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/dmu.h>
28 #include <sys/dmu_objset.h>
29 #include <sys/dmu_tx.h>
30 #include <sys/dnode.h>
31 #include <sys/zap.h>
32 #include <sys/zfeature.h>
33 #include <sys/dsl_dataset.h>
34 
35 /*
36  * Each of the concurrent object allocators will grab
37  * 2^dmu_object_alloc_chunk_shift dnode slots at a time.  The default is to
38  * grab 128 slots, which is 4 blocks worth.  This was experimentally
39  * determined to be the lowest value that eliminates the measurable effect
40  * of lock contention from this code path.
41  */
42 int dmu_object_alloc_chunk_shift = 7;
43 
44 static uint64_t
dmu_object_alloc_impl(objset_t * os,dmu_object_type_t ot,int blocksize,int indirect_blockshift,dmu_object_type_t bonustype,int bonuslen,int dnodesize,dmu_tx_t * tx)45 dmu_object_alloc_impl(objset_t *os, dmu_object_type_t ot, int blocksize,
46     int indirect_blockshift, dmu_object_type_t bonustype, int bonuslen,
47     int dnodesize, dmu_tx_t *tx)
48 {
49 	uint64_t object;
50 	uint64_t L1_dnode_count = DNODES_PER_BLOCK <<
51 	    (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT);
52 	dnode_t *dn = NULL;
53 	int dn_slots = dnodesize >> DNODE_SHIFT;
54 	boolean_t restarted = B_FALSE;
55 	uint64_t *cpuobj = &os->os_obj_next_percpu[CPU_SEQID %
56 	    os->os_obj_next_percpu_len];
57 	int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
58 	int error;
59 
60 	if (dn_slots == 0) {
61 		dn_slots = DNODE_MIN_SLOTS;
62 	} else {
63 		ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
64 		ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
65 	}
66 
67 	/*
68 	 * The "chunk" of dnodes that is assigned to a CPU-specific
69 	 * allocator needs to be at least one block's worth, to avoid
70 	 * lock contention on the dbuf.  It can be at most one L1 block's
71 	 * worth, so that the "rescan after polishing off a L1's worth"
72 	 * logic below will be sure to kick in.
73 	 */
74 	if (dnodes_per_chunk < DNODES_PER_BLOCK)
75 		dnodes_per_chunk = DNODES_PER_BLOCK;
76 	if (dnodes_per_chunk > L1_dnode_count)
77 		dnodes_per_chunk = L1_dnode_count;
78 
79 #ifdef __FreeBSD__
80 	object = atomic_load_64(cpuobj);
81 #else
82 	object = *cpuobj;
83 #endif
84 
85 	for (;;) {
86 		/*
87 		 * If we finished a chunk of dnodes, get a new one from
88 		 * the global allocator.
89 		 */
90 		if ((P2PHASE(object, dnodes_per_chunk) == 0) ||
91 		    (P2PHASE(object + dn_slots - 1, dnodes_per_chunk) <
92 		    dn_slots)) {
93 			DNODE_STAT_BUMP(dnode_alloc_next_chunk);
94 			mutex_enter(&os->os_obj_lock);
95 			ASSERT0(P2PHASE(os->os_obj_next_chunk,
96 			    dnodes_per_chunk));
97 			object = os->os_obj_next_chunk;
98 
99 			/*
100 			 * Each time we polish off a L1 bp worth of dnodes
101 			 * (2^12 objects), move to another L1 bp that's
102 			 * still reasonably sparse (at most 1/4 full). Look
103 			 * from the beginning at most once per txg. If we
104 			 * still can't allocate from that L1 block, search
105 			 * for an empty L0 block, which will quickly skip
106 			 * to the end of the metadnode if the no nearby L0
107 			 * blocks are empty. This fallback avoids a
108 			 * pathology where full dnode blocks containing
109 			 * large dnodes appear sparse because they have a
110 			 * low blk_fill, leading to many failed allocation
111 			 * attempts. In the long term a better mechanism to
112 			 * search for sparse metadnode regions, such as
113 			 * spacemaps, could be implemented.
114 			 *
115 			 * os_scan_dnodes is set during txg sync if enough
116 			 * objects have been freed since the previous
117 			 * rescan to justify backfilling again.
118 			 *
119 			 * Note that dmu_traverse depends on the behavior
120 			 * that we use multiple blocks of the dnode object
121 			 * before going back to reuse objects. Any change
122 			 * to this algorithm should preserve that property
123 			 * or find another solution to the issues described
124 			 * in traverse_visitbp.
125 			 */
126 			if (P2PHASE(object, L1_dnode_count) == 0) {
127 				uint64_t offset;
128 				uint64_t blkfill;
129 				int minlvl;
130 				if (os->os_rescan_dnodes) {
131 					offset = 0;
132 					os->os_rescan_dnodes = B_FALSE;
133 				} else {
134 					offset = object << DNODE_SHIFT;
135 				}
136 				blkfill = restarted ? 1 : DNODES_PER_BLOCK >> 2;
137 				minlvl = restarted ? 1 : 2;
138 				restarted = B_TRUE;
139 				error = dnode_next_offset(DMU_META_DNODE(os),
140 				    DNODE_FIND_HOLE, &offset, minlvl,
141 				    blkfill, 0);
142 				if (error == 0) {
143 					object = offset >> DNODE_SHIFT;
144 				}
145 			}
146 			/*
147 			 * Note: if "restarted", we may find a L0 that
148 			 * is not suitably aligned.
149 			 */
150 			os->os_obj_next_chunk =
151 			    P2ALIGN(object, dnodes_per_chunk) +
152 			    dnodes_per_chunk;
153 			(void) atomic_swap_64(cpuobj, object);
154 			mutex_exit(&os->os_obj_lock);
155 		}
156 
157 		/*
158 		 * The value of (*cpuobj) before adding dn_slots is the object
159 		 * ID assigned to us.  The value afterwards is the object ID
160 		 * assigned to whoever wants to do an allocation next.
161 		 */
162 		object = atomic_add_64_nv(cpuobj, dn_slots) - dn_slots;
163 
164 		/*
165 		 * XXX We should check for an i/o error here and return
166 		 * up to our caller.  Actually we should pre-read it in
167 		 * dmu_tx_assign(), but there is currently no mechanism
168 		 * to do so.
169 		 */
170 		error = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE,
171 		    dn_slots, FTAG, &dn);
172 		if (error == 0) {
173 			rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
174 			/*
175 			 * Another thread could have allocated it; check
176 			 * again now that we have the struct lock.
177 			 */
178 			if (dn->dn_type == DMU_OT_NONE) {
179 				dnode_allocate(dn, ot, blocksize, 0,
180 				    bonustype, bonuslen, dn_slots, tx);
181 				rw_exit(&dn->dn_struct_rwlock);
182 				dmu_tx_add_new_object(tx, dn);
183 				dnode_rele(dn, FTAG);
184 				return (object);
185 			}
186 			rw_exit(&dn->dn_struct_rwlock);
187 			dnode_rele(dn, FTAG);
188 			DNODE_STAT_BUMP(dnode_alloc_race);
189 		}
190 
191 		/*
192 		 * Skip to next known valid starting point on error. This
193 		 * is the start of the next block of dnodes.
194 		 */
195 		if (dmu_object_next(os, &object, B_TRUE, 0) != 0) {
196 			object = P2ROUNDUP(object + 1, DNODES_PER_BLOCK);
197 			DNODE_STAT_BUMP(dnode_alloc_next_block);
198 		}
199 		(void) atomic_swap_64(cpuobj, object);
200 	}
201 }
202 
203 uint64_t
dmu_object_alloc(objset_t * os,dmu_object_type_t ot,int blocksize,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)204 dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
205     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
206 {
207 	return (dmu_object_alloc_impl(os, ot, blocksize, 0, bonustype,
208 	    bonuslen, 0, tx));
209 }
210 
211 uint64_t
dmu_object_alloc_ibs(objset_t * os,dmu_object_type_t ot,int blocksize,int indirect_blockshift,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)212 dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
213     int indirect_blockshift, dmu_object_type_t bonustype, int bonuslen,
214     dmu_tx_t *tx)
215 {
216 	return (dmu_object_alloc_impl(os, ot, blocksize, indirect_blockshift,
217 	    bonustype, bonuslen, 0, tx));
218 }
219 
220 uint64_t
dmu_object_alloc_dnsize(objset_t * os,dmu_object_type_t ot,int blocksize,dmu_object_type_t bonustype,int bonuslen,int dnodesize,dmu_tx_t * tx)221 dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot, int blocksize,
222     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
223 {
224 	return (dmu_object_alloc_impl(os, ot, blocksize, 0, bonustype,
225 	    bonuslen, dnodesize, tx));
226 }
227 
228 int
dmu_object_claim(objset_t * os,uint64_t object,dmu_object_type_t ot,int blocksize,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)229 dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
230     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
231 {
232 	return (dmu_object_claim_dnsize(os, object, ot, blocksize, bonustype,
233 	    bonuslen, 0, tx));
234 }
235 
236 int
dmu_object_claim_dnsize(objset_t * os,uint64_t object,dmu_object_type_t ot,int blocksize,dmu_object_type_t bonustype,int bonuslen,int dnodesize,dmu_tx_t * tx)237 dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
238     int blocksize, dmu_object_type_t bonustype, int bonuslen,
239     int dnodesize, dmu_tx_t *tx)
240 {
241 	dnode_t *dn;
242 	int dn_slots = dnodesize >> DNODE_SHIFT;
243 	int err;
244 
245 	if (dn_slots == 0)
246 		dn_slots = DNODE_MIN_SLOTS;
247 	ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
248 	ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
249 
250 	if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx))
251 		return (SET_ERROR(EBADF));
252 
253 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, dn_slots,
254 	    FTAG, &dn);
255 	if (err)
256 		return (err);
257 	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, dn_slots, tx);
258 	dmu_tx_add_new_object(tx, dn);
259 
260 	dnode_rele(dn, FTAG);
261 
262 	return (0);
263 }
264 
265 int
dmu_object_reclaim(objset_t * os,uint64_t object,dmu_object_type_t ot,int blocksize,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)266 dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
267     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
268 {
269 	return (dmu_object_reclaim_dnsize(os, object, ot, blocksize, bonustype,
270 	    bonuslen, DNODE_MIN_SIZE, tx));
271 }
272 
273 int
dmu_object_reclaim_dnsize(objset_t * os,uint64_t object,dmu_object_type_t ot,int blocksize,dmu_object_type_t bonustype,int bonuslen,int dnodesize,dmu_tx_t * tx)274 dmu_object_reclaim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
275     int blocksize, dmu_object_type_t bonustype, int bonuslen, int dnodesize,
276     dmu_tx_t *tx)
277 {
278 	dnode_t *dn;
279 	int dn_slots = dnodesize >> DNODE_SHIFT;
280 	int err;
281 
282 	if (dn_slots == 0)
283 		dn_slots = DNODE_MIN_SLOTS;
284 
285 	if (object == DMU_META_DNODE_OBJECT)
286 		return (SET_ERROR(EBADF));
287 
288 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
289 	    FTAG, &dn);
290 	if (err)
291 		return (err);
292 
293 	dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, dn_slots, tx);
294 
295 	dnode_rele(dn, FTAG);
296 	return (err);
297 }
298 
299 
300 int
dmu_object_free(objset_t * os,uint64_t object,dmu_tx_t * tx)301 dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
302 {
303 	dnode_t *dn;
304 	int err;
305 
306 	ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
307 
308 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
309 	    FTAG, &dn);
310 	if (err)
311 		return (err);
312 
313 	ASSERT(dn->dn_type != DMU_OT_NONE);
314 	/*
315 	 * If we don't create this free range, we'll leak indirect blocks when
316 	 * we get to freeing the dnode in syncing context.
317 	 */
318 	dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
319 	dnode_free(dn, tx);
320 	dnode_rele(dn, FTAG);
321 
322 	return (0);
323 }
324 
325 /*
326  * Return (in *objectp) the next object which is allocated (or a hole)
327  * after *object, taking into account only objects that may have been modified
328  * after the specified txg.
329  */
330 int
dmu_object_next(objset_t * os,uint64_t * objectp,boolean_t hole,uint64_t txg)331 dmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg)
332 {
333 	uint64_t offset;
334 	uint64_t start_obj;
335 	struct dsl_dataset *ds = os->os_dsl_dataset;
336 	int error;
337 
338 	if (*objectp == 0) {
339 		start_obj = 1;
340 	} else if (ds && ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE]) {
341 		uint64_t i = *objectp + 1;
342 		uint64_t last_obj = *objectp | (DNODES_PER_BLOCK - 1);
343 		dmu_object_info_t doi;
344 
345 		/*
346 		 * Scan through the remaining meta dnode block. The contents
347 		 * of each slot in the block are known so it can be quickly
348 		 * checked. If the block is exhausted without a match then
349 		 * hand off to dnode_next_offset() for further scanning.
350 		 */
351 		while (i <= last_obj) {
352 			error = dmu_object_info(os, i, &doi);
353 			if (error == ENOENT) {
354 				if (hole) {
355 					*objectp = i;
356 					return (0);
357 				} else {
358 					i++;
359 				}
360 			} else if (error == EEXIST) {
361 				i++;
362 			} else if (error == 0) {
363 				if (hole) {
364 					i += doi.doi_dnodesize >> DNODE_SHIFT;
365 				} else {
366 					*objectp = i;
367 					return (0);
368 				}
369 			} else {
370 				return (error);
371 			}
372 		}
373 
374 		start_obj = i;
375 	} else {
376 		start_obj = *objectp + 1;
377 	}
378 
379 	offset = start_obj << DNODE_SHIFT;
380 
381 	error = dnode_next_offset(DMU_META_DNODE(os),
382 	    (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg);
383 
384 	*objectp = offset >> DNODE_SHIFT;
385 
386 	return (error);
387 }
388 
389 /*
390  * Turn this object from old_type into DMU_OTN_ZAP_METADATA, and bump the
391  * refcount on SPA_FEATURE_EXTENSIBLE_DATASET.
392  *
393  * Only for use from syncing context, on MOS objects.
394  */
395 void
dmu_object_zapify(objset_t * mos,uint64_t object,dmu_object_type_t old_type,dmu_tx_t * tx)396 dmu_object_zapify(objset_t *mos, uint64_t object, dmu_object_type_t old_type,
397     dmu_tx_t *tx)
398 {
399 	dnode_t *dn;
400 
401 	ASSERT(dmu_tx_is_syncing(tx));
402 
403 	VERIFY0(dnode_hold(mos, object, FTAG, &dn));
404 	if (dn->dn_type == DMU_OTN_ZAP_METADATA) {
405 		dnode_rele(dn, FTAG);
406 		return;
407 	}
408 	ASSERT3U(dn->dn_type, ==, old_type);
409 	ASSERT0(dn->dn_maxblkid);
410 
411 	/*
412 	 * We must initialize the ZAP data before changing the type,
413 	 * so that concurrent calls to *_is_zapified() can determine if
414 	 * the object has been completely zapified by checking the type.
415 	 */
416 	mzap_create_impl(mos, object, 0, 0, tx);
417 
418 	dn->dn_next_type[tx->tx_txg & TXG_MASK] = dn->dn_type =
419 	    DMU_OTN_ZAP_METADATA;
420 	dnode_setdirty(dn, tx);
421 	dnode_rele(dn, FTAG);
422 
423 	spa_feature_incr(dmu_objset_spa(mos),
424 	    SPA_FEATURE_EXTENSIBLE_DATASET, tx);
425 }
426 
427 void
dmu_object_free_zapified(objset_t * mos,uint64_t object,dmu_tx_t * tx)428 dmu_object_free_zapified(objset_t *mos, uint64_t object, dmu_tx_t *tx)
429 {
430 	dnode_t *dn;
431 	dmu_object_type_t t;
432 
433 	ASSERT(dmu_tx_is_syncing(tx));
434 
435 	VERIFY0(dnode_hold(mos, object, FTAG, &dn));
436 	t = dn->dn_type;
437 	dnode_rele(dn, FTAG);
438 
439 	if (t == DMU_OTN_ZAP_METADATA) {
440 		spa_feature_decr(dmu_objset_spa(mos),
441 		    SPA_FEATURE_EXTENSIBLE_DATASET, tx);
442 	}
443 	VERIFY0(dmu_object_free(mos, object, tx));
444 }
445