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]
23 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24 * Portions Copyright 2011 iXsystems, Inc
25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_objset.h>
38 #include <sys/dnode.h>
41 #include <sys/sunddi.h>
42 #include <sys/sa_impl.h>
43 #include <sys/dnode.h>
44 #include <sys/errno.h>
45 #include <sys/zfs_context.h>
48 * ZFS System attributes:
50 * A generic mechanism to allow for arbitrary attributes
51 * to be stored in a dnode. The data will be stored in the bonus buffer of
52 * the dnode and if necessary a special "spill" block will be used to handle
53 * overflow situations. The spill block will be sized to fit the data
54 * from 512 - 128K. When a spill block is used the BP (blkptr_t) for the
55 * spill block is stored at the end of the current bonus buffer. Any
56 * attributes that would be in the way of the blkptr_t will be relocated
57 * into the spill block.
59 * Attribute registration:
61 * Stored persistently on a per dataset basis
62 * a mapping between attribute "string" names and their actual attribute
63 * numeric values, length, and byteswap function. The names are only used
64 * during registration. All attributes are known by their unique attribute
65 * id value. If an attribute can have a variable size then the value
66 * 0 will be used to indicate this.
70 * Attribute layouts are a way to compactly store multiple attributes, but
71 * without taking the overhead associated with managing each attribute
72 * individually. Since you will typically have the same set of attributes
73 * stored in the same order a single table will be used to represent that
74 * layout. The ZPL for example will usually have only about 10 different
75 * layouts (regular files, device files, symlinks,
76 * regular files + scanstamp, files/dir with extended attributes, and then
77 * you have the possibility of all of those minus ACL, because it would
78 * be kicked out into the spill block)
80 * Layouts are simply an array of the attributes and their
81 * ordering i.e. [0, 1, 4, 5, 2]
83 * Each distinct layout is given a unique layout number and that is whats
84 * stored in the header at the beginning of the SA data buffer.
86 * A layout only covers a single dbuf (bonus or spill). If a set of
87 * attributes is split up between the bonus buffer and a spill buffer then
88 * two different layouts will be used. This allows us to byteswap the
89 * spill without looking at the bonus buffer and keeps the on disk format of
90 * the bonus and spill buffer the same.
92 * Adding a single attribute will cause the entire set of attributes to
93 * be rewritten and could result in a new layout number being constructed
94 * as part of the rewrite if no such layout exists for the new set of
95 * attribues. The new attribute will be appended to the end of the already
96 * existing attributes.
98 * Both the attribute registration and attribute layout information are
99 * stored in normal ZAP attributes. Their should be a small number of
100 * known layouts and the set of attributes is assumed to typically be quite
103 * The registered attributes and layout "table" information is maintained
104 * in core and a special "sa_os_t" is attached to the objset_t.
106 * A special interface is provided to allow for quickly applying
107 * a large set of attributes at once. sa_replace_all_by_template() is
108 * used to set an array of attributes. This is used by the ZPL when
109 * creating a brand new file. The template that is passed into the function
110 * specifies the attribute, size for variable length attributes, location of
111 * data and special "data locator" function if the data isn't in a contiguous
114 * Byteswap implications:
116 * Since the SA attributes are not entirely self describing we can't do
117 * the normal byteswap processing. The special ZAP layout attribute and
118 * attribute registration attributes define the byteswap function and the
119 * size of the attributes, unless it is variable sized.
120 * The normal ZFS byteswapping infrastructure assumes you don't need
121 * to read any objects in order to do the necessary byteswapping. Whereas
122 * SA attributes can only be properly byteswapped if the dataset is opened
123 * and the layout/attribute ZAP attributes are available. Because of this
124 * the SA attributes will be byteswapped when they are first accessed by
125 * the SA code that will read the SA data.
128 typedef void (sa_iterfunc_t
)(void *hdr
, void *addr
, sa_attr_type_t
,
129 uint16_t length
, int length_idx
, boolean_t
, void *userp
);
131 static int sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
);
132 static void sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
);
133 static void *sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
,
135 static void sa_idx_tab_rele(objset_t
*os
, void *arg
);
136 static void sa_copy_data(sa_data_locator_t
*func
, void *start
, void *target
,
138 static int sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
139 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
140 uint16_t buflen
, dmu_tx_t
*tx
);
142 arc_byteswap_func_t
*sa_bswap_table
[] = {
143 byteswap_uint64_array
,
144 byteswap_uint32_array
,
145 byteswap_uint16_array
,
146 byteswap_uint8_array
,
150 #define SA_COPY_DATA(f, s, t, l) \
154 *(uint64_t *)t = *(uint64_t *)s; \
155 } else if (l == 16) { \
156 *(uint64_t *)t = *(uint64_t *)s; \
157 *(uint64_t *)((uintptr_t)t + 8) = \
158 *(uint64_t *)((uintptr_t)s + 8); \
163 sa_copy_data(f, s, t, l); \
167 * This table is fixed and cannot be changed. Its purpose is to
168 * allow the SA code to work with both old/new ZPL file systems.
169 * It contains the list of legacy attributes. These attributes aren't
170 * stored in the "attribute" registry zap objects, since older ZPL file systems
171 * won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
172 * use this static table.
174 sa_attr_reg_t sa_legacy_attrs
[] = {
175 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 0},
176 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 1},
177 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 2},
178 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 3},
179 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY
, 4},
180 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY
, 5},
181 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY
, 6},
182 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY
, 7},
183 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY
, 8},
184 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY
, 9},
185 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY
, 10},
186 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY
, 11},
187 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY
, 12},
188 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY
, 13},
189 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY
, 14},
190 {"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY
, 15},
194 * This is only used for objects of type DMU_OT_ZNODE
196 sa_attr_type_t sa_legacy_zpl_layout
[] = {
197 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
201 * Special dummy layout used for buffers with no attributes.
203 sa_attr_type_t sa_dummy_zpl_layout
[] = { 0 };
205 static int sa_legacy_attr_count
= 16;
206 static kmem_cache_t
*sa_cache
= NULL
;
210 sa_cache_constructor(void *buf
, void *unused
, int kmflag
)
212 sa_handle_t
*hdl
= buf
;
214 hdl
->sa_dbu
.dbu_evict_func
= NULL
;
215 hdl
->sa_bonus_tab
= NULL
;
216 hdl
->sa_spill_tab
= NULL
;
218 hdl
->sa_userp
= NULL
;
219 hdl
->sa_bonus
= NULL
;
220 hdl
->sa_spill
= NULL
;
221 mutex_init(&hdl
->sa_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
227 sa_cache_destructor(void *buf
, void *unused
)
229 sa_handle_t
*hdl
= buf
;
230 hdl
->sa_dbu
.dbu_evict_func
= NULL
;
231 mutex_destroy(&hdl
->sa_lock
);
237 sa_cache
= kmem_cache_create("sa_cache",
238 sizeof (sa_handle_t
), 0, sa_cache_constructor
,
239 sa_cache_destructor
, NULL
, NULL
, NULL
, 0);
246 kmem_cache_destroy(sa_cache
);
250 layout_num_compare(const void *arg1
, const void *arg2
)
252 const sa_lot_t
*node1
= arg1
;
253 const sa_lot_t
*node2
= arg2
;
255 if (node1
->lot_num
> node2
->lot_num
)
257 else if (node1
->lot_num
< node2
->lot_num
)
263 layout_hash_compare(const void *arg1
, const void *arg2
)
265 const sa_lot_t
*node1
= arg1
;
266 const sa_lot_t
*node2
= arg2
;
268 if (node1
->lot_hash
> node2
->lot_hash
)
270 if (node1
->lot_hash
< node2
->lot_hash
)
272 if (node1
->lot_instance
> node2
->lot_instance
)
274 if (node1
->lot_instance
< node2
->lot_instance
)
280 sa_layout_equal(sa_lot_t
*tbf
, sa_attr_type_t
*attrs
, int count
)
284 if (count
!= tbf
->lot_attr_count
)
287 for (i
= 0; i
!= count
; i
++) {
288 if (attrs
[i
] != tbf
->lot_attrs
[i
])
294 #define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
297 sa_layout_info_hash(sa_attr_type_t
*attrs
, int attr_count
)
300 uint64_t crc
= -1ULL;
302 for (i
= 0; i
!= attr_count
; i
++)
303 crc
^= SA_ATTR_HASH(attrs
[i
]);
309 sa_get_spill(sa_handle_t
*hdl
)
312 if (hdl
->sa_spill
== NULL
) {
313 if ((rc
= dmu_spill_hold_existing(hdl
->sa_bonus
, NULL
,
314 &hdl
->sa_spill
)) == 0)
315 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
324 * Main attribute lookup/update function
325 * returns 0 for success or non zero for failures
327 * Operates on bulk array, first failure will abort further processing
330 sa_attr_op(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
331 sa_data_op_t data_op
, dmu_tx_t
*tx
)
333 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
336 sa_buf_type_t buftypes
;
341 for (i
= 0; i
!= count
; i
++) {
342 ASSERT(bulk
[i
].sa_attr
<= hdl
->sa_os
->os_sa
->sa_num_attrs
);
344 bulk
[i
].sa_addr
= NULL
;
345 /* First check the bonus buffer */
347 if (hdl
->sa_bonus_tab
&& TOC_ATTR_PRESENT(
348 hdl
->sa_bonus_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
349 SA_ATTR_INFO(sa
, hdl
->sa_bonus_tab
,
350 SA_GET_HDR(hdl
, SA_BONUS
),
351 bulk
[i
].sa_attr
, bulk
[i
], SA_BONUS
, hdl
);
352 if (tx
&& !(buftypes
& SA_BONUS
)) {
353 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
354 buftypes
|= SA_BONUS
;
357 if (bulk
[i
].sa_addr
== NULL
&&
358 ((error
= sa_get_spill(hdl
)) == 0)) {
359 if (TOC_ATTR_PRESENT(
360 hdl
->sa_spill_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
361 SA_ATTR_INFO(sa
, hdl
->sa_spill_tab
,
362 SA_GET_HDR(hdl
, SA_SPILL
),
363 bulk
[i
].sa_attr
, bulk
[i
], SA_SPILL
, hdl
);
364 if (tx
&& !(buftypes
& SA_SPILL
) &&
365 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
366 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
367 buftypes
|= SA_SPILL
;
371 if (error
&& error
!= ENOENT
) {
372 return ((error
== ECKSUM
) ? EIO
: error
);
377 if (bulk
[i
].sa_addr
== NULL
)
378 return (SET_ERROR(ENOENT
));
379 if (bulk
[i
].sa_data
) {
380 SA_COPY_DATA(bulk
[i
].sa_data_func
,
381 bulk
[i
].sa_addr
, bulk
[i
].sa_data
,
387 /* existing rewrite of attr */
388 if (bulk
[i
].sa_addr
&&
389 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
390 SA_COPY_DATA(bulk
[i
].sa_data_func
,
391 bulk
[i
].sa_data
, bulk
[i
].sa_addr
,
394 } else if (bulk
[i
].sa_addr
) { /* attr size change */
395 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
396 SA_REPLACE
, bulk
[i
].sa_data_func
,
397 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
398 } else { /* adding new attribute */
399 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
400 SA_ADD
, bulk
[i
].sa_data_func
,
401 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
412 sa_add_layout_entry(objset_t
*os
, sa_attr_type_t
*attrs
, int attr_count
,
413 uint64_t lot_num
, uint64_t hash
, boolean_t zapadd
, dmu_tx_t
*tx
)
415 sa_os_t
*sa
= os
->os_sa
;
416 sa_lot_t
*tb
, *findtb
;
420 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
421 tb
= kmem_zalloc(sizeof (sa_lot_t
), KM_SLEEP
);
422 tb
->lot_attr_count
= attr_count
;
423 tb
->lot_attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
425 bcopy(attrs
, tb
->lot_attrs
, sizeof (sa_attr_type_t
) * attr_count
);
426 tb
->lot_num
= lot_num
;
428 tb
->lot_instance
= 0;
433 if (sa
->sa_layout_attr_obj
== 0) {
434 sa
->sa_layout_attr_obj
= zap_create_link(os
,
435 DMU_OT_SA_ATTR_LAYOUTS
,
436 sa
->sa_master_obj
, SA_LAYOUTS
, tx
);
439 (void) snprintf(attr_name
, sizeof (attr_name
),
441 VERIFY(0 == zap_update(os
, os
->os_sa
->sa_layout_attr_obj
,
442 attr_name
, 2, attr_count
, attrs
, tx
));
445 list_create(&tb
->lot_idx_tab
, sizeof (sa_idx_tab_t
),
446 offsetof(sa_idx_tab_t
, sa_next
));
448 for (i
= 0; i
!= attr_count
; i
++) {
449 if (sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
== 0)
453 avl_add(&sa
->sa_layout_num_tree
, tb
);
455 /* verify we don't have a hash collision */
456 if ((findtb
= avl_find(&sa
->sa_layout_hash_tree
, tb
, &loc
)) != NULL
) {
457 for (; findtb
&& findtb
->lot_hash
== hash
;
458 findtb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, findtb
)) {
459 if (findtb
->lot_instance
!= tb
->lot_instance
)
464 avl_add(&sa
->sa_layout_hash_tree
, tb
);
469 sa_find_layout(objset_t
*os
, uint64_t hash
, sa_attr_type_t
*attrs
,
470 int count
, dmu_tx_t
*tx
, sa_lot_t
**lot
)
472 sa_lot_t
*tb
, tbsearch
;
474 sa_os_t
*sa
= os
->os_sa
;
475 boolean_t found
= B_FALSE
;
477 mutex_enter(&sa
->sa_lock
);
478 tbsearch
.lot_hash
= hash
;
479 tbsearch
.lot_instance
= 0;
480 tb
= avl_find(&sa
->sa_layout_hash_tree
, &tbsearch
, &loc
);
482 for (; tb
&& tb
->lot_hash
== hash
;
483 tb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, tb
)) {
484 if (sa_layout_equal(tb
, attrs
, count
) == 0) {
491 tb
= sa_add_layout_entry(os
, attrs
, count
,
492 avl_numnodes(&sa
->sa_layout_num_tree
), hash
, B_TRUE
, tx
);
494 mutex_exit(&sa
->sa_lock
);
499 sa_resize_spill(sa_handle_t
*hdl
, uint32_t size
, dmu_tx_t
*tx
)
505 blocksize
= SPA_MINBLOCKSIZE
;
506 } else if (size
> SPA_OLD_MAXBLOCKSIZE
) {
508 return (SET_ERROR(EFBIG
));
510 blocksize
= P2ROUNDUP_TYPED(size
, SPA_MINBLOCKSIZE
, uint32_t);
513 error
= dbuf_spill_set_blksz(hdl
->sa_spill
, blocksize
, tx
);
519 sa_copy_data(sa_data_locator_t
*func
, void *datastart
, void *target
, int buflen
)
522 bcopy(datastart
, target
, buflen
);
527 void *saptr
= target
;
532 while (bytes
< buflen
) {
533 func(&dataptr
, &length
, buflen
, start
, datastart
);
534 bcopy(dataptr
, saptr
, length
);
535 saptr
= (void *)((caddr_t
)saptr
+ length
);
543 * Determine several different sizes
544 * first the sa header size
545 * the number of bytes to be stored
546 * if spill would occur the index in the attribute array is returned
548 * the boolean will_spill will be set when spilling is necessary. It
549 * is only set when the buftype is SA_BONUS
552 sa_find_sizes(sa_os_t
*sa
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
553 dmu_buf_t
*db
, sa_buf_type_t buftype
, int *index
, int *total
,
554 boolean_t
*will_spill
)
561 boolean_t done
= B_FALSE
;
563 if (buftype
== SA_BONUS
&& sa
->sa_force_spill
) {
566 *will_spill
= B_TRUE
;
573 if (buftype
== SA_BONUS
)
574 *will_spill
= B_FALSE
;
576 hdrsize
= (SA_BONUSTYPE_FROM_DB(db
) == DMU_OT_ZNODE
) ? 0 :
577 sizeof (sa_hdr_phys_t
);
579 full_space
= (buftype
== SA_BONUS
) ? DN_MAX_BONUSLEN
: db
->db_size
;
580 ASSERT(IS_P2ALIGNED(full_space
, 8));
582 for (i
= 0; i
!= attr_count
; i
++) {
585 *total
= P2ROUNDUP(*total
, 8);
586 *total
+= attr_desc
[i
].sa_length
;
590 is_var_sz
= (SA_REGISTERED_LEN(sa
, attr_desc
[i
].sa_attr
) == 0);
595 if (is_var_sz
&& var_size
> 1) {
596 if (P2ROUNDUP(hdrsize
+ sizeof (uint16_t), 8) +
597 *total
< full_space
) {
599 * Account for header space used by array of
600 * optional sizes of variable-length attributes.
601 * Record the index in case this increase needs
602 * to be reversed due to spill-over.
604 hdrsize
+= sizeof (uint16_t);
609 if (buftype
== SA_BONUS
)
610 *will_spill
= B_TRUE
;
616 * find index of where spill *could* occur.
617 * Then continue to count of remainder attribute
618 * space. The sum is used later for sizing bonus
621 if (buftype
== SA_BONUS
&& *index
== -1 &&
622 *total
+ P2ROUNDUP(hdrsize
, 8) >
623 (full_space
- sizeof (blkptr_t
))) {
629 if (*total
+ P2ROUNDUP(hdrsize
, 8) > full_space
&&
631 *will_spill
= B_TRUE
;
635 * j holds the index of the last variable-sized attribute for
636 * which hdrsize was increased. Reverse the increase if that
637 * attribute will be relocated to the spill block.
639 if (*will_spill
&& j
== *index
)
640 hdrsize
-= sizeof (uint16_t);
642 hdrsize
= P2ROUNDUP(hdrsize
, 8);
646 #define BUF_SPACE_NEEDED(total, header) (total + header)
649 * Find layout that corresponds to ordering of attributes
650 * If not found a new layout number is created and added to
651 * persistent layout tables.
654 sa_build_layouts(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
657 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
659 sa_buf_type_t buftype
;
660 sa_hdr_phys_t
*sahdr
;
663 sa_attr_type_t
*attrs
, *attrs_start
;
666 int spillhdrsize
= 0;
668 dmu_object_type_t bonustype
;
674 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
675 bonustype
= SA_BONUSTYPE_FROM_DB(hdl
->sa_bonus
);
677 /* first determine bonus header size and sum of all attributes */
678 hdrsize
= sa_find_sizes(sa
, attr_desc
, attr_count
, hdl
->sa_bonus
,
679 SA_BONUS
, &i
, &used
, &spilling
);
681 if (used
> SPA_OLD_MAXBLOCKSIZE
)
682 return (SET_ERROR(EFBIG
));
684 VERIFY(0 == dmu_set_bonus(hdl
->sa_bonus
, spilling
?
685 MIN(DN_MAX_BONUSLEN
- sizeof (blkptr_t
), used
+ hdrsize
) :
686 used
+ hdrsize
, tx
));
688 ASSERT((bonustype
== DMU_OT_ZNODE
&& spilling
== 0) ||
689 bonustype
== DMU_OT_SA
);
691 /* setup and size spill buffer when needed */
695 if (hdl
->sa_spill
== NULL
) {
696 VERIFY(dmu_spill_hold_by_bonus(hdl
->sa_bonus
, NULL
,
697 &hdl
->sa_spill
) == 0);
699 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
701 spillhdrsize
= sa_find_sizes(sa
, &attr_desc
[i
],
702 attr_count
- i
, hdl
->sa_spill
, SA_SPILL
, &i
,
703 &spill_used
, &dummy
);
705 if (spill_used
> SPA_OLD_MAXBLOCKSIZE
)
706 return (SET_ERROR(EFBIG
));
708 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
709 if (BUF_SPACE_NEEDED(spill_used
, spillhdrsize
) >
710 hdl
->sa_spill
->db_size
)
711 VERIFY(0 == sa_resize_spill(hdl
,
712 BUF_SPACE_NEEDED(spill_used
, spillhdrsize
), tx
));
715 /* setup starting pointers to lay down data */
716 data_start
= (void *)((uintptr_t)hdl
->sa_bonus
->db_data
+ hdrsize
);
717 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_bonus
->db_data
;
721 buf_space
= (sa
->sa_force_spill
) ?
722 0 : SA_BLKPTR_SPACE
- hdrsize
;
724 buf_space
= hdl
->sa_bonus
->db_size
- hdrsize
;
726 attrs_start
= attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
730 for (i
= 0, len_idx
= 0, hash
= -1ULL; i
!= attr_count
; i
++) {
733 ASSERT(IS_P2ALIGNED(data_start
, 8));
734 ASSERT(IS_P2ALIGNED(buf_space
, 8));
735 attrs
[i
] = attr_desc
[i
].sa_attr
;
736 length
= SA_REGISTERED_LEN(sa
, attrs
[i
]);
738 length
= attr_desc
[i
].sa_length
;
740 if (buf_space
< length
) { /* switch to spill buffer */
742 VERIFY(bonustype
== DMU_OT_SA
);
743 if (buftype
== SA_BONUS
&& !sa
->sa_force_spill
) {
744 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
,
745 lot_count
, tx
, &lot
);
746 SA_SET_HDR(sahdr
, lot
->lot_num
, hdrsize
);
753 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_spill
->db_data
;
754 sahdr
->sa_magic
= SA_MAGIC
;
755 data_start
= (void *)((uintptr_t)sahdr
+
757 attrs_start
= &attrs
[i
];
758 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
761 hash
^= SA_ATTR_HASH(attrs
[i
]);
762 attr_desc
[i
].sa_addr
= data_start
;
763 attr_desc
[i
].sa_size
= length
;
764 SA_COPY_DATA(attr_desc
[i
].sa_data_func
, attr_desc
[i
].sa_data
,
766 if (sa
->sa_attr_table
[attrs
[i
]].sa_length
== 0) {
767 sahdr
->sa_lengths
[len_idx
++] = length
;
769 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
771 buf_space
-= P2ROUNDUP(length
, 8);
775 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
, lot_count
, tx
, &lot
);
778 * Verify that old znodes always have layout number 0.
779 * Must be DMU_OT_SA for arbitrary layouts
781 VERIFY((bonustype
== DMU_OT_ZNODE
&& lot
->lot_num
== 0) ||
782 (bonustype
== DMU_OT_SA
&& lot
->lot_num
> 1));
784 if (bonustype
== DMU_OT_SA
) {
785 SA_SET_HDR(sahdr
, lot
->lot_num
,
786 buftype
== SA_BONUS
? hdrsize
: spillhdrsize
);
789 kmem_free(attrs
, sizeof (sa_attr_type_t
) * attr_count
);
790 if (hdl
->sa_bonus_tab
) {
791 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
792 hdl
->sa_bonus_tab
= NULL
;
794 if (!sa
->sa_force_spill
)
795 VERIFY(0 == sa_build_index(hdl
, SA_BONUS
));
797 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
800 * remove spill block that is no longer needed.
802 dmu_buf_rele(hdl
->sa_spill
, NULL
);
803 hdl
->sa_spill
= NULL
;
804 hdl
->sa_spill_tab
= NULL
;
805 VERIFY(0 == dmu_rm_spill(hdl
->sa_os
,
806 sa_handle_object(hdl
), tx
));
808 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
816 sa_free_attr_table(sa_os_t
*sa
)
820 if (sa
->sa_attr_table
== NULL
)
823 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
824 if (sa
->sa_attr_table
[i
].sa_name
)
825 kmem_free(sa
->sa_attr_table
[i
].sa_name
,
826 strlen(sa
->sa_attr_table
[i
].sa_name
) + 1);
829 kmem_free(sa
->sa_attr_table
,
830 sizeof (sa_attr_table_t
) * sa
->sa_num_attrs
);
832 sa
->sa_attr_table
= NULL
;
836 sa_attr_table_setup(objset_t
*os
, sa_attr_reg_t
*reg_attrs
, int count
)
838 sa_os_t
*sa
= os
->os_sa
;
839 uint64_t sa_attr_count
= 0;
840 uint64_t sa_reg_count
= 0;
846 int registered_count
= 0;
848 dmu_objset_type_t ostype
= dmu_objset_type(os
);
851 kmem_zalloc(count
* sizeof (sa_attr_type_t
), KM_SLEEP
);
852 sa
->sa_user_table_sz
= count
* sizeof (sa_attr_type_t
);
854 if (sa
->sa_reg_attr_obj
!= 0) {
855 error
= zap_count(os
, sa
->sa_reg_attr_obj
,
859 * Make sure we retrieved a count and that it isn't zero
861 if (error
|| (error
== 0 && sa_attr_count
== 0)) {
863 error
= SET_ERROR(EINVAL
);
866 sa_reg_count
= sa_attr_count
;
869 if (ostype
== DMU_OST_ZFS
&& sa_attr_count
== 0)
870 sa_attr_count
+= sa_legacy_attr_count
;
872 /* Allocate attribute numbers for attributes that aren't registered */
873 for (i
= 0; i
!= count
; i
++) {
874 boolean_t found
= B_FALSE
;
877 if (ostype
== DMU_OST_ZFS
) {
878 for (j
= 0; j
!= sa_legacy_attr_count
; j
++) {
879 if (strcmp(reg_attrs
[i
].sa_name
,
880 sa_legacy_attrs
[j
].sa_name
) == 0) {
881 sa
->sa_user_table
[i
] =
882 sa_legacy_attrs
[j
].sa_attr
;
890 if (sa
->sa_reg_attr_obj
)
891 error
= zap_lookup(os
, sa
->sa_reg_attr_obj
,
892 reg_attrs
[i
].sa_name
, 8, 1, &attr_value
);
894 error
= SET_ERROR(ENOENT
);
897 sa
->sa_user_table
[i
] = (sa_attr_type_t
)sa_attr_count
;
901 sa
->sa_user_table
[i
] = ATTR_NUM(attr_value
);
908 sa
->sa_num_attrs
= sa_attr_count
;
909 tb
= sa
->sa_attr_table
=
910 kmem_zalloc(sizeof (sa_attr_table_t
) * sa_attr_count
, KM_SLEEP
);
913 * Attribute table is constructed from requested attribute list,
914 * previously foreign registered attributes, and also the legacy
915 * ZPL set of attributes.
918 if (sa
->sa_reg_attr_obj
) {
919 for (zap_cursor_init(&zc
, os
, sa
->sa_reg_attr_obj
);
920 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
921 zap_cursor_advance(&zc
)) {
923 value
= za
.za_first_integer
;
926 tb
[ATTR_NUM(value
)].sa_attr
= ATTR_NUM(value
);
927 tb
[ATTR_NUM(value
)].sa_length
= ATTR_LENGTH(value
);
928 tb
[ATTR_NUM(value
)].sa_byteswap
= ATTR_BSWAP(value
);
929 tb
[ATTR_NUM(value
)].sa_registered
= B_TRUE
;
931 if (tb
[ATTR_NUM(value
)].sa_name
) {
934 tb
[ATTR_NUM(value
)].sa_name
=
935 kmem_zalloc(strlen(za
.za_name
) +1, KM_SLEEP
);
936 (void) strlcpy(tb
[ATTR_NUM(value
)].sa_name
, za
.za_name
,
937 strlen(za
.za_name
) +1);
939 zap_cursor_fini(&zc
);
941 * Make sure we processed the correct number of registered
944 if (registered_count
!= sa_reg_count
) {
951 if (ostype
== DMU_OST_ZFS
) {
952 for (i
= 0; i
!= sa_legacy_attr_count
; i
++) {
955 tb
[i
].sa_attr
= sa_legacy_attrs
[i
].sa_attr
;
956 tb
[i
].sa_length
= sa_legacy_attrs
[i
].sa_length
;
957 tb
[i
].sa_byteswap
= sa_legacy_attrs
[i
].sa_byteswap
;
958 tb
[i
].sa_registered
= B_FALSE
;
960 kmem_zalloc(strlen(sa_legacy_attrs
[i
].sa_name
) +1,
962 (void) strlcpy(tb
[i
].sa_name
,
963 sa_legacy_attrs
[i
].sa_name
,
964 strlen(sa_legacy_attrs
[i
].sa_name
) + 1);
968 for (i
= 0; i
!= count
; i
++) {
969 sa_attr_type_t attr_id
;
971 attr_id
= sa
->sa_user_table
[i
];
972 if (tb
[attr_id
].sa_name
)
975 tb
[attr_id
].sa_length
= reg_attrs
[i
].sa_length
;
976 tb
[attr_id
].sa_byteswap
= reg_attrs
[i
].sa_byteswap
;
977 tb
[attr_id
].sa_attr
= attr_id
;
978 tb
[attr_id
].sa_name
=
979 kmem_zalloc(strlen(reg_attrs
[i
].sa_name
) + 1, KM_SLEEP
);
980 (void) strlcpy(tb
[attr_id
].sa_name
, reg_attrs
[i
].sa_name
,
981 strlen(reg_attrs
[i
].sa_name
) + 1);
984 sa
->sa_need_attr_registration
=
985 (sa_attr_count
!= registered_count
);
989 kmem_free(sa
->sa_user_table
, count
* sizeof (sa_attr_type_t
));
990 sa
->sa_user_table
= NULL
;
991 sa_free_attr_table(sa
);
992 return ((error
!= 0) ? error
: EINVAL
);
996 sa_setup(objset_t
*os
, uint64_t sa_obj
, sa_attr_reg_t
*reg_attrs
, int count
,
997 sa_attr_type_t
**user_table
)
1002 dmu_objset_type_t ostype
= dmu_objset_type(os
);
1006 mutex_enter(&os
->os_user_ptr_lock
);
1008 mutex_enter(&os
->os_sa
->sa_lock
);
1009 mutex_exit(&os
->os_user_ptr_lock
);
1010 tb
= os
->os_sa
->sa_user_table
;
1011 mutex_exit(&os
->os_sa
->sa_lock
);
1016 sa
= kmem_zalloc(sizeof (sa_os_t
), KM_SLEEP
);
1017 mutex_init(&sa
->sa_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1018 sa
->sa_master_obj
= sa_obj
;
1021 mutex_enter(&sa
->sa_lock
);
1022 mutex_exit(&os
->os_user_ptr_lock
);
1023 avl_create(&sa
->sa_layout_num_tree
, layout_num_compare
,
1024 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_num_node
));
1025 avl_create(&sa
->sa_layout_hash_tree
, layout_hash_compare
,
1026 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_hash_node
));
1029 error
= zap_lookup(os
, sa_obj
, SA_LAYOUTS
,
1030 8, 1, &sa
->sa_layout_attr_obj
);
1031 if (error
!= 0 && error
!= ENOENT
)
1033 error
= zap_lookup(os
, sa_obj
, SA_REGISTRY
,
1034 8, 1, &sa
->sa_reg_attr_obj
);
1035 if (error
!= 0 && error
!= ENOENT
)
1039 if ((error
= sa_attr_table_setup(os
, reg_attrs
, count
)) != 0)
1042 if (sa
->sa_layout_attr_obj
!= 0) {
1043 uint64_t layout_count
;
1045 error
= zap_count(os
, sa
->sa_layout_attr_obj
,
1049 * Layout number count should be > 0
1051 if (error
|| (error
== 0 && layout_count
== 0)) {
1053 error
= SET_ERROR(EINVAL
);
1057 for (zap_cursor_init(&zc
, os
, sa
->sa_layout_attr_obj
);
1058 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
1059 zap_cursor_advance(&zc
)) {
1060 sa_attr_type_t
*lot_attrs
;
1063 lot_attrs
= kmem_zalloc(sizeof (sa_attr_type_t
) *
1064 za
.za_num_integers
, KM_SLEEP
);
1066 if ((error
= (zap_lookup(os
, sa
->sa_layout_attr_obj
,
1067 za
.za_name
, 2, za
.za_num_integers
,
1068 lot_attrs
))) != 0) {
1069 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1070 za
.za_num_integers
);
1073 VERIFY(ddi_strtoull(za
.za_name
, NULL
, 10,
1074 (unsigned long long *)&lot_num
) == 0);
1076 (void) sa_add_layout_entry(os
, lot_attrs
,
1077 za
.za_num_integers
, lot_num
,
1078 sa_layout_info_hash(lot_attrs
,
1079 za
.za_num_integers
), B_FALSE
, NULL
);
1080 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1081 za
.za_num_integers
);
1083 zap_cursor_fini(&zc
);
1086 * Make sure layout count matches number of entries added
1089 if (avl_numnodes(&sa
->sa_layout_num_tree
) != layout_count
) {
1095 /* Add special layout number for old ZNODES */
1096 if (ostype
== DMU_OST_ZFS
) {
1097 (void) sa_add_layout_entry(os
, sa_legacy_zpl_layout
,
1098 sa_legacy_attr_count
, 0,
1099 sa_layout_info_hash(sa_legacy_zpl_layout
,
1100 sa_legacy_attr_count
), B_FALSE
, NULL
);
1102 (void) sa_add_layout_entry(os
, sa_dummy_zpl_layout
, 0, 1,
1105 *user_table
= os
->os_sa
->sa_user_table
;
1106 mutex_exit(&sa
->sa_lock
);
1110 sa_free_attr_table(sa
);
1111 if (sa
->sa_user_table
)
1112 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1113 mutex_exit(&sa
->sa_lock
);
1114 avl_destroy(&sa
->sa_layout_hash_tree
);
1115 avl_destroy(&sa
->sa_layout_num_tree
);
1116 mutex_destroy(&sa
->sa_lock
);
1117 kmem_free(sa
, sizeof (sa_os_t
));
1118 return ((error
== ECKSUM
) ? EIO
: error
);
1122 sa_tear_down(objset_t
*os
)
1124 sa_os_t
*sa
= os
->os_sa
;
1128 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1130 /* Free up attr table */
1132 sa_free_attr_table(sa
);
1135 while (layout
= avl_destroy_nodes(&sa
->sa_layout_hash_tree
, &cookie
)) {
1137 while (tab
= list_head(&layout
->lot_idx_tab
)) {
1138 ASSERT(refcount_count(&tab
->sa_refcount
));
1139 sa_idx_tab_rele(os
, tab
);
1144 while (layout
= avl_destroy_nodes(&sa
->sa_layout_num_tree
, &cookie
)) {
1145 kmem_free(layout
->lot_attrs
,
1146 sizeof (sa_attr_type_t
) * layout
->lot_attr_count
);
1147 kmem_free(layout
, sizeof (sa_lot_t
));
1150 avl_destroy(&sa
->sa_layout_hash_tree
);
1151 avl_destroy(&sa
->sa_layout_num_tree
);
1152 mutex_destroy(&sa
->sa_lock
);
1154 kmem_free(sa
, sizeof (sa_os_t
));
1159 sa_build_idx_tab(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1160 uint16_t length
, int length_idx
, boolean_t var_length
, void *userp
)
1162 sa_idx_tab_t
*idx_tab
= userp
;
1165 ASSERT(idx_tab
->sa_variable_lengths
);
1166 idx_tab
->sa_variable_lengths
[length_idx
] = length
;
1168 TOC_ATTR_ENCODE(idx_tab
->sa_idx_tab
[attr
], length_idx
,
1169 (uint32_t)((uintptr_t)attr_addr
- (uintptr_t)hdr
));
1173 sa_attr_iter(objset_t
*os
, sa_hdr_phys_t
*hdr
, dmu_object_type_t type
,
1174 sa_iterfunc_t func
, sa_lot_t
*tab
, void *userp
)
1180 sa_os_t
*sa
= os
->os_sa
;
1182 uint16_t *length_start
= NULL
;
1183 uint8_t length_idx
= 0;
1186 search
.lot_num
= SA_LAYOUT_NUM(hdr
, type
);
1187 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1191 if (IS_SA_BONUSTYPE(type
)) {
1192 data_start
= (void *)P2ROUNDUP(((uintptr_t)hdr
+
1193 offsetof(sa_hdr_phys_t
, sa_lengths
) +
1194 (sizeof (uint16_t) * tb
->lot_var_sizes
)), 8);
1195 length_start
= hdr
->sa_lengths
;
1200 for (i
= 0; i
!= tb
->lot_attr_count
; i
++) {
1201 int attr_length
, reg_length
;
1204 reg_length
= sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
;
1206 attr_length
= reg_length
;
1209 attr_length
= length_start
[length_idx
];
1210 idx_len
= length_idx
++;
1213 func(hdr
, data_start
, tb
->lot_attrs
[i
], attr_length
,
1214 idx_len
, reg_length
== 0 ? B_TRUE
: B_FALSE
, userp
);
1216 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
1223 sa_byteswap_cb(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1224 uint16_t length
, int length_idx
, boolean_t variable_length
, void *userp
)
1226 sa_handle_t
*hdl
= userp
;
1227 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1229 sa_bswap_table
[sa
->sa_attr_table
[attr
].sa_byteswap
](attr_addr
, length
);
1233 sa_byteswap(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1235 sa_hdr_phys_t
*sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1237 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1238 int num_lengths
= 1;
1241 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1242 if (sa_hdr_phys
->sa_magic
== SA_MAGIC
)
1245 db
= SA_GET_DB(hdl
, buftype
);
1247 if (buftype
== SA_SPILL
) {
1248 arc_release(db
->db_buf
, NULL
);
1249 arc_buf_thaw(db
->db_buf
);
1252 sa_hdr_phys
->sa_magic
= BSWAP_32(sa_hdr_phys
->sa_magic
);
1253 sa_hdr_phys
->sa_layout_info
= BSWAP_16(sa_hdr_phys
->sa_layout_info
);
1256 * Determine number of variable lenghts in header
1257 * The standard 8 byte header has one for free and a
1258 * 16 byte header would have 4 + 1;
1260 if (SA_HDR_SIZE(sa_hdr_phys
) > 8)
1261 num_lengths
+= (SA_HDR_SIZE(sa_hdr_phys
) - 8) >> 1;
1262 for (i
= 0; i
!= num_lengths
; i
++)
1263 sa_hdr_phys
->sa_lengths
[i
] =
1264 BSWAP_16(sa_hdr_phys
->sa_lengths
[i
]);
1266 sa_attr_iter(hdl
->sa_os
, sa_hdr_phys
, DMU_OT_SA
,
1267 sa_byteswap_cb
, NULL
, hdl
);
1269 if (buftype
== SA_SPILL
)
1270 arc_buf_freeze(((dmu_buf_impl_t
*)hdl
->sa_spill
)->db_buf
);
1274 sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1276 sa_hdr_phys_t
*sa_hdr_phys
;
1277 dmu_buf_impl_t
*db
= SA_GET_DB(hdl
, buftype
);
1278 dmu_object_type_t bonustype
= SA_BONUSTYPE_FROM_DB(db
);
1279 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1280 sa_idx_tab_t
*idx_tab
;
1282 sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1284 mutex_enter(&sa
->sa_lock
);
1286 /* Do we need to byteswap? */
1288 /* only check if not old znode */
1289 if (IS_SA_BONUSTYPE(bonustype
) && sa_hdr_phys
->sa_magic
!= SA_MAGIC
&&
1290 sa_hdr_phys
->sa_magic
!= 0) {
1291 VERIFY(BSWAP_32(sa_hdr_phys
->sa_magic
) == SA_MAGIC
);
1292 sa_byteswap(hdl
, buftype
);
1295 idx_tab
= sa_find_idx_tab(hdl
->sa_os
, bonustype
, sa_hdr_phys
);
1297 if (buftype
== SA_BONUS
)
1298 hdl
->sa_bonus_tab
= idx_tab
;
1300 hdl
->sa_spill_tab
= idx_tab
;
1302 mutex_exit(&sa
->sa_lock
);
1310 panic("evicting sa dbuf\n");
1314 sa_idx_tab_rele(objset_t
*os
, void *arg
)
1316 sa_os_t
*sa
= os
->os_sa
;
1317 sa_idx_tab_t
*idx_tab
= arg
;
1319 if (idx_tab
== NULL
)
1322 mutex_enter(&sa
->sa_lock
);
1323 if (refcount_remove(&idx_tab
->sa_refcount
, NULL
) == 0) {
1324 list_remove(&idx_tab
->sa_layout
->lot_idx_tab
, idx_tab
);
1325 if (idx_tab
->sa_variable_lengths
)
1326 kmem_free(idx_tab
->sa_variable_lengths
,
1328 idx_tab
->sa_layout
->lot_var_sizes
);
1329 refcount_destroy(&idx_tab
->sa_refcount
);
1330 kmem_free(idx_tab
->sa_idx_tab
,
1331 sizeof (uint32_t) * sa
->sa_num_attrs
);
1332 kmem_free(idx_tab
, sizeof (sa_idx_tab_t
));
1334 mutex_exit(&sa
->sa_lock
);
1338 sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
)
1340 sa_os_t
*sa
= os
->os_sa
;
1342 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1343 (void) refcount_add(&idx_tab
->sa_refcount
, NULL
);
1347 sa_handle_destroy(sa_handle_t
*hdl
)
1349 dmu_buf_t
*db
= hdl
->sa_bonus
;
1351 mutex_enter(&hdl
->sa_lock
);
1352 (void) dmu_buf_remove_user(db
, &hdl
->sa_dbu
);
1354 if (hdl
->sa_bonus_tab
) {
1355 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
1356 hdl
->sa_bonus_tab
= NULL
;
1358 if (hdl
->sa_spill_tab
) {
1359 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1360 hdl
->sa_spill_tab
= NULL
;
1363 dmu_buf_rele(hdl
->sa_bonus
, NULL
);
1366 dmu_buf_rele((dmu_buf_t
*)hdl
->sa_spill
, NULL
);
1367 mutex_exit(&hdl
->sa_lock
);
1369 kmem_cache_free(sa_cache
, hdl
);
1373 sa_handle_get_from_db(objset_t
*os
, dmu_buf_t
*db
, void *userp
,
1374 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1377 dmu_object_info_t doi
;
1378 sa_handle_t
*handle
= NULL
;
1381 dmu_object_info_from_db(db
, &doi
);
1382 ASSERT(doi
.doi_bonus_type
== DMU_OT_SA
||
1383 doi
.doi_bonus_type
== DMU_OT_ZNODE
);
1385 /* find handle, if it exists */
1386 /* if one doesn't exist then create a new one, and initialize it */
1388 if (hdl_type
== SA_HDL_SHARED
)
1389 handle
= dmu_buf_get_user(db
);
1391 if (handle
== NULL
) {
1392 sa_handle_t
*winner
= NULL
;
1394 handle
= kmem_cache_alloc(sa_cache
, KM_SLEEP
);
1395 handle
->sa_userp
= userp
;
1396 handle
->sa_bonus
= db
;
1398 handle
->sa_spill
= NULL
;
1400 error
= sa_build_index(handle
, SA_BONUS
);
1402 if (hdl_type
== SA_HDL_SHARED
) {
1403 dmu_buf_init_user(&handle
->sa_dbu
, sa_evict
, NULL
);
1404 winner
= dmu_buf_set_user_ie(db
, &handle
->sa_dbu
);
1407 if (winner
!= NULL
) {
1408 kmem_cache_free(sa_cache
, handle
);
1418 sa_handle_get(objset_t
*objset
, uint64_t objid
, void *userp
,
1419 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1424 if (error
= dmu_bonus_hold(objset
, objid
, NULL
, &db
))
1427 return (sa_handle_get_from_db(objset
, db
, userp
, hdl_type
,
1432 sa_buf_hold(objset_t
*objset
, uint64_t obj_num
, void *tag
, dmu_buf_t
**db
)
1434 return (dmu_bonus_hold(objset
, obj_num
, tag
, db
));
1438 sa_buf_rele(dmu_buf_t
*db
, void *tag
)
1440 dmu_buf_rele(db
, tag
);
1444 sa_lookup_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
)
1447 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1448 return (sa_attr_op(hdl
, bulk
, count
, SA_LOOKUP
, NULL
));
1452 sa_lookup(sa_handle_t
*hdl
, sa_attr_type_t attr
, void *buf
, uint32_t buflen
)
1455 sa_bulk_attr_t bulk
;
1457 bulk
.sa_attr
= attr
;
1459 bulk
.sa_length
= buflen
;
1460 bulk
.sa_data_func
= NULL
;
1463 mutex_enter(&hdl
->sa_lock
);
1464 error
= sa_lookup_impl(hdl
, &bulk
, 1);
1465 mutex_exit(&hdl
->sa_lock
);
1471 sa_lookup_uio(sa_handle_t
*hdl
, sa_attr_type_t attr
, uio_t
*uio
)
1474 sa_bulk_attr_t bulk
;
1476 bulk
.sa_data
= NULL
;
1477 bulk
.sa_attr
= attr
;
1478 bulk
.sa_data_func
= NULL
;
1482 mutex_enter(&hdl
->sa_lock
);
1483 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) == 0) {
1484 error
= uiomove((void *)bulk
.sa_addr
, MIN(bulk
.sa_size
,
1485 uio
->uio_resid
), UIO_READ
, uio
);
1487 mutex_exit(&hdl
->sa_lock
);
1494 sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
, void *data
)
1496 sa_idx_tab_t
*idx_tab
;
1497 sa_hdr_phys_t
*hdr
= (sa_hdr_phys_t
*)data
;
1498 sa_os_t
*sa
= os
->os_sa
;
1499 sa_lot_t
*tb
, search
;
1503 * Deterimine layout number. If SA node and header == 0 then
1504 * force the index table to the dummy "1" empty layout.
1506 * The layout number would only be zero for a newly created file
1507 * that has not added any attributes yet, or with crypto enabled which
1508 * doesn't write any attributes to the bonus buffer.
1511 search
.lot_num
= SA_LAYOUT_NUM(hdr
, bonustype
);
1513 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1515 /* Verify header size is consistent with layout information */
1517 ASSERT(IS_SA_BONUSTYPE(bonustype
) &&
1518 SA_HDR_SIZE_MATCH_LAYOUT(hdr
, tb
) || !IS_SA_BONUSTYPE(bonustype
) ||
1519 (IS_SA_BONUSTYPE(bonustype
) && hdr
->sa_layout_info
== 0));
1522 * See if any of the already existing TOC entries can be reused?
1525 for (idx_tab
= list_head(&tb
->lot_idx_tab
); idx_tab
;
1526 idx_tab
= list_next(&tb
->lot_idx_tab
, idx_tab
)) {
1527 boolean_t valid_idx
= B_TRUE
;
1530 if (tb
->lot_var_sizes
!= 0 &&
1531 idx_tab
->sa_variable_lengths
!= NULL
) {
1532 for (i
= 0; i
!= tb
->lot_var_sizes
; i
++) {
1533 if (hdr
->sa_lengths
[i
] !=
1534 idx_tab
->sa_variable_lengths
[i
]) {
1535 valid_idx
= B_FALSE
;
1541 sa_idx_tab_hold(os
, idx_tab
);
1546 /* No such luck, create a new entry */
1547 idx_tab
= kmem_zalloc(sizeof (sa_idx_tab_t
), KM_SLEEP
);
1548 idx_tab
->sa_idx_tab
=
1549 kmem_zalloc(sizeof (uint32_t) * sa
->sa_num_attrs
, KM_SLEEP
);
1550 idx_tab
->sa_layout
= tb
;
1551 refcount_create(&idx_tab
->sa_refcount
);
1552 if (tb
->lot_var_sizes
)
1553 idx_tab
->sa_variable_lengths
= kmem_alloc(sizeof (uint16_t) *
1554 tb
->lot_var_sizes
, KM_SLEEP
);
1556 sa_attr_iter(os
, hdr
, bonustype
, sa_build_idx_tab
,
1558 sa_idx_tab_hold(os
, idx_tab
); /* one hold for consumer */
1559 sa_idx_tab_hold(os
, idx_tab
); /* one for layout */
1560 list_insert_tail(&tb
->lot_idx_tab
, idx_tab
);
1565 sa_default_locator(void **dataptr
, uint32_t *len
, uint32_t total_len
,
1566 boolean_t start
, void *userdata
)
1570 *dataptr
= userdata
;
1575 sa_attr_register_sync(sa_handle_t
*hdl
, dmu_tx_t
*tx
)
1577 uint64_t attr_value
= 0;
1578 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1579 sa_attr_table_t
*tb
= sa
->sa_attr_table
;
1582 mutex_enter(&sa
->sa_lock
);
1584 if (!sa
->sa_need_attr_registration
|| sa
->sa_master_obj
== NULL
) {
1585 mutex_exit(&sa
->sa_lock
);
1589 if (sa
->sa_reg_attr_obj
== NULL
) {
1590 sa
->sa_reg_attr_obj
= zap_create_link(hdl
->sa_os
,
1591 DMU_OT_SA_ATTR_REGISTRATION
,
1592 sa
->sa_master_obj
, SA_REGISTRY
, tx
);
1594 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
1595 if (sa
->sa_attr_table
[i
].sa_registered
)
1597 ATTR_ENCODE(attr_value
, tb
[i
].sa_attr
, tb
[i
].sa_length
,
1599 VERIFY(0 == zap_update(hdl
->sa_os
, sa
->sa_reg_attr_obj
,
1600 tb
[i
].sa_name
, 8, 1, &attr_value
, tx
));
1601 tb
[i
].sa_registered
= B_TRUE
;
1603 sa
->sa_need_attr_registration
= B_FALSE
;
1604 mutex_exit(&sa
->sa_lock
);
1608 * Replace all attributes with attributes specified in template.
1609 * If dnode had a spill buffer then those attributes will be
1610 * also be replaced, possibly with just an empty spill block
1612 * This interface is intended to only be used for bulk adding of
1613 * attributes for a new file. It will also be used by the ZPL
1614 * when converting and old formatted znode to native SA support.
1617 sa_replace_all_by_template_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1618 int attr_count
, dmu_tx_t
*tx
)
1620 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1622 if (sa
->sa_need_attr_registration
)
1623 sa_attr_register_sync(hdl
, tx
);
1624 return (sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
));
1628 sa_replace_all_by_template(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1629 int attr_count
, dmu_tx_t
*tx
)
1633 mutex_enter(&hdl
->sa_lock
);
1634 error
= sa_replace_all_by_template_locked(hdl
, attr_desc
,
1636 mutex_exit(&hdl
->sa_lock
);
1641 * add/remove/replace a single attribute and then rewrite the entire set
1645 sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
1646 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
1647 uint16_t buflen
, dmu_tx_t
*tx
)
1649 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1650 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)hdl
->sa_bonus
;
1652 sa_bulk_attr_t
*attr_desc
;
1654 int bonus_attr_count
= 0;
1655 int bonus_data_size
= 0;
1656 int spill_data_size
= 0;
1657 int spill_attr_count
= 0;
1660 int i
, j
, k
, length_idx
;
1662 sa_idx_tab_t
*idx_tab
;
1666 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1668 /* First make of copy of the old data */
1672 if (dn
->dn_bonuslen
!= 0) {
1673 bonus_data_size
= hdl
->sa_bonus
->db_size
;
1674 old_data
[0] = kmem_alloc(bonus_data_size
, KM_SLEEP
);
1675 bcopy(hdl
->sa_bonus
->db_data
, old_data
[0],
1676 hdl
->sa_bonus
->db_size
);
1677 bonus_attr_count
= hdl
->sa_bonus_tab
->sa_layout
->lot_attr_count
;
1683 /* Bring spill buffer online if it isn't currently */
1685 if ((error
= sa_get_spill(hdl
)) == 0) {
1686 spill_data_size
= hdl
->sa_spill
->db_size
;
1687 old_data
[1] = kmem_alloc(spill_data_size
, KM_SLEEP
);
1688 bcopy(hdl
->sa_spill
->db_data
, old_data
[1],
1689 hdl
->sa_spill
->db_size
);
1691 hdl
->sa_spill_tab
->sa_layout
->lot_attr_count
;
1692 } else if (error
&& error
!= ENOENT
) {
1694 kmem_free(old_data
[0], bonus_data_size
);
1700 /* build descriptor of all attributes */
1702 attr_count
= bonus_attr_count
+ spill_attr_count
;
1703 if (action
== SA_ADD
)
1705 else if (action
== SA_REMOVE
)
1708 attr_desc
= kmem_zalloc(sizeof (sa_bulk_attr_t
) * attr_count
, KM_SLEEP
);
1711 * loop through bonus and spill buffer if it exists, and
1712 * build up new attr_descriptor to reset the attributes
1715 count
= bonus_attr_count
;
1716 hdr
= SA_GET_HDR(hdl
, SA_BONUS
);
1717 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_BONUS
);
1718 for (; k
!= 2; k
++) {
1719 /* iterate over each attribute in layout */
1720 for (i
= 0, length_idx
= 0; i
!= count
; i
++) {
1721 sa_attr_type_t attr
;
1723 attr
= idx_tab
->sa_layout
->lot_attrs
[i
];
1724 if (attr
== newattr
) {
1725 if (action
== SA_REMOVE
) {
1729 ASSERT(SA_REGISTERED_LEN(sa
, attr
) == 0);
1730 ASSERT(action
== SA_REPLACE
);
1731 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1732 locator
, datastart
, buflen
);
1734 length
= SA_REGISTERED_LEN(sa
, attr
);
1736 length
= hdr
->sa_lengths
[length_idx
++];
1739 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1741 (TOC_OFF(idx_tab
->sa_idx_tab
[attr
]) +
1742 (uintptr_t)old_data
[k
]), length
);
1745 if (k
== 0 && hdl
->sa_spill
) {
1746 hdr
= SA_GET_HDR(hdl
, SA_SPILL
);
1747 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_SPILL
);
1748 count
= spill_attr_count
;
1753 if (action
== SA_ADD
) {
1754 length
= SA_REGISTERED_LEN(sa
, newattr
);
1758 SA_ADD_BULK_ATTR(attr_desc
, j
, newattr
, locator
,
1762 error
= sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
);
1765 kmem_free(old_data
[0], bonus_data_size
);
1767 kmem_free(old_data
[1], spill_data_size
);
1768 kmem_free(attr_desc
, sizeof (sa_bulk_attr_t
) * attr_count
);
1774 sa_bulk_update_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
1778 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1779 dmu_object_type_t bonustype
;
1781 bonustype
= SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl
, SA_BONUS
));
1784 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1786 /* sync out registration table if necessary */
1787 if (sa
->sa_need_attr_registration
)
1788 sa_attr_register_sync(hdl
, tx
);
1790 error
= sa_attr_op(hdl
, bulk
, count
, SA_UPDATE
, tx
);
1791 if (error
== 0 && !IS_SA_BONUSTYPE(bonustype
) && sa
->sa_update_cb
)
1792 sa
->sa_update_cb(hdl
, tx
);
1798 * update or add new attribute
1801 sa_update(sa_handle_t
*hdl
, sa_attr_type_t type
,
1802 void *buf
, uint32_t buflen
, dmu_tx_t
*tx
)
1805 sa_bulk_attr_t bulk
;
1807 bulk
.sa_attr
= type
;
1808 bulk
.sa_data_func
= NULL
;
1809 bulk
.sa_length
= buflen
;
1812 mutex_enter(&hdl
->sa_lock
);
1813 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1814 mutex_exit(&hdl
->sa_lock
);
1819 sa_update_from_cb(sa_handle_t
*hdl
, sa_attr_type_t attr
,
1820 uint32_t buflen
, sa_data_locator_t
*locator
, void *userdata
, dmu_tx_t
*tx
)
1823 sa_bulk_attr_t bulk
;
1825 bulk
.sa_attr
= attr
;
1826 bulk
.sa_data
= userdata
;
1827 bulk
.sa_data_func
= locator
;
1828 bulk
.sa_length
= buflen
;
1830 mutex_enter(&hdl
->sa_lock
);
1831 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1832 mutex_exit(&hdl
->sa_lock
);
1837 * Return size of an attribute
1841 sa_size(sa_handle_t
*hdl
, sa_attr_type_t attr
, int *size
)
1843 sa_bulk_attr_t bulk
;
1846 bulk
.sa_data
= NULL
;
1847 bulk
.sa_attr
= attr
;
1848 bulk
.sa_data_func
= NULL
;
1851 mutex_enter(&hdl
->sa_lock
);
1852 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) != 0) {
1853 mutex_exit(&hdl
->sa_lock
);
1856 *size
= bulk
.sa_size
;
1858 mutex_exit(&hdl
->sa_lock
);
1863 sa_bulk_lookup_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1866 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1867 return (sa_lookup_impl(hdl
, attrs
, count
));
1871 sa_bulk_lookup(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1876 mutex_enter(&hdl
->sa_lock
);
1877 error
= sa_bulk_lookup_locked(hdl
, attrs
, count
);
1878 mutex_exit(&hdl
->sa_lock
);
1883 sa_bulk_update(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
, dmu_tx_t
*tx
)
1888 mutex_enter(&hdl
->sa_lock
);
1889 error
= sa_bulk_update_impl(hdl
, attrs
, count
, tx
);
1890 mutex_exit(&hdl
->sa_lock
);
1895 sa_remove(sa_handle_t
*hdl
, sa_attr_type_t attr
, dmu_tx_t
*tx
)
1899 mutex_enter(&hdl
->sa_lock
);
1900 error
= sa_modify_attrs(hdl
, attr
, SA_REMOVE
, NULL
,
1902 mutex_exit(&hdl
->sa_lock
);
1907 sa_object_info(sa_handle_t
*hdl
, dmu_object_info_t
*doi
)
1909 dmu_object_info_from_db((dmu_buf_t
*)hdl
->sa_bonus
, doi
);
1913 sa_object_size(sa_handle_t
*hdl
, uint32_t *blksize
, u_longlong_t
*nblocks
)
1915 dmu_object_size_from_db((dmu_buf_t
*)hdl
->sa_bonus
,
1920 sa_set_userp(sa_handle_t
*hdl
, void *ptr
)
1922 hdl
->sa_userp
= ptr
;
1926 sa_get_db(sa_handle_t
*hdl
)
1928 return ((dmu_buf_t
*)hdl
->sa_bonus
);
1932 sa_get_userdata(sa_handle_t
*hdl
)
1934 return (hdl
->sa_userp
);
1938 sa_register_update_callback_locked(objset_t
*os
, sa_update_cb_t
*func
)
1940 ASSERT(MUTEX_HELD(&os
->os_sa
->sa_lock
));
1941 os
->os_sa
->sa_update_cb
= func
;
1945 sa_register_update_callback(objset_t
*os
, sa_update_cb_t
*func
)
1948 mutex_enter(&os
->os_sa
->sa_lock
);
1949 sa_register_update_callback_locked(os
, func
);
1950 mutex_exit(&os
->os_sa
->sa_lock
);
1954 sa_handle_object(sa_handle_t
*hdl
)
1956 return (hdl
->sa_bonus
->db_object
);
1960 sa_enabled(objset_t
*os
)
1962 return (os
->os_sa
== NULL
);
1966 sa_set_sa_object(objset_t
*os
, uint64_t sa_object
)
1968 sa_os_t
*sa
= os
->os_sa
;
1970 if (sa
->sa_master_obj
)
1973 sa
->sa_master_obj
= sa_object
;
1979 sa_hdrsize(void *arg
)
1981 sa_hdr_phys_t
*hdr
= arg
;
1983 return (SA_HDR_SIZE(hdr
));
1987 sa_handle_lock(sa_handle_t
*hdl
)
1990 mutex_enter(&hdl
->sa_lock
);
1994 sa_handle_unlock(sa_handle_t
*hdl
)
1997 mutex_exit(&hdl
->sa_lock
);