config/dracut/90zfs: handle cases where hostid(1) returns all zeros
[zfs.git] / include / sys / arc_impl.h
blobb08a42c75e8d731b518e3c2872250e6fef22a168
1 /*
2 * CDDL HEADER START
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 http://www.opensolaris.org/os/licensing.
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]
19 * CDDL HEADER END
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
32 #include <sys/arc.h>
33 #include <sys/zio_crypt.h>
34 #include <sys/zthr.h>
35 #include <sys/aggsum.h>
37 #ifdef __cplusplus
38 extern "C" {
39 #endif
42 * Note that buffers can be in one of 6 states:
43 * ARC_anon - anonymous (discussed below)
44 * ARC_mru - recently used, currently cached
45 * ARC_mru_ghost - recently used, no longer in cache
46 * ARC_mfu - frequently used, currently cached
47 * ARC_mfu_ghost - frequently used, no longer in cache
48 * ARC_l2c_only - exists in L2ARC but not other states
49 * When there are no active references to the buffer, they are
50 * are linked onto a list in one of these arc states. These are
51 * the only buffers that can be evicted or deleted. Within each
52 * state there are multiple lists, one for meta-data and one for
53 * non-meta-data. Meta-data (indirect blocks, blocks of dnodes,
54 * etc.) is tracked separately so that it can be managed more
55 * explicitly: favored over data, limited explicitly.
57 * Anonymous buffers are buffers that are not associated with
58 * a DVA. These are buffers that hold dirty block copies
59 * before they are written to stable storage. By definition,
60 * they are "ref'd" and are considered part of arc_mru
61 * that cannot be freed. Generally, they will acquire a DVA
62 * as they are written and migrate onto the arc_mru list.
64 * The ARC_l2c_only state is for buffers that are in the second
65 * level ARC but no longer in any of the ARC_m* lists. The second
66 * level ARC itself may also contain buffers that are in any of
67 * the ARC_m* states - meaning that a buffer can exist in two
68 * places. The reason for the ARC_l2c_only state is to keep the
69 * buffer header in the hash table, so that reads that hit the
70 * second level ARC benefit from these fast lookups.
73 typedef struct arc_state {
75 * list of evictable buffers
77 multilist_t *arcs_list[ARC_BUFC_NUMTYPES];
79 * total amount of evictable data in this state
81 zfs_refcount_t arcs_esize[ARC_BUFC_NUMTYPES];
83 * total amount of data in this state; this includes: evictable,
84 * non-evictable, ARC_BUFC_DATA, and ARC_BUFC_METADATA.
86 zfs_refcount_t arcs_size;
88 * supports the "dbufs" kstat
90 arc_state_type_t arcs_state;
91 } arc_state_t;
93 typedef struct arc_callback arc_callback_t;
95 struct arc_callback {
96 void *acb_private;
97 arc_read_done_func_t *acb_done;
98 arc_buf_t *acb_buf;
99 boolean_t acb_encrypted;
100 boolean_t acb_compressed;
101 boolean_t acb_noauth;
102 zbookmark_phys_t acb_zb;
103 zio_t *acb_zio_dummy;
104 zio_t *acb_zio_head;
105 arc_callback_t *acb_next;
108 typedef struct arc_write_callback arc_write_callback_t;
110 struct arc_write_callback {
111 void *awcb_private;
112 arc_write_done_func_t *awcb_ready;
113 arc_write_done_func_t *awcb_children_ready;
114 arc_write_done_func_t *awcb_physdone;
115 arc_write_done_func_t *awcb_done;
116 arc_buf_t *awcb_buf;
120 * ARC buffers are separated into multiple structs as a memory saving measure:
121 * - Common fields struct, always defined, and embedded within it:
122 * - L2-only fields, always allocated but undefined when not in L2ARC
123 * - L1-only fields, only allocated when in L1ARC
125 * Buffer in L1 Buffer only in L2
126 * +------------------------+ +------------------------+
127 * | arc_buf_hdr_t | | arc_buf_hdr_t |
128 * | | | |
129 * | | | |
130 * | | | |
131 * +------------------------+ +------------------------+
132 * | l2arc_buf_hdr_t | | l2arc_buf_hdr_t |
133 * | (undefined if L1-only) | | |
134 * +------------------------+ +------------------------+
135 * | l1arc_buf_hdr_t |
136 * | |
137 * | |
138 * | |
139 * | |
140 * +------------------------+
142 * Because it's possible for the L2ARC to become extremely large, we can wind
143 * up eating a lot of memory in L2ARC buffer headers, so the size of a header
144 * is minimized by only allocating the fields necessary for an L1-cached buffer
145 * when a header is actually in the L1 cache. The sub-headers (l1arc_buf_hdr and
146 * l2arc_buf_hdr) are embedded rather than allocated separately to save a couple
147 * words in pointers. arc_hdr_realloc() is used to switch a header between
148 * these two allocation states.
150 typedef struct l1arc_buf_hdr {
151 kmutex_t b_freeze_lock;
152 zio_cksum_t *b_freeze_cksum;
154 arc_buf_t *b_buf;
155 uint32_t b_bufcnt;
156 /* for waiting on writes to complete */
157 kcondvar_t b_cv;
158 uint8_t b_byteswap;
161 /* protected by arc state mutex */
162 arc_state_t *b_state;
163 multilist_node_t b_arc_node;
165 /* updated atomically */
166 clock_t b_arc_access;
167 uint32_t b_mru_hits;
168 uint32_t b_mru_ghost_hits;
169 uint32_t b_mfu_hits;
170 uint32_t b_mfu_ghost_hits;
171 uint32_t b_l2_hits;
173 /* self protecting */
174 zfs_refcount_t b_refcnt;
176 arc_callback_t *b_acb;
177 abd_t *b_pabd;
178 } l1arc_buf_hdr_t;
180 typedef enum l2arc_dev_hdr_flags_t {
181 L2ARC_DEV_HDR_EVICT_FIRST = (1 << 0) /* mirror of l2ad_first */
182 } l2arc_dev_hdr_flags_t;
185 * Pointer used in persistent L2ARC (for pointing to log blocks).
187 typedef struct l2arc_log_blkptr {
189 * Offset of log block within the device, in bytes
191 uint64_t lbp_daddr;
193 * Aligned payload size (in bytes) of the log block
195 uint64_t lbp_payload_asize;
197 * Offset in bytes of the first buffer in the payload
199 uint64_t lbp_payload_start;
201 * lbp_prop has the following format:
202 * * logical size (in bytes)
203 * * aligned (after compression) size (in bytes)
204 * * compression algorithm (we always LZ4-compress l2arc logs)
205 * * checksum algorithm (used for lbp_cksum)
207 uint64_t lbp_prop;
208 zio_cksum_t lbp_cksum; /* checksum of log */
209 } l2arc_log_blkptr_t;
212 * The persistent L2ARC device header.
213 * Byte order of magic determines whether 64-bit bswap of fields is necessary.
215 typedef struct l2arc_dev_hdr_phys {
216 uint64_t dh_magic; /* L2ARC_DEV_HDR_MAGIC */
217 uint64_t dh_version; /* Persistent L2ARC version */
220 * Global L2ARC device state and metadata.
222 uint64_t dh_spa_guid;
223 uint64_t dh_vdev_guid;
224 uint64_t dh_log_entries; /* mirror of l2ad_log_entries */
225 uint64_t dh_evict; /* evicted offset in bytes */
226 uint64_t dh_flags; /* l2arc_dev_hdr_flags_t */
228 * Used in zdb.c for determining if a log block is valid, in the same
229 * way that l2arc_rebuild() does.
231 uint64_t dh_start; /* mirror of l2ad_start */
232 uint64_t dh_end; /* mirror of l2ad_end */
234 * Start of log block chain. [0] -> newest log, [1] -> one older (used
235 * for initiating prefetch).
237 l2arc_log_blkptr_t dh_start_lbps[2];
239 * Aligned size of all log blocks as accounted by vdev_space_update().
241 uint64_t dh_lb_asize; /* mirror of l2ad_lb_asize */
242 uint64_t dh_lb_count; /* mirror of l2ad_lb_count */
244 * Mirrors of vdev_trim_action_time and vdev_trim_state, used to
245 * display when the cache device was fully trimmed for the last
246 * time.
248 uint64_t dh_trim_action_time;
249 uint64_t dh_trim_state;
250 const uint64_t dh_pad[30]; /* pad to 512 bytes */
251 zio_eck_t dh_tail;
252 } l2arc_dev_hdr_phys_t;
253 CTASSERT_GLOBAL(sizeof (l2arc_dev_hdr_phys_t) == SPA_MINBLOCKSIZE);
256 * A single ARC buffer header entry in a l2arc_log_blk_phys_t.
258 typedef struct l2arc_log_ent_phys {
259 dva_t le_dva; /* dva of buffer */
260 uint64_t le_birth; /* birth txg of buffer */
262 * le_prop has the following format:
263 * * logical size (in bytes)
264 * * physical (compressed) size (in bytes)
265 * * compression algorithm
266 * * object type (used to restore arc_buf_contents_t)
267 * * protected status (used for encryption)
268 * * prefetch status (used in l2arc_read_done())
270 uint64_t le_prop;
271 uint64_t le_daddr; /* buf location on l2dev */
272 uint64_t le_complevel;
274 * We pad the size of each entry to a power of 2 so that the size of
275 * l2arc_log_blk_phys_t is power-of-2 aligned with SPA_MINBLOCKSHIFT,
276 * because of the L2ARC_SET_*SIZE macros.
278 const uint64_t le_pad[2]; /* pad to 64 bytes */
279 } l2arc_log_ent_phys_t;
281 #define L2ARC_LOG_BLK_MAX_ENTRIES (1022)
284 * A log block of up to 1022 ARC buffer log entries, chained into the
285 * persistent L2ARC metadata linked list. Byte order of magic determines
286 * whether 64-bit bswap of fields is necessary.
288 typedef struct l2arc_log_blk_phys {
289 uint64_t lb_magic; /* L2ARC_LOG_BLK_MAGIC */
291 * There are 2 chains (headed by dh_start_lbps[2]), and this field
292 * points back to the previous block in this chain. We alternate
293 * which chain we append to, so they are time-wise and offset-wise
294 * interleaved, but that is an optimization rather than for
295 * correctness.
297 l2arc_log_blkptr_t lb_prev_lbp; /* pointer to prev log block */
299 * Pad header section to 128 bytes
301 uint64_t lb_pad[7];
302 /* Payload */
303 l2arc_log_ent_phys_t lb_entries[L2ARC_LOG_BLK_MAX_ENTRIES];
304 } l2arc_log_blk_phys_t; /* 64K total */
307 * The size of l2arc_log_blk_phys_t has to be power-of-2 aligned with
308 * SPA_MINBLOCKSHIFT because of L2BLK_SET_*SIZE macros.
310 CTASSERT_GLOBAL(IS_P2ALIGNED(sizeof (l2arc_log_blk_phys_t),
311 1ULL << SPA_MINBLOCKSHIFT));
312 CTASSERT_GLOBAL(sizeof (l2arc_log_blk_phys_t) >= SPA_MINBLOCKSIZE);
313 CTASSERT_GLOBAL(sizeof (l2arc_log_blk_phys_t) <= SPA_MAXBLOCKSIZE);
316 * These structures hold in-flight abd buffers for log blocks as they're being
317 * written to the L2ARC device.
319 typedef struct l2arc_lb_abd_buf {
320 abd_t *abd;
321 list_node_t node;
322 } l2arc_lb_abd_buf_t;
325 * These structures hold pointers to log blocks present on the L2ARC device.
327 typedef struct l2arc_lb_ptr_buf {
328 l2arc_log_blkptr_t *lb_ptr;
329 list_node_t node;
330 } l2arc_lb_ptr_buf_t;
332 /* Macros for setting fields in le_prop and lbp_prop */
333 #define L2BLK_GET_LSIZE(field) \
334 BF64_GET_SB((field), 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1)
335 #define L2BLK_SET_LSIZE(field, x) \
336 BF64_SET_SB((field), 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x)
337 #define L2BLK_GET_PSIZE(field) \
338 BF64_GET_SB((field), 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1)
339 #define L2BLK_SET_PSIZE(field, x) \
340 BF64_SET_SB((field), 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x)
341 #define L2BLK_GET_COMPRESS(field) \
342 BF64_GET((field), 32, SPA_COMPRESSBITS)
343 #define L2BLK_SET_COMPRESS(field, x) \
344 BF64_SET((field), 32, SPA_COMPRESSBITS, x)
345 #define L2BLK_GET_PREFETCH(field) BF64_GET((field), 39, 1)
346 #define L2BLK_SET_PREFETCH(field, x) BF64_SET((field), 39, 1, x)
347 #define L2BLK_GET_CHECKSUM(field) BF64_GET((field), 40, 8)
348 #define L2BLK_SET_CHECKSUM(field, x) BF64_SET((field), 40, 8, x)
349 #define L2BLK_GET_TYPE(field) BF64_GET((field), 48, 8)
350 #define L2BLK_SET_TYPE(field, x) BF64_SET((field), 48, 8, x)
351 #define L2BLK_GET_PROTECTED(field) BF64_GET((field), 56, 1)
352 #define L2BLK_SET_PROTECTED(field, x) BF64_SET((field), 56, 1, x)
353 #define L2BLK_GET_STATE(field) BF64_GET((field), 57, 4)
354 #define L2BLK_SET_STATE(field, x) BF64_SET((field), 57, 4, x)
356 #define PTR_SWAP(x, y) \
357 do { \
358 void *tmp = (x);\
359 x = y; \
360 y = tmp; \
361 _NOTE(CONSTCOND)\
362 } while (0)
364 #define L2ARC_DEV_HDR_MAGIC 0x5a46534341434845LLU /* ASCII: "ZFSCACHE" */
365 #define L2ARC_LOG_BLK_MAGIC 0x4c4f47424c4b4844LLU /* ASCII: "LOGBLKHD" */
368 * L2ARC Internals
370 typedef struct l2arc_dev {
371 vdev_t *l2ad_vdev; /* vdev */
372 spa_t *l2ad_spa; /* spa */
373 uint64_t l2ad_hand; /* next write location */
374 uint64_t l2ad_start; /* first addr on device */
375 uint64_t l2ad_end; /* last addr on device */
376 boolean_t l2ad_first; /* first sweep through */
377 boolean_t l2ad_writing; /* currently writing */
378 kmutex_t l2ad_mtx; /* lock for buffer list */
379 list_t l2ad_buflist; /* buffer list */
380 list_node_t l2ad_node; /* device list node */
381 zfs_refcount_t l2ad_alloc; /* allocated bytes */
383 * Persistence-related stuff
385 l2arc_dev_hdr_phys_t *l2ad_dev_hdr; /* persistent device header */
386 uint64_t l2ad_dev_hdr_asize; /* aligned hdr size */
387 l2arc_log_blk_phys_t l2ad_log_blk; /* currently open log block */
388 int l2ad_log_ent_idx; /* index into cur log blk */
389 /* Number of bytes in current log block's payload */
390 uint64_t l2ad_log_blk_payload_asize;
392 * Offset (in bytes) of the first buffer in current log block's
393 * payload.
395 uint64_t l2ad_log_blk_payload_start;
396 /* Flag indicating whether a rebuild is scheduled or is going on */
397 boolean_t l2ad_rebuild;
398 boolean_t l2ad_rebuild_cancel;
399 boolean_t l2ad_rebuild_began;
400 uint64_t l2ad_log_entries; /* entries per log blk */
401 uint64_t l2ad_evict; /* evicted offset in bytes */
402 /* List of pointers to log blocks present in the L2ARC device */
403 list_t l2ad_lbptr_list;
405 * Aligned size of all log blocks as accounted by vdev_space_update().
407 zfs_refcount_t l2ad_lb_asize;
409 * Number of log blocks present on the device.
411 zfs_refcount_t l2ad_lb_count;
412 boolean_t l2ad_trim_all; /* TRIM whole device */
413 } l2arc_dev_t;
416 * Encrypted blocks will need to be stored encrypted on the L2ARC
417 * disk as they appear in the main pool. In order for this to work we
418 * need to pass around the encryption parameters so they can be used
419 * to write data to the L2ARC. This struct is only defined in the
420 * arc_buf_hdr_t if the L1 header is defined and has the ARC_FLAG_ENCRYPTED
421 * flag set.
423 typedef struct arc_buf_hdr_crypt {
424 abd_t *b_rabd; /* raw encrypted data */
425 dmu_object_type_t b_ot; /* object type */
426 uint32_t b_ebufcnt; /* count of encrypted buffers */
428 /* dsobj for looking up encryption key for l2arc encryption */
429 uint64_t b_dsobj;
431 /* encryption parameters */
432 uint8_t b_salt[ZIO_DATA_SALT_LEN];
433 uint8_t b_iv[ZIO_DATA_IV_LEN];
436 * Technically this could be removed since we will always be able to
437 * get the mac from the bp when we need it. However, it is inconvenient
438 * for callers of arc code to have to pass a bp in all the time. This
439 * also allows us to assert that L2ARC data is properly encrypted to
440 * match the data in the main storage pool.
442 uint8_t b_mac[ZIO_DATA_MAC_LEN];
443 } arc_buf_hdr_crypt_t;
445 typedef struct l2arc_buf_hdr {
446 /* protected by arc_buf_hdr mutex */
447 l2arc_dev_t *b_dev; /* L2ARC device */
448 uint64_t b_daddr; /* disk address, offset byte */
449 uint32_t b_hits;
450 arc_state_type_t b_arcs_state;
451 list_node_t b_l2node;
452 } l2arc_buf_hdr_t;
454 typedef struct l2arc_write_callback {
455 l2arc_dev_t *l2wcb_dev; /* device info */
456 arc_buf_hdr_t *l2wcb_head; /* head of write buflist */
457 /* in-flight list of log blocks */
458 list_t l2wcb_abd_list;
459 } l2arc_write_callback_t;
461 struct arc_buf_hdr {
462 /* protected by hash lock */
463 dva_t b_dva;
464 uint64_t b_birth;
466 arc_buf_contents_t b_type;
467 uint8_t b_complevel;
468 uint8_t b_reserved1; /* used for 4 byte alignment */
469 uint16_t b_reserved2; /* used for 4 byte alignment */
470 arc_buf_hdr_t *b_hash_next;
471 arc_flags_t b_flags;
474 * This field stores the size of the data buffer after
475 * compression, and is set in the arc's zio completion handlers.
476 * It is in units of SPA_MINBLOCKSIZE (e.g. 1 == 512 bytes).
478 * While the block pointers can store up to 32MB in their psize
479 * field, we can only store up to 32MB minus 512B. This is due
480 * to the bp using a bias of 1, whereas we use a bias of 0 (i.e.
481 * a field of zeros represents 512B in the bp). We can't use a
482 * bias of 1 since we need to reserve a psize of zero, here, to
483 * represent holes and embedded blocks.
485 * This isn't a problem in practice, since the maximum size of a
486 * buffer is limited to 16MB, so we never need to store 32MB in
487 * this field. Even in the upstream illumos code base, the
488 * maximum size of a buffer is limited to 16MB.
490 uint16_t b_psize;
493 * This field stores the size of the data buffer before
494 * compression, and cannot change once set. It is in units
495 * of SPA_MINBLOCKSIZE (e.g. 2 == 1024 bytes)
497 uint16_t b_lsize; /* immutable */
498 uint64_t b_spa; /* immutable */
500 /* L2ARC fields. Undefined when not in L2ARC. */
501 l2arc_buf_hdr_t b_l2hdr;
502 /* L1ARC fields. Undefined when in l2arc_only state */
503 l1arc_buf_hdr_t b_l1hdr;
505 * Encryption parameters. Defined only when ARC_FLAG_ENCRYPTED
506 * is set and the L1 header exists.
508 arc_buf_hdr_crypt_t b_crypt_hdr;
511 typedef struct arc_stats {
512 kstat_named_t arcstat_hits;
513 kstat_named_t arcstat_misses;
514 kstat_named_t arcstat_demand_data_hits;
515 kstat_named_t arcstat_demand_data_misses;
516 kstat_named_t arcstat_demand_metadata_hits;
517 kstat_named_t arcstat_demand_metadata_misses;
518 kstat_named_t arcstat_prefetch_data_hits;
519 kstat_named_t arcstat_prefetch_data_misses;
520 kstat_named_t arcstat_prefetch_metadata_hits;
521 kstat_named_t arcstat_prefetch_metadata_misses;
522 kstat_named_t arcstat_mru_hits;
523 kstat_named_t arcstat_mru_ghost_hits;
524 kstat_named_t arcstat_mfu_hits;
525 kstat_named_t arcstat_mfu_ghost_hits;
526 kstat_named_t arcstat_deleted;
528 * Number of buffers that could not be evicted because the hash lock
529 * was held by another thread. The lock may not necessarily be held
530 * by something using the same buffer, since hash locks are shared
531 * by multiple buffers.
533 kstat_named_t arcstat_mutex_miss;
535 * Number of buffers skipped when updating the access state due to the
536 * header having already been released after acquiring the hash lock.
538 kstat_named_t arcstat_access_skip;
540 * Number of buffers skipped because they have I/O in progress, are
541 * indirect prefetch buffers that have not lived long enough, or are
542 * not from the spa we're trying to evict from.
544 kstat_named_t arcstat_evict_skip;
546 * Number of times arc_evict_state() was unable to evict enough
547 * buffers to reach its target amount.
549 kstat_named_t arcstat_evict_not_enough;
550 kstat_named_t arcstat_evict_l2_cached;
551 kstat_named_t arcstat_evict_l2_eligible;
552 kstat_named_t arcstat_evict_l2_eligible_mfu;
553 kstat_named_t arcstat_evict_l2_eligible_mru;
554 kstat_named_t arcstat_evict_l2_ineligible;
555 kstat_named_t arcstat_evict_l2_skip;
556 kstat_named_t arcstat_hash_elements;
557 kstat_named_t arcstat_hash_elements_max;
558 kstat_named_t arcstat_hash_collisions;
559 kstat_named_t arcstat_hash_chains;
560 kstat_named_t arcstat_hash_chain_max;
561 kstat_named_t arcstat_p;
562 kstat_named_t arcstat_c;
563 kstat_named_t arcstat_c_min;
564 kstat_named_t arcstat_c_max;
565 /* Not updated directly; only synced in arc_kstat_update. */
566 kstat_named_t arcstat_size;
568 * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd.
569 * Note that the compressed bytes may match the uncompressed bytes
570 * if the block is either not compressed or compressed arc is disabled.
572 kstat_named_t arcstat_compressed_size;
574 * Uncompressed size of the data stored in b_pabd. If compressed
575 * arc is disabled then this value will be identical to the stat
576 * above.
578 kstat_named_t arcstat_uncompressed_size;
580 * Number of bytes stored in all the arc_buf_t's. This is classified
581 * as "overhead" since this data is typically short-lived and will
582 * be evicted from the arc when it becomes unreferenced unless the
583 * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level
584 * values have been set (see comment in dbuf.c for more information).
586 kstat_named_t arcstat_overhead_size;
588 * Number of bytes consumed by internal ARC structures necessary
589 * for tracking purposes; these structures are not actually
590 * backed by ARC buffers. This includes arc_buf_hdr_t structures
591 * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only
592 * caches), and arc_buf_t structures (allocated via arc_buf_t
593 * cache).
594 * Not updated directly; only synced in arc_kstat_update.
596 kstat_named_t arcstat_hdr_size;
598 * Number of bytes consumed by ARC buffers of type equal to
599 * ARC_BUFC_DATA. This is generally consumed by buffers backing
600 * on disk user data (e.g. plain file contents).
601 * Not updated directly; only synced in arc_kstat_update.
603 kstat_named_t arcstat_data_size;
605 * Number of bytes consumed by ARC buffers of type equal to
606 * ARC_BUFC_METADATA. This is generally consumed by buffers
607 * backing on disk data that is used for internal ZFS
608 * structures (e.g. ZAP, dnode, indirect blocks, etc).
609 * Not updated directly; only synced in arc_kstat_update.
611 kstat_named_t arcstat_metadata_size;
613 * Number of bytes consumed by dmu_buf_impl_t objects.
614 * Not updated directly; only synced in arc_kstat_update.
616 kstat_named_t arcstat_dbuf_size;
618 * Number of bytes consumed by dnode_t objects.
619 * Not updated directly; only synced in arc_kstat_update.
621 kstat_named_t arcstat_dnode_size;
623 * Number of bytes consumed by bonus buffers.
624 * Not updated directly; only synced in arc_kstat_update.
626 kstat_named_t arcstat_bonus_size;
627 #if defined(COMPAT_FREEBSD11)
629 * Sum of the previous three counters, provided for compatibility.
631 kstat_named_t arcstat_other_size;
632 #endif
635 * Total number of bytes consumed by ARC buffers residing in the
636 * arc_anon state. This includes *all* buffers in the arc_anon
637 * state; e.g. data, metadata, evictable, and unevictable buffers
638 * are all included in this value.
639 * Not updated directly; only synced in arc_kstat_update.
641 kstat_named_t arcstat_anon_size;
643 * Number of bytes consumed by ARC buffers that meet the
644 * following criteria: backing buffers of type ARC_BUFC_DATA,
645 * residing in the arc_anon state, and are eligible for eviction
646 * (e.g. have no outstanding holds on the buffer).
647 * Not updated directly; only synced in arc_kstat_update.
649 kstat_named_t arcstat_anon_evictable_data;
651 * Number of bytes consumed by ARC buffers that meet the
652 * following criteria: backing buffers of type ARC_BUFC_METADATA,
653 * residing in the arc_anon state, and are eligible for eviction
654 * (e.g. have no outstanding holds on the buffer).
655 * Not updated directly; only synced in arc_kstat_update.
657 kstat_named_t arcstat_anon_evictable_metadata;
659 * Total number of bytes consumed by ARC buffers residing in the
660 * arc_mru state. This includes *all* buffers in the arc_mru
661 * state; e.g. data, metadata, evictable, and unevictable buffers
662 * are all included in this value.
663 * Not updated directly; only synced in arc_kstat_update.
665 kstat_named_t arcstat_mru_size;
667 * Number of bytes consumed by ARC buffers that meet the
668 * following criteria: backing buffers of type ARC_BUFC_DATA,
669 * residing in the arc_mru state, and are eligible for eviction
670 * (e.g. have no outstanding holds on the buffer).
671 * Not updated directly; only synced in arc_kstat_update.
673 kstat_named_t arcstat_mru_evictable_data;
675 * Number of bytes consumed by ARC buffers that meet the
676 * following criteria: backing buffers of type ARC_BUFC_METADATA,
677 * residing in the arc_mru state, and are eligible for eviction
678 * (e.g. have no outstanding holds on the buffer).
679 * Not updated directly; only synced in arc_kstat_update.
681 kstat_named_t arcstat_mru_evictable_metadata;
683 * Total number of bytes that *would have been* consumed by ARC
684 * buffers in the arc_mru_ghost state. The key thing to note
685 * here, is the fact that this size doesn't actually indicate
686 * RAM consumption. The ghost lists only consist of headers and
687 * don't actually have ARC buffers linked off of these headers.
688 * Thus, *if* the headers had associated ARC buffers, these
689 * buffers *would have* consumed this number of bytes.
690 * Not updated directly; only synced in arc_kstat_update.
692 kstat_named_t arcstat_mru_ghost_size;
694 * Number of bytes that *would have been* consumed by ARC
695 * buffers that are eligible for eviction, of type
696 * ARC_BUFC_DATA, and linked off the arc_mru_ghost state.
697 * Not updated directly; only synced in arc_kstat_update.
699 kstat_named_t arcstat_mru_ghost_evictable_data;
701 * Number of bytes that *would have been* consumed by ARC
702 * buffers that are eligible for eviction, of type
703 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
704 * Not updated directly; only synced in arc_kstat_update.
706 kstat_named_t arcstat_mru_ghost_evictable_metadata;
708 * Total number of bytes consumed by ARC buffers residing in the
709 * arc_mfu state. This includes *all* buffers in the arc_mfu
710 * state; e.g. data, metadata, evictable, and unevictable buffers
711 * are all included in this value.
712 * Not updated directly; only synced in arc_kstat_update.
714 kstat_named_t arcstat_mfu_size;
716 * Number of bytes consumed by ARC buffers that are eligible for
717 * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu
718 * state.
719 * Not updated directly; only synced in arc_kstat_update.
721 kstat_named_t arcstat_mfu_evictable_data;
723 * Number of bytes consumed by ARC buffers that are eligible for
724 * eviction, of type ARC_BUFC_METADATA, and reside in the
725 * arc_mfu state.
726 * Not updated directly; only synced in arc_kstat_update.
728 kstat_named_t arcstat_mfu_evictable_metadata;
730 * Total number of bytes that *would have been* consumed by ARC
731 * buffers in the arc_mfu_ghost state. See the comment above
732 * arcstat_mru_ghost_size for more details.
733 * Not updated directly; only synced in arc_kstat_update.
735 kstat_named_t arcstat_mfu_ghost_size;
737 * Number of bytes that *would have been* consumed by ARC
738 * buffers that are eligible for eviction, of type
739 * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state.
740 * Not updated directly; only synced in arc_kstat_update.
742 kstat_named_t arcstat_mfu_ghost_evictable_data;
744 * Number of bytes that *would have been* consumed by ARC
745 * buffers that are eligible for eviction, of type
746 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
747 * Not updated directly; only synced in arc_kstat_update.
749 kstat_named_t arcstat_mfu_ghost_evictable_metadata;
750 kstat_named_t arcstat_l2_hits;
751 kstat_named_t arcstat_l2_misses;
753 * Allocated size (in bytes) of L2ARC cached buffers by ARC state.
755 kstat_named_t arcstat_l2_prefetch_asize;
756 kstat_named_t arcstat_l2_mru_asize;
757 kstat_named_t arcstat_l2_mfu_asize;
759 * Allocated size (in bytes) of L2ARC cached buffers by buffer content
760 * type.
762 kstat_named_t arcstat_l2_bufc_data_asize;
763 kstat_named_t arcstat_l2_bufc_metadata_asize;
764 kstat_named_t arcstat_l2_feeds;
765 kstat_named_t arcstat_l2_rw_clash;
766 kstat_named_t arcstat_l2_read_bytes;
767 kstat_named_t arcstat_l2_write_bytes;
768 kstat_named_t arcstat_l2_writes_sent;
769 kstat_named_t arcstat_l2_writes_done;
770 kstat_named_t arcstat_l2_writes_error;
771 kstat_named_t arcstat_l2_writes_lock_retry;
772 kstat_named_t arcstat_l2_evict_lock_retry;
773 kstat_named_t arcstat_l2_evict_reading;
774 kstat_named_t arcstat_l2_evict_l1cached;
775 kstat_named_t arcstat_l2_free_on_write;
776 kstat_named_t arcstat_l2_abort_lowmem;
777 kstat_named_t arcstat_l2_cksum_bad;
778 kstat_named_t arcstat_l2_io_error;
779 kstat_named_t arcstat_l2_lsize;
780 kstat_named_t arcstat_l2_psize;
781 /* Not updated directly; only synced in arc_kstat_update. */
782 kstat_named_t arcstat_l2_hdr_size;
784 * Number of L2ARC log blocks written. These are used for restoring the
785 * L2ARC. Updated during writing of L2ARC log blocks.
787 kstat_named_t arcstat_l2_log_blk_writes;
789 * Moving average of the aligned size of the L2ARC log blocks, in
790 * bytes. Updated during L2ARC rebuild and during writing of L2ARC
791 * log blocks.
793 kstat_named_t arcstat_l2_log_blk_avg_asize;
794 /* Aligned size of L2ARC log blocks on L2ARC devices. */
795 kstat_named_t arcstat_l2_log_blk_asize;
796 /* Number of L2ARC log blocks present on L2ARC devices. */
797 kstat_named_t arcstat_l2_log_blk_count;
799 * Moving average of the aligned size of L2ARC restored data, in bytes,
800 * to the aligned size of their metadata in L2ARC, in bytes.
801 * Updated during L2ARC rebuild and during writing of L2ARC log blocks.
803 kstat_named_t arcstat_l2_data_to_meta_ratio;
805 * Number of times the L2ARC rebuild was successful for an L2ARC device.
807 kstat_named_t arcstat_l2_rebuild_success;
809 * Number of times the L2ARC rebuild failed because the device header
810 * was in an unsupported format or corrupted.
812 kstat_named_t arcstat_l2_rebuild_abort_unsupported;
814 * Number of times the L2ARC rebuild failed because of IO errors
815 * while reading a log block.
817 kstat_named_t arcstat_l2_rebuild_abort_io_errors;
819 * Number of times the L2ARC rebuild failed because of IO errors when
820 * reading the device header.
822 kstat_named_t arcstat_l2_rebuild_abort_dh_errors;
824 * Number of L2ARC log blocks which failed to be restored due to
825 * checksum errors.
827 kstat_named_t arcstat_l2_rebuild_abort_cksum_lb_errors;
829 * Number of times the L2ARC rebuild was aborted due to low system
830 * memory.
832 kstat_named_t arcstat_l2_rebuild_abort_lowmem;
833 /* Logical size of L2ARC restored data, in bytes. */
834 kstat_named_t arcstat_l2_rebuild_size;
835 /* Aligned size of L2ARC restored data, in bytes. */
836 kstat_named_t arcstat_l2_rebuild_asize;
838 * Number of L2ARC log entries (buffers) that were successfully
839 * restored in ARC.
841 kstat_named_t arcstat_l2_rebuild_bufs;
843 * Number of L2ARC log entries (buffers) already cached in ARC. These
844 * were not restored again.
846 kstat_named_t arcstat_l2_rebuild_bufs_precached;
848 * Number of L2ARC log blocks that were restored successfully. Each
849 * log block may hold up to L2ARC_LOG_BLK_MAX_ENTRIES buffers.
851 kstat_named_t arcstat_l2_rebuild_log_blks;
852 kstat_named_t arcstat_memory_throttle_count;
853 kstat_named_t arcstat_memory_direct_count;
854 kstat_named_t arcstat_memory_indirect_count;
855 kstat_named_t arcstat_memory_all_bytes;
856 kstat_named_t arcstat_memory_free_bytes;
857 kstat_named_t arcstat_memory_available_bytes;
858 kstat_named_t arcstat_no_grow;
859 kstat_named_t arcstat_tempreserve;
860 kstat_named_t arcstat_loaned_bytes;
861 kstat_named_t arcstat_prune;
862 /* Not updated directly; only synced in arc_kstat_update. */
863 kstat_named_t arcstat_meta_used;
864 kstat_named_t arcstat_meta_limit;
865 kstat_named_t arcstat_dnode_limit;
866 kstat_named_t arcstat_meta_max;
867 kstat_named_t arcstat_meta_min;
868 kstat_named_t arcstat_async_upgrade_sync;
869 kstat_named_t arcstat_demand_hit_predictive_prefetch;
870 kstat_named_t arcstat_demand_hit_prescient_prefetch;
871 kstat_named_t arcstat_need_free;
872 kstat_named_t arcstat_sys_free;
873 kstat_named_t arcstat_raw_size;
874 kstat_named_t arcstat_cached_only_in_progress;
875 kstat_named_t arcstat_abd_chunk_waste_size;
876 } arc_stats_t;
878 typedef struct arc_evict_waiter {
879 list_node_t aew_node;
880 kcondvar_t aew_cv;
881 uint64_t aew_count;
882 } arc_evict_waiter_t;
884 #define ARCSTAT(stat) (arc_stats.stat.value.ui64)
886 #define ARCSTAT_INCR(stat, val) \
887 atomic_add_64(&arc_stats.stat.value.ui64, (val))
889 #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1)
890 #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1)
892 #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */
893 #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */
894 #define arc_c ARCSTAT(arcstat_c) /* target size of cache */
895 #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */
896 #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */
897 #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */
899 extern taskq_t *arc_prune_taskq;
900 extern arc_stats_t arc_stats;
901 extern hrtime_t arc_growtime;
902 extern boolean_t arc_warm;
903 extern int arc_grow_retry;
904 extern int arc_no_grow_shift;
905 extern int arc_shrink_shift;
906 extern kmutex_t arc_prune_mtx;
907 extern list_t arc_prune_list;
908 extern aggsum_t arc_size;
909 extern arc_state_t *arc_mfu;
910 extern arc_state_t *arc_mru;
911 extern uint_t zfs_arc_pc_percent;
912 extern int arc_lotsfree_percent;
913 extern unsigned long zfs_arc_min;
914 extern unsigned long zfs_arc_max;
916 extern void arc_reduce_target_size(int64_t to_free);
917 extern boolean_t arc_reclaim_needed(void);
918 extern void arc_kmem_reap_soon(void);
919 extern boolean_t arc_is_overflowing(void);
920 extern void arc_wait_for_eviction(uint64_t);
922 extern void arc_lowmem_init(void);
923 extern void arc_lowmem_fini(void);
924 extern void arc_prune_async(int64_t);
925 extern int arc_memory_throttle(spa_t *spa, uint64_t reserve, uint64_t txg);
926 extern uint64_t arc_free_memory(void);
927 extern int64_t arc_available_memory(void);
928 extern void arc_tuning_update(boolean_t);
930 extern int param_set_arc_long(ZFS_MODULE_PARAM_ARGS);
931 extern int param_set_arc_int(ZFS_MODULE_PARAM_ARGS);
933 /* used in zdb.c */
934 boolean_t l2arc_log_blkptr_valid(l2arc_dev_t *dev,
935 const l2arc_log_blkptr_t *lbp);
937 /* used in vdev_trim.c */
938 void l2arc_dev_hdr_update(l2arc_dev_t *dev);
939 l2arc_dev_t *l2arc_vdev_get(vdev_t *vd);
941 #ifdef __cplusplus
943 #endif
945 #endif /* _SYS_ARC_IMPL_H */