]>
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 2010 Sun Microsystems, Inc. All rights reserved. |
34dc7c2f BB |
23 | * Use is subject to license terms. |
24 | */ | |
25 | ||
34dc7c2f BB |
26 | #include <sys/zfs_context.h> |
27 | #include <sys/spa.h> | |
28 | #include <sys/vdev_impl.h> | |
29 | #include <sys/zio.h> | |
30 | #include <sys/fs/zfs.h> | |
31 | ||
32 | /* | |
33 | * Virtual device vector for mirroring. | |
34 | */ | |
35 | ||
36 | typedef struct mirror_child { | |
37 | vdev_t *mc_vd; | |
38 | uint64_t mc_offset; | |
39 | int mc_error; | |
b128c09f BB |
40 | uint8_t mc_tried; |
41 | uint8_t mc_skipped; | |
42 | uint8_t mc_speculative; | |
34dc7c2f BB |
43 | } mirror_child_t; |
44 | ||
45 | typedef struct mirror_map { | |
46 | int mm_children; | |
47 | int mm_replacing; | |
48 | int mm_preferred; | |
49 | int mm_root; | |
50 | mirror_child_t mm_child[1]; | |
51 | } mirror_map_t; | |
52 | ||
53 | int vdev_mirror_shift = 21; | |
54 | ||
b128c09f BB |
55 | static void |
56 | vdev_mirror_map_free(zio_t *zio) | |
57 | { | |
58 | mirror_map_t *mm = zio->io_vsd; | |
59 | ||
60 | kmem_free(mm, offsetof(mirror_map_t, mm_child[mm->mm_children])); | |
61 | } | |
62 | ||
428870ff BB |
63 | static const zio_vsd_ops_t vdev_mirror_vsd_ops = { |
64 | vdev_mirror_map_free, | |
65 | zio_vsd_default_cksum_report | |
66 | }; | |
67 | ||
34dc7c2f BB |
68 | static mirror_map_t * |
69 | vdev_mirror_map_alloc(zio_t *zio) | |
70 | { | |
71 | mirror_map_t *mm = NULL; | |
72 | mirror_child_t *mc; | |
73 | vdev_t *vd = zio->io_vd; | |
74 | int c, d; | |
75 | ||
76 | if (vd == NULL) { | |
77 | dva_t *dva = zio->io_bp->blk_dva; | |
78 | spa_t *spa = zio->io_spa; | |
79 | ||
80 | c = BP_GET_NDVAS(zio->io_bp); | |
81 | ||
82 | mm = kmem_zalloc(offsetof(mirror_map_t, mm_child[c]), KM_SLEEP); | |
83 | mm->mm_children = c; | |
84 | mm->mm_replacing = B_FALSE; | |
85 | mm->mm_preferred = spa_get_random(c); | |
86 | mm->mm_root = B_TRUE; | |
87 | ||
88 | /* | |
89 | * Check the other, lower-index DVAs to see if they're on | |
90 | * the same vdev as the child we picked. If they are, use | |
91 | * them since they are likely to have been allocated from | |
92 | * the primary metaslab in use at the time, and hence are | |
93 | * more likely to have locality with single-copy data. | |
94 | */ | |
95 | for (c = mm->mm_preferred, d = c - 1; d >= 0; d--) { | |
96 | if (DVA_GET_VDEV(&dva[d]) == DVA_GET_VDEV(&dva[c])) | |
97 | mm->mm_preferred = d; | |
98 | } | |
99 | ||
100 | for (c = 0; c < mm->mm_children; c++) { | |
101 | mc = &mm->mm_child[c]; | |
102 | ||
103 | mc->mc_vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[c])); | |
104 | mc->mc_offset = DVA_GET_OFFSET(&dva[c]); | |
105 | } | |
106 | } else { | |
107 | c = vd->vdev_children; | |
108 | ||
109 | mm = kmem_zalloc(offsetof(mirror_map_t, mm_child[c]), KM_SLEEP); | |
110 | mm->mm_children = c; | |
111 | mm->mm_replacing = (vd->vdev_ops == &vdev_replacing_ops || | |
112 | vd->vdev_ops == &vdev_spare_ops); | |
113 | mm->mm_preferred = mm->mm_replacing ? 0 : | |
114 | (zio->io_offset >> vdev_mirror_shift) % c; | |
115 | mm->mm_root = B_FALSE; | |
116 | ||
117 | for (c = 0; c < mm->mm_children; c++) { | |
118 | mc = &mm->mm_child[c]; | |
119 | mc->mc_vd = vd->vdev_child[c]; | |
120 | mc->mc_offset = zio->io_offset; | |
121 | } | |
122 | } | |
123 | ||
124 | zio->io_vsd = mm; | |
428870ff | 125 | zio->io_vsd_ops = &vdev_mirror_vsd_ops; |
34dc7c2f BB |
126 | return (mm); |
127 | } | |
128 | ||
34dc7c2f BB |
129 | static int |
130 | vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift) | |
131 | { | |
34dc7c2f | 132 | int numerrors = 0; |
45d1cae3 | 133 | int lasterror = 0; |
d6320ddb | 134 | int c; |
34dc7c2f BB |
135 | |
136 | if (vd->vdev_children == 0) { | |
137 | vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; | |
138 | return (EINVAL); | |
139 | } | |
140 | ||
45d1cae3 | 141 | vdev_open_children(vd); |
34dc7c2f | 142 | |
d6320ddb | 143 | for (c = 0; c < vd->vdev_children; c++) { |
45d1cae3 BB |
144 | vdev_t *cvd = vd->vdev_child[c]; |
145 | ||
146 | if (cvd->vdev_open_error) { | |
147 | lasterror = cvd->vdev_open_error; | |
34dc7c2f BB |
148 | numerrors++; |
149 | continue; | |
150 | } | |
151 | ||
152 | *asize = MIN(*asize - 1, cvd->vdev_asize - 1) + 1; | |
153 | *ashift = MAX(*ashift, cvd->vdev_ashift); | |
154 | } | |
155 | ||
156 | if (numerrors == vd->vdev_children) { | |
157 | vd->vdev_stat.vs_aux = VDEV_AUX_NO_REPLICAS; | |
158 | return (lasterror); | |
159 | } | |
160 | ||
161 | return (0); | |
162 | } | |
163 | ||
164 | static void | |
165 | vdev_mirror_close(vdev_t *vd) | |
166 | { | |
d6320ddb BB |
167 | int c; |
168 | ||
169 | for (c = 0; c < vd->vdev_children; c++) | |
34dc7c2f BB |
170 | vdev_close(vd->vdev_child[c]); |
171 | } | |
172 | ||
173 | static void | |
174 | vdev_mirror_child_done(zio_t *zio) | |
175 | { | |
176 | mirror_child_t *mc = zio->io_private; | |
177 | ||
178 | mc->mc_error = zio->io_error; | |
179 | mc->mc_tried = 1; | |
180 | mc->mc_skipped = 0; | |
181 | } | |
182 | ||
183 | static void | |
184 | vdev_mirror_scrub_done(zio_t *zio) | |
185 | { | |
186 | mirror_child_t *mc = zio->io_private; | |
187 | ||
188 | if (zio->io_error == 0) { | |
d164b209 BB |
189 | zio_t *pio; |
190 | ||
191 | mutex_enter(&zio->io_lock); | |
192 | while ((pio = zio_walk_parents(zio)) != NULL) { | |
193 | mutex_enter(&pio->io_lock); | |
194 | ASSERT3U(zio->io_size, >=, pio->io_size); | |
195 | bcopy(zio->io_data, pio->io_data, pio->io_size); | |
196 | mutex_exit(&pio->io_lock); | |
197 | } | |
198 | mutex_exit(&zio->io_lock); | |
34dc7c2f BB |
199 | } |
200 | ||
201 | zio_buf_free(zio->io_data, zio->io_size); | |
202 | ||
203 | mc->mc_error = zio->io_error; | |
204 | mc->mc_tried = 1; | |
205 | mc->mc_skipped = 0; | |
206 | } | |
207 | ||
34dc7c2f BB |
208 | /* |
209 | * Try to find a child whose DTL doesn't contain the block we want to read. | |
210 | * If we can't, try the read on any vdev we haven't already tried. | |
211 | */ | |
212 | static int | |
213 | vdev_mirror_child_select(zio_t *zio) | |
214 | { | |
215 | mirror_map_t *mm = zio->io_vsd; | |
216 | mirror_child_t *mc; | |
217 | uint64_t txg = zio->io_txg; | |
218 | int i, c; | |
219 | ||
428870ff | 220 | ASSERT(zio->io_bp == NULL || BP_PHYSICAL_BIRTH(zio->io_bp) == txg); |
34dc7c2f BB |
221 | |
222 | /* | |
223 | * Try to find a child whose DTL doesn't contain the block to read. | |
224 | * If a child is known to be completely inaccessible (indicated by | |
225 | * vdev_readable() returning B_FALSE), don't even try. | |
226 | */ | |
227 | for (i = 0, c = mm->mm_preferred; i < mm->mm_children; i++, c++) { | |
228 | if (c >= mm->mm_children) | |
229 | c = 0; | |
230 | mc = &mm->mm_child[c]; | |
231 | if (mc->mc_tried || mc->mc_skipped) | |
232 | continue; | |
b128c09f | 233 | if (!vdev_readable(mc->mc_vd)) { |
34dc7c2f BB |
234 | mc->mc_error = ENXIO; |
235 | mc->mc_tried = 1; /* don't even try */ | |
236 | mc->mc_skipped = 1; | |
237 | continue; | |
238 | } | |
fb5f0bc8 | 239 | if (!vdev_dtl_contains(mc->mc_vd, DTL_MISSING, txg, 1)) |
34dc7c2f BB |
240 | return (c); |
241 | mc->mc_error = ESTALE; | |
242 | mc->mc_skipped = 1; | |
b128c09f | 243 | mc->mc_speculative = 1; |
34dc7c2f BB |
244 | } |
245 | ||
246 | /* | |
247 | * Every device is either missing or has this txg in its DTL. | |
248 | * Look for any child we haven't already tried before giving up. | |
249 | */ | |
250 | for (c = 0; c < mm->mm_children; c++) | |
251 | if (!mm->mm_child[c].mc_tried) | |
252 | return (c); | |
253 | ||
254 | /* | |
255 | * Every child failed. There's no place left to look. | |
256 | */ | |
257 | return (-1); | |
258 | } | |
259 | ||
260 | static int | |
261 | vdev_mirror_io_start(zio_t *zio) | |
262 | { | |
263 | mirror_map_t *mm; | |
264 | mirror_child_t *mc; | |
265 | int c, children; | |
266 | ||
267 | mm = vdev_mirror_map_alloc(zio); | |
268 | ||
269 | if (zio->io_type == ZIO_TYPE_READ) { | |
270 | if ((zio->io_flags & ZIO_FLAG_SCRUB) && !mm->mm_replacing) { | |
271 | /* | |
272 | * For scrubbing reads we need to allocate a read | |
273 | * buffer for each child and issue reads to all | |
274 | * children. If any child succeeds, it will copy its | |
275 | * data into zio->io_data in vdev_mirror_scrub_done. | |
276 | */ | |
277 | for (c = 0; c < mm->mm_children; c++) { | |
278 | mc = &mm->mm_child[c]; | |
279 | zio_nowait(zio_vdev_child_io(zio, zio->io_bp, | |
280 | mc->mc_vd, mc->mc_offset, | |
281 | zio_buf_alloc(zio->io_size), zio->io_size, | |
b128c09f | 282 | zio->io_type, zio->io_priority, 0, |
34dc7c2f BB |
283 | vdev_mirror_scrub_done, mc)); |
284 | } | |
b128c09f | 285 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
286 | } |
287 | /* | |
288 | * For normal reads just pick one child. | |
289 | */ | |
290 | c = vdev_mirror_child_select(zio); | |
291 | children = (c >= 0); | |
292 | } else { | |
293 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
294 | ||
295 | /* | |
fb5f0bc8 | 296 | * Writes go to all children. |
34dc7c2f | 297 | */ |
fb5f0bc8 BB |
298 | c = 0; |
299 | children = mm->mm_children; | |
34dc7c2f BB |
300 | } |
301 | ||
302 | while (children--) { | |
303 | mc = &mm->mm_child[c]; | |
304 | zio_nowait(zio_vdev_child_io(zio, zio->io_bp, | |
b128c09f BB |
305 | mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size, |
306 | zio->io_type, zio->io_priority, 0, | |
307 | vdev_mirror_child_done, mc)); | |
34dc7c2f BB |
308 | c++; |
309 | } | |
310 | ||
b128c09f | 311 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
312 | } |
313 | ||
314 | static int | |
b128c09f BB |
315 | vdev_mirror_worst_error(mirror_map_t *mm) |
316 | { | |
d6320ddb | 317 | int c, error[2] = { 0, 0 }; |
b128c09f | 318 | |
d6320ddb | 319 | for (c = 0; c < mm->mm_children; c++) { |
b128c09f BB |
320 | mirror_child_t *mc = &mm->mm_child[c]; |
321 | int s = mc->mc_speculative; | |
322 | error[s] = zio_worst_error(error[s], mc->mc_error); | |
323 | } | |
324 | ||
325 | return (error[0] ? error[0] : error[1]); | |
326 | } | |
327 | ||
328 | static void | |
34dc7c2f BB |
329 | vdev_mirror_io_done(zio_t *zio) |
330 | { | |
331 | mirror_map_t *mm = zio->io_vsd; | |
332 | mirror_child_t *mc; | |
333 | int c; | |
334 | int good_copies = 0; | |
335 | int unexpected_errors = 0; | |
336 | ||
34dc7c2f BB |
337 | for (c = 0; c < mm->mm_children; c++) { |
338 | mc = &mm->mm_child[c]; | |
339 | ||
34dc7c2f | 340 | if (mc->mc_error) { |
34dc7c2f BB |
341 | if (!mc->mc_skipped) |
342 | unexpected_errors++; | |
b128c09f BB |
343 | } else if (mc->mc_tried) { |
344 | good_copies++; | |
34dc7c2f BB |
345 | } |
346 | } | |
347 | ||
348 | if (zio->io_type == ZIO_TYPE_WRITE) { | |
349 | /* | |
350 | * XXX -- for now, treat partial writes as success. | |
b128c09f BB |
351 | * |
352 | * Now that we support write reallocation, it would be better | |
353 | * to treat partial failure as real failure unless there are | |
354 | * no non-degraded top-level vdevs left, and not update DTLs | |
355 | * if we intend to reallocate. | |
34dc7c2f BB |
356 | */ |
357 | /* XXPOLICY */ | |
b128c09f BB |
358 | if (good_copies != mm->mm_children) { |
359 | /* | |
360 | * Always require at least one good copy. | |
361 | * | |
362 | * For ditto blocks (io_vd == NULL), require | |
363 | * all copies to be good. | |
364 | * | |
365 | * XXX -- for replacing vdevs, there's no great answer. | |
366 | * If the old device is really dead, we may not even | |
367 | * be able to access it -- so we only want to | |
368 | * require good writes to the new device. But if | |
369 | * the new device turns out to be flaky, we want | |
370 | * to be able to detach it -- which requires all | |
371 | * writes to the old device to have succeeded. | |
372 | */ | |
373 | if (good_copies == 0 || zio->io_vd == NULL) | |
374 | zio->io_error = vdev_mirror_worst_error(mm); | |
375 | } | |
376 | return; | |
34dc7c2f BB |
377 | } |
378 | ||
379 | ASSERT(zio->io_type == ZIO_TYPE_READ); | |
380 | ||
381 | /* | |
382 | * If we don't have a good copy yet, keep trying other children. | |
383 | */ | |
384 | /* XXPOLICY */ | |
385 | if (good_copies == 0 && (c = vdev_mirror_child_select(zio)) != -1) { | |
386 | ASSERT(c >= 0 && c < mm->mm_children); | |
387 | mc = &mm->mm_child[c]; | |
34dc7c2f BB |
388 | zio_vdev_io_redone(zio); |
389 | zio_nowait(zio_vdev_child_io(zio, zio->io_bp, | |
390 | mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size, | |
b128c09f | 391 | ZIO_TYPE_READ, zio->io_priority, 0, |
34dc7c2f | 392 | vdev_mirror_child_done, mc)); |
b128c09f | 393 | return; |
34dc7c2f BB |
394 | } |
395 | ||
396 | /* XXPOLICY */ | |
b128c09f BB |
397 | if (good_copies == 0) { |
398 | zio->io_error = vdev_mirror_worst_error(mm); | |
34dc7c2f | 399 | ASSERT(zio->io_error != 0); |
b128c09f | 400 | } |
34dc7c2f | 401 | |
fb5f0bc8 | 402 | if (good_copies && spa_writeable(zio->io_spa) && |
34dc7c2f BB |
403 | (unexpected_errors || |
404 | (zio->io_flags & ZIO_FLAG_RESILVER) || | |
405 | ((zio->io_flags & ZIO_FLAG_SCRUB) && mm->mm_replacing))) { | |
34dc7c2f BB |
406 | /* |
407 | * Use the good data we have in hand to repair damaged children. | |
34dc7c2f | 408 | */ |
34dc7c2f BB |
409 | for (c = 0; c < mm->mm_children; c++) { |
410 | /* | |
411 | * Don't rewrite known good children. | |
412 | * Not only is it unnecessary, it could | |
413 | * actually be harmful: if the system lost | |
414 | * power while rewriting the only good copy, | |
415 | * there would be no good copies left! | |
416 | */ | |
417 | mc = &mm->mm_child[c]; | |
418 | ||
419 | if (mc->mc_error == 0) { | |
420 | if (mc->mc_tried) | |
421 | continue; | |
422 | if (!(zio->io_flags & ZIO_FLAG_SCRUB) && | |
fb5f0bc8 | 423 | !vdev_dtl_contains(mc->mc_vd, DTL_PARTIAL, |
34dc7c2f BB |
424 | zio->io_txg, 1)) |
425 | continue; | |
426 | mc->mc_error = ESTALE; | |
427 | } | |
428 | ||
b128c09f BB |
429 | zio_nowait(zio_vdev_child_io(zio, zio->io_bp, |
430 | mc->mc_vd, mc->mc_offset, | |
431 | zio->io_data, zio->io_size, | |
34dc7c2f | 432 | ZIO_TYPE_WRITE, zio->io_priority, |
fb5f0bc8 BB |
433 | ZIO_FLAG_IO_REPAIR | (unexpected_errors ? |
434 | ZIO_FLAG_SELF_HEAL : 0), NULL, NULL)); | |
34dc7c2f | 435 | } |
34dc7c2f | 436 | } |
34dc7c2f BB |
437 | } |
438 | ||
439 | static void | |
440 | vdev_mirror_state_change(vdev_t *vd, int faulted, int degraded) | |
441 | { | |
442 | if (faulted == vd->vdev_children) | |
443 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
444 | VDEV_AUX_NO_REPLICAS); | |
445 | else if (degraded + faulted != 0) | |
446 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, VDEV_AUX_NONE); | |
447 | else | |
448 | vdev_set_state(vd, B_FALSE, VDEV_STATE_HEALTHY, VDEV_AUX_NONE); | |
449 | } | |
450 | ||
451 | vdev_ops_t vdev_mirror_ops = { | |
452 | vdev_mirror_open, | |
453 | vdev_mirror_close, | |
34dc7c2f BB |
454 | vdev_default_asize, |
455 | vdev_mirror_io_start, | |
456 | vdev_mirror_io_done, | |
457 | vdev_mirror_state_change, | |
428870ff BB |
458 | NULL, |
459 | NULL, | |
34dc7c2f BB |
460 | VDEV_TYPE_MIRROR, /* name of this vdev type */ |
461 | B_FALSE /* not a leaf vdev */ | |
462 | }; | |
463 | ||
464 | vdev_ops_t vdev_replacing_ops = { | |
465 | vdev_mirror_open, | |
466 | vdev_mirror_close, | |
34dc7c2f BB |
467 | vdev_default_asize, |
468 | vdev_mirror_io_start, | |
469 | vdev_mirror_io_done, | |
470 | vdev_mirror_state_change, | |
428870ff BB |
471 | NULL, |
472 | NULL, | |
34dc7c2f BB |
473 | VDEV_TYPE_REPLACING, /* name of this vdev type */ |
474 | B_FALSE /* not a leaf vdev */ | |
475 | }; | |
476 | ||
477 | vdev_ops_t vdev_spare_ops = { | |
478 | vdev_mirror_open, | |
479 | vdev_mirror_close, | |
34dc7c2f BB |
480 | vdev_default_asize, |
481 | vdev_mirror_io_start, | |
482 | vdev_mirror_io_done, | |
483 | vdev_mirror_state_change, | |
428870ff BB |
484 | NULL, |
485 | NULL, | |
34dc7c2f BB |
486 | VDEV_TYPE_SPARE, /* name of this vdev type */ |
487 | B_FALSE /* not a leaf vdev */ | |
488 | }; |