]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright 2009 Sun Microsystems, Inc. All rights reserved. |
34dc7c2f BB |
23 | * Use is subject to license terms. |
24 | */ | |
25 | ||
a08ee875 LG |
26 | /* |
27 | * Copyright (c) 2013 by Delphix. All rights reserved. | |
28 | */ | |
29 | ||
34dc7c2f BB |
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> | |
428870ff | 36 | #include <sys/kstat.h> |
34dc7c2f BB |
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 | ||
43 | int zfs_prefetch_disable = 0; | |
44 | ||
45 | /* max # of streams per zfetch */ | |
c409e464 | 46 | unsigned int zfetch_max_streams = 8; |
34dc7c2f | 47 | /* min time before stream reclaim */ |
c409e464 | 48 | unsigned int zfetch_min_sec_reap = 2; |
34dc7c2f | 49 | /* max number of blocks to fetch at a time */ |
c409e464 | 50 | unsigned int zfetch_block_cap = 256; |
34dc7c2f | 51 | /* number of bytes in a array_read at which we stop prefetching (1Mb) */ |
c409e464 | 52 | unsigned long zfetch_array_rd_sz = 1024 * 1024; |
34dc7c2f BB |
53 | |
54 | /* forward decls for static routines */ | |
a08ee875 | 55 | static boolean_t dmu_zfetch_colinear(zfetch_t *, zstream_t *); |
34dc7c2f BB |
56 | static void dmu_zfetch_dofetch(zfetch_t *, zstream_t *); |
57 | static uint64_t dmu_zfetch_fetch(dnode_t *, uint64_t, uint64_t); | |
58 | static uint64_t dmu_zfetch_fetchsz(dnode_t *, uint64_t, uint64_t); | |
a08ee875 | 59 | static boolean_t dmu_zfetch_find(zfetch_t *, zstream_t *, int); |
34dc7c2f BB |
60 | static int dmu_zfetch_stream_insert(zfetch_t *, zstream_t *); |
61 | static zstream_t *dmu_zfetch_stream_reclaim(zfetch_t *); | |
62 | static void dmu_zfetch_stream_remove(zfetch_t *, zstream_t *); | |
63 | static int dmu_zfetch_streams_equal(zstream_t *, zstream_t *); | |
64 | ||
428870ff BB |
65 | typedef struct zfetch_stats { |
66 | kstat_named_t zfetchstat_hits; | |
67 | kstat_named_t zfetchstat_misses; | |
68 | kstat_named_t zfetchstat_colinear_hits; | |
69 | kstat_named_t zfetchstat_colinear_misses; | |
70 | kstat_named_t zfetchstat_stride_hits; | |
71 | kstat_named_t zfetchstat_stride_misses; | |
72 | kstat_named_t zfetchstat_reclaim_successes; | |
73 | kstat_named_t zfetchstat_reclaim_failures; | |
74 | kstat_named_t zfetchstat_stream_resets; | |
75 | kstat_named_t zfetchstat_stream_noresets; | |
76 | kstat_named_t zfetchstat_bogus_streams; | |
77 | } zfetch_stats_t; | |
78 | ||
79 | static zfetch_stats_t zfetch_stats = { | |
80 | { "hits", KSTAT_DATA_UINT64 }, | |
81 | { "misses", KSTAT_DATA_UINT64 }, | |
82 | { "colinear_hits", KSTAT_DATA_UINT64 }, | |
83 | { "colinear_misses", KSTAT_DATA_UINT64 }, | |
84 | { "stride_hits", KSTAT_DATA_UINT64 }, | |
85 | { "stride_misses", KSTAT_DATA_UINT64 }, | |
86 | { "reclaim_successes", KSTAT_DATA_UINT64 }, | |
87 | { "reclaim_failures", KSTAT_DATA_UINT64 }, | |
88 | { "streams_resets", KSTAT_DATA_UINT64 }, | |
89 | { "streams_noresets", KSTAT_DATA_UINT64 }, | |
90 | { "bogus_streams", KSTAT_DATA_UINT64 }, | |
91 | }; | |
92 | ||
93 | #define ZFETCHSTAT_INCR(stat, val) \ | |
94 | atomic_add_64(&zfetch_stats.stat.value.ui64, (val)); | |
95 | ||
96 | #define ZFETCHSTAT_BUMP(stat) ZFETCHSTAT_INCR(stat, 1); | |
97 | ||
98 | kstat_t *zfetch_ksp; | |
99 | ||
34dc7c2f BB |
100 | /* |
101 | * Given a zfetch structure and a zstream structure, determine whether the | |
102 | * blocks to be read are part of a co-linear pair of existing prefetch | |
103 | * streams. If a set is found, coalesce the streams, removing one, and | |
104 | * configure the prefetch so it looks for a strided access pattern. | |
105 | * | |
106 | * In other words: if we find two sequential access streams that are | |
107 | * the same length and distance N appart, and this read is N from the | |
108 | * last stream, then we are probably in a strided access pattern. So | |
109 | * combine the two sequential streams into a single strided stream. | |
110 | * | |
a08ee875 | 111 | * Returns whether co-linear streams were found. |
34dc7c2f | 112 | */ |
a08ee875 | 113 | static boolean_t |
34dc7c2f BB |
114 | dmu_zfetch_colinear(zfetch_t *zf, zstream_t *zh) |
115 | { | |
116 | zstream_t *z_walk; | |
117 | zstream_t *z_comp; | |
118 | ||
119 | if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER)) | |
120 | return (0); | |
121 | ||
122 | if (zh == NULL) { | |
123 | rw_exit(&zf->zf_rwlock); | |
124 | return (0); | |
125 | } | |
126 | ||
127 | for (z_walk = list_head(&zf->zf_stream); z_walk; | |
128 | z_walk = list_next(&zf->zf_stream, z_walk)) { | |
129 | for (z_comp = list_next(&zf->zf_stream, z_walk); z_comp; | |
130 | z_comp = list_next(&zf->zf_stream, z_comp)) { | |
131 | int64_t diff; | |
132 | ||
133 | if (z_walk->zst_len != z_walk->zst_stride || | |
134 | z_comp->zst_len != z_comp->zst_stride) { | |
135 | continue; | |
136 | } | |
137 | ||
138 | diff = z_comp->zst_offset - z_walk->zst_offset; | |
139 | if (z_comp->zst_offset + diff == zh->zst_offset) { | |
140 | z_walk->zst_offset = zh->zst_offset; | |
a08ee875 LG |
141 | z_walk->zst_direction = diff < 0 ? |
142 | ZFETCH_BACKWARD : ZFETCH_FORWARD; | |
34dc7c2f BB |
143 | z_walk->zst_stride = |
144 | diff * z_walk->zst_direction; | |
145 | z_walk->zst_ph_offset = | |
146 | zh->zst_offset + z_walk->zst_stride; | |
147 | dmu_zfetch_stream_remove(zf, z_comp); | |
148 | mutex_destroy(&z_comp->zst_lock); | |
149 | kmem_free(z_comp, sizeof (zstream_t)); | |
150 | ||
151 | dmu_zfetch_dofetch(zf, z_walk); | |
152 | ||
153 | rw_exit(&zf->zf_rwlock); | |
154 | return (1); | |
155 | } | |
156 | ||
157 | diff = z_walk->zst_offset - z_comp->zst_offset; | |
158 | if (z_walk->zst_offset + diff == zh->zst_offset) { | |
159 | z_walk->zst_offset = zh->zst_offset; | |
a08ee875 LG |
160 | z_walk->zst_direction = diff < 0 ? |
161 | ZFETCH_BACKWARD : ZFETCH_FORWARD; | |
34dc7c2f BB |
162 | z_walk->zst_stride = |
163 | diff * z_walk->zst_direction; | |
164 | z_walk->zst_ph_offset = | |
165 | zh->zst_offset + z_walk->zst_stride; | |
166 | dmu_zfetch_stream_remove(zf, z_comp); | |
167 | mutex_destroy(&z_comp->zst_lock); | |
168 | kmem_free(z_comp, sizeof (zstream_t)); | |
169 | ||
170 | dmu_zfetch_dofetch(zf, z_walk); | |
171 | ||
172 | rw_exit(&zf->zf_rwlock); | |
173 | return (1); | |
174 | } | |
175 | } | |
176 | } | |
177 | ||
178 | rw_exit(&zf->zf_rwlock); | |
179 | return (0); | |
180 | } | |
181 | ||
182 | /* | |
183 | * Given a zstream_t, determine the bounds of the prefetch. Then call the | |
184 | * routine that actually prefetches the individual blocks. | |
185 | */ | |
186 | static void | |
187 | dmu_zfetch_dofetch(zfetch_t *zf, zstream_t *zs) | |
188 | { | |
189 | uint64_t prefetch_tail; | |
190 | uint64_t prefetch_limit; | |
191 | uint64_t prefetch_ofst; | |
192 | uint64_t prefetch_len; | |
193 | uint64_t blocks_fetched; | |
194 | ||
195 | zs->zst_stride = MAX((int64_t)zs->zst_stride, zs->zst_len); | |
196 | zs->zst_cap = MIN(zfetch_block_cap, 2 * zs->zst_cap); | |
197 | ||
198 | prefetch_tail = MAX((int64_t)zs->zst_ph_offset, | |
199 | (int64_t)(zs->zst_offset + zs->zst_stride)); | |
200 | /* | |
201 | * XXX: use a faster division method? | |
202 | */ | |
203 | prefetch_limit = zs->zst_offset + zs->zst_len + | |
204 | (zs->zst_cap * zs->zst_stride) / zs->zst_len; | |
205 | ||
206 | while (prefetch_tail < prefetch_limit) { | |
207 | prefetch_ofst = zs->zst_offset + zs->zst_direction * | |
208 | (prefetch_tail - zs->zst_offset); | |
209 | ||
210 | prefetch_len = zs->zst_len; | |
211 | ||
212 | /* | |
213 | * Don't prefetch beyond the end of the file, if working | |
214 | * backwards. | |
215 | */ | |
216 | if ((zs->zst_direction == ZFETCH_BACKWARD) && | |
217 | (prefetch_ofst > prefetch_tail)) { | |
218 | prefetch_len += prefetch_ofst; | |
219 | prefetch_ofst = 0; | |
220 | } | |
221 | ||
222 | /* don't prefetch more than we're supposed to */ | |
223 | if (prefetch_len > zs->zst_len) | |
224 | break; | |
225 | ||
226 | blocks_fetched = dmu_zfetch_fetch(zf->zf_dnode, | |
227 | prefetch_ofst, zs->zst_len); | |
228 | ||
229 | prefetch_tail += zs->zst_stride; | |
230 | /* stop if we've run out of stuff to prefetch */ | |
231 | if (blocks_fetched < zs->zst_len) | |
232 | break; | |
233 | } | |
234 | zs->zst_ph_offset = prefetch_tail; | |
428870ff BB |
235 | zs->zst_last = ddi_get_lbolt(); |
236 | } | |
237 | ||
238 | void | |
239 | zfetch_init(void) | |
240 | { | |
241 | ||
242 | zfetch_ksp = kstat_create("zfs", 0, "zfetchstats", "misc", | |
243 | KSTAT_TYPE_NAMED, sizeof (zfetch_stats) / sizeof (kstat_named_t), | |
244 | KSTAT_FLAG_VIRTUAL); | |
245 | ||
246 | if (zfetch_ksp != NULL) { | |
247 | zfetch_ksp->ks_data = &zfetch_stats; | |
248 | kstat_install(zfetch_ksp); | |
249 | } | |
250 | } | |
251 | ||
252 | void | |
253 | zfetch_fini(void) | |
254 | { | |
255 | if (zfetch_ksp != NULL) { | |
256 | kstat_delete(zfetch_ksp); | |
257 | zfetch_ksp = NULL; | |
258 | } | |
34dc7c2f BB |
259 | } |
260 | ||
261 | /* | |
262 | * This takes a pointer to a zfetch structure and a dnode. It performs the | |
263 | * necessary setup for the zfetch structure, grokking data from the | |
264 | * associated dnode. | |
265 | */ | |
266 | void | |
267 | dmu_zfetch_init(zfetch_t *zf, dnode_t *dno) | |
268 | { | |
269 | if (zf == NULL) { | |
270 | return; | |
271 | } | |
272 | ||
273 | zf->zf_dnode = dno; | |
274 | zf->zf_stream_cnt = 0; | |
275 | zf->zf_alloc_fail = 0; | |
276 | ||
277 | list_create(&zf->zf_stream, sizeof (zstream_t), | |
278 | offsetof(zstream_t, zst_node)); | |
279 | ||
280 | rw_init(&zf->zf_rwlock, NULL, RW_DEFAULT, NULL); | |
281 | } | |
282 | ||
283 | /* | |
284 | * This function computes the actual size, in blocks, that can be prefetched, | |
285 | * and fetches it. | |
286 | */ | |
287 | static uint64_t | |
288 | dmu_zfetch_fetch(dnode_t *dn, uint64_t blkid, uint64_t nblks) | |
289 | { | |
290 | uint64_t fetchsz; | |
291 | uint64_t i; | |
292 | ||
293 | fetchsz = dmu_zfetch_fetchsz(dn, blkid, nblks); | |
294 | ||
295 | for (i = 0; i < fetchsz; i++) { | |
a08ee875 | 296 | dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_ASYNC_READ); |
34dc7c2f BB |
297 | } |
298 | ||
299 | return (fetchsz); | |
300 | } | |
301 | ||
302 | /* | |
303 | * this function returns the number of blocks that would be prefetched, based | |
304 | * upon the supplied dnode, blockid, and nblks. This is used so that we can | |
305 | * update streams in place, and then prefetch with their old value after the | |
306 | * fact. This way, we can delay the prefetch, but subsequent accesses to the | |
307 | * stream won't result in the same data being prefetched multiple times. | |
308 | */ | |
309 | static uint64_t | |
310 | dmu_zfetch_fetchsz(dnode_t *dn, uint64_t blkid, uint64_t nblks) | |
311 | { | |
312 | uint64_t fetchsz; | |
313 | ||
314 | if (blkid > dn->dn_maxblkid) { | |
315 | return (0); | |
316 | } | |
317 | ||
318 | /* compute fetch size */ | |
319 | if (blkid + nblks + 1 > dn->dn_maxblkid) { | |
320 | fetchsz = (dn->dn_maxblkid - blkid) + 1; | |
321 | ASSERT(blkid + fetchsz - 1 <= dn->dn_maxblkid); | |
322 | } else { | |
323 | fetchsz = nblks; | |
324 | } | |
325 | ||
326 | ||
327 | return (fetchsz); | |
328 | } | |
329 | ||
330 | /* | |
428870ff | 331 | * given a zfetch and a zstream structure, see if there is an associated zstream |
34dc7c2f BB |
332 | * for this block read. If so, it starts a prefetch for the stream it |
333 | * located and returns true, otherwise it returns false | |
334 | */ | |
a08ee875 | 335 | static boolean_t |
34dc7c2f BB |
336 | dmu_zfetch_find(zfetch_t *zf, zstream_t *zh, int prefetched) |
337 | { | |
338 | zstream_t *zs; | |
339 | int64_t diff; | |
340 | int reset = !prefetched; | |
341 | int rc = 0; | |
342 | ||
343 | if (zh == NULL) | |
344 | return (0); | |
345 | ||
346 | /* | |
347 | * XXX: This locking strategy is a bit coarse; however, it's impact has | |
348 | * yet to be tested. If this turns out to be an issue, it can be | |
349 | * modified in a number of different ways. | |
350 | */ | |
351 | ||
352 | rw_enter(&zf->zf_rwlock, RW_READER); | |
353 | top: | |
354 | ||
355 | for (zs = list_head(&zf->zf_stream); zs; | |
356 | zs = list_next(&zf->zf_stream, zs)) { | |
357 | ||
358 | /* | |
359 | * XXX - should this be an assert? | |
360 | */ | |
361 | if (zs->zst_len == 0) { | |
362 | /* bogus stream */ | |
428870ff | 363 | ZFETCHSTAT_BUMP(zfetchstat_bogus_streams); |
34dc7c2f BB |
364 | continue; |
365 | } | |
366 | ||
367 | /* | |
368 | * We hit this case when we are in a strided prefetch stream: | |
369 | * we will read "len" blocks before "striding". | |
370 | */ | |
371 | if (zh->zst_offset >= zs->zst_offset && | |
372 | zh->zst_offset < zs->zst_offset + zs->zst_len) { | |
428870ff BB |
373 | if (prefetched) { |
374 | /* already fetched */ | |
375 | ZFETCHSTAT_BUMP(zfetchstat_stride_hits); | |
376 | rc = 1; | |
377 | goto out; | |
378 | } else { | |
379 | ZFETCHSTAT_BUMP(zfetchstat_stride_misses); | |
380 | } | |
34dc7c2f BB |
381 | } |
382 | ||
383 | /* | |
384 | * This is the forward sequential read case: we increment | |
385 | * len by one each time we hit here, so we will enter this | |
386 | * case on every read. | |
387 | */ | |
388 | if (zh->zst_offset == zs->zst_offset + zs->zst_len) { | |
389 | ||
390 | reset = !prefetched && zs->zst_len > 1; | |
391 | ||
392 | mutex_enter(&zs->zst_lock); | |
393 | ||
394 | if (zh->zst_offset != zs->zst_offset + zs->zst_len) { | |
395 | mutex_exit(&zs->zst_lock); | |
396 | goto top; | |
397 | } | |
398 | zs->zst_len += zh->zst_len; | |
399 | diff = zs->zst_len - zfetch_block_cap; | |
400 | if (diff > 0) { | |
401 | zs->zst_offset += diff; | |
402 | zs->zst_len = zs->zst_len > diff ? | |
403 | zs->zst_len - diff : 0; | |
404 | } | |
405 | zs->zst_direction = ZFETCH_FORWARD; | |
406 | ||
407 | break; | |
408 | ||
409 | /* | |
410 | * Same as above, but reading backwards through the file. | |
411 | */ | |
412 | } else if (zh->zst_offset == zs->zst_offset - zh->zst_len) { | |
413 | /* backwards sequential access */ | |
414 | ||
415 | reset = !prefetched && zs->zst_len > 1; | |
416 | ||
417 | mutex_enter(&zs->zst_lock); | |
418 | ||
419 | if (zh->zst_offset != zs->zst_offset - zh->zst_len) { | |
420 | mutex_exit(&zs->zst_lock); | |
421 | goto top; | |
422 | } | |
423 | ||
424 | zs->zst_offset = zs->zst_offset > zh->zst_len ? | |
425 | zs->zst_offset - zh->zst_len : 0; | |
426 | zs->zst_ph_offset = zs->zst_ph_offset > zh->zst_len ? | |
427 | zs->zst_ph_offset - zh->zst_len : 0; | |
428 | zs->zst_len += zh->zst_len; | |
429 | ||
430 | diff = zs->zst_len - zfetch_block_cap; | |
431 | if (diff > 0) { | |
432 | zs->zst_ph_offset = zs->zst_ph_offset > diff ? | |
433 | zs->zst_ph_offset - diff : 0; | |
434 | zs->zst_len = zs->zst_len > diff ? | |
435 | zs->zst_len - diff : zs->zst_len; | |
436 | } | |
437 | zs->zst_direction = ZFETCH_BACKWARD; | |
438 | ||
439 | break; | |
440 | ||
441 | } else if ((zh->zst_offset - zs->zst_offset - zs->zst_stride < | |
442 | zs->zst_len) && (zs->zst_len != zs->zst_stride)) { | |
443 | /* strided forward access */ | |
444 | ||
445 | mutex_enter(&zs->zst_lock); | |
446 | ||
447 | if ((zh->zst_offset - zs->zst_offset - zs->zst_stride >= | |
448 | zs->zst_len) || (zs->zst_len == zs->zst_stride)) { | |
449 | mutex_exit(&zs->zst_lock); | |
450 | goto top; | |
451 | } | |
452 | ||
453 | zs->zst_offset += zs->zst_stride; | |
454 | zs->zst_direction = ZFETCH_FORWARD; | |
455 | ||
456 | break; | |
457 | ||
458 | } else if ((zh->zst_offset - zs->zst_offset + zs->zst_stride < | |
459 | zs->zst_len) && (zs->zst_len != zs->zst_stride)) { | |
460 | /* strided reverse access */ | |
461 | ||
462 | mutex_enter(&zs->zst_lock); | |
463 | ||
464 | if ((zh->zst_offset - zs->zst_offset + zs->zst_stride >= | |
465 | zs->zst_len) || (zs->zst_len == zs->zst_stride)) { | |
466 | mutex_exit(&zs->zst_lock); | |
467 | goto top; | |
468 | } | |
469 | ||
470 | zs->zst_offset = zs->zst_offset > zs->zst_stride ? | |
471 | zs->zst_offset - zs->zst_stride : 0; | |
472 | zs->zst_ph_offset = (zs->zst_ph_offset > | |
473 | (2 * zs->zst_stride)) ? | |
474 | (zs->zst_ph_offset - (2 * zs->zst_stride)) : 0; | |
475 | zs->zst_direction = ZFETCH_BACKWARD; | |
476 | ||
477 | break; | |
478 | } | |
479 | } | |
480 | ||
481 | if (zs) { | |
482 | if (reset) { | |
483 | zstream_t *remove = zs; | |
484 | ||
428870ff | 485 | ZFETCHSTAT_BUMP(zfetchstat_stream_resets); |
34dc7c2f BB |
486 | rc = 0; |
487 | mutex_exit(&zs->zst_lock); | |
488 | rw_exit(&zf->zf_rwlock); | |
489 | rw_enter(&zf->zf_rwlock, RW_WRITER); | |
490 | /* | |
491 | * Relocate the stream, in case someone removes | |
492 | * it while we were acquiring the WRITER lock. | |
493 | */ | |
494 | for (zs = list_head(&zf->zf_stream); zs; | |
495 | zs = list_next(&zf->zf_stream, zs)) { | |
496 | if (zs == remove) { | |
497 | dmu_zfetch_stream_remove(zf, zs); | |
498 | mutex_destroy(&zs->zst_lock); | |
499 | kmem_free(zs, sizeof (zstream_t)); | |
500 | break; | |
501 | } | |
502 | } | |
503 | } else { | |
428870ff | 504 | ZFETCHSTAT_BUMP(zfetchstat_stream_noresets); |
34dc7c2f BB |
505 | rc = 1; |
506 | dmu_zfetch_dofetch(zf, zs); | |
507 | mutex_exit(&zs->zst_lock); | |
508 | } | |
509 | } | |
510 | out: | |
511 | rw_exit(&zf->zf_rwlock); | |
512 | return (rc); | |
513 | } | |
514 | ||
515 | /* | |
516 | * Clean-up state associated with a zfetch structure. This frees allocated | |
517 | * structure members, empties the zf_stream tree, and generally makes things | |
518 | * nice. This doesn't free the zfetch_t itself, that's left to the caller. | |
519 | */ | |
520 | void | |
521 | dmu_zfetch_rele(zfetch_t *zf) | |
522 | { | |
523 | zstream_t *zs; | |
524 | zstream_t *zs_next; | |
525 | ||
526 | ASSERT(!RW_LOCK_HELD(&zf->zf_rwlock)); | |
527 | ||
528 | for (zs = list_head(&zf->zf_stream); zs; zs = zs_next) { | |
529 | zs_next = list_next(&zf->zf_stream, zs); | |
530 | ||
531 | list_remove(&zf->zf_stream, zs); | |
532 | mutex_destroy(&zs->zst_lock); | |
533 | kmem_free(zs, sizeof (zstream_t)); | |
534 | } | |
535 | list_destroy(&zf->zf_stream); | |
536 | rw_destroy(&zf->zf_rwlock); | |
537 | ||
538 | zf->zf_dnode = NULL; | |
539 | } | |
540 | ||
541 | /* | |
542 | * Given a zfetch and zstream structure, insert the zstream structure into the | |
543 | * AVL tree contained within the zfetch structure. Peform the appropriate | |
544 | * book-keeping. It is possible that another thread has inserted a stream which | |
545 | * matches one that we are about to insert, so we must be sure to check for this | |
546 | * case. If one is found, return failure, and let the caller cleanup the | |
547 | * duplicates. | |
548 | */ | |
549 | static int | |
550 | dmu_zfetch_stream_insert(zfetch_t *zf, zstream_t *zs) | |
551 | { | |
552 | zstream_t *zs_walk; | |
553 | zstream_t *zs_next; | |
554 | ||
555 | ASSERT(RW_WRITE_HELD(&zf->zf_rwlock)); | |
556 | ||
557 | for (zs_walk = list_head(&zf->zf_stream); zs_walk; zs_walk = zs_next) { | |
558 | zs_next = list_next(&zf->zf_stream, zs_walk); | |
559 | ||
560 | if (dmu_zfetch_streams_equal(zs_walk, zs)) { | |
428870ff | 561 | return (0); |
34dc7c2f BB |
562 | } |
563 | } | |
564 | ||
565 | list_insert_head(&zf->zf_stream, zs); | |
566 | zf->zf_stream_cnt++; | |
34dc7c2f BB |
567 | return (1); |
568 | } | |
569 | ||
570 | ||
571 | /* | |
572 | * Walk the list of zstreams in the given zfetch, find an old one (by time), and | |
573 | * reclaim it for use by the caller. | |
574 | */ | |
575 | static zstream_t * | |
576 | dmu_zfetch_stream_reclaim(zfetch_t *zf) | |
577 | { | |
578 | zstream_t *zs; | |
579 | ||
580 | if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER)) | |
581 | return (0); | |
582 | ||
583 | for (zs = list_head(&zf->zf_stream); zs; | |
584 | zs = list_next(&zf->zf_stream, zs)) { | |
585 | ||
428870ff | 586 | if (((ddi_get_lbolt() - zs->zst_last)/hz) > zfetch_min_sec_reap) |
34dc7c2f BB |
587 | break; |
588 | } | |
589 | ||
590 | if (zs) { | |
591 | dmu_zfetch_stream_remove(zf, zs); | |
592 | mutex_destroy(&zs->zst_lock); | |
593 | bzero(zs, sizeof (zstream_t)); | |
594 | } else { | |
595 | zf->zf_alloc_fail++; | |
596 | } | |
597 | rw_exit(&zf->zf_rwlock); | |
598 | ||
599 | return (zs); | |
600 | } | |
601 | ||
602 | /* | |
603 | * Given a zfetch and zstream structure, remove the zstream structure from its | |
604 | * container in the zfetch structure. Perform the appropriate book-keeping. | |
605 | */ | |
606 | static void | |
607 | dmu_zfetch_stream_remove(zfetch_t *zf, zstream_t *zs) | |
608 | { | |
609 | ASSERT(RW_WRITE_HELD(&zf->zf_rwlock)); | |
610 | ||
611 | list_remove(&zf->zf_stream, zs); | |
612 | zf->zf_stream_cnt--; | |
613 | } | |
614 | ||
615 | static int | |
616 | dmu_zfetch_streams_equal(zstream_t *zs1, zstream_t *zs2) | |
617 | { | |
618 | if (zs1->zst_offset != zs2->zst_offset) | |
619 | return (0); | |
620 | ||
621 | if (zs1->zst_len != zs2->zst_len) | |
622 | return (0); | |
623 | ||
624 | if (zs1->zst_stride != zs2->zst_stride) | |
625 | return (0); | |
626 | ||
627 | if (zs1->zst_ph_offset != zs2->zst_ph_offset) | |
628 | return (0); | |
629 | ||
630 | if (zs1->zst_cap != zs2->zst_cap) | |
631 | return (0); | |
632 | ||
633 | if (zs1->zst_direction != zs2->zst_direction) | |
634 | return (0); | |
635 | ||
636 | return (1); | |
637 | } | |
638 | ||
639 | /* | |
640 | * This is the prefetch entry point. It calls all of the other dmu_zfetch | |
641 | * routines to create, delete, find, or operate upon prefetch streams. | |
642 | */ | |
643 | void | |
644 | dmu_zfetch(zfetch_t *zf, uint64_t offset, uint64_t size, int prefetched) | |
645 | { | |
646 | zstream_t zst; | |
647 | zstream_t *newstream; | |
a08ee875 | 648 | boolean_t fetched; |
34dc7c2f BB |
649 | int inserted; |
650 | unsigned int blkshft; | |
651 | uint64_t blksz; | |
652 | ||
653 | if (zfs_prefetch_disable) | |
654 | return; | |
655 | ||
656 | /* files that aren't ln2 blocksz are only one block -- nothing to do */ | |
657 | if (!zf->zf_dnode->dn_datablkshift) | |
658 | return; | |
659 | ||
660 | /* convert offset and size, into blockid and nblocks */ | |
661 | blkshft = zf->zf_dnode->dn_datablkshift; | |
662 | blksz = (1 << blkshft); | |
663 | ||
664 | bzero(&zst, sizeof (zstream_t)); | |
665 | zst.zst_offset = offset >> blkshft; | |
666 | zst.zst_len = (P2ROUNDUP(offset + size, blksz) - | |
667 | P2ALIGN(offset, blksz)) >> blkshft; | |
668 | ||
669 | fetched = dmu_zfetch_find(zf, &zst, prefetched); | |
428870ff BB |
670 | if (fetched) { |
671 | ZFETCHSTAT_BUMP(zfetchstat_hits); | |
672 | } else { | |
673 | ZFETCHSTAT_BUMP(zfetchstat_misses); | |
c65aa5b2 | 674 | if ((fetched = dmu_zfetch_colinear(zf, &zst))) { |
428870ff BB |
675 | ZFETCHSTAT_BUMP(zfetchstat_colinear_hits); |
676 | } else { | |
677 | ZFETCHSTAT_BUMP(zfetchstat_colinear_misses); | |
678 | } | |
34dc7c2f BB |
679 | } |
680 | ||
681 | if (!fetched) { | |
682 | newstream = dmu_zfetch_stream_reclaim(zf); | |
683 | ||
684 | /* | |
685 | * we still couldn't find a stream, drop the lock, and allocate | |
686 | * one if possible. Otherwise, give up and go home. | |
687 | */ | |
428870ff BB |
688 | if (newstream) { |
689 | ZFETCHSTAT_BUMP(zfetchstat_reclaim_successes); | |
690 | } else { | |
34dc7c2f BB |
691 | uint64_t maxblocks; |
692 | uint32_t max_streams; | |
693 | uint32_t cur_streams; | |
694 | ||
428870ff | 695 | ZFETCHSTAT_BUMP(zfetchstat_reclaim_failures); |
34dc7c2f BB |
696 | cur_streams = zf->zf_stream_cnt; |
697 | maxblocks = zf->zf_dnode->dn_maxblkid; | |
698 | ||
699 | max_streams = MIN(zfetch_max_streams, | |
700 | (maxblocks / zfetch_block_cap)); | |
701 | if (max_streams == 0) { | |
702 | max_streams++; | |
703 | } | |
704 | ||
705 | if (cur_streams >= max_streams) { | |
706 | return; | |
707 | } | |
a08ee875 | 708 | newstream = |
ea04106b | 709 | kmem_zalloc(sizeof (zstream_t), KM_SLEEP); |
34dc7c2f BB |
710 | } |
711 | ||
712 | newstream->zst_offset = zst.zst_offset; | |
713 | newstream->zst_len = zst.zst_len; | |
714 | newstream->zst_stride = zst.zst_len; | |
715 | newstream->zst_ph_offset = zst.zst_len + zst.zst_offset; | |
716 | newstream->zst_cap = zst.zst_len; | |
717 | newstream->zst_direction = ZFETCH_FORWARD; | |
428870ff | 718 | newstream->zst_last = ddi_get_lbolt(); |
34dc7c2f BB |
719 | |
720 | mutex_init(&newstream->zst_lock, NULL, MUTEX_DEFAULT, NULL); | |
721 | ||
722 | rw_enter(&zf->zf_rwlock, RW_WRITER); | |
723 | inserted = dmu_zfetch_stream_insert(zf, newstream); | |
724 | rw_exit(&zf->zf_rwlock); | |
725 | ||
726 | if (!inserted) { | |
727 | mutex_destroy(&newstream->zst_lock); | |
728 | kmem_free(newstream, sizeof (zstream_t)); | |
729 | } | |
730 | } | |
731 | } | |
c28b2279 BB |
732 | |
733 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
734 | module_param(zfs_prefetch_disable, int, 0644); | |
735 | MODULE_PARM_DESC(zfs_prefetch_disable, "Disable all ZFS prefetching"); | |
c409e464 BB |
736 | |
737 | module_param(zfetch_max_streams, uint, 0644); | |
738 | MODULE_PARM_DESC(zfetch_max_streams, "Max number of streams per zfetch"); | |
739 | ||
740 | module_param(zfetch_min_sec_reap, uint, 0644); | |
741 | MODULE_PARM_DESC(zfetch_min_sec_reap, "Min time before stream reclaim"); | |
742 | ||
743 | module_param(zfetch_block_cap, uint, 0644); | |
744 | MODULE_PARM_DESC(zfetch_block_cap, "Max number of blocks to fetch at a time"); | |
745 | ||
746 | module_param(zfetch_array_rd_sz, ulong, 0644); | |
747 | MODULE_PARM_DESC(zfetch_array_rd_sz, "Number of bytes in a array_read"); | |
c28b2279 | 748 | #endif |