]>
Commit | Line | Data |
---|---|---|
70e083d2 TG |
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | * Copyright (c) 2012, 2016 by Delphix. All rights reserved. | |
24 | */ | |
25 | ||
26 | #include <sys/zfs_context.h> | |
27 | #include <sys/dmu_objset.h> | |
28 | #include <sys/dmu_traverse.h> | |
29 | #include <sys/dsl_dataset.h> | |
30 | #include <sys/dsl_dir.h> | |
31 | #include <sys/dsl_pool.h> | |
32 | #include <sys/dnode.h> | |
33 | #include <sys/spa.h> | |
2dae45ea | 34 | #include <sys/spa_impl.h> |
70e083d2 TG |
35 | #include <sys/zio.h> |
36 | #include <sys/dmu_impl.h> | |
37 | #include <sys/sa.h> | |
38 | #include <sys/sa_impl.h> | |
39 | #include <sys/callb.h> | |
40 | #include <sys/zfeature.h> | |
41 | ||
42 | int32_t zfs_pd_bytes_max = 50 * 1024 * 1024; /* 50MB */ | |
86e3c28a | 43 | int32_t send_holes_without_birth_time = 1; |
70e083d2 TG |
44 | |
45 | typedef struct prefetch_data { | |
46 | kmutex_t pd_mtx; | |
47 | kcondvar_t pd_cv; | |
48 | int32_t pd_bytes_fetched; | |
49 | int pd_flags; | |
50 | boolean_t pd_cancel; | |
51 | boolean_t pd_exited; | |
86e3c28a | 52 | zbookmark_phys_t pd_resume; |
70e083d2 TG |
53 | } prefetch_data_t; |
54 | ||
55 | typedef struct traverse_data { | |
56 | spa_t *td_spa; | |
57 | uint64_t td_objset; | |
58 | blkptr_t *td_rootbp; | |
59 | uint64_t td_min_txg; | |
60 | zbookmark_phys_t *td_resume; | |
61 | int td_flags; | |
62 | prefetch_data_t *td_pfd; | |
63 | boolean_t td_paused; | |
64 | uint64_t td_hole_birth_enabled_txg; | |
65 | blkptr_cb_t *td_func; | |
66 | void *td_arg; | |
67 | boolean_t td_realloc_possible; | |
68 | } traverse_data_t; | |
69 | ||
70 | static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, | |
71 | uint64_t objset, uint64_t object); | |
72 | static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *, | |
73 | uint64_t objset, uint64_t object); | |
74 | ||
75 | static int | |
76 | traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) | |
77 | { | |
78 | traverse_data_t *td = arg; | |
79 | zbookmark_phys_t zb; | |
80 | ||
81 | if (BP_IS_HOLE(bp)) | |
82 | return (0); | |
83 | ||
84 | if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa)) | |
85 | return (0); | |
86 | ||
87 | SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, | |
88 | bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
89 | ||
90 | (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg); | |
91 | ||
92 | return (0); | |
93 | } | |
94 | ||
95 | static int | |
96 | traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) | |
97 | { | |
98 | traverse_data_t *td = arg; | |
99 | ||
100 | if (lrc->lrc_txtype == TX_WRITE) { | |
101 | lr_write_t *lr = (lr_write_t *)lrc; | |
102 | blkptr_t *bp = &lr->lr_blkptr; | |
103 | zbookmark_phys_t zb; | |
104 | ||
105 | if (BP_IS_HOLE(bp)) | |
106 | return (0); | |
107 | ||
108 | if (claim_txg == 0 || bp->blk_birth < claim_txg) | |
109 | return (0); | |
110 | ||
111 | SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid, | |
112 | ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); | |
113 | ||
114 | (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, | |
115 | td->td_arg); | |
116 | } | |
117 | return (0); | |
118 | } | |
119 | ||
120 | static void | |
121 | traverse_zil(traverse_data_t *td, zil_header_t *zh) | |
122 | { | |
123 | uint64_t claim_txg = zh->zh_claim_txg; | |
124 | zilog_t *zilog; | |
125 | ||
126 | /* | |
127 | * We only want to visit blocks that have been claimed but not yet | |
128 | * replayed; plus, in read-only mode, blocks that are already stable. | |
129 | */ | |
130 | if (claim_txg == 0 && spa_writeable(td->td_spa)) | |
131 | return; | |
132 | ||
133 | zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh); | |
134 | ||
135 | (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td, | |
136 | claim_txg); | |
137 | ||
138 | zil_free(zilog); | |
139 | } | |
140 | ||
141 | typedef enum resume_skip { | |
142 | RESUME_SKIP_ALL, | |
143 | RESUME_SKIP_NONE, | |
144 | RESUME_SKIP_CHILDREN | |
145 | } resume_skip_t; | |
146 | ||
147 | /* | |
148 | * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and | |
149 | * the block indicated by zb does not need to be visited at all. Returns | |
150 | * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the | |
151 | * resume point. This indicates that this block should be visited but not its | |
152 | * children (since they must have been visited in a previous traversal). | |
153 | * Otherwise returns RESUME_SKIP_NONE. | |
154 | */ | |
155 | static resume_skip_t | |
156 | resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp, | |
157 | const zbookmark_phys_t *zb) | |
158 | { | |
159 | if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) { | |
160 | /* | |
161 | * If we already visited this bp & everything below, | |
162 | * don't bother doing it again. | |
163 | */ | |
86e3c28a | 164 | if (zbookmark_subtree_completed(dnp, zb, td->td_resume)) |
70e083d2 TG |
165 | return (RESUME_SKIP_ALL); |
166 | ||
167 | /* | |
168 | * If we found the block we're trying to resume from, zero | |
169 | * the bookmark out to indicate that we have resumed. | |
170 | */ | |
171 | if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) { | |
172 | bzero(td->td_resume, sizeof (*zb)); | |
173 | if (td->td_flags & TRAVERSE_POST) | |
174 | return (RESUME_SKIP_CHILDREN); | |
175 | } | |
176 | } | |
177 | return (RESUME_SKIP_NONE); | |
178 | } | |
179 | ||
180 | static void | |
181 | traverse_prefetch_metadata(traverse_data_t *td, | |
182 | const blkptr_t *bp, const zbookmark_phys_t *zb) | |
183 | { | |
184 | arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; | |
185 | ||
186 | if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA)) | |
187 | return; | |
188 | /* | |
189 | * If we are in the process of resuming, don't prefetch, because | |
190 | * some children will not be needed (and in fact may have already | |
191 | * been freed). | |
192 | */ | |
193 | if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) | |
194 | return; | |
195 | if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg) | |
196 | return; | |
197 | if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE) | |
198 | return; | |
199 | ||
200 | (void) arc_read(NULL, td->td_spa, bp, NULL, NULL, | |
201 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); | |
202 | } | |
203 | ||
204 | static boolean_t | |
205 | prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp) | |
206 | { | |
207 | ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA); | |
208 | if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) || | |
209 | BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) | |
210 | return (B_FALSE); | |
211 | return (B_TRUE); | |
212 | } | |
213 | ||
214 | static int | |
215 | traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, | |
216 | const blkptr_t *bp, const zbookmark_phys_t *zb) | |
217 | { | |
218 | int err = 0; | |
219 | arc_buf_t *buf = NULL; | |
220 | prefetch_data_t *pd = td->td_pfd; | |
221 | ||
222 | switch (resume_skip_check(td, dnp, zb)) { | |
223 | case RESUME_SKIP_ALL: | |
224 | return (0); | |
225 | case RESUME_SKIP_CHILDREN: | |
226 | goto post; | |
227 | case RESUME_SKIP_NONE: | |
228 | break; | |
229 | default: | |
230 | ASSERT(0); | |
231 | } | |
232 | ||
233 | if (bp->blk_birth == 0) { | |
234 | /* | |
235 | * Since this block has a birth time of 0 it must be one of | |
236 | * two things: a hole created before the | |
237 | * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole | |
238 | * which has always been a hole in an object. | |
239 | * | |
240 | * If a file is written sparsely, then the unwritten parts of | |
241 | * the file were "always holes" -- that is, they have been | |
242 | * holes since this object was allocated. However, we (and | |
243 | * our callers) can not necessarily tell when an object was | |
244 | * allocated. Therefore, if it's possible that this object | |
245 | * was freed and then its object number reused, we need to | |
246 | * visit all the holes with birth==0. | |
247 | * | |
248 | * If it isn't possible that the object number was reused, | |
249 | * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote | |
250 | * all the blocks we will visit as part of this traversal, | |
251 | * then this hole must have always existed, so we can skip | |
252 | * it. We visit blocks born after (exclusive) td_min_txg. | |
253 | * | |
254 | * Note that the meta-dnode cannot be reallocated. | |
255 | */ | |
86e3c28a CIK |
256 | if (!send_holes_without_birth_time && |
257 | (!td->td_realloc_possible || | |
258 | zb->zb_object == DMU_META_DNODE_OBJECT) && | |
259 | td->td_hole_birth_enabled_txg <= td->td_min_txg) | |
70e083d2 TG |
260 | return (0); |
261 | } else if (bp->blk_birth <= td->td_min_txg) { | |
262 | return (0); | |
263 | } | |
264 | ||
265 | if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) { | |
266 | uint64_t size = BP_GET_LSIZE(bp); | |
267 | mutex_enter(&pd->pd_mtx); | |
268 | ASSERT(pd->pd_bytes_fetched >= 0); | |
269 | while (pd->pd_bytes_fetched < size && !pd->pd_exited) | |
270 | cv_wait_sig(&pd->pd_cv, &pd->pd_mtx); | |
271 | pd->pd_bytes_fetched -= size; | |
272 | cv_broadcast(&pd->pd_cv); | |
273 | mutex_exit(&pd->pd_mtx); | |
274 | } | |
275 | ||
276 | if (BP_IS_HOLE(bp)) { | |
277 | err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); | |
278 | if (err != 0) | |
279 | goto post; | |
280 | return (0); | |
281 | } | |
282 | ||
283 | if (td->td_flags & TRAVERSE_PRE) { | |
284 | err = td->td_func(td->td_spa, NULL, bp, zb, dnp, | |
285 | td->td_arg); | |
286 | if (err == TRAVERSE_VISIT_NO_CHILDREN) | |
287 | return (0); | |
288 | if (err != 0) | |
289 | goto post; | |
290 | } | |
291 | ||
292 | if (BP_GET_LEVEL(bp) > 0) { | |
293 | uint32_t flags = ARC_FLAG_WAIT; | |
294 | int32_t i; | |
295 | int32_t epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; | |
296 | zbookmark_phys_t *czb; | |
297 | ||
298 | err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, | |
299 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); | |
300 | if (err != 0) | |
301 | goto post; | |
302 | ||
303 | czb = kmem_alloc(sizeof (zbookmark_phys_t), KM_SLEEP); | |
304 | ||
305 | for (i = 0; i < epb; i++) { | |
306 | SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object, | |
307 | zb->zb_level - 1, | |
308 | zb->zb_blkid * epb + i); | |
309 | traverse_prefetch_metadata(td, | |
310 | &((blkptr_t *)buf->b_data)[i], czb); | |
311 | } | |
312 | ||
313 | /* recursively visitbp() blocks below this */ | |
314 | for (i = 0; i < epb; i++) { | |
315 | SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object, | |
316 | zb->zb_level - 1, | |
317 | zb->zb_blkid * epb + i); | |
318 | err = traverse_visitbp(td, dnp, | |
319 | &((blkptr_t *)buf->b_data)[i], czb); | |
320 | if (err != 0) | |
321 | break; | |
322 | } | |
323 | ||
324 | kmem_free(czb, sizeof (zbookmark_phys_t)); | |
325 | ||
326 | } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { | |
327 | uint32_t flags = ARC_FLAG_WAIT; | |
328 | int32_t i; | |
329 | int32_t epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; | |
86e3c28a | 330 | dnode_phys_t *child_dnp; |
70e083d2 TG |
331 | |
332 | err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, | |
333 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); | |
334 | if (err != 0) | |
335 | goto post; | |
86e3c28a | 336 | child_dnp = buf->b_data; |
70e083d2 | 337 | |
86e3c28a CIK |
338 | for (i = 0; i < epb; i += child_dnp[i].dn_extra_slots + 1) { |
339 | prefetch_dnode_metadata(td, &child_dnp[i], | |
340 | zb->zb_objset, zb->zb_blkid * epb + i); | |
70e083d2 TG |
341 | } |
342 | ||
343 | /* recursively visitbp() blocks below this */ | |
86e3c28a CIK |
344 | for (i = 0; i < epb; i += child_dnp[i].dn_extra_slots + 1) { |
345 | err = traverse_dnode(td, &child_dnp[i], | |
346 | zb->zb_objset, zb->zb_blkid * epb + i); | |
70e083d2 TG |
347 | if (err != 0) |
348 | break; | |
349 | } | |
350 | } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { | |
351 | arc_flags_t flags = ARC_FLAG_WAIT; | |
352 | objset_phys_t *osp; | |
70e083d2 TG |
353 | |
354 | err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, | |
355 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); | |
356 | if (err != 0) | |
357 | goto post; | |
358 | ||
359 | osp = buf->b_data; | |
86e3c28a | 360 | prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset, |
70e083d2 TG |
361 | DMU_META_DNODE_OBJECT); |
362 | /* | |
363 | * See the block comment above for the goal of this variable. | |
364 | * If the maxblkid of the meta-dnode is 0, then we know that | |
365 | * we've never had more than DNODES_PER_BLOCK objects in the | |
366 | * dataset, which means we can't have reused any object ids. | |
367 | */ | |
368 | if (osp->os_meta_dnode.dn_maxblkid == 0) | |
369 | td->td_realloc_possible = B_FALSE; | |
370 | ||
371 | if (arc_buf_size(buf) >= sizeof (objset_phys_t)) { | |
86e3c28a CIK |
372 | prefetch_dnode_metadata(td, &osp->os_groupused_dnode, |
373 | zb->zb_objset, DMU_GROUPUSED_OBJECT); | |
374 | prefetch_dnode_metadata(td, &osp->os_userused_dnode, | |
375 | zb->zb_objset, DMU_USERUSED_OBJECT); | |
70e083d2 TG |
376 | } |
377 | ||
86e3c28a | 378 | err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset, |
70e083d2 TG |
379 | DMU_META_DNODE_OBJECT); |
380 | if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { | |
86e3c28a CIK |
381 | err = traverse_dnode(td, &osp->os_groupused_dnode, |
382 | zb->zb_objset, DMU_GROUPUSED_OBJECT); | |
70e083d2 TG |
383 | } |
384 | if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { | |
86e3c28a CIK |
385 | err = traverse_dnode(td, &osp->os_userused_dnode, |
386 | zb->zb_objset, DMU_USERUSED_OBJECT); | |
70e083d2 TG |
387 | } |
388 | } | |
389 | ||
390 | if (buf) | |
86e3c28a | 391 | arc_buf_destroy(buf, &buf); |
70e083d2 TG |
392 | |
393 | post: | |
394 | if (err == 0 && (td->td_flags & TRAVERSE_POST)) | |
395 | err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); | |
396 | ||
397 | if ((td->td_flags & TRAVERSE_HARD) && (err == EIO || err == ECKSUM)) { | |
398 | /* | |
399 | * Ignore this disk error as requested by the HARD flag, | |
400 | * and continue traversal. | |
401 | */ | |
402 | err = 0; | |
403 | } | |
404 | ||
405 | /* | |
406 | * If we are stopping here, set td_resume. | |
407 | */ | |
408 | if (td->td_resume != NULL && err != 0 && !td->td_paused) { | |
409 | td->td_resume->zb_objset = zb->zb_objset; | |
410 | td->td_resume->zb_object = zb->zb_object; | |
411 | td->td_resume->zb_level = 0; | |
412 | /* | |
413 | * If we have stopped on an indirect block (e.g. due to | |
414 | * i/o error), we have not visited anything below it. | |
415 | * Set the bookmark to the first level-0 block that we need | |
416 | * to visit. This way, the resuming code does not need to | |
417 | * deal with resuming from indirect blocks. | |
86e3c28a CIK |
418 | * |
419 | * Note, if zb_level <= 0, dnp may be NULL, so we don't want | |
420 | * to dereference it. | |
70e083d2 | 421 | */ |
86e3c28a CIK |
422 | td->td_resume->zb_blkid = zb->zb_blkid; |
423 | if (zb->zb_level > 0) { | |
424 | td->td_resume->zb_blkid <<= zb->zb_level * | |
425 | (dnp->dn_indblkshift - SPA_BLKPTRSHIFT); | |
426 | } | |
70e083d2 TG |
427 | td->td_paused = B_TRUE; |
428 | } | |
429 | ||
430 | return (err); | |
431 | } | |
432 | ||
433 | static void | |
434 | prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp, | |
435 | uint64_t objset, uint64_t object) | |
436 | { | |
437 | int j; | |
438 | zbookmark_phys_t czb; | |
439 | ||
440 | for (j = 0; j < dnp->dn_nblkptr; j++) { | |
441 | SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); | |
442 | traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb); | |
443 | } | |
444 | ||
445 | if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { | |
446 | SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); | |
86e3c28a | 447 | traverse_prefetch_metadata(td, DN_SPILL_BLKPTR(dnp), &czb); |
70e083d2 TG |
448 | } |
449 | } | |
450 | ||
451 | static int | |
452 | traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, | |
453 | uint64_t objset, uint64_t object) | |
454 | { | |
455 | int j, err = 0; | |
456 | zbookmark_phys_t czb; | |
457 | ||
86e3c28a CIK |
458 | if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL && |
459 | object < td->td_resume->zb_object) | |
460 | return (0); | |
461 | ||
462 | if (td->td_flags & TRAVERSE_PRE) { | |
463 | SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL, | |
464 | ZB_DNODE_BLKID); | |
465 | err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp, | |
466 | td->td_arg); | |
467 | if (err == TRAVERSE_VISIT_NO_CHILDREN) | |
468 | return (0); | |
469 | if (err != 0) | |
470 | return (err); | |
471 | } | |
472 | ||
70e083d2 TG |
473 | for (j = 0; j < dnp->dn_nblkptr; j++) { |
474 | SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); | |
475 | err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb); | |
476 | if (err != 0) | |
477 | break; | |
478 | } | |
479 | ||
86e3c28a | 480 | if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { |
70e083d2 | 481 | SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); |
86e3c28a CIK |
482 | err = traverse_visitbp(td, dnp, DN_SPILL_BLKPTR(dnp), &czb); |
483 | } | |
484 | ||
485 | if (err == 0 && (td->td_flags & TRAVERSE_POST)) { | |
486 | SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL, | |
487 | ZB_DNODE_BLKID); | |
488 | err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp, | |
489 | td->td_arg); | |
490 | if (err == TRAVERSE_VISIT_NO_CHILDREN) | |
491 | return (0); | |
492 | if (err != 0) | |
493 | return (err); | |
70e083d2 TG |
494 | } |
495 | return (err); | |
496 | } | |
497 | ||
498 | /* ARGSUSED */ | |
499 | static int | |
500 | traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
501 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
502 | { | |
503 | prefetch_data_t *pfd = arg; | |
504 | arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; | |
505 | ||
506 | ASSERT(pfd->pd_bytes_fetched >= 0); | |
86e3c28a CIK |
507 | if (bp == NULL) |
508 | return (0); | |
70e083d2 TG |
509 | if (pfd->pd_cancel) |
510 | return (SET_ERROR(EINTR)); | |
511 | ||
512 | if (!prefetch_needed(pfd, bp)) | |
513 | return (0); | |
514 | ||
515 | mutex_enter(&pfd->pd_mtx); | |
516 | while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max) | |
517 | cv_wait_sig(&pfd->pd_cv, &pfd->pd_mtx); | |
518 | pfd->pd_bytes_fetched += BP_GET_LSIZE(bp); | |
519 | cv_broadcast(&pfd->pd_cv); | |
520 | mutex_exit(&pfd->pd_mtx); | |
521 | ||
522 | (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ, | |
523 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb); | |
524 | ||
525 | return (0); | |
526 | } | |
527 | ||
528 | static void | |
529 | traverse_prefetch_thread(void *arg) | |
530 | { | |
531 | traverse_data_t *td_main = arg; | |
532 | traverse_data_t td = *td_main; | |
533 | zbookmark_phys_t czb; | |
534 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
535 | ||
536 | td.td_func = traverse_prefetcher; | |
537 | td.td_arg = td_main->td_pfd; | |
538 | td.td_pfd = NULL; | |
86e3c28a | 539 | td.td_resume = &td_main->td_pfd->pd_resume; |
70e083d2 TG |
540 | |
541 | SET_BOOKMARK(&czb, td.td_objset, | |
542 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
543 | (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb); | |
544 | ||
545 | mutex_enter(&td_main->td_pfd->pd_mtx); | |
546 | td_main->td_pfd->pd_exited = B_TRUE; | |
547 | cv_broadcast(&td_main->td_pfd->pd_cv); | |
548 | mutex_exit(&td_main->td_pfd->pd_mtx); | |
549 | spl_fstrans_unmark(cookie); | |
550 | } | |
551 | ||
552 | /* | |
553 | * NB: dataset must not be changing on-disk (eg, is a snapshot or we are | |
554 | * in syncing context). | |
555 | */ | |
556 | static int | |
557 | traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp, | |
558 | uint64_t txg_start, zbookmark_phys_t *resume, int flags, | |
559 | blkptr_cb_t func, void *arg) | |
560 | { | |
561 | traverse_data_t *td; | |
562 | prefetch_data_t *pd; | |
563 | zbookmark_phys_t *czb; | |
564 | int err; | |
565 | ||
566 | ASSERT(ds == NULL || objset == ds->ds_object); | |
567 | ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST)); | |
568 | ||
70e083d2 TG |
569 | td = kmem_alloc(sizeof (traverse_data_t), KM_SLEEP); |
570 | pd = kmem_zalloc(sizeof (prefetch_data_t), KM_SLEEP); | |
571 | czb = kmem_alloc(sizeof (zbookmark_phys_t), KM_SLEEP); | |
572 | ||
573 | td->td_spa = spa; | |
574 | td->td_objset = objset; | |
575 | td->td_rootbp = rootbp; | |
576 | td->td_min_txg = txg_start; | |
577 | td->td_resume = resume; | |
578 | td->td_func = func; | |
579 | td->td_arg = arg; | |
580 | td->td_pfd = pd; | |
581 | td->td_flags = flags; | |
582 | td->td_paused = B_FALSE; | |
583 | td->td_realloc_possible = (txg_start == 0 ? B_FALSE : B_TRUE); | |
584 | ||
585 | if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { | |
586 | VERIFY(spa_feature_enabled_txg(spa, | |
587 | SPA_FEATURE_HOLE_BIRTH, &td->td_hole_birth_enabled_txg)); | |
588 | } else { | |
589 | td->td_hole_birth_enabled_txg = UINT64_MAX; | |
590 | } | |
591 | ||
592 | pd->pd_flags = flags; | |
86e3c28a CIK |
593 | if (resume != NULL) |
594 | pd->pd_resume = *resume; | |
70e083d2 TG |
595 | mutex_init(&pd->pd_mtx, NULL, MUTEX_DEFAULT, NULL); |
596 | cv_init(&pd->pd_cv, NULL, CV_DEFAULT, NULL); | |
597 | ||
598 | SET_BOOKMARK(czb, td->td_objset, | |
599 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
600 | ||
601 | /* See comment on ZIL traversal in dsl_scan_visitds. */ | |
602 | if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) { | |
603 | uint32_t flags = ARC_FLAG_WAIT; | |
604 | objset_phys_t *osp; | |
605 | arc_buf_t *buf; | |
606 | ||
607 | err = arc_read(NULL, td->td_spa, rootbp, | |
608 | arc_getbuf_func, &buf, | |
609 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, czb); | |
610 | if (err != 0) | |
611 | return (err); | |
612 | ||
b49151d6 CIK |
613 | if (err != 0) { |
614 | /* | |
615 | * If both TRAVERSE_HARD and TRAVERSE_PRE are set, | |
616 | * continue to visitbp so that td_func can be called | |
617 | * in pre stage, and err will reset to zero. | |
618 | */ | |
619 | if (!(td->td_flags & TRAVERSE_HARD) || | |
620 | !(td->td_flags & TRAVERSE_PRE)) | |
621 | return (err); | |
622 | } else { | |
623 | osp = buf->b_data; | |
624 | traverse_zil(td, &osp->os_zil_header); | |
625 | arc_buf_destroy(buf, &buf); | |
626 | } | |
70e083d2 TG |
627 | } |
628 | ||
629 | if (!(flags & TRAVERSE_PREFETCH_DATA) || | |
2dae45ea | 630 | taskq_dispatch(spa->spa_prefetch_taskq, traverse_prefetch_thread, |
86e3c28a | 631 | td, TQ_NOQUEUE) == TASKQID_INVALID) |
70e083d2 TG |
632 | pd->pd_exited = B_TRUE; |
633 | ||
634 | err = traverse_visitbp(td, NULL, rootbp, czb); | |
635 | ||
636 | mutex_enter(&pd->pd_mtx); | |
637 | pd->pd_cancel = B_TRUE; | |
638 | cv_broadcast(&pd->pd_cv); | |
639 | while (!pd->pd_exited) | |
640 | cv_wait_sig(&pd->pd_cv, &pd->pd_mtx); | |
641 | mutex_exit(&pd->pd_mtx); | |
642 | ||
643 | mutex_destroy(&pd->pd_mtx); | |
644 | cv_destroy(&pd->pd_cv); | |
645 | ||
646 | kmem_free(czb, sizeof (zbookmark_phys_t)); | |
647 | kmem_free(pd, sizeof (struct prefetch_data)); | |
648 | kmem_free(td, sizeof (struct traverse_data)); | |
649 | ||
650 | return (err); | |
651 | } | |
652 | ||
653 | /* | |
654 | * NB: dataset must not be changing on-disk (eg, is a snapshot or we are | |
655 | * in syncing context). | |
656 | */ | |
657 | int | |
86e3c28a CIK |
658 | traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start, |
659 | zbookmark_phys_t *resume, | |
660 | int flags, blkptr_cb_t func, void *arg) | |
70e083d2 TG |
661 | { |
662 | return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object, | |
86e3c28a CIK |
663 | &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg)); |
664 | } | |
665 | ||
666 | int | |
667 | traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, | |
668 | int flags, blkptr_cb_t func, void *arg) | |
669 | { | |
670 | return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg)); | |
70e083d2 TG |
671 | } |
672 | ||
673 | int | |
674 | traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr, | |
675 | uint64_t txg_start, zbookmark_phys_t *resume, int flags, | |
676 | blkptr_cb_t func, void *arg) | |
677 | { | |
678 | return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET, | |
679 | blkptr, txg_start, resume, flags, func, arg)); | |
680 | } | |
681 | ||
682 | /* | |
683 | * NB: pool must not be changing on-disk (eg, from zdb or sync context). | |
684 | */ | |
685 | int | |
686 | traverse_pool(spa_t *spa, uint64_t txg_start, int flags, | |
687 | blkptr_cb_t func, void *arg) | |
688 | { | |
689 | int err; | |
690 | uint64_t obj; | |
691 | dsl_pool_t *dp = spa_get_dsl(spa); | |
692 | objset_t *mos = dp->dp_meta_objset; | |
693 | boolean_t hard = (flags & TRAVERSE_HARD); | |
694 | ||
695 | /* visit the MOS */ | |
696 | err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa), | |
697 | txg_start, NULL, flags, func, arg); | |
698 | if (err != 0) | |
699 | return (err); | |
700 | ||
701 | /* visit each dataset */ | |
702 | for (obj = 1; err == 0; | |
86e3c28a | 703 | err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) { |
70e083d2 TG |
704 | dmu_object_info_t doi; |
705 | ||
706 | err = dmu_object_info(mos, obj, &doi); | |
707 | if (err != 0) { | |
708 | if (hard) | |
709 | continue; | |
710 | break; | |
711 | } | |
712 | ||
713 | if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) { | |
714 | dsl_dataset_t *ds; | |
715 | uint64_t txg = txg_start; | |
716 | ||
717 | dsl_pool_config_enter(dp, FTAG); | |
718 | err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); | |
719 | dsl_pool_config_exit(dp, FTAG); | |
720 | if (err != 0) { | |
721 | if (hard) | |
722 | continue; | |
723 | break; | |
724 | } | |
725 | if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg) | |
726 | txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; | |
727 | err = traverse_dataset(ds, txg, flags, func, arg); | |
728 | dsl_dataset_rele(ds, FTAG); | |
729 | if (err != 0) | |
730 | break; | |
731 | } | |
732 | } | |
733 | if (err == ESRCH) | |
734 | err = 0; | |
735 | return (err); | |
736 | } | |
737 | ||
738 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
739 | EXPORT_SYMBOL(traverse_dataset); | |
740 | EXPORT_SYMBOL(traverse_pool); | |
741 | ||
742 | module_param(zfs_pd_bytes_max, int, 0644); | |
743 | MODULE_PARM_DESC(zfs_pd_bytes_max, "Max number of bytes to prefetch"); | |
744 | ||
86e3c28a CIK |
745 | module_param_named(ignore_hole_birth, send_holes_without_birth_time, int, 0644); |
746 | MODULE_PARM_DESC(ignore_hole_birth, "Alias for send_holes_without_birth_time"); | |
747 | ||
748 | module_param_named(send_holes_without_birth_time, | |
749 | send_holes_without_birth_time, int, 0644); | |
750 | MODULE_PARM_DESC(send_holes_without_birth_time, | |
751 | "Ignore hole_birth txg for zfs send"); | |
70e083d2 | 752 | #endif |