Print this page
4046 dsl_dataset_t ds_dir->dd_lock is highly contended
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/dsl_dir.c
+++ new/usr/src/uts/common/fs/zfs/dsl_dir.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 *
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 23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
25 25 */
26 26
27 27 #include <sys/dmu.h>
28 28 #include <sys/dmu_objset.h>
29 29 #include <sys/dmu_tx.h>
30 30 #include <sys/dsl_dataset.h>
31 31 #include <sys/dsl_dir.h>
32 32 #include <sys/dsl_prop.h>
33 33 #include <sys/dsl_synctask.h>
34 34 #include <sys/dsl_deleg.h>
35 35 #include <sys/spa.h>
36 36 #include <sys/metaslab.h>
37 37 #include <sys/zap.h>
38 38 #include <sys/zio.h>
39 39 #include <sys/arc.h>
40 40 #include <sys/sunddi.h>
41 41 #include "zfs_namecheck.h"
42 42
43 43 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
44 44
45 45 /* ARGSUSED */
46 46 static void
47 47 dsl_dir_evict(dmu_buf_t *db, void *arg)
48 48 {
49 49 dsl_dir_t *dd = arg;
50 50 dsl_pool_t *dp = dd->dd_pool;
51 51 int t;
52 52
53 53 for (t = 0; t < TXG_SIZE; t++) {
54 54 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
55 55 ASSERT(dd->dd_tempreserved[t] == 0);
56 56 ASSERT(dd->dd_space_towrite[t] == 0);
57 57 }
58 58
59 59 if (dd->dd_parent)
60 60 dsl_dir_rele(dd->dd_parent, dd);
61 61
62 62 spa_close(dd->dd_pool->dp_spa, dd);
63 63
64 64 /*
65 65 * The props callback list should have been cleaned up by
66 66 * objset_evict().
67 67 */
68 68 list_destroy(&dd->dd_prop_cbs);
69 69 mutex_destroy(&dd->dd_lock);
70 70 kmem_free(dd, sizeof (dsl_dir_t));
71 71 }
72 72
73 73 int
74 74 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
75 75 const char *tail, void *tag, dsl_dir_t **ddp)
76 76 {
77 77 dmu_buf_t *dbuf;
78 78 dsl_dir_t *dd;
79 79 int err;
80 80
81 81 ASSERT(dsl_pool_config_held(dp));
82 82
83 83 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
84 84 if (err != 0)
85 85 return (err);
86 86 dd = dmu_buf_get_user(dbuf);
87 87 #ifdef ZFS_DEBUG
88 88 {
89 89 dmu_object_info_t doi;
90 90 dmu_object_info_from_db(dbuf, &doi);
91 91 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
92 92 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
93 93 }
94 94 #endif
95 95 if (dd == NULL) {
96 96 dsl_dir_t *winner;
97 97
98 98 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
99 99 dd->dd_object = ddobj;
100 100 dd->dd_dbuf = dbuf;
101 101 dd->dd_pool = dp;
102 102 dd->dd_phys = dbuf->db_data;
103 103 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
104 104
105 105 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
106 106 offsetof(dsl_prop_cb_record_t, cbr_node));
107 107
108 108 dsl_dir_snap_cmtime_update(dd);
109 109
110 110 if (dd->dd_phys->dd_parent_obj) {
111 111 err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
112 112 NULL, dd, &dd->dd_parent);
113 113 if (err != 0)
114 114 goto errout;
115 115 if (tail) {
116 116 #ifdef ZFS_DEBUG
117 117 uint64_t foundobj;
118 118
119 119 err = zap_lookup(dp->dp_meta_objset,
120 120 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
121 121 tail, sizeof (foundobj), 1, &foundobj);
122 122 ASSERT(err || foundobj == ddobj);
123 123 #endif
124 124 (void) strcpy(dd->dd_myname, tail);
125 125 } else {
126 126 err = zap_value_search(dp->dp_meta_objset,
127 127 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
128 128 ddobj, 0, dd->dd_myname);
129 129 }
130 130 if (err != 0)
131 131 goto errout;
132 132 } else {
133 133 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
134 134 }
135 135
136 136 if (dsl_dir_is_clone(dd)) {
137 137 dmu_buf_t *origin_bonus;
138 138 dsl_dataset_phys_t *origin_phys;
139 139
140 140 /*
141 141 * We can't open the origin dataset, because
142 142 * that would require opening this dsl_dir.
143 143 * Just look at its phys directly instead.
144 144 */
145 145 err = dmu_bonus_hold(dp->dp_meta_objset,
146 146 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
147 147 if (err != 0)
148 148 goto errout;
149 149 origin_phys = origin_bonus->db_data;
150 150 dd->dd_origin_txg =
151 151 origin_phys->ds_creation_txg;
152 152 dmu_buf_rele(origin_bonus, FTAG);
153 153 }
154 154
155 155 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
156 156 dsl_dir_evict);
157 157 if (winner) {
158 158 if (dd->dd_parent)
159 159 dsl_dir_rele(dd->dd_parent, dd);
160 160 mutex_destroy(&dd->dd_lock);
161 161 kmem_free(dd, sizeof (dsl_dir_t));
162 162 dd = winner;
163 163 } else {
164 164 spa_open_ref(dp->dp_spa, dd);
165 165 }
166 166 }
167 167
168 168 /*
169 169 * The dsl_dir_t has both open-to-close and instantiate-to-evict
170 170 * holds on the spa. We need the open-to-close holds because
171 171 * otherwise the spa_refcnt wouldn't change when we open a
172 172 * dir which the spa also has open, so we could incorrectly
173 173 * think it was OK to unload/export/destroy the pool. We need
174 174 * the instantiate-to-evict hold because the dsl_dir_t has a
175 175 * pointer to the dd_pool, which has a pointer to the spa_t.
176 176 */
177 177 spa_open_ref(dp->dp_spa, tag);
178 178 ASSERT3P(dd->dd_pool, ==, dp);
179 179 ASSERT3U(dd->dd_object, ==, ddobj);
180 180 ASSERT3P(dd->dd_dbuf, ==, dbuf);
181 181 *ddp = dd;
182 182 return (0);
183 183
184 184 errout:
185 185 if (dd->dd_parent)
186 186 dsl_dir_rele(dd->dd_parent, dd);
187 187 mutex_destroy(&dd->dd_lock);
188 188 kmem_free(dd, sizeof (dsl_dir_t));
189 189 dmu_buf_rele(dbuf, tag);
190 190 return (err);
191 191 }
192 192
193 193 void
194 194 dsl_dir_rele(dsl_dir_t *dd, void *tag)
195 195 {
196 196 dprintf_dd(dd, "%s\n", "");
197 197 spa_close(dd->dd_pool->dp_spa, tag);
198 198 dmu_buf_rele(dd->dd_dbuf, tag);
199 199 }
200 200
201 201 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
202 202 void
203 203 dsl_dir_name(dsl_dir_t *dd, char *buf)
204 204 {
205 205 if (dd->dd_parent) {
206 206 dsl_dir_name(dd->dd_parent, buf);
207 207 (void) strcat(buf, "/");
208 208 } else {
209 209 buf[0] = '\0';
210 210 }
211 211 if (!MUTEX_HELD(&dd->dd_lock)) {
212 212 /*
213 213 * recursive mutex so that we can use
214 214 * dprintf_dd() with dd_lock held
215 215 */
216 216 mutex_enter(&dd->dd_lock);
217 217 (void) strcat(buf, dd->dd_myname);
218 218 mutex_exit(&dd->dd_lock);
219 219 } else {
220 220 (void) strcat(buf, dd->dd_myname);
221 221 }
222 222 }
223 223
224 224 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
225 225 int
226 226 dsl_dir_namelen(dsl_dir_t *dd)
227 227 {
228 228 int result = 0;
229 229
230 230 if (dd->dd_parent) {
231 231 /* parent's name + 1 for the "/" */
232 232 result = dsl_dir_namelen(dd->dd_parent) + 1;
233 233 }
234 234
235 235 if (!MUTEX_HELD(&dd->dd_lock)) {
236 236 /* see dsl_dir_name */
237 237 mutex_enter(&dd->dd_lock);
238 238 result += strlen(dd->dd_myname);
239 239 mutex_exit(&dd->dd_lock);
240 240 } else {
241 241 result += strlen(dd->dd_myname);
242 242 }
243 243
244 244 return (result);
245 245 }
246 246
247 247 static int
248 248 getcomponent(const char *path, char *component, const char **nextp)
249 249 {
250 250 char *p;
251 251
252 252 if ((path == NULL) || (path[0] == '\0'))
253 253 return (SET_ERROR(ENOENT));
254 254 /* This would be a good place to reserve some namespace... */
255 255 p = strpbrk(path, "/@");
256 256 if (p && (p[1] == '/' || p[1] == '@')) {
257 257 /* two separators in a row */
258 258 return (SET_ERROR(EINVAL));
259 259 }
260 260 if (p == NULL || p == path) {
261 261 /*
262 262 * if the first thing is an @ or /, it had better be an
263 263 * @ and it had better not have any more ats or slashes,
264 264 * and it had better have something after the @.
265 265 */
266 266 if (p != NULL &&
267 267 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
268 268 return (SET_ERROR(EINVAL));
269 269 if (strlen(path) >= MAXNAMELEN)
270 270 return (SET_ERROR(ENAMETOOLONG));
271 271 (void) strcpy(component, path);
272 272 p = NULL;
273 273 } else if (p[0] == '/') {
274 274 if (p - path >= MAXNAMELEN)
275 275 return (SET_ERROR(ENAMETOOLONG));
276 276 (void) strncpy(component, path, p - path);
277 277 component[p - path] = '\0';
278 278 p++;
279 279 } else if (p[0] == '@') {
280 280 /*
281 281 * if the next separator is an @, there better not be
282 282 * any more slashes.
283 283 */
284 284 if (strchr(path, '/'))
285 285 return (SET_ERROR(EINVAL));
286 286 if (p - path >= MAXNAMELEN)
287 287 return (SET_ERROR(ENAMETOOLONG));
288 288 (void) strncpy(component, path, p - path);
289 289 component[p - path] = '\0';
290 290 } else {
291 291 panic("invalid p=%p", (void *)p);
292 292 }
293 293 *nextp = p;
294 294 return (0);
295 295 }
296 296
297 297 /*
298 298 * Return the dsl_dir_t, and possibly the last component which couldn't
299 299 * be found in *tail. The name must be in the specified dsl_pool_t. This
300 300 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
301 301 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
302 302 * (*tail)[0] == '@' means that the last component is a snapshot.
303 303 */
304 304 int
305 305 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
306 306 dsl_dir_t **ddp, const char **tailp)
307 307 {
308 308 char buf[MAXNAMELEN];
309 309 const char *spaname, *next, *nextnext = NULL;
310 310 int err;
311 311 dsl_dir_t *dd;
312 312 uint64_t ddobj;
313 313
314 314 err = getcomponent(name, buf, &next);
315 315 if (err != 0)
316 316 return (err);
317 317
318 318 /* Make sure the name is in the specified pool. */
319 319 spaname = spa_name(dp->dp_spa);
320 320 if (strcmp(buf, spaname) != 0)
321 321 return (SET_ERROR(EINVAL));
322 322
323 323 ASSERT(dsl_pool_config_held(dp));
324 324
325 325 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
326 326 if (err != 0) {
327 327 return (err);
328 328 }
329 329
330 330 while (next != NULL) {
331 331 dsl_dir_t *child_ds;
332 332 err = getcomponent(next, buf, &nextnext);
333 333 if (err != 0)
334 334 break;
335 335 ASSERT(next[0] != '\0');
336 336 if (next[0] == '@')
337 337 break;
338 338 dprintf("looking up %s in obj%lld\n",
339 339 buf, dd->dd_phys->dd_child_dir_zapobj);
340 340
341 341 err = zap_lookup(dp->dp_meta_objset,
342 342 dd->dd_phys->dd_child_dir_zapobj,
343 343 buf, sizeof (ddobj), 1, &ddobj);
344 344 if (err != 0) {
345 345 if (err == ENOENT)
346 346 err = 0;
347 347 break;
348 348 }
349 349
350 350 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
351 351 if (err != 0)
352 352 break;
353 353 dsl_dir_rele(dd, tag);
354 354 dd = child_ds;
355 355 next = nextnext;
356 356 }
357 357
358 358 if (err != 0) {
359 359 dsl_dir_rele(dd, tag);
360 360 return (err);
361 361 }
362 362
363 363 /*
364 364 * It's an error if there's more than one component left, or
365 365 * tailp==NULL and there's any component left.
366 366 */
367 367 if (next != NULL &&
368 368 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
369 369 /* bad path name */
370 370 dsl_dir_rele(dd, tag);
371 371 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
372 372 err = SET_ERROR(ENOENT);
373 373 }
374 374 if (tailp != NULL)
375 375 *tailp = next;
376 376 *ddp = dd;
377 377 return (err);
378 378 }
379 379
380 380 uint64_t
381 381 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
382 382 dmu_tx_t *tx)
383 383 {
384 384 objset_t *mos = dp->dp_meta_objset;
385 385 uint64_t ddobj;
386 386 dsl_dir_phys_t *ddphys;
387 387 dmu_buf_t *dbuf;
388 388
389 389 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
390 390 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
391 391 if (pds) {
392 392 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
393 393 name, sizeof (uint64_t), 1, &ddobj, tx));
394 394 } else {
395 395 /* it's the root dir */
396 396 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
397 397 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
398 398 }
399 399 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
400 400 dmu_buf_will_dirty(dbuf, tx);
401 401 ddphys = dbuf->db_data;
402 402
403 403 ddphys->dd_creation_time = gethrestime_sec();
404 404 if (pds)
405 405 ddphys->dd_parent_obj = pds->dd_object;
406 406 ddphys->dd_props_zapobj = zap_create(mos,
407 407 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
408 408 ddphys->dd_child_dir_zapobj = zap_create(mos,
409 409 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
410 410 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
411 411 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
412 412 dmu_buf_rele(dbuf, FTAG);
413 413
414 414 return (ddobj);
415 415 }
416 416
417 417 boolean_t
418 418 dsl_dir_is_clone(dsl_dir_t *dd)
419 419 {
420 420 return (dd->dd_phys->dd_origin_obj &&
421 421 (dd->dd_pool->dp_origin_snap == NULL ||
422 422 dd->dd_phys->dd_origin_obj !=
423 423 dd->dd_pool->dp_origin_snap->ds_object));
424 424 }
425 425
426 426 void
427 427 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
428 428 {
429 429 mutex_enter(&dd->dd_lock);
430 430 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
431 431 dd->dd_phys->dd_used_bytes);
432 432 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
433 433 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
434 434 dd->dd_phys->dd_reserved);
435 435 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
436 436 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
437 437 (dd->dd_phys->dd_uncompressed_bytes * 100 /
438 438 dd->dd_phys->dd_compressed_bytes));
439 439 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
440 440 dd->dd_phys->dd_uncompressed_bytes);
441 441 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
442 442 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
443 443 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
444 444 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
445 445 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
446 446 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
447 447 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
448 448 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
449 449 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
450 450 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
451 451 }
452 452 mutex_exit(&dd->dd_lock);
453 453
454 454 if (dsl_dir_is_clone(dd)) {
455 455 dsl_dataset_t *ds;
456 456 char buf[MAXNAMELEN];
457 457
458 458 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
459 459 dd->dd_phys->dd_origin_obj, FTAG, &ds));
460 460 dsl_dataset_name(ds, buf);
461 461 dsl_dataset_rele(ds, FTAG);
462 462 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
463 463 }
464 464 }
465 465
466 466 void
467 467 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
468 468 {
469 469 dsl_pool_t *dp = dd->dd_pool;
470 470
471 471 ASSERT(dd->dd_phys);
472 472
473 473 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
474 474 /* up the hold count until we can be written out */
475 475 dmu_buf_add_ref(dd->dd_dbuf, dd);
476 476 }
477 477 }
478 478
479 479 static int64_t
480 480 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
481 481 {
482 482 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
483 483 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
484 484 return (new_accounted - old_accounted);
485 485 }
486 486
487 487 void
488 488 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
489 489 {
490 490 ASSERT(dmu_tx_is_syncing(tx));
491 491
492 492 mutex_enter(&dd->dd_lock);
493 493 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
494 494 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
495 495 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
496 496 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
497 497 mutex_exit(&dd->dd_lock);
498 498
499 499 /* release the hold from dsl_dir_dirty */
500 500 dmu_buf_rele(dd->dd_dbuf, dd);
501 501 }
502 502
503 503 static uint64_t
504 504 dsl_dir_space_towrite(dsl_dir_t *dd)
505 505 {
506 506 uint64_t space = 0;
507 507 int i;
508 508
509 509 ASSERT(MUTEX_HELD(&dd->dd_lock));
510 510
511 511 for (i = 0; i < TXG_SIZE; i++) {
512 512 space += dd->dd_space_towrite[i&TXG_MASK];
513 513 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
514 514 }
515 515 return (space);
516 516 }
517 517
518 518 /*
519 519 * How much space would dd have available if ancestor had delta applied
520 520 * to it? If ondiskonly is set, we're only interested in what's
521 521 * on-disk, not estimated pending changes.
522 522 */
523 523 uint64_t
524 524 dsl_dir_space_available(dsl_dir_t *dd,
525 525 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
526 526 {
527 527 uint64_t parentspace, myspace, quota, used;
528 528
529 529 /*
530 530 * If there are no restrictions otherwise, assume we have
531 531 * unlimited space available.
532 532 */
533 533 quota = UINT64_MAX;
534 534 parentspace = UINT64_MAX;
535 535
536 536 if (dd->dd_parent != NULL) {
537 537 parentspace = dsl_dir_space_available(dd->dd_parent,
538 538 ancestor, delta, ondiskonly);
539 539 }
540 540
541 541 mutex_enter(&dd->dd_lock);
542 542 if (dd->dd_phys->dd_quota != 0)
543 543 quota = dd->dd_phys->dd_quota;
544 544 used = dd->dd_phys->dd_used_bytes;
545 545 if (!ondiskonly)
546 546 used += dsl_dir_space_towrite(dd);
547 547
548 548 if (dd->dd_parent == NULL) {
549 549 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
550 550 quota = MIN(quota, poolsize);
551 551 }
552 552
553 553 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
554 554 /*
555 555 * We have some space reserved, in addition to what our
556 556 * parent gave us.
557 557 */
558 558 parentspace += dd->dd_phys->dd_reserved - used;
559 559 }
560 560
561 561 if (dd == ancestor) {
562 562 ASSERT(delta <= 0);
563 563 ASSERT(used >= -delta);
564 564 used += delta;
565 565 if (parentspace != UINT64_MAX)
566 566 parentspace -= delta;
567 567 }
568 568
569 569 if (used > quota) {
570 570 /* over quota */
571 571 myspace = 0;
572 572 } else {
573 573 /*
574 574 * the lesser of the space provided by our parent and
575 575 * the space left in our quota
576 576 */
577 577 myspace = MIN(parentspace, quota - used);
578 578 }
579 579
580 580 mutex_exit(&dd->dd_lock);
581 581
582 582 return (myspace);
583 583 }
584 584
585 585 struct tempreserve {
586 586 list_node_t tr_node;
587 587 dsl_pool_t *tr_dp;
588 588 dsl_dir_t *tr_ds;
589 589 uint64_t tr_size;
590 590 };
591 591
592 592 static int
593 593 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
594 594 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
595 595 dmu_tx_t *tx, boolean_t first)
596 596 {
597 597 uint64_t txg = tx->tx_txg;
598 598 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
599 599 uint64_t deferred = 0;
600 600 struct tempreserve *tr;
601 601 int retval = EDQUOT;
602 602 int txgidx = txg & TXG_MASK;
603 603 int i;
604 604 uint64_t ref_rsrv = 0;
605 605
606 606 ASSERT3U(txg, !=, 0);
607 607 ASSERT3S(asize, >, 0);
608 608
609 609 mutex_enter(&dd->dd_lock);
610 610
611 611 /*
612 612 * Check against the dsl_dir's quota. We don't add in the delta
613 613 * when checking for over-quota because they get one free hit.
614 614 */
615 615 est_inflight = dsl_dir_space_towrite(dd);
616 616 for (i = 0; i < TXG_SIZE; i++)
617 617 est_inflight += dd->dd_tempreserved[i];
618 618 used_on_disk = dd->dd_phys->dd_used_bytes;
619 619
620 620 /*
621 621 * On the first iteration, fetch the dataset's used-on-disk and
622 622 * refreservation values. Also, if checkrefquota is set, test if
623 623 * allocating this space would exceed the dataset's refquota.
624 624 */
625 625 if (first && tx->tx_objset) {
626 626 int error;
627 627 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
628 628
629 629 error = dsl_dataset_check_quota(ds, checkrefquota,
630 630 asize, est_inflight, &used_on_disk, &ref_rsrv);
631 631 if (error) {
632 632 mutex_exit(&dd->dd_lock);
633 633 return (error);
634 634 }
635 635 }
636 636
637 637 /*
638 638 * If this transaction will result in a net free of space,
639 639 * we want to let it through.
640 640 */
641 641 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
642 642 quota = UINT64_MAX;
643 643 else
644 644 quota = dd->dd_phys->dd_quota;
645 645
646 646 /*
647 647 * Adjust the quota against the actual pool size at the root
648 648 * minus any outstanding deferred frees.
649 649 * To ensure that it's possible to remove files from a full
650 650 * pool without inducing transient overcommits, we throttle
651 651 * netfree transactions against a quota that is slightly larger,
652 652 * but still within the pool's allocation slop. In cases where
653 653 * we're very close to full, this will allow a steady trickle of
654 654 * removes to get through.
655 655 */
656 656 if (dd->dd_parent == NULL) {
657 657 spa_t *spa = dd->dd_pool->dp_spa;
658 658 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
659 659 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
660 660 if (poolsize - deferred < quota) {
661 661 quota = poolsize - deferred;
662 662 retval = ENOSPC;
663 663 }
664 664 }
665 665
666 666 /*
667 667 * If they are requesting more space, and our current estimate
668 668 * is over quota, they get to try again unless the actual
669 669 * on-disk is over quota and there are no pending changes (which
670 670 * may free up space for us).
671 671 */
672 672 if (used_on_disk + est_inflight >= quota) {
673 673 if (est_inflight > 0 || used_on_disk < quota ||
674 674 (retval == ENOSPC && used_on_disk < quota + deferred))
675 675 retval = ERESTART;
676 676 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
677 677 "quota=%lluK tr=%lluK err=%d\n",
678 678 used_on_disk>>10, est_inflight>>10,
679 679 quota>>10, asize>>10, retval);
680 680 mutex_exit(&dd->dd_lock);
681 681 return (SET_ERROR(retval));
682 682 }
683 683
684 684 /* We need to up our estimated delta before dropping dd_lock */
685 685 dd->dd_tempreserved[txgidx] += asize;
686 686
687 687 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
688 688 asize - ref_rsrv);
689 689 mutex_exit(&dd->dd_lock);
690 690
691 691 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
692 692 tr->tr_ds = dd;
693 693 tr->tr_size = asize;
694 694 list_insert_tail(tr_list, tr);
695 695
696 696 /* see if it's OK with our parent */
697 697 if (dd->dd_parent && parent_rsrv) {
698 698 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
699 699
700 700 return (dsl_dir_tempreserve_impl(dd->dd_parent,
701 701 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
702 702 } else {
703 703 return (0);
704 704 }
705 705 }
706 706
707 707 /*
708 708 * Reserve space in this dsl_dir, to be used in this tx's txg.
709 709 * After the space has been dirtied (and dsl_dir_willuse_space()
710 710 * has been called), the reservation should be canceled, using
711 711 * dsl_dir_tempreserve_clear().
712 712 */
713 713 int
714 714 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
715 715 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
716 716 {
717 717 int err;
718 718 list_t *tr_list;
719 719
720 720 if (asize == 0) {
721 721 *tr_cookiep = NULL;
722 722 return (0);
723 723 }
724 724
725 725 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
726 726 list_create(tr_list, sizeof (struct tempreserve),
727 727 offsetof(struct tempreserve, tr_node));
728 728 ASSERT3S(asize, >, 0);
729 729 ASSERT3S(fsize, >=, 0);
730 730
731 731 err = arc_tempreserve_space(lsize, tx->tx_txg);
732 732 if (err == 0) {
733 733 struct tempreserve *tr;
734 734
735 735 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
736 736 tr->tr_size = lsize;
737 737 list_insert_tail(tr_list, tr);
738 738
739 739 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
740 740 } else {
741 741 if (err == EAGAIN) {
742 742 txg_delay(dd->dd_pool, tx->tx_txg,
743 743 MSEC2NSEC(10), MSEC2NSEC(10));
744 744 err = SET_ERROR(ERESTART);
745 745 }
746 746 dsl_pool_memory_pressure(dd->dd_pool);
747 747 }
748 748
749 749 if (err == 0) {
750 750 struct tempreserve *tr;
751 751
752 752 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
753 753 tr->tr_dp = dd->dd_pool;
754 754 tr->tr_size = asize;
755 755 list_insert_tail(tr_list, tr);
756 756
757 757 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
758 758 FALSE, asize > usize, tr_list, tx, TRUE);
759 759 }
760 760
761 761 if (err != 0)
762 762 dsl_dir_tempreserve_clear(tr_list, tx);
763 763 else
764 764 *tr_cookiep = tr_list;
765 765
766 766 return (err);
767 767 }
768 768
769 769 /*
770 770 * Clear a temporary reservation that we previously made with
771 771 * dsl_dir_tempreserve_space().
772 772 */
773 773 void
774 774 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
775 775 {
776 776 int txgidx = tx->tx_txg & TXG_MASK;
777 777 list_t *tr_list = tr_cookie;
778 778 struct tempreserve *tr;
779 779
780 780 ASSERT3U(tx->tx_txg, !=, 0);
781 781
782 782 if (tr_cookie == NULL)
783 783 return;
784 784
785 785 while (tr = list_head(tr_list)) {
786 786 if (tr->tr_dp) {
787 787 dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
788 788 } else if (tr->tr_ds) {
789 789 mutex_enter(&tr->tr_ds->dd_lock);
790 790 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
791 791 tr->tr_size);
792 792 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
793 793 mutex_exit(&tr->tr_ds->dd_lock);
794 794 } else {
795 795 arc_tempreserve_clear(tr->tr_size);
796 796 }
797 797 list_remove(tr_list, tr);
798 798 kmem_free(tr, sizeof (struct tempreserve));
799 799 }
800 800
801 801 kmem_free(tr_list, sizeof (list_t));
802 802 }
803 803
804 804 static void
805 805 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
806 806 {
807 807 int64_t parent_space;
808 808 uint64_t est_used;
809 809
810 810 mutex_enter(&dd->dd_lock);
811 811 if (space > 0)
812 812 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
813 813
814 814 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
815 815 parent_space = parent_delta(dd, est_used, space);
816 816 mutex_exit(&dd->dd_lock);
817 817
818 818 /* Make sure that we clean up dd_space_to* */
819 819 dsl_dir_dirty(dd, tx);
820 820
821 821 /* XXX this is potentially expensive and unnecessary... */
822 822 if (parent_space && dd->dd_parent)
823 823 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
824 824 }
825 825
826 826 /*
827 827 * Call in open context when we think we're going to write/free space,
828 828 * eg. when dirtying data. Be conservative (ie. OK to write less than
829 829 * this or free more than this, but don't write more or free less).
830 830 */
831 831 void
832 832 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
833 833 {
|
↓ open down ↓ |
833 lines elided |
↑ open up ↑ |
834 834 dsl_pool_willuse_space(dd->dd_pool, space, tx);
835 835 dsl_dir_willuse_space_impl(dd, space, tx);
836 836 }
837 837
838 838 /* call from syncing context when we actually write/free space for this dd */
839 839 void
840 840 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
841 841 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
842 842 {
843 843 int64_t accounted_delta;
844 +
845 + /*
846 + * dsl_dataset_set_refreservation_sync_impl() calls this with
847 + * dd_lock held, so that it can atomically update
848 + * ds->ds_reserved and the dsl_dir accounting, so that
849 + * dsl_dataset_check_quota() can see dataset and dir accounting
850 + * consistently.
851 + */
844 852 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
845 853
846 854 ASSERT(dmu_tx_is_syncing(tx));
847 855 ASSERT(type < DD_USED_NUM);
848 856
857 + dmu_buf_will_dirty(dd->dd_dbuf, tx);
858 +
849 859 if (needlock)
850 860 mutex_enter(&dd->dd_lock);
851 861 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
852 862 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
853 863 ASSERT(compressed >= 0 ||
854 864 dd->dd_phys->dd_compressed_bytes >= -compressed);
855 865 ASSERT(uncompressed >= 0 ||
856 866 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
857 - dmu_buf_will_dirty(dd->dd_dbuf, tx);
858 867 dd->dd_phys->dd_used_bytes += used;
859 868 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
860 869 dd->dd_phys->dd_compressed_bytes += compressed;
861 870
862 871 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
863 872 ASSERT(used > 0 ||
864 873 dd->dd_phys->dd_used_breakdown[type] >= -used);
865 874 dd->dd_phys->dd_used_breakdown[type] += used;
866 875 #ifdef DEBUG
867 876 dd_used_t t;
868 877 uint64_t u = 0;
869 878 for (t = 0; t < DD_USED_NUM; t++)
870 879 u += dd->dd_phys->dd_used_breakdown[t];
871 880 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
872 881 #endif
873 882 }
874 883 if (needlock)
875 884 mutex_exit(&dd->dd_lock);
876 885
877 886 if (dd->dd_parent != NULL) {
878 887 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
879 888 accounted_delta, compressed, uncompressed, tx);
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
880 889 dsl_dir_transfer_space(dd->dd_parent,
881 890 used - accounted_delta,
882 891 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
883 892 }
884 893 }
885 894
886 895 void
887 896 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
888 897 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
889 898 {
890 - boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
891 -
892 899 ASSERT(dmu_tx_is_syncing(tx));
893 900 ASSERT(oldtype < DD_USED_NUM);
894 901 ASSERT(newtype < DD_USED_NUM);
895 902
896 903 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
897 904 return;
898 905
899 - if (needlock)
900 - mutex_enter(&dd->dd_lock);
906 + dmu_buf_will_dirty(dd->dd_dbuf, tx);
907 + mutex_enter(&dd->dd_lock);
901 908 ASSERT(delta > 0 ?
902 909 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
903 910 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
904 911 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
905 - dmu_buf_will_dirty(dd->dd_dbuf, tx);
906 912 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
907 913 dd->dd_phys->dd_used_breakdown[newtype] += delta;
908 - if (needlock)
909 - mutex_exit(&dd->dd_lock);
914 + mutex_exit(&dd->dd_lock);
910 915 }
911 916
912 917 typedef struct dsl_dir_set_qr_arg {
913 918 const char *ddsqra_name;
914 919 zprop_source_t ddsqra_source;
915 920 uint64_t ddsqra_value;
916 921 } dsl_dir_set_qr_arg_t;
917 922
918 923 static int
919 924 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
920 925 {
921 926 dsl_dir_set_qr_arg_t *ddsqra = arg;
922 927 dsl_pool_t *dp = dmu_tx_pool(tx);
923 928 dsl_dataset_t *ds;
924 929 int error;
925 930 uint64_t towrite, newval;
926 931
927 932 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
928 933 if (error != 0)
929 934 return (error);
930 935
931 936 error = dsl_prop_predict(ds->ds_dir, "quota",
932 937 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
933 938 if (error != 0) {
934 939 dsl_dataset_rele(ds, FTAG);
935 940 return (error);
936 941 }
937 942
938 943 if (newval == 0) {
939 944 dsl_dataset_rele(ds, FTAG);
940 945 return (0);
941 946 }
942 947
943 948 mutex_enter(&ds->ds_dir->dd_lock);
944 949 /*
945 950 * If we are doing the preliminary check in open context, and
946 951 * there are pending changes, then don't fail it, since the
947 952 * pending changes could under-estimate the amount of space to be
948 953 * freed up.
949 954 */
950 955 towrite = dsl_dir_space_towrite(ds->ds_dir);
951 956 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
952 957 (newval < ds->ds_dir->dd_phys->dd_reserved ||
953 958 newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
954 959 error = SET_ERROR(ENOSPC);
955 960 }
956 961 mutex_exit(&ds->ds_dir->dd_lock);
957 962 dsl_dataset_rele(ds, FTAG);
958 963 return (error);
959 964 }
960 965
961 966 static void
962 967 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
963 968 {
964 969 dsl_dir_set_qr_arg_t *ddsqra = arg;
965 970 dsl_pool_t *dp = dmu_tx_pool(tx);
966 971 dsl_dataset_t *ds;
967 972 uint64_t newval;
968 973
969 974 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
970 975
971 976 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
972 977 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
973 978 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
974 979 &ddsqra->ddsqra_value, tx);
975 980
976 981 VERIFY0(dsl_prop_get_int_ds(ds,
977 982 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
978 983 } else {
979 984 newval = ddsqra->ddsqra_value;
980 985 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
981 986 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
982 987 }
983 988
984 989 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
985 990 mutex_enter(&ds->ds_dir->dd_lock);
986 991 ds->ds_dir->dd_phys->dd_quota = newval;
987 992 mutex_exit(&ds->ds_dir->dd_lock);
988 993 dsl_dataset_rele(ds, FTAG);
989 994 }
990 995
991 996 int
992 997 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
993 998 {
994 999 dsl_dir_set_qr_arg_t ddsqra;
995 1000
996 1001 ddsqra.ddsqra_name = ddname;
997 1002 ddsqra.ddsqra_source = source;
998 1003 ddsqra.ddsqra_value = quota;
999 1004
1000 1005 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1001 1006 dsl_dir_set_quota_sync, &ddsqra, 0));
1002 1007 }
1003 1008
1004 1009 int
1005 1010 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1006 1011 {
1007 1012 dsl_dir_set_qr_arg_t *ddsqra = arg;
1008 1013 dsl_pool_t *dp = dmu_tx_pool(tx);
1009 1014 dsl_dataset_t *ds;
1010 1015 dsl_dir_t *dd;
1011 1016 uint64_t newval, used, avail;
1012 1017 int error;
1013 1018
1014 1019 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1015 1020 if (error != 0)
1016 1021 return (error);
1017 1022 dd = ds->ds_dir;
1018 1023
1019 1024 /*
1020 1025 * If we are doing the preliminary check in open context, the
1021 1026 * space estimates may be inaccurate.
1022 1027 */
1023 1028 if (!dmu_tx_is_syncing(tx)) {
1024 1029 dsl_dataset_rele(ds, FTAG);
1025 1030 return (0);
1026 1031 }
1027 1032
1028 1033 error = dsl_prop_predict(ds->ds_dir,
1029 1034 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1030 1035 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1031 1036 if (error != 0) {
1032 1037 dsl_dataset_rele(ds, FTAG);
1033 1038 return (error);
1034 1039 }
1035 1040
1036 1041 mutex_enter(&dd->dd_lock);
1037 1042 used = dd->dd_phys->dd_used_bytes;
1038 1043 mutex_exit(&dd->dd_lock);
1039 1044
1040 1045 if (dd->dd_parent) {
1041 1046 avail = dsl_dir_space_available(dd->dd_parent,
1042 1047 NULL, 0, FALSE);
1043 1048 } else {
1044 1049 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1045 1050 }
1046 1051
1047 1052 if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1048 1053 uint64_t delta = MAX(used, newval) -
1049 1054 MAX(used, dd->dd_phys->dd_reserved);
1050 1055
1051 1056 if (delta > avail ||
1052 1057 (dd->dd_phys->dd_quota > 0 &&
1053 1058 newval > dd->dd_phys->dd_quota))
1054 1059 error = SET_ERROR(ENOSPC);
1055 1060 }
1056 1061
1057 1062 dsl_dataset_rele(ds, FTAG);
1058 1063 return (error);
1059 1064 }
1060 1065
1061 1066 void
1062 1067 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1063 1068 {
1064 1069 uint64_t used;
1065 1070 int64_t delta;
1066 1071
1067 1072 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1068 1073
1069 1074 mutex_enter(&dd->dd_lock);
1070 1075 used = dd->dd_phys->dd_used_bytes;
1071 1076 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1072 1077 dd->dd_phys->dd_reserved = value;
1073 1078
1074 1079 if (dd->dd_parent != NULL) {
1075 1080 /* Roll up this additional usage into our ancestors */
1076 1081 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1077 1082 delta, 0, 0, tx);
1078 1083 }
1079 1084 mutex_exit(&dd->dd_lock);
1080 1085 }
1081 1086
1082 1087
1083 1088 static void
1084 1089 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1085 1090 {
1086 1091 dsl_dir_set_qr_arg_t *ddsqra = arg;
1087 1092 dsl_pool_t *dp = dmu_tx_pool(tx);
1088 1093 dsl_dataset_t *ds;
1089 1094 uint64_t newval;
1090 1095
1091 1096 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1092 1097
1093 1098 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1094 1099 dsl_prop_set_sync_impl(ds,
1095 1100 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1096 1101 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1097 1102 &ddsqra->ddsqra_value, tx);
1098 1103
1099 1104 VERIFY0(dsl_prop_get_int_ds(ds,
1100 1105 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1101 1106 } else {
1102 1107 newval = ddsqra->ddsqra_value;
1103 1108 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1104 1109 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1105 1110 (longlong_t)newval);
1106 1111 }
1107 1112
1108 1113 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1109 1114 dsl_dataset_rele(ds, FTAG);
1110 1115 }
1111 1116
1112 1117 int
1113 1118 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1114 1119 uint64_t reservation)
1115 1120 {
1116 1121 dsl_dir_set_qr_arg_t ddsqra;
1117 1122
1118 1123 ddsqra.ddsqra_name = ddname;
1119 1124 ddsqra.ddsqra_source = source;
1120 1125 ddsqra.ddsqra_value = reservation;
1121 1126
1122 1127 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1123 1128 dsl_dir_set_reservation_sync, &ddsqra, 0));
1124 1129 }
1125 1130
1126 1131 static dsl_dir_t *
1127 1132 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1128 1133 {
1129 1134 for (; ds1; ds1 = ds1->dd_parent) {
1130 1135 dsl_dir_t *dd;
1131 1136 for (dd = ds2; dd; dd = dd->dd_parent) {
1132 1137 if (ds1 == dd)
1133 1138 return (dd);
1134 1139 }
1135 1140 }
1136 1141 return (NULL);
1137 1142 }
1138 1143
1139 1144 /*
1140 1145 * If delta is applied to dd, how much of that delta would be applied to
1141 1146 * ancestor? Syncing context only.
1142 1147 */
1143 1148 static int64_t
1144 1149 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1145 1150 {
1146 1151 if (dd == ancestor)
1147 1152 return (delta);
1148 1153
1149 1154 mutex_enter(&dd->dd_lock);
1150 1155 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1151 1156 mutex_exit(&dd->dd_lock);
1152 1157 return (would_change(dd->dd_parent, delta, ancestor));
1153 1158 }
1154 1159
1155 1160 typedef struct dsl_dir_rename_arg {
1156 1161 const char *ddra_oldname;
1157 1162 const char *ddra_newname;
1158 1163 } dsl_dir_rename_arg_t;
1159 1164
1160 1165 /* ARGSUSED */
1161 1166 static int
1162 1167 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1163 1168 {
1164 1169 int *deltap = arg;
1165 1170 char namebuf[MAXNAMELEN];
1166 1171
1167 1172 dsl_dataset_name(ds, namebuf);
1168 1173
1169 1174 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1170 1175 return (SET_ERROR(ENAMETOOLONG));
1171 1176 return (0);
1172 1177 }
1173 1178
1174 1179 static int
1175 1180 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1176 1181 {
1177 1182 dsl_dir_rename_arg_t *ddra = arg;
1178 1183 dsl_pool_t *dp = dmu_tx_pool(tx);
1179 1184 dsl_dir_t *dd, *newparent;
1180 1185 const char *mynewname;
1181 1186 int error;
1182 1187 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1183 1188
1184 1189 /* target dir should exist */
1185 1190 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1186 1191 if (error != 0)
1187 1192 return (error);
1188 1193
1189 1194 /* new parent should exist */
1190 1195 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1191 1196 &newparent, &mynewname);
1192 1197 if (error != 0) {
1193 1198 dsl_dir_rele(dd, FTAG);
1194 1199 return (error);
1195 1200 }
1196 1201
1197 1202 /* can't rename to different pool */
1198 1203 if (dd->dd_pool != newparent->dd_pool) {
1199 1204 dsl_dir_rele(newparent, FTAG);
1200 1205 dsl_dir_rele(dd, FTAG);
1201 1206 return (SET_ERROR(ENXIO));
1202 1207 }
1203 1208
1204 1209 /* new name should not already exist */
1205 1210 if (mynewname == NULL) {
1206 1211 dsl_dir_rele(newparent, FTAG);
1207 1212 dsl_dir_rele(dd, FTAG);
1208 1213 return (SET_ERROR(EEXIST));
1209 1214 }
1210 1215
1211 1216 /* if the name length is growing, validate child name lengths */
1212 1217 if (delta > 0) {
1213 1218 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1214 1219 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1215 1220 if (error != 0) {
1216 1221 dsl_dir_rele(newparent, FTAG);
1217 1222 dsl_dir_rele(dd, FTAG);
1218 1223 return (error);
1219 1224 }
1220 1225 }
1221 1226
1222 1227 if (newparent != dd->dd_parent) {
1223 1228 /* is there enough space? */
1224 1229 uint64_t myspace =
1225 1230 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1226 1231
1227 1232 /* no rename into our descendant */
1228 1233 if (closest_common_ancestor(dd, newparent) == dd) {
1229 1234 dsl_dir_rele(newparent, FTAG);
1230 1235 dsl_dir_rele(dd, FTAG);
1231 1236 return (SET_ERROR(EINVAL));
1232 1237 }
1233 1238
1234 1239 error = dsl_dir_transfer_possible(dd->dd_parent,
1235 1240 newparent, myspace);
1236 1241 if (error != 0) {
1237 1242 dsl_dir_rele(newparent, FTAG);
1238 1243 dsl_dir_rele(dd, FTAG);
1239 1244 return (error);
1240 1245 }
1241 1246 }
1242 1247
1243 1248 dsl_dir_rele(newparent, FTAG);
1244 1249 dsl_dir_rele(dd, FTAG);
1245 1250 return (0);
1246 1251 }
1247 1252
1248 1253 static void
1249 1254 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1250 1255 {
1251 1256 dsl_dir_rename_arg_t *ddra = arg;
1252 1257 dsl_pool_t *dp = dmu_tx_pool(tx);
1253 1258 dsl_dir_t *dd, *newparent;
1254 1259 const char *mynewname;
1255 1260 int error;
1256 1261 objset_t *mos = dp->dp_meta_objset;
1257 1262
1258 1263 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1259 1264 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1260 1265 &mynewname));
1261 1266
1262 1267 /* Log this before we change the name. */
1263 1268 spa_history_log_internal_dd(dd, "rename", tx,
1264 1269 "-> %s", ddra->ddra_newname);
1265 1270
1266 1271 if (newparent != dd->dd_parent) {
1267 1272 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1268 1273 -dd->dd_phys->dd_used_bytes,
1269 1274 -dd->dd_phys->dd_compressed_bytes,
1270 1275 -dd->dd_phys->dd_uncompressed_bytes, tx);
1271 1276 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1272 1277 dd->dd_phys->dd_used_bytes,
1273 1278 dd->dd_phys->dd_compressed_bytes,
1274 1279 dd->dd_phys->dd_uncompressed_bytes, tx);
1275 1280
1276 1281 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1277 1282 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1278 1283 dd->dd_phys->dd_used_bytes;
1279 1284
1280 1285 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1281 1286 -unused_rsrv, 0, 0, tx);
1282 1287 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1283 1288 unused_rsrv, 0, 0, tx);
1284 1289 }
1285 1290 }
1286 1291
1287 1292 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1288 1293
1289 1294 /* remove from old parent zapobj */
1290 1295 error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1291 1296 dd->dd_myname, tx);
1292 1297 ASSERT0(error);
1293 1298
1294 1299 (void) strcpy(dd->dd_myname, mynewname);
1295 1300 dsl_dir_rele(dd->dd_parent, dd);
1296 1301 dd->dd_phys->dd_parent_obj = newparent->dd_object;
1297 1302 VERIFY0(dsl_dir_hold_obj(dp,
1298 1303 newparent->dd_object, NULL, dd, &dd->dd_parent));
1299 1304
1300 1305 /* add to new parent zapobj */
1301 1306 VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1302 1307 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1303 1308
1304 1309 dsl_prop_notify_all(dd);
1305 1310
1306 1311 dsl_dir_rele(newparent, FTAG);
1307 1312 dsl_dir_rele(dd, FTAG);
1308 1313 }
1309 1314
1310 1315 int
1311 1316 dsl_dir_rename(const char *oldname, const char *newname)
1312 1317 {
1313 1318 dsl_dir_rename_arg_t ddra;
1314 1319
1315 1320 ddra.ddra_oldname = oldname;
1316 1321 ddra.ddra_newname = newname;
1317 1322
1318 1323 return (dsl_sync_task(oldname,
1319 1324 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1320 1325 }
1321 1326
1322 1327 int
1323 1328 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1324 1329 {
1325 1330 dsl_dir_t *ancestor;
1326 1331 int64_t adelta;
1327 1332 uint64_t avail;
1328 1333
1329 1334 ancestor = closest_common_ancestor(sdd, tdd);
1330 1335 adelta = would_change(sdd, -space, ancestor);
1331 1336 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1332 1337 if (avail < space)
1333 1338 return (SET_ERROR(ENOSPC));
1334 1339
1335 1340 return (0);
1336 1341 }
1337 1342
1338 1343 timestruc_t
1339 1344 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1340 1345 {
1341 1346 timestruc_t t;
1342 1347
1343 1348 mutex_enter(&dd->dd_lock);
1344 1349 t = dd->dd_snap_cmtime;
1345 1350 mutex_exit(&dd->dd_lock);
1346 1351
1347 1352 return (t);
1348 1353 }
1349 1354
1350 1355 void
1351 1356 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1352 1357 {
1353 1358 timestruc_t t;
1354 1359
1355 1360 gethrestime(&t);
1356 1361 mutex_enter(&dd->dd_lock);
1357 1362 dd->dd_snap_cmtime = t;
1358 1363 mutex_exit(&dd->dd_lock);
1359 1364 }
|
↓ open down ↓ |
440 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX