4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013, Delphix. All rights reserved.
24 * Copyright (c) 2013, Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2020, George Amanakis. All rights reserved.
29 #ifndef _SYS_ARC_IMPL_H
30 #define _SYS_ARC_IMPL_H
33 #include <sys/multilist.h>
34 #include <sys/zio_crypt.h>
36 #include <sys/aggsum.h>
37 #include <sys/wmsum.h>
44 * Note that buffers can be in one of 6 states:
45 * ARC_anon - anonymous (discussed below)
46 * ARC_mru - recently used, currently cached
47 * ARC_mru_ghost - recently used, no longer in cache
48 * ARC_mfu - frequently used, currently cached
49 * ARC_mfu_ghost - frequently used, no longer in cache
50 * ARC_uncached - uncacheable prefetch, to be evicted
51 * ARC_l2c_only - exists in L2ARC but not other states
52 * When there are no active references to the buffer, they are
53 * are linked onto a list in one of these arc states. These are
54 * the only buffers that can be evicted or deleted. Within each
55 * state there are multiple lists, one for meta-data and one for
56 * non-meta-data. Meta-data (indirect blocks, blocks of dnodes,
57 * etc.) is tracked separately so that it can be managed more
58 * explicitly: favored over data, limited explicitly.
60 * Anonymous buffers are buffers that are not associated with
61 * a DVA. These are buffers that hold dirty block copies
62 * before they are written to stable storage. By definition,
63 * they are "ref'd" and are considered part of arc_mru
64 * that cannot be freed. Generally, they will acquire a DVA
65 * as they are written and migrate onto the arc_mru list.
67 * The ARC_l2c_only state is for buffers that are in the second
68 * level ARC but no longer in any of the ARC_m* lists. The second
69 * level ARC itself may also contain buffers that are in any of
70 * the ARC_m* states - meaning that a buffer can exist in two
71 * places. The reason for the ARC_l2c_only state is to keep the
72 * buffer header in the hash table, so that reads that hit the
73 * second level ARC benefit from these fast lookups.
76 typedef struct arc_state
{
78 * list of evictable buffers
80 multilist_t arcs_list
[ARC_BUFC_NUMTYPES
];
82 * supports the "dbufs" kstat
84 arc_state_type_t arcs_state
;
86 * total amount of data in this state.
88 zfs_refcount_t arcs_size
[ARC_BUFC_NUMTYPES
] ____cacheline_aligned
;
90 * total amount of evictable data in this state
92 zfs_refcount_t arcs_esize
[ARC_BUFC_NUMTYPES
];
94 * amount of hit bytes for this state (counted only for ghost states)
96 wmsum_t arcs_hits
[ARC_BUFC_NUMTYPES
];
99 typedef struct arc_callback arc_callback_t
;
101 struct arc_callback
{
103 arc_read_done_func_t
*acb_done
;
105 boolean_t acb_encrypted
;
106 boolean_t acb_compressed
;
107 boolean_t acb_noauth
;
111 kmutex_t acb_wait_lock
;
112 kcondvar_t acb_wait_cv
;
113 zbookmark_phys_t acb_zb
;
114 zio_t
*acb_zio_dummy
;
116 arc_callback_t
*acb_prev
;
117 arc_callback_t
*acb_next
;
120 typedef struct arc_write_callback arc_write_callback_t
;
122 struct arc_write_callback
{
124 arc_write_done_func_t
*awcb_ready
;
125 arc_write_done_func_t
*awcb_children_ready
;
126 arc_write_done_func_t
*awcb_done
;
131 * ARC buffers are separated into multiple structs as a memory saving measure:
132 * - Common fields struct, always defined, and embedded within it:
133 * - L2-only fields, always allocated but undefined when not in L2ARC
134 * - L1-only fields, only allocated when in L1ARC
136 * Buffer in L1 Buffer only in L2
137 * +------------------------+ +------------------------+
138 * | arc_buf_hdr_t | | arc_buf_hdr_t |
142 * +------------------------+ +------------------------+
143 * | l2arc_buf_hdr_t | | l2arc_buf_hdr_t |
144 * | (undefined if L1-only) | | |
145 * +------------------------+ +------------------------+
146 * | l1arc_buf_hdr_t |
151 * +------------------------+
153 * Because it's possible for the L2ARC to become extremely large, we can wind
154 * up eating a lot of memory in L2ARC buffer headers, so the size of a header
155 * is minimized by only allocating the fields necessary for an L1-cached buffer
156 * when a header is actually in the L1 cache. The sub-headers (l1arc_buf_hdr and
157 * l2arc_buf_hdr) are embedded rather than allocated separately to save a couple
158 * words in pointers. arc_hdr_realloc() is used to switch a header between
159 * these two allocation states.
161 typedef struct l1arc_buf_hdr
{
164 /* protected by arc state mutex */
165 arc_state_t
*b_state
;
166 multilist_node_t b_arc_node
;
168 /* protected by hash lock */
169 clock_t b_arc_access
;
171 uint32_t b_mru_ghost_hits
;
173 uint32_t b_mfu_ghost_hits
;
177 /* self protecting */
178 zfs_refcount_t b_refcnt
;
180 arc_callback_t
*b_acb
;
184 zio_cksum_t
*b_freeze_cksum
;
185 kmutex_t b_freeze_lock
;
189 typedef enum l2arc_dev_hdr_flags_t
{
190 L2ARC_DEV_HDR_EVICT_FIRST
= (1 << 0) /* mirror of l2ad_first */
191 } l2arc_dev_hdr_flags_t
;
194 * Pointer used in persistent L2ARC (for pointing to log blocks).
196 typedef struct l2arc_log_blkptr
{
198 * Offset of log block within the device, in bytes
202 * Aligned payload size (in bytes) of the log block
204 uint64_t lbp_payload_asize
;
206 * Offset in bytes of the first buffer in the payload
208 uint64_t lbp_payload_start
;
210 * lbp_prop has the following format:
211 * * logical size (in bytes)
212 * * aligned (after compression) size (in bytes)
213 * * compression algorithm (we always LZ4-compress l2arc logs)
214 * * checksum algorithm (used for lbp_cksum)
217 zio_cksum_t lbp_cksum
; /* checksum of log */
218 } l2arc_log_blkptr_t
;
221 * The persistent L2ARC device header.
222 * Byte order of magic determines whether 64-bit bswap of fields is necessary.
224 typedef struct l2arc_dev_hdr_phys
{
225 uint64_t dh_magic
; /* L2ARC_DEV_HDR_MAGIC */
226 uint64_t dh_version
; /* Persistent L2ARC version */
229 * Global L2ARC device state and metadata.
231 uint64_t dh_spa_guid
;
232 uint64_t dh_vdev_guid
;
233 uint64_t dh_log_entries
; /* mirror of l2ad_log_entries */
234 uint64_t dh_evict
; /* evicted offset in bytes */
235 uint64_t dh_flags
; /* l2arc_dev_hdr_flags_t */
237 * Used in zdb.c for determining if a log block is valid, in the same
238 * way that l2arc_rebuild() does.
240 uint64_t dh_start
; /* mirror of l2ad_start */
241 uint64_t dh_end
; /* mirror of l2ad_end */
243 * Start of log block chain. [0] -> newest log, [1] -> one older (used
244 * for initiating prefetch).
246 l2arc_log_blkptr_t dh_start_lbps
[2];
248 * Aligned size of all log blocks as accounted by vdev_space_update().
250 uint64_t dh_lb_asize
; /* mirror of l2ad_lb_asize */
251 uint64_t dh_lb_count
; /* mirror of l2ad_lb_count */
253 * Mirrors of vdev_trim_action_time and vdev_trim_state, used to
254 * display when the cache device was fully trimmed for the last
257 uint64_t dh_trim_action_time
;
258 uint64_t dh_trim_state
;
259 const uint64_t dh_pad
[30]; /* pad to 512 bytes */
261 } l2arc_dev_hdr_phys_t
;
262 _Static_assert(sizeof (l2arc_dev_hdr_phys_t
) == SPA_MINBLOCKSIZE
,
263 "l2arc_dev_hdr_phys_t wrong size");
266 * A single ARC buffer header entry in a l2arc_log_blk_phys_t.
268 typedef struct l2arc_log_ent_phys
{
269 dva_t le_dva
; /* dva of buffer */
270 uint64_t le_birth
; /* birth txg of buffer */
272 * le_prop has the following format:
273 * * logical size (in bytes)
274 * * physical (compressed) size (in bytes)
275 * * compression algorithm
276 * * object type (used to restore arc_buf_contents_t)
277 * * protected status (used for encryption)
278 * * prefetch status (used in l2arc_read_done())
281 uint64_t le_daddr
; /* buf location on l2dev */
282 uint64_t le_complevel
;
284 * We pad the size of each entry to a power of 2 so that the size of
285 * l2arc_log_blk_phys_t is power-of-2 aligned with SPA_MINBLOCKSHIFT,
286 * because of the L2ARC_SET_*SIZE macros.
288 const uint64_t le_pad
[2]; /* pad to 64 bytes */
289 } l2arc_log_ent_phys_t
;
291 #define L2ARC_LOG_BLK_MAX_ENTRIES (1022)
294 * A log block of up to 1022 ARC buffer log entries, chained into the
295 * persistent L2ARC metadata linked list. Byte order of magic determines
296 * whether 64-bit bswap of fields is necessary.
298 typedef struct l2arc_log_blk_phys
{
299 uint64_t lb_magic
; /* L2ARC_LOG_BLK_MAGIC */
301 * There are 2 chains (headed by dh_start_lbps[2]), and this field
302 * points back to the previous block in this chain. We alternate
303 * which chain we append to, so they are time-wise and offset-wise
304 * interleaved, but that is an optimization rather than for
307 l2arc_log_blkptr_t lb_prev_lbp
; /* pointer to prev log block */
309 * Pad header section to 128 bytes
313 l2arc_log_ent_phys_t lb_entries
[L2ARC_LOG_BLK_MAX_ENTRIES
];
314 } l2arc_log_blk_phys_t
; /* 64K total */
317 * The size of l2arc_log_blk_phys_t has to be power-of-2 aligned with
318 * SPA_MINBLOCKSHIFT because of L2BLK_SET_*SIZE macros.
320 _Static_assert(IS_P2ALIGNED(sizeof (l2arc_log_blk_phys_t
),
321 1ULL << SPA_MINBLOCKSHIFT
), "l2arc_log_blk_phys_t misaligned");
322 _Static_assert(sizeof (l2arc_log_blk_phys_t
) >= SPA_MINBLOCKSIZE
,
323 "l2arc_log_blk_phys_t too small");
324 _Static_assert(sizeof (l2arc_log_blk_phys_t
) <= SPA_MAXBLOCKSIZE
,
325 "l2arc_log_blk_phys_t too big");
328 * These structures hold in-flight abd buffers for log blocks as they're being
329 * written to the L2ARC device.
331 typedef struct l2arc_lb_abd_buf
{
334 } l2arc_lb_abd_buf_t
;
337 * These structures hold pointers to log blocks present on the L2ARC device.
339 typedef struct l2arc_lb_ptr_buf
{
340 l2arc_log_blkptr_t
*lb_ptr
;
342 } l2arc_lb_ptr_buf_t
;
344 /* Macros for setting fields in le_prop and lbp_prop */
345 #define L2BLK_GET_LSIZE(field) \
346 BF64_GET_SB((field), 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1)
347 #define L2BLK_SET_LSIZE(field, x) \
348 BF64_SET_SB((field), 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x)
349 #define L2BLK_GET_PSIZE(field) \
350 BF64_GET_SB((field), 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1)
351 #define L2BLK_SET_PSIZE(field, x) \
352 BF64_SET_SB((field), 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x)
353 #define L2BLK_GET_COMPRESS(field) \
354 BF64_GET((field), 32, SPA_COMPRESSBITS)
355 #define L2BLK_SET_COMPRESS(field, x) \
356 BF64_SET((field), 32, SPA_COMPRESSBITS, x)
357 #define L2BLK_GET_PREFETCH(field) BF64_GET((field), 39, 1)
358 #define L2BLK_SET_PREFETCH(field, x) BF64_SET((field), 39, 1, x)
359 #define L2BLK_GET_CHECKSUM(field) BF64_GET((field), 40, 8)
360 #define L2BLK_SET_CHECKSUM(field, x) BF64_SET((field), 40, 8, x)
361 /* +/- 1 here are to keep compatibility after ARC_BUFC_INVALID removal. */
362 #define L2BLK_GET_TYPE(field) (BF64_GET((field), 48, 8) - 1)
363 #define L2BLK_SET_TYPE(field, x) BF64_SET((field), 48, 8, (x) + 1)
364 #define L2BLK_GET_PROTECTED(field) BF64_GET((field), 56, 1)
365 #define L2BLK_SET_PROTECTED(field, x) BF64_SET((field), 56, 1, x)
366 #define L2BLK_GET_STATE(field) BF64_GET((field), 57, 4)
367 #define L2BLK_SET_STATE(field, x) BF64_SET((field), 57, 4, x)
369 #define PTR_SWAP(x, y) \
376 #define L2ARC_DEV_HDR_MAGIC 0x5a46534341434845LLU /* ASCII: "ZFSCACHE" */
377 #define L2ARC_LOG_BLK_MAGIC 0x4c4f47424c4b4844LLU /* ASCII: "LOGBLKHD" */
382 typedef struct l2arc_dev
{
383 vdev_t
*l2ad_vdev
; /* vdev */
384 spa_t
*l2ad_spa
; /* spa */
385 uint64_t l2ad_hand
; /* next write location */
386 uint64_t l2ad_start
; /* first addr on device */
387 uint64_t l2ad_end
; /* last addr on device */
388 boolean_t l2ad_first
; /* first sweep through */
389 boolean_t l2ad_writing
; /* currently writing */
390 kmutex_t l2ad_mtx
; /* lock for buffer list */
391 list_t l2ad_buflist
; /* buffer list */
392 list_node_t l2ad_node
; /* device list node */
393 zfs_refcount_t l2ad_alloc
; /* allocated bytes */
395 * Persistence-related stuff
397 l2arc_dev_hdr_phys_t
*l2ad_dev_hdr
; /* persistent device header */
398 uint64_t l2ad_dev_hdr_asize
; /* aligned hdr size */
399 l2arc_log_blk_phys_t l2ad_log_blk
; /* currently open log block */
400 int l2ad_log_ent_idx
; /* index into cur log blk */
401 /* Number of bytes in current log block's payload */
402 uint64_t l2ad_log_blk_payload_asize
;
404 * Offset (in bytes) of the first buffer in current log block's
407 uint64_t l2ad_log_blk_payload_start
;
408 /* Flag indicating whether a rebuild is scheduled or is going on */
409 boolean_t l2ad_rebuild
;
410 boolean_t l2ad_rebuild_cancel
;
411 boolean_t l2ad_rebuild_began
;
412 uint64_t l2ad_log_entries
; /* entries per log blk */
413 uint64_t l2ad_evict
; /* evicted offset in bytes */
414 /* List of pointers to log blocks present in the L2ARC device */
415 list_t l2ad_lbptr_list
;
417 * Aligned size of all log blocks as accounted by vdev_space_update().
419 zfs_refcount_t l2ad_lb_asize
;
421 * Number of log blocks present on the device.
423 zfs_refcount_t l2ad_lb_count
;
424 boolean_t l2ad_trim_all
; /* TRIM whole device */
428 * Encrypted blocks will need to be stored encrypted on the L2ARC
429 * disk as they appear in the main pool. In order for this to work we
430 * need to pass around the encryption parameters so they can be used
431 * to write data to the L2ARC. This struct is only defined in the
432 * arc_buf_hdr_t if the L1 header is defined and has the ARC_FLAG_ENCRYPTED
435 typedef struct arc_buf_hdr_crypt
{
436 abd_t
*b_rabd
; /* raw encrypted data */
437 dmu_object_type_t b_ot
; /* object type */
438 uint32_t b_ebufcnt
; /* count of encrypted buffers */
440 /* dsobj for looking up encryption key for l2arc encryption */
443 /* encryption parameters */
444 uint8_t b_salt
[ZIO_DATA_SALT_LEN
];
445 uint8_t b_iv
[ZIO_DATA_IV_LEN
];
448 * Technically this could be removed since we will always be able to
449 * get the mac from the bp when we need it. However, it is inconvenient
450 * for callers of arc code to have to pass a bp in all the time. This
451 * also allows us to assert that L2ARC data is properly encrypted to
452 * match the data in the main storage pool.
454 uint8_t b_mac
[ZIO_DATA_MAC_LEN
];
455 } arc_buf_hdr_crypt_t
;
457 typedef struct l2arc_buf_hdr
{
458 /* protected by arc_buf_hdr mutex */
459 l2arc_dev_t
*b_dev
; /* L2ARC device */
460 uint64_t b_daddr
; /* disk address, offset byte */
462 arc_state_type_t b_arcs_state
;
463 list_node_t b_l2node
;
466 typedef struct l2arc_write_callback
{
467 l2arc_dev_t
*l2wcb_dev
; /* device info */
468 arc_buf_hdr_t
*l2wcb_head
; /* head of write buflist */
469 /* in-flight list of log blocks */
470 list_t l2wcb_abd_list
;
471 } l2arc_write_callback_t
;
474 /* protected by hash lock */
478 arc_buf_contents_t b_type
;
480 uint8_t b_reserved1
; /* used for 4 byte alignment */
481 uint16_t b_reserved2
; /* used for 4 byte alignment */
482 arc_buf_hdr_t
*b_hash_next
;
486 * This field stores the size of the data buffer after
487 * compression, and is set in the arc's zio completion handlers.
488 * It is in units of SPA_MINBLOCKSIZE (e.g. 1 == 512 bytes).
490 * While the block pointers can store up to 32MB in their psize
491 * field, we can only store up to 32MB minus 512B. This is due
492 * to the bp using a bias of 1, whereas we use a bias of 0 (i.e.
493 * a field of zeros represents 512B in the bp). We can't use a
494 * bias of 1 since we need to reserve a psize of zero, here, to
495 * represent holes and embedded blocks.
497 * This isn't a problem in practice, since the maximum size of a
498 * buffer is limited to 16MB, so we never need to store 32MB in
499 * this field. Even in the upstream illumos code base, the
500 * maximum size of a buffer is limited to 16MB.
505 * This field stores the size of the data buffer before
506 * compression, and cannot change once set. It is in units
507 * of SPA_MINBLOCKSIZE (e.g. 2 == 1024 bytes)
509 uint16_t b_lsize
; /* immutable */
510 uint64_t b_spa
; /* immutable */
512 /* L2ARC fields. Undefined when not in L2ARC. */
513 l2arc_buf_hdr_t b_l2hdr
;
514 /* L1ARC fields. Undefined when in l2arc_only state */
515 l1arc_buf_hdr_t b_l1hdr
;
517 * Encryption parameters. Defined only when ARC_FLAG_ENCRYPTED
518 * is set and the L1 header exists.
520 arc_buf_hdr_crypt_t b_crypt_hdr
;
523 typedef struct arc_stats
{
524 /* Number of requests that were satisfied without I/O. */
525 kstat_named_t arcstat_hits
;
526 /* Number of requests for which I/O was already running. */
527 kstat_named_t arcstat_iohits
;
528 /* Number of requests for which I/O has to be issued. */
529 kstat_named_t arcstat_misses
;
530 /* Same three, but specifically for demand data. */
531 kstat_named_t arcstat_demand_data_hits
;
532 kstat_named_t arcstat_demand_data_iohits
;
533 kstat_named_t arcstat_demand_data_misses
;
534 /* Same three, but specifically for demand metadata. */
535 kstat_named_t arcstat_demand_metadata_hits
;
536 kstat_named_t arcstat_demand_metadata_iohits
;
537 kstat_named_t arcstat_demand_metadata_misses
;
538 /* Same three, but specifically for prefetch data. */
539 kstat_named_t arcstat_prefetch_data_hits
;
540 kstat_named_t arcstat_prefetch_data_iohits
;
541 kstat_named_t arcstat_prefetch_data_misses
;
542 /* Same three, but specifically for prefetch metadata. */
543 kstat_named_t arcstat_prefetch_metadata_hits
;
544 kstat_named_t arcstat_prefetch_metadata_iohits
;
545 kstat_named_t arcstat_prefetch_metadata_misses
;
546 kstat_named_t arcstat_mru_hits
;
547 kstat_named_t arcstat_mru_ghost_hits
;
548 kstat_named_t arcstat_mfu_hits
;
549 kstat_named_t arcstat_mfu_ghost_hits
;
550 kstat_named_t arcstat_uncached_hits
;
551 kstat_named_t arcstat_deleted
;
553 * Number of buffers that could not be evicted because the hash lock
554 * was held by another thread. The lock may not necessarily be held
555 * by something using the same buffer, since hash locks are shared
556 * by multiple buffers.
558 kstat_named_t arcstat_mutex_miss
;
560 * Number of buffers skipped when updating the access state due to the
561 * header having already been released after acquiring the hash lock.
563 kstat_named_t arcstat_access_skip
;
565 * Number of buffers skipped because they have I/O in progress, are
566 * indirect prefetch buffers that have not lived long enough, or are
567 * not from the spa we're trying to evict from.
569 kstat_named_t arcstat_evict_skip
;
571 * Number of times arc_evict_state() was unable to evict enough
572 * buffers to reach its target amount.
574 kstat_named_t arcstat_evict_not_enough
;
575 kstat_named_t arcstat_evict_l2_cached
;
576 kstat_named_t arcstat_evict_l2_eligible
;
577 kstat_named_t arcstat_evict_l2_eligible_mfu
;
578 kstat_named_t arcstat_evict_l2_eligible_mru
;
579 kstat_named_t arcstat_evict_l2_ineligible
;
580 kstat_named_t arcstat_evict_l2_skip
;
581 kstat_named_t arcstat_hash_elements
;
582 kstat_named_t arcstat_hash_elements_max
;
583 kstat_named_t arcstat_hash_collisions
;
584 kstat_named_t arcstat_hash_chains
;
585 kstat_named_t arcstat_hash_chain_max
;
586 kstat_named_t arcstat_meta
;
587 kstat_named_t arcstat_pd
;
588 kstat_named_t arcstat_pm
;
589 kstat_named_t arcstat_c
;
590 kstat_named_t arcstat_c_min
;
591 kstat_named_t arcstat_c_max
;
592 kstat_named_t arcstat_size
;
594 * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd.
595 * Note that the compressed bytes may match the uncompressed bytes
596 * if the block is either not compressed or compressed arc is disabled.
598 kstat_named_t arcstat_compressed_size
;
600 * Uncompressed size of the data stored in b_pabd. If compressed
601 * arc is disabled then this value will be identical to the stat
604 kstat_named_t arcstat_uncompressed_size
;
606 * Number of bytes stored in all the arc_buf_t's. This is classified
607 * as "overhead" since this data is typically short-lived and will
608 * be evicted from the arc when it becomes unreferenced unless the
609 * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level
610 * values have been set (see comment in dbuf.c for more information).
612 kstat_named_t arcstat_overhead_size
;
614 * Number of bytes consumed by internal ARC structures necessary
615 * for tracking purposes; these structures are not actually
616 * backed by ARC buffers. This includes arc_buf_hdr_t structures
617 * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only
618 * caches), and arc_buf_t structures (allocated via arc_buf_t
621 kstat_named_t arcstat_hdr_size
;
623 * Number of bytes consumed by ARC buffers of type equal to
624 * ARC_BUFC_DATA. This is generally consumed by buffers backing
625 * on disk user data (e.g. plain file contents).
627 kstat_named_t arcstat_data_size
;
629 * Number of bytes consumed by ARC buffers of type equal to
630 * ARC_BUFC_METADATA. This is generally consumed by buffers
631 * backing on disk data that is used for internal ZFS
632 * structures (e.g. ZAP, dnode, indirect blocks, etc).
634 kstat_named_t arcstat_metadata_size
;
636 * Number of bytes consumed by dmu_buf_impl_t objects.
638 kstat_named_t arcstat_dbuf_size
;
640 * Number of bytes consumed by dnode_t objects.
642 kstat_named_t arcstat_dnode_size
;
644 * Number of bytes consumed by bonus buffers.
646 kstat_named_t arcstat_bonus_size
;
647 #if defined(COMPAT_FREEBSD11)
649 * Sum of the previous three counters, provided for compatibility.
651 kstat_named_t arcstat_other_size
;
655 * Total number of bytes consumed by ARC buffers residing in the
656 * arc_anon state. This includes *all* buffers in the arc_anon
657 * state; e.g. data, metadata, evictable, and unevictable buffers
658 * are all included in this value.
660 kstat_named_t arcstat_anon_size
;
661 kstat_named_t arcstat_anon_data
;
662 kstat_named_t arcstat_anon_metadata
;
664 * Number of bytes consumed by ARC buffers that meet the
665 * following criteria: backing buffers of type ARC_BUFC_DATA,
666 * residing in the arc_anon state, and are eligible for eviction
667 * (e.g. have no outstanding holds on the buffer).
669 kstat_named_t arcstat_anon_evictable_data
;
671 * Number of bytes consumed by ARC buffers that meet the
672 * following criteria: backing buffers of type ARC_BUFC_METADATA,
673 * residing in the arc_anon state, and are eligible for eviction
674 * (e.g. have no outstanding holds on the buffer).
676 kstat_named_t arcstat_anon_evictable_metadata
;
678 * Total number of bytes consumed by ARC buffers residing in the
679 * arc_mru state. This includes *all* buffers in the arc_mru
680 * state; e.g. data, metadata, evictable, and unevictable buffers
681 * are all included in this value.
683 kstat_named_t arcstat_mru_size
;
684 kstat_named_t arcstat_mru_data
;
685 kstat_named_t arcstat_mru_metadata
;
687 * Number of bytes consumed by ARC buffers that meet the
688 * following criteria: backing buffers of type ARC_BUFC_DATA,
689 * residing in the arc_mru state, and are eligible for eviction
690 * (e.g. have no outstanding holds on the buffer).
692 kstat_named_t arcstat_mru_evictable_data
;
694 * Number of bytes consumed by ARC buffers that meet the
695 * following criteria: backing buffers of type ARC_BUFC_METADATA,
696 * residing in the arc_mru state, and are eligible for eviction
697 * (e.g. have no outstanding holds on the buffer).
699 kstat_named_t arcstat_mru_evictable_metadata
;
701 * Total number of bytes that *would have been* consumed by ARC
702 * buffers in the arc_mru_ghost state. The key thing to note
703 * here, is the fact that this size doesn't actually indicate
704 * RAM consumption. The ghost lists only consist of headers and
705 * don't actually have ARC buffers linked off of these headers.
706 * Thus, *if* the headers had associated ARC buffers, these
707 * buffers *would have* consumed this number of bytes.
709 kstat_named_t arcstat_mru_ghost_size
;
710 kstat_named_t arcstat_mru_ghost_data
;
711 kstat_named_t arcstat_mru_ghost_metadata
;
713 * Number of bytes that *would have been* consumed by ARC
714 * buffers that are eligible for eviction, of type
715 * ARC_BUFC_DATA, and linked off the arc_mru_ghost state.
717 kstat_named_t arcstat_mru_ghost_evictable_data
;
719 * Number of bytes that *would have been* consumed by ARC
720 * buffers that are eligible for eviction, of type
721 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
723 kstat_named_t arcstat_mru_ghost_evictable_metadata
;
725 * Total number of bytes consumed by ARC buffers residing in the
726 * arc_mfu state. This includes *all* buffers in the arc_mfu
727 * state; e.g. data, metadata, evictable, and unevictable buffers
728 * are all included in this value.
730 kstat_named_t arcstat_mfu_size
;
731 kstat_named_t arcstat_mfu_data
;
732 kstat_named_t arcstat_mfu_metadata
;
734 * Number of bytes consumed by ARC buffers that are eligible for
735 * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu
738 kstat_named_t arcstat_mfu_evictable_data
;
740 * Number of bytes consumed by ARC buffers that are eligible for
741 * eviction, of type ARC_BUFC_METADATA, and reside in the
744 kstat_named_t arcstat_mfu_evictable_metadata
;
746 * Total number of bytes that *would have been* consumed by ARC
747 * buffers in the arc_mfu_ghost state. See the comment above
748 * arcstat_mru_ghost_size for more details.
750 kstat_named_t arcstat_mfu_ghost_size
;
751 kstat_named_t arcstat_mfu_ghost_data
;
752 kstat_named_t arcstat_mfu_ghost_metadata
;
754 * Number of bytes that *would have been* consumed by ARC
755 * buffers that are eligible for eviction, of type
756 * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state.
758 kstat_named_t arcstat_mfu_ghost_evictable_data
;
760 * Number of bytes that *would have been* consumed by ARC
761 * buffers that are eligible for eviction, of type
762 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
764 kstat_named_t arcstat_mfu_ghost_evictable_metadata
;
766 * Total number of bytes that are going to be evicted from ARC due to
767 * ARC_FLAG_UNCACHED being set.
769 kstat_named_t arcstat_uncached_size
;
770 kstat_named_t arcstat_uncached_data
;
771 kstat_named_t arcstat_uncached_metadata
;
773 * Number of data bytes that are going to be evicted from ARC due to
774 * ARC_FLAG_UNCACHED being set.
776 kstat_named_t arcstat_uncached_evictable_data
;
778 * Number of metadata bytes that that are going to be evicted from ARC
779 * due to ARC_FLAG_UNCACHED being set.
781 kstat_named_t arcstat_uncached_evictable_metadata
;
782 kstat_named_t arcstat_l2_hits
;
783 kstat_named_t arcstat_l2_misses
;
785 * Allocated size (in bytes) of L2ARC cached buffers by ARC state.
787 kstat_named_t arcstat_l2_prefetch_asize
;
788 kstat_named_t arcstat_l2_mru_asize
;
789 kstat_named_t arcstat_l2_mfu_asize
;
791 * Allocated size (in bytes) of L2ARC cached buffers by buffer content
794 kstat_named_t arcstat_l2_bufc_data_asize
;
795 kstat_named_t arcstat_l2_bufc_metadata_asize
;
796 kstat_named_t arcstat_l2_feeds
;
797 kstat_named_t arcstat_l2_rw_clash
;
798 kstat_named_t arcstat_l2_read_bytes
;
799 kstat_named_t arcstat_l2_write_bytes
;
800 kstat_named_t arcstat_l2_writes_sent
;
801 kstat_named_t arcstat_l2_writes_done
;
802 kstat_named_t arcstat_l2_writes_error
;
803 kstat_named_t arcstat_l2_writes_lock_retry
;
804 kstat_named_t arcstat_l2_evict_lock_retry
;
805 kstat_named_t arcstat_l2_evict_reading
;
806 kstat_named_t arcstat_l2_evict_l1cached
;
807 kstat_named_t arcstat_l2_free_on_write
;
808 kstat_named_t arcstat_l2_abort_lowmem
;
809 kstat_named_t arcstat_l2_cksum_bad
;
810 kstat_named_t arcstat_l2_io_error
;
811 kstat_named_t arcstat_l2_lsize
;
812 kstat_named_t arcstat_l2_psize
;
813 kstat_named_t arcstat_l2_hdr_size
;
815 * Number of L2ARC log blocks written. These are used for restoring the
816 * L2ARC. Updated during writing of L2ARC log blocks.
818 kstat_named_t arcstat_l2_log_blk_writes
;
820 * Moving average of the aligned size of the L2ARC log blocks, in
821 * bytes. Updated during L2ARC rebuild and during writing of L2ARC
824 kstat_named_t arcstat_l2_log_blk_avg_asize
;
825 /* Aligned size of L2ARC log blocks on L2ARC devices. */
826 kstat_named_t arcstat_l2_log_blk_asize
;
827 /* Number of L2ARC log blocks present on L2ARC devices. */
828 kstat_named_t arcstat_l2_log_blk_count
;
830 * Moving average of the aligned size of L2ARC restored data, in bytes,
831 * to the aligned size of their metadata in L2ARC, in bytes.
832 * Updated during L2ARC rebuild and during writing of L2ARC log blocks.
834 kstat_named_t arcstat_l2_data_to_meta_ratio
;
836 * Number of times the L2ARC rebuild was successful for an L2ARC device.
838 kstat_named_t arcstat_l2_rebuild_success
;
840 * Number of times the L2ARC rebuild failed because the device header
841 * was in an unsupported format or corrupted.
843 kstat_named_t arcstat_l2_rebuild_abort_unsupported
;
845 * Number of times the L2ARC rebuild failed because of IO errors
846 * while reading a log block.
848 kstat_named_t arcstat_l2_rebuild_abort_io_errors
;
850 * Number of times the L2ARC rebuild failed because of IO errors when
851 * reading the device header.
853 kstat_named_t arcstat_l2_rebuild_abort_dh_errors
;
855 * Number of L2ARC log blocks which failed to be restored due to
858 kstat_named_t arcstat_l2_rebuild_abort_cksum_lb_errors
;
860 * Number of times the L2ARC rebuild was aborted due to low system
863 kstat_named_t arcstat_l2_rebuild_abort_lowmem
;
864 /* Logical size of L2ARC restored data, in bytes. */
865 kstat_named_t arcstat_l2_rebuild_size
;
866 /* Aligned size of L2ARC restored data, in bytes. */
867 kstat_named_t arcstat_l2_rebuild_asize
;
869 * Number of L2ARC log entries (buffers) that were successfully
872 kstat_named_t arcstat_l2_rebuild_bufs
;
874 * Number of L2ARC log entries (buffers) already cached in ARC. These
875 * were not restored again.
877 kstat_named_t arcstat_l2_rebuild_bufs_precached
;
879 * Number of L2ARC log blocks that were restored successfully. Each
880 * log block may hold up to L2ARC_LOG_BLK_MAX_ENTRIES buffers.
882 kstat_named_t arcstat_l2_rebuild_log_blks
;
883 kstat_named_t arcstat_memory_throttle_count
;
884 kstat_named_t arcstat_memory_direct_count
;
885 kstat_named_t arcstat_memory_indirect_count
;
886 kstat_named_t arcstat_memory_all_bytes
;
887 kstat_named_t arcstat_memory_free_bytes
;
888 kstat_named_t arcstat_memory_available_bytes
;
889 kstat_named_t arcstat_no_grow
;
890 kstat_named_t arcstat_tempreserve
;
891 kstat_named_t arcstat_loaned_bytes
;
892 kstat_named_t arcstat_prune
;
893 kstat_named_t arcstat_meta_used
;
894 kstat_named_t arcstat_dnode_limit
;
895 kstat_named_t arcstat_async_upgrade_sync
;
896 /* Number of predictive prefetch requests. */
897 kstat_named_t arcstat_predictive_prefetch
;
898 /* Number of requests for which predictive prefetch has completed. */
899 kstat_named_t arcstat_demand_hit_predictive_prefetch
;
900 /* Number of requests for which predictive prefetch was running. */
901 kstat_named_t arcstat_demand_iohit_predictive_prefetch
;
902 /* Number of prescient prefetch requests. */
903 kstat_named_t arcstat_prescient_prefetch
;
904 /* Number of requests for which prescient prefetch has completed. */
905 kstat_named_t arcstat_demand_hit_prescient_prefetch
;
906 /* Number of requests for which prescient prefetch was running. */
907 kstat_named_t arcstat_demand_iohit_prescient_prefetch
;
908 kstat_named_t arcstat_need_free
;
909 kstat_named_t arcstat_sys_free
;
910 kstat_named_t arcstat_raw_size
;
911 kstat_named_t arcstat_cached_only_in_progress
;
912 kstat_named_t arcstat_abd_chunk_waste_size
;
915 typedef struct arc_sums
{
916 wmsum_t arcstat_hits
;
917 wmsum_t arcstat_iohits
;
918 wmsum_t arcstat_misses
;
919 wmsum_t arcstat_demand_data_hits
;
920 wmsum_t arcstat_demand_data_iohits
;
921 wmsum_t arcstat_demand_data_misses
;
922 wmsum_t arcstat_demand_metadata_hits
;
923 wmsum_t arcstat_demand_metadata_iohits
;
924 wmsum_t arcstat_demand_metadata_misses
;
925 wmsum_t arcstat_prefetch_data_hits
;
926 wmsum_t arcstat_prefetch_data_iohits
;
927 wmsum_t arcstat_prefetch_data_misses
;
928 wmsum_t arcstat_prefetch_metadata_hits
;
929 wmsum_t arcstat_prefetch_metadata_iohits
;
930 wmsum_t arcstat_prefetch_metadata_misses
;
931 wmsum_t arcstat_mru_hits
;
932 wmsum_t arcstat_mru_ghost_hits
;
933 wmsum_t arcstat_mfu_hits
;
934 wmsum_t arcstat_mfu_ghost_hits
;
935 wmsum_t arcstat_uncached_hits
;
936 wmsum_t arcstat_deleted
;
937 wmsum_t arcstat_mutex_miss
;
938 wmsum_t arcstat_access_skip
;
939 wmsum_t arcstat_evict_skip
;
940 wmsum_t arcstat_evict_not_enough
;
941 wmsum_t arcstat_evict_l2_cached
;
942 wmsum_t arcstat_evict_l2_eligible
;
943 wmsum_t arcstat_evict_l2_eligible_mfu
;
944 wmsum_t arcstat_evict_l2_eligible_mru
;
945 wmsum_t arcstat_evict_l2_ineligible
;
946 wmsum_t arcstat_evict_l2_skip
;
947 wmsum_t arcstat_hash_collisions
;
948 wmsum_t arcstat_hash_chains
;
949 aggsum_t arcstat_size
;
950 wmsum_t arcstat_compressed_size
;
951 wmsum_t arcstat_uncompressed_size
;
952 wmsum_t arcstat_overhead_size
;
953 wmsum_t arcstat_hdr_size
;
954 wmsum_t arcstat_data_size
;
955 wmsum_t arcstat_metadata_size
;
956 wmsum_t arcstat_dbuf_size
;
957 wmsum_t arcstat_dnode_size
;
958 wmsum_t arcstat_bonus_size
;
959 wmsum_t arcstat_l2_hits
;
960 wmsum_t arcstat_l2_misses
;
961 wmsum_t arcstat_l2_prefetch_asize
;
962 wmsum_t arcstat_l2_mru_asize
;
963 wmsum_t arcstat_l2_mfu_asize
;
964 wmsum_t arcstat_l2_bufc_data_asize
;
965 wmsum_t arcstat_l2_bufc_metadata_asize
;
966 wmsum_t arcstat_l2_feeds
;
967 wmsum_t arcstat_l2_rw_clash
;
968 wmsum_t arcstat_l2_read_bytes
;
969 wmsum_t arcstat_l2_write_bytes
;
970 wmsum_t arcstat_l2_writes_sent
;
971 wmsum_t arcstat_l2_writes_done
;
972 wmsum_t arcstat_l2_writes_error
;
973 wmsum_t arcstat_l2_writes_lock_retry
;
974 wmsum_t arcstat_l2_evict_lock_retry
;
975 wmsum_t arcstat_l2_evict_reading
;
976 wmsum_t arcstat_l2_evict_l1cached
;
977 wmsum_t arcstat_l2_free_on_write
;
978 wmsum_t arcstat_l2_abort_lowmem
;
979 wmsum_t arcstat_l2_cksum_bad
;
980 wmsum_t arcstat_l2_io_error
;
981 wmsum_t arcstat_l2_lsize
;
982 wmsum_t arcstat_l2_psize
;
983 aggsum_t arcstat_l2_hdr_size
;
984 wmsum_t arcstat_l2_log_blk_writes
;
985 wmsum_t arcstat_l2_log_blk_asize
;
986 wmsum_t arcstat_l2_log_blk_count
;
987 wmsum_t arcstat_l2_rebuild_success
;
988 wmsum_t arcstat_l2_rebuild_abort_unsupported
;
989 wmsum_t arcstat_l2_rebuild_abort_io_errors
;
990 wmsum_t arcstat_l2_rebuild_abort_dh_errors
;
991 wmsum_t arcstat_l2_rebuild_abort_cksum_lb_errors
;
992 wmsum_t arcstat_l2_rebuild_abort_lowmem
;
993 wmsum_t arcstat_l2_rebuild_size
;
994 wmsum_t arcstat_l2_rebuild_asize
;
995 wmsum_t arcstat_l2_rebuild_bufs
;
996 wmsum_t arcstat_l2_rebuild_bufs_precached
;
997 wmsum_t arcstat_l2_rebuild_log_blks
;
998 wmsum_t arcstat_memory_throttle_count
;
999 wmsum_t arcstat_memory_direct_count
;
1000 wmsum_t arcstat_memory_indirect_count
;
1001 wmsum_t arcstat_prune
;
1002 wmsum_t arcstat_meta_used
;
1003 wmsum_t arcstat_async_upgrade_sync
;
1004 wmsum_t arcstat_predictive_prefetch
;
1005 wmsum_t arcstat_demand_hit_predictive_prefetch
;
1006 wmsum_t arcstat_demand_iohit_predictive_prefetch
;
1007 wmsum_t arcstat_prescient_prefetch
;
1008 wmsum_t arcstat_demand_hit_prescient_prefetch
;
1009 wmsum_t arcstat_demand_iohit_prescient_prefetch
;
1010 wmsum_t arcstat_raw_size
;
1011 wmsum_t arcstat_cached_only_in_progress
;
1012 wmsum_t arcstat_abd_chunk_waste_size
;
1015 typedef struct arc_evict_waiter
{
1016 list_node_t aew_node
;
1019 } arc_evict_waiter_t
;
1021 #define ARCSTAT(stat) (arc_stats.stat.value.ui64)
1023 #define ARCSTAT_INCR(stat, val) \
1024 wmsum_add(&arc_sums.stat, (val))
1026 #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1)
1027 #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1)
1029 #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */
1030 #define arc_meta ARCSTAT(arcstat_meta) /* target frac of metadata */
1031 #define arc_pd ARCSTAT(arcstat_pd) /* target frac of data MRU */
1032 #define arc_pm ARCSTAT(arcstat_pm) /* target frac of meta MRU */
1033 #define arc_c ARCSTAT(arcstat_c) /* target size of cache */
1034 #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */
1035 #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */
1036 #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */
1038 #define arc_anon (&ARC_anon)
1039 #define arc_mru (&ARC_mru)
1040 #define arc_mru_ghost (&ARC_mru_ghost)
1041 #define arc_mfu (&ARC_mfu)
1042 #define arc_mfu_ghost (&ARC_mfu_ghost)
1043 #define arc_l2c_only (&ARC_l2c_only)
1044 #define arc_uncached (&ARC_uncached)
1046 extern taskq_t
*arc_prune_taskq
;
1047 extern arc_stats_t arc_stats
;
1048 extern arc_sums_t arc_sums
;
1049 extern hrtime_t arc_growtime
;
1050 extern boolean_t arc_warm
;
1051 extern uint_t arc_grow_retry
;
1052 extern uint_t arc_no_grow_shift
;
1053 extern uint_t arc_shrink_shift
;
1054 extern kmutex_t arc_prune_mtx
;
1055 extern list_t arc_prune_list
;
1056 extern arc_state_t ARC_mfu
;
1057 extern arc_state_t ARC_mru
;
1058 extern uint_t zfs_arc_pc_percent
;
1059 extern uint_t arc_lotsfree_percent
;
1060 extern uint64_t zfs_arc_min
;
1061 extern uint64_t zfs_arc_max
;
1063 extern void arc_reduce_target_size(int64_t to_free
);
1064 extern boolean_t
arc_reclaim_needed(void);
1065 extern void arc_kmem_reap_soon(void);
1066 extern void arc_wait_for_eviction(uint64_t, boolean_t
);
1068 extern void arc_lowmem_init(void);
1069 extern void arc_lowmem_fini(void);
1070 extern void arc_prune_async(uint64_t);
1071 extern int arc_memory_throttle(spa_t
*spa
, uint64_t reserve
, uint64_t txg
);
1072 extern uint64_t arc_free_memory(void);
1073 extern int64_t arc_available_memory(void);
1074 extern void arc_tuning_update(boolean_t
);
1075 extern void arc_register_hotplug(void);
1076 extern void arc_unregister_hotplug(void);
1078 extern int param_set_arc_u64(ZFS_MODULE_PARAM_ARGS
);
1079 extern int param_set_arc_int(ZFS_MODULE_PARAM_ARGS
);
1080 extern int param_set_arc_min(ZFS_MODULE_PARAM_ARGS
);
1081 extern int param_set_arc_max(ZFS_MODULE_PARAM_ARGS
);
1084 boolean_t
l2arc_log_blkptr_valid(l2arc_dev_t
*dev
,
1085 const l2arc_log_blkptr_t
*lbp
);
1087 /* used in vdev_trim.c */
1088 void l2arc_dev_hdr_update(l2arc_dev_t
*dev
);
1089 l2arc_dev_t
*l2arc_vdev_get(vdev_t
*vd
);
1095 #endif /* _SYS_ARC_IMPL_H */