Print this page
NEX-9752 backport illumos 6950 ARC should cache compressed data
Reviewed by: Saso Kiselkov <saso.kiselkov@nexenta.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
6950 ARC should cache compressed data
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Reviewed by: Don Brady <don.brady@intel.com>
Reviewed by: Richard Elling <Richard.Elling@RichardElling.com>
Approved by: Richard Lowe <richlowe@richlowe.net>
NEX-7603 Back port OpenZFS #188 Create tunable to ignore hole_birth
feature
Reviewed by: Roman Strashkin <roman.strashkin@nexenta.com>
Reviewed by: Dan Fields <dan.fields@nexenta.com>
2605 want to resume interrupted zfs send
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Reviewed by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed by: Xin Li <delphij@freebsd.org>
Reviewed by: Arne Jansen <sensille@gmx.net>
Approved by: Dan McDonald <danmcd@omniti.com>
NEX-4582 update wrc test cases for allow to use write back cache per tree of datasets
Reviewed by: Steve Peng <steve.peng@nexenta.com>
Reviewed by: Alex Aizman <alex.aizman@nexenta.com>
5960 zfs recv should prefetch indirect blocks
5925 zfs receive -o origin=
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
NEX-4207 WRC and dedup on the same pool cause system-panic
Reviewed by: Alek Pinchuk <alek.pinchuk@nexenta.com>
Reviewed by: Alex Aizman <alex.aizman@nexenta.com>
NEX-4193 WRC does not migrate data that belong to intermediate snapshots
Reviewed by: Alek Pinchuk <alek.pinchuk@nexenta.com>
Reviewed by: Alex Aizman <alex.aizman@nexenta.com>
NEX-3710 WRC improvements and bug-fixes
* refactored WRC move-logic to use zio kmem_cashes
* replace size and compression fields by blk_prop field
(the same in blkptr_t) to little reduce size of wrc_block_t
and use similar macros as for blkptr_t to get PSIZE, LSIZE
and COMPRESSION
* make CPU more happy by reduce atomic calls
* removed unused code
* fixed naming of variables
* fixed possible system panic after restart system
with enabled WRC
* fixed a race that causes system panic
Reviewed by: Alek Pinchuk <alek@nexenta.com>
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
NEX-3558 KRRP Integration
4459 Typo in itadm(1m) usage message: delete-inititator
Reviewed by: Milan Jurik <milan.jurik@xylab.cz>
Reviewed by: Marcel Telka <marcel@telka.sk>
Approved by: Robert Mustacchi <rm@joyent.com>
4504 traverse_visitbp: visit DMU_GROUPUSED_OBJECT before DMU_USERUSED_OBJECT
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Andriy Gapon <andriy.gapon@hybridcluster.com>
Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com>
Approved by: Robert Mustacchi <rm@joyent.com>
4391 panic system rather than corrupting pool if we hit bug 4390
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Approved by: Gordon Ross <gwr@nexenta.com>
4370 avoid transmitting holes during zfs send
4371 DMU code clean up
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
Approved by: Garrett D'Amore <garrett@damore.org>
re #12619 rb4429 More dp->dp_config_rwlock holds
re #12585 rb4049 ZFS++ work port - refactoring to improve separation of open/closed code, bug fixes, performance improvements - open code
Bug 11205: add missing libzfs_closed_stubs.c to fix opensource-only build.
ZFS plus work: special vdevs, cos, cos/vdev properties
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/dmu_traverse.c
+++ new/usr/src/uts/common/fs/zfs/dmu_traverse.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 + * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
23 24 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
24 25 */
25 26
26 27 #include <sys/zfs_context.h>
27 28 #include <sys/dmu_objset.h>
28 29 #include <sys/dmu_traverse.h>
29 30 #include <sys/dsl_dataset.h>
30 31 #include <sys/dsl_dir.h>
31 32 #include <sys/dsl_pool.h>
32 33 #include <sys/dnode.h>
33 34 #include <sys/spa.h>
34 35 #include <sys/zio.h>
35 36 #include <sys/dmu_impl.h>
36 37 #include <sys/sa.h>
37 38 #include <sys/sa_impl.h>
38 39 #include <sys/callb.h>
39 40 #include <sys/zfeature.h>
40 41
41 42 int32_t zfs_pd_bytes_max = 50 * 1024 * 1024; /* 50MB */
42 43 boolean_t send_holes_without_birth_time = B_TRUE;
43 44
44 45 typedef struct prefetch_data {
45 46 kmutex_t pd_mtx;
46 47 kcondvar_t pd_cv;
47 48 int32_t pd_bytes_fetched;
48 49 int pd_flags;
|
↓ open down ↓ |
16 lines elided |
↑ open up ↑ |
49 50 boolean_t pd_cancel;
50 51 boolean_t pd_exited;
51 52 zbookmark_phys_t pd_resume;
52 53 } prefetch_data_t;
53 54
54 55 typedef struct traverse_data {
55 56 spa_t *td_spa;
56 57 uint64_t td_objset;
57 58 blkptr_t *td_rootbp;
58 59 uint64_t td_min_txg;
60 + uint64_t td_max_txg;
59 61 zbookmark_phys_t *td_resume;
60 62 int td_flags;
61 63 prefetch_data_t *td_pfd;
62 64 boolean_t td_paused;
63 65 uint64_t td_hole_birth_enabled_txg;
64 66 blkptr_cb_t *td_func;
65 67 void *td_arg;
66 68 boolean_t td_realloc_possible;
67 69 } traverse_data_t;
68 70
69 71 static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
70 72 uint64_t objset, uint64_t object);
71 73 static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *,
72 74 uint64_t objset, uint64_t object);
73 75
74 76 static int
75 77 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
76 78 {
77 79 traverse_data_t *td = arg;
78 80 zbookmark_phys_t zb;
79 81
80 82 if (BP_IS_HOLE(bp))
81 83 return (0);
82 84
83 85 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
84 86 return (0);
85 87
86 88 SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
87 89 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
88 90
89 91 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg);
90 92
91 93 return (0);
92 94 }
93 95
94 96 static int
95 97 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
96 98 {
97 99 traverse_data_t *td = arg;
98 100
99 101 if (lrc->lrc_txtype == TX_WRITE) {
100 102 lr_write_t *lr = (lr_write_t *)lrc;
101 103 blkptr_t *bp = &lr->lr_blkptr;
102 104 zbookmark_phys_t zb;
103 105
104 106 if (BP_IS_HOLE(bp))
105 107 return (0);
106 108
107 109 if (claim_txg == 0 || bp->blk_birth < claim_txg)
108 110 return (0);
109 111
110 112 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
111 113 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
112 114
113 115 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL,
114 116 td->td_arg);
115 117 }
116 118 return (0);
117 119 }
118 120
119 121 static void
120 122 traverse_zil(traverse_data_t *td, zil_header_t *zh)
121 123 {
122 124 uint64_t claim_txg = zh->zh_claim_txg;
123 125 zilog_t *zilog;
124 126
125 127 /*
126 128 * We only want to visit blocks that have been claimed but not yet
127 129 * replayed; plus, in read-only mode, blocks that are already stable.
128 130 */
129 131 if (claim_txg == 0 && spa_writeable(td->td_spa))
130 132 return;
131 133
132 134 zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
133 135
134 136 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
135 137 claim_txg);
136 138
137 139 zil_free(zilog);
138 140 }
139 141
140 142 typedef enum resume_skip {
141 143 RESUME_SKIP_ALL,
142 144 RESUME_SKIP_NONE,
143 145 RESUME_SKIP_CHILDREN
144 146 } resume_skip_t;
145 147
146 148 /*
147 149 * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
148 150 * the block indicated by zb does not need to be visited at all. Returns
149 151 * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
150 152 * resume point. This indicates that this block should be visited but not its
151 153 * children (since they must have been visited in a previous traversal).
152 154 * Otherwise returns RESUME_SKIP_NONE.
153 155 */
154 156 static resume_skip_t
155 157 resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
156 158 const zbookmark_phys_t *zb)
157 159 {
158 160 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
159 161 /*
160 162 * If we already visited this bp & everything below,
161 163 * don't bother doing it again.
162 164 */
163 165 if (zbookmark_subtree_completed(dnp, zb, td->td_resume))
164 166 return (RESUME_SKIP_ALL);
165 167
166 168 /*
167 169 * If we found the block we're trying to resume from, zero
168 170 * the bookmark out to indicate that we have resumed.
169 171 */
170 172 if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
171 173 bzero(td->td_resume, sizeof (*zb));
172 174 if (td->td_flags & TRAVERSE_POST)
173 175 return (RESUME_SKIP_CHILDREN);
174 176 }
175 177 }
176 178 return (RESUME_SKIP_NONE);
177 179 }
178 180
179 181 static void
180 182 traverse_prefetch_metadata(traverse_data_t *td,
181 183 const blkptr_t *bp, const zbookmark_phys_t *zb)
182 184 {
183 185 arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
|
↓ open down ↓ |
115 lines elided |
↑ open up ↑ |
184 186
185 187 if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
186 188 return;
187 189 /*
188 190 * If we are in the process of resuming, don't prefetch, because
189 191 * some children will not be needed (and in fact may have already
190 192 * been freed).
191 193 */
192 194 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
193 195 return;
194 - if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
196 + if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg ||
197 + bp->blk_birth >= td->td_max_txg)
195 198 return;
196 199 if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
197 200 return;
198 201
199 202 (void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
200 203 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
201 204 }
202 205
203 206 static boolean_t
204 207 prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp)
205 208 {
206 209 ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA);
207 210 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) ||
208 211 BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
209 212 return (B_FALSE);
210 213 return (B_TRUE);
211 214 }
212 215
213 216 static int
214 217 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
215 218 const blkptr_t *bp, const zbookmark_phys_t *zb)
216 219 {
217 220 zbookmark_phys_t czb;
218 221 int err = 0;
219 222 arc_buf_t *buf = NULL;
220 223 prefetch_data_t *pd = td->td_pfd;
221 224 boolean_t hard = td->td_flags & TRAVERSE_HARD;
222 225
223 226 switch (resume_skip_check(td, dnp, zb)) {
224 227 case RESUME_SKIP_ALL:
225 228 return (0);
226 229 case RESUME_SKIP_CHILDREN:
227 230 goto post;
228 231 case RESUME_SKIP_NONE:
229 232 break;
230 233 default:
231 234 ASSERT(0);
232 235 }
233 236
234 237 if (bp->blk_birth == 0) {
235 238 /*
236 239 * Since this block has a birth time of 0 it must be one of
237 240 * two things: a hole created before the
238 241 * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole
239 242 * which has always been a hole in an object.
240 243 *
241 244 * If a file is written sparsely, then the unwritten parts of
242 245 * the file were "always holes" -- that is, they have been
243 246 * holes since this object was allocated. However, we (and
244 247 * our callers) can not necessarily tell when an object was
245 248 * allocated. Therefore, if it's possible that this object
246 249 * was freed and then its object number reused, we need to
247 250 * visit all the holes with birth==0.
248 251 *
249 252 * If it isn't possible that the object number was reused,
|
↓ open down ↓ |
45 lines elided |
↑ open up ↑ |
250 253 * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote
251 254 * all the blocks we will visit as part of this traversal,
252 255 * then this hole must have always existed, so we can skip
253 256 * it. We visit blocks born after (exclusive) td_min_txg.
254 257 *
255 258 * Note that the meta-dnode cannot be reallocated.
256 259 */
257 260 if (!send_holes_without_birth_time &&
258 261 (!td->td_realloc_possible ||
259 262 zb->zb_object == DMU_META_DNODE_OBJECT) &&
260 - td->td_hole_birth_enabled_txg <= td->td_min_txg)
263 + (td->td_hole_birth_enabled_txg <= td->td_min_txg ||
264 + td->td_hole_birth_enabled_txg > td->td_max_txg))
261 265 return (0);
262 - } else if (bp->blk_birth <= td->td_min_txg) {
266 + } else if (bp->blk_birth <= td->td_min_txg ||
267 + bp->blk_birth >= td->td_max_txg) {
263 268 return (0);
264 269 }
265 270
266 271 if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) {
267 272 uint64_t size = BP_GET_LSIZE(bp);
268 273 mutex_enter(&pd->pd_mtx);
269 274 ASSERT(pd->pd_bytes_fetched >= 0);
270 275 while (pd->pd_bytes_fetched < size && !pd->pd_exited)
271 276 cv_wait(&pd->pd_cv, &pd->pd_mtx);
272 277 pd->pd_bytes_fetched -= size;
273 278 cv_broadcast(&pd->pd_cv);
274 279 mutex_exit(&pd->pd_mtx);
275 280 }
276 281
277 282 if (BP_IS_HOLE(bp)) {
278 283 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
|
↓ open down ↓ |
6 lines elided |
↑ open up ↑ |
279 284 if (err != 0)
280 285 goto post;
281 286 return (0);
282 287 }
283 288
284 289 if (td->td_flags & TRAVERSE_PRE) {
285 290 err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
286 291 td->td_arg);
287 292 if (err == TRAVERSE_VISIT_NO_CHILDREN)
288 293 return (0);
294 + /* handle pausing at a common point */
295 + if (err == ERESTART)
296 + td->td_paused = B_TRUE;
289 297 if (err != 0)
290 298 goto post;
291 299 }
292 300
293 301 if (BP_GET_LEVEL(bp) > 0) {
294 302 arc_flags_t flags = ARC_FLAG_WAIT;
295 303 int i;
296 304 blkptr_t *cbp;
297 305 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
298 306
299 307 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
300 308 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
301 309 if (err != 0)
302 310 goto post;
303 311 cbp = buf->b_data;
304 312
305 313 for (i = 0; i < epb; i++) {
306 314 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
307 315 zb->zb_level - 1,
308 316 zb->zb_blkid * epb + i);
309 317 traverse_prefetch_metadata(td, &cbp[i], &czb);
310 318 }
311 319
312 320 /* recursively visitbp() blocks below this */
313 321 for (i = 0; i < epb; i++) {
314 322 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
315 323 zb->zb_level - 1,
316 324 zb->zb_blkid * epb + i);
317 325 err = traverse_visitbp(td, dnp, &cbp[i], &czb);
318 326 if (err != 0)
319 327 break;
320 328 }
321 329 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
322 330 arc_flags_t flags = ARC_FLAG_WAIT;
323 331 int i;
324 332 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
325 333
326 334 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
327 335 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
328 336 if (err != 0)
329 337 goto post;
330 338 dnode_phys_t *child_dnp = buf->b_data;
331 339
332 340 for (i = 0; i < epb; i++) {
333 341 prefetch_dnode_metadata(td, &child_dnp[i],
334 342 zb->zb_objset, zb->zb_blkid * epb + i);
335 343 }
336 344
337 345 /* recursively visitbp() blocks below this */
338 346 for (i = 0; i < epb; i++) {
339 347 err = traverse_dnode(td, &child_dnp[i],
340 348 zb->zb_objset, zb->zb_blkid * epb + i);
341 349 if (err != 0)
342 350 break;
343 351 }
344 352 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
345 353 arc_flags_t flags = ARC_FLAG_WAIT;
346 354
347 355 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
348 356 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
349 357 if (err != 0)
350 358 goto post;
351 359
352 360 objset_phys_t *osp = buf->b_data;
353 361 prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset,
354 362 DMU_META_DNODE_OBJECT);
355 363 /*
356 364 * See the block comment above for the goal of this variable.
357 365 * If the maxblkid of the meta-dnode is 0, then we know that
358 366 * we've never had more than DNODES_PER_BLOCK objects in the
359 367 * dataset, which means we can't have reused any object ids.
360 368 */
361 369 if (osp->os_meta_dnode.dn_maxblkid == 0)
362 370 td->td_realloc_possible = B_FALSE;
363 371
364 372 if (arc_buf_size(buf) >= sizeof (objset_phys_t)) {
365 373 prefetch_dnode_metadata(td, &osp->os_groupused_dnode,
366 374 zb->zb_objset, DMU_GROUPUSED_OBJECT);
367 375 prefetch_dnode_metadata(td, &osp->os_userused_dnode,
368 376 zb->zb_objset, DMU_USERUSED_OBJECT);
369 377 }
370 378
371 379 err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset,
372 380 DMU_META_DNODE_OBJECT);
373 381 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
374 382 err = traverse_dnode(td, &osp->os_groupused_dnode,
375 383 zb->zb_objset, DMU_GROUPUSED_OBJECT);
376 384 }
377 385 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
378 386 err = traverse_dnode(td, &osp->os_userused_dnode,
379 387 zb->zb_objset, DMU_USERUSED_OBJECT);
380 388 }
381 389 }
382 390
383 391 if (buf)
384 392 arc_buf_destroy(buf, &buf);
385 393
386 394 post:
387 395 if (err == 0 && (td->td_flags & TRAVERSE_POST))
388 396 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
389 397
390 398 if (hard && (err == EIO || err == ECKSUM)) {
391 399 /*
392 400 * Ignore this disk error as requested by the HARD flag,
393 401 * and continue traversal.
394 402 */
395 403 err = 0;
396 404 }
397 405
398 406 /*
399 407 * If we are stopping here, set td_resume.
400 408 */
401 409 if (td->td_resume != NULL && err != 0 && !td->td_paused) {
402 410 td->td_resume->zb_objset = zb->zb_objset;
403 411 td->td_resume->zb_object = zb->zb_object;
404 412 td->td_resume->zb_level = 0;
405 413 /*
406 414 * If we have stopped on an indirect block (e.g. due to
407 415 * i/o error), we have not visited anything below it.
408 416 * Set the bookmark to the first level-0 block that we need
409 417 * to visit. This way, the resuming code does not need to
410 418 * deal with resuming from indirect blocks.
411 419 *
412 420 * Note, if zb_level <= 0, dnp may be NULL, so we don't want
|
↓ open down ↓ |
114 lines elided |
↑ open up ↑ |
413 421 * to dereference it.
414 422 */
415 423 td->td_resume->zb_blkid = zb->zb_blkid;
416 424 if (zb->zb_level > 0) {
417 425 td->td_resume->zb_blkid <<= zb->zb_level *
418 426 (dnp->dn_indblkshift - SPA_BLKPTRSHIFT);
419 427 }
420 428 td->td_paused = B_TRUE;
421 429 }
422 430
431 + /* if we walked over all bp bookmark must be cleared */
432 + if (!err && !td->td_paused && td->td_resume != NULL &&
433 + bp == td->td_rootbp && td->td_pfd != NULL) {
434 + bzero(td->td_resume, sizeof (*td->td_resume));
435 + }
436 +
423 437 return (err);
424 438 }
425 439
426 440 static void
427 441 prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
428 442 uint64_t objset, uint64_t object)
429 443 {
430 444 int j;
431 445 zbookmark_phys_t czb;
432 446
433 447 for (j = 0; j < dnp->dn_nblkptr; j++) {
434 448 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
435 449 traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb);
436 450 }
437 451
438 452 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
439 453 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
440 454 traverse_prefetch_metadata(td, &dnp->dn_spill, &czb);
441 455 }
442 456 }
443 457
444 458 static int
445 459 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
446 460 uint64_t objset, uint64_t object)
447 461 {
448 462 int j, err = 0;
449 463 zbookmark_phys_t czb;
450 464
451 465 if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL &&
452 466 object < td->td_resume->zb_object)
453 467 return (0);
454 468
455 469 if (td->td_flags & TRAVERSE_PRE) {
456 470 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
457 471 ZB_DNODE_BLKID);
458 472 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
459 473 td->td_arg);
460 474 if (err == TRAVERSE_VISIT_NO_CHILDREN)
461 475 return (0);
462 476 if (err != 0)
463 477 return (err);
464 478 }
465 479
466 480 for (j = 0; j < dnp->dn_nblkptr; j++) {
467 481 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
468 482 err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb);
469 483 if (err != 0)
470 484 break;
471 485 }
472 486
473 487 if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
474 488 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
475 489 err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb);
476 490 }
477 491
478 492 if (err == 0 && (td->td_flags & TRAVERSE_POST)) {
479 493 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
480 494 ZB_DNODE_BLKID);
481 495 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
482 496 td->td_arg);
483 497 if (err == TRAVERSE_VISIT_NO_CHILDREN)
484 498 return (0);
485 499 if (err != 0)
486 500 return (err);
487 501 }
488 502 return (err);
489 503 }
490 504
491 505 /* ARGSUSED */
492 506 static int
493 507 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
494 508 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
495 509 {
496 510 prefetch_data_t *pfd = arg;
497 511 arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
498 512
499 513 ASSERT(pfd->pd_bytes_fetched >= 0);
500 514 if (bp == NULL)
501 515 return (0);
502 516 if (pfd->pd_cancel)
503 517 return (SET_ERROR(EINTR));
504 518
505 519 if (!prefetch_needed(pfd, bp))
506 520 return (0);
507 521
508 522 mutex_enter(&pfd->pd_mtx);
509 523 while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max)
510 524 cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
511 525 pfd->pd_bytes_fetched += BP_GET_LSIZE(bp);
512 526 cv_broadcast(&pfd->pd_cv);
513 527 mutex_exit(&pfd->pd_mtx);
514 528
515 529 (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
516 530 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb);
517 531
518 532 return (0);
519 533 }
520 534
521 535 static void
522 536 traverse_prefetch_thread(void *arg)
523 537 {
524 538 traverse_data_t *td_main = arg;
525 539 traverse_data_t td = *td_main;
526 540 zbookmark_phys_t czb;
527 541
528 542 td.td_func = traverse_prefetcher;
529 543 td.td_arg = td_main->td_pfd;
530 544 td.td_pfd = NULL;
531 545 td.td_resume = &td_main->td_pfd->pd_resume;
532 546
533 547 SET_BOOKMARK(&czb, td.td_objset,
534 548 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
535 549 (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb);
536 550
537 551 mutex_enter(&td_main->td_pfd->pd_mtx);
538 552 td_main->td_pfd->pd_exited = B_TRUE;
|
↓ open down ↓ |
106 lines elided |
↑ open up ↑ |
539 553 cv_broadcast(&td_main->td_pfd->pd_cv);
540 554 mutex_exit(&td_main->td_pfd->pd_mtx);
541 555 }
542 556
543 557 /*
544 558 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
545 559 * in syncing context).
546 560 */
547 561 static int
548 562 traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
549 - uint64_t txg_start, zbookmark_phys_t *resume, int flags,
550 - blkptr_cb_t func, void *arg)
563 + uint64_t txg_start, uint64_t txg_finish, zbookmark_phys_t *resume,
564 + int flags, blkptr_cb_t func, void *arg)
551 565 {
552 566 traverse_data_t td;
553 567 prefetch_data_t pd = { 0 };
554 568 zbookmark_phys_t czb;
555 569 int err;
556 570
557 571 ASSERT(ds == NULL || objset == ds->ds_object);
558 572 ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
559 573
560 574 td.td_spa = spa;
561 575 td.td_objset = objset;
562 576 td.td_rootbp = rootbp;
563 577 td.td_min_txg = txg_start;
578 + td.td_max_txg = txg_finish;
564 579 td.td_resume = resume;
565 580 td.td_func = func;
566 581 td.td_arg = arg;
567 582 td.td_pfd = &pd;
568 583 td.td_flags = flags;
569 584 td.td_paused = B_FALSE;
570 585 td.td_realloc_possible = (txg_start == 0 ? B_FALSE : B_TRUE);
571 586
572 587 if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) {
573 588 VERIFY(spa_feature_enabled_txg(spa,
574 589 SPA_FEATURE_HOLE_BIRTH, &td.td_hole_birth_enabled_txg));
575 590 } else {
576 591 td.td_hole_birth_enabled_txg = UINT64_MAX;
577 592 }
578 593
579 594 pd.pd_flags = flags;
580 595 if (resume != NULL)
581 596 pd.pd_resume = *resume;
582 597 mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
583 598 cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);
584 599
585 600 /* See comment on ZIL traversal in dsl_scan_visitds. */
586 601 if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) {
587 602 arc_flags_t flags = ARC_FLAG_WAIT;
588 603 objset_phys_t *osp;
589 604 arc_buf_t *buf;
590 605
591 606 err = arc_read(NULL, td.td_spa, rootbp,
592 607 arc_getbuf_func, &buf,
593 608 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL);
594 609 if (err != 0)
595 610 return (err);
596 611
597 612 osp = buf->b_data;
598 613 traverse_zil(&td, &osp->os_zil_header);
599 614 arc_buf_destroy(buf, &buf);
600 615 }
601 616
602 617 if (!(flags & TRAVERSE_PREFETCH_DATA) ||
603 618 0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
604 619 &td, TQ_NOQUEUE))
605 620 pd.pd_exited = B_TRUE;
606 621
607 622 SET_BOOKMARK(&czb, td.td_objset,
608 623 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
609 624 err = traverse_visitbp(&td, NULL, rootbp, &czb);
610 625
611 626 mutex_enter(&pd.pd_mtx);
612 627 pd.pd_cancel = B_TRUE;
613 628 cv_broadcast(&pd.pd_cv);
614 629 while (!pd.pd_exited)
615 630 cv_wait(&pd.pd_cv, &pd.pd_mtx);
616 631 mutex_exit(&pd.pd_mtx);
617 632
618 633 mutex_destroy(&pd.pd_mtx);
619 634 cv_destroy(&pd.pd_cv);
620 635
621 636 return (err);
622 637 }
623 638
|
↓ open down ↓ |
50 lines elided |
↑ open up ↑ |
624 639 /*
625 640 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
626 641 * in syncing context).
627 642 */
628 643 int
629 644 traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start,
630 645 zbookmark_phys_t *resume,
631 646 int flags, blkptr_cb_t func, void *arg)
632 647 {
633 648 return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
634 - &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg));
649 + &dsl_dataset_phys(ds)->ds_bp, txg_start, UINT64_MAX, resume, flags,
650 + func, arg));
635 651 }
636 652
637 653 int
638 654 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start,
639 655 int flags, blkptr_cb_t func, void *arg)
640 656 {
641 657 return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg));
642 658 }
643 659
644 660 int
645 661 traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
646 662 uint64_t txg_start, zbookmark_phys_t *resume, int flags,
647 663 blkptr_cb_t func, void *arg)
648 664 {
649 665 return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
650 - blkptr, txg_start, resume, flags, func, arg));
666 + blkptr, txg_start, UINT64_MAX, resume, flags, func, arg));
651 667 }
652 668
653 669 /*
654 670 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
655 671 */
656 672 int
657 -traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
658 - blkptr_cb_t func, void *arg)
673 +traverse_pool(spa_t *spa, uint64_t txg_start, uint64_t txg_finish, int flags,
674 + blkptr_cb_t func, void *arg, zbookmark_phys_t *zb)
659 675 {
660 - int err;
676 + int err = 0, lasterr = 0;
661 677 dsl_pool_t *dp = spa_get_dsl(spa);
662 678 objset_t *mos = dp->dp_meta_objset;
663 679 boolean_t hard = (flags & TRAVERSE_HARD);
664 680
665 681 /* visit the MOS */
666 - err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
667 - txg_start, NULL, flags, func, arg);
668 - if (err != 0)
669 - return (err);
682 + if (!zb || (zb->zb_objset == 0 && zb->zb_object == 0)) {
683 + err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
684 + txg_start, txg_finish, NULL, flags, func, arg);
685 + if (err != 0)
686 + return (err);
687 + }
670 688
671 689 /* visit each dataset */
672 - for (uint64_t obj = 1; err == 0;
690 + for (uint64_t obj = (zb && !ZB_IS_ZERO(zb))? zb->zb_objset : 1;
691 + err == 0 || (err != ESRCH && hard);
673 692 err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) {
674 693 dmu_object_info_t doi;
675 694
676 695 err = dmu_object_info(mos, obj, &doi);
677 696 if (err != 0) {
678 697 if (hard)
679 698 continue;
680 699 break;
681 700 }
682 701
683 702 if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) {
684 703 dsl_dataset_t *ds;
704 + objset_t *os;
705 + boolean_t os_is_snapshot = B_FALSE;
685 706 uint64_t txg = txg_start;
707 + uint64_t ctxg;
708 + uint64_t max_txg = txg_finish;
686 709
687 710 dsl_pool_config_enter(dp, FTAG);
688 711 err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
689 712 dsl_pool_config_exit(dp, FTAG);
690 713 if (err != 0) {
691 714 if (hard)
692 715 continue;
693 716 break;
694 717 }
695 - if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg)
718 +
719 + dsl_pool_config_enter(dp, FTAG);
720 + err = dmu_objset_from_ds(ds, &os);
721 + if (err == 0)
722 + os_is_snapshot = dmu_objset_is_snapshot(os);
723 +
724 + dsl_pool_config_exit(dp, FTAG);
725 + if (err != 0) {
726 + dsl_dataset_rele(ds, FTAG);
727 + if (hard)
728 + continue;
729 + break;
730 + }
731 + ctxg = dsl_dataset_phys(ds)->ds_creation_txg;
732 +
733 + /* uplimited traverse walks over shapshots only */
734 + if (max_txg != UINT64_MAX && !os_is_snapshot) {
735 + dsl_dataset_rele(ds, FTAG);
736 + continue;
737 + }
738 + if (max_txg != UINT64_MAX && ctxg >= max_txg) {
739 + dsl_dataset_rele(ds, FTAG);
740 + continue;
741 + }
742 + if (os_is_snapshot && ctxg <= txg_start) {
743 + dsl_dataset_rele(ds, FTAG);
744 + continue;
745 + }
746 + if (max_txg == UINT64_MAX &&
747 + dsl_dataset_phys(ds)->ds_prev_snap_txg > txg)
696 748 txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
697 - err = traverse_dataset(ds, txg, flags, func, arg);
749 + if (txg > max_txg)
750 + max_txg = txg;
751 + err = traverse_impl(spa, ds, ds->ds_object,
752 + &dsl_dataset_phys(ds)->ds_bp,
753 + txg, max_txg, zb, flags, func, arg);
698 754 dsl_dataset_rele(ds, FTAG);
699 - if (err != 0)
755 + if (err != 0) {
756 + if (!hard)
757 + return (err);
758 + lasterr = err;
759 + }
760 + if (zb && !ZB_IS_ZERO(zb))
700 761 break;
701 762 }
702 763 }
703 - if (err == ESRCH)
764 + if (err == ESRCH) {
765 + /* zero bookmark means we are done */
766 + if (zb)
767 + bzero(zb, sizeof (*zb));
704 768 err = 0;
705 - return (err);
769 + }
770 + return (err != 0 ? err : lasterr);
706 771 }
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX