xref: /illumos-gate/usr/src/uts/common/fs/zfs/dmu_object.c (revision 8a2b682e57a046b828f37bcde1776f131ef4629f)
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) 2013, 2017 by Delphix. All rights reserved.
24  * Copyright 2014 HybridCluster. All rights reserved.
25  */
26 
27 #include <sys/dbuf.h>
28 #include <sys/dmu.h>
29 #include <sys/dmu_objset.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dnode.h>
32 #include <sys/zap.h>
33 #include <sys/zfeature.h>
34 #include <sys/dsl_dataset.h>
35 
36 /*
37  * Each of the concurrent object allocators will grab
38  * 2^dmu_object_alloc_chunk_shift dnode slots at a time.  The default is to
39  * grab 128 slots, which is 4 blocks worth.  This was experimentally
40  * determined to be the lowest value that eliminates the measurable effect
41  * of lock contention from this code path.
42  */
43 int dmu_object_alloc_chunk_shift = 7;
44 
45 static uint64_t
46 dmu_object_alloc_impl(objset_t *os, dmu_object_type_t ot, int blocksize,
47     int indirect_blockshift, dmu_object_type_t bonustype, int bonuslen,
48     int dnodesize, dmu_tx_t *tx)
49 {
50 	uint64_t object;
51 	uint64_t L1_dnode_count = DNODES_PER_BLOCK <<
52 	    (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT);
53 	dnode_t *dn = NULL;
54 	int dn_slots = dnodesize >> DNODE_SHIFT;
55 	boolean_t restarted = B_FALSE;
56 	uint64_t *cpuobj = &os->os_obj_next_percpu[CPU_SEQID %
57 	    os->os_obj_next_percpu_len];
58 	int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
59 	int error;
60 
61 	if (dn_slots == 0) {
62 		dn_slots = DNODE_MIN_SLOTS;
63 	} else {
64 		ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
65 		ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
66 	}
67 
68 	/*
69 	 * The "chunk" of dnodes that is assigned to a CPU-specific
70 	 * allocator needs to be at least one block's worth, to avoid
71 	 * lock contention on the dbuf.  It can be at most one L1 block's
72 	 * worth, so that the "rescan after polishing off a L1's worth"
73 	 * logic below will be sure to kick in.
74 	 */
75 	if (dnodes_per_chunk < DNODES_PER_BLOCK)
76 		dnodes_per_chunk = DNODES_PER_BLOCK;
77 	if (dnodes_per_chunk > L1_dnode_count)
78 		dnodes_per_chunk = L1_dnode_count;
79 
80 	object = *cpuobj;
81 
82 	for (;;) {
83 		/*
84 		 * If we finished a chunk of dnodes, get a new one from
85 		 * the global allocator.
86 		 */
87 		if ((P2PHASE(object, dnodes_per_chunk) == 0) ||
88 		    (P2PHASE(object + dn_slots - 1, dnodes_per_chunk) <
89 		    dn_slots)) {
90 			DNODE_STAT_BUMP(dnode_alloc_next_chunk);
91 			mutex_enter(&os->os_obj_lock);
92 			ASSERT0(P2PHASE(os->os_obj_next_chunk,
93 			    dnodes_per_chunk));
94 			object = os->os_obj_next_chunk;
95 
96 			/*
97 			 * Each time we polish off a L1 bp worth of dnodes
98 			 * (2^12 objects), move to another L1 bp that's
99 			 * still reasonably sparse (at most 1/4 full). Look
100 			 * from the beginning at most once per txg. If we
101 			 * still can't allocate from that L1 block, search
102 			 * for an empty L0 block, which will quickly skip
103 			 * to the end of the metadnode if the no nearby L0
104 			 * blocks are empty. This fallback avoids a
105 			 * pathology where full dnode blocks containing
106 			 * large dnodes appear sparse because they have a
107 			 * low blk_fill, leading to many failed allocation
108 			 * attempts. In the long term a better mechanism to
109 			 * search for sparse metadnode regions, such as
110 			 * spacemaps, could be implemented.
111 			 *
112 			 * os_scan_dnodes is set during txg sync if enough
113 			 * objects have been freed since the previous
114 			 * rescan to justify backfilling again.
115 			 *
116 			 * Note that dmu_traverse depends on the behavior
117 			 * that we use multiple blocks of the dnode object
118 			 * before going back to reuse objects. Any change
119 			 * to this algorithm should preserve that property
120 			 * or find another solution to the issues described
121 			 * in traverse_visitbp.
122 			 */
123 			if (P2PHASE(object, L1_dnode_count) == 0) {
124 				uint64_t offset;
125 				uint64_t blkfill;
126 				int minlvl;
127 				if (os->os_rescan_dnodes) {
128 					offset = 0;
129 					os->os_rescan_dnodes = B_FALSE;
130 				} else {
131 					offset = object << DNODE_SHIFT;
132 				}
133 				blkfill = restarted ? 1 : DNODES_PER_BLOCK >> 2;
134 				minlvl = restarted ? 1 : 2;
135 				restarted = B_TRUE;
136 				error = dnode_next_offset(DMU_META_DNODE(os),
137 				    DNODE_FIND_HOLE, &offset, minlvl,
138 				    blkfill, 0);
139 				if (error == 0) {
140 					object = offset >> DNODE_SHIFT;
141 				}
142 			}
143 			/*
144 			 * Note: if "restarted", we may find a L0 that
145 			 * is not suitably aligned.
146 			 */
147 			os->os_obj_next_chunk =
148 			    P2ALIGN(object, dnodes_per_chunk) +
149 			    dnodes_per_chunk;
150 			(void) atomic_swap_64(cpuobj, object);
151 			mutex_exit(&os->os_obj_lock);
152 		}
153 
154 		/*
155 		 * The value of (*cpuobj) before adding dn_slots is the object
156 		 * ID assigned to us.  The value afterwards is the object ID
157 		 * assigned to whoever wants to do an allocation next.
158 		 */
159 		object = atomic_add_64_nv(cpuobj, dn_slots) - dn_slots;
160 
161 		/*
162 		 * XXX We should check for an i/o error here and return
163 		 * up to our caller.  Actually we should pre-read it in
164 		 * dmu_tx_assign(), but there is currently no mechanism
165 		 * to do so.
166 		 */
167 		error = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE,
168 		    dn_slots, FTAG, &dn);
169 		if (error == 0) {
170 			rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
171 			/*
172 			 * Another thread could have allocated it; check
173 			 * again now that we have the struct lock.
174 			 */
175 			if (dn->dn_type == DMU_OT_NONE) {
176 				dnode_allocate(dn, ot, blocksize, 0,
177 				    bonustype, bonuslen, dn_slots, tx);
178 				rw_exit(&dn->dn_struct_rwlock);
179 				dmu_tx_add_new_object(tx, dn);
180 				dnode_rele(dn, FTAG);
181 				return (object);
182 			}
183 			rw_exit(&dn->dn_struct_rwlock);
184 			dnode_rele(dn, FTAG);
185 			DNODE_STAT_BUMP(dnode_alloc_race);
186 		}
187 
188 		/*
189 		 * Skip to next known valid starting point on error. This
190 		 * is the start of the next block of dnodes.
191 		 */
192 		if (dmu_object_next(os, &object, B_TRUE, 0) != 0) {
193 			object = P2ROUNDUP(object + 1, DNODES_PER_BLOCK);
194 			DNODE_STAT_BUMP(dnode_alloc_next_block);
195 		}
196 		(void) atomic_swap_64(cpuobj, object);
197 	}
198 }
199 
200 uint64_t
201 dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
202     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
203 {
204 	return (dmu_object_alloc_impl(os, ot, blocksize, 0, bonustype,
205 	    bonuslen, 0, tx));
206 }
207 
208 uint64_t
209 dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
210     int indirect_blockshift, dmu_object_type_t bonustype, int bonuslen,
211     dmu_tx_t *tx)
212 {
213 	return (dmu_object_alloc_impl(os, ot, blocksize, indirect_blockshift,
214 	    bonustype, bonuslen, 0, tx));
215 }
216 
217 uint64_t
218 dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot, int blocksize,
219     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
220 {
221 	return (dmu_object_alloc_impl(os, ot, blocksize, 0, bonustype,
222 	    bonuslen, dnodesize, tx));
223 }
224 
225 int
226 dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
227     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
228 {
229 	return (dmu_object_claim_dnsize(os, object, ot, blocksize, bonustype,
230 	    bonuslen, 0, tx));
231 }
232 
233 int
234 dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
235     int blocksize, dmu_object_type_t bonustype, int bonuslen,
236     int dnodesize, dmu_tx_t *tx)
237 {
238 	dnode_t *dn;
239 	int dn_slots = dnodesize >> DNODE_SHIFT;
240 	int err;
241 
242 	if (dn_slots == 0)
243 		dn_slots = DNODE_MIN_SLOTS;
244 	ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
245 	ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
246 
247 	if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx))
248 		return (SET_ERROR(EBADF));
249 
250 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, dn_slots,
251 	    FTAG, &dn);
252 	if (err)
253 		return (err);
254 	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, dn_slots, tx);
255 	dmu_tx_add_new_object(tx, dn);
256 
257 	dnode_rele(dn, FTAG);
258 
259 	return (0);
260 }
261 
262 int
263 dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
264     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
265 {
266 	return (dmu_object_reclaim_dnsize(os, object, ot, blocksize, bonustype,
267 	    bonuslen, DNODE_MIN_SIZE, B_FALSE, tx));
268 }
269 
270 int
271 dmu_object_reclaim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
272     int blocksize, dmu_object_type_t bonustype, int bonuslen, int dnodesize,
273     boolean_t keep_spill, dmu_tx_t *tx)
274 {
275 	dnode_t *dn;
276 	int dn_slots = dnodesize >> DNODE_SHIFT;
277 	int err;
278 
279 	if (dn_slots == 0)
280 		dn_slots = DNODE_MIN_SLOTS;
281 
282 	if (object == DMU_META_DNODE_OBJECT)
283 		return (SET_ERROR(EBADF));
284 
285 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
286 	    FTAG, &dn);
287 	if (err)
288 		return (err);
289 
290 	dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, dn_slots,
291 	    keep_spill, tx);
292 
293 	dnode_rele(dn, FTAG);
294 	return (err);
295 }
296 
297 int
298 dmu_object_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
299 {
300 	dnode_t *dn;
301 	int err;
302 
303 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
304 	    FTAG, &dn);
305 	if (err)
306 		return (err);
307 
308 	rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
309 	if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
310 		dbuf_rm_spill(dn, tx);
311 		dnode_rm_spill(dn, tx);
312 	}
313 	rw_exit(&dn->dn_struct_rwlock);
314 
315 	dnode_rele(dn, FTAG);
316 	return (err);
317 }
318 
319 int
320 dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
321 {
322 	dnode_t *dn;
323 	int err;
324 
325 	ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
326 
327 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
328 	    FTAG, &dn);
329 	if (err)
330 		return (err);
331 
332 	ASSERT(dn->dn_type != DMU_OT_NONE);
333 	/*
334 	 * If we don't create this free range, we'll leak indirect blocks when
335 	 * we get to freeing the dnode in syncing context.
336 	 */
337 	dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
338 	dnode_free(dn, tx);
339 	dnode_rele(dn, FTAG);
340 
341 	return (0);
342 }
343 
344 /*
345  * Return (in *objectp) the next object which is allocated (or a hole)
346  * after *object, taking into account only objects that may have been modified
347  * after the specified txg.
348  */
349 int
350 dmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg)
351 {
352 	uint64_t offset;
353 	uint64_t start_obj;
354 	struct dsl_dataset *ds = os->os_dsl_dataset;
355 	int error;
356 
357 	if (*objectp == 0) {
358 		start_obj = 1;
359 	} else if (ds && ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE]) {
360 		uint64_t i = *objectp + 1;
361 		uint64_t last_obj = *objectp | (DNODES_PER_BLOCK - 1);
362 		dmu_object_info_t doi;
363 
364 		/*
365 		 * Scan through the remaining meta dnode block. The contents
366 		 * of each slot in the block are known so it can be quickly
367 		 * checked. If the block is exhausted without a match then
368 		 * hand off to dnode_next_offset() for further scanning.
369 		 */
370 		while (i <= last_obj) {
371 			error = dmu_object_info(os, i, &doi);
372 			if (error == ENOENT) {
373 				if (hole) {
374 					*objectp = i;
375 					return (0);
376 				} else {
377 					i++;
378 				}
379 			} else if (error == EEXIST) {
380 				i++;
381 			} else if (error == 0) {
382 				if (hole) {
383 					i += doi.doi_dnodesize >> DNODE_SHIFT;
384 				} else {
385 					*objectp = i;
386 					return (0);
387 				}
388 			} else {
389 				return (error);
390 			}
391 		}
392 
393 		start_obj = i;
394 	} else {
395 		start_obj = *objectp + 1;
396 	}
397 
398 	offset = start_obj << DNODE_SHIFT;
399 
400 	error = dnode_next_offset(DMU_META_DNODE(os),
401 	    (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg);
402 
403 	*objectp = offset >> DNODE_SHIFT;
404 
405 	return (error);
406 }
407 
408 /*
409  * Turn this object from old_type into DMU_OTN_ZAP_METADATA, and bump the
410  * refcount on SPA_FEATURE_EXTENSIBLE_DATASET.
411  *
412  * Only for use from syncing context, on MOS objects.
413  */
414 void
415 dmu_object_zapify(objset_t *mos, uint64_t object, dmu_object_type_t old_type,
416     dmu_tx_t *tx)
417 {
418 	dnode_t *dn;
419 
420 	ASSERT(dmu_tx_is_syncing(tx));
421 
422 	VERIFY0(dnode_hold(mos, object, FTAG, &dn));
423 	if (dn->dn_type == DMU_OTN_ZAP_METADATA) {
424 		dnode_rele(dn, FTAG);
425 		return;
426 	}
427 	ASSERT3U(dn->dn_type, ==, old_type);
428 	ASSERT0(dn->dn_maxblkid);
429 
430 	/*
431 	 * We must initialize the ZAP data before changing the type,
432 	 * so that concurrent calls to *_is_zapified() can determine if
433 	 * the object has been completely zapified by checking the type.
434 	 */
435 	mzap_create_impl(mos, object, 0, 0, tx);
436 
437 	dn->dn_next_type[tx->tx_txg & TXG_MASK] = dn->dn_type =
438 	    DMU_OTN_ZAP_METADATA;
439 	dnode_setdirty(dn, tx);
440 	dnode_rele(dn, FTAG);
441 
442 	spa_feature_incr(dmu_objset_spa(mos),
443 	    SPA_FEATURE_EXTENSIBLE_DATASET, tx);
444 }
445 
446 void
447 dmu_object_free_zapified(objset_t *mos, uint64_t object, dmu_tx_t *tx)
448 {
449 	dnode_t *dn;
450 	dmu_object_type_t t;
451 
452 	ASSERT(dmu_tx_is_syncing(tx));
453 
454 	VERIFY0(dnode_hold(mos, object, FTAG, &dn));
455 	t = dn->dn_type;
456 	dnode_rele(dn, FTAG);
457 
458 	if (t == DMU_OTN_ZAP_METADATA) {
459 		spa_feature_decr(dmu_objset_spa(mos),
460 		    SPA_FEATURE_EXTENSIBLE_DATASET, tx);
461 	}
462 	VERIFY0(dmu_object_free(mos, object, tx));
463 }
464