 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);
- size = arc_hdr_size(hdr);
-
- (void) refcount_add_many(&dev->l2ad_alloc, size, hdr);
+ (void) zfs_refcount_add_many(&dev->l2ad_alloc,
+ arc_hdr_size(hdr), 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. To
- * ensure that this 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.
- */
- if (!HDR_SHARED_DATA(hdr)) {
- to_write = hdr->b_l1hdr.b_pabd;
- } else {
- to_write = abd_alloc_for_io(size,
- HDR_ISTYPE_METADATA(hdr));
- abd_copy(to_write, hdr->b_l1hdr.b_pabd, size);
- l2arc_free_abd_on_write(to_write, size,
- arc_buf_type(hdr));
- }
wzio = zio_write_phys(pio, dev->l2ad_vdev,
- hdr->b_l2hdr.b_daddr, size, to_write,
+ hdr->b_l2hdr.b_daddr, asize, to_write,
ZIO_CHECKSUM_OFF, NULL, hdr,
ZIO_PRIORITY_ASYNC_WRITE,
ZIO_FLAG_CANFAIL, B_FALSE);
- write_sz += HDR_GET_LSIZE(hdr);
+ write_lsize += HDR_GET_LSIZE(hdr);
DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev,
zio_t *, wzio);
- write_asize += size;
- /*
- * Keep the clock hand suitably device-aligned.
- */
- asize = vdev_psize_to_asize(dev->l2ad_vdev, size);
- write_psize += asize;
+ write_psize += psize;
+ write_asize += asize;
dev->l2ad_hand += asize;
+ vdev_space_update(dev->l2ad_vdev, asize, 0, 0);
mutex_exit(hash_lock);
/* No buffers selected for writing? */
if (pio == NULL) {
- ASSERT0(write_sz);
+ ASSERT0(write_lsize);
ASSERT(!HDR_HAS_L1HDR(head));
kmem_cache_free(hdr_l2only_cache, head);
return (0);
ASSERT3U(write_asize, <=, target_sz);
ARCSTAT_BUMP(arcstat_l2_writes_sent);
- ARCSTAT_INCR(arcstat_l2_write_bytes, write_asize);
- ARCSTAT_INCR(arcstat_l2_size, write_sz);
- ARCSTAT_INCR(arcstat_l2_asize, write_asize);
- vdev_space_update(dev->l2ad_vdev, write_asize, 0, 0);
+ ARCSTAT_INCR(arcstat_l2_write_bytes, write_psize);
+ ARCSTAT_INCR(arcstat_l2_lsize, write_lsize);
+ ARCSTAT_INCR(arcstat_l2_psize, write_psize);
/*
* Bump device hand to the device start if it is approaching the end.
* This thread feeds the L2ARC at regular intervals. This is the beating
* heart of the L2ARC.
*/
+/* ARGSUSED */
static void
-l2arc_feed_thread(void)
+l2arc_feed_thread(void *unused)
{
callb_cpr_t cpr;
l2arc_dev_t *dev;
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);
+ zfs_refcount_create(&adddev->l2ad_alloc);
/*
* Add device to global list
l2arc_evict(remdev, 0, B_TRUE);
list_destroy(&remdev->l2ad_buflist);
mutex_destroy(&remdev->l2ad_mtx);
- refcount_destroy(&remdev->l2ad_alloc);
+ zfs_refcount_destroy(&remdev->l2ad_alloc);
kmem_free(remdev, sizeof (l2arc_dev_t));
}
mutex_exit(&l2arc_feed_thr_lock);
}
-#if defined(_KERNEL) && defined(HAVE_SPL)
+#if defined(_KERNEL)
EXPORT_SYMBOL(arc_buf_size);
EXPORT_SYMBOL(arc_write);
EXPORT_SYMBOL(arc_read);
module_param(zfs_arc_grow_retry, int, 0644);
MODULE_PARM_DESC(zfs_arc_grow_retry, "Seconds before growing arc size");
-module_param(zfs_arc_p_aggressive_disable, int, 0644);
-MODULE_PARM_DESC(zfs_arc_p_aggressive_disable, "disable aggressive arc_p grow");
-
module_param(zfs_arc_p_dampener_disable, int, 0644);
MODULE_PARM_DESC(zfs_arc_p_dampener_disable, "disable arc_p adapt dampener");
module_param(zfs_arc_shrink_shift, int, 0644);
MODULE_PARM_DESC(zfs_arc_shrink_shift, "log2(fraction of arc to reclaim)");
+module_param(zfs_arc_pc_percent, uint, 0644);
+MODULE_PARM_DESC(zfs_arc_pc_percent,
+ "Percent of pagecache to reclaim arc to");
+
module_param(zfs_arc_p_min_shift, int, 0644);
MODULE_PARM_DESC(zfs_arc_p_min_shift, "arc_c shift to calc min/max arc_p");
module_param(zfs_compressed_arc_enabled, int, 0644);
MODULE_PARM_DESC(zfs_compressed_arc_enabled, "Disable compressed arc buffers");
-module_param(zfs_arc_min_prefetch_lifespan, int, 0644);
-MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan, "Min life of prefetch block");
+module_param(zfs_arc_min_prefetch_ms, int, 0644);
+MODULE_PARM_DESC(zfs_arc_min_prefetch_ms, "Min life of prefetch block in ms");
+
+module_param(zfs_arc_min_prescient_prefetch_ms, int, 0644);
+MODULE_PARM_DESC(zfs_arc_min_prescient_prefetch_ms,
+ "Min life of prescient prefetched block in ms");
module_param(l2arc_write_max, ulong, 0644);
MODULE_PARM_DESC(l2arc_write_max, "Max write bytes per interval");