xref: /netbsd/external/cddl/osnet/dist/uts/common/fs/zfs/dmu.c (revision eada09ac)
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) 2011, 2016 by Delphix. All rights reserved.
24  */
25 /* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */
26 /* Copyright (c) 2013, Joyent, Inc. All rights reserved. */
27 /* Copyright (c) 2014, Nexenta Systems, Inc. All rights reserved. */
28 
29 #include <sys/dmu.h>
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dbuf.h>
33 #include <sys/dnode.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_pool.h>
40 #include <sys/dsl_synctask.h>
41 #include <sys/dsl_prop.h>
42 #include <sys/dmu_zfetch.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/zap.h>
45 #include <sys/zio_checksum.h>
46 #include <sys/zio_compress.h>
47 #include <sys/sa.h>
48 #include <sys/zfeature.h>
49 #ifdef _KERNEL
50 #include <sys/racct.h>
51 #include <sys/vm.h>
52 #include <sys/zfs_znode.h>
53 #endif
54 
55 /*
56  * Enable/disable nopwrite feature.
57  */
58 int zfs_nopwrite_enabled = 1;
59 SYSCTL_DECL(_vfs_zfs);
60 SYSCTL_INT(_vfs_zfs, OID_AUTO, nopwrite_enabled, CTLFLAG_RDTUN,
61     &zfs_nopwrite_enabled, 0, "Enable nopwrite feature");
62 
63 /*
64  * Tunable to control percentage of dirtied blocks from frees in one TXG.
65  * After this threshold is crossed, additional dirty blocks from frees
66  * wait until the next TXG.
67  * A value of zero will disable this throttle.
68  */
69 uint32_t zfs_per_txg_dirty_frees_percent = 30;
70 SYSCTL_INT(_vfs_zfs, OID_AUTO, per_txg_dirty_frees_percent, CTLFLAG_RWTUN,
71 	&zfs_per_txg_dirty_frees_percent, 0, "Percentage of dirtied blocks from frees in one txg");
72 
73 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
74 	{	DMU_BSWAP_UINT8,	TRUE,	"unallocated"		},
75 	{	DMU_BSWAP_ZAP,		TRUE,	"object directory"	},
76 	{	DMU_BSWAP_UINT64,	TRUE,	"object array"		},
77 	{	DMU_BSWAP_UINT8,	TRUE,	"packed nvlist"		},
78 	{	DMU_BSWAP_UINT64,	TRUE,	"packed nvlist size"	},
79 	{	DMU_BSWAP_UINT64,	TRUE,	"bpobj"			},
80 	{	DMU_BSWAP_UINT64,	TRUE,	"bpobj header"		},
81 	{	DMU_BSWAP_UINT64,	TRUE,	"SPA space map header"	},
82 	{	DMU_BSWAP_UINT64,	TRUE,	"SPA space map"		},
83 	{	DMU_BSWAP_UINT64,	TRUE,	"ZIL intent log"	},
84 	{	DMU_BSWAP_DNODE,	TRUE,	"DMU dnode"		},
85 	{	DMU_BSWAP_OBJSET,	TRUE,	"DMU objset"		},
86 	{	DMU_BSWAP_UINT64,	TRUE,	"DSL directory"		},
87 	{	DMU_BSWAP_ZAP,		TRUE,	"DSL directory child map"},
88 	{	DMU_BSWAP_ZAP,		TRUE,	"DSL dataset snap map"	},
89 	{	DMU_BSWAP_ZAP,		TRUE,	"DSL props"		},
90 	{	DMU_BSWAP_UINT64,	TRUE,	"DSL dataset"		},
91 	{	DMU_BSWAP_ZNODE,	TRUE,	"ZFS znode"		},
92 	{	DMU_BSWAP_OLDACL,	TRUE,	"ZFS V0 ACL"		},
93 	{	DMU_BSWAP_UINT8,	FALSE,	"ZFS plain file"	},
94 	{	DMU_BSWAP_ZAP,		TRUE,	"ZFS directory"		},
95 	{	DMU_BSWAP_ZAP,		TRUE,	"ZFS master node"	},
96 	{	DMU_BSWAP_ZAP,		TRUE,	"ZFS delete queue"	},
97 	{	DMU_BSWAP_UINT8,	FALSE,	"zvol object"		},
98 	{	DMU_BSWAP_ZAP,		TRUE,	"zvol prop"		},
99 	{	DMU_BSWAP_UINT8,	FALSE,	"other uint8[]"		},
100 	{	DMU_BSWAP_UINT64,	FALSE,	"other uint64[]"	},
101 	{	DMU_BSWAP_ZAP,		TRUE,	"other ZAP"		},
102 	{	DMU_BSWAP_ZAP,		TRUE,	"persistent error log"	},
103 	{	DMU_BSWAP_UINT8,	TRUE,	"SPA history"		},
104 	{	DMU_BSWAP_UINT64,	TRUE,	"SPA history offsets"	},
105 	{	DMU_BSWAP_ZAP,		TRUE,	"Pool properties"	},
106 	{	DMU_BSWAP_ZAP,		TRUE,	"DSL permissions"	},
107 	{	DMU_BSWAP_ACL,		TRUE,	"ZFS ACL"		},
108 	{	DMU_BSWAP_UINT8,	TRUE,	"ZFS SYSACL"		},
109 	{	DMU_BSWAP_UINT8,	TRUE,	"FUID table"		},
110 	{	DMU_BSWAP_UINT64,	TRUE,	"FUID table size"	},
111 	{	DMU_BSWAP_ZAP,		TRUE,	"DSL dataset next clones"},
112 	{	DMU_BSWAP_ZAP,		TRUE,	"scan work queue"	},
113 	{	DMU_BSWAP_ZAP,		TRUE,	"ZFS user/group used"	},
114 	{	DMU_BSWAP_ZAP,		TRUE,	"ZFS user/group quota"	},
115 	{	DMU_BSWAP_ZAP,		TRUE,	"snapshot refcount tags"},
116 	{	DMU_BSWAP_ZAP,		TRUE,	"DDT ZAP algorithm"	},
117 	{	DMU_BSWAP_ZAP,		TRUE,	"DDT statistics"	},
118 	{	DMU_BSWAP_UINT8,	TRUE,	"System attributes"	},
119 	{	DMU_BSWAP_ZAP,		TRUE,	"SA master node"	},
120 	{	DMU_BSWAP_ZAP,		TRUE,	"SA attr registration"	},
121 	{	DMU_BSWAP_ZAP,		TRUE,	"SA attr layouts"	},
122 	{	DMU_BSWAP_ZAP,		TRUE,	"scan translations"	},
123 	{	DMU_BSWAP_UINT8,	FALSE,	"deduplicated block"	},
124 	{	DMU_BSWAP_ZAP,		TRUE,	"DSL deadlist map"	},
125 	{	DMU_BSWAP_UINT64,	TRUE,	"DSL deadlist map hdr"	},
126 	{	DMU_BSWAP_ZAP,		TRUE,	"DSL dir clones"	},
127 	{	DMU_BSWAP_UINT64,	TRUE,	"bpobj subobj"		}
128 };
129 
130 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
131 	{	byteswap_uint8_array,	"uint8"		},
132 	{	byteswap_uint16_array,	"uint16"	},
133 	{	byteswap_uint32_array,	"uint32"	},
134 	{	byteswap_uint64_array,	"uint64"	},
135 	{	zap_byteswap,		"zap"		},
136 	{	dnode_buf_byteswap,	"dnode"		},
137 	{	dmu_objset_byteswap,	"objset"	},
138 	{	zfs_znode_byteswap,	"znode"		},
139 	{	zfs_oldacl_byteswap,	"oldacl"	},
140 	{	zfs_acl_byteswap,	"acl"		}
141 };
142 
143 int
dmu_buf_hold_noread_by_dnode(dnode_t * dn,uint64_t offset,void * tag,dmu_buf_t ** dbp)144 dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset,
145     void *tag, dmu_buf_t **dbp)
146 {
147 	uint64_t blkid;
148 	dmu_buf_impl_t *db;
149 
150 	blkid = dbuf_whichblock(dn, 0, offset);
151 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
152 	db = dbuf_hold(dn, blkid, tag);
153 	rw_exit(&dn->dn_struct_rwlock);
154 
155 	if (db == NULL) {
156 		*dbp = NULL;
157 		return (SET_ERROR(EIO));
158 	}
159 
160 	*dbp = &db->db;
161 	return (0);
162 }
163 int
dmu_buf_hold_noread(objset_t * os,uint64_t object,uint64_t offset,void * tag,dmu_buf_t ** dbp)164 dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
165     void *tag, dmu_buf_t **dbp)
166 {
167 	dnode_t *dn;
168 	uint64_t blkid;
169 	dmu_buf_impl_t *db;
170 	int err;
171 
172 	err = dnode_hold(os, object, FTAG, &dn);
173 	if (err)
174 		return (err);
175 	blkid = dbuf_whichblock(dn, 0, offset);
176 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
177 	db = dbuf_hold(dn, blkid, tag);
178 	rw_exit(&dn->dn_struct_rwlock);
179 	dnode_rele(dn, FTAG);
180 
181 	if (db == NULL) {
182 		*dbp = NULL;
183 		return (SET_ERROR(EIO));
184 	}
185 
186 	*dbp = &db->db;
187 	return (err);
188 }
189 
190 int
dmu_buf_hold_by_dnode(dnode_t * dn,uint64_t offset,void * tag,dmu_buf_t ** dbp,int flags)191 dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
192     void *tag, dmu_buf_t **dbp, int flags)
193 {
194 	int err;
195 	int db_flags = DB_RF_CANFAIL;
196 
197 	if (flags & DMU_READ_NO_PREFETCH)
198 		db_flags |= DB_RF_NOPREFETCH;
199 
200 	err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp);
201 	if (err == 0) {
202 		dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
203 		err = dbuf_read(db, NULL, db_flags);
204 		if (err != 0) {
205 			dbuf_rele(db, tag);
206 			*dbp = NULL;
207 		}
208 	}
209 
210 	return (err);
211 }
212 
213 int
dmu_buf_hold(objset_t * os,uint64_t object,uint64_t offset,void * tag,dmu_buf_t ** dbp,int flags)214 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
215     void *tag, dmu_buf_t **dbp, int flags)
216 {
217 	int err;
218 	int db_flags = DB_RF_CANFAIL;
219 
220 	if (flags & DMU_READ_NO_PREFETCH)
221 		db_flags |= DB_RF_NOPREFETCH;
222 
223 	err = dmu_buf_hold_noread(os, object, offset, tag, dbp);
224 	if (err == 0) {
225 		dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
226 		err = dbuf_read(db, NULL, db_flags);
227 		if (err != 0) {
228 			dbuf_rele(db, tag);
229 			*dbp = NULL;
230 		}
231 	}
232 
233 	return (err);
234 }
235 
236 int
dmu_bonus_max(void)237 dmu_bonus_max(void)
238 {
239 	return (DN_MAX_BONUSLEN);
240 }
241 
242 int
dmu_set_bonus(dmu_buf_t * db_fake,int newsize,dmu_tx_t * tx)243 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
244 {
245 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
246 	dnode_t *dn;
247 	int error;
248 
249 	DB_DNODE_ENTER(db);
250 	dn = DB_DNODE(db);
251 
252 	if (dn->dn_bonus != db) {
253 		error = SET_ERROR(EINVAL);
254 	} else if (newsize < 0 || newsize > db_fake->db_size) {
255 		error = SET_ERROR(EINVAL);
256 	} else {
257 		dnode_setbonuslen(dn, newsize, tx);
258 		error = 0;
259 	}
260 
261 	DB_DNODE_EXIT(db);
262 	return (error);
263 }
264 
265 int
dmu_set_bonustype(dmu_buf_t * db_fake,dmu_object_type_t type,dmu_tx_t * tx)266 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
267 {
268 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
269 	dnode_t *dn;
270 	int error;
271 
272 	DB_DNODE_ENTER(db);
273 	dn = DB_DNODE(db);
274 
275 	if (!DMU_OT_IS_VALID(type)) {
276 		error = SET_ERROR(EINVAL);
277 	} else if (dn->dn_bonus != db) {
278 		error = SET_ERROR(EINVAL);
279 	} else {
280 		dnode_setbonus_type(dn, type, tx);
281 		error = 0;
282 	}
283 
284 	DB_DNODE_EXIT(db);
285 	return (error);
286 }
287 
288 dmu_object_type_t
dmu_get_bonustype(dmu_buf_t * db_fake)289 dmu_get_bonustype(dmu_buf_t *db_fake)
290 {
291 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
292 	dnode_t *dn;
293 	dmu_object_type_t type;
294 
295 	DB_DNODE_ENTER(db);
296 	dn = DB_DNODE(db);
297 	type = dn->dn_bonustype;
298 	DB_DNODE_EXIT(db);
299 
300 	return (type);
301 }
302 
303 int
dmu_rm_spill(objset_t * os,uint64_t object,dmu_tx_t * tx)304 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
305 {
306 	dnode_t *dn;
307 	int error;
308 
309 	error = dnode_hold(os, object, FTAG, &dn);
310 	dbuf_rm_spill(dn, tx);
311 	rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
312 	dnode_rm_spill(dn, tx);
313 	rw_exit(&dn->dn_struct_rwlock);
314 	dnode_rele(dn, FTAG);
315 	return (error);
316 }
317 
318 /*
319  * returns ENOENT, EIO, or 0.
320  */
321 int
dmu_bonus_hold(objset_t * os,uint64_t object,void * tag,dmu_buf_t ** dbp)322 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
323 {
324 	dnode_t *dn;
325 	dmu_buf_impl_t *db;
326 	int error;
327 
328 	error = dnode_hold(os, object, FTAG, &dn);
329 	if (error)
330 		return (error);
331 
332 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
333 	if (dn->dn_bonus == NULL) {
334 		rw_exit(&dn->dn_struct_rwlock);
335 		rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
336 		if (dn->dn_bonus == NULL)
337 			dbuf_create_bonus(dn);
338 	}
339 	db = dn->dn_bonus;
340 
341 	/* as long as the bonus buf is held, the dnode will be held */
342 	if (refcount_add(&db->db_holds, tag) == 1) {
343 		VERIFY(dnode_add_ref(dn, db));
344 		atomic_inc_32(&dn->dn_dbufs_count);
345 	}
346 
347 	/*
348 	 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
349 	 * hold and incrementing the dbuf count to ensure that dnode_move() sees
350 	 * a dnode hold for every dbuf.
351 	 */
352 	rw_exit(&dn->dn_struct_rwlock);
353 
354 	dnode_rele(dn, FTAG);
355 
356 	VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
357 
358 	*dbp = &db->db;
359 	return (0);
360 }
361 
362 /*
363  * returns ENOENT, EIO, or 0.
364  *
365  * This interface will allocate a blank spill dbuf when a spill blk
366  * doesn't already exist on the dnode.
367  *
368  * if you only want to find an already existing spill db, then
369  * dmu_spill_hold_existing() should be used.
370  */
371 int
dmu_spill_hold_by_dnode(dnode_t * dn,uint32_t flags,void * tag,dmu_buf_t ** dbp)372 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
373 {
374 	dmu_buf_impl_t *db = NULL;
375 	int err;
376 
377 	if ((flags & DB_RF_HAVESTRUCT) == 0)
378 		rw_enter(&dn->dn_struct_rwlock, RW_READER);
379 
380 	db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
381 
382 	if ((flags & DB_RF_HAVESTRUCT) == 0)
383 		rw_exit(&dn->dn_struct_rwlock);
384 
385 	ASSERT(db != NULL);
386 	err = dbuf_read(db, NULL, flags);
387 	if (err == 0)
388 		*dbp = &db->db;
389 	else
390 		dbuf_rele(db, tag);
391 	return (err);
392 }
393 
394 int
dmu_spill_hold_existing(dmu_buf_t * bonus,void * tag,dmu_buf_t ** dbp)395 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
396 {
397 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
398 	dnode_t *dn;
399 	int err;
400 
401 	DB_DNODE_ENTER(db);
402 	dn = DB_DNODE(db);
403 
404 	if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
405 		err = SET_ERROR(EINVAL);
406 	} else {
407 		rw_enter(&dn->dn_struct_rwlock, RW_READER);
408 
409 		if (!dn->dn_have_spill) {
410 			err = SET_ERROR(ENOENT);
411 		} else {
412 			err = dmu_spill_hold_by_dnode(dn,
413 			    DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
414 		}
415 
416 		rw_exit(&dn->dn_struct_rwlock);
417 	}
418 
419 	DB_DNODE_EXIT(db);
420 	return (err);
421 }
422 
423 int
dmu_spill_hold_by_bonus(dmu_buf_t * bonus,void * tag,dmu_buf_t ** dbp)424 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
425 {
426 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
427 	dnode_t *dn;
428 	int err;
429 
430 	DB_DNODE_ENTER(db);
431 	dn = DB_DNODE(db);
432 	err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
433 	DB_DNODE_EXIT(db);
434 
435 	return (err);
436 }
437 
438 /*
439  * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
440  * to take a held dnode rather than <os, object> -- the lookup is wasteful,
441  * and can induce severe lock contention when writing to several files
442  * whose dnodes are in the same block.
443  */
444 static int
dmu_buf_hold_array_by_dnode(dnode_t * dn,uint64_t offset,uint64_t length,boolean_t read,void * tag,int * numbufsp,dmu_buf_t *** dbpp,uint32_t flags)445 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
446     boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
447 {
448 	dmu_buf_t **dbp;
449 	uint64_t blkid, nblks, i;
450 	uint32_t dbuf_flags;
451 	int err;
452 	zio_t *zio;
453 
454 	ASSERT(length <= DMU_MAX_ACCESS);
455 
456 	/*
457 	 * Note: We directly notify the prefetch code of this read, so that
458 	 * we can tell it about the multi-block read.  dbuf_read() only knows
459 	 * about the one block it is accessing.
460 	 */
461 	dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT |
462 	    DB_RF_NOPREFETCH;
463 
464 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
465 	if (dn->dn_datablkshift) {
466 		int blkshift = dn->dn_datablkshift;
467 		nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) -
468 		    P2ALIGN(offset, 1ULL << blkshift)) >> blkshift;
469 	} else {
470 		if (offset + length > dn->dn_datablksz) {
471 			zfs_panic_recover("zfs: accessing past end of object "
472 			    "%llx/%llx (size=%u access=%llu+%llu)",
473 			    (longlong_t)dn->dn_objset->
474 			    os_dsl_dataset->ds_object,
475 			    (longlong_t)dn->dn_object, dn->dn_datablksz,
476 			    (longlong_t)offset, (longlong_t)length);
477 			rw_exit(&dn->dn_struct_rwlock);
478 			return (SET_ERROR(EIO));
479 		}
480 		nblks = 1;
481 	}
482 	dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
483 
484 #if defined(_KERNEL) && defined(RACCT)
485 	if (racct_enable && !read) {
486 		PROC_LOCK(curproc);
487 		racct_add_force(curproc, RACCT_WRITEBPS, length);
488 		racct_add_force(curproc, RACCT_WRITEIOPS, nblks);
489 		PROC_UNLOCK(curproc);
490 	}
491 #endif
492 
493 	zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
494 	blkid = dbuf_whichblock(dn, 0, offset);
495 	for (i = 0; i < nblks; i++) {
496 		dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag);
497 		if (db == NULL) {
498 			rw_exit(&dn->dn_struct_rwlock);
499 			dmu_buf_rele_array(dbp, nblks, tag);
500 			zio_nowait(zio);
501 			return (SET_ERROR(EIO));
502 		}
503 
504 		/* initiate async i/o */
505 		if (read)
506 			(void) dbuf_read(db, zio, dbuf_flags);
507 
508 #ifdef _KERNEL
509 		else
510 			curthread->td_ru.ru_oublock++;
511 #endif
512 		dbp[i] = &db->db;
513 	}
514 
515 	if ((flags & DMU_READ_NO_PREFETCH) == 0 &&
516 	    DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) {
517 		dmu_zfetch(&dn->dn_zfetch, blkid, nblks,
518 		    read && DNODE_IS_CACHEABLE(dn));
519 	}
520 	rw_exit(&dn->dn_struct_rwlock);
521 
522 	/* wait for async i/o */
523 	err = zio_wait(zio);
524 	if (err) {
525 		dmu_buf_rele_array(dbp, nblks, tag);
526 		return (err);
527 	}
528 
529 	/* wait for other io to complete */
530 	if (read) {
531 		for (i = 0; i < nblks; i++) {
532 			dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
533 			mutex_enter(&db->db_mtx);
534 			while (db->db_state == DB_READ ||
535 			    db->db_state == DB_FILL)
536 				cv_wait(&db->db_changed, &db->db_mtx);
537 			if (db->db_state == DB_UNCACHED)
538 				err = SET_ERROR(EIO);
539 			mutex_exit(&db->db_mtx);
540 			if (err) {
541 				dmu_buf_rele_array(dbp, nblks, tag);
542 				return (err);
543 			}
544 		}
545 	}
546 
547 	*numbufsp = nblks;
548 	*dbpp = dbp;
549 	return (0);
550 }
551 
552 static int
dmu_buf_hold_array(objset_t * os,uint64_t object,uint64_t offset,uint64_t length,int read,void * tag,int * numbufsp,dmu_buf_t *** dbpp)553 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
554     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
555 {
556 	dnode_t *dn;
557 	int err;
558 
559 	err = dnode_hold(os, object, FTAG, &dn);
560 	if (err)
561 		return (err);
562 
563 	err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
564 	    numbufsp, dbpp, DMU_READ_PREFETCH);
565 
566 	dnode_rele(dn, FTAG);
567 
568 	return (err);
569 }
570 
571 int
dmu_buf_hold_array_by_bonus(dmu_buf_t * db_fake,uint64_t offset,uint64_t length,boolean_t read,void * tag,int * numbufsp,dmu_buf_t *** dbpp)572 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
573     uint64_t length, boolean_t read, void *tag, int *numbufsp,
574     dmu_buf_t ***dbpp)
575 {
576 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
577 	dnode_t *dn;
578 	int err;
579 
580 	DB_DNODE_ENTER(db);
581 	dn = DB_DNODE(db);
582 	err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
583 	    numbufsp, dbpp, DMU_READ_PREFETCH);
584 	DB_DNODE_EXIT(db);
585 
586 	return (err);
587 }
588 
589 void
dmu_buf_rele_array(dmu_buf_t ** dbp_fake,int numbufs,void * tag)590 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
591 {
592 	int i;
593 	dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
594 
595 	if (numbufs == 0)
596 		return;
597 
598 	for (i = 0; i < numbufs; i++) {
599 		if (dbp[i])
600 			dbuf_rele(dbp[i], tag);
601 	}
602 
603 	kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
604 }
605 
606 /*
607  * Issue prefetch i/os for the given blocks.  If level is greater than 0, the
608  * indirect blocks prefeteched will be those that point to the blocks containing
609  * the data starting at offset, and continuing to offset + len.
610  *
611  * Note that if the indirect blocks above the blocks being prefetched are not in
612  * cache, they will be asychronously read in.
613  */
614 void
dmu_prefetch(objset_t * os,uint64_t object,int64_t level,uint64_t offset,uint64_t len,zio_priority_t pri)615 dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
616     uint64_t len, zio_priority_t pri)
617 {
618 	dnode_t *dn;
619 	uint64_t blkid;
620 	int nblks, err;
621 
622 	if (len == 0) {  /* they're interested in the bonus buffer */
623 		dn = DMU_META_DNODE(os);
624 
625 		if (object == 0 || object >= DN_MAX_OBJECT)
626 			return;
627 
628 		rw_enter(&dn->dn_struct_rwlock, RW_READER);
629 		blkid = dbuf_whichblock(dn, level,
630 		    object * sizeof (dnode_phys_t));
631 		dbuf_prefetch(dn, level, blkid, pri, 0);
632 		rw_exit(&dn->dn_struct_rwlock);
633 		return;
634 	}
635 
636 	/*
637 	 * XXX - Note, if the dnode for the requested object is not
638 	 * already cached, we will do a *synchronous* read in the
639 	 * dnode_hold() call.  The same is true for any indirects.
640 	 */
641 	err = dnode_hold(os, object, FTAG, &dn);
642 	if (err != 0)
643 		return;
644 
645 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
646 	/*
647 	 * offset + len - 1 is the last byte we want to prefetch for, and offset
648 	 * is the first.  Then dbuf_whichblk(dn, level, off + len - 1) is the
649 	 * last block we want to prefetch, and dbuf_whichblock(dn, level,
650 	 * offset)  is the first.  Then the number we need to prefetch is the
651 	 * last - first + 1.
652 	 */
653 	if (level > 0 || dn->dn_datablkshift != 0) {
654 		nblks = dbuf_whichblock(dn, level, offset + len - 1) -
655 		    dbuf_whichblock(dn, level, offset) + 1;
656 	} else {
657 		nblks = (offset < dn->dn_datablksz);
658 	}
659 
660 	if (nblks != 0) {
661 		blkid = dbuf_whichblock(dn, level, offset);
662 		for (int i = 0; i < nblks; i++)
663 			dbuf_prefetch(dn, level, blkid + i, pri, 0);
664 	}
665 
666 	rw_exit(&dn->dn_struct_rwlock);
667 
668 	dnode_rele(dn, FTAG);
669 }
670 
671 /*
672  * Get the next "chunk" of file data to free.  We traverse the file from
673  * the end so that the file gets shorter over time (if we crashes in the
674  * middle, this will leave us in a better state).  We find allocated file
675  * data by simply searching the allocated level 1 indirects.
676  *
677  * On input, *start should be the first offset that does not need to be
678  * freed (e.g. "offset + length").  On return, *start will be the first
679  * offset that should be freed.
680  */
681 static int
get_next_chunk(dnode_t * dn,uint64_t * start,uint64_t minimum)682 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum)
683 {
684 	uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
685 	/* bytes of data covered by a level-1 indirect block */
686 	uint64_t iblkrange =
687 	    dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
688 
689 	ASSERT3U(minimum, <=, *start);
690 
691 	if (*start - minimum <= iblkrange * maxblks) {
692 		*start = minimum;
693 		return (0);
694 	}
695 	ASSERT(ISP2(iblkrange));
696 
697 	for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) {
698 		int err;
699 
700 		/*
701 		 * dnode_next_offset(BACKWARDS) will find an allocated L1
702 		 * indirect block at or before the input offset.  We must
703 		 * decrement *start so that it is at the end of the region
704 		 * to search.
705 		 */
706 		(*start)--;
707 		err = dnode_next_offset(dn,
708 		    DNODE_FIND_BACKWARDS, start, 2, 1, 0);
709 
710 		/* if there are no indirect blocks before start, we are done */
711 		if (err == ESRCH) {
712 			*start = minimum;
713 			break;
714 		} else if (err != 0) {
715 			return (err);
716 		}
717 
718 		/* set start to the beginning of this L1 indirect */
719 		*start = P2ALIGN(*start, iblkrange);
720 	}
721 	if (*start < minimum)
722 		*start = minimum;
723 	return (0);
724 }
725 
726 static int
dmu_free_long_range_impl(objset_t * os,dnode_t * dn,uint64_t offset,uint64_t length)727 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
728     uint64_t length)
729 {
730 	uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
731 	int err;
732 	uint64_t dirty_frees_threshold;
733 	dsl_pool_t *dp = dmu_objset_pool(os);
734 
735 	if (offset >= object_size)
736 		return (0);
737 
738 	if (zfs_per_txg_dirty_frees_percent <= 100)
739 		dirty_frees_threshold =
740 		    zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100;
741 	else
742 		dirty_frees_threshold = zfs_dirty_data_max / 4;
743 
744 	if (length == DMU_OBJECT_END || offset + length > object_size)
745 		length = object_size - offset;
746 
747 	while (length != 0) {
748 		uint64_t chunk_end, chunk_begin, chunk_len;
749 		uint64_t long_free_dirty_all_txgs = 0;
750 		dmu_tx_t *tx;
751 
752 		chunk_end = chunk_begin = offset + length;
753 
754 		/* move chunk_begin backwards to the beginning of this chunk */
755 		err = get_next_chunk(dn, &chunk_begin, offset);
756 		if (err)
757 			return (err);
758 		ASSERT3U(chunk_begin, >=, offset);
759 		ASSERT3U(chunk_begin, <=, chunk_end);
760 
761 		chunk_len = chunk_end - chunk_begin;
762 
763 		mutex_enter(&dp->dp_lock);
764 		for (int t = 0; t < TXG_SIZE; t++) {
765 			long_free_dirty_all_txgs +=
766 			    dp->dp_long_free_dirty_pertxg[t];
767 		}
768 		mutex_exit(&dp->dp_lock);
769 
770 		/*
771 		 * To avoid filling up a TXG with just frees wait for
772 		 * the next TXG to open before freeing more chunks if
773 		 * we have reached the threshold of frees
774 		 */
775 		if (dirty_frees_threshold != 0 &&
776 		    long_free_dirty_all_txgs >= dirty_frees_threshold) {
777 			txg_wait_open(dp, 0);
778 			continue;
779 		}
780 
781 		tx = dmu_tx_create(os);
782 		dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len);
783 
784 		/*
785 		 * Mark this transaction as typically resulting in a net
786 		 * reduction in space used.
787 		 */
788 		dmu_tx_mark_netfree(tx);
789 		err = dmu_tx_assign(tx, TXG_WAIT);
790 		if (err) {
791 			dmu_tx_abort(tx);
792 			return (err);
793 		}
794 
795 		mutex_enter(&dp->dp_lock);
796 		dp->dp_long_free_dirty_pertxg[dmu_tx_get_txg(tx) & TXG_MASK] +=
797 		    chunk_len;
798 		mutex_exit(&dp->dp_lock);
799 		DTRACE_PROBE3(free__long__range,
800 		    uint64_t, long_free_dirty_all_txgs, uint64_t, chunk_len,
801 		    uint64_t, dmu_tx_get_txg(tx));
802 		dnode_free_range(dn, chunk_begin, chunk_len, tx);
803 		dmu_tx_commit(tx);
804 
805 		length -= chunk_len;
806 	}
807 	return (0);
808 }
809 
810 int
dmu_free_long_range(objset_t * os,uint64_t object,uint64_t offset,uint64_t length)811 dmu_free_long_range(objset_t *os, uint64_t object,
812     uint64_t offset, uint64_t length)
813 {
814 	dnode_t *dn;
815 	int err;
816 
817 	err = dnode_hold(os, object, FTAG, &dn);
818 	if (err != 0)
819 		return (err);
820 	err = dmu_free_long_range_impl(os, dn, offset, length);
821 
822 	/*
823 	 * It is important to zero out the maxblkid when freeing the entire
824 	 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
825 	 * will take the fast path, and (b) dnode_reallocate() can verify
826 	 * that the entire file has been freed.
827 	 */
828 	if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
829 		dn->dn_maxblkid = 0;
830 
831 	dnode_rele(dn, FTAG);
832 	return (err);
833 }
834 
835 int
dmu_free_long_object(objset_t * os,uint64_t object)836 dmu_free_long_object(objset_t *os, uint64_t object)
837 {
838 	dmu_tx_t *tx;
839 	int err;
840 
841 	err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
842 	if (err != 0)
843 		return (err);
844 
845 	tx = dmu_tx_create(os);
846 	dmu_tx_hold_bonus(tx, object);
847 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
848 	dmu_tx_mark_netfree(tx);
849 	err = dmu_tx_assign(tx, TXG_WAIT);
850 	if (err == 0) {
851 		err = dmu_object_free(os, object, tx);
852 		dmu_tx_commit(tx);
853 	} else {
854 		dmu_tx_abort(tx);
855 	}
856 
857 	return (err);
858 }
859 
860 int
dmu_free_range(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,dmu_tx_t * tx)861 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
862     uint64_t size, dmu_tx_t *tx)
863 {
864 	dnode_t *dn;
865 	int err = dnode_hold(os, object, FTAG, &dn);
866 	if (err)
867 		return (err);
868 	ASSERT(offset < UINT64_MAX);
869 	ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
870 	dnode_free_range(dn, offset, size, tx);
871 	dnode_rele(dn, FTAG);
872 	return (0);
873 }
874 
875 int
dmu_read(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,void * buf,uint32_t flags)876 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
877     void *buf, uint32_t flags)
878 {
879 	dnode_t *dn;
880 	dmu_buf_t **dbp;
881 	int numbufs, err;
882 
883 	err = dnode_hold(os, object, FTAG, &dn);
884 	if (err)
885 		return (err);
886 
887 	/*
888 	 * Deal with odd block sizes, where there can't be data past the first
889 	 * block.  If we ever do the tail block optimization, we will need to
890 	 * handle that here as well.
891 	 */
892 	if (dn->dn_maxblkid == 0) {
893 		int newsz = offset > dn->dn_datablksz ? 0 :
894 		    MIN(size, dn->dn_datablksz - offset);
895 		bzero((char *)buf + newsz, size - newsz);
896 		size = newsz;
897 	}
898 
899 	while (size > 0) {
900 		uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
901 		int i;
902 
903 		/*
904 		 * NB: we could do this block-at-a-time, but it's nice
905 		 * to be reading in parallel.
906 		 */
907 		err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
908 		    TRUE, FTAG, &numbufs, &dbp, flags);
909 		if (err)
910 			break;
911 
912 		for (i = 0; i < numbufs; i++) {
913 			int tocpy;
914 			int bufoff;
915 			dmu_buf_t *db = dbp[i];
916 
917 			ASSERT(size > 0);
918 
919 			bufoff = offset - db->db_offset;
920 			tocpy = (int)MIN(db->db_size - bufoff, size);
921 
922 			bcopy((char *)db->db_data + bufoff, buf, tocpy);
923 
924 			offset += tocpy;
925 			size -= tocpy;
926 			buf = (char *)buf + tocpy;
927 		}
928 		dmu_buf_rele_array(dbp, numbufs, FTAG);
929 	}
930 	dnode_rele(dn, FTAG);
931 	return (err);
932 }
933 
934 void
dmu_write(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,const void * buf,dmu_tx_t * tx)935 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
936     const void *buf, dmu_tx_t *tx)
937 {
938 	dmu_buf_t **dbp;
939 	int numbufs, i;
940 
941 	if (size == 0)
942 		return;
943 
944 	VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
945 	    FALSE, FTAG, &numbufs, &dbp));
946 
947 	for (i = 0; i < numbufs; i++) {
948 		int tocpy;
949 		int bufoff;
950 		dmu_buf_t *db = dbp[i];
951 
952 		ASSERT(size > 0);
953 
954 		bufoff = offset - db->db_offset;
955 		tocpy = (int)MIN(db->db_size - bufoff, size);
956 
957 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
958 
959 		if (tocpy == db->db_size)
960 			dmu_buf_will_fill(db, tx);
961 		else
962 			dmu_buf_will_dirty(db, tx);
963 
964 		bcopy(buf, (char *)db->db_data + bufoff, tocpy);
965 
966 		if (tocpy == db->db_size)
967 			dmu_buf_fill_done(db, tx);
968 
969 		offset += tocpy;
970 		size -= tocpy;
971 		buf = (char *)buf + tocpy;
972 	}
973 	dmu_buf_rele_array(dbp, numbufs, FTAG);
974 }
975 
976 void
dmu_prealloc(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,dmu_tx_t * tx)977 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
978     dmu_tx_t *tx)
979 {
980 	dmu_buf_t **dbp;
981 	int numbufs, i;
982 
983 	if (size == 0)
984 		return;
985 
986 	VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
987 	    FALSE, FTAG, &numbufs, &dbp));
988 
989 	for (i = 0; i < numbufs; i++) {
990 		dmu_buf_t *db = dbp[i];
991 
992 		dmu_buf_will_not_fill(db, tx);
993 	}
994 	dmu_buf_rele_array(dbp, numbufs, FTAG);
995 }
996 
997 void
dmu_write_embedded(objset_t * os,uint64_t object,uint64_t offset,void * data,uint8_t etype,uint8_t comp,int uncompressed_size,int compressed_size,int byteorder,dmu_tx_t * tx)998 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
999     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
1000     int compressed_size, int byteorder, dmu_tx_t *tx)
1001 {
1002 	dmu_buf_t *db;
1003 
1004 	ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
1005 	ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
1006 	VERIFY0(dmu_buf_hold_noread(os, object, offset,
1007 	    FTAG, &db));
1008 
1009 	dmu_buf_write_embedded(db,
1010 	    data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
1011 	    uncompressed_size, compressed_size, byteorder, tx);
1012 
1013 	dmu_buf_rele(db, FTAG);
1014 }
1015 
1016 /*
1017  * DMU support for xuio
1018  */
1019 kstat_t *xuio_ksp = NULL;
1020 
1021 int
dmu_xuio_init(xuio_t * xuio,int nblk)1022 dmu_xuio_init(xuio_t *xuio, int nblk)
1023 {
1024 	dmu_xuio_t *priv;
1025 	uio_t *uio = &xuio->xu_uio;
1026 
1027 	uio->uio_iovcnt = nblk;
1028 	uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
1029 
1030 	priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
1031 	priv->cnt = nblk;
1032 	priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
1033 	priv->iovp = uio->uio_iov;
1034 	XUIO_XUZC_PRIV(xuio) = priv;
1035 
1036 	if (XUIO_XUZC_RW(xuio) == UIO_READ)
1037 		XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
1038 	else
1039 		XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
1040 
1041 	return (0);
1042 }
1043 
1044 void
dmu_xuio_fini(xuio_t * xuio)1045 dmu_xuio_fini(xuio_t *xuio)
1046 {
1047 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1048 	int nblk = priv->cnt;
1049 
1050 	kmem_free(priv->iovp, nblk * sizeof (iovec_t));
1051 	kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
1052 	kmem_free(priv, sizeof (dmu_xuio_t));
1053 
1054 	if (XUIO_XUZC_RW(xuio) == UIO_READ)
1055 		XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
1056 	else
1057 		XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
1058 }
1059 
1060 /*
1061  * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
1062  * and increase priv->next by 1.
1063  */
1064 int
dmu_xuio_add(xuio_t * xuio,arc_buf_t * abuf,offset_t off,size_t n)1065 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
1066 {
1067 	struct iovec *iov;
1068 	uio_t *uio = &xuio->xu_uio;
1069 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1070 	int i = priv->next++;
1071 
1072 	ASSERT(i < priv->cnt);
1073 	ASSERT(off + n <= arc_buf_size(abuf));
1074 	iov = uio->uio_iov + i;
1075 	iov->iov_base = (char *)abuf->b_data + off;
1076 	iov->iov_len = n;
1077 	priv->bufs[i] = abuf;
1078 	return (0);
1079 }
1080 
1081 int
dmu_xuio_cnt(xuio_t * xuio)1082 dmu_xuio_cnt(xuio_t *xuio)
1083 {
1084 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1085 	return (priv->cnt);
1086 }
1087 
1088 arc_buf_t *
dmu_xuio_arcbuf(xuio_t * xuio,int i)1089 dmu_xuio_arcbuf(xuio_t *xuio, int i)
1090 {
1091 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1092 
1093 	ASSERT(i < priv->cnt);
1094 	return (priv->bufs[i]);
1095 }
1096 
1097 void
dmu_xuio_clear(xuio_t * xuio,int i)1098 dmu_xuio_clear(xuio_t *xuio, int i)
1099 {
1100 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1101 
1102 	ASSERT(i < priv->cnt);
1103 	priv->bufs[i] = NULL;
1104 }
1105 
1106 static void
xuio_stat_init(void)1107 xuio_stat_init(void)
1108 {
1109 	xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
1110 	    KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
1111 	    KSTAT_FLAG_VIRTUAL);
1112 	if (xuio_ksp != NULL) {
1113 		xuio_ksp->ks_data = &xuio_stats;
1114 		kstat_install(xuio_ksp);
1115 	}
1116 }
1117 
1118 static void
xuio_stat_fini(void)1119 xuio_stat_fini(void)
1120 {
1121 	if (xuio_ksp != NULL) {
1122 		kstat_delete(xuio_ksp);
1123 		xuio_ksp = NULL;
1124 	}
1125 }
1126 
1127 void
xuio_stat_wbuf_copied()1128 xuio_stat_wbuf_copied()
1129 {
1130 	XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1131 }
1132 
1133 void
xuio_stat_wbuf_nocopy()1134 xuio_stat_wbuf_nocopy()
1135 {
1136 	XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1137 }
1138 
1139 #ifdef _KERNEL
1140 static int
dmu_read_uio_dnode(dnode_t * dn,uio_t * uio,uint64_t size)1141 dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
1142 {
1143 	dmu_buf_t **dbp;
1144 	int numbufs, i, err;
1145 	xuio_t *xuio = NULL;
1146 
1147 	/*
1148 	 * NB: we could do this block-at-a-time, but it's nice
1149 	 * to be reading in parallel.
1150 	 */
1151 	err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1152 	    TRUE, FTAG, &numbufs, &dbp, 0);
1153 	if (err)
1154 		return (err);
1155 
1156 #ifdef UIO_XUIO
1157 	if (uio->uio_extflg == UIO_XUIO)
1158 		xuio = (xuio_t *)uio;
1159 #endif
1160 
1161 	for (i = 0; i < numbufs; i++) {
1162 		int tocpy;
1163 		int bufoff;
1164 		dmu_buf_t *db = dbp[i];
1165 
1166 		ASSERT(size > 0);
1167 
1168 		bufoff = uio->uio_loffset - db->db_offset;
1169 		tocpy = (int)MIN(db->db_size - bufoff, size);
1170 
1171 		if (xuio) {
1172 			dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1173 			arc_buf_t *dbuf_abuf = dbi->db_buf;
1174 			arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1175 			err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1176 			if (!err) {
1177 				uio->uio_resid -= tocpy;
1178 				uio->uio_loffset += tocpy;
1179 			}
1180 
1181 			if (abuf == dbuf_abuf)
1182 				XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1183 			else
1184 				XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1185 		} else {
1186 #ifdef illumos
1187 			err = uiomove((char *)db->db_data + bufoff, tocpy,
1188 			    UIO_READ, uio);
1189 #endif
1190 #ifdef __FreeBSD__
1191 			err = vn_io_fault_uiomove((char *)db->db_data + bufoff,
1192 			    tocpy, uio);
1193 #endif
1194 #ifdef __NetBSD__
1195 			err = uiomove((char *)db->db_data + bufoff, tocpy,
1196 			    UIO_READ, uio);
1197 #endif
1198 		}
1199 		if (err)
1200 			break;
1201 
1202 		size -= tocpy;
1203 	}
1204 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1205 
1206 	return (err);
1207 }
1208 
1209 /*
1210  * Read 'size' bytes into the uio buffer.
1211  * From object zdb->db_object.
1212  * Starting at offset uio->uio_loffset.
1213  *
1214  * If the caller already has a dbuf in the target object
1215  * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1216  * because we don't have to find the dnode_t for the object.
1217  */
1218 int
dmu_read_uio_dbuf(dmu_buf_t * zdb,uio_t * uio,uint64_t size)1219 dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size)
1220 {
1221 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1222 	dnode_t *dn;
1223 	int err;
1224 
1225 	if (size == 0)
1226 		return (0);
1227 
1228 	DB_DNODE_ENTER(db);
1229 	dn = DB_DNODE(db);
1230 	err = dmu_read_uio_dnode(dn, uio, size);
1231 	DB_DNODE_EXIT(db);
1232 
1233 	return (err);
1234 }
1235 
1236 /*
1237  * Read 'size' bytes into the uio buffer.
1238  * From the specified object
1239  * Starting at offset uio->uio_loffset.
1240  */
1241 int
dmu_read_uio(objset_t * os,uint64_t object,uio_t * uio,uint64_t size)1242 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1243 {
1244 	dnode_t *dn;
1245 	int err;
1246 
1247 	if (size == 0)
1248 		return (0);
1249 
1250 	err = dnode_hold(os, object, FTAG, &dn);
1251 	if (err)
1252 		return (err);
1253 
1254 	err = dmu_read_uio_dnode(dn, uio, size);
1255 
1256 	dnode_rele(dn, FTAG);
1257 
1258 	return (err);
1259 }
1260 
1261 static int
dmu_write_uio_dnode(dnode_t * dn,uio_t * uio,uint64_t size,dmu_tx_t * tx)1262 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1263 {
1264 	dmu_buf_t **dbp;
1265 	int numbufs;
1266 	int err = 0;
1267 	int i;
1268 
1269 	err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1270 	    FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1271 	if (err)
1272 		return (err);
1273 
1274 	for (i = 0; i < numbufs; i++) {
1275 		int tocpy;
1276 		int bufoff;
1277 		dmu_buf_t *db = dbp[i];
1278 
1279 		ASSERT(size > 0);
1280 
1281 		bufoff = uio->uio_loffset - db->db_offset;
1282 		tocpy = (int)MIN(db->db_size - bufoff, size);
1283 
1284 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1285 
1286 		if (tocpy == db->db_size)
1287 			dmu_buf_will_fill(db, tx);
1288 		else
1289 			dmu_buf_will_dirty(db, tx);
1290 
1291 #ifdef illumos
1292 		/*
1293 		 * XXX uiomove could block forever (eg. nfs-backed
1294 		 * pages).  There needs to be a uiolockdown() function
1295 		 * to lock the pages in memory, so that uiomove won't
1296 		 * block.
1297 		 */
1298 		err = uiomove((char *)db->db_data + bufoff, tocpy,
1299 		    UIO_WRITE, uio);
1300 #endif
1301 #ifdef __FreeBSD__
1302 		err = vn_io_fault_uiomove((char *)db->db_data + bufoff, tocpy,
1303 		    uio);
1304 #endif
1305 #ifdef __NetBSD__
1306 		err = uiomove((char *)db->db_data + bufoff, tocpy,
1307 		    UIO_WRITE, uio);
1308 #endif
1309 
1310 		if (tocpy == db->db_size)
1311 			dmu_buf_fill_done(db, tx);
1312 
1313 		if (err)
1314 			break;
1315 
1316 		size -= tocpy;
1317 	}
1318 
1319 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1320 	return (err);
1321 }
1322 
1323 /*
1324  * Write 'size' bytes from the uio buffer.
1325  * To object zdb->db_object.
1326  * Starting at offset uio->uio_loffset.
1327  *
1328  * If the caller already has a dbuf in the target object
1329  * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1330  * because we don't have to find the dnode_t for the object.
1331  */
1332 int
dmu_write_uio_dbuf(dmu_buf_t * zdb,uio_t * uio,uint64_t size,dmu_tx_t * tx)1333 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1334     dmu_tx_t *tx)
1335 {
1336 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1337 	dnode_t *dn;
1338 	int err;
1339 
1340 	if (size == 0)
1341 		return (0);
1342 
1343 	DB_DNODE_ENTER(db);
1344 	dn = DB_DNODE(db);
1345 	err = dmu_write_uio_dnode(dn, uio, size, tx);
1346 	DB_DNODE_EXIT(db);
1347 
1348 	return (err);
1349 }
1350 
1351 /*
1352  * Write 'size' bytes from the uio buffer.
1353  * To the specified object.
1354  * Starting at offset uio->uio_loffset.
1355  */
1356 int
dmu_write_uio(objset_t * os,uint64_t object,uio_t * uio,uint64_t size,dmu_tx_t * tx)1357 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1358     dmu_tx_t *tx)
1359 {
1360 	dnode_t *dn;
1361 	int err;
1362 
1363 	if (size == 0)
1364 		return (0);
1365 
1366 	err = dnode_hold(os, object, FTAG, &dn);
1367 	if (err)
1368 		return (err);
1369 
1370 	err = dmu_write_uio_dnode(dn, uio, size, tx);
1371 
1372 	dnode_rele(dn, FTAG);
1373 
1374 	return (err);
1375 }
1376 
1377 #ifdef illumos
1378 int
dmu_write_pages(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,page_t * pp,dmu_tx_t * tx)1379 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1380     page_t *pp, dmu_tx_t *tx)
1381 {
1382 	dmu_buf_t **dbp;
1383 	int numbufs, i;
1384 	int err;
1385 
1386 	if (size == 0)
1387 		return (0);
1388 
1389 	err = dmu_buf_hold_array(os, object, offset, size,
1390 	    FALSE, FTAG, &numbufs, &dbp);
1391 	if (err)
1392 		return (err);
1393 
1394 	for (i = 0; i < numbufs; i++) {
1395 		int tocpy, copied, thiscpy;
1396 		int bufoff;
1397 		dmu_buf_t *db = dbp[i];
1398 		caddr_t va;
1399 
1400 		ASSERT(size > 0);
1401 		ASSERT3U(db->db_size, >=, PAGESIZE);
1402 
1403 		bufoff = offset - db->db_offset;
1404 		tocpy = (int)MIN(db->db_size - bufoff, size);
1405 
1406 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1407 
1408 		if (tocpy == db->db_size)
1409 			dmu_buf_will_fill(db, tx);
1410 		else
1411 			dmu_buf_will_dirty(db, tx);
1412 
1413 		for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1414 			ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
1415 			thiscpy = MIN(PAGESIZE, tocpy - copied);
1416 			va = zfs_map_page(pp, S_READ);
1417 			bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1418 			zfs_unmap_page(pp, va);
1419 			pp = pp->p_next;
1420 			bufoff += PAGESIZE;
1421 		}
1422 
1423 		if (tocpy == db->db_size)
1424 			dmu_buf_fill_done(db, tx);
1425 
1426 		offset += tocpy;
1427 		size -= tocpy;
1428 	}
1429 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1430 	return (err);
1431 }
1432 #endif /* illumos */
1433 
1434 #ifdef __FreeBSD__
1435 int
dmu_write_pages(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,vm_page_t * ma,dmu_tx_t * tx)1436 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1437     vm_page_t *ma, dmu_tx_t *tx)
1438 {
1439 	dmu_buf_t **dbp;
1440 	struct sf_buf *sf;
1441 	int numbufs, i;
1442 	int err;
1443 
1444 	if (size == 0)
1445 		return (0);
1446 
1447 	err = dmu_buf_hold_array(os, object, offset, size,
1448 	    FALSE, FTAG, &numbufs, &dbp);
1449 	if (err)
1450 		return (err);
1451 
1452 	for (i = 0; i < numbufs; i++) {
1453 		int tocpy, copied, thiscpy;
1454 		int bufoff;
1455 		dmu_buf_t *db = dbp[i];
1456 		caddr_t va;
1457 
1458 		ASSERT(size > 0);
1459 		ASSERT3U(db->db_size, >=, PAGESIZE);
1460 
1461 		bufoff = offset - db->db_offset;
1462 		tocpy = (int)MIN(db->db_size - bufoff, size);
1463 
1464 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1465 
1466 		if (tocpy == db->db_size)
1467 			dmu_buf_will_fill(db, tx);
1468 		else
1469 			dmu_buf_will_dirty(db, tx);
1470 
1471 		for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1472 			ASSERT3U(ptoa((*ma)->pindex), ==, db->db_offset + bufoff);
1473 			thiscpy = MIN(PAGESIZE, tocpy - copied);
1474 			va = zfs_map_page(*ma, &sf);
1475 			bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1476 			zfs_unmap_page(sf);
1477 			ma += 1;
1478 			bufoff += PAGESIZE;
1479 		}
1480 
1481 		if (tocpy == db->db_size)
1482 			dmu_buf_fill_done(db, tx);
1483 
1484 		offset += tocpy;
1485 		size -= tocpy;
1486 	}
1487 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1488 	return (err);
1489 }
1490 #endif	/* __FreeBSD__ */
1491 
1492 #ifdef __NetBSD__
1493 int
dmu_write_pages(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,struct vm_page ** pgs,dmu_tx_t * tx)1494 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1495     struct vm_page **pgs, dmu_tx_t *tx)
1496 {
1497 	dmu_buf_t **dbp;
1498 	int numbufs, i;
1499 	int err;
1500 
1501 	if (size == 0)
1502 		return (0);
1503 
1504 	err = dmu_buf_hold_array(os, object, offset, size,
1505 	    FALSE, FTAG, &numbufs, &dbp);
1506 	if (err)
1507 		return (err);
1508 
1509 	for (i = 0; i < numbufs; i++) {
1510 		int tocpy, copied, thiscpy;
1511 		int bufoff;
1512 		dmu_buf_t *db = dbp[i];
1513 		caddr_t va;
1514 
1515 		ASSERT(size > 0);
1516 		ASSERT3U(db->db_size, >=, PAGESIZE);
1517 
1518 		bufoff = offset - db->db_offset;
1519 		tocpy = (int)MIN(db->db_size - bufoff, size);
1520 
1521 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1522 
1523 		if (tocpy == db->db_size)
1524 			dmu_buf_will_fill(db, tx);
1525 		else
1526 			dmu_buf_will_dirty(db, tx);
1527 
1528 		for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1529 			ASSERT3U((*pgs)->offset, ==, db->db_offset + bufoff);
1530 			thiscpy = MIN(PAGESIZE, tocpy - copied);
1531 			va = zfs_map_page(*pgs, S_READ);
1532 			bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1533 			zfs_unmap_page(*pgs, va);
1534 			pgs++;
1535 			bufoff += PAGESIZE;
1536 		}
1537 
1538 		if (tocpy == db->db_size)
1539 			dmu_buf_fill_done(db, tx);
1540 
1541 		offset += tocpy;
1542 		size -= tocpy;
1543 	}
1544 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1545 	return (err);
1546 }
1547 #endif
1548 #endif	/* _KERNEL */
1549 
1550 /*
1551  * Allocate a loaned anonymous arc buffer.
1552  */
1553 arc_buf_t *
dmu_request_arcbuf(dmu_buf_t * handle,int size)1554 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1555 {
1556 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1557 
1558 	return (arc_loan_buf(db->db_objset->os_spa, size));
1559 }
1560 
1561 /*
1562  * Free a loaned arc buffer.
1563  */
1564 void
dmu_return_arcbuf(arc_buf_t * buf)1565 dmu_return_arcbuf(arc_buf_t *buf)
1566 {
1567 	arc_return_buf(buf, FTAG);
1568 	arc_buf_destroy(buf, FTAG);
1569 }
1570 
1571 /*
1572  * When possible directly assign passed loaned arc buffer to a dbuf.
1573  * If this is not possible copy the contents of passed arc buf via
1574  * dmu_write().
1575  */
1576 void
dmu_assign_arcbuf(dmu_buf_t * handle,uint64_t offset,arc_buf_t * buf,dmu_tx_t * tx)1577 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1578     dmu_tx_t *tx)
1579 {
1580 	dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1581 	dnode_t *dn;
1582 	dmu_buf_impl_t *db;
1583 	uint32_t blksz = (uint32_t)arc_buf_size(buf);
1584 	uint64_t blkid;
1585 
1586 	DB_DNODE_ENTER(dbuf);
1587 	dn = DB_DNODE(dbuf);
1588 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
1589 	blkid = dbuf_whichblock(dn, 0, offset);
1590 	VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1591 	rw_exit(&dn->dn_struct_rwlock);
1592 	DB_DNODE_EXIT(dbuf);
1593 
1594 	/*
1595 	 * We can only assign if the offset is aligned, the arc buf is the
1596 	 * same size as the dbuf, and the dbuf is not metadata.  It
1597 	 * can't be metadata because the loaned arc buf comes from the
1598 	 * user-data kmem arena.
1599 	 */
1600 	if (offset == db->db.db_offset && blksz == db->db.db_size &&
1601 	    DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) {
1602 #ifdef _KERNEL
1603 		curthread->td_ru.ru_oublock++;
1604 #ifdef RACCT
1605 		if (racct_enable) {
1606 			PROC_LOCK(curproc);
1607 			racct_add_force(curproc, RACCT_WRITEBPS, blksz);
1608 			racct_add_force(curproc, RACCT_WRITEIOPS, 1);
1609 			PROC_UNLOCK(curproc);
1610 		}
1611 #endif /* RACCT */
1612 #endif /* _KERNEL */
1613 		dbuf_assign_arcbuf(db, buf, tx);
1614 		dbuf_rele(db, FTAG);
1615 	} else {
1616 		objset_t *os;
1617 		uint64_t object;
1618 
1619 		DB_DNODE_ENTER(dbuf);
1620 		dn = DB_DNODE(dbuf);
1621 		os = dn->dn_objset;
1622 		object = dn->dn_object;
1623 		DB_DNODE_EXIT(dbuf);
1624 
1625 		dbuf_rele(db, FTAG);
1626 		dmu_write(os, object, offset, blksz, buf->b_data, tx);
1627 		dmu_return_arcbuf(buf);
1628 		XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1629 	}
1630 }
1631 
1632 typedef struct {
1633 	dbuf_dirty_record_t	*dsa_dr;
1634 	dmu_sync_cb_t		*dsa_done;
1635 	zgd_t			*dsa_zgd;
1636 	dmu_tx_t		*dsa_tx;
1637 } dmu_sync_arg_t;
1638 
1639 /* ARGSUSED */
1640 static void
dmu_sync_ready(zio_t * zio,arc_buf_t * buf,void * varg)1641 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1642 {
1643 	dmu_sync_arg_t *dsa = varg;
1644 	dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1645 	blkptr_t *bp = zio->io_bp;
1646 
1647 	if (zio->io_error == 0) {
1648 		if (BP_IS_HOLE(bp)) {
1649 			/*
1650 			 * A block of zeros may compress to a hole, but the
1651 			 * block size still needs to be known for replay.
1652 			 */
1653 			BP_SET_LSIZE(bp, db->db_size);
1654 		} else if (!BP_IS_EMBEDDED(bp)) {
1655 			ASSERT(BP_GET_LEVEL(bp) == 0);
1656 			bp->blk_fill = 1;
1657 		}
1658 	}
1659 }
1660 
1661 static void
dmu_sync_late_arrival_ready(zio_t * zio)1662 dmu_sync_late_arrival_ready(zio_t *zio)
1663 {
1664 	dmu_sync_ready(zio, NULL, zio->io_private);
1665 }
1666 
1667 /* ARGSUSED */
1668 static void
dmu_sync_done(zio_t * zio,arc_buf_t * buf,void * varg)1669 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1670 {
1671 	dmu_sync_arg_t *dsa = varg;
1672 	dbuf_dirty_record_t *dr = dsa->dsa_dr;
1673 	dmu_buf_impl_t *db = dr->dr_dbuf;
1674 
1675 	mutex_enter(&db->db_mtx);
1676 	ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1677 	if (zio->io_error == 0) {
1678 		dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1679 		if (dr->dt.dl.dr_nopwrite) {
1680 			blkptr_t *bp = zio->io_bp;
1681 			blkptr_t *bp_orig = &zio->io_bp_orig;
1682 			uint8_t chksum = BP_GET_CHECKSUM(bp_orig);
1683 
1684 			ASSERT(BP_EQUAL(bp, bp_orig));
1685 			ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1686 			ASSERT(zio_checksum_table[chksum].ci_flags &
1687 			    ZCHECKSUM_FLAG_NOPWRITE);
1688 		}
1689 		dr->dt.dl.dr_overridden_by = *zio->io_bp;
1690 		dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1691 		dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1692 
1693 		/*
1694 		 * Old style holes are filled with all zeros, whereas
1695 		 * new-style holes maintain their lsize, type, level,
1696 		 * and birth time (see zio_write_compress). While we
1697 		 * need to reset the BP_SET_LSIZE() call that happened
1698 		 * in dmu_sync_ready for old style holes, we do *not*
1699 		 * want to wipe out the information contained in new
1700 		 * style holes. Thus, only zero out the block pointer if
1701 		 * it's an old style hole.
1702 		 */
1703 		if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) &&
1704 		    dr->dt.dl.dr_overridden_by.blk_birth == 0)
1705 			BP_ZERO(&dr->dt.dl.dr_overridden_by);
1706 	} else {
1707 		dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1708 	}
1709 	cv_broadcast(&db->db_changed);
1710 	mutex_exit(&db->db_mtx);
1711 
1712 	dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1713 
1714 	kmem_free(dsa, sizeof (*dsa));
1715 }
1716 
1717 static void
dmu_sync_late_arrival_done(zio_t * zio)1718 dmu_sync_late_arrival_done(zio_t *zio)
1719 {
1720 	blkptr_t *bp = zio->io_bp;
1721 	dmu_sync_arg_t *dsa = zio->io_private;
1722 	blkptr_t *bp_orig = &zio->io_bp_orig;
1723 
1724 	if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1725 		/*
1726 		 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1727 		 * then there is nothing to do here. Otherwise, free the
1728 		 * newly allocated block in this txg.
1729 		 */
1730 		if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
1731 			ASSERT(BP_EQUAL(bp, bp_orig));
1732 		} else {
1733 			ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1734 			ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1735 			ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1736 			zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1737 		}
1738 	}
1739 
1740 	dmu_tx_commit(dsa->dsa_tx);
1741 
1742 	dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1743 
1744 	kmem_free(dsa, sizeof (*dsa));
1745 }
1746 
1747 static int
dmu_sync_late_arrival(zio_t * pio,objset_t * os,dmu_sync_cb_t * done,zgd_t * zgd,zio_prop_t * zp,zbookmark_phys_t * zb)1748 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1749     zio_prop_t *zp, zbookmark_phys_t *zb)
1750 {
1751 	dmu_sync_arg_t *dsa;
1752 	dmu_tx_t *tx;
1753 
1754 	tx = dmu_tx_create(os);
1755 	dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1756 	if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1757 		dmu_tx_abort(tx);
1758 		/* Make zl_get_data do txg_waited_synced() */
1759 		return (SET_ERROR(EIO));
1760 	}
1761 
1762 	dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1763 	dsa->dsa_dr = NULL;
1764 	dsa->dsa_done = done;
1765 	dsa->dsa_zgd = zgd;
1766 	dsa->dsa_tx = tx;
1767 
1768 	zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx),
1769 	    zgd->zgd_bp, zgd->zgd_db->db_data, zgd->zgd_db->db_size,
1770 	    zp, dmu_sync_late_arrival_ready, NULL,
1771 	    NULL, dmu_sync_late_arrival_done, dsa, ZIO_PRIORITY_SYNC_WRITE,
1772 	    ZIO_FLAG_CANFAIL, zb));
1773 
1774 	return (0);
1775 }
1776 
1777 /*
1778  * Intent log support: sync the block associated with db to disk.
1779  * N.B. and XXX: the caller is responsible for making sure that the
1780  * data isn't changing while dmu_sync() is writing it.
1781  *
1782  * Return values:
1783  *
1784  *	EEXIST: this txg has already been synced, so there's nothing to do.
1785  *		The caller should not log the write.
1786  *
1787  *	ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1788  *		The caller should not log the write.
1789  *
1790  *	EALREADY: this block is already in the process of being synced.
1791  *		The caller should track its progress (somehow).
1792  *
1793  *	EIO: could not do the I/O.
1794  *		The caller should do a txg_wait_synced().
1795  *
1796  *	0: the I/O has been initiated.
1797  *		The caller should log this blkptr in the done callback.
1798  *		It is possible that the I/O will fail, in which case
1799  *		the error will be reported to the done callback and
1800  *		propagated to pio from zio_done().
1801  */
1802 int
dmu_sync(zio_t * pio,uint64_t txg,dmu_sync_cb_t * done,zgd_t * zgd)1803 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1804 {
1805 	blkptr_t *bp = zgd->zgd_bp;
1806 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1807 	objset_t *os = db->db_objset;
1808 	dsl_dataset_t *ds = os->os_dsl_dataset;
1809 	dbuf_dirty_record_t *dr;
1810 	dmu_sync_arg_t *dsa;
1811 	zbookmark_phys_t zb;
1812 	zio_prop_t zp;
1813 	dnode_t *dn;
1814 
1815 	ASSERT(pio != NULL);
1816 	ASSERT(txg != 0);
1817 
1818 	SET_BOOKMARK(&zb, ds->ds_object,
1819 	    db->db.db_object, db->db_level, db->db_blkid);
1820 
1821 	DB_DNODE_ENTER(db);
1822 	dn = DB_DNODE(db);
1823 	dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1824 	DB_DNODE_EXIT(db);
1825 
1826 	/*
1827 	 * If we're frozen (running ziltest), we always need to generate a bp.
1828 	 */
1829 	if (txg > spa_freeze_txg(os->os_spa))
1830 		return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1831 
1832 	/*
1833 	 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1834 	 * and us.  If we determine that this txg is not yet syncing,
1835 	 * but it begins to sync a moment later, that's OK because the
1836 	 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1837 	 */
1838 	mutex_enter(&db->db_mtx);
1839 
1840 	if (txg <= spa_last_synced_txg(os->os_spa)) {
1841 		/*
1842 		 * This txg has already synced.  There's nothing to do.
1843 		 */
1844 		mutex_exit(&db->db_mtx);
1845 		return (SET_ERROR(EEXIST));
1846 	}
1847 
1848 	if (txg <= spa_syncing_txg(os->os_spa)) {
1849 		/*
1850 		 * This txg is currently syncing, so we can't mess with
1851 		 * the dirty record anymore; just write a new log block.
1852 		 */
1853 		mutex_exit(&db->db_mtx);
1854 		return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1855 	}
1856 
1857 	dr = db->db_last_dirty;
1858 	while (dr && dr->dr_txg != txg)
1859 		dr = dr->dr_next;
1860 
1861 	if (dr == NULL) {
1862 		/*
1863 		 * There's no dr for this dbuf, so it must have been freed.
1864 		 * There's no need to log writes to freed blocks, so we're done.
1865 		 */
1866 		mutex_exit(&db->db_mtx);
1867 		return (SET_ERROR(ENOENT));
1868 	}
1869 
1870 	ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
1871 
1872 	/*
1873 	 * Assume the on-disk data is X, the current syncing data (in
1874 	 * txg - 1) is Y, and the current in-memory data is Z (currently
1875 	 * in dmu_sync).
1876 	 *
1877 	 * We usually want to perform a nopwrite if X and Z are the
1878 	 * same.  However, if Y is different (i.e. the BP is going to
1879 	 * change before this write takes effect), then a nopwrite will
1880 	 * be incorrect - we would override with X, which could have
1881 	 * been freed when Y was written.
1882 	 *
1883 	 * (Note that this is not a concern when we are nop-writing from
1884 	 * syncing context, because X and Y must be identical, because
1885 	 * all previous txgs have been synced.)
1886 	 *
1887 	 * Therefore, we disable nopwrite if the current BP could change
1888 	 * before this TXG.  There are two ways it could change: by
1889 	 * being dirty (dr_next is non-NULL), or by being freed
1890 	 * (dnode_block_freed()).  This behavior is verified by
1891 	 * zio_done(), which VERIFYs that the override BP is identical
1892 	 * to the on-disk BP.
1893 	 */
1894 	DB_DNODE_ENTER(db);
1895 	dn = DB_DNODE(db);
1896 	if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid))
1897 		zp.zp_nopwrite = B_FALSE;
1898 	DB_DNODE_EXIT(db);
1899 
1900 	ASSERT(dr->dr_txg == txg);
1901 	if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1902 	    dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1903 		/*
1904 		 * We have already issued a sync write for this buffer,
1905 		 * or this buffer has already been synced.  It could not
1906 		 * have been dirtied since, or we would have cleared the state.
1907 		 */
1908 		mutex_exit(&db->db_mtx);
1909 		return (SET_ERROR(EALREADY));
1910 	}
1911 
1912 	ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1913 	dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1914 	mutex_exit(&db->db_mtx);
1915 
1916 	dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1917 	dsa->dsa_dr = dr;
1918 	dsa->dsa_done = done;
1919 	dsa->dsa_zgd = zgd;
1920 	dsa->dsa_tx = NULL;
1921 
1922 	zio_nowait(arc_write(pio, os->os_spa, txg,
1923 	    bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
1924 	    &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa,
1925 	    ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
1926 
1927 	return (0);
1928 }
1929 
1930 int
dmu_object_set_blocksize(objset_t * os,uint64_t object,uint64_t size,int ibs,dmu_tx_t * tx)1931 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1932     dmu_tx_t *tx)
1933 {
1934 	dnode_t *dn;
1935 	int err;
1936 
1937 	err = dnode_hold(os, object, FTAG, &dn);
1938 	if (err)
1939 		return (err);
1940 	err = dnode_set_blksz(dn, size, ibs, tx);
1941 	dnode_rele(dn, FTAG);
1942 	return (err);
1943 }
1944 
1945 void
dmu_object_set_checksum(objset_t * os,uint64_t object,uint8_t checksum,dmu_tx_t * tx)1946 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1947     dmu_tx_t *tx)
1948 {
1949 	dnode_t *dn;
1950 
1951 	/*
1952 	 * Send streams include each object's checksum function.  This
1953 	 * check ensures that the receiving system can understand the
1954 	 * checksum function transmitted.
1955 	 */
1956 	ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
1957 
1958 	VERIFY0(dnode_hold(os, object, FTAG, &dn));
1959 	ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
1960 	dn->dn_checksum = checksum;
1961 	dnode_setdirty(dn, tx);
1962 	dnode_rele(dn, FTAG);
1963 }
1964 
1965 void
dmu_object_set_compress(objset_t * os,uint64_t object,uint8_t compress,dmu_tx_t * tx)1966 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1967     dmu_tx_t *tx)
1968 {
1969 	dnode_t *dn;
1970 
1971 	/*
1972 	 * Send streams include each object's compression function.  This
1973 	 * check ensures that the receiving system can understand the
1974 	 * compression function transmitted.
1975 	 */
1976 	ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
1977 
1978 	VERIFY0(dnode_hold(os, object, FTAG, &dn));
1979 	dn->dn_compress = compress;
1980 	dnode_setdirty(dn, tx);
1981 	dnode_rele(dn, FTAG);
1982 }
1983 
1984 int zfs_mdcomp_disable = 0;
1985 SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RWTUN,
1986     &zfs_mdcomp_disable, 0, "Disable metadata compression");
1987 
1988 /*
1989  * When the "redundant_metadata" property is set to "most", only indirect
1990  * blocks of this level and higher will have an additional ditto block.
1991  */
1992 int zfs_redundant_metadata_most_ditto_level = 2;
1993 
1994 void
dmu_write_policy(objset_t * os,dnode_t * dn,int level,int wp,zio_prop_t * zp)1995 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1996 {
1997 	dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1998 	boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1999 	    (wp & WP_SPILL));
2000 	enum zio_checksum checksum = os->os_checksum;
2001 	enum zio_compress compress = os->os_compress;
2002 	enum zio_checksum dedup_checksum = os->os_dedup_checksum;
2003 	boolean_t dedup = B_FALSE;
2004 	boolean_t nopwrite = B_FALSE;
2005 	boolean_t dedup_verify = os->os_dedup_verify;
2006 	int copies = os->os_copies;
2007 
2008 	/*
2009 	 * We maintain different write policies for each of the following
2010 	 * types of data:
2011 	 *	 1. metadata
2012 	 *	 2. preallocated blocks (i.e. level-0 blocks of a dump device)
2013 	 *	 3. all other level 0 blocks
2014 	 */
2015 	if (ismd) {
2016 		if (zfs_mdcomp_disable) {
2017 			compress = ZIO_COMPRESS_EMPTY;
2018 		} else {
2019 			/*
2020 			 * XXX -- we should design a compression algorithm
2021 			 * that specializes in arrays of bps.
2022 			 */
2023 			compress = zio_compress_select(os->os_spa,
2024 			    ZIO_COMPRESS_ON, ZIO_COMPRESS_ON);
2025 		}
2026 
2027 		/*
2028 		 * Metadata always gets checksummed.  If the data
2029 		 * checksum is multi-bit correctable, and it's not a
2030 		 * ZBT-style checksum, then it's suitable for metadata
2031 		 * as well.  Otherwise, the metadata checksum defaults
2032 		 * to fletcher4.
2033 		 */
2034 		if (!(zio_checksum_table[checksum].ci_flags &
2035 		    ZCHECKSUM_FLAG_METADATA) ||
2036 		    (zio_checksum_table[checksum].ci_flags &
2037 		    ZCHECKSUM_FLAG_EMBEDDED))
2038 			checksum = ZIO_CHECKSUM_FLETCHER_4;
2039 
2040 		if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
2041 		    (os->os_redundant_metadata ==
2042 		    ZFS_REDUNDANT_METADATA_MOST &&
2043 		    (level >= zfs_redundant_metadata_most_ditto_level ||
2044 		    DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))))
2045 			copies++;
2046 	} else if (wp & WP_NOFILL) {
2047 		ASSERT(level == 0);
2048 
2049 		/*
2050 		 * If we're writing preallocated blocks, we aren't actually
2051 		 * writing them so don't set any policy properties.  These
2052 		 * blocks are currently only used by an external subsystem
2053 		 * outside of zfs (i.e. dump) and not written by the zio
2054 		 * pipeline.
2055 		 */
2056 		compress = ZIO_COMPRESS_OFF;
2057 		checksum = ZIO_CHECKSUM_NOPARITY;
2058 	} else {
2059 		compress = zio_compress_select(os->os_spa, dn->dn_compress,
2060 		    compress);
2061 
2062 		checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
2063 		    zio_checksum_select(dn->dn_checksum, checksum) :
2064 		    dedup_checksum;
2065 
2066 		/*
2067 		 * Determine dedup setting.  If we are in dmu_sync(),
2068 		 * we won't actually dedup now because that's all
2069 		 * done in syncing context; but we do want to use the
2070 		 * dedup checkum.  If the checksum is not strong
2071 		 * enough to ensure unique signatures, force
2072 		 * dedup_verify.
2073 		 */
2074 		if (dedup_checksum != ZIO_CHECKSUM_OFF) {
2075 			dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
2076 			if (!(zio_checksum_table[checksum].ci_flags &
2077 			    ZCHECKSUM_FLAG_DEDUP))
2078 				dedup_verify = B_TRUE;
2079 		}
2080 
2081 		/*
2082 		 * Enable nopwrite if we have secure enough checksum
2083 		 * algorithm (see comment in zio_nop_write) and
2084 		 * compression is enabled.  We don't enable nopwrite if
2085 		 * dedup is enabled as the two features are mutually
2086 		 * exclusive.
2087 		 */
2088 		nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags &
2089 		    ZCHECKSUM_FLAG_NOPWRITE) &&
2090 		    compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
2091 	}
2092 
2093 	zp->zp_checksum = checksum;
2094 	zp->zp_compress = compress;
2095 	zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
2096 	zp->zp_level = level;
2097 	zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
2098 	zp->zp_dedup = dedup;
2099 	zp->zp_dedup_verify = dedup && dedup_verify;
2100 	zp->zp_nopwrite = nopwrite;
2101 }
2102 
2103 int
dmu_offset_next(objset_t * os,uint64_t object,boolean_t hole,uint64_t * off)2104 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
2105 {
2106 	dnode_t *dn;
2107 	int err;
2108 
2109 	/*
2110 	 * Sync any current changes before
2111 	 * we go trundling through the block pointers.
2112 	 */
2113 	err = dmu_object_wait_synced(os, object);
2114 	if (err) {
2115 		return (err);
2116 	}
2117 
2118 	err = dnode_hold(os, object, FTAG, &dn);
2119 	if (err) {
2120 		return (err);
2121 	}
2122 
2123 	err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
2124 	dnode_rele(dn, FTAG);
2125 
2126 	return (err);
2127 }
2128 
2129 /*
2130  * Given the ZFS object, if it contains any dirty nodes
2131  * this function flushes all dirty blocks to disk. This
2132  * ensures the DMU object info is updated. A more efficient
2133  * future version might just find the TXG with the maximum
2134  * ID and wait for that to be synced.
2135  */
2136 int
dmu_object_wait_synced(objset_t * os,uint64_t object)2137 dmu_object_wait_synced(objset_t *os, uint64_t object)
2138 {
2139 	dnode_t *dn;
2140 	int error, i;
2141 
2142 	error = dnode_hold(os, object, FTAG, &dn);
2143 	if (error) {
2144 		return (error);
2145 	}
2146 
2147 	for (i = 0; i < TXG_SIZE; i++) {
2148 		if (list_link_active(&dn->dn_dirty_link[i])) {
2149 			break;
2150 		}
2151 	}
2152 	dnode_rele(dn, FTAG);
2153 	if (i != TXG_SIZE) {
2154 		txg_wait_synced(dmu_objset_pool(os), 0);
2155 	}
2156 
2157 	return (0);
2158 }
2159 
2160 void
dmu_object_info_from_dnode(dnode_t * dn,dmu_object_info_t * doi)2161 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
2162 {
2163 	dnode_phys_t *dnp;
2164 
2165 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
2166 	mutex_enter(&dn->dn_mtx);
2167 
2168 	dnp = dn->dn_phys;
2169 
2170 	doi->doi_data_block_size = dn->dn_datablksz;
2171 	doi->doi_metadata_block_size = dn->dn_indblkshift ?
2172 	    1ULL << dn->dn_indblkshift : 0;
2173 	doi->doi_type = dn->dn_type;
2174 	doi->doi_bonus_type = dn->dn_bonustype;
2175 	doi->doi_bonus_size = dn->dn_bonuslen;
2176 	doi->doi_indirection = dn->dn_nlevels;
2177 	doi->doi_checksum = dn->dn_checksum;
2178 	doi->doi_compress = dn->dn_compress;
2179 	doi->doi_nblkptr = dn->dn_nblkptr;
2180 	doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
2181 	doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
2182 	doi->doi_fill_count = 0;
2183 	for (int i = 0; i < dnp->dn_nblkptr; i++)
2184 		doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
2185 
2186 	mutex_exit(&dn->dn_mtx);
2187 	rw_exit(&dn->dn_struct_rwlock);
2188 }
2189 
2190 /*
2191  * Get information on a DMU object.
2192  * If doi is NULL, just indicates whether the object exists.
2193  */
2194 int
dmu_object_info(objset_t * os,uint64_t object,dmu_object_info_t * doi)2195 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
2196 {
2197 	dnode_t *dn;
2198 	int err = dnode_hold(os, object, FTAG, &dn);
2199 
2200 	if (err)
2201 		return (err);
2202 
2203 	if (doi != NULL)
2204 		dmu_object_info_from_dnode(dn, doi);
2205 
2206 	dnode_rele(dn, FTAG);
2207 	return (0);
2208 }
2209 
2210 /*
2211  * As above, but faster; can be used when you have a held dbuf in hand.
2212  */
2213 void
dmu_object_info_from_db(dmu_buf_t * db_fake,dmu_object_info_t * doi)2214 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
2215 {
2216 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2217 
2218 	DB_DNODE_ENTER(db);
2219 	dmu_object_info_from_dnode(DB_DNODE(db), doi);
2220 	DB_DNODE_EXIT(db);
2221 }
2222 
2223 /*
2224  * Faster still when you only care about the size.
2225  * This is specifically optimized for zfs_getattr().
2226  */
2227 void
dmu_object_size_from_db(dmu_buf_t * db_fake,uint32_t * blksize,u_longlong_t * nblk512)2228 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
2229     u_longlong_t *nblk512)
2230 {
2231 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2232 	dnode_t *dn;
2233 
2234 	DB_DNODE_ENTER(db);
2235 	dn = DB_DNODE(db);
2236 
2237 	*blksize = dn->dn_datablksz;
2238 	/* add 1 for dnode space */
2239 	*nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
2240 	    SPA_MINBLOCKSHIFT) + 1;
2241 	DB_DNODE_EXIT(db);
2242 }
2243 
2244 void
byteswap_uint64_array(void * vbuf,size_t size)2245 byteswap_uint64_array(void *vbuf, size_t size)
2246 {
2247 	uint64_t *buf = vbuf;
2248 	size_t count = size >> 3;
2249 	int i;
2250 
2251 	ASSERT((size & 7) == 0);
2252 
2253 	for (i = 0; i < count; i++)
2254 		buf[i] = BSWAP_64(buf[i]);
2255 }
2256 
2257 void
byteswap_uint32_array(void * vbuf,size_t size)2258 byteswap_uint32_array(void *vbuf, size_t size)
2259 {
2260 	uint32_t *buf = vbuf;
2261 	size_t count = size >> 2;
2262 	int i;
2263 
2264 	ASSERT((size & 3) == 0);
2265 
2266 	for (i = 0; i < count; i++)
2267 		buf[i] = BSWAP_32(buf[i]);
2268 }
2269 
2270 void
byteswap_uint16_array(void * vbuf,size_t size)2271 byteswap_uint16_array(void *vbuf, size_t size)
2272 {
2273 	uint16_t *buf = vbuf;
2274 	size_t count = size >> 1;
2275 	int i;
2276 
2277 	ASSERT((size & 1) == 0);
2278 
2279 	for (i = 0; i < count; i++)
2280 		buf[i] = BSWAP_16(buf[i]);
2281 }
2282 
2283 /* ARGSUSED */
2284 void
byteswap_uint8_array(void * vbuf,size_t size)2285 byteswap_uint8_array(void *vbuf, size_t size)
2286 {
2287 }
2288 
2289 void
dmu_init(void)2290 dmu_init(void)
2291 {
2292 	zfs_dbgmsg_init();
2293 	sa_cache_init();
2294 	xuio_stat_init();
2295 	dmu_objset_init();
2296 	dnode_init();
2297 	zfetch_init();
2298 	zio_compress_init();
2299 	l2arc_init();
2300 	arc_init();
2301 	dbuf_init();
2302 }
2303 
2304 void
dmu_fini(void)2305 dmu_fini(void)
2306 {
2307 	arc_fini(); /* arc depends on l2arc, so arc must go first */
2308 	l2arc_fini();
2309 	zfetch_fini();
2310 	zio_compress_fini();
2311 	dbuf_fini();
2312 	dnode_fini();
2313 	dmu_objset_fini();
2314 	xuio_stat_fini();
2315 	sa_cache_fini();
2316 	zfs_dbgmsg_fini();
2317 }
2318