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.
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.
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]
22 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
25 #include <sys/zfs_context.h>
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/systm.h>
29 #include <sys/sysmacros.h>
31 #include <sys/dmu_impl.h>
32 #include <sys/dmu_objset.h>
34 #include <sys/dnode.h>
37 #include <sys/sunddi.h>
38 #include <sys/sa_impl.h>
39 #include <sys/dnode.h>
40 #include <sys/errno.h>
41 #include <sys/zfs_context.h>
44 * ZFS System attributes:
46 * A generic mechanism to allow for arbitrary attributes
47 * to be stored in a dnode. The data will be stored in the bonus buffer of
48 * the dnode and if necessary a special "spill" block will be used to handle
49 * overflow situations. The spill block will be sized to fit the data
50 * from 512 - 128K. When a spill block is used the BP (blkptr_t) for the
51 * spill block is stored at the end of the current bonus buffer. Any
52 * attributes that would be in the way of the blkptr_t will be relocated
53 * into the spill block.
55 * Attribute registration:
57 * Stored persistently on a per dataset basis
58 * a mapping between attribute "string" names and their actual attribute
59 * numeric values, length, and byteswap function. The names are only used
60 * during registration. All attributes are known by their unique attribute
61 * id value. If an attribute can have a variable size then the value
62 * 0 will be used to indicate this.
66 * Attribute layouts are a way to compactly store multiple attributes, but
67 * without taking the overhead associated with managing each attribute
68 * individually. Since you will typically have the same set of attributes
69 * stored in the same order a single table will be used to represent that
70 * layout. The ZPL for example will usually have only about 10 different
71 * layouts (regular files, device files, symlinks,
72 * regular files + scanstamp, files/dir with extended attributes, and then
73 * you have the possibility of all of those minus ACL, because it would
74 * be kicked out into the spill block)
76 * Layouts are simply an array of the attributes and their
77 * ordering i.e. [0, 1, 4, 5, 2]
79 * Each distinct layout is given a unique layout number and that is whats
80 * stored in the header at the beginning of the SA data buffer.
82 * A layout only covers a single dbuf (bonus or spill). If a set of
83 * attributes is split up between the bonus buffer and a spill buffer then
84 * two different layouts will be used. This allows us to byteswap the
85 * spill without looking at the bonus buffer and keeps the on disk format of
86 * the bonus and spill buffer the same.
88 * Adding a single attribute will cause the entire set of attributes to
89 * be rewritten and could result in a new layout number being constructed
90 * as part of the rewrite if no such layout exists for the new set of
91 * attribues. The new attribute will be appended to the end of the already
92 * existing attributes.
94 * Both the attribute registration and attribute layout information are
95 * stored in normal ZAP attributes. Their should be a small number of
96 * known layouts and the set of attributes is assumed to typically be quite
99 * The registered attributes and layout "table" information is maintained
100 * in core and a special "sa_os_t" is attached to the objset_t.
102 * A special interface is provided to allow for quickly applying
103 * a large set of attributes at once. sa_replace_all_by_template() is
104 * used to set an array of attributes. This is used by the ZPL when
105 * creating a brand new file. The template that is passed into the function
106 * specifies the attribute, size for variable length attributes, location of
107 * data and special "data locator" function if the data isn't in a contiguous
110 * Byteswap implications:
111 * Since the SA attributes are not entirely self describing we can't do
112 * the normal byteswap processing. The special ZAP layout attribute and
113 * attribute registration attributes define the byteswap function and the
114 * size of the attributes, unless it is variable sized.
115 * The normal ZFS byteswapping infrastructure assumes you don't need
116 * to read any objects in order to do the necessary byteswapping. Whereas
117 * SA attributes can only be properly byteswapped if the dataset is opened
118 * and the layout/attribute ZAP attributes are available. Because of this
119 * the SA attributes will be byteswapped when they are first accessed by
120 * the SA code that will read the SA data.
123 typedef void (sa_iterfunc_t)(void *hdr, void *addr, sa_attr_type_t,
124 uint16_t length, int length_idx, boolean_t, void *userp);
126 static int sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype);
127 static void sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab);
128 static void *sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype,
130 static void sa_idx_tab_rele(objset_t *os, void *arg);
131 static void sa_copy_data(sa_data_locator_t *func, void *start, void *target,
133 static int sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
134 sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
135 uint16_t buflen, dmu_tx_t *tx);
137 arc_byteswap_func_t *sa_bswap_table[] = {
138 byteswap_uint64_array,
139 byteswap_uint32_array,
140 byteswap_uint16_array,
141 byteswap_uint8_array,
145 #define SA_COPY_DATA(f, s, t, l) \
149 *(uint64_t *)t = *(uint64_t *)s; \
150 } else if (l == 16) { \
151 *(uint64_t *)t = *(uint64_t *)s; \
152 *(uint64_t *)((uintptr_t)t + 8) = \
153 *(uint64_t *)((uintptr_t)s + 8); \
158 sa_copy_data(f, s, t, l); \
162 * This table is fixed and cannot be changed. Its purpose is to
163 * allow the SA code to work with both old/new ZPL file systems.
164 * It contains the list of legacy attributes. These attributes aren't
165 * stored in the "attribute" registry zap objects, since older ZPL file systems
166 * won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
167 * use this static table.
169 sa_attr_reg_t sa_legacy_attrs[] = {
170 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
171 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
172 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
173 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
174 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
175 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
176 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
177 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
178 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
179 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
180 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
181 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
182 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
183 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
184 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
185 {"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
190 * This is only used for objects of type DMU_OT_ZNODE
192 sa_attr_type_t sa_legacy_zpl_layout[] = {
193 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
197 * Special dummy layout used for buffers with no attributes.
200 sa_attr_type_t sa_dummy_zpl_layout[] = { 0 };
202 static int sa_legacy_attr_count = 16;
203 static kmem_cache_t *sa_cache = NULL;
207 sa_cache_constructor(void *buf, void *unused, int kmflag)
209 sa_handle_t *hdl = buf;
211 hdl->sa_bonus_tab = NULL;
212 hdl->sa_spill_tab = NULL;
214 hdl->sa_userp = NULL;
215 hdl->sa_bonus = NULL;
216 hdl->sa_spill = NULL;
217 mutex_init(&hdl->sa_lock, NULL, MUTEX_DEFAULT, NULL);
223 sa_cache_destructor(void *buf, void *unused)
225 sa_handle_t *hdl = buf;
226 mutex_destroy(&hdl->sa_lock);
232 sa_cache = kmem_cache_create("sa_cache",
233 sizeof (sa_handle_t), 0, sa_cache_constructor,
234 sa_cache_destructor, NULL, NULL, NULL, 0);
241 kmem_cache_destroy(sa_cache);
245 layout_num_compare(const void *arg1, const void *arg2)
247 const sa_lot_t *node1 = arg1;
248 const sa_lot_t *node2 = arg2;
250 if (node1->lot_num > node2->lot_num)
252 else if (node1->lot_num < node2->lot_num)
258 layout_hash_compare(const void *arg1, const void *arg2)
260 const sa_lot_t *node1 = arg1;
261 const sa_lot_t *node2 = arg2;
263 if (node1->lot_hash > node2->lot_hash)
265 if (node1->lot_hash < node2->lot_hash)
267 if (node1->lot_instance > node2->lot_instance)
269 if (node1->lot_instance < node2->lot_instance)
275 sa_layout_equal(sa_lot_t *tbf, sa_attr_type_t *attrs, int count)
279 if (count != tbf->lot_attr_count)
282 for (i = 0; i != count; i++) {
283 if (attrs[i] != tbf->lot_attrs[i])
289 #define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
292 sa_layout_info_hash(sa_attr_type_t *attrs, int attr_count)
295 uint64_t crc = -1ULL;
297 for (i = 0; i != attr_count; i++)
298 crc ^= SA_ATTR_HASH(attrs[i]);
304 sa_has_blkptr(sa_handle_t *hdl)
307 if (hdl->sa_spill == NULL) {
308 if ((rc = dmu_spill_hold_existing(hdl->sa_bonus, NULL,
309 &hdl->sa_spill)) == 0)
310 VERIFY(0 == sa_build_index(hdl, SA_SPILL));
315 return (rc == 0 ? B_TRUE : B_FALSE);
319 * Main attribute lookup/update function
320 * returns 0 for success or non zero for failures
322 * Operates on bulk array, first failure will abort further processing
325 sa_attr_op(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
326 sa_data_op_t data_op, dmu_tx_t *tx)
328 sa_os_t *sa = hdl->sa_os->os_sa;
331 sa_buf_type_t buftypes;
336 for (i = 0; i != count; i++) {
337 ASSERT(bulk[i].sa_attr <= hdl->sa_os->os_sa->sa_num_attrs);
339 bulk[i].sa_addr = NULL;
340 /* First check the bonus buffer */
342 if (hdl->sa_bonus_tab && TOC_ATTR_PRESENT(
343 hdl->sa_bonus_tab->sa_idx_tab[bulk[i].sa_attr])) {
344 SA_ATTR_INFO(sa, hdl->sa_bonus_tab,
345 SA_GET_HDR(hdl, SA_BONUS),
346 bulk[i].sa_attr, bulk[i], SA_BONUS, hdl);
347 if (tx && !(buftypes & SA_BONUS)) {
348 dmu_buf_will_dirty(hdl->sa_bonus, tx);
349 buftypes |= SA_BONUS;
352 if (bulk[i].sa_addr == NULL && sa_has_blkptr(hdl)) {
353 if (TOC_ATTR_PRESENT(
354 hdl->sa_spill_tab->sa_idx_tab[bulk[i].sa_attr])) {
355 SA_ATTR_INFO(sa, hdl->sa_spill_tab,
356 SA_GET_HDR(hdl, SA_SPILL),
357 bulk[i].sa_attr, bulk[i], SA_SPILL, hdl);
358 if (tx && !(buftypes & SA_SPILL) &&
359 bulk[i].sa_size == bulk[i].sa_length) {
360 dmu_buf_will_dirty(hdl->sa_spill, tx);
361 buftypes |= SA_SPILL;
367 if (bulk[i].sa_addr == NULL)
369 if (bulk[i].sa_data) {
370 SA_COPY_DATA(bulk[i].sa_data_func,
371 bulk[i].sa_addr, bulk[i].sa_data,
377 /* existing rewrite of attr */
378 if (bulk[i].sa_addr &&
379 bulk[i].sa_size == bulk[i].sa_length) {
380 SA_COPY_DATA(bulk[i].sa_data_func,
381 bulk[i].sa_data, bulk[i].sa_addr,
384 } else if (bulk[i].sa_addr) { /* attr size change */
385 error = sa_modify_attrs(hdl, bulk[i].sa_attr,
386 SA_REPLACE, bulk[i].sa_data_func,
387 bulk[i].sa_data, bulk[i].sa_length, tx);
388 } else { /* adding new attribute */
389 error = sa_modify_attrs(hdl, bulk[i].sa_attr,
390 SA_ADD, bulk[i].sa_data_func,
391 bulk[i].sa_data, bulk[i].sa_length, tx);
402 sa_add_layout_entry(objset_t *os, sa_attr_type_t *attrs, int attr_count,
403 uint64_t lot_num, uint64_t hash, boolean_t zapadd, dmu_tx_t *tx)
405 sa_os_t *sa = os->os_sa;
406 sa_lot_t *tb, *findtb;
410 ASSERT(MUTEX_HELD(&sa->sa_lock));
411 tb = kmem_zalloc(sizeof (sa_lot_t), KM_SLEEP);
412 tb->lot_attr_count = attr_count;
413 tb->lot_attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
415 bcopy(attrs, tb->lot_attrs, sizeof (sa_attr_type_t) * attr_count);
416 tb->lot_num = lot_num;
418 tb->lot_instance = 0;
423 if (sa->sa_layout_attr_obj == 0) {
425 sa->sa_layout_attr_obj = zap_create(os,
426 DMU_OT_SA_ATTR_LAYOUTS, DMU_OT_NONE, 0, tx);
427 error = zap_add(os, sa->sa_master_obj, SA_LAYOUTS, 8, 1,
428 &sa->sa_layout_attr_obj, tx);
429 ASSERT3U(error, ==, 0);
432 (void) snprintf(attr_name, sizeof (attr_name),
434 VERIFY(0 == zap_update(os, os->os_sa->sa_layout_attr_obj,
435 attr_name, 2, attr_count, attrs, tx));
438 list_create(&tb->lot_idx_tab, sizeof (sa_idx_tab_t),
439 offsetof(sa_idx_tab_t, sa_next));
441 for (i = 0; i != attr_count; i++) {
442 if (sa->sa_attr_table[tb->lot_attrs[i]].sa_length == 0)
446 avl_add(&sa->sa_layout_num_tree, tb);
448 /* verify we don't have a hash collision */
449 if ((findtb = avl_find(&sa->sa_layout_hash_tree, tb, &loc)) != NULL) {
450 for (; findtb && findtb->lot_hash == hash;
451 findtb = AVL_NEXT(&sa->sa_layout_hash_tree, findtb)) {
452 if (findtb->lot_instance != tb->lot_instance)
457 avl_add(&sa->sa_layout_hash_tree, tb);
462 sa_find_layout(objset_t *os, uint64_t hash, sa_attr_type_t *attrs,
463 int count, dmu_tx_t *tx, sa_lot_t **lot)
465 sa_lot_t *tb, tbsearch;
467 sa_os_t *sa = os->os_sa;
468 boolean_t found = B_FALSE;
470 mutex_enter(&sa->sa_lock);
471 tbsearch.lot_hash = hash;
472 tbsearch.lot_instance = 0;
473 tb = avl_find(&sa->sa_layout_hash_tree, &tbsearch, &loc);
475 for (; tb && tb->lot_hash == hash;
476 tb = AVL_NEXT(&sa->sa_layout_hash_tree, tb)) {
477 if (sa_layout_equal(tb, attrs, count) == 0) {
484 tb = sa_add_layout_entry(os, attrs, count,
485 avl_numnodes(&sa->sa_layout_num_tree), hash, B_TRUE, tx);
487 mutex_exit(&sa->sa_lock);
492 sa_resize_spill(sa_handle_t *hdl, uint32_t size, dmu_tx_t *tx)
498 blocksize = SPA_MINBLOCKSIZE;
499 } else if (size > SPA_MAXBLOCKSIZE) {
503 blocksize = P2ROUNDUP_TYPED(size, SPA_MINBLOCKSIZE, uint32_t);
506 error = dbuf_spill_set_blksz(hdl->sa_spill, blocksize, tx);
512 sa_copy_data(sa_data_locator_t *func, void *datastart, void *target, int buflen)
515 bcopy(datastart, target, buflen);
520 void *saptr = target;
525 while (bytes < buflen) {
526 func(&dataptr, &length, buflen, start, datastart);
527 bcopy(dataptr, saptr, length);
528 saptr = (void *)((caddr_t)saptr + length);
536 * Determine several different sizes
537 * first the sa header size
538 * the number of bytes to be stored
539 * if spill would occur the index in the attribute array is returned
541 * the boolean will_spill will be set when spilling is necessary. It
542 * is only set when the buftype is SA_BONUS
545 sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count,
546 dmu_buf_t *db, sa_buf_type_t buftype, int *index, int *total,
547 boolean_t *will_spill)
553 boolean_t done = B_FALSE;
555 if (buftype == SA_BONUS && sa->sa_force_spill) {
558 *will_spill = B_TRUE;
565 if (buftype == SA_BONUS)
566 *will_spill = B_FALSE;
568 hdrsize = (SA_BONUSTYPE_FROM_DB(db) == DMU_OT_ZNODE) ? 0 :
569 sizeof (sa_hdr_phys_t);
571 full_space = (buftype == SA_BONUS) ? DN_MAX_BONUSLEN : db->db_size;
573 for (i = 0; i != attr_count; i++) {
576 *total += attr_desc[i].sa_length;
580 is_var_sz = (SA_REGISTERED_LEN(sa, attr_desc[i].sa_attr) == 0);
585 if (is_var_sz && var_size > 1) {
586 if (P2ROUNDUP(hdrsize + sizeof (uint16_t), 8) +
587 *total < full_space) {
588 hdrsize += sizeof (uint16_t);
592 if (buftype == SA_BONUS)
593 *will_spill = B_TRUE;
599 * find index of where spill *could* occur.
600 * Then continue to count of remainder attribute
601 * space. The sum is used later for sizing bonus
604 if (buftype == SA_BONUS && *index == -1 &&
605 P2ROUNDUP(*total + hdrsize, 8) >
606 (full_space - sizeof (blkptr_t))) {
612 if (P2ROUNDUP(*total + hdrsize, 8) > full_space &&
614 *will_spill = B_TRUE;
617 hdrsize = P2ROUNDUP(hdrsize, 8);
621 #define BUF_SPACE_NEEDED(total, header) (total + header)
624 * Find layout that corresponds to ordering of attributes
625 * If not found a new layout number is created and added to
626 * persistent layout tables.
629 sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
632 sa_os_t *sa = hdl->sa_os->os_sa;
634 sa_buf_type_t buftype;
635 sa_hdr_phys_t *sahdr;
638 sa_attr_type_t *attrs, *attrs_start;
640 int hdrsize, spillhdrsize;
642 dmu_object_type_t bonustype;
648 dmu_buf_will_dirty(hdl->sa_bonus, tx);
649 bonustype = SA_BONUSTYPE_FROM_DB(hdl->sa_bonus);
651 /* first determine bonus header size and sum of all attributes */
652 hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus,
653 SA_BONUS, &i, &used, &spilling);
655 if (used > SPA_MAXBLOCKSIZE)
658 VERIFY(0 == dmu_set_bonus(hdl->sa_bonus, spilling ?
659 MIN(DN_MAX_BONUSLEN - sizeof (blkptr_t), used + hdrsize) :
660 used + hdrsize, tx));
662 ASSERT((bonustype == DMU_OT_ZNODE && spilling == 0) ||
663 bonustype == DMU_OT_SA);
665 /* setup and size spill buffer when needed */
669 if (hdl->sa_spill == NULL) {
671 error = dmu_spill_hold_by_bonus(hdl->sa_bonus, NULL,
673 ASSERT3U(error, ==, 0);
675 dmu_buf_will_dirty(hdl->sa_spill, tx);
677 spillhdrsize = sa_find_sizes(sa, &attr_desc[i],
678 attr_count - i, hdl->sa_spill, SA_SPILL, &i,
679 &spill_used, &dummy);
681 if (spill_used > SPA_MAXBLOCKSIZE)
684 buf_space = hdl->sa_spill->db_size - spillhdrsize;
685 if (BUF_SPACE_NEEDED(spill_used, spillhdrsize) >
686 hdl->sa_spill->db_size)
687 VERIFY(0 == sa_resize_spill(hdl,
688 BUF_SPACE_NEEDED(spill_used, spillhdrsize), tx));
691 /* setup starting pointers to lay down data */
692 data_start = (void *)((uintptr_t)hdl->sa_bonus->db_data + hdrsize);
693 sahdr = (sa_hdr_phys_t *)hdl->sa_bonus->db_data;
697 buf_space = (sa->sa_force_spill) ?
698 0 : SA_BLKPTR_SPACE - hdrsize;
700 buf_space = hdl->sa_bonus->db_size - hdrsize;
702 attrs_start = attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
706 for (i = 0, len_idx = 0, hash = -1ULL; i != attr_count; i++) {
709 attrs[i] = attr_desc[i].sa_attr;
710 length = SA_REGISTERED_LEN(sa, attrs[i]);
712 length = attr_desc[i].sa_length;
714 if (buf_space < length) { /* switch to spill buffer */
715 ASSERT(bonustype != DMU_OT_ZNODE);
716 if (buftype == SA_BONUS && !sa->sa_force_spill) {
717 sa_find_layout(hdl->sa_os, hash, attrs_start,
718 lot_count, tx, &lot);
719 SA_SET_HDR(sahdr, lot->lot_num, hdrsize);
726 sahdr = (sa_hdr_phys_t *)hdl->sa_spill->db_data;
727 sahdr->sa_magic = SA_MAGIC;
728 data_start = (void *)((uintptr_t)sahdr +
730 attrs_start = &attrs[i];
731 buf_space = hdl->sa_spill->db_size - spillhdrsize;
734 hash ^= SA_ATTR_HASH(attrs[i]);
735 attr_desc[i].sa_addr = data_start;
736 attr_desc[i].sa_size = length;
737 SA_COPY_DATA(attr_desc[i].sa_data_func, attr_desc[i].sa_data,
739 if (sa->sa_attr_table[attrs[i]].sa_length == 0) {
740 sahdr->sa_lengths[len_idx++] = length;
742 data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
744 buf_space -= P2ROUNDUP(length, 8);
748 sa_find_layout(hdl->sa_os, hash, attrs_start, lot_count, tx, &lot);
749 if (bonustype == DMU_OT_SA) {
750 SA_SET_HDR(sahdr, lot->lot_num,
751 buftype == SA_BONUS ? hdrsize : spillhdrsize);
754 kmem_free(attrs, sizeof (sa_attr_type_t) * attr_count);
755 if (hdl->sa_bonus_tab) {
756 sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
757 hdl->sa_bonus_tab = NULL;
759 if (!sa->sa_force_spill)
760 VERIFY(0 == sa_build_index(hdl, SA_BONUS));
762 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
765 * remove spill block that is no longer needed.
766 * set sa_spill_remove to prevent sa_attr_op
767 * from trying to retrieve spill block before its
768 * been removed. The flag will be cleared if/when
769 * the handle is destroyed recreated or
770 * sa_build_layouts() needs to spill again.
772 dmu_buf_rele(hdl->sa_spill, NULL);
773 hdl->sa_spill = NULL;
774 hdl->sa_spill_tab = NULL;
775 VERIFY(0 == dmu_rm_spill(hdl->sa_os,
776 sa_handle_object(hdl), tx));
778 VERIFY(0 == sa_build_index(hdl, SA_SPILL));
786 sa_attr_table_setup(objset_t *os, sa_attr_reg_t *reg_attrs, int count)
788 sa_os_t *sa = os->os_sa;
789 uint64_t sa_attr_count = 0;
795 int registered_count = 0;
797 dmu_objset_type_t ostype = dmu_objset_type(os);
800 kmem_zalloc(count * sizeof (sa_attr_type_t), KM_SLEEP);
801 sa->sa_user_table_sz = count * sizeof (sa_attr_type_t);
803 if (sa->sa_reg_attr_obj != 0)
804 VERIFY(zap_count(os, sa->sa_reg_attr_obj, &sa_attr_count) == 0);
806 if (ostype == DMU_OST_ZFS && sa_attr_count == 0)
807 sa_attr_count += sa_legacy_attr_count;
809 /* Allocate attribute numbers for attributes that aren't registered */
810 for (i = 0; i != count; i++) {
811 boolean_t found = B_FALSE;
814 if (ostype == DMU_OST_ZFS) {
815 for (j = 0; j != sa_legacy_attr_count; j++) {
816 if (strcmp(reg_attrs[i].sa_name,
817 sa_legacy_attrs[j].sa_name) == 0) {
818 sa->sa_user_table[i] =
819 sa_legacy_attrs[j].sa_attr;
827 if (sa->sa_reg_attr_obj)
828 error = zap_lookup(os, sa->sa_reg_attr_obj,
829 reg_attrs[i].sa_name, 8, 1, &attr_value);
835 sa->sa_user_table[i] = (sa_attr_type_t)sa_attr_count;
839 sa->sa_user_table[i] = ATTR_NUM(attr_value);
844 os->os_sa->sa_num_attrs = sa_attr_count;
845 tb = os->os_sa->sa_attr_table =
846 kmem_zalloc(sizeof (sa_attr_table_t) * sa_attr_count, KM_SLEEP);
849 * Attribute table is constructed from requested attribute list,
850 * previously foreign registered attributes, and also the legacy
851 * ZPL set of attributes.
854 if (sa->sa_reg_attr_obj) {
855 for (zap_cursor_init(&zc, os, sa->sa_reg_attr_obj);
856 zap_cursor_retrieve(&zc, &za) == 0;
857 zap_cursor_advance(&zc)) {
859 value = za.za_first_integer;
862 tb[ATTR_NUM(value)].sa_attr = ATTR_NUM(value);
863 tb[ATTR_NUM(value)].sa_length = ATTR_LENGTH(value);
864 tb[ATTR_NUM(value)].sa_byteswap = ATTR_BSWAP(value);
865 tb[ATTR_NUM(value)].sa_registered = B_TRUE;
867 if (tb[ATTR_NUM(value)].sa_name) {
870 tb[ATTR_NUM(value)].sa_name =
871 kmem_zalloc(strlen(za.za_name) +1, KM_SLEEP);
872 (void) strlcpy(tb[ATTR_NUM(value)].sa_name, za.za_name,
873 strlen(za.za_name) +1);
875 zap_cursor_fini(&zc);
878 if (ostype == DMU_OST_ZFS) {
879 for (i = 0; i != sa_legacy_attr_count; i++) {
882 tb[i].sa_attr = sa_legacy_attrs[i].sa_attr;
883 tb[i].sa_length = sa_legacy_attrs[i].sa_length;
884 tb[i].sa_byteswap = sa_legacy_attrs[i].sa_byteswap;
885 tb[i].sa_registered = B_FALSE;
887 kmem_zalloc(strlen(sa_legacy_attrs[i].sa_name) +1,
889 (void) strlcpy(tb[i].sa_name,
890 sa_legacy_attrs[i].sa_name,
891 strlen(sa_legacy_attrs[i].sa_name) + 1);
895 for (i = 0; i != count; i++) {
896 sa_attr_type_t attr_id;
898 attr_id = sa->sa_user_table[i];
899 if (tb[attr_id].sa_name)
902 tb[attr_id].sa_length = reg_attrs[i].sa_length;
903 tb[attr_id].sa_byteswap = reg_attrs[i].sa_byteswap;
904 tb[attr_id].sa_attr = attr_id;
905 tb[attr_id].sa_name =
906 kmem_zalloc(strlen(reg_attrs[i].sa_name) + 1, KM_SLEEP);
907 (void) strlcpy(tb[attr_id].sa_name, reg_attrs[i].sa_name,
908 strlen(reg_attrs[i].sa_name) + 1);
911 os->os_sa->sa_need_attr_registration =
912 (sa_attr_count != registered_count);
916 sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count)
921 dmu_objset_type_t ostype = dmu_objset_type(os);
924 mutex_enter(&os->os_lock);
926 mutex_enter(&os->os_sa->sa_lock);
927 mutex_exit(&os->os_lock);
928 tb = os->os_sa->sa_user_table;
929 mutex_exit(&os->os_sa->sa_lock);
933 sa = kmem_zalloc(sizeof (sa_os_t), KM_SLEEP);
934 mutex_init(&sa->sa_lock, NULL, MUTEX_DEFAULT, NULL);
935 sa->sa_master_obj = sa_obj;
937 mutex_enter(&sa->sa_lock);
938 mutex_exit(&os->os_lock);
939 avl_create(&sa->sa_layout_num_tree, layout_num_compare,
940 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node));
941 avl_create(&sa->sa_layout_hash_tree, layout_hash_compare,
942 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_hash_node));
946 error = zap_lookup(os, sa_obj, SA_LAYOUTS,
947 8, 1, &sa->sa_layout_attr_obj);
948 if (error != 0 && error != ENOENT) {
951 error = zap_lookup(os, sa_obj, SA_REGISTRY,
952 8, 1, &sa->sa_reg_attr_obj);
953 if (error != 0 && error != ENOENT) {
954 mutex_exit(&sa->sa_lock);
960 sa_attr_table_setup(os, reg_attrs, count);
962 if (sa->sa_layout_attr_obj != 0) {
963 for (zap_cursor_init(&zc, os, sa->sa_layout_attr_obj);
964 zap_cursor_retrieve(&zc, &za) == 0;
965 zap_cursor_advance(&zc)) {
966 sa_attr_type_t *lot_attrs;
969 lot_attrs = kmem_zalloc(sizeof (sa_attr_type_t) *
970 za.za_num_integers, KM_SLEEP);
972 VERIFY(zap_lookup(os, sa->sa_layout_attr_obj,
973 za.za_name, 2, za.za_num_integers, lot_attrs) == 0);
974 VERIFY(ddi_strtoull(za.za_name, NULL, 10,
975 (unsigned long long *)&lot_num) == 0);
977 (void) sa_add_layout_entry(os, lot_attrs,
978 za.za_num_integers, lot_num,
979 sa_layout_info_hash(lot_attrs,
980 za.za_num_integers), B_FALSE, NULL);
981 kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
984 zap_cursor_fini(&zc);
987 /* Add special layout number for old ZNODES */
988 if (ostype == DMU_OST_ZFS) {
989 (void) sa_add_layout_entry(os, sa_legacy_zpl_layout,
990 sa_legacy_attr_count, 0,
991 sa_layout_info_hash(sa_legacy_zpl_layout,
992 sa_legacy_attr_count), B_FALSE, NULL);
994 (void) sa_add_layout_entry(os, sa_dummy_zpl_layout, 0, 1,
997 mutex_exit(&sa->sa_lock);
998 return (os->os_sa->sa_user_table);
1002 sa_tear_down(objset_t *os)
1004 sa_os_t *sa = os->os_sa;
1009 kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
1011 /* Free up attr table */
1013 for (i = 0; i != sa->sa_num_attrs; i++) {
1014 if (sa->sa_attr_table[i].sa_name)
1015 kmem_free(sa->sa_attr_table[i].sa_name,
1016 strlen(sa->sa_attr_table[i].sa_name) + 1);
1019 kmem_free(sa->sa_attr_table,
1020 sizeof (sa_attr_table_t) * sa->sa_num_attrs);
1023 while (layout = avl_destroy_nodes(&sa->sa_layout_hash_tree, &cookie)) {
1025 while (tab = list_head(&layout->lot_idx_tab)) {
1026 ASSERT(refcount_count(&tab->sa_refcount));
1027 sa_idx_tab_rele(os, tab);
1032 while (layout = avl_destroy_nodes(&sa->sa_layout_num_tree, &cookie)) {
1033 kmem_free(layout->lot_attrs,
1034 sizeof (sa_attr_type_t) * layout->lot_attr_count);
1035 kmem_free(layout, sizeof (sa_lot_t));
1038 avl_destroy(&sa->sa_layout_hash_tree);
1039 avl_destroy(&sa->sa_layout_num_tree);
1041 kmem_free(sa, sizeof (sa_os_t));
1046 sa_build_idx_tab(void *hdr, void *attr_addr, sa_attr_type_t attr,
1047 uint16_t length, int length_idx, boolean_t var_length, void *userp)
1049 sa_idx_tab_t *idx_tab = userp;
1052 ASSERT(idx_tab->sa_variable_lengths);
1053 idx_tab->sa_variable_lengths[length_idx] = length;
1055 TOC_ATTR_ENCODE(idx_tab->sa_idx_tab[attr], length_idx,
1056 (uint32_t)((uintptr_t)attr_addr - (uintptr_t)hdr));
1060 sa_attr_iter(objset_t *os, sa_hdr_phys_t *hdr, dmu_object_type_t type,
1061 sa_iterfunc_t func, sa_lot_t *tab, void *userp)
1067 sa_os_t *sa = os->os_sa;
1069 uint16_t *length_start = NULL;
1070 uint8_t length_idx = 0;
1073 search.lot_num = SA_LAYOUT_NUM(hdr, type);
1074 tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
1078 if (IS_SA_BONUSTYPE(type)) {
1079 data_start = (void *)P2ROUNDUP(((uintptr_t)hdr +
1080 offsetof(sa_hdr_phys_t, sa_lengths) +
1081 (sizeof (uint16_t) * tb->lot_var_sizes)), 8);
1082 length_start = hdr->sa_lengths;
1087 for (i = 0; i != tb->lot_attr_count; i++) {
1088 int attr_length, reg_length;
1091 reg_length = sa->sa_attr_table[tb->lot_attrs[i]].sa_length;
1093 attr_length = reg_length;
1096 attr_length = length_start[length_idx];
1097 idx_len = length_idx++;
1100 func(hdr, data_start, tb->lot_attrs[i], attr_length,
1101 idx_len, reg_length == 0 ? B_TRUE : B_FALSE, userp);
1103 data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
1110 sa_byteswap_cb(void *hdr, void *attr_addr, sa_attr_type_t attr,
1111 uint16_t length, int length_idx, boolean_t variable_length, void *userp)
1113 sa_handle_t *hdl = userp;
1114 sa_os_t *sa = hdl->sa_os->os_sa;
1116 sa_bswap_table[sa->sa_attr_table[attr].sa_byteswap](attr_addr, length);
1120 sa_byteswap(sa_handle_t *hdl, sa_buf_type_t buftype)
1122 sa_hdr_phys_t *sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1124 sa_os_t *sa = hdl->sa_os->os_sa;
1125 int num_lengths = 1;
1128 ASSERT(MUTEX_HELD(&sa->sa_lock));
1129 if (sa_hdr_phys->sa_magic == SA_MAGIC)
1132 db = SA_GET_DB(hdl, buftype);
1134 if (buftype == SA_SPILL) {
1135 arc_release(db->db_buf, NULL);
1136 arc_buf_thaw(db->db_buf);
1139 sa_hdr_phys->sa_magic = BSWAP_32(sa_hdr_phys->sa_magic);
1140 sa_hdr_phys->sa_layout_info = BSWAP_16(sa_hdr_phys->sa_layout_info);
1143 * Determine number of variable lenghts in header
1144 * The standard 8 byte header has one for free and a
1145 * 16 byte header would have 4 + 1;
1147 if (SA_HDR_SIZE(sa_hdr_phys) > 8)
1148 num_lengths += (SA_HDR_SIZE(sa_hdr_phys) - 8) >> 1;
1149 for (i = 0; i != num_lengths; i++)
1150 sa_hdr_phys->sa_lengths[i] =
1151 BSWAP_16(sa_hdr_phys->sa_lengths[i]);
1153 sa_attr_iter(hdl->sa_os, sa_hdr_phys, DMU_OT_SA,
1154 sa_byteswap_cb, NULL, hdl);
1156 if (buftype == SA_SPILL)
1157 arc_buf_freeze(((dmu_buf_impl_t *)hdl->sa_spill)->db_buf);
1161 sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype)
1163 sa_hdr_phys_t *sa_hdr_phys;
1164 dmu_buf_impl_t *db = SA_GET_DB(hdl, buftype);
1165 dmu_object_type_t bonustype = SA_BONUSTYPE_FROM_DB(db);
1166 sa_os_t *sa = hdl->sa_os->os_sa;
1167 sa_idx_tab_t *idx_tab;
1169 sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1171 mutex_enter(&sa->sa_lock);
1173 /* Do we need to byteswap? */
1175 /* only check if not old znode */
1176 if (IS_SA_BONUSTYPE(bonustype) && sa_hdr_phys->sa_magic != SA_MAGIC &&
1177 sa_hdr_phys->sa_magic != 0) {
1178 VERIFY(BSWAP_32(sa_hdr_phys->sa_magic) == SA_MAGIC);
1179 sa_byteswap(hdl, buftype);
1182 idx_tab = sa_find_idx_tab(hdl->sa_os, bonustype, sa_hdr_phys);
1184 if (buftype == SA_BONUS)
1185 hdl->sa_bonus_tab = idx_tab;
1187 hdl->sa_spill_tab = idx_tab;
1189 mutex_exit(&sa->sa_lock);
1195 sa_evict(dmu_buf_t *db, void *sap)
1197 panic("evicting sa dbuf %p\n", (void *)db);
1201 sa_idx_tab_rele(objset_t *os, void *arg)
1203 sa_os_t *sa = os->os_sa;
1204 sa_idx_tab_t *idx_tab = arg;
1206 if (idx_tab == NULL)
1209 mutex_enter(&sa->sa_lock);
1210 if (refcount_remove(&idx_tab->sa_refcount, NULL) == 0) {
1211 list_remove(&idx_tab->sa_layout->lot_idx_tab, idx_tab);
1212 if (idx_tab->sa_variable_lengths)
1213 kmem_free(idx_tab->sa_variable_lengths,
1215 idx_tab->sa_layout->lot_var_sizes);
1216 refcount_destroy(&idx_tab->sa_refcount);
1217 kmem_free(idx_tab->sa_idx_tab,
1218 sizeof (uint32_t) * sa->sa_num_attrs);
1219 kmem_free(idx_tab, sizeof (sa_idx_tab_t));
1221 mutex_exit(&sa->sa_lock);
1225 sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab)
1227 sa_os_t *sa = os->os_sa;
1229 ASSERT(MUTEX_HELD(&sa->sa_lock));
1230 (void) refcount_add(&idx_tab->sa_refcount, NULL);
1234 sa_handle_destroy(sa_handle_t *hdl)
1236 mutex_enter(&hdl->sa_lock);
1237 (void) dmu_buf_update_user((dmu_buf_t *)hdl->sa_bonus, hdl,
1240 if (hdl->sa_bonus_tab) {
1241 sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
1242 hdl->sa_bonus_tab = NULL;
1244 if (hdl->sa_spill_tab) {
1245 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
1246 hdl->sa_spill_tab = NULL;
1249 dmu_buf_rele(hdl->sa_bonus, NULL);
1252 dmu_buf_rele((dmu_buf_t *)hdl->sa_spill, NULL);
1253 mutex_exit(&hdl->sa_lock);
1255 kmem_cache_free(sa_cache, hdl);
1259 sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
1260 sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1263 dmu_object_info_t doi;
1264 sa_handle_t *handle;
1267 dmu_object_info_from_db(db, &doi);
1268 ASSERT(doi.doi_bonus_type == DMU_OT_SA ||
1269 doi.doi_bonus_type == DMU_OT_ZNODE);
1271 /* find handle, if it exists */
1272 /* if one doesn't exist then create a new one, and initialize it */
1274 handle = (hdl_type == SA_HDL_SHARED) ? dmu_buf_get_user(db) : NULL;
1275 if (handle == NULL) {
1276 sa_handle_t *newhandle;
1277 handle = kmem_cache_alloc(sa_cache, KM_SLEEP);
1278 handle->sa_userp = userp;
1279 handle->sa_bonus = db;
1281 handle->sa_spill = NULL;
1283 error = sa_build_index(handle, SA_BONUS);
1284 newhandle = (hdl_type == SA_HDL_SHARED) ?
1285 dmu_buf_set_user_ie(db, handle,
1286 NULL, sa_evict) : NULL;
1288 if (newhandle != NULL) {
1289 kmem_cache_free(sa_cache, handle);
1299 sa_handle_get(objset_t *objset, uint64_t objid, void *userp,
1300 sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1305 if (error = dmu_bonus_hold(objset, objid, NULL, &db))
1308 return (sa_handle_get_from_db(objset, db, userp, hdl_type,
1313 sa_buf_hold(objset_t *objset, uint64_t obj_num, void *tag, dmu_buf_t **db)
1315 return (dmu_bonus_hold(objset, obj_num, tag, db));
1319 sa_buf_rele(dmu_buf_t *db, void *tag)
1321 dmu_buf_rele(db, tag);
1325 sa_lookup_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count)
1328 ASSERT(MUTEX_HELD(&hdl->sa_lock));
1329 return (sa_attr_op(hdl, bulk, count, SA_LOOKUP, NULL));
1333 sa_lookup(sa_handle_t *hdl, sa_attr_type_t attr, void *buf, uint32_t buflen)
1336 sa_bulk_attr_t bulk;
1338 bulk.sa_attr = attr;
1340 bulk.sa_length = buflen;
1341 bulk.sa_data_func = NULL;
1344 mutex_enter(&hdl->sa_lock);
1345 error = sa_lookup_impl(hdl, &bulk, 1);
1346 mutex_exit(&hdl->sa_lock);
1352 sa_lookup_uio(sa_handle_t *hdl, sa_attr_type_t attr, uio_t *uio)
1355 sa_bulk_attr_t bulk;
1357 bulk.sa_data = NULL;
1358 bulk.sa_attr = attr;
1359 bulk.sa_data_func = NULL;
1363 mutex_enter(&hdl->sa_lock);
1364 if (sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL) == 0) {
1365 error = uiomove((void *)bulk.sa_addr, MIN(bulk.sa_size,
1366 uio->uio_resid), UIO_READ, uio);
1370 mutex_exit(&hdl->sa_lock);
1377 * Find an already existing TOC from given os and data
1378 * This is a special interface to be used by the ZPL for
1379 * finding the uid/gid/gen attributes.
1382 sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype, void *data)
1384 sa_idx_tab_t *idx_tab;
1385 sa_hdr_phys_t *hdr = (sa_hdr_phys_t *)data;
1386 sa_os_t *sa = os->os_sa;
1387 sa_lot_t *tb, search;
1391 * Deterimine layout number. If SA node and header == 0 then
1392 * force the index table to the dummy "1" empty layout.
1394 * The layout number would only be zero for a newly created file
1395 * that has not added any attributes yet, or with crypto enabled which
1396 * doesn't write any attributes to the bonus buffer.
1399 search.lot_num = SA_LAYOUT_NUM(hdr, bonustype);
1401 tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
1403 /* Verify header size is consistent with layout information */
1405 ASSERT(IS_SA_BONUSTYPE(bonustype) &&
1406 SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb) || !IS_SA_BONUSTYPE(bonustype) ||
1407 (IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0));
1410 * See if any of the already existing TOC entries can be reused?
1413 for (idx_tab = list_head(&tb->lot_idx_tab); idx_tab;
1414 idx_tab = list_next(&tb->lot_idx_tab, idx_tab)) {
1415 boolean_t valid_idx = B_TRUE;
1418 if (tb->lot_var_sizes != 0 &&
1419 idx_tab->sa_variable_lengths != NULL) {
1420 for (i = 0; i != tb->lot_var_sizes; i++) {
1421 if (hdr->sa_lengths[i] !=
1422 idx_tab->sa_variable_lengths[i]) {
1423 valid_idx = B_FALSE;
1429 sa_idx_tab_hold(os, idx_tab);
1434 /* No such luck, create a new entry */
1435 idx_tab = kmem_zalloc(sizeof (sa_idx_tab_t), KM_SLEEP);
1436 idx_tab->sa_idx_tab =
1437 kmem_zalloc(sizeof (uint32_t) * sa->sa_num_attrs, KM_SLEEP);
1438 idx_tab->sa_layout = tb;
1439 refcount_create(&idx_tab->sa_refcount);
1440 if (tb->lot_var_sizes)
1441 idx_tab->sa_variable_lengths = kmem_alloc(sizeof (uint16_t) *
1442 tb->lot_var_sizes, KM_SLEEP);
1444 sa_attr_iter(os, hdr, bonustype, sa_build_idx_tab,
1446 sa_idx_tab_hold(os, idx_tab); /* one hold for consumer */
1447 sa_idx_tab_hold(os, idx_tab); /* one for layout */
1448 list_insert_tail(&tb->lot_idx_tab, idx_tab);
1453 sa_default_locator(void **dataptr, uint32_t *len, uint32_t total_len,
1454 boolean_t start, void *userdata)
1458 *dataptr = userdata;
1463 sa_attr_register_sync(sa_handle_t *hdl, dmu_tx_t *tx)
1465 uint64_t attr_value = 0;
1466 sa_os_t *sa = hdl->sa_os->os_sa;
1467 sa_attr_table_t *tb = sa->sa_attr_table;
1470 mutex_enter(&sa->sa_lock);
1472 if (!sa->sa_need_attr_registration || sa->sa_master_obj == NULL) {
1473 mutex_exit(&sa->sa_lock);
1477 if (sa->sa_reg_attr_obj == NULL) {
1479 sa->sa_reg_attr_obj = zap_create(hdl->sa_os,
1480 DMU_OT_SA_ATTR_REGISTRATION, DMU_OT_NONE, 0, tx);
1481 error = zap_add(hdl->sa_os, sa->sa_master_obj,
1482 SA_REGISTRY, 8, 1, &sa->sa_reg_attr_obj, tx);
1485 for (i = 0; i != sa->sa_num_attrs; i++) {
1486 if (sa->sa_attr_table[i].sa_registered)
1488 ATTR_ENCODE(attr_value, tb[i].sa_attr, tb[i].sa_length,
1490 VERIFY(0 == zap_update(hdl->sa_os, sa->sa_reg_attr_obj,
1491 tb[i].sa_name, 8, 1, &attr_value, tx));
1492 tb[i].sa_registered = B_TRUE;
1494 sa->sa_need_attr_registration = B_FALSE;
1495 mutex_exit(&sa->sa_lock);
1499 * Replace all attributes with attributes specified in template.
1500 * If dnode had a spill buffer then those attributes will be
1501 * also be replaced, possibly with just an empty spill block
1503 * This interface is intended to only be used for bulk adding of
1504 * attributes for a new file. It will also be used by the ZPL
1505 * when converting and old formatted znode to native SA support.
1508 sa_replace_all_by_template_locked(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
1509 int attr_count, dmu_tx_t *tx)
1511 sa_os_t *sa = hdl->sa_os->os_sa;
1513 if (sa->sa_need_attr_registration)
1514 sa_attr_register_sync(hdl, tx);
1515 return (sa_build_layouts(hdl, attr_desc, attr_count, tx));
1519 sa_replace_all_by_template(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
1520 int attr_count, dmu_tx_t *tx)
1524 mutex_enter(&hdl->sa_lock);
1525 error = sa_replace_all_by_template_locked(hdl, attr_desc,
1527 mutex_exit(&hdl->sa_lock);
1532 * add/remove/replace a single attribute and then rewrite the entire set
1536 sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
1537 sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
1538 uint16_t buflen, dmu_tx_t *tx)
1540 sa_os_t *sa = hdl->sa_os->os_sa;
1541 sa_bulk_attr_t *attr_desc;
1543 int bonus_attr_count = 0;
1544 int bonus_data_size, spill_data_size;
1545 int spill_attr_count = 0;
1548 int i, j, k, length_idx;
1550 sa_idx_tab_t *idx_tab;
1554 ASSERT(MUTEX_HELD(&hdl->sa_lock));
1556 /* First make of copy of the old data */
1558 if (((dmu_buf_impl_t *)hdl->sa_bonus)->db_dnode->dn_bonuslen) {
1559 bonus_data_size = hdl->sa_bonus->db_size;
1560 old_data[0] = kmem_alloc(bonus_data_size, KM_SLEEP);
1561 bcopy(hdl->sa_bonus->db_data, old_data[0],
1562 hdl->sa_bonus->db_size);
1563 bonus_attr_count = hdl->sa_bonus_tab->sa_layout->lot_attr_count;
1568 /* Bring spill buffer online if it isn't currently */
1570 if (sa_has_blkptr(hdl)) {
1571 spill_data_size = hdl->sa_spill->db_size;
1572 old_data[1] = kmem_alloc(spill_data_size, KM_SLEEP);
1573 bcopy(hdl->sa_spill->db_data, old_data[1],
1574 hdl->sa_spill->db_size);
1576 hdl->sa_spill_tab->sa_layout->lot_attr_count;
1581 /* build descriptor of all attributes */
1583 attr_count = bonus_attr_count + spill_attr_count;
1584 if (action == SA_ADD)
1586 else if (action == SA_REMOVE)
1589 attr_desc = kmem_zalloc(sizeof (sa_bulk_attr_t) * attr_count, KM_SLEEP);
1592 * loop through bonus and spill buffer if it exists, and
1593 * build up new attr_descriptor to reset the attributes
1596 count = bonus_attr_count;
1597 hdr = SA_GET_HDR(hdl, SA_BONUS);
1598 idx_tab = SA_IDX_TAB_GET(hdl, SA_BONUS);
1599 for (; k != 2; k++) {
1600 /* iterate over each attribute in layout */
1601 for (i = 0, length_idx = 0; i != count; i++) {
1602 sa_attr_type_t attr;
1604 attr = idx_tab->sa_layout->lot_attrs[i];
1605 if (attr == newattr) {
1606 if (action == SA_REMOVE) {
1610 ASSERT(SA_REGISTERED_LEN(sa, attr) == 0);
1611 ASSERT(action == SA_REPLACE);
1612 SA_ADD_BULK_ATTR(attr_desc, j, attr,
1613 locator, datastart, buflen);
1615 length = SA_REGISTERED_LEN(sa, attr);
1617 length = hdr->sa_lengths[length_idx++];
1620 SA_ADD_BULK_ATTR(attr_desc, j, attr,
1622 (TOC_OFF(idx_tab->sa_idx_tab[attr]) +
1623 (uintptr_t)old_data[k]), length);
1626 if (k == 0 && hdl->sa_spill) {
1627 hdr = SA_GET_HDR(hdl, SA_SPILL);
1628 idx_tab = SA_IDX_TAB_GET(hdl, SA_SPILL);
1629 count = spill_attr_count;
1634 if (action == SA_ADD) {
1635 length = SA_REGISTERED_LEN(sa, newattr);
1639 SA_ADD_BULK_ATTR(attr_desc, j, newattr, locator,
1643 error = sa_build_layouts(hdl, attr_desc, attr_count, tx);
1646 kmem_free(old_data[0], bonus_data_size);
1648 kmem_free(old_data[1], spill_data_size);
1649 kmem_free(attr_desc, sizeof (sa_bulk_attr_t) * attr_count);
1655 sa_bulk_update_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
1659 sa_os_t *sa = hdl->sa_os->os_sa;
1660 dmu_object_type_t bonustype;
1662 bonustype = SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl, SA_BONUS));
1665 ASSERT(MUTEX_HELD(&hdl->sa_lock));
1667 /* sync out registration table if necessary */
1668 if (sa->sa_need_attr_registration)
1669 sa_attr_register_sync(hdl, tx);
1671 error = sa_attr_op(hdl, bulk, count, SA_UPDATE, tx);
1672 if (error == 0 && !IS_SA_BONUSTYPE(bonustype) && sa->sa_update_cb)
1673 sa->sa_update_cb(hdl, tx);
1679 * update or add new attribute
1682 sa_update(sa_handle_t *hdl, sa_attr_type_t type,
1683 void *buf, uint32_t buflen, dmu_tx_t *tx)
1686 sa_bulk_attr_t bulk;
1688 bulk.sa_attr = type;
1689 bulk.sa_data_func = NULL;
1690 bulk.sa_length = buflen;
1693 mutex_enter(&hdl->sa_lock);
1694 error = sa_bulk_update_impl(hdl, &bulk, 1, tx);
1695 mutex_exit(&hdl->sa_lock);
1700 sa_update_from_cb(sa_handle_t *hdl, sa_attr_type_t attr,
1701 uint32_t buflen, sa_data_locator_t *locator, void *userdata, dmu_tx_t *tx)
1704 sa_bulk_attr_t bulk;
1706 bulk.sa_attr = attr;
1707 bulk.sa_data = userdata;
1708 bulk.sa_data_func = locator;
1709 bulk.sa_length = buflen;
1711 mutex_enter(&hdl->sa_lock);
1712 error = sa_bulk_update_impl(hdl, &bulk, 1, tx);
1713 mutex_exit(&hdl->sa_lock);
1718 * Return size of an attribute
1722 sa_size(sa_handle_t *hdl, sa_attr_type_t attr, int *size)
1724 sa_bulk_attr_t bulk;
1726 bulk.sa_data = NULL;
1727 bulk.sa_attr = attr;
1728 bulk.sa_data_func = NULL;
1731 mutex_enter(&hdl->sa_lock);
1732 if (sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) {
1733 mutex_exit(&hdl->sa_lock);
1736 *size = bulk.sa_size;
1738 mutex_exit(&hdl->sa_lock);
1743 sa_bulk_lookup_locked(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
1746 ASSERT(MUTEX_HELD(&hdl->sa_lock));
1747 return (sa_lookup_impl(hdl, attrs, count));
1751 sa_bulk_lookup(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
1756 mutex_enter(&hdl->sa_lock);
1757 error = sa_bulk_lookup_locked(hdl, attrs, count);
1758 mutex_exit(&hdl->sa_lock);
1763 sa_bulk_update(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count, dmu_tx_t *tx)
1768 mutex_enter(&hdl->sa_lock);
1769 error = sa_bulk_update_impl(hdl, attrs, count, tx);
1770 mutex_exit(&hdl->sa_lock);
1775 sa_remove(sa_handle_t *hdl, sa_attr_type_t attr, dmu_tx_t *tx)
1779 mutex_enter(&hdl->sa_lock);
1780 error = sa_modify_attrs(hdl, attr, SA_REMOVE, NULL,
1782 mutex_exit(&hdl->sa_lock);
1787 sa_object_info(sa_handle_t *hdl, dmu_object_info_t *doi)
1789 dmu_object_info_from_db((dmu_buf_t *)hdl->sa_bonus, doi);
1793 sa_object_size(sa_handle_t *hdl, uint32_t *blksize, u_longlong_t *nblocks)
1795 dmu_object_size_from_db((dmu_buf_t *)hdl->sa_bonus,
1800 sa_update_user(sa_handle_t *newhdl, sa_handle_t *oldhdl)
1802 (void) dmu_buf_update_user((dmu_buf_t *)newhdl->sa_bonus,
1803 oldhdl, newhdl, NULL, sa_evict);
1804 oldhdl->sa_bonus = NULL;
1808 sa_set_userp(sa_handle_t *hdl, void *ptr)
1810 hdl->sa_userp = ptr;
1814 sa_get_db(sa_handle_t *hdl)
1816 return ((dmu_buf_t *)hdl->sa_bonus);
1820 sa_get_userdata(sa_handle_t *hdl)
1822 return (hdl->sa_userp);
1826 sa_register_update_callback_locked(objset_t *os, sa_update_cb_t *func)
1828 ASSERT(MUTEX_HELD(&os->os_sa->sa_lock));
1829 os->os_sa->sa_update_cb = func;
1833 sa_register_update_callback(objset_t *os, sa_update_cb_t *func)
1836 mutex_enter(&os->os_sa->sa_lock);
1837 sa_register_update_callback_locked(os, func);
1838 mutex_exit(&os->os_sa->sa_lock);
1842 sa_handle_object(sa_handle_t *hdl)
1844 return (hdl->sa_bonus->db_object);
1848 sa_enabled(objset_t *os)
1850 return (os->os_sa == NULL);
1854 sa_set_sa_object(objset_t *os, uint64_t sa_object)
1856 sa_os_t *sa = os->os_sa;
1858 if (sa->sa_master_obj)
1861 sa->sa_master_obj = sa_object;
1867 sa_hdrsize(void *arg)
1869 sa_hdr_phys_t *hdr = arg;
1871 return (SA_HDR_SIZE(hdr));
1875 sa_handle_lock(sa_handle_t *hdl)
1878 mutex_enter(&hdl->sa_lock);
1882 sa_handle_unlock(sa_handle_t *hdl)
1885 mutex_exit(&hdl->sa_lock);