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