xref: /illumos-gate/usr/src/uts/common/fs/zfs/sa.c (revision 3665ce8aeee26b1a84fb98951ef011e0779e1ae2)
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 /*
23  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24  * Portions Copyright 2011 iXsystems, Inc
25  * Copyright (c) 2013, 2017 by Delphix. All rights reserved.
26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  * Copyright (c) 2014 Integros [integros.com]
28  * Copyright 2019 Joyent, Inc.
29  */
30 
31 #include <sys/zfs_context.h>
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/dmu.h>
37 #include <sys/dmu_impl.h>
38 #include <sys/dmu_objset.h>
39 #include <sys/dmu_tx.h>
40 #include <sys/dbuf.h>
41 #include <sys/dnode.h>
42 #include <sys/zap.h>
43 #include <sys/sa.h>
44 #include <sys/sunddi.h>
45 #include <sys/sa_impl.h>
46 #include <sys/dnode.h>
47 #include <sys/errno.h>
48 #include <sys/zfs_context.h>
49 
50 #ifdef _KERNEL
51 #include <sys/zfs_znode.h>
52 #endif
53 
54 /*
55  * ZFS System attributes:
56  *
57  * A generic mechanism to allow for arbitrary attributes
58  * to be stored in a dnode.  The data will be stored in the bonus buffer of
59  * the dnode and if necessary a special "spill" block will be used to handle
60  * overflow situations.  The spill block will be sized to fit the data
61  * from 512 - 128K.  When a spill block is used the BP (blkptr_t) for the
62  * spill block is stored at the end of the current bonus buffer.  Any
63  * attributes that would be in the way of the blkptr_t will be relocated
64  * into the spill block.
65  *
66  * Attribute registration:
67  *
68  * Stored persistently on a per dataset basis
69  * a mapping between attribute "string" names and their actual attribute
70  * numeric values, length, and byteswap function.  The names are only used
71  * during registration.  All  attributes are known by their unique attribute
72  * id value.  If an attribute can have a variable size then the value
73  * 0 will be used to indicate this.
74  *
75  * Attribute Layout:
76  *
77  * Attribute layouts are a way to compactly store multiple attributes, but
78  * without taking the overhead associated with managing each attribute
79  * individually.  Since you will typically have the same set of attributes
80  * stored in the same order a single table will be used to represent that
81  * layout.  The ZPL for example will usually have only about 10 different
82  * layouts (regular files, device files, symlinks,
83  * regular files + scanstamp, files/dir with extended attributes, and then
84  * you have the possibility of all of those minus ACL, because it would
85  * be kicked out into the spill block)
86  *
87  * Layouts are simply an array of the attributes and their
88  * ordering i.e. [0, 1, 4, 5, 2]
89  *
90  * Each distinct layout is given a unique layout number and that is whats
91  * stored in the header at the beginning of the SA data buffer.
92  *
93  * A layout only covers a single dbuf (bonus or spill).  If a set of
94  * attributes is split up between the bonus buffer and a spill buffer then
95  * two different layouts will be used.  This allows us to byteswap the
96  * spill without looking at the bonus buffer and keeps the on disk format of
97  * the bonus and spill buffer the same.
98  *
99  * Adding a single attribute will cause the entire set of attributes to
100  * be rewritten and could result in a new layout number being constructed
101  * as part of the rewrite if no such layout exists for the new set of
102  * attribues.  The new attribute will be appended to the end of the already
103  * existing attributes.
104  *
105  * Both the attribute registration and attribute layout information are
106  * stored in normal ZAP attributes.  Their should be a small number of
107  * known layouts and the set of attributes is assumed to typically be quite
108  * small.
109  *
110  * The registered attributes and layout "table" information is maintained
111  * in core and a special "sa_os_t" is attached to the objset_t.
112  *
113  * A special interface is provided to allow for quickly applying
114  * a large set of attributes at once.  sa_replace_all_by_template() is
115  * used to set an array of attributes.  This is used by the ZPL when
116  * creating a brand new file.  The template that is passed into the function
117  * specifies the attribute, size for variable length attributes, location of
118  * data and special "data locator" function if the data isn't in a contiguous
119  * location.
120  *
121  * Byteswap implications:
122  *
123  * Since the SA attributes are not entirely self describing we can't do
124  * the normal byteswap processing.  The special ZAP layout attribute and
125  * attribute registration attributes define the byteswap function and the
126  * size of the attributes, unless it is variable sized.
127  * The normal ZFS byteswapping infrastructure assumes you don't need
128  * to read any objects in order to do the necessary byteswapping.  Whereas
129  * SA attributes can only be properly byteswapped if the dataset is opened
130  * and the layout/attribute ZAP attributes are available.  Because of this
131  * the SA attributes will be byteswapped when they are first accessed by
132  * the SA code that will read the SA data.
133  */
134 
135 typedef void (sa_iterfunc_t)(void *hdr, void *addr, sa_attr_type_t,
136     uint16_t length, int length_idx, boolean_t, void *userp);
137 
138 static int sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype);
139 static void sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab);
140 static sa_idx_tab_t *sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype,
141     sa_hdr_phys_t *hdr);
142 static void sa_idx_tab_rele(objset_t *os, void *arg);
143 static void sa_copy_data(sa_data_locator_t *func, void *start, void *target,
144     int buflen);
145 static int sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
146     sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
147     uint16_t buflen, dmu_tx_t *tx);
148 
149 arc_byteswap_func_t *sa_bswap_table[] = {
150 	byteswap_uint64_array,
151 	byteswap_uint32_array,
152 	byteswap_uint16_array,
153 	byteswap_uint8_array,
154 	zfs_acl_byteswap,
155 };
156 
157 #define	SA_COPY_DATA(f, s, t, l) \
158 	{ \
159 		if (f == NULL) { \
160 			if (l == 8) { \
161 				*(uint64_t *)t = *(uint64_t *)s; \
162 			} else if (l == 16) { \
163 				*(uint64_t *)t = *(uint64_t *)s; \
164 				*(uint64_t *)((uintptr_t)t + 8) = \
165 				    *(uint64_t *)((uintptr_t)s + 8); \
166 			} else { \
167 				bcopy(s, t, l); \
168 			} \
169 		} else \
170 			sa_copy_data(f, s, t, l); \
171 	}
172 
173 /*
174  * This table is fixed and cannot be changed.  Its purpose is to
175  * allow the SA code to work with both old/new ZPL file systems.
176  * It contains the list of legacy attributes.  These attributes aren't
177  * stored in the "attribute" registry zap objects, since older ZPL file systems
178  * won't have the registry.  Only objsets of type ZFS_TYPE_FILESYSTEM will
179  * use this static table.
180  */
181 sa_attr_reg_t sa_legacy_attrs[] = {
182 	{"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
183 	{"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
184 	{"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
185 	{"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
186 	{"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
187 	{"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
188 	{"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
189 	{"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
190 	{"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
191 	{"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
192 	{"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
193 	{"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
194 	{"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
195 	{"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
196 	{"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
197 	{"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
198 };
199 
200 /*
201  * This is only used for objects of type DMU_OT_ZNODE
202  */
203 sa_attr_type_t sa_legacy_zpl_layout[] = {
204     0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
205 };
206 
207 /*
208  * Special dummy layout used for buffers with no attributes.
209  */
210 sa_attr_type_t sa_dummy_zpl_layout[] = { 0 };
211 
212 static int sa_legacy_attr_count = 16;
213 static kmem_cache_t *sa_cache = NULL;
214 
215 /*ARGSUSED*/
216 static int
217 sa_cache_constructor(void *buf, void *unused, int kmflag)
218 {
219 	sa_handle_t *hdl = buf;
220 
221 	mutex_init(&hdl->sa_lock, NULL, MUTEX_DEFAULT, NULL);
222 	return (0);
223 }
224 
225 /*ARGSUSED*/
226 static void
227 sa_cache_destructor(void *buf, void *unused)
228 {
229 	sa_handle_t *hdl = buf;
230 	mutex_destroy(&hdl->sa_lock);
231 }
232 
233 void
234 sa_cache_init(void)
235 {
236 	sa_cache = kmem_cache_create("sa_cache",
237 	    sizeof (sa_handle_t), 0, sa_cache_constructor,
238 	    sa_cache_destructor, NULL, NULL, NULL, 0);
239 }
240 
241 void
242 sa_cache_fini(void)
243 {
244 	if (sa_cache)
245 		kmem_cache_destroy(sa_cache);
246 }
247 
248 static int
249 layout_num_compare(const void *arg1, const void *arg2)
250 {
251 	const sa_lot_t *node1 = (const sa_lot_t *)arg1;
252 	const sa_lot_t *node2 = (const sa_lot_t *)arg2;
253 
254 	return (TREE_CMP(node1->lot_num, node2->lot_num));
255 }
256 
257 static int
258 layout_hash_compare(const void *arg1, const void *arg2)
259 {
260 	const sa_lot_t *node1 = (const sa_lot_t *)arg1;
261 	const sa_lot_t *node2 = (const sa_lot_t *)arg2;
262 
263 	int cmp = TREE_CMP(node1->lot_hash, node2->lot_hash);
264 	if (likely(cmp))
265 		return (cmp);
266 
267 	return (TREE_CMP(node1->lot_instance, node2->lot_instance));
268 }
269 
270 boolean_t
271 sa_layout_equal(sa_lot_t *tbf, sa_attr_type_t *attrs, int count)
272 {
273 	int i;
274 
275 	if (count != tbf->lot_attr_count)
276 		return (1);
277 
278 	for (i = 0; i != count; i++) {
279 		if (attrs[i] != tbf->lot_attrs[i])
280 			return (1);
281 	}
282 	return (0);
283 }
284 
285 #define	SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
286 
287 static uint64_t
288 sa_layout_info_hash(sa_attr_type_t *attrs, int attr_count)
289 {
290 	int i;
291 	uint64_t crc = -1ULL;
292 
293 	for (i = 0; i != attr_count; i++)
294 		crc ^= SA_ATTR_HASH(attrs[i]);
295 
296 	return (crc);
297 }
298 
299 static int
300 sa_get_spill(sa_handle_t *hdl)
301 {
302 	int rc;
303 	if (hdl->sa_spill == NULL) {
304 		if ((rc = dmu_spill_hold_existing(hdl->sa_bonus, NULL,
305 		    &hdl->sa_spill)) == 0)
306 			VERIFY(0 == sa_build_index(hdl, SA_SPILL));
307 	} else {
308 		rc = 0;
309 	}
310 
311 	return (rc);
312 }
313 
314 /*
315  * Main attribute lookup/update function
316  * returns 0 for success or non zero for failures
317  *
318  * Operates on bulk array, first failure will abort further processing
319  */
320 int
321 sa_attr_op(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
322     sa_data_op_t data_op, dmu_tx_t *tx)
323 {
324 	sa_os_t *sa = hdl->sa_os->os_sa;
325 	int i;
326 	int error = 0;
327 	sa_buf_type_t buftypes;
328 
329 	buftypes = 0;
330 
331 	ASSERT(count > 0);
332 	for (i = 0; i != count; i++) {
333 		ASSERT(bulk[i].sa_attr <= hdl->sa_os->os_sa->sa_num_attrs);
334 
335 		bulk[i].sa_addr = NULL;
336 		/* First check the bonus buffer */
337 
338 		if (hdl->sa_bonus_tab && TOC_ATTR_PRESENT(
339 		    hdl->sa_bonus_tab->sa_idx_tab[bulk[i].sa_attr])) {
340 			SA_ATTR_INFO(sa, hdl->sa_bonus_tab,
341 			    SA_GET_HDR(hdl, SA_BONUS),
342 			    bulk[i].sa_attr, bulk[i], SA_BONUS, hdl);
343 			if (tx && !(buftypes & SA_BONUS)) {
344 				dmu_buf_will_dirty(hdl->sa_bonus, tx);
345 				buftypes |= SA_BONUS;
346 			}
347 		}
348 		if (bulk[i].sa_addr == NULL &&
349 		    ((error = sa_get_spill(hdl)) == 0)) {
350 			if (TOC_ATTR_PRESENT(
351 			    hdl->sa_spill_tab->sa_idx_tab[bulk[i].sa_attr])) {
352 				SA_ATTR_INFO(sa, hdl->sa_spill_tab,
353 				    SA_GET_HDR(hdl, SA_SPILL),
354 				    bulk[i].sa_attr, bulk[i], SA_SPILL, hdl);
355 				if (tx && !(buftypes & SA_SPILL) &&
356 				    bulk[i].sa_size == bulk[i].sa_length) {
357 					dmu_buf_will_dirty(hdl->sa_spill, tx);
358 					buftypes |= SA_SPILL;
359 				}
360 			}
361 		}
362 		if (error && error != ENOENT) {
363 			return ((error == ECKSUM) ? EIO : error);
364 		}
365 
366 		switch (data_op) {
367 		case SA_LOOKUP:
368 			if (bulk[i].sa_addr == NULL)
369 				return (SET_ERROR(ENOENT));
370 			if (bulk[i].sa_data) {
371 				SA_COPY_DATA(bulk[i].sa_data_func,
372 				    bulk[i].sa_addr, bulk[i].sa_data,
373 				    bulk[i].sa_size);
374 			}
375 			continue;
376 
377 		case SA_UPDATE:
378 			/* existing rewrite of attr */
379 			if (bulk[i].sa_addr &&
380 			    bulk[i].sa_size == bulk[i].sa_length) {
381 				SA_COPY_DATA(bulk[i].sa_data_func,
382 				    bulk[i].sa_data, bulk[i].sa_addr,
383 				    bulk[i].sa_length);
384 				continue;
385 			} else if (bulk[i].sa_addr) { /* attr size change */
386 				error = sa_modify_attrs(hdl, bulk[i].sa_attr,
387 				    SA_REPLACE, bulk[i].sa_data_func,
388 				    bulk[i].sa_data, bulk[i].sa_length, tx);
389 			} else { /* adding new attribute */
390 				error = sa_modify_attrs(hdl, bulk[i].sa_attr,
391 				    SA_ADD, bulk[i].sa_data_func,
392 				    bulk[i].sa_data, bulk[i].sa_length, tx);
393 			}
394 			if (error)
395 				return (error);
396 			break;
397 		}
398 	}
399 	return (error);
400 }
401 
402 static sa_lot_t *
403 sa_add_layout_entry(objset_t *os, sa_attr_type_t *attrs, int attr_count,
404     uint64_t lot_num, uint64_t hash, boolean_t zapadd, dmu_tx_t *tx)
405 {
406 	sa_os_t *sa = os->os_sa;
407 	sa_lot_t *tb, *findtb;
408 	int i, size;
409 	avl_index_t loc;
410 
411 	ASSERT(MUTEX_HELD(&sa->sa_lock));
412 	tb = kmem_zalloc(sizeof (sa_lot_t), KM_SLEEP);
413 	tb->lot_attr_count = attr_count;
414 
415 	if ((size = sizeof (sa_attr_type_t) * attr_count) != 0) {
416 		tb->lot_attrs = kmem_alloc(size, KM_SLEEP);
417 		bcopy(attrs, tb->lot_attrs, size);
418 	}
419 
420 	tb->lot_num = lot_num;
421 	tb->lot_hash = hash;
422 	tb->lot_instance = 0;
423 
424 	if (zapadd) {
425 		char attr_name[8];
426 
427 		if (sa->sa_layout_attr_obj == 0) {
428 			sa->sa_layout_attr_obj = zap_create_link(os,
429 			    DMU_OT_SA_ATTR_LAYOUTS,
430 			    sa->sa_master_obj, SA_LAYOUTS, tx);
431 		}
432 
433 		(void) snprintf(attr_name, sizeof (attr_name),
434 		    "%d", (int)lot_num);
435 		VERIFY(0 == zap_update(os, os->os_sa->sa_layout_attr_obj,
436 		    attr_name, 2, attr_count, attrs, tx));
437 	}
438 
439 	list_create(&tb->lot_idx_tab, sizeof (sa_idx_tab_t),
440 	    offsetof(sa_idx_tab_t, sa_next));
441 
442 	for (i = 0; i != attr_count; i++) {
443 		if (sa->sa_attr_table[tb->lot_attrs[i]].sa_length == 0)
444 			tb->lot_var_sizes++;
445 	}
446 
447 	avl_add(&sa->sa_layout_num_tree, tb);
448 
449 	/* verify we don't have a hash collision */
450 	if ((findtb = avl_find(&sa->sa_layout_hash_tree, tb, &loc)) != NULL) {
451 		for (; findtb && findtb->lot_hash == hash;
452 		    findtb = AVL_NEXT(&sa->sa_layout_hash_tree, findtb)) {
453 			if (findtb->lot_instance != tb->lot_instance)
454 				break;
455 			tb->lot_instance++;
456 		}
457 	}
458 	avl_add(&sa->sa_layout_hash_tree, tb);
459 	return (tb);
460 }
461 
462 static void
463 sa_find_layout(objset_t *os, uint64_t hash, sa_attr_type_t *attrs,
464     int count, dmu_tx_t *tx, sa_lot_t **lot)
465 {
466 	sa_lot_t *tb, tbsearch;
467 	avl_index_t loc;
468 	sa_os_t *sa = os->os_sa;
469 	boolean_t found = B_FALSE;
470 
471 	mutex_enter(&sa->sa_lock);
472 	tbsearch.lot_hash = hash;
473 	tbsearch.lot_instance = 0;
474 	tb = avl_find(&sa->sa_layout_hash_tree, &tbsearch, &loc);
475 	if (tb) {
476 		for (; tb && tb->lot_hash == hash;
477 		    tb = AVL_NEXT(&sa->sa_layout_hash_tree, tb)) {
478 			if (sa_layout_equal(tb, attrs, count) == 0) {
479 				found = B_TRUE;
480 				break;
481 			}
482 		}
483 	}
484 	if (!found) {
485 		tb = sa_add_layout_entry(os, attrs, count,
486 		    avl_numnodes(&sa->sa_layout_num_tree), hash, B_TRUE, tx);
487 	}
488 	mutex_exit(&sa->sa_lock);
489 	*lot = tb;
490 }
491 
492 static int
493 sa_resize_spill(sa_handle_t *hdl, uint32_t size, dmu_tx_t *tx)
494 {
495 	int error;
496 	uint32_t blocksize;
497 
498 	if (size == 0) {
499 		blocksize = SPA_MINBLOCKSIZE;
500 	} else if (size > SPA_OLD_MAXBLOCKSIZE) {
501 		ASSERT(0);
502 		return (SET_ERROR(EFBIG));
503 	} else {
504 		blocksize = P2ROUNDUP_TYPED(size, SPA_MINBLOCKSIZE, uint32_t);
505 	}
506 
507 	error = dbuf_spill_set_blksz(hdl->sa_spill, blocksize, tx);
508 	ASSERT(error == 0);
509 	return (error);
510 }
511 
512 static void
513 sa_copy_data(sa_data_locator_t *func, void *datastart, void *target, int buflen)
514 {
515 	if (func == NULL) {
516 		bcopy(datastart, target, buflen);
517 	} else {
518 		boolean_t start;
519 		int bytes;
520 		void *dataptr;
521 		void *saptr = target;
522 		uint32_t length;
523 
524 		start = B_TRUE;
525 		bytes = 0;
526 		while (bytes < buflen) {
527 			func(&dataptr, &length, buflen, start, datastart);
528 			bcopy(dataptr, saptr, length);
529 			saptr = (void *)((caddr_t)saptr + length);
530 			bytes += length;
531 			start = B_FALSE;
532 		}
533 	}
534 }
535 
536 /*
537  * Determine several different sizes
538  * first the sa header size
539  * the number of bytes to be stored
540  * if spill would occur the index in the attribute array is returned
541  *
542  * the boolean will_spill will be set when spilling is necessary.  It
543  * is only set when the buftype is SA_BONUS
544  */
545 static int
546 sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count,
547     dmu_buf_t *db, sa_buf_type_t buftype, int full_space, int *index,
548     int *total, boolean_t *will_spill)
549 {
550 	int var_size = 0;
551 	int i;
552 	int hdrsize;
553 	int extra_hdrsize;
554 
555 	if (buftype == SA_BONUS && sa->sa_force_spill) {
556 		*total = 0;
557 		*index = 0;
558 		*will_spill = B_TRUE;
559 		return (0);
560 	}
561 
562 	*index = -1;
563 	*total = 0;
564 	*will_spill = B_FALSE;
565 
566 	extra_hdrsize = 0;
567 	hdrsize = (SA_BONUSTYPE_FROM_DB(db) == DMU_OT_ZNODE) ? 0 :
568 	    sizeof (sa_hdr_phys_t);
569 
570 	ASSERT(IS_P2ALIGNED(full_space, 8));
571 
572 	for (i = 0; i != attr_count; i++) {
573 		boolean_t is_var_sz;
574 
575 		*total = P2ROUNDUP(*total, 8);
576 		*total += attr_desc[i].sa_length;
577 		if (*will_spill)
578 			continue;
579 
580 		is_var_sz = (SA_REGISTERED_LEN(sa, attr_desc[i].sa_attr) == 0);
581 		if (is_var_sz) {
582 			var_size++;
583 		}
584 
585 		if (is_var_sz && var_size > 1) {
586 			/*
587 			 * Don't worry that the spill block might overflow.
588 			 * It will be resized if needed in sa_build_layouts().
589 			 */
590 			if (buftype == SA_SPILL ||
591 			    P2ROUNDUP(hdrsize + sizeof (uint16_t), 8) +
592 			    *total < full_space) {
593 				/*
594 				 * Account for header space used by array of
595 				 * optional sizes of variable-length attributes.
596 				 * Record the extra header size in case this
597 				 * increase needs to be reversed due to
598 				 * spill-over.
599 				 */
600 				hdrsize += sizeof (uint16_t);
601 				if (*index != -1)
602 					extra_hdrsize += sizeof (uint16_t);
603 			} else {
604 				ASSERT(buftype == SA_BONUS);
605 				if (*index == -1)
606 					*index = i;
607 				*will_spill = B_TRUE;
608 				continue;
609 			}
610 		}
611 
612 		/*
613 		 * find index of where spill *could* occur.
614 		 * Then continue to count of remainder attribute
615 		 * space.  The sum is used later for sizing bonus
616 		 * and spill buffer.
617 		 */
618 		if (buftype == SA_BONUS && *index == -1 &&
619 		    *total + P2ROUNDUP(hdrsize, 8) >
620 		    (full_space - sizeof (blkptr_t))) {
621 			*index = i;
622 		}
623 
624 		if (*total + P2ROUNDUP(hdrsize, 8) > full_space &&
625 		    buftype == SA_BONUS)
626 			*will_spill = B_TRUE;
627 	}
628 
629 	if (*will_spill)
630 		hdrsize -= extra_hdrsize;
631 
632 	hdrsize = P2ROUNDUP(hdrsize, 8);
633 	return (hdrsize);
634 }
635 
636 #define	BUF_SPACE_NEEDED(total, header) (total + header)
637 
638 /*
639  * Find layout that corresponds to ordering of attributes
640  * If not found a new layout number is created and added to
641  * persistent layout tables.
642  */
643 static int
644 sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
645     dmu_tx_t *tx)
646 {
647 	sa_os_t *sa = hdl->sa_os->os_sa;
648 	uint64_t hash;
649 	sa_buf_type_t buftype;
650 	sa_hdr_phys_t *sahdr;
651 	void *data_start;
652 	int buf_space;
653 	sa_attr_type_t *attrs, *attrs_start;
654 	int i, lot_count;
655 	int dnodesize;
656 	int hdrsize;
657 	int spillhdrsize = 0;
658 	int used;
659 	dmu_object_type_t bonustype;
660 	sa_lot_t *lot;
661 	int len_idx;
662 	int spill_used;
663 	int bonuslen;
664 	boolean_t spilling;
665 
666 	dmu_buf_will_dirty(hdl->sa_bonus, tx);
667 	bonustype = SA_BONUSTYPE_FROM_DB(hdl->sa_bonus);
668 
669 	dmu_object_dnsize_from_db(hdl->sa_bonus, &dnodesize);
670 	bonuslen = DN_BONUS_SIZE(dnodesize);
671 
672 	/* first determine bonus header size and sum of all attributes */
673 	hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus,
674 	    SA_BONUS, bonuslen, &i, &used, &spilling);
675 
676 	if (used > SPA_OLD_MAXBLOCKSIZE)
677 		return (SET_ERROR(EFBIG));
678 
679 	VERIFY(0 == dmu_set_bonus(hdl->sa_bonus, spilling ?
680 	    MIN(bonuslen - sizeof (blkptr_t), used + hdrsize) :
681 	    used + hdrsize, tx));
682 
683 	ASSERT((bonustype == DMU_OT_ZNODE && spilling == 0) ||
684 	    bonustype == DMU_OT_SA);
685 
686 	/* setup and size spill buffer when needed */
687 	if (spilling) {
688 		boolean_t dummy;
689 
690 		if (hdl->sa_spill == NULL) {
691 			VERIFY(dmu_spill_hold_by_bonus(hdl->sa_bonus, 0, NULL,
692 			    &hdl->sa_spill) == 0);
693 		}
694 		dmu_buf_will_dirty(hdl->sa_spill, tx);
695 
696 		spillhdrsize = sa_find_sizes(sa, &attr_desc[i],
697 		    attr_count - i, hdl->sa_spill, SA_SPILL,
698 		    hdl->sa_spill->db_size, &i, &spill_used, &dummy);
699 
700 		if (spill_used > SPA_OLD_MAXBLOCKSIZE)
701 			return (SET_ERROR(EFBIG));
702 
703 		buf_space = hdl->sa_spill->db_size - spillhdrsize;
704 		if (BUF_SPACE_NEEDED(spill_used, spillhdrsize) >
705 		    hdl->sa_spill->db_size)
706 			VERIFY(0 == sa_resize_spill(hdl,
707 			    BUF_SPACE_NEEDED(spill_used, spillhdrsize), tx));
708 	}
709 
710 	/* setup starting pointers to lay down data */
711 	data_start = (void *)((uintptr_t)hdl->sa_bonus->db_data + hdrsize);
712 	sahdr = (sa_hdr_phys_t *)hdl->sa_bonus->db_data;
713 	buftype = SA_BONUS;
714 
715 	if (spilling)
716 		buf_space = (sa->sa_force_spill) ?
717 		    0 : SA_BLKPTR_SPACE - hdrsize;
718 	else
719 		buf_space = hdl->sa_bonus->db_size - hdrsize;
720 
721 	attrs_start = attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
722 	    KM_SLEEP);
723 	lot_count = 0;
724 
725 	for (i = 0, len_idx = 0, hash = -1ULL; i != attr_count; i++) {
726 		uint16_t length;
727 
728 		ASSERT(IS_P2ALIGNED(data_start, 8));
729 		ASSERT(IS_P2ALIGNED(buf_space, 8));
730 		attrs[i] = attr_desc[i].sa_attr;
731 		length = SA_REGISTERED_LEN(sa, attrs[i]);
732 		if (length == 0)
733 			length = attr_desc[i].sa_length;
734 
735 		if (buf_space < length) {  /* switch to spill buffer */
736 			VERIFY(spilling);
737 			VERIFY(bonustype == DMU_OT_SA);
738 			if (buftype == SA_BONUS && !sa->sa_force_spill) {
739 				sa_find_layout(hdl->sa_os, hash, attrs_start,
740 				    lot_count, tx, &lot);
741 				SA_SET_HDR(sahdr, lot->lot_num, hdrsize);
742 			}
743 
744 			buftype = SA_SPILL;
745 			hash = -1ULL;
746 			len_idx = 0;
747 
748 			sahdr = (sa_hdr_phys_t *)hdl->sa_spill->db_data;
749 			sahdr->sa_magic = SA_MAGIC;
750 			data_start = (void *)((uintptr_t)sahdr +
751 			    spillhdrsize);
752 			attrs_start = &attrs[i];
753 			buf_space = hdl->sa_spill->db_size - spillhdrsize;
754 			lot_count = 0;
755 		}
756 		hash ^= SA_ATTR_HASH(attrs[i]);
757 		attr_desc[i].sa_addr = data_start;
758 		attr_desc[i].sa_size = length;
759 		SA_COPY_DATA(attr_desc[i].sa_data_func, attr_desc[i].sa_data,
760 		    data_start, length);
761 		if (sa->sa_attr_table[attrs[i]].sa_length == 0) {
762 			sahdr->sa_lengths[len_idx++] = length;
763 		}
764 		data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
765 		    length), 8);
766 		buf_space -= P2ROUNDUP(length, 8);
767 		lot_count++;
768 	}
769 
770 	sa_find_layout(hdl->sa_os, hash, attrs_start, lot_count, tx, &lot);
771 
772 	/*
773 	 * Verify that old znodes always have layout number 0.
774 	 * Must be DMU_OT_SA for arbitrary layouts
775 	 */
776 	VERIFY((bonustype == DMU_OT_ZNODE && lot->lot_num == 0) ||
777 	    (bonustype == DMU_OT_SA && lot->lot_num > 1));
778 
779 	if (bonustype == DMU_OT_SA) {
780 		SA_SET_HDR(sahdr, lot->lot_num,
781 		    buftype == SA_BONUS ? hdrsize : spillhdrsize);
782 	}
783 
784 	kmem_free(attrs, sizeof (sa_attr_type_t) * attr_count);
785 	if (hdl->sa_bonus_tab) {
786 		sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
787 		hdl->sa_bonus_tab = NULL;
788 	}
789 	if (!sa->sa_force_spill)
790 		VERIFY(0 == sa_build_index(hdl, SA_BONUS));
791 	if (hdl->sa_spill) {
792 		sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
793 		if (!spilling) {
794 			/*
795 			 * remove spill block that is no longer needed.
796 			 */
797 			dmu_buf_rele(hdl->sa_spill, NULL);
798 			hdl->sa_spill = NULL;
799 			hdl->sa_spill_tab = NULL;
800 			VERIFY(0 == dmu_rm_spill(hdl->sa_os,
801 			    sa_handle_object(hdl), tx));
802 		} else {
803 			VERIFY(0 == sa_build_index(hdl, SA_SPILL));
804 		}
805 	}
806 
807 	return (0);
808 }
809 
810 static void
811 sa_free_attr_table(sa_os_t *sa)
812 {
813 	int i;
814 
815 	if (sa->sa_attr_table == NULL)
816 		return;
817 
818 	for (i = 0; i != sa->sa_num_attrs; i++) {
819 		if (sa->sa_attr_table[i].sa_name)
820 			kmem_free(sa->sa_attr_table[i].sa_name,
821 			    strlen(sa->sa_attr_table[i].sa_name) + 1);
822 	}
823 
824 	kmem_free(sa->sa_attr_table,
825 	    sizeof (sa_attr_table_t) * sa->sa_num_attrs);
826 
827 	sa->sa_attr_table = NULL;
828 }
829 
830 static int
831 sa_attr_table_setup(objset_t *os, sa_attr_reg_t *reg_attrs, int count)
832 {
833 	sa_os_t *sa = os->os_sa;
834 	uint64_t sa_attr_count = 0;
835 	uint64_t sa_reg_count = 0;
836 	int error = 0;
837 	uint64_t attr_value;
838 	sa_attr_table_t *tb;
839 	zap_cursor_t zc;
840 	zap_attribute_t za;
841 	int registered_count = 0;
842 	int i;
843 	dmu_objset_type_t ostype = dmu_objset_type(os);
844 
845 	sa->sa_user_table =
846 	    kmem_zalloc(count * sizeof (sa_attr_type_t), KM_SLEEP);
847 	sa->sa_user_table_sz = count * sizeof (sa_attr_type_t);
848 
849 	if (sa->sa_reg_attr_obj != 0) {
850 		error = zap_count(os, sa->sa_reg_attr_obj,
851 		    &sa_attr_count);
852 
853 		/*
854 		 * Make sure we retrieved a count and that it isn't zero
855 		 */
856 		if (error || (error == 0 && sa_attr_count == 0)) {
857 			if (error == 0)
858 				error = SET_ERROR(EINVAL);
859 			goto bail;
860 		}
861 		sa_reg_count = sa_attr_count;
862 	}
863 
864 	if (ostype == DMU_OST_ZFS && sa_attr_count == 0)
865 		sa_attr_count += sa_legacy_attr_count;
866 
867 	/* Allocate attribute numbers for attributes that aren't registered */
868 	for (i = 0; i != count; i++) {
869 		boolean_t found = B_FALSE;
870 		int j;
871 
872 		if (ostype == DMU_OST_ZFS) {
873 			for (j = 0; j != sa_legacy_attr_count; j++) {
874 				if (strcmp(reg_attrs[i].sa_name,
875 				    sa_legacy_attrs[j].sa_name) == 0) {
876 					sa->sa_user_table[i] =
877 					    sa_legacy_attrs[j].sa_attr;
878 					found = B_TRUE;
879 				}
880 			}
881 		}
882 		if (found)
883 			continue;
884 
885 		if (sa->sa_reg_attr_obj)
886 			error = zap_lookup(os, sa->sa_reg_attr_obj,
887 			    reg_attrs[i].sa_name, 8, 1, &attr_value);
888 		else
889 			error = SET_ERROR(ENOENT);
890 		switch (error) {
891 		case ENOENT:
892 			sa->sa_user_table[i] = (sa_attr_type_t)sa_attr_count;
893 			sa_attr_count++;
894 			break;
895 		case 0:
896 			sa->sa_user_table[i] = ATTR_NUM(attr_value);
897 			break;
898 		default:
899 			goto bail;
900 		}
901 	}
902 
903 	sa->sa_num_attrs = sa_attr_count;
904 	tb = sa->sa_attr_table =
905 	    kmem_zalloc(sizeof (sa_attr_table_t) * sa_attr_count, KM_SLEEP);
906 
907 	/*
908 	 * Attribute table is constructed from requested attribute list,
909 	 * previously foreign registered attributes, and also the legacy
910 	 * ZPL set of attributes.
911 	 */
912 
913 	if (sa->sa_reg_attr_obj) {
914 		for (zap_cursor_init(&zc, os, sa->sa_reg_attr_obj);
915 		    (error = zap_cursor_retrieve(&zc, &za)) == 0;
916 		    zap_cursor_advance(&zc)) {
917 			uint64_t value;
918 			value  = za.za_first_integer;
919 
920 			registered_count++;
921 			tb[ATTR_NUM(value)].sa_attr = ATTR_NUM(value);
922 			tb[ATTR_NUM(value)].sa_length = ATTR_LENGTH(value);
923 			tb[ATTR_NUM(value)].sa_byteswap = ATTR_BSWAP(value);
924 			tb[ATTR_NUM(value)].sa_registered = B_TRUE;
925 
926 			if (tb[ATTR_NUM(value)].sa_name) {
927 				continue;
928 			}
929 			tb[ATTR_NUM(value)].sa_name =
930 			    kmem_zalloc(strlen(za.za_name) +1, KM_SLEEP);
931 			(void) strlcpy(tb[ATTR_NUM(value)].sa_name, za.za_name,
932 			    strlen(za.za_name) +1);
933 		}
934 		zap_cursor_fini(&zc);
935 		/*
936 		 * Make sure we processed the correct number of registered
937 		 * attributes
938 		 */
939 		if (registered_count != sa_reg_count) {
940 			ASSERT(error != 0);
941 			goto bail;
942 		}
943 
944 	}
945 
946 	if (ostype == DMU_OST_ZFS) {
947 		for (i = 0; i != sa_legacy_attr_count; i++) {
948 			if (tb[i].sa_name)
949 				continue;
950 			tb[i].sa_attr = sa_legacy_attrs[i].sa_attr;
951 			tb[i].sa_length = sa_legacy_attrs[i].sa_length;
952 			tb[i].sa_byteswap = sa_legacy_attrs[i].sa_byteswap;
953 			tb[i].sa_registered = B_FALSE;
954 			tb[i].sa_name =
955 			    kmem_zalloc(strlen(sa_legacy_attrs[i].sa_name) +1,
956 			    KM_SLEEP);
957 			(void) strlcpy(tb[i].sa_name,
958 			    sa_legacy_attrs[i].sa_name,
959 			    strlen(sa_legacy_attrs[i].sa_name) + 1);
960 		}
961 	}
962 
963 	for (i = 0; i != count; i++) {
964 		sa_attr_type_t attr_id;
965 
966 		attr_id = sa->sa_user_table[i];
967 		if (tb[attr_id].sa_name)
968 			continue;
969 
970 		tb[attr_id].sa_length = reg_attrs[i].sa_length;
971 		tb[attr_id].sa_byteswap = reg_attrs[i].sa_byteswap;
972 		tb[attr_id].sa_attr = attr_id;
973 		tb[attr_id].sa_name =
974 		    kmem_zalloc(strlen(reg_attrs[i].sa_name) + 1, KM_SLEEP);
975 		(void) strlcpy(tb[attr_id].sa_name, reg_attrs[i].sa_name,
976 		    strlen(reg_attrs[i].sa_name) + 1);
977 	}
978 
979 	sa->sa_need_attr_registration =
980 	    (sa_attr_count != registered_count);
981 
982 	return (0);
983 bail:
984 	kmem_free(sa->sa_user_table, count * sizeof (sa_attr_type_t));
985 	sa->sa_user_table = NULL;
986 	sa_free_attr_table(sa);
987 	return ((error != 0) ? error : EINVAL);
988 }
989 
990 int
991 sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count,
992     sa_attr_type_t **user_table)
993 {
994 	zap_cursor_t zc;
995 	zap_attribute_t za;
996 	sa_os_t *sa;
997 	dmu_objset_type_t ostype = dmu_objset_type(os);
998 	sa_attr_type_t *tb;
999 	int error;
1000 
1001 	mutex_enter(&os->os_user_ptr_lock);
1002 	if (os->os_sa) {
1003 		mutex_enter(&os->os_sa->sa_lock);
1004 		mutex_exit(&os->os_user_ptr_lock);
1005 		tb = os->os_sa->sa_user_table;
1006 		mutex_exit(&os->os_sa->sa_lock);
1007 		*user_table = tb;
1008 		return (0);
1009 	}
1010 
1011 	sa = kmem_zalloc(sizeof (sa_os_t), KM_SLEEP);
1012 	mutex_init(&sa->sa_lock, NULL, MUTEX_DEFAULT, NULL);
1013 	sa->sa_master_obj = sa_obj;
1014 
1015 	os->os_sa = sa;
1016 	mutex_enter(&sa->sa_lock);
1017 	mutex_exit(&os->os_user_ptr_lock);
1018 	avl_create(&sa->sa_layout_num_tree, layout_num_compare,
1019 	    sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node));
1020 	avl_create(&sa->sa_layout_hash_tree, layout_hash_compare,
1021 	    sizeof (sa_lot_t), offsetof(sa_lot_t, lot_hash_node));
1022 
1023 	if (sa_obj) {
1024 		error = zap_lookup(os, sa_obj, SA_LAYOUTS,
1025 		    8, 1, &sa->sa_layout_attr_obj);
1026 		if (error != 0 && error != ENOENT)
1027 			goto fail;
1028 		error = zap_lookup(os, sa_obj, SA_REGISTRY,
1029 		    8, 1, &sa->sa_reg_attr_obj);
1030 		if (error != 0 && error != ENOENT)
1031 			goto fail;
1032 	}
1033 
1034 	if ((error = sa_attr_table_setup(os, reg_attrs, count)) != 0)
1035 		goto fail;
1036 
1037 	if (sa->sa_layout_attr_obj != 0) {
1038 		uint64_t layout_count;
1039 
1040 		error = zap_count(os, sa->sa_layout_attr_obj,
1041 		    &layout_count);
1042 
1043 		/*
1044 		 * Layout number count should be > 0
1045 		 */
1046 		if (error || (error == 0 && layout_count == 0)) {
1047 			if (error == 0)
1048 				error = SET_ERROR(EINVAL);
1049 			goto fail;
1050 		}
1051 
1052 		for (zap_cursor_init(&zc, os, sa->sa_layout_attr_obj);
1053 		    (error = zap_cursor_retrieve(&zc, &za)) == 0;
1054 		    zap_cursor_advance(&zc)) {
1055 			sa_attr_type_t *lot_attrs;
1056 			uint64_t lot_num;
1057 
1058 			lot_attrs = kmem_zalloc(sizeof (sa_attr_type_t) *
1059 			    za.za_num_integers, KM_SLEEP);
1060 
1061 			if ((error = (zap_lookup(os, sa->sa_layout_attr_obj,
1062 			    za.za_name, 2, za.za_num_integers,
1063 			    lot_attrs))) != 0) {
1064 				kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
1065 				    za.za_num_integers);
1066 				break;
1067 			}
1068 			VERIFY(ddi_strtoull(za.za_name, NULL, 10,
1069 			    (unsigned long long *)&lot_num) == 0);
1070 
1071 			(void) sa_add_layout_entry(os, lot_attrs,
1072 			    za.za_num_integers, lot_num,
1073 			    sa_layout_info_hash(lot_attrs,
1074 			    za.za_num_integers), B_FALSE, NULL);
1075 			kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
1076 			    za.za_num_integers);
1077 		}
1078 		zap_cursor_fini(&zc);
1079 
1080 		/*
1081 		 * Make sure layout count matches number of entries added
1082 		 * to AVL tree
1083 		 */
1084 		if (avl_numnodes(&sa->sa_layout_num_tree) != layout_count) {
1085 			ASSERT(error != 0);
1086 			goto fail;
1087 		}
1088 	}
1089 
1090 	/* Add special layout number for old ZNODES */
1091 	if (ostype == DMU_OST_ZFS) {
1092 		(void) sa_add_layout_entry(os, sa_legacy_zpl_layout,
1093 		    sa_legacy_attr_count, 0,
1094 		    sa_layout_info_hash(sa_legacy_zpl_layout,
1095 		    sa_legacy_attr_count), B_FALSE, NULL);
1096 
1097 		(void) sa_add_layout_entry(os, sa_dummy_zpl_layout, 0, 1,
1098 		    0, B_FALSE, NULL);
1099 	}
1100 	*user_table = os->os_sa->sa_user_table;
1101 	mutex_exit(&sa->sa_lock);
1102 	return (0);
1103 fail:
1104 	os->os_sa = NULL;
1105 	sa_free_attr_table(sa);
1106 	if (sa->sa_user_table)
1107 		kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
1108 	mutex_exit(&sa->sa_lock);
1109 	avl_destroy(&sa->sa_layout_hash_tree);
1110 	avl_destroy(&sa->sa_layout_num_tree);
1111 	mutex_destroy(&sa->sa_lock);
1112 	kmem_free(sa, sizeof (sa_os_t));
1113 	return ((error == ECKSUM) ? EIO : error);
1114 }
1115 
1116 void
1117 sa_tear_down(objset_t *os)
1118 {
1119 	sa_os_t *sa = os->os_sa;
1120 	sa_lot_t *layout;
1121 	void *cookie;
1122 
1123 	kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
1124 
1125 	/* Free up attr table */
1126 
1127 	sa_free_attr_table(sa);
1128 
1129 	cookie = NULL;
1130 	while (layout = avl_destroy_nodes(&sa->sa_layout_hash_tree, &cookie)) {
1131 		sa_idx_tab_t *tab;
1132 		while (tab = list_head(&layout->lot_idx_tab)) {
1133 			ASSERT(zfs_refcount_count(&tab->sa_refcount));
1134 			sa_idx_tab_rele(os, tab);
1135 		}
1136 	}
1137 
1138 	cookie = NULL;
1139 	while (layout = avl_destroy_nodes(&sa->sa_layout_num_tree, &cookie)) {
1140 		kmem_free(layout->lot_attrs,
1141 		    sizeof (sa_attr_type_t) * layout->lot_attr_count);
1142 		kmem_free(layout, sizeof (sa_lot_t));
1143 	}
1144 
1145 	avl_destroy(&sa->sa_layout_hash_tree);
1146 	avl_destroy(&sa->sa_layout_num_tree);
1147 	mutex_destroy(&sa->sa_lock);
1148 
1149 	kmem_free(sa, sizeof (sa_os_t));
1150 	os->os_sa = NULL;
1151 }
1152 
1153 void
1154 sa_build_idx_tab(void *hdr, void *attr_addr, sa_attr_type_t attr,
1155     uint16_t length, int length_idx, boolean_t var_length, void *userp)
1156 {
1157 	sa_idx_tab_t *idx_tab = userp;
1158 
1159 	if (var_length) {
1160 		ASSERT(idx_tab->sa_variable_lengths);
1161 		idx_tab->sa_variable_lengths[length_idx] = length;
1162 	}
1163 	TOC_ATTR_ENCODE(idx_tab->sa_idx_tab[attr], length_idx,
1164 	    (uint32_t)((uintptr_t)attr_addr - (uintptr_t)hdr));
1165 }
1166 
1167 static void
1168 sa_attr_iter(objset_t *os, sa_hdr_phys_t *hdr, dmu_object_type_t type,
1169     sa_iterfunc_t func, sa_lot_t *tab, void *userp)
1170 {
1171 	void *data_start;
1172 	sa_lot_t *tb = tab;
1173 	sa_lot_t search;
1174 	avl_index_t loc;
1175 	sa_os_t *sa = os->os_sa;
1176 	int i;
1177 	uint16_t *length_start = NULL;
1178 	uint8_t length_idx = 0;
1179 
1180 	if (tab == NULL) {
1181 		search.lot_num = SA_LAYOUT_NUM(hdr, type);
1182 		tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
1183 		ASSERT(tb);
1184 	}
1185 
1186 	if (IS_SA_BONUSTYPE(type)) {
1187 		data_start = (void *)P2ROUNDUP(((uintptr_t)hdr +
1188 		    offsetof(sa_hdr_phys_t, sa_lengths) +
1189 		    (sizeof (uint16_t) * tb->lot_var_sizes)), 8);
1190 		length_start = hdr->sa_lengths;
1191 	} else {
1192 		data_start = hdr;
1193 	}
1194 
1195 	for (i = 0; i != tb->lot_attr_count; i++) {
1196 		int attr_length, reg_length;
1197 		uint8_t idx_len;
1198 
1199 		reg_length = sa->sa_attr_table[tb->lot_attrs[i]].sa_length;
1200 		if (reg_length) {
1201 			attr_length = reg_length;
1202 			idx_len = 0;
1203 		} else {
1204 			attr_length = length_start[length_idx];
1205 			idx_len = length_idx++;
1206 		}
1207 
1208 		func(hdr, data_start, tb->lot_attrs[i], attr_length,
1209 		    idx_len, reg_length == 0 ? B_TRUE : B_FALSE, userp);
1210 
1211 		data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
1212 		    attr_length), 8);
1213 	}
1214 }
1215 
1216 /*ARGSUSED*/
1217 void
1218 sa_byteswap_cb(void *hdr, void *attr_addr, sa_attr_type_t attr,
1219     uint16_t length, int length_idx, boolean_t variable_length, void *userp)
1220 {
1221 	sa_handle_t *hdl = userp;
1222 	sa_os_t *sa = hdl->sa_os->os_sa;
1223 
1224 	sa_bswap_table[sa->sa_attr_table[attr].sa_byteswap](attr_addr, length);
1225 }
1226 
1227 void
1228 sa_byteswap(sa_handle_t *hdl, sa_buf_type_t buftype)
1229 {
1230 	sa_hdr_phys_t *sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1231 	dmu_buf_impl_t *db;
1232 	sa_os_t *sa = hdl->sa_os->os_sa;
1233 	int num_lengths = 1;
1234 	int i;
1235 
1236 	ASSERT(MUTEX_HELD(&sa->sa_lock));
1237 	if (sa_hdr_phys->sa_magic == SA_MAGIC)
1238 		return;
1239 
1240 	db = SA_GET_DB(hdl, buftype);
1241 
1242 	if (buftype == SA_SPILL) {
1243 		arc_release(db->db_buf, NULL);
1244 		arc_buf_thaw(db->db_buf);
1245 	}
1246 
1247 	sa_hdr_phys->sa_magic = BSWAP_32(sa_hdr_phys->sa_magic);
1248 	sa_hdr_phys->sa_layout_info = BSWAP_16(sa_hdr_phys->sa_layout_info);
1249 
1250 	/*
1251 	 * Determine number of variable lenghts in header
1252 	 * The standard 8 byte header has one for free and a
1253 	 * 16 byte header would have 4 + 1;
1254 	 */
1255 	if (SA_HDR_SIZE(sa_hdr_phys) > 8)
1256 		num_lengths += (SA_HDR_SIZE(sa_hdr_phys) - 8) >> 1;
1257 	for (i = 0; i != num_lengths; i++)
1258 		sa_hdr_phys->sa_lengths[i] =
1259 		    BSWAP_16(sa_hdr_phys->sa_lengths[i]);
1260 
1261 	sa_attr_iter(hdl->sa_os, sa_hdr_phys, DMU_OT_SA,
1262 	    sa_byteswap_cb, NULL, hdl);
1263 
1264 	if (buftype == SA_SPILL)
1265 		arc_buf_freeze(((dmu_buf_impl_t *)hdl->sa_spill)->db_buf);
1266 }
1267 
1268 static int
1269 sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype)
1270 {
1271 	sa_hdr_phys_t *sa_hdr_phys;
1272 	dmu_buf_impl_t *db = SA_GET_DB(hdl, buftype);
1273 	dmu_object_type_t bonustype = SA_BONUSTYPE_FROM_DB(db);
1274 	sa_os_t *sa = hdl->sa_os->os_sa;
1275 	sa_idx_tab_t *idx_tab;
1276 
1277 	sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1278 
1279 	mutex_enter(&sa->sa_lock);
1280 
1281 	/* Do we need to byteswap? */
1282 
1283 	/* only check if not old znode */
1284 	if (IS_SA_BONUSTYPE(bonustype) && sa_hdr_phys->sa_magic != SA_MAGIC &&
1285 	    sa_hdr_phys->sa_magic != 0) {
1286 		VERIFY(BSWAP_32(sa_hdr_phys->sa_magic) == SA_MAGIC);
1287 		sa_byteswap(hdl, buftype);
1288 	}
1289 
1290 	idx_tab = sa_find_idx_tab(hdl->sa_os, bonustype, sa_hdr_phys);
1291 
1292 	if (buftype == SA_BONUS)
1293 		hdl->sa_bonus_tab = idx_tab;
1294 	else
1295 		hdl->sa_spill_tab = idx_tab;
1296 
1297 	mutex_exit(&sa->sa_lock);
1298 	return (0);
1299 }
1300 
1301 /*ARGSUSED*/
1302 static void
1303 sa_evict_sync(void *dbu)
1304 {
1305 	panic("evicting sa dbuf\n");
1306 }
1307 
1308 static void
1309 sa_idx_tab_rele(objset_t *os, void *arg)
1310 {
1311 	sa_os_t *sa = os->os_sa;
1312 	sa_idx_tab_t *idx_tab = arg;
1313 
1314 	if (idx_tab == NULL)
1315 		return;
1316 
1317 	mutex_enter(&sa->sa_lock);
1318 	if (zfs_refcount_remove(&idx_tab->sa_refcount, NULL) == 0) {
1319 		list_remove(&idx_tab->sa_layout->lot_idx_tab, idx_tab);
1320 		if (idx_tab->sa_variable_lengths)
1321 			kmem_free(idx_tab->sa_variable_lengths,
1322 			    sizeof (uint16_t) *
1323 			    idx_tab->sa_layout->lot_var_sizes);
1324 		zfs_refcount_destroy(&idx_tab->sa_refcount);
1325 		kmem_free(idx_tab->sa_idx_tab,
1326 		    sizeof (uint32_t) * sa->sa_num_attrs);
1327 		kmem_free(idx_tab, sizeof (sa_idx_tab_t));
1328 	}
1329 	mutex_exit(&sa->sa_lock);
1330 }
1331 
1332 static void
1333 sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab)
1334 {
1335 	sa_os_t *sa = os->os_sa;
1336 
1337 	ASSERT(MUTEX_HELD(&sa->sa_lock));
1338 	(void) zfs_refcount_add(&idx_tab->sa_refcount, NULL);
1339 }
1340 
1341 void
1342 sa_handle_destroy(sa_handle_t *hdl)
1343 {
1344 	dmu_buf_t *db = hdl->sa_bonus;
1345 
1346 	mutex_enter(&hdl->sa_lock);
1347 	(void) dmu_buf_remove_user(db, &hdl->sa_dbu);
1348 
1349 	if (hdl->sa_bonus_tab)
1350 		sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
1351 
1352 	if (hdl->sa_spill_tab)
1353 		sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
1354 
1355 	dmu_buf_rele(hdl->sa_bonus, NULL);
1356 
1357 	if (hdl->sa_spill)
1358 		dmu_buf_rele((dmu_buf_t *)hdl->sa_spill, NULL);
1359 	mutex_exit(&hdl->sa_lock);
1360 
1361 	kmem_cache_free(sa_cache, hdl);
1362 }
1363 
1364 int
1365 sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
1366     sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1367 {
1368 	int error = 0;
1369 	dmu_object_info_t doi;
1370 	sa_handle_t *handle = NULL;
1371 
1372 #ifdef ZFS_DEBUG
1373 	dmu_object_info_from_db(db, &doi);
1374 	ASSERT(doi.doi_bonus_type == DMU_OT_SA ||
1375 	    doi.doi_bonus_type == DMU_OT_ZNODE);
1376 #endif
1377 	/* find handle, if it exists */
1378 	/* if one doesn't exist then create a new one, and initialize it */
1379 
1380 	if (hdl_type == SA_HDL_SHARED)
1381 		handle = dmu_buf_get_user(db);
1382 
1383 	if (handle == NULL) {
1384 		sa_handle_t *winner = NULL;
1385 
1386 		handle = kmem_cache_alloc(sa_cache, KM_SLEEP);
1387 		handle->sa_dbu.dbu_evict_func_sync = NULL;
1388 		handle->sa_dbu.dbu_evict_func_async = NULL;
1389 		handle->sa_userp = userp;
1390 		handle->sa_bonus = db;
1391 		handle->sa_os = os;
1392 		handle->sa_spill = NULL;
1393 		handle->sa_bonus_tab = NULL;
1394 		handle->sa_spill_tab = NULL;
1395 
1396 		error = sa_build_index(handle, SA_BONUS);
1397 
1398 		if (hdl_type == SA_HDL_SHARED) {
1399 			dmu_buf_init_user(&handle->sa_dbu, sa_evict_sync, NULL,
1400 			    NULL);
1401 			winner = dmu_buf_set_user_ie(db, &handle->sa_dbu);
1402 		}
1403 
1404 		if (winner != NULL) {
1405 			kmem_cache_free(sa_cache, handle);
1406 			handle = winner;
1407 		}
1408 	}
1409 	*handlepp = handle;
1410 
1411 	return (error);
1412 }
1413 
1414 int
1415 sa_handle_get(objset_t *objset, uint64_t objid, void *userp,
1416     sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1417 {
1418 	dmu_buf_t *db;
1419 	int error;
1420 
1421 	if (error = dmu_bonus_hold(objset, objid, NULL, &db))
1422 		return (error);
1423 
1424 	return (sa_handle_get_from_db(objset, db, userp, hdl_type,
1425 	    handlepp));
1426 }
1427 
1428 int
1429 sa_buf_hold(objset_t *objset, uint64_t obj_num, void *tag, dmu_buf_t **db)
1430 {
1431 	return (dmu_bonus_hold(objset, obj_num, tag, db));
1432 }
1433 
1434 void
1435 sa_buf_rele(dmu_buf_t *db, void *tag)
1436 {
1437 	dmu_buf_rele(db, tag);
1438 }
1439 
1440 int
1441 sa_lookup_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count)
1442 {
1443 	ASSERT(hdl);
1444 	ASSERT(MUTEX_HELD(&hdl->sa_lock));
1445 	return (sa_attr_op(hdl, bulk, count, SA_LOOKUP, NULL));
1446 }
1447 
1448 static int
1449 sa_lookup_locked(sa_handle_t *hdl, sa_attr_type_t attr, void *buf,
1450     uint32_t buflen)
1451 {
1452 	int error;
1453 	sa_bulk_attr_t bulk;
1454 
1455 	bulk.sa_attr = attr;
1456 	bulk.sa_data = buf;
1457 	bulk.sa_length = buflen;
1458 	bulk.sa_data_func = NULL;
1459 
1460 	ASSERT(hdl);
1461 	error = sa_lookup_impl(hdl, &bulk, 1);
1462 	return (error);
1463 }
1464 
1465 int
1466 sa_lookup(sa_handle_t *hdl, sa_attr_type_t attr, void *buf, uint32_t buflen)
1467 {
1468 	int error;
1469 
1470 	mutex_enter(&hdl->sa_lock);
1471 	error = sa_lookup_locked(hdl, attr, buf, buflen);
1472 	mutex_exit(&hdl->sa_lock);
1473 
1474 	return (error);
1475 }
1476 
1477 #ifdef _KERNEL
1478 int
1479 sa_lookup_uio(sa_handle_t *hdl, sa_attr_type_t attr, uio_t *uio)
1480 {
1481 	int error;
1482 	sa_bulk_attr_t bulk;
1483 
1484 	bulk.sa_data = NULL;
1485 	bulk.sa_attr = attr;
1486 	bulk.sa_data_func = NULL;
1487 
1488 	ASSERT(hdl);
1489 
1490 	mutex_enter(&hdl->sa_lock);
1491 	if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) == 0) {
1492 		error = uiomove((void *)bulk.sa_addr, MIN(bulk.sa_size,
1493 		    uio->uio_resid), UIO_READ, uio);
1494 	}
1495 	mutex_exit(&hdl->sa_lock);
1496 	return (error);
1497 
1498 }
1499 
1500 /*
1501  * For the existing object that is upgraded from old system, its ondisk layout
1502  * has no slot for the project ID attribute. But quota accounting logic needs
1503  * to access related slots by offset directly. So we need to adjust these old
1504  * objects' layout to make the project ID to some unified and fixed offset.
1505  */
1506 int
1507 sa_add_projid(sa_handle_t *hdl, dmu_tx_t *tx, uint64_t projid)
1508 {
1509 	znode_t *zp = sa_get_userdata(hdl);
1510 	dmu_buf_t *db = sa_get_db(hdl);
1511 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1512 	int count = 0, err = 0;
1513 	sa_bulk_attr_t *bulk, *attrs;
1514 	zfs_acl_locator_cb_t locate = { 0 };
1515 	uint64_t uid, gid, mode, rdev, xattr = 0, parent, gen, links;
1516 	uint64_t crtime[2], mtime[2], ctime[2], atime[2];
1517 	zfs_acl_phys_t znode_acl = { 0 };
1518 	char scanstamp[AV_SCANSTAMP_SZ];
1519 
1520 	if (zp->z_acl_cached == NULL) {
1521 		zfs_acl_t *aclp;
1522 
1523 		mutex_enter(&zp->z_acl_lock);
1524 		err = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1525 		mutex_exit(&zp->z_acl_lock);
1526 		if (err != 0 && err != ENOENT)
1527 			return (err);
1528 	}
1529 
1530 	bulk = kmem_zalloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
1531 	attrs = kmem_zalloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
1532 	mutex_enter(&hdl->sa_lock);
1533 	mutex_enter(&zp->z_lock);
1534 
1535 	err = sa_lookup_locked(hdl, SA_ZPL_PROJID(zfsvfs), &projid,
1536 	    sizeof (uint64_t));
1537 	if (unlikely(err == 0))
1538 		/* Someone has added project ID attr by race. */
1539 		err = EEXIST;
1540 	if (err != ENOENT)
1541 		goto out;
1542 
1543 	/* First do a bulk query of the attributes that aren't cached */
1544 	if (zp->z_is_sa) {
1545 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1546 		    &mode, 8);
1547 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1548 		    &gen, 8);
1549 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1550 		    &uid, 8);
1551 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1552 		    &gid, 8);
1553 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
1554 		    &parent, 8);
1555 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1556 		    &atime, 16);
1557 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1558 		    &mtime, 16);
1559 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1560 		    &ctime, 16);
1561 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1562 		    &crtime, 16);
1563 		if (S_ISBLK(zp->z_mode) || S_ISCHR(zp->z_mode))
1564 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1565 			    &rdev, 8);
1566 	} else {
1567 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1568 		    &atime, 16);
1569 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1570 		    &mtime, 16);
1571 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1572 		    &ctime, 16);
1573 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1574 		    &crtime, 16);
1575 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1576 		    &gen, 8);
1577 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1578 		    &mode, 8);
1579 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
1580 		    &parent, 8);
1581 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL,
1582 		    &xattr, 8);
1583 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1584 		    &rdev, 8);
1585 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1586 		    &uid, 8);
1587 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1588 		    &gid, 8);
1589 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1590 		    &znode_acl, 88);
1591 	}
1592 	err = sa_bulk_lookup_locked(hdl, bulk, count);
1593 	if (err != 0)
1594 		goto out;
1595 
1596 	err = sa_lookup_locked(hdl, SA_ZPL_XATTR(zfsvfs), &xattr, 8);
1597 	if (err != 0 && err != ENOENT)
1598 		goto out;
1599 
1600 	zp->z_projid = projid;
1601 	zp->z_pflags |= ZFS_PROJID;
1602 	links = zp->z_links;
1603 	count = 0;
1604 	err = 0;
1605 
1606 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
1607 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
1608 	    &zp->z_size, 8);
1609 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_GEN(zfsvfs), NULL, &gen, 8);
1610 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
1611 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
1612 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
1613 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1614 	    &zp->z_pflags, 8);
1615 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
1616 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
1617 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
1618 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1619 	    &crtime, 16);
1620 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
1621 	SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_PROJID(zfsvfs), NULL, &projid, 8);
1622 
1623 	if (S_ISBLK(zp->z_mode) || S_ISCHR(zp->z_mode))
1624 		SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
1625 		    &rdev, 8);
1626 
1627 	if (zp->z_acl_cached != NULL) {
1628 		SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
1629 		    &zp->z_acl_cached->z_acl_count, 8);
1630 		if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
1631 			zfs_acl_xform(zp, zp->z_acl_cached, CRED());
1632 		locate.cb_aclp = zp->z_acl_cached;
1633 		SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
1634 		    zfs_acl_data_locator, &locate,
1635 		    zp->z_acl_cached->z_acl_bytes);
1636 	}
1637 
1638 	if (xattr)
1639 		SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_XATTR(zfsvfs), NULL,
1640 		    &xattr, 8);
1641 
1642 	if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
1643 		bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
1644 		    scanstamp, AV_SCANSTAMP_SZ);
1645 		SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_SCANSTAMP(zfsvfs), NULL,
1646 		    scanstamp, AV_SCANSTAMP_SZ);
1647 		zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
1648 	}
1649 
1650 	VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
1651 	VERIFY(sa_replace_all_by_template_locked(hdl, attrs, count, tx) == 0);
1652 	if (znode_acl.z_acl_extern_obj) {
1653 		VERIFY(0 == dmu_object_free(zfsvfs->z_os,
1654 		    znode_acl.z_acl_extern_obj, tx));
1655 	}
1656 
1657 	zp->z_is_sa = B_TRUE;
1658 
1659 out:
1660 	mutex_exit(&zp->z_lock);
1661 	mutex_exit(&hdl->sa_lock);
1662 	kmem_free(attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
1663 	kmem_free(bulk, sizeof (sa_bulk_attr_t) * ZPL_END);
1664 	return (err);
1665 }
1666 #endif
1667 
1668 static sa_idx_tab_t *
1669 sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype, sa_hdr_phys_t *hdr)
1670 {
1671 	sa_idx_tab_t *idx_tab;
1672 	sa_os_t *sa = os->os_sa;
1673 	sa_lot_t *tb, search;
1674 	avl_index_t loc;
1675 
1676 	/*
1677 	 * Deterimine layout number.  If SA node and header == 0 then
1678 	 * force the index table to the dummy "1" empty layout.
1679 	 *
1680 	 * The layout number would only be zero for a newly created file
1681 	 * that has not added any attributes yet, or with crypto enabled which
1682 	 * doesn't write any attributes to the bonus buffer.
1683 	 */
1684 
1685 	search.lot_num = SA_LAYOUT_NUM(hdr, bonustype);
1686 
1687 	tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
1688 
1689 	/* Verify header size is consistent with layout information */
1690 	ASSERT(tb);
1691 	ASSERT(IS_SA_BONUSTYPE(bonustype) &&
1692 	    SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb) || !IS_SA_BONUSTYPE(bonustype) ||
1693 	    (IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0));
1694 
1695 	/*
1696 	 * See if any of the already existing TOC entries can be reused?
1697 	 */
1698 
1699 	for (idx_tab = list_head(&tb->lot_idx_tab); idx_tab;
1700 	    idx_tab = list_next(&tb->lot_idx_tab, idx_tab)) {
1701 		boolean_t valid_idx = B_TRUE;
1702 		int i;
1703 
1704 		if (tb->lot_var_sizes != 0 &&
1705 		    idx_tab->sa_variable_lengths != NULL) {
1706 			for (i = 0; i != tb->lot_var_sizes; i++) {
1707 				if (hdr->sa_lengths[i] !=
1708 				    idx_tab->sa_variable_lengths[i]) {
1709 					valid_idx = B_FALSE;
1710 					break;
1711 				}
1712 			}
1713 		}
1714 		if (valid_idx) {
1715 			sa_idx_tab_hold(os, idx_tab);
1716 			return (idx_tab);
1717 		}
1718 	}
1719 
1720 	/* No such luck, create a new entry */
1721 	idx_tab = kmem_zalloc(sizeof (sa_idx_tab_t), KM_SLEEP);
1722 	idx_tab->sa_idx_tab =
1723 	    kmem_zalloc(sizeof (uint32_t) * sa->sa_num_attrs, KM_SLEEP);
1724 	idx_tab->sa_layout = tb;
1725 	zfs_refcount_create(&idx_tab->sa_refcount);
1726 	if (tb->lot_var_sizes)
1727 		idx_tab->sa_variable_lengths = kmem_alloc(sizeof (uint16_t) *
1728 		    tb->lot_var_sizes, KM_SLEEP);
1729 
1730 	sa_attr_iter(os, hdr, bonustype, sa_build_idx_tab,
1731 	    tb, idx_tab);
1732 	sa_idx_tab_hold(os, idx_tab);   /* one hold for consumer */
1733 	sa_idx_tab_hold(os, idx_tab);	/* one for layout */
1734 	list_insert_tail(&tb->lot_idx_tab, idx_tab);
1735 	return (idx_tab);
1736 }
1737 
1738 void
1739 sa_default_locator(void **dataptr, uint32_t *len, uint32_t total_len,
1740     boolean_t start, void *userdata)
1741 {
1742 	ASSERT(start);
1743 
1744 	*dataptr = userdata;
1745 	*len = total_len;
1746 }
1747 
1748 static void
1749 sa_attr_register_sync(sa_handle_t *hdl, dmu_tx_t *tx)
1750 {
1751 	uint64_t attr_value = 0;
1752 	sa_os_t *sa = hdl->sa_os->os_sa;
1753 	sa_attr_table_t *tb = sa->sa_attr_table;
1754 	int i;
1755 
1756 	mutex_enter(&sa->sa_lock);
1757 
1758 	if (!sa->sa_need_attr_registration || sa->sa_master_obj == 0) {
1759 		mutex_exit(&sa->sa_lock);
1760 		return;
1761 	}
1762 
1763 	if (sa->sa_reg_attr_obj == 0) {
1764 		sa->sa_reg_attr_obj = zap_create_link(hdl->sa_os,
1765 		    DMU_OT_SA_ATTR_REGISTRATION,
1766 		    sa->sa_master_obj, SA_REGISTRY, tx);
1767 	}
1768 	for (i = 0; i != sa->sa_num_attrs; i++) {
1769 		if (sa->sa_attr_table[i].sa_registered)
1770 			continue;
1771 		ATTR_ENCODE(attr_value, tb[i].sa_attr, tb[i].sa_length,
1772 		    tb[i].sa_byteswap);
1773 		VERIFY(0 == zap_update(hdl->sa_os, sa->sa_reg_attr_obj,
1774 		    tb[i].sa_name, 8, 1, &attr_value, tx));
1775 		tb[i].sa_registered = B_TRUE;
1776 	}
1777 	sa->sa_need_attr_registration = B_FALSE;
1778 	mutex_exit(&sa->sa_lock);
1779 }
1780 
1781 /*
1782  * Replace all attributes with attributes specified in template.
1783  * If dnode had a spill buffer then those attributes will be
1784  * also be replaced, possibly with just an empty spill block
1785  *
1786  * This interface is intended to only be used for bulk adding of
1787  * attributes for a new file.  It will also be used by the ZPL
1788  * when converting and old formatted znode to native SA support.
1789  */
1790 int
1791 sa_replace_all_by_template_locked(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
1792     int attr_count, dmu_tx_t *tx)
1793 {
1794 	sa_os_t *sa = hdl->sa_os->os_sa;
1795 
1796 	if (sa->sa_need_attr_registration)
1797 		sa_attr_register_sync(hdl, tx);
1798 	return (sa_build_layouts(hdl, attr_desc, attr_count, tx));
1799 }
1800 
1801 int
1802 sa_replace_all_by_template(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
1803     int attr_count, dmu_tx_t *tx)
1804 {
1805 	int error;
1806 
1807 	mutex_enter(&hdl->sa_lock);
1808 	error = sa_replace_all_by_template_locked(hdl, attr_desc,
1809 	    attr_count, tx);
1810 	mutex_exit(&hdl->sa_lock);
1811 	return (error);
1812 }
1813 
1814 /*
1815  * Add/remove a single attribute or replace a variable-sized attribute value
1816  * with a value of a different size, and then rewrite the entire set
1817  * of attributes.
1818  * Same-length attribute value replacement (including fixed-length attributes)
1819  * is handled more efficiently by the upper layers.
1820  */
1821 static int
1822 sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
1823     sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
1824     uint16_t buflen, dmu_tx_t *tx)
1825 {
1826 	sa_os_t *sa = hdl->sa_os->os_sa;
1827 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)hdl->sa_bonus;
1828 	dnode_t *dn;
1829 	sa_bulk_attr_t *attr_desc;
1830 	void *old_data[2];
1831 	int bonus_attr_count = 0;
1832 	int bonus_data_size = 0;
1833 	int spill_data_size = 0;
1834 	int spill_attr_count = 0;
1835 	int error;
1836 	uint16_t length, reg_length;
1837 	int i, j, k, length_idx;
1838 	sa_hdr_phys_t *hdr;
1839 	sa_idx_tab_t *idx_tab;
1840 	int attr_count;
1841 	int count;
1842 
1843 	ASSERT(MUTEX_HELD(&hdl->sa_lock));
1844 
1845 	/* First make of copy of the old data */
1846 
1847 	DB_DNODE_ENTER(db);
1848 	dn = DB_DNODE(db);
1849 	if (dn->dn_bonuslen != 0) {
1850 		bonus_data_size = hdl->sa_bonus->db_size;
1851 		old_data[0] = kmem_alloc(bonus_data_size, KM_SLEEP);
1852 		bcopy(hdl->sa_bonus->db_data, old_data[0],
1853 		    hdl->sa_bonus->db_size);
1854 		bonus_attr_count = hdl->sa_bonus_tab->sa_layout->lot_attr_count;
1855 	} else {
1856 		old_data[0] = NULL;
1857 	}
1858 	DB_DNODE_EXIT(db);
1859 
1860 	/* Bring spill buffer online if it isn't currently */
1861 
1862 	if ((error = sa_get_spill(hdl)) == 0) {
1863 		spill_data_size = hdl->sa_spill->db_size;
1864 		old_data[1] = kmem_alloc(spill_data_size, KM_SLEEP);
1865 		bcopy(hdl->sa_spill->db_data, old_data[1],
1866 		    hdl->sa_spill->db_size);
1867 		spill_attr_count =
1868 		    hdl->sa_spill_tab->sa_layout->lot_attr_count;
1869 	} else if (error && error != ENOENT) {
1870 		if (old_data[0])
1871 			kmem_free(old_data[0], bonus_data_size);
1872 		return (error);
1873 	} else {
1874 		old_data[1] = NULL;
1875 	}
1876 
1877 	/* build descriptor of all attributes */
1878 
1879 	attr_count = bonus_attr_count + spill_attr_count;
1880 	if (action == SA_ADD)
1881 		attr_count++;
1882 	else if (action == SA_REMOVE)
1883 		attr_count--;
1884 
1885 	attr_desc = kmem_zalloc(sizeof (sa_bulk_attr_t) * attr_count, KM_SLEEP);
1886 
1887 	/*
1888 	 * loop through bonus and spill buffer if it exists, and
1889 	 * build up new attr_descriptor to reset the attributes
1890 	 */
1891 	k = j = 0;
1892 	count = bonus_attr_count;
1893 	hdr = SA_GET_HDR(hdl, SA_BONUS);
1894 	idx_tab = SA_IDX_TAB_GET(hdl, SA_BONUS);
1895 	for (; k != 2; k++) {
1896 		/*
1897 		 * Iterate over each attribute in layout.  Fetch the
1898 		 * size of variable-length attributes needing rewrite
1899 		 * from sa_lengths[].
1900 		 */
1901 		for (i = 0, length_idx = 0; i != count; i++) {
1902 			sa_attr_type_t attr;
1903 
1904 			attr = idx_tab->sa_layout->lot_attrs[i];
1905 			reg_length = SA_REGISTERED_LEN(sa, attr);
1906 			if (reg_length == 0) {
1907 				length = hdr->sa_lengths[length_idx];
1908 				length_idx++;
1909 			} else {
1910 				length = reg_length;
1911 			}
1912 			if (attr == newattr) {
1913 				/*
1914 				 * There is nothing to do for SA_REMOVE,
1915 				 * so it is just skipped.
1916 				 */
1917 				if (action == SA_REMOVE)
1918 					continue;
1919 
1920 				/*
1921 				 * Duplicate attributes are not allowed, so the
1922 				 * action can not be SA_ADD here.
1923 				 */
1924 				ASSERT3S(action, ==, SA_REPLACE);
1925 
1926 				/*
1927 				 * Only a variable-sized attribute can be
1928 				 * replaced here, and its size must be changing.
1929 				 */
1930 				ASSERT3U(reg_length, ==, 0);
1931 				ASSERT3U(length, !=, buflen);
1932 				SA_ADD_BULK_ATTR(attr_desc, j, attr,
1933 				    locator, datastart, buflen);
1934 			} else {
1935 				SA_ADD_BULK_ATTR(attr_desc, j, attr,
1936 				    NULL, (void *)
1937 				    (TOC_OFF(idx_tab->sa_idx_tab[attr]) +
1938 				    (uintptr_t)old_data[k]), length);
1939 			}
1940 		}
1941 		if (k == 0 && hdl->sa_spill) {
1942 			hdr = SA_GET_HDR(hdl, SA_SPILL);
1943 			idx_tab = SA_IDX_TAB_GET(hdl, SA_SPILL);
1944 			count = spill_attr_count;
1945 		} else {
1946 			break;
1947 		}
1948 	}
1949 	if (action == SA_ADD) {
1950 		reg_length = SA_REGISTERED_LEN(sa, newattr);
1951 		IMPLY(reg_length != 0, reg_length == buflen);
1952 		SA_ADD_BULK_ATTR(attr_desc, j, newattr, locator,
1953 		    datastart, buflen);
1954 	}
1955 	ASSERT3U(j, ==, attr_count);
1956 
1957 	error = sa_build_layouts(hdl, attr_desc, attr_count, tx);
1958 
1959 	if (old_data[0])
1960 		kmem_free(old_data[0], bonus_data_size);
1961 	if (old_data[1])
1962 		kmem_free(old_data[1], spill_data_size);
1963 	kmem_free(attr_desc, sizeof (sa_bulk_attr_t) * attr_count);
1964 
1965 	return (error);
1966 }
1967 
1968 static int
1969 sa_bulk_update_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
1970     dmu_tx_t *tx)
1971 {
1972 	int error;
1973 	sa_os_t *sa = hdl->sa_os->os_sa;
1974 	dmu_object_type_t bonustype;
1975 
1976 	bonustype = SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl, SA_BONUS));
1977 
1978 	ASSERT(hdl);
1979 	ASSERT(MUTEX_HELD(&hdl->sa_lock));
1980 
1981 	/* sync out registration table if necessary */
1982 	if (sa->sa_need_attr_registration)
1983 		sa_attr_register_sync(hdl, tx);
1984 
1985 	error = sa_attr_op(hdl, bulk, count, SA_UPDATE, tx);
1986 	if (error == 0 && !IS_SA_BONUSTYPE(bonustype) && sa->sa_update_cb)
1987 		sa->sa_update_cb(hdl, tx);
1988 
1989 	return (error);
1990 }
1991 
1992 /*
1993  * update or add new attribute
1994  */
1995 int
1996 sa_update(sa_handle_t *hdl, sa_attr_type_t type,
1997     void *buf, uint32_t buflen, dmu_tx_t *tx)
1998 {
1999 	int error;
2000 	sa_bulk_attr_t bulk;
2001 
2002 	bulk.sa_attr = type;
2003 	bulk.sa_data_func = NULL;
2004 	bulk.sa_length = buflen;
2005 	bulk.sa_data = buf;
2006 
2007 	mutex_enter(&hdl->sa_lock);
2008 	error = sa_bulk_update_impl(hdl, &bulk, 1, tx);
2009 	mutex_exit(&hdl->sa_lock);
2010 	return (error);
2011 }
2012 
2013 int
2014 sa_update_from_cb(sa_handle_t *hdl, sa_attr_type_t attr,
2015     uint32_t buflen, sa_data_locator_t *locator, void *userdata, dmu_tx_t *tx)
2016 {
2017 	int error;
2018 	sa_bulk_attr_t bulk;
2019 
2020 	bulk.sa_attr = attr;
2021 	bulk.sa_data = userdata;
2022 	bulk.sa_data_func = locator;
2023 	bulk.sa_length = buflen;
2024 
2025 	mutex_enter(&hdl->sa_lock);
2026 	error = sa_bulk_update_impl(hdl, &bulk, 1, tx);
2027 	mutex_exit(&hdl->sa_lock);
2028 	return (error);
2029 }
2030 
2031 /*
2032  * Return size of an attribute
2033  */
2034 
2035 int
2036 sa_size(sa_handle_t *hdl, sa_attr_type_t attr, int *size)
2037 {
2038 	sa_bulk_attr_t bulk;
2039 	int error;
2040 
2041 	bulk.sa_data = NULL;
2042 	bulk.sa_attr = attr;
2043 	bulk.sa_data_func = NULL;
2044 
2045 	ASSERT(hdl);
2046 	mutex_enter(&hdl->sa_lock);
2047 	if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) != 0) {
2048 		mutex_exit(&hdl->sa_lock);
2049 		return (error);
2050 	}
2051 	*size = bulk.sa_size;
2052 
2053 	mutex_exit(&hdl->sa_lock);
2054 	return (0);
2055 }
2056 
2057 int
2058 sa_bulk_lookup_locked(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
2059 {
2060 	ASSERT(hdl);
2061 	ASSERT(MUTEX_HELD(&hdl->sa_lock));
2062 	return (sa_lookup_impl(hdl, attrs, count));
2063 }
2064 
2065 int
2066 sa_bulk_lookup(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
2067 {
2068 	int error;
2069 
2070 	ASSERT(hdl);
2071 	mutex_enter(&hdl->sa_lock);
2072 	error = sa_bulk_lookup_locked(hdl, attrs, count);
2073 	mutex_exit(&hdl->sa_lock);
2074 	return (error);
2075 }
2076 
2077 int
2078 sa_bulk_update(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count, dmu_tx_t *tx)
2079 {
2080 	int error;
2081 
2082 	ASSERT(hdl);
2083 	mutex_enter(&hdl->sa_lock);
2084 	error = sa_bulk_update_impl(hdl, attrs, count, tx);
2085 	mutex_exit(&hdl->sa_lock);
2086 	return (error);
2087 }
2088 
2089 int
2090 sa_remove(sa_handle_t *hdl, sa_attr_type_t attr, dmu_tx_t *tx)
2091 {
2092 	int error;
2093 
2094 	mutex_enter(&hdl->sa_lock);
2095 	error = sa_modify_attrs(hdl, attr, SA_REMOVE, NULL,
2096 	    NULL, 0, tx);
2097 	mutex_exit(&hdl->sa_lock);
2098 	return (error);
2099 }
2100 
2101 void
2102 sa_object_info(sa_handle_t *hdl, dmu_object_info_t *doi)
2103 {
2104 	dmu_object_info_from_db((dmu_buf_t *)hdl->sa_bonus, doi);
2105 }
2106 
2107 void
2108 sa_object_size(sa_handle_t *hdl, uint32_t *blksize, u_longlong_t *nblocks)
2109 {
2110 	dmu_object_size_from_db((dmu_buf_t *)hdl->sa_bonus,
2111 	    blksize, nblocks);
2112 }
2113 
2114 void
2115 sa_set_userp(sa_handle_t *hdl, void *ptr)
2116 {
2117 	hdl->sa_userp = ptr;
2118 }
2119 
2120 dmu_buf_t *
2121 sa_get_db(sa_handle_t *hdl)
2122 {
2123 	return ((dmu_buf_t *)hdl->sa_bonus);
2124 }
2125 
2126 void *
2127 sa_get_userdata(sa_handle_t *hdl)
2128 {
2129 	return (hdl->sa_userp);
2130 }
2131 
2132 void
2133 sa_register_update_callback_locked(objset_t *os, sa_update_cb_t *func)
2134 {
2135 	ASSERT(MUTEX_HELD(&os->os_sa->sa_lock));
2136 	os->os_sa->sa_update_cb = func;
2137 }
2138 
2139 void
2140 sa_register_update_callback(objset_t *os, sa_update_cb_t *func)
2141 {
2142 
2143 	mutex_enter(&os->os_sa->sa_lock);
2144 	sa_register_update_callback_locked(os, func);
2145 	mutex_exit(&os->os_sa->sa_lock);
2146 }
2147 
2148 uint64_t
2149 sa_handle_object(sa_handle_t *hdl)
2150 {
2151 	return (hdl->sa_bonus->db_object);
2152 }
2153 
2154 boolean_t
2155 sa_enabled(objset_t *os)
2156 {
2157 	return (os->os_sa == NULL);
2158 }
2159 
2160 int
2161 sa_set_sa_object(objset_t *os, uint64_t sa_object)
2162 {
2163 	sa_os_t *sa = os->os_sa;
2164 
2165 	if (sa->sa_master_obj)
2166 		return (1);
2167 
2168 	sa->sa_master_obj = sa_object;
2169 
2170 	return (0);
2171 }
2172 
2173 int
2174 sa_hdrsize(void *arg)
2175 {
2176 	sa_hdr_phys_t *hdr = arg;
2177 
2178 	return (SA_HDR_SIZE(hdr));
2179 }
2180 
2181 void
2182 sa_handle_lock(sa_handle_t *hdl)
2183 {
2184 	ASSERT(hdl);
2185 	mutex_enter(&hdl->sa_lock);
2186 }
2187 
2188 void
2189 sa_handle_unlock(sa_handle_t *hdl)
2190 {
2191 	ASSERT(hdl);
2192 	mutex_exit(&hdl->sa_lock);
2193 }
2194