dmu_zfetch.c revision 288571
1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26/*
27 * Copyright (c) 2013 by Delphix. All rights reserved.
28 */
29
30#include <sys/zfs_context.h>
31#include <sys/dnode.h>
32#include <sys/dmu_objset.h>
33#include <sys/dmu_zfetch.h>
34#include <sys/dmu.h>
35#include <sys/dbuf.h>
36#include <sys/kstat.h>
37
38/*
39 * I'm against tune-ables, but these should probably exist as tweakable globals
40 * until we can get this working the way we want it to.
41 */
42
43int zfs_prefetch_disable = 0;
44
45/* max # of streams per zfetch */
46uint32_t	zfetch_max_streams = 8;
47/* min time before stream reclaim */
48uint32_t	zfetch_min_sec_reap = 2;
49/* max number of blocks to fetch at a time */
50uint32_t	zfetch_block_cap = 256;
51/* number of bytes in a array_read at which we stop prefetching (1Mb) */
52uint64_t	zfetch_array_rd_sz = 1024 * 1024;
53
54SYSCTL_DECL(_vfs_zfs);
55SYSCTL_INT(_vfs_zfs, OID_AUTO, prefetch_disable, CTLFLAG_RW,
56    &zfs_prefetch_disable, 0, "Disable prefetch");
57SYSCTL_NODE(_vfs_zfs, OID_AUTO, zfetch, CTLFLAG_RW, 0, "ZFS ZFETCH");
58TUNABLE_INT("vfs.zfs.zfetch.max_streams", &zfetch_max_streams);
59SYSCTL_UINT(_vfs_zfs_zfetch, OID_AUTO, max_streams, CTLFLAG_RW,
60    &zfetch_max_streams, 0, "Max # of streams per zfetch");
61TUNABLE_INT("vfs.zfs.zfetch.min_sec_reap", &zfetch_min_sec_reap);
62SYSCTL_UINT(_vfs_zfs_zfetch, OID_AUTO, min_sec_reap, CTLFLAG_RWTUN,
63    &zfetch_min_sec_reap, 0, "Min time before stream reclaim");
64TUNABLE_INT("vfs.zfs.zfetch.block_cap", &zfetch_block_cap);
65SYSCTL_UINT(_vfs_zfs_zfetch, OID_AUTO, block_cap, CTLFLAG_RWTUN,
66    &zfetch_block_cap, 0, "Max number of blocks to fetch at a time");
67TUNABLE_QUAD("vfs.zfs.zfetch.array_rd_sz", &zfetch_array_rd_sz);
68SYSCTL_UQUAD(_vfs_zfs_zfetch, OID_AUTO, array_rd_sz, CTLFLAG_RWTUN,
69    &zfetch_array_rd_sz, 0,
70    "Number of bytes in a array_read at which we stop prefetching");
71
72/* forward decls for static routines */
73static boolean_t	dmu_zfetch_colinear(zfetch_t *, zstream_t *);
74static void		dmu_zfetch_dofetch(zfetch_t *, zstream_t *);
75static uint64_t		dmu_zfetch_fetch(dnode_t *, uint64_t, uint64_t);
76static uint64_t		dmu_zfetch_fetchsz(dnode_t *, uint64_t, uint64_t);
77static boolean_t	dmu_zfetch_find(zfetch_t *, zstream_t *, int);
78static int		dmu_zfetch_stream_insert(zfetch_t *, zstream_t *);
79static zstream_t	*dmu_zfetch_stream_reclaim(zfetch_t *);
80static void		dmu_zfetch_stream_remove(zfetch_t *, zstream_t *);
81static int		dmu_zfetch_streams_equal(zstream_t *, zstream_t *);
82
83typedef struct zfetch_stats {
84	kstat_named_t zfetchstat_hits;
85	kstat_named_t zfetchstat_misses;
86	kstat_named_t zfetchstat_colinear_hits;
87	kstat_named_t zfetchstat_colinear_misses;
88	kstat_named_t zfetchstat_stride_hits;
89	kstat_named_t zfetchstat_stride_misses;
90	kstat_named_t zfetchstat_reclaim_successes;
91	kstat_named_t zfetchstat_reclaim_failures;
92	kstat_named_t zfetchstat_stream_resets;
93	kstat_named_t zfetchstat_stream_noresets;
94	kstat_named_t zfetchstat_bogus_streams;
95} zfetch_stats_t;
96
97static zfetch_stats_t zfetch_stats = {
98	{ "hits",			KSTAT_DATA_UINT64 },
99	{ "misses",			KSTAT_DATA_UINT64 },
100	{ "colinear_hits",		KSTAT_DATA_UINT64 },
101	{ "colinear_misses",		KSTAT_DATA_UINT64 },
102	{ "stride_hits",		KSTAT_DATA_UINT64 },
103	{ "stride_misses",		KSTAT_DATA_UINT64 },
104	{ "reclaim_successes",		KSTAT_DATA_UINT64 },
105	{ "reclaim_failures",		KSTAT_DATA_UINT64 },
106	{ "streams_resets",		KSTAT_DATA_UINT64 },
107	{ "streams_noresets",		KSTAT_DATA_UINT64 },
108	{ "bogus_streams",		KSTAT_DATA_UINT64 },
109};
110
111#define	ZFETCHSTAT_INCR(stat, val) \
112	atomic_add_64(&zfetch_stats.stat.value.ui64, (val));
113
114#define	ZFETCHSTAT_BUMP(stat)		ZFETCHSTAT_INCR(stat, 1);
115
116kstat_t		*zfetch_ksp;
117
118/*
119 * Given a zfetch structure and a zstream structure, determine whether the
120 * blocks to be read are part of a co-linear pair of existing prefetch
121 * streams.  If a set is found, coalesce the streams, removing one, and
122 * configure the prefetch so it looks for a strided access pattern.
123 *
124 * In other words: if we find two sequential access streams that are
125 * the same length and distance N appart, and this read is N from the
126 * last stream, then we are probably in a strided access pattern.  So
127 * combine the two sequential streams into a single strided stream.
128 *
129 * Returns whether co-linear streams were found.
130 */
131static boolean_t
132dmu_zfetch_colinear(zfetch_t *zf, zstream_t *zh)
133{
134	zstream_t	*z_walk;
135	zstream_t	*z_comp;
136
137	if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER))
138		return (0);
139
140	if (zh == NULL) {
141		rw_exit(&zf->zf_rwlock);
142		return (0);
143	}
144
145	for (z_walk = list_head(&zf->zf_stream); z_walk;
146	    z_walk = list_next(&zf->zf_stream, z_walk)) {
147		for (z_comp = list_next(&zf->zf_stream, z_walk); z_comp;
148		    z_comp = list_next(&zf->zf_stream, z_comp)) {
149			int64_t		diff;
150
151			if (z_walk->zst_len != z_walk->zst_stride ||
152			    z_comp->zst_len != z_comp->zst_stride) {
153				continue;
154			}
155
156			diff = z_comp->zst_offset - z_walk->zst_offset;
157			if (z_comp->zst_offset + diff == zh->zst_offset) {
158				z_walk->zst_offset = zh->zst_offset;
159				z_walk->zst_direction = diff < 0 ? -1 : 1;
160				z_walk->zst_stride =
161				    diff * z_walk->zst_direction;
162				z_walk->zst_ph_offset =
163				    zh->zst_offset + z_walk->zst_stride;
164				dmu_zfetch_stream_remove(zf, z_comp);
165				mutex_destroy(&z_comp->zst_lock);
166				kmem_free(z_comp, sizeof (zstream_t));
167
168				dmu_zfetch_dofetch(zf, z_walk);
169
170				rw_exit(&zf->zf_rwlock);
171				return (1);
172			}
173
174			diff = z_walk->zst_offset - z_comp->zst_offset;
175			if (z_walk->zst_offset + diff == zh->zst_offset) {
176				z_walk->zst_offset = zh->zst_offset;
177				z_walk->zst_direction = diff < 0 ? -1 : 1;
178				z_walk->zst_stride =
179				    diff * z_walk->zst_direction;
180				z_walk->zst_ph_offset =
181				    zh->zst_offset + z_walk->zst_stride;
182				dmu_zfetch_stream_remove(zf, z_comp);
183				mutex_destroy(&z_comp->zst_lock);
184				kmem_free(z_comp, sizeof (zstream_t));
185
186				dmu_zfetch_dofetch(zf, z_walk);
187
188				rw_exit(&zf->zf_rwlock);
189				return (1);
190			}
191		}
192	}
193
194	rw_exit(&zf->zf_rwlock);
195	return (0);
196}
197
198/*
199 * Given a zstream_t, determine the bounds of the prefetch.  Then call the
200 * routine that actually prefetches the individual blocks.
201 */
202static void
203dmu_zfetch_dofetch(zfetch_t *zf, zstream_t *zs)
204{
205	uint64_t	prefetch_tail;
206	uint64_t	prefetch_limit;
207	uint64_t	prefetch_ofst;
208	uint64_t	prefetch_len;
209	uint64_t	blocks_fetched;
210
211	zs->zst_stride = MAX((int64_t)zs->zst_stride, zs->zst_len);
212	zs->zst_cap = MIN(zfetch_block_cap, 2 * zs->zst_cap);
213
214	prefetch_tail = MAX((int64_t)zs->zst_ph_offset,
215	    (int64_t)(zs->zst_offset + zs->zst_stride));
216	/*
217	 * XXX: use a faster division method?
218	 */
219	prefetch_limit = zs->zst_offset + zs->zst_len +
220	    (zs->zst_cap * zs->zst_stride) / zs->zst_len;
221
222	while (prefetch_tail < prefetch_limit) {
223		prefetch_ofst = zs->zst_offset + zs->zst_direction *
224		    (prefetch_tail - zs->zst_offset);
225
226		prefetch_len = zs->zst_len;
227
228		/*
229		 * Don't prefetch beyond the end of the file, if working
230		 * backwards.
231		 */
232		if ((zs->zst_direction == ZFETCH_BACKWARD) &&
233		    (prefetch_ofst > prefetch_tail)) {
234			prefetch_len += prefetch_ofst;
235			prefetch_ofst = 0;
236		}
237
238		/* don't prefetch more than we're supposed to */
239		if (prefetch_len > zs->zst_len)
240			break;
241
242		blocks_fetched = dmu_zfetch_fetch(zf->zf_dnode,
243		    prefetch_ofst, zs->zst_len);
244
245		prefetch_tail += zs->zst_stride;
246		/* stop if we've run out of stuff to prefetch */
247		if (blocks_fetched < zs->zst_len)
248			break;
249	}
250	zs->zst_ph_offset = prefetch_tail;
251	zs->zst_last = ddi_get_lbolt();
252}
253
254void
255zfetch_init(void)
256{
257
258	zfetch_ksp = kstat_create("zfs", 0, "zfetchstats", "misc",
259	    KSTAT_TYPE_NAMED, sizeof (zfetch_stats) / sizeof (kstat_named_t),
260	    KSTAT_FLAG_VIRTUAL);
261
262	if (zfetch_ksp != NULL) {
263		zfetch_ksp->ks_data = &zfetch_stats;
264		kstat_install(zfetch_ksp);
265	}
266}
267
268void
269zfetch_fini(void)
270{
271	if (zfetch_ksp != NULL) {
272		kstat_delete(zfetch_ksp);
273		zfetch_ksp = NULL;
274	}
275}
276
277/*
278 * This takes a pointer to a zfetch structure and a dnode.  It performs the
279 * necessary setup for the zfetch structure, grokking data from the
280 * associated dnode.
281 */
282void
283dmu_zfetch_init(zfetch_t *zf, dnode_t *dno)
284{
285	if (zf == NULL) {
286		return;
287	}
288
289	zf->zf_dnode = dno;
290	zf->zf_stream_cnt = 0;
291	zf->zf_alloc_fail = 0;
292
293	list_create(&zf->zf_stream, sizeof (zstream_t),
294	    offsetof(zstream_t, zst_node));
295
296	rw_init(&zf->zf_rwlock, NULL, RW_DEFAULT, NULL);
297}
298
299/*
300 * This function computes the actual size, in blocks, that can be prefetched,
301 * and fetches it.
302 */
303static uint64_t
304dmu_zfetch_fetch(dnode_t *dn, uint64_t blkid, uint64_t nblks)
305{
306	uint64_t	fetchsz;
307	uint64_t	i;
308
309	fetchsz = dmu_zfetch_fetchsz(dn, blkid, nblks);
310
311	for (i = 0; i < fetchsz; i++) {
312		dbuf_prefetch(dn, 0, blkid + i, ZIO_PRIORITY_ASYNC_READ,
313		    ARC_FLAG_PREFETCH);
314	}
315
316	return (fetchsz);
317}
318
319/*
320 * this function returns the number of blocks that would be prefetched, based
321 * upon the supplied dnode, blockid, and nblks.  This is used so that we can
322 * update streams in place, and then prefetch with their old value after the
323 * fact.  This way, we can delay the prefetch, but subsequent accesses to the
324 * stream won't result in the same data being prefetched multiple times.
325 */
326static uint64_t
327dmu_zfetch_fetchsz(dnode_t *dn, uint64_t blkid, uint64_t nblks)
328{
329	uint64_t	fetchsz;
330
331	if (blkid > dn->dn_maxblkid) {
332		return (0);
333	}
334
335	/* compute fetch size */
336	if (blkid + nblks + 1 > dn->dn_maxblkid) {
337		fetchsz = (dn->dn_maxblkid - blkid) + 1;
338		ASSERT(blkid + fetchsz - 1 <= dn->dn_maxblkid);
339	} else {
340		fetchsz = nblks;
341	}
342
343
344	return (fetchsz);
345}
346
347/*
348 * given a zfetch and a zstream structure, see if there is an associated zstream
349 * for this block read.  If so, it starts a prefetch for the stream it
350 * located and returns true, otherwise it returns false
351 */
352static boolean_t
353dmu_zfetch_find(zfetch_t *zf, zstream_t *zh, int prefetched)
354{
355	zstream_t	*zs;
356	int64_t		diff;
357	int		reset = !prefetched;
358	int		rc = 0;
359
360	if (zh == NULL)
361		return (0);
362
363	/*
364	 * XXX: This locking strategy is a bit coarse; however, it's impact has
365	 * yet to be tested.  If this turns out to be an issue, it can be
366	 * modified in a number of different ways.
367	 */
368
369	rw_enter(&zf->zf_rwlock, RW_READER);
370top:
371
372	for (zs = list_head(&zf->zf_stream); zs;
373	    zs = list_next(&zf->zf_stream, zs)) {
374
375		/*
376		 * XXX - should this be an assert?
377		 */
378		if (zs->zst_len == 0) {
379			/* bogus stream */
380			ZFETCHSTAT_BUMP(zfetchstat_bogus_streams);
381			continue;
382		}
383
384		/*
385		 * We hit this case when we are in a strided prefetch stream:
386		 * we will read "len" blocks before "striding".
387		 */
388		if (zh->zst_offset >= zs->zst_offset &&
389		    zh->zst_offset < zs->zst_offset + zs->zst_len) {
390			if (prefetched) {
391				/* already fetched */
392				ZFETCHSTAT_BUMP(zfetchstat_stride_hits);
393				rc = 1;
394				goto out;
395			} else {
396				ZFETCHSTAT_BUMP(zfetchstat_stride_misses);
397			}
398		}
399
400		/*
401		 * This is the forward sequential read case: we increment
402		 * len by one each time we hit here, so we will enter this
403		 * case on every read.
404		 */
405		if (zh->zst_offset == zs->zst_offset + zs->zst_len) {
406
407			reset = !prefetched && zs->zst_len > 1;
408
409			if (mutex_tryenter(&zs->zst_lock) == 0) {
410				rc = 1;
411				goto out;
412			}
413
414			if (zh->zst_offset != zs->zst_offset + zs->zst_len) {
415				mutex_exit(&zs->zst_lock);
416				goto top;
417			}
418			zs->zst_len += zh->zst_len;
419			diff = zs->zst_len - zfetch_block_cap;
420			if (diff > 0) {
421				zs->zst_offset += diff;
422				zs->zst_len = zs->zst_len > diff ?
423				    zs->zst_len - diff : 0;
424			}
425			zs->zst_direction = ZFETCH_FORWARD;
426
427			break;
428
429		/*
430		 * Same as above, but reading backwards through the file.
431		 */
432		} else if (zh->zst_offset == zs->zst_offset - zh->zst_len) {
433			/* backwards sequential access */
434
435			reset = !prefetched && zs->zst_len > 1;
436
437			if (mutex_tryenter(&zs->zst_lock) == 0) {
438				rc = 1;
439				goto out;
440			}
441
442			if (zh->zst_offset != zs->zst_offset - zh->zst_len) {
443				mutex_exit(&zs->zst_lock);
444				goto top;
445			}
446
447			zs->zst_offset = zs->zst_offset > zh->zst_len ?
448			    zs->zst_offset - zh->zst_len : 0;
449			zs->zst_ph_offset = zs->zst_ph_offset > zh->zst_len ?
450			    zs->zst_ph_offset - zh->zst_len : 0;
451			zs->zst_len += zh->zst_len;
452
453			diff = zs->zst_len - zfetch_block_cap;
454			if (diff > 0) {
455				zs->zst_ph_offset = zs->zst_ph_offset > diff ?
456				    zs->zst_ph_offset - diff : 0;
457				zs->zst_len = zs->zst_len > diff ?
458				    zs->zst_len - diff : zs->zst_len;
459			}
460			zs->zst_direction = ZFETCH_BACKWARD;
461
462			break;
463
464		} else if ((zh->zst_offset - zs->zst_offset - zs->zst_stride <
465		    zs->zst_len) && (zs->zst_len != zs->zst_stride)) {
466			/* strided forward access */
467
468			if (mutex_tryenter(&zs->zst_lock) == 0) {
469				rc = 1;
470				goto out;
471			}
472
473			if ((zh->zst_offset - zs->zst_offset - zs->zst_stride >=
474			    zs->zst_len) || (zs->zst_len == zs->zst_stride)) {
475				mutex_exit(&zs->zst_lock);
476				goto top;
477			}
478
479			zs->zst_offset += zs->zst_stride;
480			zs->zst_direction = ZFETCH_FORWARD;
481
482			break;
483
484		} else if ((zh->zst_offset - zs->zst_offset + zs->zst_stride <
485		    zs->zst_len) && (zs->zst_len != zs->zst_stride)) {
486			/* strided reverse access */
487
488			if (mutex_tryenter(&zs->zst_lock) == 0) {
489				rc = 1;
490				goto out;
491			}
492
493			if ((zh->zst_offset - zs->zst_offset + zs->zst_stride >=
494			    zs->zst_len) || (zs->zst_len == zs->zst_stride)) {
495				mutex_exit(&zs->zst_lock);
496				goto top;
497			}
498
499			zs->zst_offset = zs->zst_offset > zs->zst_stride ?
500			    zs->zst_offset - zs->zst_stride : 0;
501			zs->zst_ph_offset = (zs->zst_ph_offset >
502			    (2 * zs->zst_stride)) ?
503			    (zs->zst_ph_offset - (2 * zs->zst_stride)) : 0;
504			zs->zst_direction = ZFETCH_BACKWARD;
505
506			break;
507		}
508	}
509
510	if (zs) {
511		if (reset) {
512			zstream_t *remove = zs;
513
514			ZFETCHSTAT_BUMP(zfetchstat_stream_resets);
515			rc = 0;
516			mutex_exit(&zs->zst_lock);
517			rw_exit(&zf->zf_rwlock);
518			rw_enter(&zf->zf_rwlock, RW_WRITER);
519			/*
520			 * Relocate the stream, in case someone removes
521			 * it while we were acquiring the WRITER lock.
522			 */
523			for (zs = list_head(&zf->zf_stream); zs;
524			    zs = list_next(&zf->zf_stream, zs)) {
525				if (zs == remove) {
526					dmu_zfetch_stream_remove(zf, zs);
527					mutex_destroy(&zs->zst_lock);
528					kmem_free(zs, sizeof (zstream_t));
529					break;
530				}
531			}
532		} else {
533			ZFETCHSTAT_BUMP(zfetchstat_stream_noresets);
534			rc = 1;
535			dmu_zfetch_dofetch(zf, zs);
536			mutex_exit(&zs->zst_lock);
537		}
538	}
539out:
540	rw_exit(&zf->zf_rwlock);
541	return (rc);
542}
543
544/*
545 * Clean-up state associated with a zfetch structure.  This frees allocated
546 * structure members, empties the zf_stream tree, and generally makes things
547 * nice.  This doesn't free the zfetch_t itself, that's left to the caller.
548 */
549void
550dmu_zfetch_rele(zfetch_t *zf)
551{
552	zstream_t	*zs;
553	zstream_t	*zs_next;
554
555	ASSERT(!RW_LOCK_HELD(&zf->zf_rwlock));
556
557	for (zs = list_head(&zf->zf_stream); zs; zs = zs_next) {
558		zs_next = list_next(&zf->zf_stream, zs);
559
560		list_remove(&zf->zf_stream, zs);
561		mutex_destroy(&zs->zst_lock);
562		kmem_free(zs, sizeof (zstream_t));
563	}
564	list_destroy(&zf->zf_stream);
565	rw_destroy(&zf->zf_rwlock);
566
567	zf->zf_dnode = NULL;
568}
569
570/*
571 * Given a zfetch and zstream structure, insert the zstream structure into the
572 * AVL tree contained within the zfetch structure.  Peform the appropriate
573 * book-keeping.  It is possible that another thread has inserted a stream which
574 * matches one that we are about to insert, so we must be sure to check for this
575 * case.  If one is found, return failure, and let the caller cleanup the
576 * duplicates.
577 */
578static int
579dmu_zfetch_stream_insert(zfetch_t *zf, zstream_t *zs)
580{
581	zstream_t	*zs_walk;
582	zstream_t	*zs_next;
583
584	ASSERT(RW_WRITE_HELD(&zf->zf_rwlock));
585
586	for (zs_walk = list_head(&zf->zf_stream); zs_walk; zs_walk = zs_next) {
587		zs_next = list_next(&zf->zf_stream, zs_walk);
588
589		if (dmu_zfetch_streams_equal(zs_walk, zs)) {
590			return (0);
591		}
592	}
593
594	list_insert_head(&zf->zf_stream, zs);
595	zf->zf_stream_cnt++;
596	return (1);
597}
598
599
600/*
601 * Walk the list of zstreams in the given zfetch, find an old one (by time), and
602 * reclaim it for use by the caller.
603 */
604static zstream_t *
605dmu_zfetch_stream_reclaim(zfetch_t *zf)
606{
607	zstream_t	*zs;
608	clock_t		ticks;
609
610	ticks = zfetch_min_sec_reap * hz;
611	if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER))
612		return (0);
613
614	for (zs = list_head(&zf->zf_stream); zs;
615	    zs = list_next(&zf->zf_stream, zs)) {
616
617		if (ddi_get_lbolt() - zs->zst_last > ticks)
618			break;
619	}
620
621	if (zs) {
622		dmu_zfetch_stream_remove(zf, zs);
623		mutex_destroy(&zs->zst_lock);
624		bzero(zs, sizeof (zstream_t));
625	} else {
626		zf->zf_alloc_fail++;
627	}
628	rw_exit(&zf->zf_rwlock);
629
630	return (zs);
631}
632
633/*
634 * Given a zfetch and zstream structure, remove the zstream structure from its
635 * container in the zfetch structure.  Perform the appropriate book-keeping.
636 */
637static void
638dmu_zfetch_stream_remove(zfetch_t *zf, zstream_t *zs)
639{
640	ASSERT(RW_WRITE_HELD(&zf->zf_rwlock));
641
642	list_remove(&zf->zf_stream, zs);
643	zf->zf_stream_cnt--;
644}
645
646static int
647dmu_zfetch_streams_equal(zstream_t *zs1, zstream_t *zs2)
648{
649	if (zs1->zst_offset != zs2->zst_offset)
650		return (0);
651
652	if (zs1->zst_len != zs2->zst_len)
653		return (0);
654
655	if (zs1->zst_stride != zs2->zst_stride)
656		return (0);
657
658	if (zs1->zst_ph_offset != zs2->zst_ph_offset)
659		return (0);
660
661	if (zs1->zst_cap != zs2->zst_cap)
662		return (0);
663
664	if (zs1->zst_direction != zs2->zst_direction)
665		return (0);
666
667	return (1);
668}
669
670/*
671 * This is the prefetch entry point.  It calls all of the other dmu_zfetch
672 * routines to create, delete, find, or operate upon prefetch streams.
673 */
674void
675dmu_zfetch(zfetch_t *zf, uint64_t offset, uint64_t size, int prefetched)
676{
677	zstream_t	zst;
678	zstream_t	*newstream;
679	boolean_t	fetched;
680	int		inserted;
681	unsigned int	blkshft;
682	uint64_t	blksz;
683
684	if (zfs_prefetch_disable)
685		return;
686
687	/* files that aren't ln2 blocksz are only one block -- nothing to do */
688	if (!zf->zf_dnode->dn_datablkshift)
689		return;
690
691	/* convert offset and size, into blockid and nblocks */
692	blkshft = zf->zf_dnode->dn_datablkshift;
693	blksz = (1 << blkshft);
694
695	bzero(&zst, sizeof (zstream_t));
696	zst.zst_offset = offset >> blkshft;
697	zst.zst_len = (P2ROUNDUP(offset + size, blksz) -
698	    P2ALIGN(offset, blksz)) >> blkshft;
699
700	fetched = dmu_zfetch_find(zf, &zst, prefetched);
701	if (fetched) {
702		ZFETCHSTAT_BUMP(zfetchstat_hits);
703	} else {
704		ZFETCHSTAT_BUMP(zfetchstat_misses);
705		fetched = dmu_zfetch_colinear(zf, &zst);
706		if (fetched) {
707			ZFETCHSTAT_BUMP(zfetchstat_colinear_hits);
708		} else {
709			ZFETCHSTAT_BUMP(zfetchstat_colinear_misses);
710		}
711	}
712
713	if (!fetched) {
714		newstream = dmu_zfetch_stream_reclaim(zf);
715
716		/*
717		 * we still couldn't find a stream, drop the lock, and allocate
718		 * one if possible.  Otherwise, give up and go home.
719		 */
720		if (newstream) {
721			ZFETCHSTAT_BUMP(zfetchstat_reclaim_successes);
722		} else {
723			uint64_t	maxblocks;
724			uint32_t	max_streams;
725			uint32_t	cur_streams;
726
727			ZFETCHSTAT_BUMP(zfetchstat_reclaim_failures);
728			cur_streams = zf->zf_stream_cnt;
729			maxblocks = zf->zf_dnode->dn_maxblkid;
730
731			max_streams = MIN(zfetch_max_streams,
732			    (maxblocks / zfetch_block_cap));
733			if (max_streams == 0) {
734				max_streams++;
735			}
736
737			if (cur_streams >= max_streams) {
738				return;
739			}
740			newstream = kmem_zalloc(sizeof (zstream_t), KM_SLEEP);
741		}
742
743		newstream->zst_offset = zst.zst_offset;
744		newstream->zst_len = zst.zst_len;
745		newstream->zst_stride = zst.zst_len;
746		newstream->zst_ph_offset = zst.zst_len + zst.zst_offset;
747		newstream->zst_cap = zst.zst_len;
748		newstream->zst_direction = ZFETCH_FORWARD;
749		newstream->zst_last = ddi_get_lbolt();
750
751		mutex_init(&newstream->zst_lock, NULL, MUTEX_DEFAULT, NULL);
752
753		rw_enter(&zf->zf_rwlock, RW_WRITER);
754		inserted = dmu_zfetch_stream_insert(zf, newstream);
755		rw_exit(&zf->zf_rwlock);
756
757		if (!inserted) {
758			mutex_destroy(&newstream->zst_lock);
759			kmem_free(newstream, sizeof (zstream_t));
760		}
761	}
762}
763