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.
28 #include <sys/zfs_context.h>
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
34 #include <sys/dmu_impl.h>
35 #include <sys/dmu_objset.h>
37 #include <sys/dnode.h>
40 #include <sys/sunddi.h>
41 #include <sys/sa_impl.h>
42 #include <sys/dnode.h>
43 #include <sys/errno.h>
44 #include <sys/zfs_context.h>
47 * ZFS System attributes:
49 * A generic mechanism to allow for arbitrary attributes
50 * to be stored in a dnode. The data will be stored in the bonus buffer of
51 * the dnode and if necessary a special "spill" block will be used to handle
52 * overflow situations. The spill block will be sized to fit the data
53 * from 512 - 128K. When a spill block is used the BP (blkptr_t) for the
54 * spill block is stored at the end of the current bonus buffer. Any
55 * attributes that would be in the way of the blkptr_t will be relocated
56 * into the spill block.
58 * Attribute registration:
60 * Stored persistently on a per dataset basis
61 * a mapping between attribute "string" names and their actual attribute
62 * numeric values, length, and byteswap function. The names are only used
63 * during registration. All attributes are known by their unique attribute
64 * id value. If an attribute can have a variable size then the value
65 * 0 will be used to indicate this.
69 * Attribute layouts are a way to compactly store multiple attributes, but
70 * without taking the overhead associated with managing each attribute
71 * individually. Since you will typically have the same set of attributes
72 * stored in the same order a single table will be used to represent that
73 * layout. The ZPL for example will usually have only about 10 different
74 * layouts (regular files, device files, symlinks,
75 * regular files + scanstamp, files/dir with extended attributes, and then
76 * you have the possibility of all of those minus ACL, because it would
77 * be kicked out into the spill block)
79 * Layouts are simply an array of the attributes and their
80 * ordering i.e. [0, 1, 4, 5, 2]
82 * Each distinct layout is given a unique layout number and that is whats
83 * stored in the header at the beginning of the SA data buffer.
85 * A layout only covers a single dbuf (bonus or spill). If a set of
86 * attributes is split up between the bonus buffer and a spill buffer then
87 * two different layouts will be used. This allows us to byteswap the
88 * spill without looking at the bonus buffer and keeps the on disk format of
89 * the bonus and spill buffer the same.
91 * Adding a single attribute will cause the entire set of attributes to
92 * be rewritten and could result in a new layout number being constructed
93 * as part of the rewrite if no such layout exists for the new set of
94 * attribues. The new attribute will be appended to the end of the already
95 * existing attributes.
97 * Both the attribute registration and attribute layout information are
98 * stored in normal ZAP attributes. Their should be a small number of
99 * known layouts and the set of attributes is assumed to typically be quite
102 * The registered attributes and layout "table" information is maintained
103 * in core and a special "sa_os_t" is attached to the objset_t.
105 * A special interface is provided to allow for quickly applying
106 * a large set of attributes at once. sa_replace_all_by_template() is
107 * used to set an array of attributes. This is used by the ZPL when
108 * creating a brand new file. The template that is passed into the function
109 * specifies the attribute, size for variable length attributes, location of
110 * data and special "data locator" function if the data isn't in a contiguous
113 * Byteswap implications:
115 * Since the SA attributes are not entirely self describing we can't do
116 * the normal byteswap processing. The special ZAP layout attribute and
117 * attribute registration attributes define the byteswap function and the
118 * size of the attributes, unless it is variable sized.
119 * The normal ZFS byteswapping infrastructure assumes you don't need
120 * to read any objects in order to do the necessary byteswapping. Whereas
121 * SA attributes can only be properly byteswapped if the dataset is opened
122 * and the layout/attribute ZAP attributes are available. Because of this
123 * the SA attributes will be byteswapped when they are first accessed by
124 * the SA code that will read the SA data.
127 typedef void (sa_iterfunc_t
)(void *hdr
, void *addr
, sa_attr_type_t
,
128 uint16_t length
, int length_idx
, boolean_t
, void *userp
);
130 static int sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
);
131 static void sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
);
132 static void *sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
,
134 static void sa_idx_tab_rele(objset_t
*os
, void *arg
);
135 static void sa_copy_data(sa_data_locator_t
*func
, void *start
, void *target
,
137 static int sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
138 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
139 uint16_t buflen
, dmu_tx_t
*tx
);
141 arc_byteswap_func_t
*sa_bswap_table
[] = {
142 byteswap_uint64_array
,
143 byteswap_uint32_array
,
144 byteswap_uint16_array
,
145 byteswap_uint8_array
,
149 #define SA_COPY_DATA(f, s, t, l) \
153 *(uint64_t *)t = *(uint64_t *)s; \
154 } else if (l == 16) { \
155 *(uint64_t *)t = *(uint64_t *)s; \
156 *(uint64_t *)((uintptr_t)t + 8) = \
157 *(uint64_t *)((uintptr_t)s + 8); \
162 sa_copy_data(f, s, t, l); \
166 * This table is fixed and cannot be changed. Its purpose is to
167 * allow the SA code to work with both old/new ZPL file systems.
168 * It contains the list of legacy attributes. These attributes aren't
169 * stored in the "attribute" registry zap objects, since older ZPL file systems
170 * won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
171 * use this static table.
173 sa_attr_reg_t sa_legacy_attrs
[] = {
174 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 0},
175 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 1},
176 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 2},
177 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 3},
178 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY
, 4},
179 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY
, 5},
180 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY
, 6},
181 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY
, 7},
182 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY
, 8},
183 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY
, 9},
184 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY
, 10},
185 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY
, 11},
186 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY
, 12},
187 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY
, 13},
188 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY
, 14},
189 {"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY
, 15},
193 * This is only used for objects of type DMU_OT_ZNODE
195 sa_attr_type_t sa_legacy_zpl_layout
[] = {
196 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
200 * Special dummy layout used for buffers with no attributes.
202 sa_attr_type_t sa_dummy_zpl_layout
[] = { 0 };
204 static int sa_legacy_attr_count
= 16;
205 static kmem_cache_t
*sa_cache
= NULL
;
209 sa_cache_constructor(void *buf
, void *unused
, int kmflag
)
211 sa_handle_t
*hdl
= buf
;
213 hdl
->sa_bonus_tab
= NULL
;
214 hdl
->sa_spill_tab
= NULL
;
216 hdl
->sa_userp
= NULL
;
217 hdl
->sa_bonus
= NULL
;
218 hdl
->sa_spill
= NULL
;
219 mutex_init(&hdl
->sa_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
225 sa_cache_destructor(void *buf
, void *unused
)
227 sa_handle_t
*hdl
= buf
;
228 mutex_destroy(&hdl
->sa_lock
);
234 sa_cache
= kmem_cache_create("sa_cache",
235 sizeof (sa_handle_t
), 0, sa_cache_constructor
,
236 sa_cache_destructor
, NULL
, NULL
, NULL
, 0);
243 kmem_cache_destroy(sa_cache
);
247 layout_num_compare(const void *arg1
, const void *arg2
)
249 const sa_lot_t
*node1
= arg1
;
250 const sa_lot_t
*node2
= arg2
;
252 if (node1
->lot_num
> node2
->lot_num
)
254 else if (node1
->lot_num
< node2
->lot_num
)
260 layout_hash_compare(const void *arg1
, const void *arg2
)
262 const sa_lot_t
*node1
= arg1
;
263 const sa_lot_t
*node2
= arg2
;
265 if (node1
->lot_hash
> node2
->lot_hash
)
267 if (node1
->lot_hash
< node2
->lot_hash
)
269 if (node1
->lot_instance
> node2
->lot_instance
)
271 if (node1
->lot_instance
< node2
->lot_instance
)
277 sa_layout_equal(sa_lot_t
*tbf
, sa_attr_type_t
*attrs
, int count
)
281 if (count
!= tbf
->lot_attr_count
)
284 for (i
= 0; i
!= count
; i
++) {
285 if (attrs
[i
] != tbf
->lot_attrs
[i
])
291 #define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
294 sa_layout_info_hash(sa_attr_type_t
*attrs
, int attr_count
)
297 uint64_t crc
= -1ULL;
299 for (i
= 0; i
!= attr_count
; i
++)
300 crc
^= SA_ATTR_HASH(attrs
[i
]);
306 sa_get_spill(sa_handle_t
*hdl
)
309 if (hdl
->sa_spill
== NULL
) {
310 if ((rc
= dmu_spill_hold_existing(hdl
->sa_bonus
, NULL
,
311 &hdl
->sa_spill
)) == 0)
312 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
321 * Main attribute lookup/update function
322 * returns 0 for success or non zero for failures
324 * Operates on bulk array, first failure will abort further processing
327 sa_attr_op(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
328 sa_data_op_t data_op
, dmu_tx_t
*tx
)
330 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
333 sa_buf_type_t buftypes
;
338 for (i
= 0; i
!= count
; i
++) {
339 ASSERT(bulk
[i
].sa_attr
<= hdl
->sa_os
->os_sa
->sa_num_attrs
);
341 bulk
[i
].sa_addr
= NULL
;
342 /* First check the bonus buffer */
344 if (hdl
->sa_bonus_tab
&& TOC_ATTR_PRESENT(
345 hdl
->sa_bonus_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
346 SA_ATTR_INFO(sa
, hdl
->sa_bonus_tab
,
347 SA_GET_HDR(hdl
, SA_BONUS
),
348 bulk
[i
].sa_attr
, bulk
[i
], SA_BONUS
, hdl
);
349 if (tx
&& !(buftypes
& SA_BONUS
)) {
350 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
351 buftypes
|= SA_BONUS
;
354 if (bulk
[i
].sa_addr
== NULL
&&
355 ((error
= sa_get_spill(hdl
)) == 0)) {
356 if (TOC_ATTR_PRESENT(
357 hdl
->sa_spill_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
358 SA_ATTR_INFO(sa
, hdl
->sa_spill_tab
,
359 SA_GET_HDR(hdl
, SA_SPILL
),
360 bulk
[i
].sa_attr
, bulk
[i
], SA_SPILL
, hdl
);
361 if (tx
&& !(buftypes
& SA_SPILL
) &&
362 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
363 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
364 buftypes
|= SA_SPILL
;
368 if (error
&& error
!= ENOENT
) {
369 return ((error
== ECKSUM
) ? EIO
: error
);
374 if (bulk
[i
].sa_addr
== NULL
)
375 return (SET_ERROR(ENOENT
));
376 if (bulk
[i
].sa_data
) {
377 SA_COPY_DATA(bulk
[i
].sa_data_func
,
378 bulk
[i
].sa_addr
, bulk
[i
].sa_data
,
384 /* existing rewrite of attr */
385 if (bulk
[i
].sa_addr
&&
386 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
387 SA_COPY_DATA(bulk
[i
].sa_data_func
,
388 bulk
[i
].sa_data
, bulk
[i
].sa_addr
,
391 } else if (bulk
[i
].sa_addr
) { /* attr size change */
392 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
393 SA_REPLACE
, bulk
[i
].sa_data_func
,
394 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
395 } else { /* adding new attribute */
396 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
397 SA_ADD
, bulk
[i
].sa_data_func
,
398 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
409 sa_add_layout_entry(objset_t
*os
, sa_attr_type_t
*attrs
, int attr_count
,
410 uint64_t lot_num
, uint64_t hash
, boolean_t zapadd
, dmu_tx_t
*tx
)
412 sa_os_t
*sa
= os
->os_sa
;
413 sa_lot_t
*tb
, *findtb
;
417 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
418 tb
= kmem_zalloc(sizeof (sa_lot_t
), KM_SLEEP
);
419 tb
->lot_attr_count
= attr_count
;
420 tb
->lot_attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
422 bcopy(attrs
, tb
->lot_attrs
, sizeof (sa_attr_type_t
) * attr_count
);
423 tb
->lot_num
= lot_num
;
425 tb
->lot_instance
= 0;
430 if (sa
->sa_layout_attr_obj
== 0) {
431 sa
->sa_layout_attr_obj
= zap_create_link(os
,
432 DMU_OT_SA_ATTR_LAYOUTS
,
433 sa
->sa_master_obj
, SA_LAYOUTS
, tx
);
436 (void) snprintf(attr_name
, sizeof (attr_name
),
438 VERIFY(0 == zap_update(os
, os
->os_sa
->sa_layout_attr_obj
,
439 attr_name
, 2, attr_count
, attrs
, tx
));
442 list_create(&tb
->lot_idx_tab
, sizeof (sa_idx_tab_t
),
443 offsetof(sa_idx_tab_t
, sa_next
));
445 for (i
= 0; i
!= attr_count
; i
++) {
446 if (sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
== 0)
450 avl_add(&sa
->sa_layout_num_tree
, tb
);
452 /* verify we don't have a hash collision */
453 if ((findtb
= avl_find(&sa
->sa_layout_hash_tree
, tb
, &loc
)) != NULL
) {
454 for (; findtb
&& findtb
->lot_hash
== hash
;
455 findtb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, findtb
)) {
456 if (findtb
->lot_instance
!= tb
->lot_instance
)
461 avl_add(&sa
->sa_layout_hash_tree
, tb
);
466 sa_find_layout(objset_t
*os
, uint64_t hash
, sa_attr_type_t
*attrs
,
467 int count
, dmu_tx_t
*tx
, sa_lot_t
**lot
)
469 sa_lot_t
*tb
, tbsearch
;
471 sa_os_t
*sa
= os
->os_sa
;
472 boolean_t found
= B_FALSE
;
474 mutex_enter(&sa
->sa_lock
);
475 tbsearch
.lot_hash
= hash
;
476 tbsearch
.lot_instance
= 0;
477 tb
= avl_find(&sa
->sa_layout_hash_tree
, &tbsearch
, &loc
);
479 for (; tb
&& tb
->lot_hash
== hash
;
480 tb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, tb
)) {
481 if (sa_layout_equal(tb
, attrs
, count
) == 0) {
488 tb
= sa_add_layout_entry(os
, attrs
, count
,
489 avl_numnodes(&sa
->sa_layout_num_tree
), hash
, B_TRUE
, tx
);
491 mutex_exit(&sa
->sa_lock
);
496 sa_resize_spill(sa_handle_t
*hdl
, uint32_t size
, dmu_tx_t
*tx
)
502 blocksize
= SPA_MINBLOCKSIZE
;
503 } else if (size
> SPA_OLD_MAXBLOCKSIZE
) {
505 return (SET_ERROR(EFBIG
));
507 blocksize
= P2ROUNDUP_TYPED(size
, SPA_MINBLOCKSIZE
, uint32_t);
510 error
= dbuf_spill_set_blksz(hdl
->sa_spill
, blocksize
, tx
);
516 sa_copy_data(sa_data_locator_t
*func
, void *datastart
, void *target
, int buflen
)
519 bcopy(datastart
, target
, buflen
);
524 void *saptr
= target
;
529 while (bytes
< buflen
) {
530 func(&dataptr
, &length
, buflen
, start
, datastart
);
531 bcopy(dataptr
, saptr
, length
);
532 saptr
= (void *)((caddr_t
)saptr
+ length
);
540 * Determine several different sizes
541 * first the sa header size
542 * the number of bytes to be stored
543 * if spill would occur the index in the attribute array is returned
545 * the boolean will_spill will be set when spilling is necessary. It
546 * is only set when the buftype is SA_BONUS
549 sa_find_sizes(sa_os_t
*sa
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
550 dmu_buf_t
*db
, sa_buf_type_t buftype
, int *index
, int *total
,
551 boolean_t
*will_spill
)
558 boolean_t done
= B_FALSE
;
560 if (buftype
== SA_BONUS
&& sa
->sa_force_spill
) {
563 *will_spill
= B_TRUE
;
570 if (buftype
== SA_BONUS
)
571 *will_spill
= B_FALSE
;
573 hdrsize
= (SA_BONUSTYPE_FROM_DB(db
) == DMU_OT_ZNODE
) ? 0 :
574 sizeof (sa_hdr_phys_t
);
576 full_space
= (buftype
== SA_BONUS
) ? DN_MAX_BONUSLEN
: db
->db_size
;
577 ASSERT(IS_P2ALIGNED(full_space
, 8));
579 for (i
= 0; i
!= attr_count
; i
++) {
582 *total
= P2ROUNDUP(*total
, 8);
583 *total
+= attr_desc
[i
].sa_length
;
587 is_var_sz
= (SA_REGISTERED_LEN(sa
, attr_desc
[i
].sa_attr
) == 0);
592 if (is_var_sz
&& var_size
> 1) {
593 if (P2ROUNDUP(hdrsize
+ sizeof (uint16_t), 8) +
594 *total
< full_space
) {
596 * Account for header space used by array of
597 * optional sizes of variable-length attributes.
598 * Record the index in case this increase needs
599 * to be reversed due to spill-over.
601 hdrsize
+= sizeof (uint16_t);
606 if (buftype
== SA_BONUS
)
607 *will_spill
= B_TRUE
;
613 * find index of where spill *could* occur.
614 * Then continue to count of remainder attribute
615 * space. The sum is used later for sizing bonus
618 if (buftype
== SA_BONUS
&& *index
== -1 &&
619 *total
+ P2ROUNDUP(hdrsize
, 8) >
620 (full_space
- sizeof (blkptr_t
))) {
626 if (*total
+ P2ROUNDUP(hdrsize
, 8) > full_space
&&
628 *will_spill
= B_TRUE
;
632 * j holds the index of the last variable-sized attribute for
633 * which hdrsize was increased. Reverse the increase if that
634 * attribute will be relocated to the spill block.
636 if (*will_spill
&& j
== *index
)
637 hdrsize
-= sizeof (uint16_t);
639 hdrsize
= P2ROUNDUP(hdrsize
, 8);
643 #define BUF_SPACE_NEEDED(total, header) (total + header)
646 * Find layout that corresponds to ordering of attributes
647 * If not found a new layout number is created and added to
648 * persistent layout tables.
651 sa_build_layouts(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
654 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
656 sa_buf_type_t buftype
;
657 sa_hdr_phys_t
*sahdr
;
660 sa_attr_type_t
*attrs
, *attrs_start
;
663 int spillhdrsize
= 0;
665 dmu_object_type_t bonustype
;
671 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
672 bonustype
= SA_BONUSTYPE_FROM_DB(hdl
->sa_bonus
);
674 /* first determine bonus header size and sum of all attributes */
675 hdrsize
= sa_find_sizes(sa
, attr_desc
, attr_count
, hdl
->sa_bonus
,
676 SA_BONUS
, &i
, &used
, &spilling
);
678 if (used
> SPA_OLD_MAXBLOCKSIZE
)
679 return (SET_ERROR(EFBIG
));
681 VERIFY(0 == dmu_set_bonus(hdl
->sa_bonus
, spilling
?
682 MIN(DN_MAX_BONUSLEN
- sizeof (blkptr_t
), used
+ hdrsize
) :
683 used
+ hdrsize
, tx
));
685 ASSERT((bonustype
== DMU_OT_ZNODE
&& spilling
== 0) ||
686 bonustype
== DMU_OT_SA
);
688 /* setup and size spill buffer when needed */
692 if (hdl
->sa_spill
== NULL
) {
693 VERIFY(dmu_spill_hold_by_bonus(hdl
->sa_bonus
, NULL
,
694 &hdl
->sa_spill
) == 0);
696 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
698 spillhdrsize
= sa_find_sizes(sa
, &attr_desc
[i
],
699 attr_count
- i
, hdl
->sa_spill
, SA_SPILL
, &i
,
700 &spill_used
, &dummy
);
702 if (spill_used
> SPA_OLD_MAXBLOCKSIZE
)
703 return (SET_ERROR(EFBIG
));
705 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
706 if (BUF_SPACE_NEEDED(spill_used
, spillhdrsize
) >
707 hdl
->sa_spill
->db_size
)
708 VERIFY(0 == sa_resize_spill(hdl
,
709 BUF_SPACE_NEEDED(spill_used
, spillhdrsize
), tx
));
712 /* setup starting pointers to lay down data */
713 data_start
= (void *)((uintptr_t)hdl
->sa_bonus
->db_data
+ hdrsize
);
714 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_bonus
->db_data
;
718 buf_space
= (sa
->sa_force_spill
) ?
719 0 : SA_BLKPTR_SPACE
- hdrsize
;
721 buf_space
= hdl
->sa_bonus
->db_size
- hdrsize
;
723 attrs_start
= attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
727 for (i
= 0, len_idx
= 0, hash
= -1ULL; i
!= attr_count
; i
++) {
730 ASSERT(IS_P2ALIGNED(data_start
, 8));
731 ASSERT(IS_P2ALIGNED(buf_space
, 8));
732 attrs
[i
] = attr_desc
[i
].sa_attr
;
733 length
= SA_REGISTERED_LEN(sa
, attrs
[i
]);
735 length
= attr_desc
[i
].sa_length
;
737 if (buf_space
< length
) { /* switch to spill buffer */
739 VERIFY(bonustype
== DMU_OT_SA
);
740 if (buftype
== SA_BONUS
&& !sa
->sa_force_spill
) {
741 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
,
742 lot_count
, tx
, &lot
);
743 SA_SET_HDR(sahdr
, lot
->lot_num
, hdrsize
);
750 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_spill
->db_data
;
751 sahdr
->sa_magic
= SA_MAGIC
;
752 data_start
= (void *)((uintptr_t)sahdr
+
754 attrs_start
= &attrs
[i
];
755 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
758 hash
^= SA_ATTR_HASH(attrs
[i
]);
759 attr_desc
[i
].sa_addr
= data_start
;
760 attr_desc
[i
].sa_size
= length
;
761 SA_COPY_DATA(attr_desc
[i
].sa_data_func
, attr_desc
[i
].sa_data
,
763 if (sa
->sa_attr_table
[attrs
[i
]].sa_length
== 0) {
764 sahdr
->sa_lengths
[len_idx
++] = length
;
766 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
768 buf_space
-= P2ROUNDUP(length
, 8);
772 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
, lot_count
, tx
, &lot
);
775 * Verify that old znodes always have layout number 0.
776 * Must be DMU_OT_SA for arbitrary layouts
778 VERIFY((bonustype
== DMU_OT_ZNODE
&& lot
->lot_num
== 0) ||
779 (bonustype
== DMU_OT_SA
&& lot
->lot_num
> 1));
781 if (bonustype
== DMU_OT_SA
) {
782 SA_SET_HDR(sahdr
, lot
->lot_num
,
783 buftype
== SA_BONUS
? hdrsize
: spillhdrsize
);
786 kmem_free(attrs
, sizeof (sa_attr_type_t
) * attr_count
);
787 if (hdl
->sa_bonus_tab
) {
788 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
789 hdl
->sa_bonus_tab
= NULL
;
791 if (!sa
->sa_force_spill
)
792 VERIFY(0 == sa_build_index(hdl
, SA_BONUS
));
794 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
797 * remove spill block that is no longer needed.
799 dmu_buf_rele(hdl
->sa_spill
, NULL
);
800 hdl
->sa_spill
= NULL
;
801 hdl
->sa_spill_tab
= NULL
;
802 VERIFY(0 == dmu_rm_spill(hdl
->sa_os
,
803 sa_handle_object(hdl
), tx
));
805 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
813 sa_free_attr_table(sa_os_t
*sa
)
817 if (sa
->sa_attr_table
== NULL
)
820 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
821 if (sa
->sa_attr_table
[i
].sa_name
)
822 kmem_free(sa
->sa_attr_table
[i
].sa_name
,
823 strlen(sa
->sa_attr_table
[i
].sa_name
) + 1);
826 kmem_free(sa
->sa_attr_table
,
827 sizeof (sa_attr_table_t
) * sa
->sa_num_attrs
);
829 sa
->sa_attr_table
= NULL
;
833 sa_attr_table_setup(objset_t
*os
, sa_attr_reg_t
*reg_attrs
, int count
)
835 sa_os_t
*sa
= os
->os_sa
;
836 uint64_t sa_attr_count
= 0;
837 uint64_t sa_reg_count
= 0;
843 int registered_count
= 0;
845 dmu_objset_type_t ostype
= dmu_objset_type(os
);
848 kmem_zalloc(count
* sizeof (sa_attr_type_t
), KM_SLEEP
);
849 sa
->sa_user_table_sz
= count
* sizeof (sa_attr_type_t
);
851 if (sa
->sa_reg_attr_obj
!= 0) {
852 error
= zap_count(os
, sa
->sa_reg_attr_obj
,
856 * Make sure we retrieved a count and that it isn't zero
858 if (error
|| (error
== 0 && sa_attr_count
== 0)) {
860 error
= SET_ERROR(EINVAL
);
863 sa_reg_count
= sa_attr_count
;
866 if (ostype
== DMU_OST_ZFS
&& sa_attr_count
== 0)
867 sa_attr_count
+= sa_legacy_attr_count
;
869 /* Allocate attribute numbers for attributes that aren't registered */
870 for (i
= 0; i
!= count
; i
++) {
871 boolean_t found
= B_FALSE
;
874 if (ostype
== DMU_OST_ZFS
) {
875 for (j
= 0; j
!= sa_legacy_attr_count
; j
++) {
876 if (strcmp(reg_attrs
[i
].sa_name
,
877 sa_legacy_attrs
[j
].sa_name
) == 0) {
878 sa
->sa_user_table
[i
] =
879 sa_legacy_attrs
[j
].sa_attr
;
887 if (sa
->sa_reg_attr_obj
)
888 error
= zap_lookup(os
, sa
->sa_reg_attr_obj
,
889 reg_attrs
[i
].sa_name
, 8, 1, &attr_value
);
891 error
= SET_ERROR(ENOENT
);
894 sa
->sa_user_table
[i
] = (sa_attr_type_t
)sa_attr_count
;
898 sa
->sa_user_table
[i
] = ATTR_NUM(attr_value
);
905 sa
->sa_num_attrs
= sa_attr_count
;
906 tb
= sa
->sa_attr_table
=
907 kmem_zalloc(sizeof (sa_attr_table_t
) * sa_attr_count
, KM_SLEEP
);
910 * Attribute table is constructed from requested attribute list,
911 * previously foreign registered attributes, and also the legacy
912 * ZPL set of attributes.
915 if (sa
->sa_reg_attr_obj
) {
916 for (zap_cursor_init(&zc
, os
, sa
->sa_reg_attr_obj
);
917 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
918 zap_cursor_advance(&zc
)) {
920 value
= za
.za_first_integer
;
923 tb
[ATTR_NUM(value
)].sa_attr
= ATTR_NUM(value
);
924 tb
[ATTR_NUM(value
)].sa_length
= ATTR_LENGTH(value
);
925 tb
[ATTR_NUM(value
)].sa_byteswap
= ATTR_BSWAP(value
);
926 tb
[ATTR_NUM(value
)].sa_registered
= B_TRUE
;
928 if (tb
[ATTR_NUM(value
)].sa_name
) {
931 tb
[ATTR_NUM(value
)].sa_name
=
932 kmem_zalloc(strlen(za
.za_name
) +1, KM_SLEEP
);
933 (void) strlcpy(tb
[ATTR_NUM(value
)].sa_name
, za
.za_name
,
934 strlen(za
.za_name
) +1);
936 zap_cursor_fini(&zc
);
938 * Make sure we processed the correct number of registered
941 if (registered_count
!= sa_reg_count
) {
948 if (ostype
== DMU_OST_ZFS
) {
949 for (i
= 0; i
!= sa_legacy_attr_count
; i
++) {
952 tb
[i
].sa_attr
= sa_legacy_attrs
[i
].sa_attr
;
953 tb
[i
].sa_length
= sa_legacy_attrs
[i
].sa_length
;
954 tb
[i
].sa_byteswap
= sa_legacy_attrs
[i
].sa_byteswap
;
955 tb
[i
].sa_registered
= B_FALSE
;
957 kmem_zalloc(strlen(sa_legacy_attrs
[i
].sa_name
) +1,
959 (void) strlcpy(tb
[i
].sa_name
,
960 sa_legacy_attrs
[i
].sa_name
,
961 strlen(sa_legacy_attrs
[i
].sa_name
) + 1);
965 for (i
= 0; i
!= count
; i
++) {
966 sa_attr_type_t attr_id
;
968 attr_id
= sa
->sa_user_table
[i
];
969 if (tb
[attr_id
].sa_name
)
972 tb
[attr_id
].sa_length
= reg_attrs
[i
].sa_length
;
973 tb
[attr_id
].sa_byteswap
= reg_attrs
[i
].sa_byteswap
;
974 tb
[attr_id
].sa_attr
= attr_id
;
975 tb
[attr_id
].sa_name
=
976 kmem_zalloc(strlen(reg_attrs
[i
].sa_name
) + 1, KM_SLEEP
);
977 (void) strlcpy(tb
[attr_id
].sa_name
, reg_attrs
[i
].sa_name
,
978 strlen(reg_attrs
[i
].sa_name
) + 1);
981 sa
->sa_need_attr_registration
=
982 (sa_attr_count
!= registered_count
);
986 kmem_free(sa
->sa_user_table
, count
* sizeof (sa_attr_type_t
));
987 sa
->sa_user_table
= NULL
;
988 sa_free_attr_table(sa
);
989 return ((error
!= 0) ? error
: EINVAL
);
993 sa_setup(objset_t
*os
, uint64_t sa_obj
, sa_attr_reg_t
*reg_attrs
, int count
,
994 sa_attr_type_t
**user_table
)
999 dmu_objset_type_t ostype
= dmu_objset_type(os
);
1003 mutex_enter(&os
->os_user_ptr_lock
);
1005 mutex_enter(&os
->os_sa
->sa_lock
);
1006 mutex_exit(&os
->os_user_ptr_lock
);
1007 tb
= os
->os_sa
->sa_user_table
;
1008 mutex_exit(&os
->os_sa
->sa_lock
);
1013 sa
= kmem_zalloc(sizeof (sa_os_t
), KM_SLEEP
);
1014 mutex_init(&sa
->sa_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1015 sa
->sa_master_obj
= sa_obj
;
1018 mutex_enter(&sa
->sa_lock
);
1019 mutex_exit(&os
->os_user_ptr_lock
);
1020 avl_create(&sa
->sa_layout_num_tree
, layout_num_compare
,
1021 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_num_node
));
1022 avl_create(&sa
->sa_layout_hash_tree
, layout_hash_compare
,
1023 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_hash_node
));
1026 error
= zap_lookup(os
, sa_obj
, SA_LAYOUTS
,
1027 8, 1, &sa
->sa_layout_attr_obj
);
1028 if (error
!= 0 && error
!= ENOENT
)
1030 error
= zap_lookup(os
, sa_obj
, SA_REGISTRY
,
1031 8, 1, &sa
->sa_reg_attr_obj
);
1032 if (error
!= 0 && error
!= ENOENT
)
1036 if ((error
= sa_attr_table_setup(os
, reg_attrs
, count
)) != 0)
1039 if (sa
->sa_layout_attr_obj
!= 0) {
1040 uint64_t layout_count
;
1042 error
= zap_count(os
, sa
->sa_layout_attr_obj
,
1046 * Layout number count should be > 0
1048 if (error
|| (error
== 0 && layout_count
== 0)) {
1050 error
= SET_ERROR(EINVAL
);
1054 for (zap_cursor_init(&zc
, os
, sa
->sa_layout_attr_obj
);
1055 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
1056 zap_cursor_advance(&zc
)) {
1057 sa_attr_type_t
*lot_attrs
;
1060 lot_attrs
= kmem_zalloc(sizeof (sa_attr_type_t
) *
1061 za
.za_num_integers
, KM_SLEEP
);
1063 if ((error
= (zap_lookup(os
, sa
->sa_layout_attr_obj
,
1064 za
.za_name
, 2, za
.za_num_integers
,
1065 lot_attrs
))) != 0) {
1066 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1067 za
.za_num_integers
);
1070 VERIFY(ddi_strtoull(za
.za_name
, NULL
, 10,
1071 (unsigned long long *)&lot_num
) == 0);
1073 (void) sa_add_layout_entry(os
, lot_attrs
,
1074 za
.za_num_integers
, lot_num
,
1075 sa_layout_info_hash(lot_attrs
,
1076 za
.za_num_integers
), B_FALSE
, NULL
);
1077 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1078 za
.za_num_integers
);
1080 zap_cursor_fini(&zc
);
1083 * Make sure layout count matches number of entries added
1086 if (avl_numnodes(&sa
->sa_layout_num_tree
) != layout_count
) {
1092 /* Add special layout number for old ZNODES */
1093 if (ostype
== DMU_OST_ZFS
) {
1094 (void) sa_add_layout_entry(os
, sa_legacy_zpl_layout
,
1095 sa_legacy_attr_count
, 0,
1096 sa_layout_info_hash(sa_legacy_zpl_layout
,
1097 sa_legacy_attr_count
), B_FALSE
, NULL
);
1099 (void) sa_add_layout_entry(os
, sa_dummy_zpl_layout
, 0, 1,
1102 *user_table
= os
->os_sa
->sa_user_table
;
1103 mutex_exit(&sa
->sa_lock
);
1107 sa_free_attr_table(sa
);
1108 if (sa
->sa_user_table
)
1109 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1110 mutex_exit(&sa
->sa_lock
);
1111 avl_destroy(&sa
->sa_layout_hash_tree
);
1112 avl_destroy(&sa
->sa_layout_num_tree
);
1113 mutex_destroy(&sa
->sa_lock
);
1114 kmem_free(sa
, sizeof (sa_os_t
));
1115 return ((error
== ECKSUM
) ? EIO
: error
);
1119 sa_tear_down(objset_t
*os
)
1121 sa_os_t
*sa
= os
->os_sa
;
1125 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1127 /* Free up attr table */
1129 sa_free_attr_table(sa
);
1132 while (layout
= avl_destroy_nodes(&sa
->sa_layout_hash_tree
, &cookie
)) {
1134 while (tab
= list_head(&layout
->lot_idx_tab
)) {
1135 ASSERT(refcount_count(&tab
->sa_refcount
));
1136 sa_idx_tab_rele(os
, tab
);
1141 while (layout
= avl_destroy_nodes(&sa
->sa_layout_num_tree
, &cookie
)) {
1142 kmem_free(layout
->lot_attrs
,
1143 sizeof (sa_attr_type_t
) * layout
->lot_attr_count
);
1144 kmem_free(layout
, sizeof (sa_lot_t
));
1147 avl_destroy(&sa
->sa_layout_hash_tree
);
1148 avl_destroy(&sa
->sa_layout_num_tree
);
1149 mutex_destroy(&sa
->sa_lock
);
1151 kmem_free(sa
, sizeof (sa_os_t
));
1156 sa_build_idx_tab(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1157 uint16_t length
, int length_idx
, boolean_t var_length
, void *userp
)
1159 sa_idx_tab_t
*idx_tab
= userp
;
1162 ASSERT(idx_tab
->sa_variable_lengths
);
1163 idx_tab
->sa_variable_lengths
[length_idx
] = length
;
1165 TOC_ATTR_ENCODE(idx_tab
->sa_idx_tab
[attr
], length_idx
,
1166 (uint32_t)((uintptr_t)attr_addr
- (uintptr_t)hdr
));
1170 sa_attr_iter(objset_t
*os
, sa_hdr_phys_t
*hdr
, dmu_object_type_t type
,
1171 sa_iterfunc_t func
, sa_lot_t
*tab
, void *userp
)
1177 sa_os_t
*sa
= os
->os_sa
;
1179 uint16_t *length_start
= NULL
;
1180 uint8_t length_idx
= 0;
1183 search
.lot_num
= SA_LAYOUT_NUM(hdr
, type
);
1184 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1188 if (IS_SA_BONUSTYPE(type
)) {
1189 data_start
= (void *)P2ROUNDUP(((uintptr_t)hdr
+
1190 offsetof(sa_hdr_phys_t
, sa_lengths
) +
1191 (sizeof (uint16_t) * tb
->lot_var_sizes
)), 8);
1192 length_start
= hdr
->sa_lengths
;
1197 for (i
= 0; i
!= tb
->lot_attr_count
; i
++) {
1198 int attr_length
, reg_length
;
1201 reg_length
= sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
;
1203 attr_length
= reg_length
;
1206 attr_length
= length_start
[length_idx
];
1207 idx_len
= length_idx
++;
1210 func(hdr
, data_start
, tb
->lot_attrs
[i
], attr_length
,
1211 idx_len
, reg_length
== 0 ? B_TRUE
: B_FALSE
, userp
);
1213 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
1220 sa_byteswap_cb(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1221 uint16_t length
, int length_idx
, boolean_t variable_length
, void *userp
)
1223 sa_handle_t
*hdl
= userp
;
1224 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1226 sa_bswap_table
[sa
->sa_attr_table
[attr
].sa_byteswap
](attr_addr
, length
);
1230 sa_byteswap(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1232 sa_hdr_phys_t
*sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1234 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1235 int num_lengths
= 1;
1238 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1239 if (sa_hdr_phys
->sa_magic
== SA_MAGIC
)
1242 db
= SA_GET_DB(hdl
, buftype
);
1244 if (buftype
== SA_SPILL
) {
1245 arc_release(db
->db_buf
, NULL
);
1246 arc_buf_thaw(db
->db_buf
);
1249 sa_hdr_phys
->sa_magic
= BSWAP_32(sa_hdr_phys
->sa_magic
);
1250 sa_hdr_phys
->sa_layout_info
= BSWAP_16(sa_hdr_phys
->sa_layout_info
);
1253 * Determine number of variable lenghts in header
1254 * The standard 8 byte header has one for free and a
1255 * 16 byte header would have 4 + 1;
1257 if (SA_HDR_SIZE(sa_hdr_phys
) > 8)
1258 num_lengths
+= (SA_HDR_SIZE(sa_hdr_phys
) - 8) >> 1;
1259 for (i
= 0; i
!= num_lengths
; i
++)
1260 sa_hdr_phys
->sa_lengths
[i
] =
1261 BSWAP_16(sa_hdr_phys
->sa_lengths
[i
]);
1263 sa_attr_iter(hdl
->sa_os
, sa_hdr_phys
, DMU_OT_SA
,
1264 sa_byteswap_cb
, NULL
, hdl
);
1266 if (buftype
== SA_SPILL
)
1267 arc_buf_freeze(((dmu_buf_impl_t
*)hdl
->sa_spill
)->db_buf
);
1271 sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1273 sa_hdr_phys_t
*sa_hdr_phys
;
1274 dmu_buf_impl_t
*db
= SA_GET_DB(hdl
, buftype
);
1275 dmu_object_type_t bonustype
= SA_BONUSTYPE_FROM_DB(db
);
1276 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1277 sa_idx_tab_t
*idx_tab
;
1279 sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1281 mutex_enter(&sa
->sa_lock
);
1283 /* Do we need to byteswap? */
1285 /* only check if not old znode */
1286 if (IS_SA_BONUSTYPE(bonustype
) && sa_hdr_phys
->sa_magic
!= SA_MAGIC
&&
1287 sa_hdr_phys
->sa_magic
!= 0) {
1288 VERIFY(BSWAP_32(sa_hdr_phys
->sa_magic
) == SA_MAGIC
);
1289 sa_byteswap(hdl
, buftype
);
1292 idx_tab
= sa_find_idx_tab(hdl
->sa_os
, bonustype
, sa_hdr_phys
);
1294 if (buftype
== SA_BONUS
)
1295 hdl
->sa_bonus_tab
= idx_tab
;
1297 hdl
->sa_spill_tab
= idx_tab
;
1299 mutex_exit(&sa
->sa_lock
);
1305 sa_evict(dmu_buf_t
*db
, void *sap
)
1307 panic("evicting sa dbuf %p\n", (void *)db
);
1311 sa_idx_tab_rele(objset_t
*os
, void *arg
)
1313 sa_os_t
*sa
= os
->os_sa
;
1314 sa_idx_tab_t
*idx_tab
= arg
;
1316 if (idx_tab
== NULL
)
1319 mutex_enter(&sa
->sa_lock
);
1320 if (refcount_remove(&idx_tab
->sa_refcount
, NULL
) == 0) {
1321 list_remove(&idx_tab
->sa_layout
->lot_idx_tab
, idx_tab
);
1322 if (idx_tab
->sa_variable_lengths
)
1323 kmem_free(idx_tab
->sa_variable_lengths
,
1325 idx_tab
->sa_layout
->lot_var_sizes
);
1326 refcount_destroy(&idx_tab
->sa_refcount
);
1327 kmem_free(idx_tab
->sa_idx_tab
,
1328 sizeof (uint32_t) * sa
->sa_num_attrs
);
1329 kmem_free(idx_tab
, sizeof (sa_idx_tab_t
));
1331 mutex_exit(&sa
->sa_lock
);
1335 sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
)
1337 sa_os_t
*sa
= os
->os_sa
;
1339 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1340 (void) refcount_add(&idx_tab
->sa_refcount
, NULL
);
1344 sa_handle_destroy(sa_handle_t
*hdl
)
1346 mutex_enter(&hdl
->sa_lock
);
1347 (void) dmu_buf_update_user((dmu_buf_t
*)hdl
->sa_bonus
, hdl
,
1350 if (hdl
->sa_bonus_tab
) {
1351 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
1352 hdl
->sa_bonus_tab
= NULL
;
1354 if (hdl
->sa_spill_tab
) {
1355 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1356 hdl
->sa_spill_tab
= NULL
;
1359 dmu_buf_rele(hdl
->sa_bonus
, NULL
);
1362 dmu_buf_rele((dmu_buf_t
*)hdl
->sa_spill
, NULL
);
1363 mutex_exit(&hdl
->sa_lock
);
1365 kmem_cache_free(sa_cache
, hdl
);
1369 sa_handle_get_from_db(objset_t
*os
, dmu_buf_t
*db
, void *userp
,
1370 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1373 dmu_object_info_t doi
;
1374 sa_handle_t
*handle
;
1377 dmu_object_info_from_db(db
, &doi
);
1378 ASSERT(doi
.doi_bonus_type
== DMU_OT_SA
||
1379 doi
.doi_bonus_type
== DMU_OT_ZNODE
);
1381 /* find handle, if it exists */
1382 /* if one doesn't exist then create a new one, and initialize it */
1384 handle
= (hdl_type
== SA_HDL_SHARED
) ? dmu_buf_get_user(db
) : NULL
;
1385 if (handle
== NULL
) {
1386 sa_handle_t
*newhandle
;
1387 handle
= kmem_cache_alloc(sa_cache
, KM_SLEEP
);
1388 handle
->sa_userp
= userp
;
1389 handle
->sa_bonus
= db
;
1391 handle
->sa_spill
= NULL
;
1393 error
= sa_build_index(handle
, SA_BONUS
);
1394 newhandle
= (hdl_type
== SA_HDL_SHARED
) ?
1395 dmu_buf_set_user_ie(db
, handle
, sa_evict
) : NULL
;
1397 if (newhandle
!= NULL
) {
1398 kmem_cache_free(sa_cache
, handle
);
1408 sa_handle_get(objset_t
*objset
, uint64_t objid
, void *userp
,
1409 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1414 if (error
= dmu_bonus_hold(objset
, objid
, NULL
, &db
))
1417 return (sa_handle_get_from_db(objset
, db
, userp
, hdl_type
,
1422 sa_buf_hold(objset_t
*objset
, uint64_t obj_num
, void *tag
, dmu_buf_t
**db
)
1424 return (dmu_bonus_hold(objset
, obj_num
, tag
, db
));
1428 sa_buf_rele(dmu_buf_t
*db
, void *tag
)
1430 dmu_buf_rele(db
, tag
);
1434 sa_lookup_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
)
1437 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1438 return (sa_attr_op(hdl
, bulk
, count
, SA_LOOKUP
, NULL
));
1442 sa_lookup(sa_handle_t
*hdl
, sa_attr_type_t attr
, void *buf
, uint32_t buflen
)
1445 sa_bulk_attr_t bulk
;
1447 bulk
.sa_attr
= attr
;
1449 bulk
.sa_length
= buflen
;
1450 bulk
.sa_data_func
= NULL
;
1453 mutex_enter(&hdl
->sa_lock
);
1454 error
= sa_lookup_impl(hdl
, &bulk
, 1);
1455 mutex_exit(&hdl
->sa_lock
);
1461 sa_lookup_uio(sa_handle_t
*hdl
, sa_attr_type_t attr
, uio_t
*uio
)
1464 sa_bulk_attr_t bulk
;
1466 bulk
.sa_data
= NULL
;
1467 bulk
.sa_attr
= attr
;
1468 bulk
.sa_data_func
= NULL
;
1472 mutex_enter(&hdl
->sa_lock
);
1473 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) == 0) {
1474 error
= uiomove((void *)bulk
.sa_addr
, MIN(bulk
.sa_size
,
1475 uio
->uio_resid
), UIO_READ
, uio
);
1477 mutex_exit(&hdl
->sa_lock
);
1484 sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
, void *data
)
1486 sa_idx_tab_t
*idx_tab
;
1487 sa_hdr_phys_t
*hdr
= (sa_hdr_phys_t
*)data
;
1488 sa_os_t
*sa
= os
->os_sa
;
1489 sa_lot_t
*tb
, search
;
1493 * Deterimine layout number. If SA node and header == 0 then
1494 * force the index table to the dummy "1" empty layout.
1496 * The layout number would only be zero for a newly created file
1497 * that has not added any attributes yet, or with crypto enabled which
1498 * doesn't write any attributes to the bonus buffer.
1501 search
.lot_num
= SA_LAYOUT_NUM(hdr
, bonustype
);
1503 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1505 /* Verify header size is consistent with layout information */
1507 ASSERT(IS_SA_BONUSTYPE(bonustype
) &&
1508 SA_HDR_SIZE_MATCH_LAYOUT(hdr
, tb
) || !IS_SA_BONUSTYPE(bonustype
) ||
1509 (IS_SA_BONUSTYPE(bonustype
) && hdr
->sa_layout_info
== 0));
1512 * See if any of the already existing TOC entries can be reused?
1515 for (idx_tab
= list_head(&tb
->lot_idx_tab
); idx_tab
;
1516 idx_tab
= list_next(&tb
->lot_idx_tab
, idx_tab
)) {
1517 boolean_t valid_idx
= B_TRUE
;
1520 if (tb
->lot_var_sizes
!= 0 &&
1521 idx_tab
->sa_variable_lengths
!= NULL
) {
1522 for (i
= 0; i
!= tb
->lot_var_sizes
; i
++) {
1523 if (hdr
->sa_lengths
[i
] !=
1524 idx_tab
->sa_variable_lengths
[i
]) {
1525 valid_idx
= B_FALSE
;
1531 sa_idx_tab_hold(os
, idx_tab
);
1536 /* No such luck, create a new entry */
1537 idx_tab
= kmem_zalloc(sizeof (sa_idx_tab_t
), KM_SLEEP
);
1538 idx_tab
->sa_idx_tab
=
1539 kmem_zalloc(sizeof (uint32_t) * sa
->sa_num_attrs
, KM_SLEEP
);
1540 idx_tab
->sa_layout
= tb
;
1541 refcount_create(&idx_tab
->sa_refcount
);
1542 if (tb
->lot_var_sizes
)
1543 idx_tab
->sa_variable_lengths
= kmem_alloc(sizeof (uint16_t) *
1544 tb
->lot_var_sizes
, KM_SLEEP
);
1546 sa_attr_iter(os
, hdr
, bonustype
, sa_build_idx_tab
,
1548 sa_idx_tab_hold(os
, idx_tab
); /* one hold for consumer */
1549 sa_idx_tab_hold(os
, idx_tab
); /* one for layout */
1550 list_insert_tail(&tb
->lot_idx_tab
, idx_tab
);
1555 sa_default_locator(void **dataptr
, uint32_t *len
, uint32_t total_len
,
1556 boolean_t start
, void *userdata
)
1560 *dataptr
= userdata
;
1565 sa_attr_register_sync(sa_handle_t
*hdl
, dmu_tx_t
*tx
)
1567 uint64_t attr_value
= 0;
1568 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1569 sa_attr_table_t
*tb
= sa
->sa_attr_table
;
1572 mutex_enter(&sa
->sa_lock
);
1574 if (!sa
->sa_need_attr_registration
|| sa
->sa_master_obj
== NULL
) {
1575 mutex_exit(&sa
->sa_lock
);
1579 if (sa
->sa_reg_attr_obj
== NULL
) {
1580 sa
->sa_reg_attr_obj
= zap_create_link(hdl
->sa_os
,
1581 DMU_OT_SA_ATTR_REGISTRATION
,
1582 sa
->sa_master_obj
, SA_REGISTRY
, tx
);
1584 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
1585 if (sa
->sa_attr_table
[i
].sa_registered
)
1587 ATTR_ENCODE(attr_value
, tb
[i
].sa_attr
, tb
[i
].sa_length
,
1589 VERIFY(0 == zap_update(hdl
->sa_os
, sa
->sa_reg_attr_obj
,
1590 tb
[i
].sa_name
, 8, 1, &attr_value
, tx
));
1591 tb
[i
].sa_registered
= B_TRUE
;
1593 sa
->sa_need_attr_registration
= B_FALSE
;
1594 mutex_exit(&sa
->sa_lock
);
1598 * Replace all attributes with attributes specified in template.
1599 * If dnode had a spill buffer then those attributes will be
1600 * also be replaced, possibly with just an empty spill block
1602 * This interface is intended to only be used for bulk adding of
1603 * attributes for a new file. It will also be used by the ZPL
1604 * when converting and old formatted znode to native SA support.
1607 sa_replace_all_by_template_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1608 int attr_count
, dmu_tx_t
*tx
)
1610 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1612 if (sa
->sa_need_attr_registration
)
1613 sa_attr_register_sync(hdl
, tx
);
1614 return (sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
));
1618 sa_replace_all_by_template(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1619 int attr_count
, dmu_tx_t
*tx
)
1623 mutex_enter(&hdl
->sa_lock
);
1624 error
= sa_replace_all_by_template_locked(hdl
, attr_desc
,
1626 mutex_exit(&hdl
->sa_lock
);
1631 * add/remove/replace a single attribute and then rewrite the entire set
1635 sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
1636 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
1637 uint16_t buflen
, dmu_tx_t
*tx
)
1639 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1640 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)hdl
->sa_bonus
;
1642 sa_bulk_attr_t
*attr_desc
;
1644 int bonus_attr_count
= 0;
1645 int bonus_data_size
= 0;
1646 int spill_data_size
= 0;
1647 int spill_attr_count
= 0;
1650 int i
, j
, k
, length_idx
;
1652 sa_idx_tab_t
*idx_tab
;
1656 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1658 /* First make of copy of the old data */
1662 if (dn
->dn_bonuslen
!= 0) {
1663 bonus_data_size
= hdl
->sa_bonus
->db_size
;
1664 old_data
[0] = kmem_alloc(bonus_data_size
, KM_SLEEP
);
1665 bcopy(hdl
->sa_bonus
->db_data
, old_data
[0],
1666 hdl
->sa_bonus
->db_size
);
1667 bonus_attr_count
= hdl
->sa_bonus_tab
->sa_layout
->lot_attr_count
;
1673 /* Bring spill buffer online if it isn't currently */
1675 if ((error
= sa_get_spill(hdl
)) == 0) {
1676 spill_data_size
= hdl
->sa_spill
->db_size
;
1677 old_data
[1] = kmem_alloc(spill_data_size
, KM_SLEEP
);
1678 bcopy(hdl
->sa_spill
->db_data
, old_data
[1],
1679 hdl
->sa_spill
->db_size
);
1681 hdl
->sa_spill_tab
->sa_layout
->lot_attr_count
;
1682 } else if (error
&& error
!= ENOENT
) {
1684 kmem_free(old_data
[0], bonus_data_size
);
1690 /* build descriptor of all attributes */
1692 attr_count
= bonus_attr_count
+ spill_attr_count
;
1693 if (action
== SA_ADD
)
1695 else if (action
== SA_REMOVE
)
1698 attr_desc
= kmem_zalloc(sizeof (sa_bulk_attr_t
) * attr_count
, KM_SLEEP
);
1701 * loop through bonus and spill buffer if it exists, and
1702 * build up new attr_descriptor to reset the attributes
1705 count
= bonus_attr_count
;
1706 hdr
= SA_GET_HDR(hdl
, SA_BONUS
);
1707 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_BONUS
);
1708 for (; k
!= 2; k
++) {
1709 /* iterate over each attribute in layout */
1710 for (i
= 0, length_idx
= 0; i
!= count
; i
++) {
1711 sa_attr_type_t attr
;
1713 attr
= idx_tab
->sa_layout
->lot_attrs
[i
];
1714 if (attr
== newattr
) {
1715 if (action
== SA_REMOVE
) {
1719 ASSERT(SA_REGISTERED_LEN(sa
, attr
) == 0);
1720 ASSERT(action
== SA_REPLACE
);
1721 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1722 locator
, datastart
, buflen
);
1724 length
= SA_REGISTERED_LEN(sa
, attr
);
1726 length
= hdr
->sa_lengths
[length_idx
++];
1729 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1731 (TOC_OFF(idx_tab
->sa_idx_tab
[attr
]) +
1732 (uintptr_t)old_data
[k
]), length
);
1735 if (k
== 0 && hdl
->sa_spill
) {
1736 hdr
= SA_GET_HDR(hdl
, SA_SPILL
);
1737 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_SPILL
);
1738 count
= spill_attr_count
;
1743 if (action
== SA_ADD
) {
1744 length
= SA_REGISTERED_LEN(sa
, newattr
);
1748 SA_ADD_BULK_ATTR(attr_desc
, j
, newattr
, locator
,
1752 error
= sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
);
1755 kmem_free(old_data
[0], bonus_data_size
);
1757 kmem_free(old_data
[1], spill_data_size
);
1758 kmem_free(attr_desc
, sizeof (sa_bulk_attr_t
) * attr_count
);
1764 sa_bulk_update_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
1768 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1769 dmu_object_type_t bonustype
;
1771 bonustype
= SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl
, SA_BONUS
));
1774 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1776 /* sync out registration table if necessary */
1777 if (sa
->sa_need_attr_registration
)
1778 sa_attr_register_sync(hdl
, tx
);
1780 error
= sa_attr_op(hdl
, bulk
, count
, SA_UPDATE
, tx
);
1781 if (error
== 0 && !IS_SA_BONUSTYPE(bonustype
) && sa
->sa_update_cb
)
1782 sa
->sa_update_cb(hdl
, tx
);
1788 * update or add new attribute
1791 sa_update(sa_handle_t
*hdl
, sa_attr_type_t type
,
1792 void *buf
, uint32_t buflen
, dmu_tx_t
*tx
)
1795 sa_bulk_attr_t bulk
;
1797 bulk
.sa_attr
= type
;
1798 bulk
.sa_data_func
= NULL
;
1799 bulk
.sa_length
= buflen
;
1802 mutex_enter(&hdl
->sa_lock
);
1803 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1804 mutex_exit(&hdl
->sa_lock
);
1809 sa_update_from_cb(sa_handle_t
*hdl
, sa_attr_type_t attr
,
1810 uint32_t buflen
, sa_data_locator_t
*locator
, void *userdata
, dmu_tx_t
*tx
)
1813 sa_bulk_attr_t bulk
;
1815 bulk
.sa_attr
= attr
;
1816 bulk
.sa_data
= userdata
;
1817 bulk
.sa_data_func
= locator
;
1818 bulk
.sa_length
= buflen
;
1820 mutex_enter(&hdl
->sa_lock
);
1821 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1822 mutex_exit(&hdl
->sa_lock
);
1827 * Return size of an attribute
1831 sa_size(sa_handle_t
*hdl
, sa_attr_type_t attr
, int *size
)
1833 sa_bulk_attr_t bulk
;
1836 bulk
.sa_data
= NULL
;
1837 bulk
.sa_attr
= attr
;
1838 bulk
.sa_data_func
= NULL
;
1841 mutex_enter(&hdl
->sa_lock
);
1842 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) != 0) {
1843 mutex_exit(&hdl
->sa_lock
);
1846 *size
= bulk
.sa_size
;
1848 mutex_exit(&hdl
->sa_lock
);
1853 sa_bulk_lookup_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1856 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1857 return (sa_lookup_impl(hdl
, attrs
, count
));
1861 sa_bulk_lookup(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1866 mutex_enter(&hdl
->sa_lock
);
1867 error
= sa_bulk_lookup_locked(hdl
, attrs
, count
);
1868 mutex_exit(&hdl
->sa_lock
);
1873 sa_bulk_update(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
, dmu_tx_t
*tx
)
1878 mutex_enter(&hdl
->sa_lock
);
1879 error
= sa_bulk_update_impl(hdl
, attrs
, count
, tx
);
1880 mutex_exit(&hdl
->sa_lock
);
1885 sa_remove(sa_handle_t
*hdl
, sa_attr_type_t attr
, dmu_tx_t
*tx
)
1889 mutex_enter(&hdl
->sa_lock
);
1890 error
= sa_modify_attrs(hdl
, attr
, SA_REMOVE
, NULL
,
1892 mutex_exit(&hdl
->sa_lock
);
1897 sa_object_info(sa_handle_t
*hdl
, dmu_object_info_t
*doi
)
1899 dmu_object_info_from_db((dmu_buf_t
*)hdl
->sa_bonus
, doi
);
1903 sa_object_size(sa_handle_t
*hdl
, uint32_t *blksize
, u_longlong_t
*nblocks
)
1905 dmu_object_size_from_db((dmu_buf_t
*)hdl
->sa_bonus
,
1910 sa_update_user(sa_handle_t
*newhdl
, sa_handle_t
*oldhdl
)
1912 (void) dmu_buf_update_user((dmu_buf_t
*)newhdl
->sa_bonus
,
1913 oldhdl
, newhdl
, sa_evict
);
1914 oldhdl
->sa_bonus
= NULL
;
1918 sa_set_userp(sa_handle_t
*hdl
, void *ptr
)
1920 hdl
->sa_userp
= ptr
;
1924 sa_get_db(sa_handle_t
*hdl
)
1926 return ((dmu_buf_t
*)hdl
->sa_bonus
);
1930 sa_get_userdata(sa_handle_t
*hdl
)
1932 return (hdl
->sa_userp
);
1936 sa_register_update_callback_locked(objset_t
*os
, sa_update_cb_t
*func
)
1938 ASSERT(MUTEX_HELD(&os
->os_sa
->sa_lock
));
1939 os
->os_sa
->sa_update_cb
= func
;
1943 sa_register_update_callback(objset_t
*os
, sa_update_cb_t
*func
)
1946 mutex_enter(&os
->os_sa
->sa_lock
);
1947 sa_register_update_callback_locked(os
, func
);
1948 mutex_exit(&os
->os_sa
->sa_lock
);
1952 sa_handle_object(sa_handle_t
*hdl
)
1954 return (hdl
->sa_bonus
->db_object
);
1958 sa_enabled(objset_t
*os
)
1960 return (os
->os_sa
== NULL
);
1964 sa_set_sa_object(objset_t
*os
, uint64_t sa_object
)
1966 sa_os_t
*sa
= os
->os_sa
;
1968 if (sa
->sa_master_obj
)
1971 sa
->sa_master_obj
= sa_object
;
1977 sa_hdrsize(void *arg
)
1979 sa_hdr_phys_t
*hdr
= arg
;
1981 return (SA_HDR_SIZE(hdr
));
1985 sa_handle_lock(sa_handle_t
*hdl
)
1988 mutex_enter(&hdl
->sa_lock
);
1992 sa_handle_unlock(sa_handle_t
*hdl
)
1995 mutex_exit(&hdl
->sa_lock
);