dmu_zfetch.c revision 2391:2fa3fd1db808
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 2006 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28#include <sys/zfs_context.h>
29#include <sys/dnode.h>
30#include <sys/dmu_objset.h>
31#include <sys/dmu_zfetch.h>
32#include <sys/dmu.h>
33#include <sys/dbuf.h>
34
35/*
36 * I'm against tune-ables, but these should probably exist as tweakable globals
37 * until we can get this working the way we want it to.
38 */
39
40/* max # of streams per zfetch */
41uint32_t	zfetch_max_streams = 8;
42/* min time before stream reclaim */
43uint32_t	zfetch_min_sec_reap = 2;
44/* max number of blocks to fetch at a time */
45uint32_t	zfetch_block_cap = 256;
46/* number of bytes in a array_read at which we stop prefetching (1Mb) */
47uint64_t	zfetch_array_rd_sz = 1024 * 1024;
48
49/* forward decls for static routines */
50static int		dmu_zfetch_colinear(zfetch_t *, zstream_t *);
51static void		dmu_zfetch_dofetch(zfetch_t *, zstream_t *);
52static uint64_t		dmu_zfetch_fetch(dnode_t *, uint64_t, uint64_t);
53static uint64_t		dmu_zfetch_fetchsz(dnode_t *, uint64_t, uint64_t);
54static int		dmu_zfetch_find(zfetch_t *, zstream_t *, int);
55static int		dmu_zfetch_stream_insert(zfetch_t *, zstream_t *);
56static zstream_t	*dmu_zfetch_stream_reclaim(zfetch_t *);
57static void		dmu_zfetch_stream_remove(zfetch_t *, zstream_t *);
58static void		dmu_zfetch_stream_update(zfetch_t *, zstream_t *);
59static int		dmu_zfetch_streams_equal(zstream_t *, zstream_t *);
60
61/*
62 * Given a zfetch structure and a zstream structure, determine whether the
63 * blocks to be read are part of a co-linear pair of existing prefetch
64 * streams.  If a set is found, coalesce the streams, removing one, and
65 * configure the prefetch so it looks for a strided access pattern.
66 *
67 * In other words: if we find two sequential access streams that are
68 * the same length and distance N appart, and this read is N from the
69 * last stream, then we are probably in a strided access pattern.  So
70 * combine the two sequential streams into a single strided stream.
71 *
72 * If no co-linear streams are found, return NULL.
73 */
74static int
75dmu_zfetch_colinear(zfetch_t *zf, zstream_t *zh)
76{
77	zstream_t	*z_walk;
78	zstream_t	*z_comp;
79
80	if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER))
81		return (0);
82
83	if (zh == NULL) {
84		rw_exit(&zf->zf_rwlock);
85		return (0);
86	}
87
88	for (z_walk = list_head(&zf->zf_stream); z_walk;
89	    z_walk = list_next(&zf->zf_stream, z_walk)) {
90		for (z_comp = list_next(&zf->zf_stream, z_walk); z_comp;
91		    z_comp = list_next(&zf->zf_stream, z_comp)) {
92			int64_t		diff;
93
94			if (z_walk->zst_len != z_walk->zst_stride ||
95			    z_comp->zst_len != z_comp->zst_stride) {
96				continue;
97			}
98
99			diff = z_comp->zst_offset - z_walk->zst_offset;
100			if (z_comp->zst_offset + diff == zh->zst_offset) {
101				z_walk->zst_offset = zh->zst_offset;
102				z_walk->zst_direction = diff < 0 ? -1 : 1;
103				z_walk->zst_stride =
104				    diff * z_walk->zst_direction;
105				z_walk->zst_ph_offset =
106				    zh->zst_offset + z_walk->zst_stride;
107				dmu_zfetch_stream_remove(zf, z_comp);
108				mutex_destroy(&z_comp->zst_lock);
109				kmem_free(z_comp, sizeof (zstream_t));
110
111				dmu_zfetch_dofetch(zf, z_walk);
112
113				rw_exit(&zf->zf_rwlock);
114				return (1);
115			}
116
117			diff = z_walk->zst_offset - z_comp->zst_offset;
118			if (z_walk->zst_offset + diff == zh->zst_offset) {
119				z_walk->zst_offset = zh->zst_offset;
120				z_walk->zst_direction = diff < 0 ? -1 : 1;
121				z_walk->zst_stride =
122				    diff * z_walk->zst_direction;
123				z_walk->zst_ph_offset =
124				    zh->zst_offset + z_walk->zst_stride;
125				dmu_zfetch_stream_remove(zf, z_comp);
126				mutex_destroy(&z_comp->zst_lock);
127				kmem_free(z_comp, sizeof (zstream_t));
128
129				dmu_zfetch_dofetch(zf, z_walk);
130
131				rw_exit(&zf->zf_rwlock);
132				return (1);
133			}
134		}
135	}
136
137	rw_exit(&zf->zf_rwlock);
138	return (0);
139}
140
141/*
142 * Given a zstream_t, determine the bounds of the prefetch.  Then call the
143 * routine that actually prefetches the individual blocks.
144 */
145static void
146dmu_zfetch_dofetch(zfetch_t *zf, zstream_t *zs)
147{
148	uint64_t	prefetch_tail;
149	uint64_t	prefetch_limit;
150	uint64_t	prefetch_ofst;
151	uint64_t	prefetch_len;
152	uint64_t	blocks_fetched;
153
154	zs->zst_stride = MAX((int64_t)zs->zst_stride, zs->zst_len);
155	zs->zst_cap = MIN(zfetch_block_cap, 2 * zs->zst_cap);
156
157	prefetch_tail = MAX((int64_t)zs->zst_ph_offset,
158	    (int64_t)(zs->zst_offset + zs->zst_stride));
159	/*
160	 * XXX: use a faster division method?
161	 */
162	prefetch_limit = zs->zst_offset + zs->zst_len +
163	    (zs->zst_cap * zs->zst_stride) / zs->zst_len;
164
165	while (prefetch_tail < prefetch_limit) {
166		prefetch_ofst = zs->zst_offset + zs->zst_direction *
167		    (prefetch_tail - zs->zst_offset);
168
169		prefetch_len = zs->zst_len;
170
171		/*
172		 * Don't prefetch beyond the end of the file, if working
173		 * backwards.
174		 */
175		if ((zs->zst_direction == ZFETCH_BACKWARD) &&
176		    (prefetch_ofst > prefetch_tail)) {
177			prefetch_len += prefetch_ofst;
178			prefetch_ofst = 0;
179		}
180
181		/* don't prefetch more than we're supposed to */
182		if (prefetch_len > zs->zst_len)
183			break;
184
185		blocks_fetched = dmu_zfetch_fetch(zf->zf_dnode,
186		    prefetch_ofst, zs->zst_len);
187
188		prefetch_tail += zs->zst_stride;
189		/* stop if we've run out of stuff to prefetch */
190		if (blocks_fetched < zs->zst_len)
191			break;
192	}
193	zs->zst_ph_offset = prefetch_tail;
194	zs->zst_last = lbolt;
195}
196
197/*
198 * This takes a pointer to a zfetch structure and a dnode.  It performs the
199 * necessary setup for the zfetch structure, grokking data from the
200 * associated dnode.
201 */
202void
203dmu_zfetch_init(zfetch_t *zf, dnode_t *dno)
204{
205	if (zf == NULL) {
206		return;
207	}
208
209	zf->zf_dnode = dno;
210	zf->zf_stream_cnt = 0;
211	zf->zf_alloc_fail = 0;
212
213	list_create(&zf->zf_stream, sizeof (zstream_t),
214	    offsetof(zstream_t, zst_node));
215
216	rw_init(&zf->zf_rwlock, NULL, RW_DEFAULT, NULL);
217}
218
219/*
220 * This function computes the actual size, in blocks, that can be prefetched,
221 * and fetches it.
222 */
223static uint64_t
224dmu_zfetch_fetch(dnode_t *dn, uint64_t blkid, uint64_t nblks)
225{
226	uint64_t	fetchsz;
227	uint64_t	i;
228
229	fetchsz = dmu_zfetch_fetchsz(dn, blkid, nblks);
230
231	for (i = 0; i < fetchsz; i++) {
232		dbuf_prefetch(dn, blkid + i);
233	}
234
235	return (fetchsz);
236}
237
238/*
239 * this function returns the number of blocks that would be prefetched, based
240 * upon the supplied dnode, blockid, and nblks.  This is used so that we can
241 * update streams in place, and then prefetch with their old value after the
242 * fact.  This way, we can delay the prefetch, but subsequent accesses to the
243 * stream won't result in the same data being prefetched multiple times.
244 */
245static uint64_t
246dmu_zfetch_fetchsz(dnode_t *dn, uint64_t blkid, uint64_t nblks)
247{
248	uint64_t	fetchsz;
249
250	if (blkid > dn->dn_maxblkid) {
251		return (0);
252	}
253
254	/* compute fetch size */
255	if (blkid + nblks + 1 > dn->dn_maxblkid) {
256		fetchsz = (dn->dn_maxblkid - blkid) + 1;
257		ASSERT(blkid + fetchsz - 1 <= dn->dn_maxblkid);
258	} else {
259		fetchsz = nblks;
260	}
261
262
263	return (fetchsz);
264}
265
266/*
267 * given a zfetch and a zsearch structure, see if there is an associated zstream
268 * for this block read.  If so, it starts a prefetch for the stream it
269 * located and returns true, otherwise it returns false
270 */
271static int
272dmu_zfetch_find(zfetch_t *zf, zstream_t *zh, int prefetched)
273{
274	zstream_t	*zs;
275	int64_t		diff;
276	int		reset = !prefetched;
277	int		rc = 0;
278
279	if (zh == NULL)
280		return (0);
281
282	/*
283	 * XXX: This locking strategy is a bit coarse; however, it's impact has
284	 * yet to be tested.  If this turns out to be an issue, it can be
285	 * modified in a number of different ways.
286	 */
287
288	rw_enter(&zf->zf_rwlock, RW_READER);
289top:
290
291	for (zs = list_head(&zf->zf_stream); zs;
292	    zs = list_next(&zf->zf_stream, zs)) {
293
294		/*
295		 * XXX - should this be an assert?
296		 */
297		if (zs->zst_len == 0) {
298			/* bogus stream */
299			continue;
300		}
301
302		/*
303		 * We hit this case when we are in a strided prefetch stream:
304		 * we will read "len" blocks before "striding".
305		 */
306		if (zh->zst_offset >= zs->zst_offset &&
307		    zh->zst_offset < zs->zst_offset + zs->zst_len) {
308			/* already fetched */
309			rc = 1;
310			goto out;
311		}
312
313		/*
314		 * This is the forward sequential read case: we increment
315		 * len by one each time we hit here, so we will enter this
316		 * case on every read.
317		 */
318		if (zh->zst_offset == zs->zst_offset + zs->zst_len) {
319
320			reset = !prefetched && zs->zst_len > 1;
321
322			mutex_enter(&zs->zst_lock);
323
324			if (zh->zst_offset != zs->zst_offset + zs->zst_len) {
325				mutex_exit(&zs->zst_lock);
326				goto top;
327			}
328			zs->zst_len += zh->zst_len;
329			diff = zs->zst_len - zfetch_block_cap;
330			if (diff > 0) {
331				zs->zst_offset += diff;
332				zs->zst_len = zs->zst_len > diff ?
333				    zs->zst_len - diff : 0;
334			}
335			zs->zst_direction = ZFETCH_FORWARD;
336
337			break;
338
339		/*
340		 * Same as above, but reading backwards through the file.
341		 */
342		} else if (zh->zst_offset == zs->zst_offset - zh->zst_len) {
343			/* backwards sequential access */
344
345			reset = !prefetched && zs->zst_len > 1;
346
347			mutex_enter(&zs->zst_lock);
348
349			if (zh->zst_offset != zs->zst_offset - zh->zst_len) {
350				mutex_exit(&zs->zst_lock);
351				goto top;
352			}
353
354			zs->zst_offset = zs->zst_offset > zh->zst_len ?
355			    zs->zst_offset - zh->zst_len : 0;
356			zs->zst_ph_offset = zs->zst_ph_offset > zh->zst_len ?
357			    zs->zst_ph_offset - zh->zst_len : 0;
358			zs->zst_len += zh->zst_len;
359
360			diff = zs->zst_len - zfetch_block_cap;
361			if (diff > 0) {
362				zs->zst_ph_offset = zs->zst_ph_offset > diff ?
363				    zs->zst_ph_offset - diff : 0;
364				zs->zst_len = zs->zst_len > diff ?
365				    zs->zst_len - diff : zs->zst_len;
366			}
367			zs->zst_direction = ZFETCH_BACKWARD;
368
369			break;
370
371		} else if ((zh->zst_offset - zs->zst_offset - zs->zst_stride <
372		    zs->zst_len) && (zs->zst_len != zs->zst_stride)) {
373			/* strided forward access */
374
375			mutex_enter(&zs->zst_lock);
376
377			if ((zh->zst_offset - zs->zst_offset - zs->zst_stride >=
378			    zs->zst_len) || (zs->zst_len == zs->zst_stride)) {
379				mutex_exit(&zs->zst_lock);
380				goto top;
381			}
382
383			zs->zst_offset += zs->zst_stride;
384			zs->zst_direction = ZFETCH_FORWARD;
385
386			break;
387
388		} else if ((zh->zst_offset - zs->zst_offset + zs->zst_stride <
389		    zs->zst_len) && (zs->zst_len != zs->zst_stride)) {
390			/* strided reverse access */
391
392			mutex_enter(&zs->zst_lock);
393
394			if ((zh->zst_offset - zs->zst_offset + zs->zst_stride >=
395			    zs->zst_len) || (zs->zst_len == zs->zst_stride)) {
396				mutex_exit(&zs->zst_lock);
397				goto top;
398			}
399
400			zs->zst_offset = zs->zst_offset > zs->zst_stride ?
401			    zs->zst_offset - zs->zst_stride : 0;
402			zs->zst_ph_offset = (zs->zst_ph_offset >
403			    (2 * zs->zst_stride)) ?
404			    (zs->zst_ph_offset - (2 * zs->zst_stride)) : 0;
405			zs->zst_direction = ZFETCH_BACKWARD;
406
407			break;
408		}
409	}
410
411	if (zs) {
412		if (reset) {
413			zstream_t *remove = zs;
414
415			rc = 0;
416			mutex_exit(&zs->zst_lock);
417			rw_exit(&zf->zf_rwlock);
418			rw_enter(&zf->zf_rwlock, RW_WRITER);
419			/*
420			 * Relocate the stream, in case someone removes
421			 * it while we were acquiring the WRITER lock.
422			 */
423			for (zs = list_head(&zf->zf_stream); zs;
424			    zs = list_next(&zf->zf_stream, zs)) {
425				if (zs == remove) {
426					dmu_zfetch_stream_remove(zf, zs);
427					mutex_destroy(&zs->zst_lock);
428					kmem_free(zs, sizeof (zstream_t));
429					break;
430				}
431			}
432		} else {
433			rc = 1;
434			dmu_zfetch_dofetch(zf, zs);
435			mutex_exit(&zs->zst_lock);
436		}
437	}
438out:
439	rw_exit(&zf->zf_rwlock);
440	return (rc);
441}
442
443/*
444 * Clean-up state associated with a zfetch structure.  This frees allocated
445 * structure members, empties the zf_stream tree, and generally makes things
446 * nice.  This doesn't free the zfetch_t itself, that's left to the caller.
447 */
448void
449dmu_zfetch_rele(zfetch_t *zf)
450{
451	zstream_t	*zs;
452	zstream_t	*zs_next;
453
454	ASSERT(!RW_LOCK_HELD(&zf->zf_rwlock));
455
456	for (zs = list_head(&zf->zf_stream); zs; zs = zs_next) {
457		zs_next = list_next(&zf->zf_stream, zs);
458
459		list_remove(&zf->zf_stream, zs);
460		mutex_destroy(&zs->zst_lock);
461		kmem_free(zs, sizeof (zstream_t));
462	}
463	list_destroy(&zf->zf_stream);
464	rw_destroy(&zf->zf_rwlock);
465
466	zf->zf_dnode = NULL;
467}
468
469/*
470 * Given a zfetch and zstream structure, insert the zstream structure into the
471 * AVL tree contained within the zfetch structure.  Peform the appropriate
472 * book-keeping.  It is possible that another thread has inserted a stream which
473 * matches one that we are about to insert, so we must be sure to check for this
474 * case.  If one is found, return failure, and let the caller cleanup the
475 * duplicates.
476 */
477static int
478dmu_zfetch_stream_insert(zfetch_t *zf, zstream_t *zs)
479{
480	zstream_t	*zs_walk;
481	zstream_t	*zs_next;
482
483	ASSERT(RW_WRITE_HELD(&zf->zf_rwlock));
484
485	for (zs_walk = list_head(&zf->zf_stream); zs_walk; zs_walk = zs_next) {
486		zs_next = list_next(&zf->zf_stream, zs_walk);
487
488		if (dmu_zfetch_streams_equal(zs_walk, zs)) {
489		    return (0);
490		}
491	}
492
493	list_insert_head(&zf->zf_stream, zs);
494	zf->zf_stream_cnt++;
495
496	return (1);
497}
498
499
500/*
501 * Walk the list of zstreams in the given zfetch, find an old one (by time), and
502 * reclaim it for use by the caller.
503 */
504static zstream_t *
505dmu_zfetch_stream_reclaim(zfetch_t *zf)
506{
507	zstream_t	*zs;
508
509	if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER))
510		return (0);
511
512	for (zs = list_head(&zf->zf_stream); zs;
513	    zs = list_next(&zf->zf_stream, zs)) {
514
515		if (((lbolt - zs->zst_last) / hz) > zfetch_min_sec_reap)
516			break;
517	}
518
519	if (zs) {
520		dmu_zfetch_stream_remove(zf, zs);
521		mutex_destroy(&zs->zst_lock);
522		bzero(zs, sizeof (zstream_t));
523	} else {
524		zf->zf_alloc_fail++;
525	}
526	rw_exit(&zf->zf_rwlock);
527
528	return (zs);
529}
530
531/*
532 * Given a zfetch and zstream structure, remove the zstream structure from its
533 * container in the zfetch structure.  Perform the appropriate book-keeping.
534 */
535static void
536dmu_zfetch_stream_remove(zfetch_t *zf, zstream_t *zs)
537{
538	ASSERT(RW_WRITE_HELD(&zf->zf_rwlock));
539
540	list_remove(&zf->zf_stream, zs);
541	zf->zf_stream_cnt--;
542}
543
544static int
545dmu_zfetch_streams_equal(zstream_t *zs1, zstream_t *zs2)
546{
547	if (zs1->zst_offset != zs2->zst_offset)
548		return (0);
549
550	if (zs1->zst_len != zs2->zst_len)
551		return (0);
552
553	if (zs1->zst_stride != zs2->zst_stride)
554		return (0);
555
556	if (zs1->zst_ph_offset != zs2->zst_ph_offset)
557		return (0);
558
559	if (zs1->zst_cap != zs2->zst_cap)
560		return (0);
561
562	if (zs1->zst_direction != zs2->zst_direction)
563		return (0);
564
565	return (1);
566}
567
568/*
569 * This is the prefetch entry point.  It calls all of the other dmu_zfetch
570 * routines to create, delete, find, or operate upon prefetch streams.
571 */
572void
573dmu_zfetch(zfetch_t *zf, uint64_t offset, uint64_t size, int prefetched)
574{
575	zstream_t	zst;
576	zstream_t	*newstream;
577	int		fetched;
578	int		inserted;
579	unsigned int	blkshft;
580	uint64_t	blksz;
581
582	/* files that aren't ln2 blocksz are only one block -- nothing to do */
583	if (!zf->zf_dnode->dn_datablkshift) {
584		return;
585	}
586
587	/* convert offset and size, into blockid and nblocks */
588	blkshft = zf->zf_dnode->dn_datablkshift;
589	blksz = (1 << blkshft);
590
591	bzero(&zst, sizeof (zstream_t));
592	zst.zst_offset = offset >> blkshft;
593	zst.zst_len = (P2ROUNDUP(offset + size, blksz) -
594	    P2ALIGN(offset, blksz)) >> blkshft;
595
596	fetched = dmu_zfetch_find(zf, &zst, prefetched);
597	if (!fetched) {
598		fetched = dmu_zfetch_colinear(zf, &zst);
599	}
600
601	if (!fetched) {
602		newstream = dmu_zfetch_stream_reclaim(zf);
603
604		/*
605		 * we still couldn't find a stream, drop the lock, and allocate
606		 * one if possible.  Otherwise, give up and go home.
607		 */
608		if (newstream == NULL) {
609			uint64_t	maxblocks;
610			uint32_t	max_streams;
611			uint32_t	cur_streams;
612
613			cur_streams = zf->zf_stream_cnt;
614			maxblocks = zf->zf_dnode->dn_maxblkid;
615
616			max_streams = MIN(zfetch_max_streams,
617			    (maxblocks / zfetch_block_cap));
618			if (max_streams == 0) {
619				max_streams++;
620			}
621
622			if (cur_streams >= max_streams) {
623				return;
624			}
625
626			newstream = kmem_zalloc(sizeof (zstream_t), KM_SLEEP);
627		}
628
629		newstream->zst_offset = zst.zst_offset;
630		newstream->zst_len = zst.zst_len;
631		newstream->zst_stride = zst.zst_len;
632		newstream->zst_ph_offset = zst.zst_len + zst.zst_offset;
633		newstream->zst_cap = zst.zst_len;
634		newstream->zst_direction = ZFETCH_FORWARD;
635		newstream->zst_last = lbolt;
636
637		mutex_init(&newstream->zst_lock, NULL, MUTEX_DEFAULT, NULL);
638
639		rw_enter(&zf->zf_rwlock, RW_WRITER);
640		inserted = dmu_zfetch_stream_insert(zf, newstream);
641		rw_exit(&zf->zf_rwlock);
642
643		if (!inserted) {
644			mutex_destroy(&newstream->zst_lock);
645			kmem_free(newstream, sizeof (zstream_t));
646		}
647	}
648}
649