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 * Copyright (c) 2013 by Delphix. All rights reserved.
27 #include <sys/zfs_context.h>
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_objset.h>
36 #include <sys/dnode.h>
39 #include <sys/sunddi.h>
40 #include <sys/sa_impl.h>
41 #include <sys/dnode.h>
42 #include <sys/errno.h>
43 #include <sys/zfs_context.h>
46 * ZFS System attributes:
48 * A generic mechanism to allow for arbitrary attributes
49 * to be stored in a dnode. The data will be stored in the bonus buffer of
50 * the dnode and if necessary a special "spill" block will be used to handle
51 * overflow situations. The spill block will be sized to fit the data
52 * from 512 - 128K. When a spill block is used the BP (blkptr_t) for the
53 * spill block is stored at the end of the current bonus buffer. Any
54 * attributes that would be in the way of the blkptr_t will be relocated
55 * into the spill block.
57 * Attribute registration:
59 * Stored persistently on a per dataset basis
60 * a mapping between attribute "string" names and their actual attribute
61 * numeric values, length, and byteswap function. The names are only used
62 * during registration. All attributes are known by their unique attribute
63 * id value. If an attribute can have a variable size then the value
64 * 0 will be used to indicate this.
68 * Attribute layouts are a way to compactly store multiple attributes, but
69 * without taking the overhead associated with managing each attribute
70 * individually. Since you will typically have the same set of attributes
71 * stored in the same order a single table will be used to represent that
72 * layout. The ZPL for example will usually have only about 10 different
73 * layouts (regular files, device files, symlinks,
74 * regular files + scanstamp, files/dir with extended attributes, and then
75 * you have the possibility of all of those minus ACL, because it would
76 * be kicked out into the spill block)
78 * Layouts are simply an array of the attributes and their
79 * ordering i.e. [0, 1, 4, 5, 2]
81 * Each distinct layout is given a unique layout number and that is whats
82 * stored in the header at the beginning of the SA data buffer.
84 * A layout only covers a single dbuf (bonus or spill). If a set of
85 * attributes is split up between the bonus buffer and a spill buffer then
86 * two different layouts will be used. This allows us to byteswap the
87 * spill without looking at the bonus buffer and keeps the on disk format of
88 * the bonus and spill buffer the same.
90 * Adding a single attribute will cause the entire set of attributes to
91 * be rewritten and could result in a new layout number being constructed
92 * as part of the rewrite if no such layout exists for the new set of
93 * attribues. The new attribute will be appended to the end of the already
94 * existing attributes.
96 * Both the attribute registration and attribute layout information are
97 * stored in normal ZAP attributes. Their should be a small number of
98 * known layouts and the set of attributes is assumed to typically be quite
101 * The registered attributes and layout "table" information is maintained
102 * in core and a special "sa_os_t" is attached to the objset_t.
104 * A special interface is provided to allow for quickly applying
105 * a large set of attributes at once. sa_replace_all_by_template() is
106 * used to set an array of attributes. This is used by the ZPL when
107 * creating a brand new file. The template that is passed into the function
108 * specifies the attribute, size for variable length attributes, location of
109 * data and special "data locator" function if the data isn't in a contiguous
112 * Byteswap implications:
114 * Since the SA attributes are not entirely self describing we can't do
115 * the normal byteswap processing. The special ZAP layout attribute and
116 * attribute registration attributes define the byteswap function and the
117 * size of the attributes, unless it is variable sized.
118 * The normal ZFS byteswapping infrastructure assumes you don't need
119 * to read any objects in order to do the necessary byteswapping. Whereas
120 * SA attributes can only be properly byteswapped if the dataset is opened
121 * and the layout/attribute ZAP attributes are available. Because of this
122 * the SA attributes will be byteswapped when they are first accessed by
123 * the SA code that will read the SA data.
126 typedef void (sa_iterfunc_t
)(void *hdr
, void *addr
, sa_attr_type_t
,
127 uint16_t length
, int length_idx
, boolean_t
, void *userp
);
129 static int sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
);
130 static void sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
);
131 static void *sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
,
133 static void sa_idx_tab_rele(objset_t
*os
, void *arg
);
134 static void sa_copy_data(sa_data_locator_t
*func
, void *start
, void *target
,
136 static int sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
137 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
138 uint16_t buflen
, dmu_tx_t
*tx
);
140 arc_byteswap_func_t sa_bswap_table
[] = {
141 byteswap_uint64_array
,
142 byteswap_uint32_array
,
143 byteswap_uint16_array
,
144 byteswap_uint8_array
,
148 #define SA_COPY_DATA(f, s, t, l) \
152 *(uint64_t *)t = *(uint64_t *)s; \
153 } else if (l == 16) { \
154 *(uint64_t *)t = *(uint64_t *)s; \
155 *(uint64_t *)((uintptr_t)t + 8) = \
156 *(uint64_t *)((uintptr_t)s + 8); \
161 sa_copy_data(f, s, t, l); \
165 * This table is fixed and cannot be changed. Its purpose is to
166 * allow the SA code to work with both old/new ZPL file systems.
167 * It contains the list of legacy attributes. These attributes aren't
168 * stored in the "attribute" registry zap objects, since older ZPL file systems
169 * won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
170 * use this static table.
172 sa_attr_reg_t sa_legacy_attrs
[] = {
173 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 0},
174 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 1},
175 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 2},
176 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 3},
177 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY
, 4},
178 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY
, 5},
179 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY
, 6},
180 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY
, 7},
181 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY
, 8},
182 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY
, 9},
183 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY
, 10},
184 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY
, 11},
185 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY
, 12},
186 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY
, 13},
187 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY
, 14},
188 {"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY
, 15},
192 * This is only used for objects of type DMU_OT_ZNODE
194 sa_attr_type_t sa_legacy_zpl_layout
[] = {
195 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
199 * Special dummy layout used for buffers with no attributes.
201 sa_attr_type_t sa_dummy_zpl_layout
[] = { 0 };
203 static int sa_legacy_attr_count
= 16;
204 static kmem_cache_t
*sa_cache
= NULL
;
205 static kmem_cache_t
*spill_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);
237 spill_cache
= kmem_cache_create("spill_cache",
238 SPA_MAXBLOCKSIZE
, 0, NULL
, NULL
, NULL
, NULL
, NULL
, 0);
245 kmem_cache_destroy(sa_cache
);
248 kmem_cache_destroy(spill_cache
);
252 sa_spill_alloc(int flags
)
254 return (kmem_cache_alloc(spill_cache
, flags
));
258 sa_spill_free(void *obj
)
260 kmem_cache_free(spill_cache
, obj
);
264 layout_num_compare(const void *arg1
, const void *arg2
)
266 const sa_lot_t
*node1
= arg1
;
267 const sa_lot_t
*node2
= arg2
;
269 if (node1
->lot_num
> node2
->lot_num
)
271 else if (node1
->lot_num
< node2
->lot_num
)
277 layout_hash_compare(const void *arg1
, const void *arg2
)
279 const sa_lot_t
*node1
= arg1
;
280 const sa_lot_t
*node2
= arg2
;
282 if (node1
->lot_hash
> node2
->lot_hash
)
284 if (node1
->lot_hash
< node2
->lot_hash
)
286 if (node1
->lot_instance
> node2
->lot_instance
)
288 if (node1
->lot_instance
< node2
->lot_instance
)
294 sa_layout_equal(sa_lot_t
*tbf
, sa_attr_type_t
*attrs
, int count
)
298 if (count
!= tbf
->lot_attr_count
)
301 for (i
= 0; i
!= count
; i
++) {
302 if (attrs
[i
] != tbf
->lot_attrs
[i
])
308 #define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
311 sa_layout_info_hash(sa_attr_type_t
*attrs
, int attr_count
)
314 uint64_t crc
= -1ULL;
316 for (i
= 0; i
!= attr_count
; i
++)
317 crc
^= SA_ATTR_HASH(attrs
[i
]);
323 sa_get_spill(sa_handle_t
*hdl
)
326 if (hdl
->sa_spill
== NULL
) {
327 if ((rc
= dmu_spill_hold_existing(hdl
->sa_bonus
, NULL
,
328 &hdl
->sa_spill
)) == 0)
329 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
338 * Main attribute lookup/update function
339 * returns 0 for success or non zero for failures
341 * Operates on bulk array, first failure will abort further processing
344 sa_attr_op(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
345 sa_data_op_t data_op
, dmu_tx_t
*tx
)
347 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
350 sa_buf_type_t buftypes
;
355 for (i
= 0; i
!= count
; i
++) {
356 ASSERT(bulk
[i
].sa_attr
<= hdl
->sa_os
->os_sa
->sa_num_attrs
);
358 bulk
[i
].sa_addr
= NULL
;
359 /* First check the bonus buffer */
361 if (hdl
->sa_bonus_tab
&& TOC_ATTR_PRESENT(
362 hdl
->sa_bonus_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
363 SA_ATTR_INFO(sa
, hdl
->sa_bonus_tab
,
364 SA_GET_HDR(hdl
, SA_BONUS
),
365 bulk
[i
].sa_attr
, bulk
[i
], SA_BONUS
, hdl
);
366 if (tx
&& !(buftypes
& SA_BONUS
)) {
367 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
368 buftypes
|= SA_BONUS
;
371 if (bulk
[i
].sa_addr
== NULL
&&
372 ((error
= sa_get_spill(hdl
)) == 0)) {
373 if (TOC_ATTR_PRESENT(
374 hdl
->sa_spill_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
375 SA_ATTR_INFO(sa
, hdl
->sa_spill_tab
,
376 SA_GET_HDR(hdl
, SA_SPILL
),
377 bulk
[i
].sa_attr
, bulk
[i
], SA_SPILL
, hdl
);
378 if (tx
&& !(buftypes
& SA_SPILL
) &&
379 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
380 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
381 buftypes
|= SA_SPILL
;
385 if (error
&& error
!= ENOENT
) {
386 return ((error
== ECKSUM
) ? EIO
: error
);
391 if (bulk
[i
].sa_addr
== NULL
)
392 return (SET_ERROR(ENOENT
));
393 if (bulk
[i
].sa_data
) {
394 SA_COPY_DATA(bulk
[i
].sa_data_func
,
395 bulk
[i
].sa_addr
, bulk
[i
].sa_data
,
401 /* existing rewrite of attr */
402 if (bulk
[i
].sa_addr
&&
403 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
404 SA_COPY_DATA(bulk
[i
].sa_data_func
,
405 bulk
[i
].sa_data
, bulk
[i
].sa_addr
,
408 } else if (bulk
[i
].sa_addr
) { /* attr size change */
409 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
410 SA_REPLACE
, bulk
[i
].sa_data_func
,
411 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
412 } else { /* adding new attribute */
413 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
414 SA_ADD
, bulk
[i
].sa_data_func
,
415 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
428 sa_add_layout_entry(objset_t
*os
, sa_attr_type_t
*attrs
, int attr_count
,
429 uint64_t lot_num
, uint64_t hash
, boolean_t zapadd
, dmu_tx_t
*tx
)
431 sa_os_t
*sa
= os
->os_sa
;
432 sa_lot_t
*tb
, *findtb
;
436 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
437 tb
= kmem_zalloc(sizeof (sa_lot_t
), KM_PUSHPAGE
);
438 tb
->lot_attr_count
= attr_count
;
439 tb
->lot_attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
441 bcopy(attrs
, tb
->lot_attrs
, sizeof (sa_attr_type_t
) * attr_count
);
442 tb
->lot_num
= lot_num
;
444 tb
->lot_instance
= 0;
449 if (sa
->sa_layout_attr_obj
== 0) {
450 sa
->sa_layout_attr_obj
= zap_create_link(os
,
451 DMU_OT_SA_ATTR_LAYOUTS
,
452 sa
->sa_master_obj
, SA_LAYOUTS
, tx
);
455 (void) snprintf(attr_name
, sizeof (attr_name
),
457 VERIFY(0 == zap_update(os
, os
->os_sa
->sa_layout_attr_obj
,
458 attr_name
, 2, attr_count
, attrs
, tx
));
461 list_create(&tb
->lot_idx_tab
, sizeof (sa_idx_tab_t
),
462 offsetof(sa_idx_tab_t
, sa_next
));
464 for (i
= 0; i
!= attr_count
; i
++) {
465 if (sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
== 0)
469 avl_add(&sa
->sa_layout_num_tree
, tb
);
471 /* verify we don't have a hash collision */
472 if ((findtb
= avl_find(&sa
->sa_layout_hash_tree
, tb
, &loc
)) != NULL
) {
473 for (; findtb
&& findtb
->lot_hash
== hash
;
474 findtb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, findtb
)) {
475 if (findtb
->lot_instance
!= tb
->lot_instance
)
480 avl_add(&sa
->sa_layout_hash_tree
, tb
);
485 sa_find_layout(objset_t
*os
, uint64_t hash
, sa_attr_type_t
*attrs
,
486 int count
, dmu_tx_t
*tx
, sa_lot_t
**lot
)
488 sa_lot_t
*tb
, tbsearch
;
490 sa_os_t
*sa
= os
->os_sa
;
491 boolean_t found
= B_FALSE
;
493 mutex_enter(&sa
->sa_lock
);
494 tbsearch
.lot_hash
= hash
;
495 tbsearch
.lot_instance
= 0;
496 tb
= avl_find(&sa
->sa_layout_hash_tree
, &tbsearch
, &loc
);
498 for (; tb
&& tb
->lot_hash
== hash
;
499 tb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, tb
)) {
500 if (sa_layout_equal(tb
, attrs
, count
) == 0) {
507 tb
= sa_add_layout_entry(os
, attrs
, count
,
508 avl_numnodes(&sa
->sa_layout_num_tree
), hash
, B_TRUE
, tx
);
510 mutex_exit(&sa
->sa_lock
);
515 sa_resize_spill(sa_handle_t
*hdl
, uint32_t size
, dmu_tx_t
*tx
)
521 blocksize
= SPA_MINBLOCKSIZE
;
522 } else if (size
> SPA_MAXBLOCKSIZE
) {
524 return (SET_ERROR(EFBIG
));
526 blocksize
= P2ROUNDUP_TYPED(size
, SPA_MINBLOCKSIZE
, uint32_t);
529 error
= dbuf_spill_set_blksz(hdl
->sa_spill
, blocksize
, tx
);
535 sa_copy_data(sa_data_locator_t
*func
, void *datastart
, void *target
, int buflen
)
538 bcopy(datastart
, target
, buflen
);
543 void *saptr
= target
;
548 while (bytes
< buflen
) {
549 func(&dataptr
, &length
, buflen
, start
, datastart
);
550 bcopy(dataptr
, saptr
, length
);
551 saptr
= (void *)((caddr_t
)saptr
+ length
);
559 * Determine several different sizes
560 * first the sa header size
561 * the number of bytes to be stored
562 * if spill would occur the index in the attribute array is returned
564 * the boolean will_spill will be set when spilling is necessary. It
565 * is only set when the buftype is SA_BONUS
568 sa_find_sizes(sa_os_t
*sa
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
569 dmu_buf_t
*db
, sa_buf_type_t buftype
, int *index
, int *total
,
570 boolean_t
*will_spill
)
578 if (buftype
== SA_BONUS
&& sa
->sa_force_spill
) {
581 *will_spill
= B_TRUE
;
587 *will_spill
= B_FALSE
;
590 hdrsize
= (SA_BONUSTYPE_FROM_DB(db
) == DMU_OT_ZNODE
) ? 0 :
591 sizeof (sa_hdr_phys_t
);
593 full_space
= (buftype
== SA_BONUS
) ? DN_MAX_BONUSLEN
: db
->db_size
;
594 ASSERT(IS_P2ALIGNED(full_space
, 8));
596 for (i
= 0; i
!= attr_count
; i
++) {
599 *total
= P2ROUNDUP(*total
, 8);
600 *total
+= attr_desc
[i
].sa_length
;
604 is_var_sz
= (SA_REGISTERED_LEN(sa
, attr_desc
[i
].sa_attr
) == 0);
609 if (is_var_sz
&& var_size
> 1) {
611 * Don't worry that the spill block might overflow.
612 * It will be resized if needed in sa_build_layouts().
614 if (buftype
== SA_SPILL
||
615 P2ROUNDUP(hdrsize
+ sizeof (uint16_t), 8) +
616 *total
< full_space
) {
618 * Account for header space used by array of
619 * optional sizes of variable-length attributes.
620 * Record the extra header size in case this
621 * increase needs to be reversed due to
624 hdrsize
+= sizeof (uint16_t);
626 extra_hdrsize
+= sizeof (uint16_t);
628 ASSERT(buftype
== SA_BONUS
);
631 *will_spill
= B_TRUE
;
637 * find index of where spill *could* occur.
638 * Then continue to count of remainder attribute
639 * space. The sum is used later for sizing bonus
642 if (buftype
== SA_BONUS
&& *index
== -1 &&
643 (*total
+ P2ROUNDUP(hdrsize
, 8)) >
644 (full_space
- sizeof (blkptr_t
))) {
648 if ((*total
+ P2ROUNDUP(hdrsize
, 8)) > full_space
&&
650 *will_spill
= B_TRUE
;
654 hdrsize
-= extra_hdrsize
;
656 hdrsize
= P2ROUNDUP(hdrsize
, 8);
660 #define BUF_SPACE_NEEDED(total, header) (total + header)
663 * Find layout that corresponds to ordering of attributes
664 * If not found a new layout number is created and added to
665 * persistent layout tables.
668 sa_build_layouts(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
671 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
673 sa_buf_type_t buftype
;
674 sa_hdr_phys_t
*sahdr
;
677 sa_attr_type_t
*attrs
, *attrs_start
;
680 int spillhdrsize
= 0;
682 dmu_object_type_t bonustype
;
688 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
689 bonustype
= SA_BONUSTYPE_FROM_DB(hdl
->sa_bonus
);
691 /* first determine bonus header size and sum of all attributes */
692 hdrsize
= sa_find_sizes(sa
, attr_desc
, attr_count
, hdl
->sa_bonus
,
693 SA_BONUS
, &i
, &used
, &spilling
);
695 if (used
> SPA_MAXBLOCKSIZE
)
696 return (SET_ERROR(EFBIG
));
698 VERIFY(0 == dmu_set_bonus(hdl
->sa_bonus
, spilling
?
699 MIN(DN_MAX_BONUSLEN
- sizeof (blkptr_t
), used
+ hdrsize
) :
700 used
+ hdrsize
, tx
));
702 ASSERT((bonustype
== DMU_OT_ZNODE
&& spilling
== 0) ||
703 bonustype
== DMU_OT_SA
);
705 /* setup and size spill buffer when needed */
709 if (hdl
->sa_spill
== NULL
) {
710 VERIFY(dmu_spill_hold_by_bonus(hdl
->sa_bonus
, NULL
,
711 &hdl
->sa_spill
) == 0);
713 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
715 spillhdrsize
= sa_find_sizes(sa
, &attr_desc
[i
],
716 attr_count
- i
, hdl
->sa_spill
, SA_SPILL
, &i
,
717 &spill_used
, &dummy
);
719 if (spill_used
> SPA_MAXBLOCKSIZE
)
720 return (SET_ERROR(EFBIG
));
722 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
723 if (BUF_SPACE_NEEDED(spill_used
, spillhdrsize
) >
724 hdl
->sa_spill
->db_size
)
725 VERIFY(0 == sa_resize_spill(hdl
,
726 BUF_SPACE_NEEDED(spill_used
, spillhdrsize
), tx
));
729 /* setup starting pointers to lay down data */
730 data_start
= (void *)((uintptr_t)hdl
->sa_bonus
->db_data
+ hdrsize
);
731 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_bonus
->db_data
;
735 buf_space
= (sa
->sa_force_spill
) ?
736 0 : SA_BLKPTR_SPACE
- hdrsize
;
738 buf_space
= hdl
->sa_bonus
->db_size
- hdrsize
;
740 attrs_start
= attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
744 for (i
= 0, len_idx
= 0, hash
= -1ULL; i
!= attr_count
; i
++) {
747 ASSERT(IS_P2ALIGNED(data_start
, 8));
748 ASSERT(IS_P2ALIGNED(buf_space
, 8));
749 attrs
[i
] = attr_desc
[i
].sa_attr
;
750 length
= SA_REGISTERED_LEN(sa
, attrs
[i
]);
752 length
= attr_desc
[i
].sa_length
;
754 if (buf_space
< length
) { /* switch to spill buffer */
756 VERIFY(bonustype
== DMU_OT_SA
);
757 if (buftype
== SA_BONUS
&& !sa
->sa_force_spill
) {
758 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
,
759 lot_count
, tx
, &lot
);
760 SA_SET_HDR(sahdr
, lot
->lot_num
, hdrsize
);
767 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_spill
->db_data
;
768 sahdr
->sa_magic
= SA_MAGIC
;
769 data_start
= (void *)((uintptr_t)sahdr
+
771 attrs_start
= &attrs
[i
];
772 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
775 hash
^= SA_ATTR_HASH(attrs
[i
]);
776 attr_desc
[i
].sa_addr
= data_start
;
777 attr_desc
[i
].sa_size
= length
;
778 SA_COPY_DATA(attr_desc
[i
].sa_data_func
, attr_desc
[i
].sa_data
,
780 if (sa
->sa_attr_table
[attrs
[i
]].sa_length
== 0) {
781 sahdr
->sa_lengths
[len_idx
++] = length
;
783 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
785 buf_space
-= P2ROUNDUP(length
, 8);
789 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
, lot_count
, tx
, &lot
);
792 * Verify that old znodes always have layout number 0.
793 * Must be DMU_OT_SA for arbitrary layouts
795 VERIFY((bonustype
== DMU_OT_ZNODE
&& lot
->lot_num
== 0) ||
796 (bonustype
== DMU_OT_SA
&& lot
->lot_num
> 1));
798 if (bonustype
== DMU_OT_SA
) {
799 SA_SET_HDR(sahdr
, lot
->lot_num
,
800 buftype
== SA_BONUS
? hdrsize
: spillhdrsize
);
803 kmem_free(attrs
, sizeof (sa_attr_type_t
) * attr_count
);
804 if (hdl
->sa_bonus_tab
) {
805 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
806 hdl
->sa_bonus_tab
= NULL
;
808 if (!sa
->sa_force_spill
)
809 VERIFY(0 == sa_build_index(hdl
, SA_BONUS
));
811 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
814 * remove spill block that is no longer needed.
816 dmu_buf_rele(hdl
->sa_spill
, NULL
);
817 hdl
->sa_spill
= NULL
;
818 hdl
->sa_spill_tab
= NULL
;
819 VERIFY(0 == dmu_rm_spill(hdl
->sa_os
,
820 sa_handle_object(hdl
), tx
));
822 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
830 sa_free_attr_table(sa_os_t
*sa
)
834 if (sa
->sa_attr_table
== NULL
)
837 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
838 if (sa
->sa_attr_table
[i
].sa_name
)
839 kmem_free(sa
->sa_attr_table
[i
].sa_name
,
840 strlen(sa
->sa_attr_table
[i
].sa_name
) + 1);
843 kmem_free(sa
->sa_attr_table
,
844 sizeof (sa_attr_table_t
) * sa
->sa_num_attrs
);
846 sa
->sa_attr_table
= NULL
;
850 sa_attr_table_setup(objset_t
*os
, sa_attr_reg_t
*reg_attrs
, int count
)
852 sa_os_t
*sa
= os
->os_sa
;
853 uint64_t sa_attr_count
= 0;
854 uint64_t sa_reg_count
= 0;
860 int registered_count
= 0;
862 dmu_objset_type_t ostype
= dmu_objset_type(os
);
865 kmem_zalloc(count
* sizeof (sa_attr_type_t
), KM_PUSHPAGE
);
866 sa
->sa_user_table_sz
= count
* sizeof (sa_attr_type_t
);
868 if (sa
->sa_reg_attr_obj
!= 0) {
869 error
= zap_count(os
, sa
->sa_reg_attr_obj
,
873 * Make sure we retrieved a count and that it isn't zero
875 if (error
|| (error
== 0 && sa_attr_count
== 0)) {
877 error
= SET_ERROR(EINVAL
);
880 sa_reg_count
= sa_attr_count
;
883 if (ostype
== DMU_OST_ZFS
&& sa_attr_count
== 0)
884 sa_attr_count
+= sa_legacy_attr_count
;
886 /* Allocate attribute numbers for attributes that aren't registered */
887 for (i
= 0; i
!= count
; i
++) {
888 boolean_t found
= B_FALSE
;
891 if (ostype
== DMU_OST_ZFS
) {
892 for (j
= 0; j
!= sa_legacy_attr_count
; j
++) {
893 if (strcmp(reg_attrs
[i
].sa_name
,
894 sa_legacy_attrs
[j
].sa_name
) == 0) {
895 sa
->sa_user_table
[i
] =
896 sa_legacy_attrs
[j
].sa_attr
;
904 if (sa
->sa_reg_attr_obj
)
905 error
= zap_lookup(os
, sa
->sa_reg_attr_obj
,
906 reg_attrs
[i
].sa_name
, 8, 1, &attr_value
);
908 error
= SET_ERROR(ENOENT
);
911 sa
->sa_user_table
[i
] = (sa_attr_type_t
)sa_attr_count
;
915 sa
->sa_user_table
[i
] = ATTR_NUM(attr_value
);
922 sa
->sa_num_attrs
= sa_attr_count
;
923 tb
= sa
->sa_attr_table
=
924 kmem_zalloc(sizeof (sa_attr_table_t
) * sa_attr_count
, KM_PUSHPAGE
);
927 * Attribute table is constructed from requested attribute list,
928 * previously foreign registered attributes, and also the legacy
929 * ZPL set of attributes.
932 if (sa
->sa_reg_attr_obj
) {
933 for (zap_cursor_init(&zc
, os
, sa
->sa_reg_attr_obj
);
934 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
935 zap_cursor_advance(&zc
)) {
937 value
= za
.za_first_integer
;
940 tb
[ATTR_NUM(value
)].sa_attr
= ATTR_NUM(value
);
941 tb
[ATTR_NUM(value
)].sa_length
= ATTR_LENGTH(value
);
942 tb
[ATTR_NUM(value
)].sa_byteswap
= ATTR_BSWAP(value
);
943 tb
[ATTR_NUM(value
)].sa_registered
= B_TRUE
;
945 if (tb
[ATTR_NUM(value
)].sa_name
) {
948 tb
[ATTR_NUM(value
)].sa_name
=
949 kmem_zalloc(strlen(za
.za_name
) +1, KM_PUSHPAGE
);
950 (void) strlcpy(tb
[ATTR_NUM(value
)].sa_name
, za
.za_name
,
951 strlen(za
.za_name
) +1);
953 zap_cursor_fini(&zc
);
955 * Make sure we processed the correct number of registered
958 if (registered_count
!= sa_reg_count
) {
965 if (ostype
== DMU_OST_ZFS
) {
966 for (i
= 0; i
!= sa_legacy_attr_count
; i
++) {
969 tb
[i
].sa_attr
= sa_legacy_attrs
[i
].sa_attr
;
970 tb
[i
].sa_length
= sa_legacy_attrs
[i
].sa_length
;
971 tb
[i
].sa_byteswap
= sa_legacy_attrs
[i
].sa_byteswap
;
972 tb
[i
].sa_registered
= B_FALSE
;
974 kmem_zalloc(strlen(sa_legacy_attrs
[i
].sa_name
) +1,
976 (void) strlcpy(tb
[i
].sa_name
,
977 sa_legacy_attrs
[i
].sa_name
,
978 strlen(sa_legacy_attrs
[i
].sa_name
) + 1);
982 for (i
= 0; i
!= count
; i
++) {
983 sa_attr_type_t attr_id
;
985 attr_id
= sa
->sa_user_table
[i
];
986 if (tb
[attr_id
].sa_name
)
989 tb
[attr_id
].sa_length
= reg_attrs
[i
].sa_length
;
990 tb
[attr_id
].sa_byteswap
= reg_attrs
[i
].sa_byteswap
;
991 tb
[attr_id
].sa_attr
= attr_id
;
992 tb
[attr_id
].sa_name
=
993 kmem_zalloc(strlen(reg_attrs
[i
].sa_name
) + 1, KM_PUSHPAGE
);
994 (void) strlcpy(tb
[attr_id
].sa_name
, reg_attrs
[i
].sa_name
,
995 strlen(reg_attrs
[i
].sa_name
) + 1);
998 sa
->sa_need_attr_registration
=
999 (sa_attr_count
!= registered_count
);
1003 kmem_free(sa
->sa_user_table
, count
* sizeof (sa_attr_type_t
));
1004 sa
->sa_user_table
= NULL
;
1005 sa_free_attr_table(sa
);
1006 return ((error
!= 0) ? error
: EINVAL
);
1010 sa_setup(objset_t
*os
, uint64_t sa_obj
, sa_attr_reg_t
*reg_attrs
, int count
,
1011 sa_attr_type_t
**user_table
)
1016 dmu_objset_type_t ostype
= dmu_objset_type(os
);
1020 mutex_enter(&os
->os_user_ptr_lock
);
1022 mutex_enter(&os
->os_sa
->sa_lock
);
1023 mutex_exit(&os
->os_user_ptr_lock
);
1024 tb
= os
->os_sa
->sa_user_table
;
1025 mutex_exit(&os
->os_sa
->sa_lock
);
1030 sa
= kmem_zalloc(sizeof (sa_os_t
), KM_PUSHPAGE
);
1031 mutex_init(&sa
->sa_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1032 sa
->sa_master_obj
= sa_obj
;
1035 mutex_enter(&sa
->sa_lock
);
1036 mutex_exit(&os
->os_user_ptr_lock
);
1037 avl_create(&sa
->sa_layout_num_tree
, layout_num_compare
,
1038 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_num_node
));
1039 avl_create(&sa
->sa_layout_hash_tree
, layout_hash_compare
,
1040 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_hash_node
));
1043 error
= zap_lookup(os
, sa_obj
, SA_LAYOUTS
,
1044 8, 1, &sa
->sa_layout_attr_obj
);
1045 if (error
!= 0 && error
!= ENOENT
)
1047 error
= zap_lookup(os
, sa_obj
, SA_REGISTRY
,
1048 8, 1, &sa
->sa_reg_attr_obj
);
1049 if (error
!= 0 && error
!= ENOENT
)
1053 if ((error
= sa_attr_table_setup(os
, reg_attrs
, count
)) != 0)
1056 if (sa
->sa_layout_attr_obj
!= 0) {
1057 uint64_t layout_count
;
1059 error
= zap_count(os
, sa
->sa_layout_attr_obj
,
1063 * Layout number count should be > 0
1065 if (error
|| (error
== 0 && layout_count
== 0)) {
1067 error
= SET_ERROR(EINVAL
);
1071 for (zap_cursor_init(&zc
, os
, sa
->sa_layout_attr_obj
);
1072 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
1073 zap_cursor_advance(&zc
)) {
1074 sa_attr_type_t
*lot_attrs
;
1077 lot_attrs
= kmem_zalloc(sizeof (sa_attr_type_t
) *
1078 za
.za_num_integers
, KM_PUSHPAGE
);
1080 if ((error
= (zap_lookup(os
, sa
->sa_layout_attr_obj
,
1081 za
.za_name
, 2, za
.za_num_integers
,
1082 lot_attrs
))) != 0) {
1083 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1084 za
.za_num_integers
);
1087 VERIFY(ddi_strtoull(za
.za_name
, NULL
, 10,
1088 (unsigned long long *)&lot_num
) == 0);
1090 (void) sa_add_layout_entry(os
, lot_attrs
,
1091 za
.za_num_integers
, lot_num
,
1092 sa_layout_info_hash(lot_attrs
,
1093 za
.za_num_integers
), B_FALSE
, NULL
);
1094 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1095 za
.za_num_integers
);
1097 zap_cursor_fini(&zc
);
1100 * Make sure layout count matches number of entries added
1103 if (avl_numnodes(&sa
->sa_layout_num_tree
) != layout_count
) {
1109 /* Add special layout number for old ZNODES */
1110 if (ostype
== DMU_OST_ZFS
) {
1111 (void) sa_add_layout_entry(os
, sa_legacy_zpl_layout
,
1112 sa_legacy_attr_count
, 0,
1113 sa_layout_info_hash(sa_legacy_zpl_layout
,
1114 sa_legacy_attr_count
), B_FALSE
, NULL
);
1116 (void) sa_add_layout_entry(os
, sa_dummy_zpl_layout
, 0, 1,
1119 *user_table
= os
->os_sa
->sa_user_table
;
1120 mutex_exit(&sa
->sa_lock
);
1124 sa_free_attr_table(sa
);
1125 if (sa
->sa_user_table
)
1126 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1127 mutex_exit(&sa
->sa_lock
);
1128 kmem_free(sa
, sizeof (sa_os_t
));
1129 return ((error
== ECKSUM
) ? EIO
: error
);
1133 sa_tear_down(objset_t
*os
)
1135 sa_os_t
*sa
= os
->os_sa
;
1139 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1141 /* Free up attr table */
1143 sa_free_attr_table(sa
);
1147 avl_destroy_nodes(&sa
->sa_layout_hash_tree
, &cookie
))) {
1149 while ((tab
= list_head(&layout
->lot_idx_tab
))) {
1150 ASSERT(refcount_count(&tab
->sa_refcount
));
1151 sa_idx_tab_rele(os
, tab
);
1156 while ((layout
= avl_destroy_nodes(&sa
->sa_layout_num_tree
, &cookie
))) {
1157 kmem_free(layout
->lot_attrs
,
1158 sizeof (sa_attr_type_t
) * layout
->lot_attr_count
);
1159 kmem_free(layout
, sizeof (sa_lot_t
));
1162 avl_destroy(&sa
->sa_layout_hash_tree
);
1163 avl_destroy(&sa
->sa_layout_num_tree
);
1165 kmem_free(sa
, sizeof (sa_os_t
));
1170 sa_build_idx_tab(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1171 uint16_t length
, int length_idx
, boolean_t var_length
, void *userp
)
1173 sa_idx_tab_t
*idx_tab
= userp
;
1176 ASSERT(idx_tab
->sa_variable_lengths
);
1177 idx_tab
->sa_variable_lengths
[length_idx
] = length
;
1179 TOC_ATTR_ENCODE(idx_tab
->sa_idx_tab
[attr
], length_idx
,
1180 (uint32_t)((uintptr_t)attr_addr
- (uintptr_t)hdr
));
1184 sa_attr_iter(objset_t
*os
, sa_hdr_phys_t
*hdr
, dmu_object_type_t type
,
1185 sa_iterfunc_t func
, sa_lot_t
*tab
, void *userp
)
1191 sa_os_t
*sa
= os
->os_sa
;
1193 uint16_t *length_start
= NULL
;
1194 uint8_t length_idx
= 0;
1197 search
.lot_num
= SA_LAYOUT_NUM(hdr
, type
);
1198 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1202 if (IS_SA_BONUSTYPE(type
)) {
1203 data_start
= (void *)P2ROUNDUP(((uintptr_t)hdr
+
1204 offsetof(sa_hdr_phys_t
, sa_lengths
) +
1205 (sizeof (uint16_t) * tb
->lot_var_sizes
)), 8);
1206 length_start
= hdr
->sa_lengths
;
1211 for (i
= 0; i
!= tb
->lot_attr_count
; i
++) {
1212 int attr_length
, reg_length
;
1215 reg_length
= sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
;
1217 attr_length
= reg_length
;
1220 attr_length
= length_start
[length_idx
];
1221 idx_len
= length_idx
++;
1224 func(hdr
, data_start
, tb
->lot_attrs
[i
], attr_length
,
1225 idx_len
, reg_length
== 0 ? B_TRUE
: B_FALSE
, userp
);
1227 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
1234 sa_byteswap_cb(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1235 uint16_t length
, int length_idx
, boolean_t variable_length
, void *userp
)
1237 sa_handle_t
*hdl
= userp
;
1238 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1240 sa_bswap_table
[sa
->sa_attr_table
[attr
].sa_byteswap
](attr_addr
, length
);
1244 sa_byteswap(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1246 sa_hdr_phys_t
*sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1248 int num_lengths
= 1;
1250 ASSERTV(sa_os_t
*sa
= hdl
->sa_os
->os_sa
);
1252 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1253 if (sa_hdr_phys
->sa_magic
== SA_MAGIC
)
1256 db
= SA_GET_DB(hdl
, buftype
);
1258 if (buftype
== SA_SPILL
) {
1259 arc_release(db
->db_buf
, NULL
);
1260 arc_buf_thaw(db
->db_buf
);
1263 sa_hdr_phys
->sa_magic
= BSWAP_32(sa_hdr_phys
->sa_magic
);
1264 sa_hdr_phys
->sa_layout_info
= BSWAP_16(sa_hdr_phys
->sa_layout_info
);
1267 * Determine number of variable lenghts in header
1268 * The standard 8 byte header has one for free and a
1269 * 16 byte header would have 4 + 1;
1271 if (SA_HDR_SIZE(sa_hdr_phys
) > 8)
1272 num_lengths
+= (SA_HDR_SIZE(sa_hdr_phys
) - 8) >> 1;
1273 for (i
= 0; i
!= num_lengths
; i
++)
1274 sa_hdr_phys
->sa_lengths
[i
] =
1275 BSWAP_16(sa_hdr_phys
->sa_lengths
[i
]);
1277 sa_attr_iter(hdl
->sa_os
, sa_hdr_phys
, DMU_OT_SA
,
1278 sa_byteswap_cb
, NULL
, hdl
);
1280 if (buftype
== SA_SPILL
)
1281 arc_buf_freeze(((dmu_buf_impl_t
*)hdl
->sa_spill
)->db_buf
);
1285 sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1287 sa_hdr_phys_t
*sa_hdr_phys
;
1288 dmu_buf_impl_t
*db
= SA_GET_DB(hdl
, buftype
);
1289 dmu_object_type_t bonustype
= SA_BONUSTYPE_FROM_DB(db
);
1290 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1291 sa_idx_tab_t
*idx_tab
;
1293 sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1295 mutex_enter(&sa
->sa_lock
);
1297 /* Do we need to byteswap? */
1299 /* only check if not old znode */
1300 if (IS_SA_BONUSTYPE(bonustype
) && sa_hdr_phys
->sa_magic
!= SA_MAGIC
&&
1301 sa_hdr_phys
->sa_magic
!= 0) {
1302 VERIFY(BSWAP_32(sa_hdr_phys
->sa_magic
) == SA_MAGIC
);
1303 sa_byteswap(hdl
, buftype
);
1306 idx_tab
= sa_find_idx_tab(hdl
->sa_os
, bonustype
, sa_hdr_phys
);
1308 if (buftype
== SA_BONUS
)
1309 hdl
->sa_bonus_tab
= idx_tab
;
1311 hdl
->sa_spill_tab
= idx_tab
;
1313 mutex_exit(&sa
->sa_lock
);
1319 sa_evict(dmu_buf_t
*db
, void *sap
)
1321 panic("evicting sa dbuf %p\n", (void *)db
);
1325 sa_idx_tab_rele(objset_t
*os
, void *arg
)
1327 sa_os_t
*sa
= os
->os_sa
;
1328 sa_idx_tab_t
*idx_tab
= arg
;
1330 if (idx_tab
== NULL
)
1333 mutex_enter(&sa
->sa_lock
);
1334 if (refcount_remove(&idx_tab
->sa_refcount
, NULL
) == 0) {
1335 list_remove(&idx_tab
->sa_layout
->lot_idx_tab
, idx_tab
);
1336 if (idx_tab
->sa_variable_lengths
)
1337 kmem_free(idx_tab
->sa_variable_lengths
,
1339 idx_tab
->sa_layout
->lot_var_sizes
);
1340 refcount_destroy(&idx_tab
->sa_refcount
);
1341 kmem_free(idx_tab
->sa_idx_tab
,
1342 sizeof (uint32_t) * sa
->sa_num_attrs
);
1343 kmem_free(idx_tab
, sizeof (sa_idx_tab_t
));
1345 mutex_exit(&sa
->sa_lock
);
1349 sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
)
1351 ASSERTV(sa_os_t
*sa
= os
->os_sa
);
1353 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1354 (void) refcount_add(&idx_tab
->sa_refcount
, NULL
);
1358 sa_spill_rele(sa_handle_t
*hdl
)
1360 mutex_enter(&hdl
->sa_lock
);
1361 if (hdl
->sa_spill
) {
1362 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1363 dmu_buf_rele(hdl
->sa_spill
, NULL
);
1364 hdl
->sa_spill
= NULL
;
1365 hdl
->sa_spill_tab
= NULL
;
1367 mutex_exit(&hdl
->sa_lock
);
1371 sa_handle_destroy(sa_handle_t
*hdl
)
1373 mutex_enter(&hdl
->sa_lock
);
1374 (void) dmu_buf_update_user((dmu_buf_t
*)hdl
->sa_bonus
, hdl
,
1377 if (hdl
->sa_bonus_tab
) {
1378 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
1379 hdl
->sa_bonus_tab
= NULL
;
1381 if (hdl
->sa_spill_tab
) {
1382 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1383 hdl
->sa_spill_tab
= NULL
;
1386 dmu_buf_rele(hdl
->sa_bonus
, NULL
);
1389 dmu_buf_rele((dmu_buf_t
*)hdl
->sa_spill
, NULL
);
1390 mutex_exit(&hdl
->sa_lock
);
1392 kmem_cache_free(sa_cache
, hdl
);
1396 sa_handle_get_from_db(objset_t
*os
, dmu_buf_t
*db
, void *userp
,
1397 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1400 sa_handle_t
*handle
;
1402 dmu_object_info_t doi
;
1404 dmu_object_info_from_db(db
, &doi
);
1405 ASSERT(doi
.doi_bonus_type
== DMU_OT_SA
||
1406 doi
.doi_bonus_type
== DMU_OT_ZNODE
);
1408 /* find handle, if it exists */
1409 /* if one doesn't exist then create a new one, and initialize it */
1411 handle
= (hdl_type
== SA_HDL_SHARED
) ? dmu_buf_get_user(db
) : NULL
;
1412 if (handle
== NULL
) {
1413 sa_handle_t
*newhandle
;
1414 handle
= kmem_cache_alloc(sa_cache
, KM_SLEEP
);
1415 handle
->sa_userp
= userp
;
1416 handle
->sa_bonus
= db
;
1418 handle
->sa_spill
= NULL
;
1420 error
= sa_build_index(handle
, SA_BONUS
);
1421 newhandle
= (hdl_type
== SA_HDL_SHARED
) ?
1422 dmu_buf_set_user_ie(db
, handle
,
1423 NULL
, sa_evict
) : NULL
;
1425 if (newhandle
!= NULL
) {
1426 kmem_cache_free(sa_cache
, handle
);
1436 sa_handle_get(objset_t
*objset
, uint64_t objid
, void *userp
,
1437 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1442 if ((error
= dmu_bonus_hold(objset
, objid
, NULL
, &db
)))
1445 return (sa_handle_get_from_db(objset
, db
, userp
, hdl_type
,
1450 sa_buf_hold(objset_t
*objset
, uint64_t obj_num
, void *tag
, dmu_buf_t
**db
)
1452 return (dmu_bonus_hold(objset
, obj_num
, tag
, db
));
1456 sa_buf_rele(dmu_buf_t
*db
, void *tag
)
1458 dmu_buf_rele(db
, tag
);
1462 sa_lookup_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
)
1465 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1466 return (sa_attr_op(hdl
, bulk
, count
, SA_LOOKUP
, NULL
));
1470 sa_lookup(sa_handle_t
*hdl
, sa_attr_type_t attr
, void *buf
, uint32_t buflen
)
1473 sa_bulk_attr_t bulk
;
1475 bulk
.sa_attr
= attr
;
1477 bulk
.sa_length
= buflen
;
1478 bulk
.sa_data_func
= NULL
;
1481 mutex_enter(&hdl
->sa_lock
);
1482 error
= sa_lookup_impl(hdl
, &bulk
, 1);
1483 mutex_exit(&hdl
->sa_lock
);
1489 sa_lookup_uio(sa_handle_t
*hdl
, sa_attr_type_t attr
, uio_t
*uio
)
1492 sa_bulk_attr_t bulk
;
1494 bulk
.sa_data
= NULL
;
1495 bulk
.sa_attr
= attr
;
1496 bulk
.sa_data_func
= NULL
;
1500 mutex_enter(&hdl
->sa_lock
);
1501 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) == 0) {
1502 error
= uiomove((void *)bulk
.sa_addr
, MIN(bulk
.sa_size
,
1503 uio
->uio_resid
), UIO_READ
, uio
);
1505 mutex_exit(&hdl
->sa_lock
);
1511 sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
, void *data
)
1513 sa_idx_tab_t
*idx_tab
;
1514 sa_hdr_phys_t
*hdr
= (sa_hdr_phys_t
*)data
;
1515 sa_os_t
*sa
= os
->os_sa
;
1516 sa_lot_t
*tb
, search
;
1520 * Deterimine layout number. If SA node and header == 0 then
1521 * force the index table to the dummy "1" empty layout.
1523 * The layout number would only be zero for a newly created file
1524 * that has not added any attributes yet, or with crypto enabled which
1525 * doesn't write any attributes to the bonus buffer.
1528 search
.lot_num
= SA_LAYOUT_NUM(hdr
, bonustype
);
1530 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1532 /* Verify header size is consistent with layout information */
1534 ASSERT((IS_SA_BONUSTYPE(bonustype
) &&
1535 SA_HDR_SIZE_MATCH_LAYOUT(hdr
, tb
)) || !IS_SA_BONUSTYPE(bonustype
) ||
1536 (IS_SA_BONUSTYPE(bonustype
) && hdr
->sa_layout_info
== 0));
1539 * See if any of the already existing TOC entries can be reused?
1542 for (idx_tab
= list_head(&tb
->lot_idx_tab
); idx_tab
;
1543 idx_tab
= list_next(&tb
->lot_idx_tab
, idx_tab
)) {
1544 boolean_t valid_idx
= B_TRUE
;
1547 if (tb
->lot_var_sizes
!= 0 &&
1548 idx_tab
->sa_variable_lengths
!= NULL
) {
1549 for (i
= 0; i
!= tb
->lot_var_sizes
; i
++) {
1550 if (hdr
->sa_lengths
[i
] !=
1551 idx_tab
->sa_variable_lengths
[i
]) {
1552 valid_idx
= B_FALSE
;
1558 sa_idx_tab_hold(os
, idx_tab
);
1563 /* No such luck, create a new entry */
1564 idx_tab
= kmem_zalloc(sizeof (sa_idx_tab_t
), KM_PUSHPAGE
);
1565 idx_tab
->sa_idx_tab
=
1566 kmem_zalloc(sizeof (uint32_t) * sa
->sa_num_attrs
, KM_PUSHPAGE
);
1567 idx_tab
->sa_layout
= tb
;
1568 refcount_create(&idx_tab
->sa_refcount
);
1569 if (tb
->lot_var_sizes
)
1570 idx_tab
->sa_variable_lengths
= kmem_alloc(sizeof (uint16_t) *
1571 tb
->lot_var_sizes
, KM_PUSHPAGE
);
1573 sa_attr_iter(os
, hdr
, bonustype
, sa_build_idx_tab
,
1575 sa_idx_tab_hold(os
, idx_tab
); /* one hold for consumer */
1576 sa_idx_tab_hold(os
, idx_tab
); /* one for layout */
1577 list_insert_tail(&tb
->lot_idx_tab
, idx_tab
);
1582 sa_default_locator(void **dataptr
, uint32_t *len
, uint32_t total_len
,
1583 boolean_t start
, void *userdata
)
1587 *dataptr
= userdata
;
1592 sa_attr_register_sync(sa_handle_t
*hdl
, dmu_tx_t
*tx
)
1594 uint64_t attr_value
= 0;
1595 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1596 sa_attr_table_t
*tb
= sa
->sa_attr_table
;
1599 mutex_enter(&sa
->sa_lock
);
1601 if (!sa
->sa_need_attr_registration
|| sa
->sa_master_obj
== 0) {
1602 mutex_exit(&sa
->sa_lock
);
1606 if (sa
->sa_reg_attr_obj
== 0) {
1607 sa
->sa_reg_attr_obj
= zap_create_link(hdl
->sa_os
,
1608 DMU_OT_SA_ATTR_REGISTRATION
,
1609 sa
->sa_master_obj
, SA_REGISTRY
, tx
);
1611 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
1612 if (sa
->sa_attr_table
[i
].sa_registered
)
1614 ATTR_ENCODE(attr_value
, tb
[i
].sa_attr
, tb
[i
].sa_length
,
1616 VERIFY(0 == zap_update(hdl
->sa_os
, sa
->sa_reg_attr_obj
,
1617 tb
[i
].sa_name
, 8, 1, &attr_value
, tx
));
1618 tb
[i
].sa_registered
= B_TRUE
;
1620 sa
->sa_need_attr_registration
= B_FALSE
;
1621 mutex_exit(&sa
->sa_lock
);
1625 * Replace all attributes with attributes specified in template.
1626 * If dnode had a spill buffer then those attributes will be
1627 * also be replaced, possibly with just an empty spill block
1629 * This interface is intended to only be used for bulk adding of
1630 * attributes for a new file. It will also be used by the ZPL
1631 * when converting and old formatted znode to native SA support.
1634 sa_replace_all_by_template_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1635 int attr_count
, dmu_tx_t
*tx
)
1637 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1639 if (sa
->sa_need_attr_registration
)
1640 sa_attr_register_sync(hdl
, tx
);
1641 return (sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
));
1645 sa_replace_all_by_template(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1646 int attr_count
, dmu_tx_t
*tx
)
1650 mutex_enter(&hdl
->sa_lock
);
1651 error
= sa_replace_all_by_template_locked(hdl
, attr_desc
,
1653 mutex_exit(&hdl
->sa_lock
);
1658 * add/remove/replace a single attribute and then rewrite the entire set
1662 sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
1663 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
1664 uint16_t buflen
, dmu_tx_t
*tx
)
1666 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1667 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)hdl
->sa_bonus
;
1669 sa_bulk_attr_t
*attr_desc
;
1671 int bonus_attr_count
= 0;
1672 int bonus_data_size
= 0;
1673 int spill_attr_count
= 0;
1676 int i
, j
, k
, length_idx
;
1678 sa_idx_tab_t
*idx_tab
;
1682 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1684 /* First make of copy of the old data */
1688 if (dn
->dn_bonuslen
!= 0) {
1689 bonus_data_size
= hdl
->sa_bonus
->db_size
;
1690 old_data
[0] = kmem_alloc(bonus_data_size
, KM_SLEEP
);
1691 bcopy(hdl
->sa_bonus
->db_data
, old_data
[0],
1692 hdl
->sa_bonus
->db_size
);
1693 bonus_attr_count
= hdl
->sa_bonus_tab
->sa_layout
->lot_attr_count
;
1699 /* Bring spill buffer online if it isn't currently */
1701 if ((error
= sa_get_spill(hdl
)) == 0) {
1702 ASSERT3U(hdl
->sa_spill
->db_size
, <=, SPA_MAXBLOCKSIZE
);
1703 old_data
[1] = sa_spill_alloc(KM_SLEEP
);
1704 bcopy(hdl
->sa_spill
->db_data
, old_data
[1],
1705 hdl
->sa_spill
->db_size
);
1707 hdl
->sa_spill_tab
->sa_layout
->lot_attr_count
;
1708 } else if (error
&& error
!= ENOENT
) {
1710 kmem_free(old_data
[0], bonus_data_size
);
1716 /* build descriptor of all attributes */
1718 attr_count
= bonus_attr_count
+ spill_attr_count
;
1719 if (action
== SA_ADD
)
1721 else if (action
== SA_REMOVE
)
1724 attr_desc
= kmem_zalloc(sizeof (sa_bulk_attr_t
) * attr_count
, KM_SLEEP
);
1727 * loop through bonus and spill buffer if it exists, and
1728 * build up new attr_descriptor to reset the attributes
1731 count
= bonus_attr_count
;
1732 hdr
= SA_GET_HDR(hdl
, SA_BONUS
);
1733 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_BONUS
);
1734 for (; k
!= 2; k
++) {
1735 /* iterate over each attribute in layout */
1736 for (i
= 0, length_idx
= 0; i
!= count
; i
++) {
1737 sa_attr_type_t attr
;
1739 attr
= idx_tab
->sa_layout
->lot_attrs
[i
];
1740 if (attr
== newattr
) {
1741 if (action
== SA_REMOVE
) {
1745 ASSERT(SA_REGISTERED_LEN(sa
, attr
) == 0);
1746 ASSERT(action
== SA_REPLACE
);
1747 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1748 locator
, datastart
, buflen
);
1750 length
= SA_REGISTERED_LEN(sa
, attr
);
1752 length
= hdr
->sa_lengths
[length_idx
++];
1755 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1757 (TOC_OFF(idx_tab
->sa_idx_tab
[attr
]) +
1758 (uintptr_t)old_data
[k
]), length
);
1761 if (k
== 0 && hdl
->sa_spill
) {
1762 hdr
= SA_GET_HDR(hdl
, SA_SPILL
);
1763 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_SPILL
);
1764 count
= spill_attr_count
;
1769 if (action
== SA_ADD
) {
1770 length
= SA_REGISTERED_LEN(sa
, newattr
);
1774 SA_ADD_BULK_ATTR(attr_desc
, j
, newattr
, locator
,
1778 error
= sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
);
1781 kmem_free(old_data
[0], bonus_data_size
);
1783 sa_spill_free(old_data
[1]);
1784 kmem_free(attr_desc
, sizeof (sa_bulk_attr_t
) * attr_count
);
1790 sa_bulk_update_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
1794 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1795 dmu_object_type_t bonustype
;
1796 dmu_buf_t
*saved_spill
;
1799 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1801 bonustype
= SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl
, SA_BONUS
));
1802 saved_spill
= hdl
->sa_spill
;
1804 /* sync out registration table if necessary */
1805 if (sa
->sa_need_attr_registration
)
1806 sa_attr_register_sync(hdl
, tx
);
1808 error
= sa_attr_op(hdl
, bulk
, count
, SA_UPDATE
, tx
);
1809 if (error
== 0 && !IS_SA_BONUSTYPE(bonustype
) && sa
->sa_update_cb
)
1810 sa
->sa_update_cb(hdl
, tx
);
1813 * If saved_spill is NULL and current sa_spill is not NULL that
1814 * means we increased the refcount of the spill buffer through
1815 * sa_get_spill() or dmu_spill_hold_by_dnode(). Therefore we
1816 * must release the hold before calling dmu_tx_commit() to avoid
1817 * making a copy of this buffer in dbuf_sync_leaf() due to the
1818 * reference count now being greater than 1.
1820 if (!saved_spill
&& hdl
->sa_spill
) {
1821 if (hdl
->sa_spill_tab
) {
1822 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1823 hdl
->sa_spill_tab
= NULL
;
1826 dmu_buf_rele((dmu_buf_t
*)hdl
->sa_spill
, NULL
);
1827 hdl
->sa_spill
= NULL
;
1834 * update or add new attribute
1837 sa_update(sa_handle_t
*hdl
, sa_attr_type_t type
,
1838 void *buf
, uint32_t buflen
, dmu_tx_t
*tx
)
1841 sa_bulk_attr_t bulk
;
1843 bulk
.sa_attr
= type
;
1844 bulk
.sa_data_func
= NULL
;
1845 bulk
.sa_length
= buflen
;
1848 mutex_enter(&hdl
->sa_lock
);
1849 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1850 mutex_exit(&hdl
->sa_lock
);
1855 sa_update_from_cb(sa_handle_t
*hdl
, sa_attr_type_t attr
,
1856 uint32_t buflen
, sa_data_locator_t
*locator
, void *userdata
, dmu_tx_t
*tx
)
1859 sa_bulk_attr_t bulk
;
1861 bulk
.sa_attr
= attr
;
1862 bulk
.sa_data
= userdata
;
1863 bulk
.sa_data_func
= locator
;
1864 bulk
.sa_length
= buflen
;
1866 mutex_enter(&hdl
->sa_lock
);
1867 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1868 mutex_exit(&hdl
->sa_lock
);
1873 * Return size of an attribute
1877 sa_size(sa_handle_t
*hdl
, sa_attr_type_t attr
, int *size
)
1879 sa_bulk_attr_t bulk
;
1882 bulk
.sa_data
= NULL
;
1883 bulk
.sa_attr
= attr
;
1884 bulk
.sa_data_func
= NULL
;
1887 mutex_enter(&hdl
->sa_lock
);
1888 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) != 0) {
1889 mutex_exit(&hdl
->sa_lock
);
1892 *size
= bulk
.sa_size
;
1894 mutex_exit(&hdl
->sa_lock
);
1899 sa_bulk_lookup_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1902 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1903 return (sa_lookup_impl(hdl
, attrs
, count
));
1907 sa_bulk_lookup(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1912 mutex_enter(&hdl
->sa_lock
);
1913 error
= sa_bulk_lookup_locked(hdl
, attrs
, count
);
1914 mutex_exit(&hdl
->sa_lock
);
1919 sa_bulk_update(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
, dmu_tx_t
*tx
)
1924 mutex_enter(&hdl
->sa_lock
);
1925 error
= sa_bulk_update_impl(hdl
, attrs
, count
, tx
);
1926 mutex_exit(&hdl
->sa_lock
);
1931 sa_remove(sa_handle_t
*hdl
, sa_attr_type_t attr
, dmu_tx_t
*tx
)
1935 mutex_enter(&hdl
->sa_lock
);
1936 error
= sa_modify_attrs(hdl
, attr
, SA_REMOVE
, NULL
,
1938 mutex_exit(&hdl
->sa_lock
);
1943 sa_object_info(sa_handle_t
*hdl
, dmu_object_info_t
*doi
)
1945 dmu_object_info_from_db((dmu_buf_t
*)hdl
->sa_bonus
, doi
);
1949 sa_object_size(sa_handle_t
*hdl
, uint32_t *blksize
, u_longlong_t
*nblocks
)
1951 dmu_object_size_from_db((dmu_buf_t
*)hdl
->sa_bonus
,
1956 sa_update_user(sa_handle_t
*newhdl
, sa_handle_t
*oldhdl
)
1958 (void) dmu_buf_update_user((dmu_buf_t
*)newhdl
->sa_bonus
,
1959 oldhdl
, newhdl
, NULL
, sa_evict
);
1960 oldhdl
->sa_bonus
= NULL
;
1964 sa_set_userp(sa_handle_t
*hdl
, void *ptr
)
1966 hdl
->sa_userp
= ptr
;
1970 sa_get_db(sa_handle_t
*hdl
)
1972 return ((dmu_buf_t
*)hdl
->sa_bonus
);
1976 sa_get_userdata(sa_handle_t
*hdl
)
1978 return (hdl
->sa_userp
);
1982 sa_register_update_callback_locked(objset_t
*os
, sa_update_cb_t
*func
)
1984 ASSERT(MUTEX_HELD(&os
->os_sa
->sa_lock
));
1985 os
->os_sa
->sa_update_cb
= func
;
1989 sa_register_update_callback(objset_t
*os
, sa_update_cb_t
*func
)
1992 mutex_enter(&os
->os_sa
->sa_lock
);
1993 sa_register_update_callback_locked(os
, func
);
1994 mutex_exit(&os
->os_sa
->sa_lock
);
1998 sa_handle_object(sa_handle_t
*hdl
)
2000 return (hdl
->sa_bonus
->db_object
);
2004 sa_enabled(objset_t
*os
)
2006 return (os
->os_sa
== NULL
);
2010 sa_set_sa_object(objset_t
*os
, uint64_t sa_object
)
2012 sa_os_t
*sa
= os
->os_sa
;
2014 if (sa
->sa_master_obj
)
2017 sa
->sa_master_obj
= sa_object
;
2023 sa_hdrsize(void *arg
)
2025 sa_hdr_phys_t
*hdr
= arg
;
2027 return (SA_HDR_SIZE(hdr
));
2031 sa_handle_lock(sa_handle_t
*hdl
)
2034 mutex_enter(&hdl
->sa_lock
);
2038 sa_handle_unlock(sa_handle_t
*hdl
)
2041 mutex_exit(&hdl
->sa_lock
);
2045 EXPORT_SYMBOL(sa_handle_get
);
2046 EXPORT_SYMBOL(sa_handle_get_from_db
);
2047 EXPORT_SYMBOL(sa_handle_destroy
);
2048 EXPORT_SYMBOL(sa_buf_hold
);
2049 EXPORT_SYMBOL(sa_buf_rele
);
2050 EXPORT_SYMBOL(sa_spill_rele
);
2051 EXPORT_SYMBOL(sa_lookup
);
2052 EXPORT_SYMBOL(sa_update
);
2053 EXPORT_SYMBOL(sa_remove
);
2054 EXPORT_SYMBOL(sa_bulk_lookup
);
2055 EXPORT_SYMBOL(sa_bulk_lookup_locked
);
2056 EXPORT_SYMBOL(sa_bulk_update
);
2057 EXPORT_SYMBOL(sa_size
);
2058 EXPORT_SYMBOL(sa_update_from_cb
);
2059 EXPORT_SYMBOL(sa_object_info
);
2060 EXPORT_SYMBOL(sa_object_size
);
2061 EXPORT_SYMBOL(sa_update_user
);
2062 EXPORT_SYMBOL(sa_get_userdata
);
2063 EXPORT_SYMBOL(sa_set_userp
);
2064 EXPORT_SYMBOL(sa_get_db
);
2065 EXPORT_SYMBOL(sa_handle_object
);
2066 EXPORT_SYMBOL(sa_register_update_callback
);
2067 EXPORT_SYMBOL(sa_setup
);
2068 EXPORT_SYMBOL(sa_replace_all_by_template
);
2069 EXPORT_SYMBOL(sa_replace_all_by_template_locked
);
2070 EXPORT_SYMBOL(sa_enabled
);
2071 EXPORT_SYMBOL(sa_cache_init
);
2072 EXPORT_SYMBOL(sa_cache_fini
);
2073 EXPORT_SYMBOL(sa_spill_alloc
);
2074 EXPORT_SYMBOL(sa_spill_free
);
2075 EXPORT_SYMBOL(sa_set_sa_object
);
2076 EXPORT_SYMBOL(sa_hdrsize
);
2077 EXPORT_SYMBOL(sa_handle_lock
);
2078 EXPORT_SYMBOL(sa_handle_unlock
);
2079 EXPORT_SYMBOL(sa_lookup_uio
);
2080 #endif /* _KERNEL */