]>
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 | ||
e8b96c60 MA |
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 */ | |
e49f1e20 | 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); | |
e49f1e20 | 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 | * | |
e49f1e20 | 111 | * Returns whether co-linear streams were found. |
34dc7c2f | 112 | */ |
e49f1e20 | 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; | |
141 | z_walk->zst_direction = diff < 0 ? -1 : 1; | |
142 | z_walk->zst_stride = | |
143 | diff * z_walk->zst_direction; | |
144 | z_walk->zst_ph_offset = | |
145 | zh->zst_offset + z_walk->zst_stride; | |
146 | dmu_zfetch_stream_remove(zf, z_comp); | |
147 | mutex_destroy(&z_comp->zst_lock); | |
148 | kmem_free(z_comp, sizeof (zstream_t)); | |
149 | ||
150 | dmu_zfetch_dofetch(zf, z_walk); | |
151 | ||
152 | rw_exit(&zf->zf_rwlock); | |
153 | return (1); | |
154 | } | |
155 | ||
156 | diff = z_walk->zst_offset - z_comp->zst_offset; | |
157 | if (z_walk->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 | } | |
175 | ||
176 | rw_exit(&zf->zf_rwlock); | |
177 | return (0); | |
178 | } | |
179 | ||
180 | /* | |
181 | * Given a zstream_t, determine the bounds of the prefetch. Then call the | |
182 | * routine that actually prefetches the individual blocks. | |
183 | */ | |
184 | static void | |
185 | dmu_zfetch_dofetch(zfetch_t *zf, zstream_t *zs) | |
186 | { | |
187 | uint64_t prefetch_tail; | |
188 | uint64_t prefetch_limit; | |
189 | uint64_t prefetch_ofst; | |
190 | uint64_t prefetch_len; | |
191 | uint64_t blocks_fetched; | |
192 | ||
193 | zs->zst_stride = MAX((int64_t)zs->zst_stride, zs->zst_len); | |
194 | zs->zst_cap = MIN(zfetch_block_cap, 2 * zs->zst_cap); | |
195 | ||
196 | prefetch_tail = MAX((int64_t)zs->zst_ph_offset, | |
197 | (int64_t)(zs->zst_offset + zs->zst_stride)); | |
198 | /* | |
199 | * XXX: use a faster division method? | |
200 | */ | |
201 | prefetch_limit = zs->zst_offset + zs->zst_len + | |
202 | (zs->zst_cap * zs->zst_stride) / zs->zst_len; | |
203 | ||
204 | while (prefetch_tail < prefetch_limit) { | |
205 | prefetch_ofst = zs->zst_offset + zs->zst_direction * | |
206 | (prefetch_tail - zs->zst_offset); | |
207 | ||
208 | prefetch_len = zs->zst_len; | |
209 | ||
210 | /* | |
211 | * Don't prefetch beyond the end of the file, if working | |
212 | * backwards. | |
213 | */ | |
214 | if ((zs->zst_direction == ZFETCH_BACKWARD) && | |
215 | (prefetch_ofst > prefetch_tail)) { | |
216 | prefetch_len += prefetch_ofst; | |
217 | prefetch_ofst = 0; | |
218 | } | |
219 | ||
220 | /* don't prefetch more than we're supposed to */ | |
221 | if (prefetch_len > zs->zst_len) | |
222 | break; | |
223 | ||
224 | blocks_fetched = dmu_zfetch_fetch(zf->zf_dnode, | |
225 | prefetch_ofst, zs->zst_len); | |
226 | ||
227 | prefetch_tail += zs->zst_stride; | |
228 | /* stop if we've run out of stuff to prefetch */ | |
229 | if (blocks_fetched < zs->zst_len) | |
230 | break; | |
231 | } | |
232 | zs->zst_ph_offset = prefetch_tail; | |
428870ff BB |
233 | zs->zst_last = ddi_get_lbolt(); |
234 | } | |
235 | ||
236 | void | |
237 | zfetch_init(void) | |
238 | { | |
239 | ||
240 | zfetch_ksp = kstat_create("zfs", 0, "zfetchstats", "misc", | |
241 | KSTAT_TYPE_NAMED, sizeof (zfetch_stats) / sizeof (kstat_named_t), | |
242 | KSTAT_FLAG_VIRTUAL); | |
243 | ||
244 | if (zfetch_ksp != NULL) { | |
245 | zfetch_ksp->ks_data = &zfetch_stats; | |
246 | kstat_install(zfetch_ksp); | |
247 | } | |
248 | } | |
249 | ||
250 | void | |
251 | zfetch_fini(void) | |
252 | { | |
253 | if (zfetch_ksp != NULL) { | |
254 | kstat_delete(zfetch_ksp); | |
255 | zfetch_ksp = NULL; | |
256 | } | |
34dc7c2f BB |
257 | } |
258 | ||
259 | /* | |
260 | * This takes a pointer to a zfetch structure and a dnode. It performs the | |
261 | * necessary setup for the zfetch structure, grokking data from the | |
262 | * associated dnode. | |
263 | */ | |
264 | void | |
265 | dmu_zfetch_init(zfetch_t *zf, dnode_t *dno) | |
266 | { | |
267 | if (zf == NULL) { | |
268 | return; | |
269 | } | |
270 | ||
271 | zf->zf_dnode = dno; | |
272 | zf->zf_stream_cnt = 0; | |
273 | zf->zf_alloc_fail = 0; | |
274 | ||
275 | list_create(&zf->zf_stream, sizeof (zstream_t), | |
276 | offsetof(zstream_t, zst_node)); | |
277 | ||
278 | rw_init(&zf->zf_rwlock, NULL, RW_DEFAULT, NULL); | |
279 | } | |
280 | ||
281 | /* | |
282 | * This function computes the actual size, in blocks, that can be prefetched, | |
283 | * and fetches it. | |
284 | */ | |
285 | static uint64_t | |
286 | dmu_zfetch_fetch(dnode_t *dn, uint64_t blkid, uint64_t nblks) | |
287 | { | |
288 | uint64_t fetchsz; | |
289 | uint64_t i; | |
290 | ||
291 | fetchsz = dmu_zfetch_fetchsz(dn, blkid, nblks); | |
292 | ||
293 | for (i = 0; i < fetchsz; i++) { | |
e8b96c60 | 294 | dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_ASYNC_READ); |
34dc7c2f BB |
295 | } |
296 | ||
297 | return (fetchsz); | |
298 | } | |
299 | ||
300 | /* | |
301 | * this function returns the number of blocks that would be prefetched, based | |
302 | * upon the supplied dnode, blockid, and nblks. This is used so that we can | |
303 | * update streams in place, and then prefetch with their old value after the | |
304 | * fact. This way, we can delay the prefetch, but subsequent accesses to the | |
305 | * stream won't result in the same data being prefetched multiple times. | |
306 | */ | |
307 | static uint64_t | |
308 | dmu_zfetch_fetchsz(dnode_t *dn, uint64_t blkid, uint64_t nblks) | |
309 | { | |
310 | uint64_t fetchsz; | |
311 | ||
312 | if (blkid > dn->dn_maxblkid) { | |
313 | return (0); | |
314 | } | |
315 | ||
316 | /* compute fetch size */ | |
317 | if (blkid + nblks + 1 > dn->dn_maxblkid) { | |
318 | fetchsz = (dn->dn_maxblkid - blkid) + 1; | |
319 | ASSERT(blkid + fetchsz - 1 <= dn->dn_maxblkid); | |
320 | } else { | |
321 | fetchsz = nblks; | |
322 | } | |
323 | ||
324 | ||
325 | return (fetchsz); | |
326 | } | |
327 | ||
328 | /* | |
428870ff | 329 | * given a zfetch and a zstream structure, see if there is an associated zstream |
34dc7c2f BB |
330 | * for this block read. If so, it starts a prefetch for the stream it |
331 | * located and returns true, otherwise it returns false | |
332 | */ | |
e49f1e20 | 333 | static boolean_t |
34dc7c2f BB |
334 | dmu_zfetch_find(zfetch_t *zf, zstream_t *zh, int prefetched) |
335 | { | |
336 | zstream_t *zs; | |
337 | int64_t diff; | |
338 | int reset = !prefetched; | |
339 | int rc = 0; | |
340 | ||
341 | if (zh == NULL) | |
342 | return (0); | |
343 | ||
344 | /* | |
345 | * XXX: This locking strategy is a bit coarse; however, it's impact has | |
346 | * yet to be tested. If this turns out to be an issue, it can be | |
347 | * modified in a number of different ways. | |
348 | */ | |
349 | ||
350 | rw_enter(&zf->zf_rwlock, RW_READER); | |
351 | top: | |
352 | ||
353 | for (zs = list_head(&zf->zf_stream); zs; | |
354 | zs = list_next(&zf->zf_stream, zs)) { | |
355 | ||
356 | /* | |
357 | * XXX - should this be an assert? | |
358 | */ | |
359 | if (zs->zst_len == 0) { | |
360 | /* bogus stream */ | |
428870ff | 361 | ZFETCHSTAT_BUMP(zfetchstat_bogus_streams); |
34dc7c2f BB |
362 | continue; |
363 | } | |
364 | ||
365 | /* | |
366 | * We hit this case when we are in a strided prefetch stream: | |
367 | * we will read "len" blocks before "striding". | |
368 | */ | |
369 | if (zh->zst_offset >= zs->zst_offset && | |
370 | zh->zst_offset < zs->zst_offset + zs->zst_len) { | |
428870ff BB |
371 | if (prefetched) { |
372 | /* already fetched */ | |
373 | ZFETCHSTAT_BUMP(zfetchstat_stride_hits); | |
374 | rc = 1; | |
375 | goto out; | |
376 | } else { | |
377 | ZFETCHSTAT_BUMP(zfetchstat_stride_misses); | |
378 | } | |
34dc7c2f BB |
379 | } |
380 | ||
381 | /* | |
382 | * This is the forward sequential read case: we increment | |
383 | * len by one each time we hit here, so we will enter this | |
384 | * case on every read. | |
385 | */ | |
386 | if (zh->zst_offset == zs->zst_offset + zs->zst_len) { | |
387 | ||
388 | reset = !prefetched && zs->zst_len > 1; | |
389 | ||
390 | mutex_enter(&zs->zst_lock); | |
391 | ||
392 | if (zh->zst_offset != zs->zst_offset + zs->zst_len) { | |
393 | mutex_exit(&zs->zst_lock); | |
394 | goto top; | |
395 | } | |
396 | zs->zst_len += zh->zst_len; | |
397 | diff = zs->zst_len - zfetch_block_cap; | |
398 | if (diff > 0) { | |
399 | zs->zst_offset += diff; | |
400 | zs->zst_len = zs->zst_len > diff ? | |
401 | zs->zst_len - diff : 0; | |
402 | } | |
403 | zs->zst_direction = ZFETCH_FORWARD; | |
404 | ||
405 | break; | |
406 | ||
407 | /* | |
408 | * Same as above, but reading backwards through the file. | |
409 | */ | |
410 | } else if (zh->zst_offset == zs->zst_offset - zh->zst_len) { | |
411 | /* backwards sequential access */ | |
412 | ||
413 | reset = !prefetched && zs->zst_len > 1; | |
414 | ||
415 | mutex_enter(&zs->zst_lock); | |
416 | ||
417 | if (zh->zst_offset != zs->zst_offset - zh->zst_len) { | |
418 | mutex_exit(&zs->zst_lock); | |
419 | goto top; | |
420 | } | |
421 | ||
422 | zs->zst_offset = zs->zst_offset > zh->zst_len ? | |
423 | zs->zst_offset - zh->zst_len : 0; | |
424 | zs->zst_ph_offset = zs->zst_ph_offset > zh->zst_len ? | |
425 | zs->zst_ph_offset - zh->zst_len : 0; | |
426 | zs->zst_len += zh->zst_len; | |
427 | ||
428 | diff = zs->zst_len - zfetch_block_cap; | |
429 | if (diff > 0) { | |
430 | zs->zst_ph_offset = zs->zst_ph_offset > diff ? | |
431 | zs->zst_ph_offset - diff : 0; | |
432 | zs->zst_len = zs->zst_len > diff ? | |
433 | zs->zst_len - diff : zs->zst_len; | |
434 | } | |
435 | zs->zst_direction = ZFETCH_BACKWARD; | |
436 | ||
437 | break; | |
438 | ||
439 | } else if ((zh->zst_offset - zs->zst_offset - zs->zst_stride < | |
440 | zs->zst_len) && (zs->zst_len != zs->zst_stride)) { | |
441 | /* strided forward access */ | |
442 | ||
443 | mutex_enter(&zs->zst_lock); | |
444 | ||
445 | if ((zh->zst_offset - zs->zst_offset - zs->zst_stride >= | |
446 | zs->zst_len) || (zs->zst_len == zs->zst_stride)) { | |
447 | mutex_exit(&zs->zst_lock); | |
448 | goto top; | |
449 | } | |
450 | ||
451 | zs->zst_offset += zs->zst_stride; | |
452 | zs->zst_direction = ZFETCH_FORWARD; | |
453 | ||
454 | break; | |
455 | ||
456 | } else if ((zh->zst_offset - zs->zst_offset + zs->zst_stride < | |
457 | zs->zst_len) && (zs->zst_len != zs->zst_stride)) { | |
458 | /* strided reverse access */ | |
459 | ||
460 | mutex_enter(&zs->zst_lock); | |
461 | ||
462 | if ((zh->zst_offset - zs->zst_offset + zs->zst_stride >= | |
463 | zs->zst_len) || (zs->zst_len == zs->zst_stride)) { | |
464 | mutex_exit(&zs->zst_lock); | |
465 | goto top; | |
466 | } | |
467 | ||
468 | zs->zst_offset = zs->zst_offset > zs->zst_stride ? | |
469 | zs->zst_offset - zs->zst_stride : 0; | |
470 | zs->zst_ph_offset = (zs->zst_ph_offset > | |
471 | (2 * zs->zst_stride)) ? | |
472 | (zs->zst_ph_offset - (2 * zs->zst_stride)) : 0; | |
473 | zs->zst_direction = ZFETCH_BACKWARD; | |
474 | ||
475 | break; | |
476 | } | |
477 | } | |
478 | ||
479 | if (zs) { | |
480 | if (reset) { | |
481 | zstream_t *remove = zs; | |
482 | ||
428870ff | 483 | ZFETCHSTAT_BUMP(zfetchstat_stream_resets); |
34dc7c2f BB |
484 | rc = 0; |
485 | mutex_exit(&zs->zst_lock); | |
486 | rw_exit(&zf->zf_rwlock); | |
487 | rw_enter(&zf->zf_rwlock, RW_WRITER); | |
488 | /* | |
489 | * Relocate the stream, in case someone removes | |
490 | * it while we were acquiring the WRITER lock. | |
491 | */ | |
492 | for (zs = list_head(&zf->zf_stream); zs; | |
493 | zs = list_next(&zf->zf_stream, zs)) { | |
494 | if (zs == remove) { | |
495 | dmu_zfetch_stream_remove(zf, zs); | |
496 | mutex_destroy(&zs->zst_lock); | |
497 | kmem_free(zs, sizeof (zstream_t)); | |
498 | break; | |
499 | } | |
500 | } | |
501 | } else { | |
428870ff | 502 | ZFETCHSTAT_BUMP(zfetchstat_stream_noresets); |
34dc7c2f BB |
503 | rc = 1; |
504 | dmu_zfetch_dofetch(zf, zs); | |
505 | mutex_exit(&zs->zst_lock); | |
506 | } | |
507 | } | |
508 | out: | |
509 | rw_exit(&zf->zf_rwlock); | |
510 | return (rc); | |
511 | } | |
512 | ||
513 | /* | |
514 | * Clean-up state associated with a zfetch structure. This frees allocated | |
515 | * structure members, empties the zf_stream tree, and generally makes things | |
516 | * nice. This doesn't free the zfetch_t itself, that's left to the caller. | |
517 | */ | |
518 | void | |
519 | dmu_zfetch_rele(zfetch_t *zf) | |
520 | { | |
521 | zstream_t *zs; | |
522 | zstream_t *zs_next; | |
523 | ||
524 | ASSERT(!RW_LOCK_HELD(&zf->zf_rwlock)); | |
525 | ||
526 | for (zs = list_head(&zf->zf_stream); zs; zs = zs_next) { | |
527 | zs_next = list_next(&zf->zf_stream, zs); | |
528 | ||
529 | list_remove(&zf->zf_stream, zs); | |
530 | mutex_destroy(&zs->zst_lock); | |
531 | kmem_free(zs, sizeof (zstream_t)); | |
532 | } | |
533 | list_destroy(&zf->zf_stream); | |
534 | rw_destroy(&zf->zf_rwlock); | |
535 | ||
536 | zf->zf_dnode = NULL; | |
537 | } | |
538 | ||
539 | /* | |
540 | * Given a zfetch and zstream structure, insert the zstream structure into the | |
541 | * AVL tree contained within the zfetch structure. Peform the appropriate | |
542 | * book-keeping. It is possible that another thread has inserted a stream which | |
543 | * matches one that we are about to insert, so we must be sure to check for this | |
544 | * case. If one is found, return failure, and let the caller cleanup the | |
545 | * duplicates. | |
546 | */ | |
547 | static int | |
548 | dmu_zfetch_stream_insert(zfetch_t *zf, zstream_t *zs) | |
549 | { | |
550 | zstream_t *zs_walk; | |
551 | zstream_t *zs_next; | |
552 | ||
553 | ASSERT(RW_WRITE_HELD(&zf->zf_rwlock)); | |
554 | ||
555 | for (zs_walk = list_head(&zf->zf_stream); zs_walk; zs_walk = zs_next) { | |
556 | zs_next = list_next(&zf->zf_stream, zs_walk); | |
557 | ||
558 | if (dmu_zfetch_streams_equal(zs_walk, zs)) { | |
428870ff | 559 | return (0); |
34dc7c2f BB |
560 | } |
561 | } | |
562 | ||
563 | list_insert_head(&zf->zf_stream, zs); | |
564 | zf->zf_stream_cnt++; | |
34dc7c2f BB |
565 | return (1); |
566 | } | |
567 | ||
568 | ||
569 | /* | |
570 | * Walk the list of zstreams in the given zfetch, find an old one (by time), and | |
571 | * reclaim it for use by the caller. | |
572 | */ | |
573 | static zstream_t * | |
574 | dmu_zfetch_stream_reclaim(zfetch_t *zf) | |
575 | { | |
576 | zstream_t *zs; | |
577 | ||
578 | if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER)) | |
579 | return (0); | |
580 | ||
581 | for (zs = list_head(&zf->zf_stream); zs; | |
582 | zs = list_next(&zf->zf_stream, zs)) { | |
583 | ||
428870ff | 584 | if (((ddi_get_lbolt() - zs->zst_last)/hz) > zfetch_min_sec_reap) |
34dc7c2f BB |
585 | break; |
586 | } | |
587 | ||
588 | if (zs) { | |
589 | dmu_zfetch_stream_remove(zf, zs); | |
590 | mutex_destroy(&zs->zst_lock); | |
591 | bzero(zs, sizeof (zstream_t)); | |
592 | } else { | |
593 | zf->zf_alloc_fail++; | |
594 | } | |
595 | rw_exit(&zf->zf_rwlock); | |
596 | ||
597 | return (zs); | |
598 | } | |
599 | ||
600 | /* | |
601 | * Given a zfetch and zstream structure, remove the zstream structure from its | |
602 | * container in the zfetch structure. Perform the appropriate book-keeping. | |
603 | */ | |
604 | static void | |
605 | dmu_zfetch_stream_remove(zfetch_t *zf, zstream_t *zs) | |
606 | { | |
607 | ASSERT(RW_WRITE_HELD(&zf->zf_rwlock)); | |
608 | ||
609 | list_remove(&zf->zf_stream, zs); | |
610 | zf->zf_stream_cnt--; | |
611 | } | |
612 | ||
613 | static int | |
614 | dmu_zfetch_streams_equal(zstream_t *zs1, zstream_t *zs2) | |
615 | { | |
616 | if (zs1->zst_offset != zs2->zst_offset) | |
617 | return (0); | |
618 | ||
619 | if (zs1->zst_len != zs2->zst_len) | |
620 | return (0); | |
621 | ||
622 | if (zs1->zst_stride != zs2->zst_stride) | |
623 | return (0); | |
624 | ||
625 | if (zs1->zst_ph_offset != zs2->zst_ph_offset) | |
626 | return (0); | |
627 | ||
628 | if (zs1->zst_cap != zs2->zst_cap) | |
629 | return (0); | |
630 | ||
631 | if (zs1->zst_direction != zs2->zst_direction) | |
632 | return (0); | |
633 | ||
634 | return (1); | |
635 | } | |
636 | ||
637 | /* | |
638 | * This is the prefetch entry point. It calls all of the other dmu_zfetch | |
639 | * routines to create, delete, find, or operate upon prefetch streams. | |
640 | */ | |
641 | void | |
642 | dmu_zfetch(zfetch_t *zf, uint64_t offset, uint64_t size, int prefetched) | |
643 | { | |
644 | zstream_t zst; | |
645 | zstream_t *newstream; | |
e49f1e20 | 646 | boolean_t fetched; |
34dc7c2f BB |
647 | int inserted; |
648 | unsigned int blkshft; | |
649 | uint64_t blksz; | |
650 | ||
651 | if (zfs_prefetch_disable) | |
652 | return; | |
653 | ||
654 | /* files that aren't ln2 blocksz are only one block -- nothing to do */ | |
655 | if (!zf->zf_dnode->dn_datablkshift) | |
656 | return; | |
657 | ||
658 | /* convert offset and size, into blockid and nblocks */ | |
659 | blkshft = zf->zf_dnode->dn_datablkshift; | |
660 | blksz = (1 << blkshft); | |
661 | ||
662 | bzero(&zst, sizeof (zstream_t)); | |
663 | zst.zst_offset = offset >> blkshft; | |
664 | zst.zst_len = (P2ROUNDUP(offset + size, blksz) - | |
665 | P2ALIGN(offset, blksz)) >> blkshft; | |
666 | ||
667 | fetched = dmu_zfetch_find(zf, &zst, prefetched); | |
428870ff BB |
668 | if (fetched) { |
669 | ZFETCHSTAT_BUMP(zfetchstat_hits); | |
670 | } else { | |
671 | ZFETCHSTAT_BUMP(zfetchstat_misses); | |
c65aa5b2 | 672 | if ((fetched = dmu_zfetch_colinear(zf, &zst))) { |
428870ff BB |
673 | ZFETCHSTAT_BUMP(zfetchstat_colinear_hits); |
674 | } else { | |
675 | ZFETCHSTAT_BUMP(zfetchstat_colinear_misses); | |
676 | } | |
34dc7c2f BB |
677 | } |
678 | ||
679 | if (!fetched) { | |
680 | newstream = dmu_zfetch_stream_reclaim(zf); | |
681 | ||
682 | /* | |
683 | * we still couldn't find a stream, drop the lock, and allocate | |
684 | * one if possible. Otherwise, give up and go home. | |
685 | */ | |
428870ff BB |
686 | if (newstream) { |
687 | ZFETCHSTAT_BUMP(zfetchstat_reclaim_successes); | |
688 | } else { | |
34dc7c2f BB |
689 | uint64_t maxblocks; |
690 | uint32_t max_streams; | |
691 | uint32_t cur_streams; | |
692 | ||
428870ff | 693 | ZFETCHSTAT_BUMP(zfetchstat_reclaim_failures); |
34dc7c2f BB |
694 | cur_streams = zf->zf_stream_cnt; |
695 | maxblocks = zf->zf_dnode->dn_maxblkid; | |
696 | ||
697 | max_streams = MIN(zfetch_max_streams, | |
698 | (maxblocks / zfetch_block_cap)); | |
699 | if (max_streams == 0) { | |
700 | max_streams++; | |
701 | } | |
702 | ||
703 | if (cur_streams >= max_streams) { | |
704 | return; | |
705 | } | |
d1d7e268 MK |
706 | newstream = |
707 | kmem_zalloc(sizeof (zstream_t), KM_PUSHPAGE); | |
34dc7c2f BB |
708 | } |
709 | ||
710 | newstream->zst_offset = zst.zst_offset; | |
711 | newstream->zst_len = zst.zst_len; | |
712 | newstream->zst_stride = zst.zst_len; | |
713 | newstream->zst_ph_offset = zst.zst_len + zst.zst_offset; | |
714 | newstream->zst_cap = zst.zst_len; | |
715 | newstream->zst_direction = ZFETCH_FORWARD; | |
428870ff | 716 | newstream->zst_last = ddi_get_lbolt(); |
34dc7c2f BB |
717 | |
718 | mutex_init(&newstream->zst_lock, NULL, MUTEX_DEFAULT, NULL); | |
719 | ||
720 | rw_enter(&zf->zf_rwlock, RW_WRITER); | |
721 | inserted = dmu_zfetch_stream_insert(zf, newstream); | |
722 | rw_exit(&zf->zf_rwlock); | |
723 | ||
724 | if (!inserted) { | |
725 | mutex_destroy(&newstream->zst_lock); | |
726 | kmem_free(newstream, sizeof (zstream_t)); | |
727 | } | |
728 | } | |
729 | } | |
c28b2279 BB |
730 | |
731 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
732 | module_param(zfs_prefetch_disable, int, 0644); | |
733 | MODULE_PARM_DESC(zfs_prefetch_disable, "Disable all ZFS prefetching"); | |
c409e464 BB |
734 | |
735 | module_param(zfetch_max_streams, uint, 0644); | |
736 | MODULE_PARM_DESC(zfetch_max_streams, "Max number of streams per zfetch"); | |
737 | ||
738 | module_param(zfetch_min_sec_reap, uint, 0644); | |
739 | MODULE_PARM_DESC(zfetch_min_sec_reap, "Min time before stream reclaim"); | |
740 | ||
741 | module_param(zfetch_block_cap, uint, 0644); | |
742 | MODULE_PARM_DESC(zfetch_block_cap, "Max number of blocks to fetch at a time"); | |
743 | ||
744 | module_param(zfetch_array_rd_sz, ulong, 0644); | |
745 | MODULE_PARM_DESC(zfetch_array_rd_sz, "Number of bytes in a array_read"); | |
c28b2279 | 746 | #endif |