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:
113 * Since the SA attributes are not entirely self describing we can't do
114 * the normal byteswap processing. The special ZAP layout attribute and
115 * attribute registration attributes define the byteswap function and the
116 * size of the attributes, unless it is variable sized.
117 * The normal ZFS byteswapping infrastructure assumes you don't need
118 * to read any objects in order to do the necessary byteswapping. Whereas
119 * SA attributes can only be properly byteswapped if the dataset is opened
120 * and the layout/attribute ZAP attributes are available. Because of this
121 * the SA attributes will be byteswapped when they are first accessed by
122 * the SA code that will read the SA data.
125 typedef void (sa_iterfunc_t
)(void *hdr
, void *addr
, sa_attr_type_t
,
126 uint16_t length
, int length_idx
, boolean_t
, void *userp
);
128 static int sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
);
129 static void sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
);
130 static void *sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
,
132 static void sa_idx_tab_rele(objset_t
*os
, void *arg
);
133 static void sa_copy_data(sa_data_locator_t
*func
, void *start
, void *target
,
135 static int sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
136 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
137 uint16_t buflen
, dmu_tx_t
*tx
);
139 arc_byteswap_func_t sa_bswap_table
[] = {
140 byteswap_uint64_array
,
141 byteswap_uint32_array
,
142 byteswap_uint16_array
,
143 byteswap_uint8_array
,
147 #define SA_COPY_DATA(f, s, t, l) \
151 *(uint64_t *)t = *(uint64_t *)s; \
152 } else if (l == 16) { \
153 *(uint64_t *)t = *(uint64_t *)s; \
154 *(uint64_t *)((uintptr_t)t + 8) = \
155 *(uint64_t *)((uintptr_t)s + 8); \
160 sa_copy_data(f, s, t, l); \
164 * This table is fixed and cannot be changed. Its purpose is to
165 * allow the SA code to work with both old/new ZPL file systems.
166 * It contains the list of legacy attributes. These attributes aren't
167 * stored in the "attribute" registry zap objects, since older ZPL file systems
168 * won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
169 * use this static table.
171 sa_attr_reg_t sa_legacy_attrs
[] = {
172 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 0},
173 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 1},
174 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 2},
175 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY
, 3},
176 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY
, 4},
177 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY
, 5},
178 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY
, 6},
179 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY
, 7},
180 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY
, 8},
181 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY
, 9},
182 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY
, 10},
183 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY
, 11},
184 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY
, 12},
185 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY
, 13},
186 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY
, 14},
187 {"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.
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
;
206 static kmem_cache_t
*spill_cache
= NULL
;
210 sa_cache_constructor(void *buf
, void *unused
, int kmflag
)
212 sa_handle_t
*hdl
= buf
;
214 hdl
->sa_bonus_tab
= NULL
;
215 hdl
->sa_spill_tab
= NULL
;
217 hdl
->sa_userp
= NULL
;
218 hdl
->sa_bonus
= NULL
;
219 hdl
->sa_spill
= NULL
;
220 mutex_init(&hdl
->sa_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
226 sa_cache_destructor(void *buf
, void *unused
)
228 sa_handle_t
*hdl
= buf
;
229 mutex_destroy(&hdl
->sa_lock
);
235 sa_cache
= kmem_cache_create("sa_cache",
236 sizeof (sa_handle_t
), 0, sa_cache_constructor
,
237 sa_cache_destructor
, NULL
, NULL
, NULL
, 0);
238 spill_cache
= kmem_cache_create("spill_cache",
239 SPA_MAXBLOCKSIZE
, 0, NULL
, NULL
, NULL
, NULL
, NULL
, 0);
246 kmem_cache_destroy(sa_cache
);
249 kmem_cache_destroy(spill_cache
);
253 sa_spill_alloc(int flags
)
255 return kmem_cache_alloc(spill_cache
, flags
);
259 sa_spill_free(void *obj
)
261 kmem_cache_free(spill_cache
, obj
);
265 layout_num_compare(const void *arg1
, const void *arg2
)
267 const sa_lot_t
*node1
= arg1
;
268 const sa_lot_t
*node2
= arg2
;
270 if (node1
->lot_num
> node2
->lot_num
)
272 else if (node1
->lot_num
< node2
->lot_num
)
278 layout_hash_compare(const void *arg1
, const void *arg2
)
280 const sa_lot_t
*node1
= arg1
;
281 const sa_lot_t
*node2
= arg2
;
283 if (node1
->lot_hash
> node2
->lot_hash
)
285 if (node1
->lot_hash
< node2
->lot_hash
)
287 if (node1
->lot_instance
> node2
->lot_instance
)
289 if (node1
->lot_instance
< node2
->lot_instance
)
295 sa_layout_equal(sa_lot_t
*tbf
, sa_attr_type_t
*attrs
, int count
)
299 if (count
!= tbf
->lot_attr_count
)
302 for (i
= 0; i
!= count
; i
++) {
303 if (attrs
[i
] != tbf
->lot_attrs
[i
])
309 #define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
312 sa_layout_info_hash(sa_attr_type_t
*attrs
, int attr_count
)
315 uint64_t crc
= -1ULL;
317 for (i
= 0; i
!= attr_count
; i
++)
318 crc
^= SA_ATTR_HASH(attrs
[i
]);
324 sa_get_spill(sa_handle_t
*hdl
)
327 if (hdl
->sa_spill
== NULL
) {
328 if ((rc
= dmu_spill_hold_existing(hdl
->sa_bonus
, NULL
,
329 &hdl
->sa_spill
)) == 0)
330 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
339 * Main attribute lookup/update function
340 * returns 0 for success or non zero for failures
342 * Operates on bulk array, first failure will abort further processing
345 sa_attr_op(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
346 sa_data_op_t data_op
, dmu_tx_t
*tx
)
348 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
351 sa_buf_type_t buftypes
;
356 for (i
= 0; i
!= count
; i
++) {
357 ASSERT(bulk
[i
].sa_attr
<= hdl
->sa_os
->os_sa
->sa_num_attrs
);
359 bulk
[i
].sa_addr
= NULL
;
360 /* First check the bonus buffer */
362 if (hdl
->sa_bonus_tab
&& TOC_ATTR_PRESENT(
363 hdl
->sa_bonus_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
364 SA_ATTR_INFO(sa
, hdl
->sa_bonus_tab
,
365 SA_GET_HDR(hdl
, SA_BONUS
),
366 bulk
[i
].sa_attr
, bulk
[i
], SA_BONUS
, hdl
);
367 if (tx
&& !(buftypes
& SA_BONUS
)) {
368 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
369 buftypes
|= SA_BONUS
;
372 if (bulk
[i
].sa_addr
== NULL
&&
373 ((error
= sa_get_spill(hdl
)) == 0)) {
374 if (TOC_ATTR_PRESENT(
375 hdl
->sa_spill_tab
->sa_idx_tab
[bulk
[i
].sa_attr
])) {
376 SA_ATTR_INFO(sa
, hdl
->sa_spill_tab
,
377 SA_GET_HDR(hdl
, SA_SPILL
),
378 bulk
[i
].sa_attr
, bulk
[i
], SA_SPILL
, hdl
);
379 if (tx
&& !(buftypes
& SA_SPILL
) &&
380 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
381 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
382 buftypes
|= SA_SPILL
;
386 if (error
&& error
!= ENOENT
) {
387 return ((error
== ECKSUM
) ? EIO
: error
);
392 if (bulk
[i
].sa_addr
== NULL
)
393 return (SET_ERROR(ENOENT
));
394 if (bulk
[i
].sa_data
) {
395 SA_COPY_DATA(bulk
[i
].sa_data_func
,
396 bulk
[i
].sa_addr
, bulk
[i
].sa_data
,
402 /* existing rewrite of attr */
403 if (bulk
[i
].sa_addr
&&
404 bulk
[i
].sa_size
== bulk
[i
].sa_length
) {
405 SA_COPY_DATA(bulk
[i
].sa_data_func
,
406 bulk
[i
].sa_data
, bulk
[i
].sa_addr
,
409 } else if (bulk
[i
].sa_addr
) { /* attr size change */
410 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
411 SA_REPLACE
, bulk
[i
].sa_data_func
,
412 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
413 } else { /* adding new attribute */
414 error
= sa_modify_attrs(hdl
, bulk
[i
].sa_attr
,
415 SA_ADD
, bulk
[i
].sa_data_func
,
416 bulk
[i
].sa_data
, bulk
[i
].sa_length
, tx
);
429 sa_add_layout_entry(objset_t
*os
, sa_attr_type_t
*attrs
, int attr_count
,
430 uint64_t lot_num
, uint64_t hash
, boolean_t zapadd
, dmu_tx_t
*tx
)
432 sa_os_t
*sa
= os
->os_sa
;
433 sa_lot_t
*tb
, *findtb
;
437 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
438 tb
= kmem_zalloc(sizeof (sa_lot_t
), KM_PUSHPAGE
);
439 tb
->lot_attr_count
= attr_count
;
440 tb
->lot_attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
442 bcopy(attrs
, tb
->lot_attrs
, sizeof (sa_attr_type_t
) * attr_count
);
443 tb
->lot_num
= lot_num
;
445 tb
->lot_instance
= 0;
450 if (sa
->sa_layout_attr_obj
== 0) {
451 sa
->sa_layout_attr_obj
= zap_create_link(os
,
452 DMU_OT_SA_ATTR_LAYOUTS
,
453 sa
->sa_master_obj
, SA_LAYOUTS
, tx
);
456 (void) snprintf(attr_name
, sizeof (attr_name
),
458 VERIFY(0 == zap_update(os
, os
->os_sa
->sa_layout_attr_obj
,
459 attr_name
, 2, attr_count
, attrs
, tx
));
462 list_create(&tb
->lot_idx_tab
, sizeof (sa_idx_tab_t
),
463 offsetof(sa_idx_tab_t
, sa_next
));
465 for (i
= 0; i
!= attr_count
; i
++) {
466 if (sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
== 0)
470 avl_add(&sa
->sa_layout_num_tree
, tb
);
472 /* verify we don't have a hash collision */
473 if ((findtb
= avl_find(&sa
->sa_layout_hash_tree
, tb
, &loc
)) != NULL
) {
474 for (; findtb
&& findtb
->lot_hash
== hash
;
475 findtb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, findtb
)) {
476 if (findtb
->lot_instance
!= tb
->lot_instance
)
481 avl_add(&sa
->sa_layout_hash_tree
, tb
);
486 sa_find_layout(objset_t
*os
, uint64_t hash
, sa_attr_type_t
*attrs
,
487 int count
, dmu_tx_t
*tx
, sa_lot_t
**lot
)
489 sa_lot_t
*tb
, tbsearch
;
491 sa_os_t
*sa
= os
->os_sa
;
492 boolean_t found
= B_FALSE
;
494 mutex_enter(&sa
->sa_lock
);
495 tbsearch
.lot_hash
= hash
;
496 tbsearch
.lot_instance
= 0;
497 tb
= avl_find(&sa
->sa_layout_hash_tree
, &tbsearch
, &loc
);
499 for (; tb
&& tb
->lot_hash
== hash
;
500 tb
= AVL_NEXT(&sa
->sa_layout_hash_tree
, tb
)) {
501 if (sa_layout_equal(tb
, attrs
, count
) == 0) {
508 tb
= sa_add_layout_entry(os
, attrs
, count
,
509 avl_numnodes(&sa
->sa_layout_num_tree
), hash
, B_TRUE
, tx
);
511 mutex_exit(&sa
->sa_lock
);
516 sa_resize_spill(sa_handle_t
*hdl
, uint32_t size
, dmu_tx_t
*tx
)
522 blocksize
= SPA_MINBLOCKSIZE
;
523 } else if (size
> SPA_MAXBLOCKSIZE
) {
525 return (SET_ERROR(EFBIG
));
527 blocksize
= P2ROUNDUP_TYPED(size
, SPA_MINBLOCKSIZE
, uint32_t);
530 error
= dbuf_spill_set_blksz(hdl
->sa_spill
, blocksize
, tx
);
536 sa_copy_data(sa_data_locator_t
*func
, void *datastart
, void *target
, int buflen
)
539 bcopy(datastart
, target
, buflen
);
544 void *saptr
= target
;
549 while (bytes
< buflen
) {
550 func(&dataptr
, &length
, buflen
, start
, datastart
);
551 bcopy(dataptr
, saptr
, length
);
552 saptr
= (void *)((caddr_t
)saptr
+ length
);
560 * Determine several different sizes
561 * first the sa header size
562 * the number of bytes to be stored
563 * if spill would occur the index in the attribute array is returned
565 * the boolean will_spill will be set when spilling is necessary. It
566 * is only set when the buftype is SA_BONUS
569 sa_find_sizes(sa_os_t
*sa
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
570 dmu_buf_t
*db
, sa_buf_type_t buftype
, int *index
, int *total
,
571 boolean_t
*will_spill
)
578 boolean_t done
= B_FALSE
;
580 if (buftype
== SA_BONUS
&& sa
->sa_force_spill
) {
583 *will_spill
= B_TRUE
;
590 if (buftype
== SA_BONUS
)
591 *will_spill
= B_FALSE
;
593 hdrsize
= (SA_BONUSTYPE_FROM_DB(db
) == DMU_OT_ZNODE
) ? 0 :
594 sizeof (sa_hdr_phys_t
);
596 full_space
= (buftype
== SA_BONUS
) ? DN_MAX_BONUSLEN
: db
->db_size
;
597 ASSERT(IS_P2ALIGNED(full_space
, 8));
599 for (i
= 0; i
!= attr_count
; i
++) {
602 *total
= P2ROUNDUP(*total
, 8);
603 *total
+= attr_desc
[i
].sa_length
;
607 is_var_sz
= (SA_REGISTERED_LEN(sa
, attr_desc
[i
].sa_attr
) == 0);
612 if (is_var_sz
&& var_size
> 1) {
613 if (P2ROUNDUP(hdrsize
+ sizeof (uint16_t), 8) +
614 *total
< full_space
) {
616 * Account for header space used by array of
617 * optional sizes of variable-length attributes.
618 * Record the index in case this increase needs
619 * to be reversed due to spill-over.
621 hdrsize
+= sizeof (uint16_t);
626 if (buftype
== SA_BONUS
)
627 *will_spill
= B_TRUE
;
633 * find index of where spill *could* occur.
634 * Then continue to count of remainder attribute
635 * space. The sum is used later for sizing bonus
638 if (buftype
== SA_BONUS
&& *index
== -1 &&
639 (*total
+ P2ROUNDUP(hdrsize
, 8)) >
640 (full_space
- sizeof (blkptr_t
))) {
646 if ((*total
+ P2ROUNDUP(hdrsize
, 8)) > full_space
&&
648 *will_spill
= B_TRUE
;
652 * j holds the index of the last variable-sized attribute for
653 * which hdrsize was increased. Reverse the increase if that
654 * attribute will be relocated to the spill block.
656 if (*will_spill
&& j
== *index
)
657 hdrsize
-= sizeof (uint16_t);
659 hdrsize
= P2ROUNDUP(hdrsize
, 8);
663 #define BUF_SPACE_NEEDED(total, header) (total + header)
666 * Find layout that corresponds to ordering of attributes
667 * If not found a new layout number is created and added to
668 * persistent layout tables.
671 sa_build_layouts(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
, int attr_count
,
674 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
676 sa_buf_type_t buftype
;
677 sa_hdr_phys_t
*sahdr
;
680 sa_attr_type_t
*attrs
, *attrs_start
;
683 int spillhdrsize
= 0;
685 dmu_object_type_t bonustype
;
691 dmu_buf_will_dirty(hdl
->sa_bonus
, tx
);
692 bonustype
= SA_BONUSTYPE_FROM_DB(hdl
->sa_bonus
);
694 /* first determine bonus header size and sum of all attributes */
695 hdrsize
= sa_find_sizes(sa
, attr_desc
, attr_count
, hdl
->sa_bonus
,
696 SA_BONUS
, &i
, &used
, &spilling
);
698 if (used
> SPA_MAXBLOCKSIZE
)
699 return (SET_ERROR(EFBIG
));
701 VERIFY(0 == dmu_set_bonus(hdl
->sa_bonus
, spilling
?
702 MIN(DN_MAX_BONUSLEN
- sizeof (blkptr_t
), used
+ hdrsize
) :
703 used
+ hdrsize
, tx
));
705 ASSERT((bonustype
== DMU_OT_ZNODE
&& spilling
== 0) ||
706 bonustype
== DMU_OT_SA
);
708 /* setup and size spill buffer when needed */
712 if (hdl
->sa_spill
== NULL
) {
713 VERIFY(dmu_spill_hold_by_bonus(hdl
->sa_bonus
, NULL
,
714 &hdl
->sa_spill
) == 0);
716 dmu_buf_will_dirty(hdl
->sa_spill
, tx
);
718 spillhdrsize
= sa_find_sizes(sa
, &attr_desc
[i
],
719 attr_count
- i
, hdl
->sa_spill
, SA_SPILL
, &i
,
720 &spill_used
, &dummy
);
722 if (spill_used
> SPA_MAXBLOCKSIZE
)
723 return (SET_ERROR(EFBIG
));
725 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
726 if (BUF_SPACE_NEEDED(spill_used
, spillhdrsize
) >
727 hdl
->sa_spill
->db_size
)
728 VERIFY(0 == sa_resize_spill(hdl
,
729 BUF_SPACE_NEEDED(spill_used
, spillhdrsize
), tx
));
732 /* setup starting pointers to lay down data */
733 data_start
= (void *)((uintptr_t)hdl
->sa_bonus
->db_data
+ hdrsize
);
734 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_bonus
->db_data
;
738 buf_space
= (sa
->sa_force_spill
) ?
739 0 : SA_BLKPTR_SPACE
- hdrsize
;
741 buf_space
= hdl
->sa_bonus
->db_size
- hdrsize
;
743 attrs_start
= attrs
= kmem_alloc(sizeof (sa_attr_type_t
) * attr_count
,
747 for (i
= 0, len_idx
= 0, hash
= -1ULL; i
!= attr_count
; i
++) {
750 ASSERT(IS_P2ALIGNED(data_start
, 8));
751 ASSERT(IS_P2ALIGNED(buf_space
, 8));
752 attrs
[i
] = attr_desc
[i
].sa_attr
;
753 length
= SA_REGISTERED_LEN(sa
, attrs
[i
]);
755 length
= attr_desc
[i
].sa_length
;
757 if (buf_space
< length
) { /* switch to spill buffer */
759 VERIFY(bonustype
== DMU_OT_SA
);
760 if (buftype
== SA_BONUS
&& !sa
->sa_force_spill
) {
761 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
,
762 lot_count
, tx
, &lot
);
763 SA_SET_HDR(sahdr
, lot
->lot_num
, hdrsize
);
770 sahdr
= (sa_hdr_phys_t
*)hdl
->sa_spill
->db_data
;
771 sahdr
->sa_magic
= SA_MAGIC
;
772 data_start
= (void *)((uintptr_t)sahdr
+
774 attrs_start
= &attrs
[i
];
775 buf_space
= hdl
->sa_spill
->db_size
- spillhdrsize
;
778 hash
^= SA_ATTR_HASH(attrs
[i
]);
779 attr_desc
[i
].sa_addr
= data_start
;
780 attr_desc
[i
].sa_size
= length
;
781 SA_COPY_DATA(attr_desc
[i
].sa_data_func
, attr_desc
[i
].sa_data
,
783 if (sa
->sa_attr_table
[attrs
[i
]].sa_length
== 0) {
784 sahdr
->sa_lengths
[len_idx
++] = length
;
786 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
788 buf_space
-= P2ROUNDUP(length
, 8);
792 sa_find_layout(hdl
->sa_os
, hash
, attrs_start
, lot_count
, tx
, &lot
);
795 * Verify that old znodes always have layout number 0.
796 * Must be DMU_OT_SA for arbitrary layouts
798 VERIFY((bonustype
== DMU_OT_ZNODE
&& lot
->lot_num
== 0) ||
799 (bonustype
== DMU_OT_SA
&& lot
->lot_num
> 1));
801 if (bonustype
== DMU_OT_SA
) {
802 SA_SET_HDR(sahdr
, lot
->lot_num
,
803 buftype
== SA_BONUS
? hdrsize
: spillhdrsize
);
806 kmem_free(attrs
, sizeof (sa_attr_type_t
) * attr_count
);
807 if (hdl
->sa_bonus_tab
) {
808 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
809 hdl
->sa_bonus_tab
= NULL
;
811 if (!sa
->sa_force_spill
)
812 VERIFY(0 == sa_build_index(hdl
, SA_BONUS
));
814 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
817 * remove spill block that is no longer needed.
819 dmu_buf_rele(hdl
->sa_spill
, NULL
);
820 hdl
->sa_spill
= NULL
;
821 hdl
->sa_spill_tab
= NULL
;
822 VERIFY(0 == dmu_rm_spill(hdl
->sa_os
,
823 sa_handle_object(hdl
), tx
));
825 VERIFY(0 == sa_build_index(hdl
, SA_SPILL
));
833 sa_free_attr_table(sa_os_t
*sa
)
837 if (sa
->sa_attr_table
== NULL
)
840 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
841 if (sa
->sa_attr_table
[i
].sa_name
)
842 kmem_free(sa
->sa_attr_table
[i
].sa_name
,
843 strlen(sa
->sa_attr_table
[i
].sa_name
) + 1);
846 kmem_free(sa
->sa_attr_table
,
847 sizeof (sa_attr_table_t
) * sa
->sa_num_attrs
);
849 sa
->sa_attr_table
= NULL
;
853 sa_attr_table_setup(objset_t
*os
, sa_attr_reg_t
*reg_attrs
, int count
)
855 sa_os_t
*sa
= os
->os_sa
;
856 uint64_t sa_attr_count
= 0;
857 uint64_t sa_reg_count
= 0;
863 int registered_count
= 0;
865 dmu_objset_type_t ostype
= dmu_objset_type(os
);
868 kmem_zalloc(count
* sizeof (sa_attr_type_t
), KM_PUSHPAGE
);
869 sa
->sa_user_table_sz
= count
* sizeof (sa_attr_type_t
);
871 if (sa
->sa_reg_attr_obj
!= 0) {
872 error
= zap_count(os
, sa
->sa_reg_attr_obj
,
876 * Make sure we retrieved a count and that it isn't zero
878 if (error
|| (error
== 0 && sa_attr_count
== 0)) {
880 error
= SET_ERROR(EINVAL
);
883 sa_reg_count
= sa_attr_count
;
886 if (ostype
== DMU_OST_ZFS
&& sa_attr_count
== 0)
887 sa_attr_count
+= sa_legacy_attr_count
;
889 /* Allocate attribute numbers for attributes that aren't registered */
890 for (i
= 0; i
!= count
; i
++) {
891 boolean_t found
= B_FALSE
;
894 if (ostype
== DMU_OST_ZFS
) {
895 for (j
= 0; j
!= sa_legacy_attr_count
; j
++) {
896 if (strcmp(reg_attrs
[i
].sa_name
,
897 sa_legacy_attrs
[j
].sa_name
) == 0) {
898 sa
->sa_user_table
[i
] =
899 sa_legacy_attrs
[j
].sa_attr
;
907 if (sa
->sa_reg_attr_obj
)
908 error
= zap_lookup(os
, sa
->sa_reg_attr_obj
,
909 reg_attrs
[i
].sa_name
, 8, 1, &attr_value
);
911 error
= SET_ERROR(ENOENT
);
914 sa
->sa_user_table
[i
] = (sa_attr_type_t
)sa_attr_count
;
918 sa
->sa_user_table
[i
] = ATTR_NUM(attr_value
);
925 sa
->sa_num_attrs
= sa_attr_count
;
926 tb
= sa
->sa_attr_table
=
927 kmem_zalloc(sizeof (sa_attr_table_t
) * sa_attr_count
, KM_PUSHPAGE
);
930 * Attribute table is constructed from requested attribute list,
931 * previously foreign registered attributes, and also the legacy
932 * ZPL set of attributes.
935 if (sa
->sa_reg_attr_obj
) {
936 for (zap_cursor_init(&zc
, os
, sa
->sa_reg_attr_obj
);
937 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
938 zap_cursor_advance(&zc
)) {
940 value
= za
.za_first_integer
;
943 tb
[ATTR_NUM(value
)].sa_attr
= ATTR_NUM(value
);
944 tb
[ATTR_NUM(value
)].sa_length
= ATTR_LENGTH(value
);
945 tb
[ATTR_NUM(value
)].sa_byteswap
= ATTR_BSWAP(value
);
946 tb
[ATTR_NUM(value
)].sa_registered
= B_TRUE
;
948 if (tb
[ATTR_NUM(value
)].sa_name
) {
951 tb
[ATTR_NUM(value
)].sa_name
=
952 kmem_zalloc(strlen(za
.za_name
) +1, KM_PUSHPAGE
);
953 (void) strlcpy(tb
[ATTR_NUM(value
)].sa_name
, za
.za_name
,
954 strlen(za
.za_name
) +1);
956 zap_cursor_fini(&zc
);
958 * Make sure we processed the correct number of registered
961 if (registered_count
!= sa_reg_count
) {
968 if (ostype
== DMU_OST_ZFS
) {
969 for (i
= 0; i
!= sa_legacy_attr_count
; i
++) {
972 tb
[i
].sa_attr
= sa_legacy_attrs
[i
].sa_attr
;
973 tb
[i
].sa_length
= sa_legacy_attrs
[i
].sa_length
;
974 tb
[i
].sa_byteswap
= sa_legacy_attrs
[i
].sa_byteswap
;
975 tb
[i
].sa_registered
= B_FALSE
;
977 kmem_zalloc(strlen(sa_legacy_attrs
[i
].sa_name
) +1,
979 (void) strlcpy(tb
[i
].sa_name
,
980 sa_legacy_attrs
[i
].sa_name
,
981 strlen(sa_legacy_attrs
[i
].sa_name
) + 1);
985 for (i
= 0; i
!= count
; i
++) {
986 sa_attr_type_t attr_id
;
988 attr_id
= sa
->sa_user_table
[i
];
989 if (tb
[attr_id
].sa_name
)
992 tb
[attr_id
].sa_length
= reg_attrs
[i
].sa_length
;
993 tb
[attr_id
].sa_byteswap
= reg_attrs
[i
].sa_byteswap
;
994 tb
[attr_id
].sa_attr
= attr_id
;
995 tb
[attr_id
].sa_name
=
996 kmem_zalloc(strlen(reg_attrs
[i
].sa_name
) + 1, KM_PUSHPAGE
);
997 (void) strlcpy(tb
[attr_id
].sa_name
, reg_attrs
[i
].sa_name
,
998 strlen(reg_attrs
[i
].sa_name
) + 1);
1001 sa
->sa_need_attr_registration
=
1002 (sa_attr_count
!= registered_count
);
1006 kmem_free(sa
->sa_user_table
, count
* sizeof (sa_attr_type_t
));
1007 sa
->sa_user_table
= NULL
;
1008 sa_free_attr_table(sa
);
1009 return ((error
!= 0) ? error
: EINVAL
);
1013 sa_setup(objset_t
*os
, uint64_t sa_obj
, sa_attr_reg_t
*reg_attrs
, int count
,
1014 sa_attr_type_t
**user_table
)
1019 dmu_objset_type_t ostype
= dmu_objset_type(os
);
1023 mutex_enter(&os
->os_user_ptr_lock
);
1025 mutex_enter(&os
->os_sa
->sa_lock
);
1026 mutex_exit(&os
->os_user_ptr_lock
);
1027 tb
= os
->os_sa
->sa_user_table
;
1028 mutex_exit(&os
->os_sa
->sa_lock
);
1033 sa
= kmem_zalloc(sizeof (sa_os_t
), KM_PUSHPAGE
);
1034 mutex_init(&sa
->sa_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1035 sa
->sa_master_obj
= sa_obj
;
1038 mutex_enter(&sa
->sa_lock
);
1039 mutex_exit(&os
->os_user_ptr_lock
);
1040 avl_create(&sa
->sa_layout_num_tree
, layout_num_compare
,
1041 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_num_node
));
1042 avl_create(&sa
->sa_layout_hash_tree
, layout_hash_compare
,
1043 sizeof (sa_lot_t
), offsetof(sa_lot_t
, lot_hash_node
));
1046 error
= zap_lookup(os
, sa_obj
, SA_LAYOUTS
,
1047 8, 1, &sa
->sa_layout_attr_obj
);
1048 if (error
!= 0 && error
!= ENOENT
)
1050 error
= zap_lookup(os
, sa_obj
, SA_REGISTRY
,
1051 8, 1, &sa
->sa_reg_attr_obj
);
1052 if (error
!= 0 && error
!= ENOENT
)
1056 if ((error
= sa_attr_table_setup(os
, reg_attrs
, count
)) != 0)
1059 if (sa
->sa_layout_attr_obj
!= 0) {
1060 uint64_t layout_count
;
1062 error
= zap_count(os
, sa
->sa_layout_attr_obj
,
1066 * Layout number count should be > 0
1068 if (error
|| (error
== 0 && layout_count
== 0)) {
1070 error
= SET_ERROR(EINVAL
);
1074 for (zap_cursor_init(&zc
, os
, sa
->sa_layout_attr_obj
);
1075 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
1076 zap_cursor_advance(&zc
)) {
1077 sa_attr_type_t
*lot_attrs
;
1080 lot_attrs
= kmem_zalloc(sizeof (sa_attr_type_t
) *
1081 za
.za_num_integers
, KM_PUSHPAGE
);
1083 if ((error
= (zap_lookup(os
, sa
->sa_layout_attr_obj
,
1084 za
.za_name
, 2, za
.za_num_integers
,
1085 lot_attrs
))) != 0) {
1086 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1087 za
.za_num_integers
);
1090 VERIFY(ddi_strtoull(za
.za_name
, NULL
, 10,
1091 (unsigned long long *)&lot_num
) == 0);
1093 (void) sa_add_layout_entry(os
, lot_attrs
,
1094 za
.za_num_integers
, lot_num
,
1095 sa_layout_info_hash(lot_attrs
,
1096 za
.za_num_integers
), B_FALSE
, NULL
);
1097 kmem_free(lot_attrs
, sizeof (sa_attr_type_t
) *
1098 za
.za_num_integers
);
1100 zap_cursor_fini(&zc
);
1103 * Make sure layout count matches number of entries added
1106 if (avl_numnodes(&sa
->sa_layout_num_tree
) != layout_count
) {
1112 /* Add special layout number for old ZNODES */
1113 if (ostype
== DMU_OST_ZFS
) {
1114 (void) sa_add_layout_entry(os
, sa_legacy_zpl_layout
,
1115 sa_legacy_attr_count
, 0,
1116 sa_layout_info_hash(sa_legacy_zpl_layout
,
1117 sa_legacy_attr_count
), B_FALSE
, NULL
);
1119 (void) sa_add_layout_entry(os
, sa_dummy_zpl_layout
, 0, 1,
1122 *user_table
= os
->os_sa
->sa_user_table
;
1123 mutex_exit(&sa
->sa_lock
);
1127 sa_free_attr_table(sa
);
1128 if (sa
->sa_user_table
)
1129 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1130 mutex_exit(&sa
->sa_lock
);
1131 kmem_free(sa
, sizeof (sa_os_t
));
1132 return ((error
== ECKSUM
) ? EIO
: error
);
1136 sa_tear_down(objset_t
*os
)
1138 sa_os_t
*sa
= os
->os_sa
;
1142 kmem_free(sa
->sa_user_table
, sa
->sa_user_table_sz
);
1144 /* Free up attr table */
1146 sa_free_attr_table(sa
);
1149 while ((layout
= avl_destroy_nodes(&sa
->sa_layout_hash_tree
, &cookie
))){
1151 while ((tab
= list_head(&layout
->lot_idx_tab
))) {
1152 ASSERT(refcount_count(&tab
->sa_refcount
));
1153 sa_idx_tab_rele(os
, tab
);
1158 while ((layout
= avl_destroy_nodes(&sa
->sa_layout_num_tree
, &cookie
))){
1159 kmem_free(layout
->lot_attrs
,
1160 sizeof (sa_attr_type_t
) * layout
->lot_attr_count
);
1161 kmem_free(layout
, sizeof (sa_lot_t
));
1164 avl_destroy(&sa
->sa_layout_hash_tree
);
1165 avl_destroy(&sa
->sa_layout_num_tree
);
1167 kmem_free(sa
, sizeof (sa_os_t
));
1172 sa_build_idx_tab(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1173 uint16_t length
, int length_idx
, boolean_t var_length
, void *userp
)
1175 sa_idx_tab_t
*idx_tab
= userp
;
1178 ASSERT(idx_tab
->sa_variable_lengths
);
1179 idx_tab
->sa_variable_lengths
[length_idx
] = length
;
1181 TOC_ATTR_ENCODE(idx_tab
->sa_idx_tab
[attr
], length_idx
,
1182 (uint32_t)((uintptr_t)attr_addr
- (uintptr_t)hdr
));
1186 sa_attr_iter(objset_t
*os
, sa_hdr_phys_t
*hdr
, dmu_object_type_t type
,
1187 sa_iterfunc_t func
, sa_lot_t
*tab
, void *userp
)
1193 sa_os_t
*sa
= os
->os_sa
;
1195 uint16_t *length_start
= NULL
;
1196 uint8_t length_idx
= 0;
1199 search
.lot_num
= SA_LAYOUT_NUM(hdr
, type
);
1200 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1204 if (IS_SA_BONUSTYPE(type
)) {
1205 data_start
= (void *)P2ROUNDUP(((uintptr_t)hdr
+
1206 offsetof(sa_hdr_phys_t
, sa_lengths
) +
1207 (sizeof (uint16_t) * tb
->lot_var_sizes
)), 8);
1208 length_start
= hdr
->sa_lengths
;
1213 for (i
= 0; i
!= tb
->lot_attr_count
; i
++) {
1214 int attr_length
, reg_length
;
1217 reg_length
= sa
->sa_attr_table
[tb
->lot_attrs
[i
]].sa_length
;
1219 attr_length
= reg_length
;
1222 attr_length
= length_start
[length_idx
];
1223 idx_len
= length_idx
++;
1226 func(hdr
, data_start
, tb
->lot_attrs
[i
], attr_length
,
1227 idx_len
, reg_length
== 0 ? B_TRUE
: B_FALSE
, userp
);
1229 data_start
= (void *)P2ROUNDUP(((uintptr_t)data_start
+
1236 sa_byteswap_cb(void *hdr
, void *attr_addr
, sa_attr_type_t attr
,
1237 uint16_t length
, int length_idx
, boolean_t variable_length
, void *userp
)
1239 sa_handle_t
*hdl
= userp
;
1240 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1242 sa_bswap_table
[sa
->sa_attr_table
[attr
].sa_byteswap
](attr_addr
, length
);
1246 sa_byteswap(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1248 sa_hdr_phys_t
*sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1250 int num_lengths
= 1;
1252 ASSERTV(sa_os_t
*sa
= hdl
->sa_os
->os_sa
);
1254 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1255 if (sa_hdr_phys
->sa_magic
== SA_MAGIC
)
1258 db
= SA_GET_DB(hdl
, buftype
);
1260 if (buftype
== SA_SPILL
) {
1261 arc_release(db
->db_buf
, NULL
);
1262 arc_buf_thaw(db
->db_buf
);
1265 sa_hdr_phys
->sa_magic
= BSWAP_32(sa_hdr_phys
->sa_magic
);
1266 sa_hdr_phys
->sa_layout_info
= BSWAP_16(sa_hdr_phys
->sa_layout_info
);
1269 * Determine number of variable lenghts in header
1270 * The standard 8 byte header has one for free and a
1271 * 16 byte header would have 4 + 1;
1273 if (SA_HDR_SIZE(sa_hdr_phys
) > 8)
1274 num_lengths
+= (SA_HDR_SIZE(sa_hdr_phys
) - 8) >> 1;
1275 for (i
= 0; i
!= num_lengths
; i
++)
1276 sa_hdr_phys
->sa_lengths
[i
] =
1277 BSWAP_16(sa_hdr_phys
->sa_lengths
[i
]);
1279 sa_attr_iter(hdl
->sa_os
, sa_hdr_phys
, DMU_OT_SA
,
1280 sa_byteswap_cb
, NULL
, hdl
);
1282 if (buftype
== SA_SPILL
)
1283 arc_buf_freeze(((dmu_buf_impl_t
*)hdl
->sa_spill
)->db_buf
);
1287 sa_build_index(sa_handle_t
*hdl
, sa_buf_type_t buftype
)
1289 sa_hdr_phys_t
*sa_hdr_phys
;
1290 dmu_buf_impl_t
*db
= SA_GET_DB(hdl
, buftype
);
1291 dmu_object_type_t bonustype
= SA_BONUSTYPE_FROM_DB(db
);
1292 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1293 sa_idx_tab_t
*idx_tab
;
1295 sa_hdr_phys
= SA_GET_HDR(hdl
, buftype
);
1297 mutex_enter(&sa
->sa_lock
);
1299 /* Do we need to byteswap? */
1301 /* only check if not old znode */
1302 if (IS_SA_BONUSTYPE(bonustype
) && sa_hdr_phys
->sa_magic
!= SA_MAGIC
&&
1303 sa_hdr_phys
->sa_magic
!= 0) {
1304 VERIFY(BSWAP_32(sa_hdr_phys
->sa_magic
) == SA_MAGIC
);
1305 sa_byteswap(hdl
, buftype
);
1308 idx_tab
= sa_find_idx_tab(hdl
->sa_os
, bonustype
, sa_hdr_phys
);
1310 if (buftype
== SA_BONUS
)
1311 hdl
->sa_bonus_tab
= idx_tab
;
1313 hdl
->sa_spill_tab
= idx_tab
;
1315 mutex_exit(&sa
->sa_lock
);
1321 sa_evict(dmu_buf_t
*db
, void *sap
)
1323 panic("evicting sa dbuf %p\n", (void *)db
);
1327 sa_idx_tab_rele(objset_t
*os
, void *arg
)
1329 sa_os_t
*sa
= os
->os_sa
;
1330 sa_idx_tab_t
*idx_tab
= arg
;
1332 if (idx_tab
== NULL
)
1335 mutex_enter(&sa
->sa_lock
);
1336 if (refcount_remove(&idx_tab
->sa_refcount
, NULL
) == 0) {
1337 list_remove(&idx_tab
->sa_layout
->lot_idx_tab
, idx_tab
);
1338 if (idx_tab
->sa_variable_lengths
)
1339 kmem_free(idx_tab
->sa_variable_lengths
,
1341 idx_tab
->sa_layout
->lot_var_sizes
);
1342 refcount_destroy(&idx_tab
->sa_refcount
);
1343 kmem_free(idx_tab
->sa_idx_tab
,
1344 sizeof (uint32_t) * sa
->sa_num_attrs
);
1345 kmem_free(idx_tab
, sizeof (sa_idx_tab_t
));
1347 mutex_exit(&sa
->sa_lock
);
1351 sa_idx_tab_hold(objset_t
*os
, sa_idx_tab_t
*idx_tab
)
1353 ASSERTV(sa_os_t
*sa
= os
->os_sa
);
1355 ASSERT(MUTEX_HELD(&sa
->sa_lock
));
1356 (void) refcount_add(&idx_tab
->sa_refcount
, NULL
);
1360 sa_spill_rele(sa_handle_t
*hdl
)
1362 mutex_enter(&hdl
->sa_lock
);
1363 if (hdl
->sa_spill
) {
1364 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1365 dmu_buf_rele(hdl
->sa_spill
, NULL
);
1366 hdl
->sa_spill
= NULL
;
1367 hdl
->sa_spill_tab
= NULL
;
1369 mutex_exit(&hdl
->sa_lock
);
1373 sa_handle_destroy(sa_handle_t
*hdl
)
1375 mutex_enter(&hdl
->sa_lock
);
1376 (void) dmu_buf_update_user((dmu_buf_t
*)hdl
->sa_bonus
, hdl
,
1379 if (hdl
->sa_bonus_tab
) {
1380 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_bonus_tab
);
1381 hdl
->sa_bonus_tab
= NULL
;
1383 if (hdl
->sa_spill_tab
) {
1384 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1385 hdl
->sa_spill_tab
= NULL
;
1388 dmu_buf_rele(hdl
->sa_bonus
, NULL
);
1391 dmu_buf_rele((dmu_buf_t
*)hdl
->sa_spill
, NULL
);
1392 mutex_exit(&hdl
->sa_lock
);
1394 kmem_cache_free(sa_cache
, hdl
);
1398 sa_handle_get_from_db(objset_t
*os
, dmu_buf_t
*db
, void *userp
,
1399 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1402 sa_handle_t
*handle
;
1404 dmu_object_info_t doi
;
1406 dmu_object_info_from_db(db
, &doi
);
1407 ASSERT(doi
.doi_bonus_type
== DMU_OT_SA
||
1408 doi
.doi_bonus_type
== DMU_OT_ZNODE
);
1410 /* find handle, if it exists */
1411 /* if one doesn't exist then create a new one, and initialize it */
1413 handle
= (hdl_type
== SA_HDL_SHARED
) ? dmu_buf_get_user(db
) : NULL
;
1414 if (handle
== NULL
) {
1415 sa_handle_t
*newhandle
;
1416 handle
= kmem_cache_alloc(sa_cache
, KM_SLEEP
);
1417 handle
->sa_userp
= userp
;
1418 handle
->sa_bonus
= db
;
1420 handle
->sa_spill
= NULL
;
1422 error
= sa_build_index(handle
, SA_BONUS
);
1423 newhandle
= (hdl_type
== SA_HDL_SHARED
) ?
1424 dmu_buf_set_user_ie(db
, handle
,
1425 NULL
, sa_evict
) : NULL
;
1427 if (newhandle
!= NULL
) {
1428 kmem_cache_free(sa_cache
, handle
);
1438 sa_handle_get(objset_t
*objset
, uint64_t objid
, void *userp
,
1439 sa_handle_type_t hdl_type
, sa_handle_t
**handlepp
)
1444 if ((error
= dmu_bonus_hold(objset
, objid
, NULL
, &db
)))
1447 return (sa_handle_get_from_db(objset
, db
, userp
, hdl_type
,
1452 sa_buf_hold(objset_t
*objset
, uint64_t obj_num
, void *tag
, dmu_buf_t
**db
)
1454 return (dmu_bonus_hold(objset
, obj_num
, tag
, db
));
1458 sa_buf_rele(dmu_buf_t
*db
, void *tag
)
1460 dmu_buf_rele(db
, tag
);
1464 sa_lookup_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
)
1467 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1468 return (sa_attr_op(hdl
, bulk
, count
, SA_LOOKUP
, NULL
));
1472 sa_lookup(sa_handle_t
*hdl
, sa_attr_type_t attr
, void *buf
, uint32_t buflen
)
1475 sa_bulk_attr_t bulk
;
1477 bulk
.sa_attr
= attr
;
1479 bulk
.sa_length
= buflen
;
1480 bulk
.sa_data_func
= NULL
;
1483 mutex_enter(&hdl
->sa_lock
);
1484 error
= sa_lookup_impl(hdl
, &bulk
, 1);
1485 mutex_exit(&hdl
->sa_lock
);
1491 sa_lookup_uio(sa_handle_t
*hdl
, sa_attr_type_t attr
, uio_t
*uio
)
1494 sa_bulk_attr_t bulk
;
1496 bulk
.sa_data
= NULL
;
1497 bulk
.sa_attr
= attr
;
1498 bulk
.sa_data_func
= NULL
;
1502 mutex_enter(&hdl
->sa_lock
);
1503 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) == 0) {
1504 error
= uiomove((void *)bulk
.sa_addr
, MIN(bulk
.sa_size
,
1505 uio
->uio_resid
), UIO_READ
, uio
);
1507 mutex_exit(&hdl
->sa_lock
);
1513 sa_find_idx_tab(objset_t
*os
, dmu_object_type_t bonustype
, void *data
)
1515 sa_idx_tab_t
*idx_tab
;
1516 sa_hdr_phys_t
*hdr
= (sa_hdr_phys_t
*)data
;
1517 sa_os_t
*sa
= os
->os_sa
;
1518 sa_lot_t
*tb
, search
;
1522 * Deterimine layout number. If SA node and header == 0 then
1523 * force the index table to the dummy "1" empty layout.
1525 * The layout number would only be zero for a newly created file
1526 * that has not added any attributes yet, or with crypto enabled which
1527 * doesn't write any attributes to the bonus buffer.
1530 search
.lot_num
= SA_LAYOUT_NUM(hdr
, bonustype
);
1532 tb
= avl_find(&sa
->sa_layout_num_tree
, &search
, &loc
);
1534 /* Verify header size is consistent with layout information */
1536 ASSERT((IS_SA_BONUSTYPE(bonustype
) &&
1537 SA_HDR_SIZE_MATCH_LAYOUT(hdr
, tb
)) || !IS_SA_BONUSTYPE(bonustype
) ||
1538 (IS_SA_BONUSTYPE(bonustype
) && hdr
->sa_layout_info
== 0));
1541 * See if any of the already existing TOC entries can be reused?
1544 for (idx_tab
= list_head(&tb
->lot_idx_tab
); idx_tab
;
1545 idx_tab
= list_next(&tb
->lot_idx_tab
, idx_tab
)) {
1546 boolean_t valid_idx
= B_TRUE
;
1549 if (tb
->lot_var_sizes
!= 0 &&
1550 idx_tab
->sa_variable_lengths
!= NULL
) {
1551 for (i
= 0; i
!= tb
->lot_var_sizes
; i
++) {
1552 if (hdr
->sa_lengths
[i
] !=
1553 idx_tab
->sa_variable_lengths
[i
]) {
1554 valid_idx
= B_FALSE
;
1560 sa_idx_tab_hold(os
, idx_tab
);
1565 /* No such luck, create a new entry */
1566 idx_tab
= kmem_zalloc(sizeof (sa_idx_tab_t
), KM_PUSHPAGE
);
1567 idx_tab
->sa_idx_tab
=
1568 kmem_zalloc(sizeof (uint32_t) * sa
->sa_num_attrs
, KM_PUSHPAGE
);
1569 idx_tab
->sa_layout
= tb
;
1570 refcount_create(&idx_tab
->sa_refcount
);
1571 if (tb
->lot_var_sizes
)
1572 idx_tab
->sa_variable_lengths
= kmem_alloc(sizeof (uint16_t) *
1573 tb
->lot_var_sizes
, KM_PUSHPAGE
);
1575 sa_attr_iter(os
, hdr
, bonustype
, sa_build_idx_tab
,
1577 sa_idx_tab_hold(os
, idx_tab
); /* one hold for consumer */
1578 sa_idx_tab_hold(os
, idx_tab
); /* one for layout */
1579 list_insert_tail(&tb
->lot_idx_tab
, idx_tab
);
1584 sa_default_locator(void **dataptr
, uint32_t *len
, uint32_t total_len
,
1585 boolean_t start
, void *userdata
)
1589 *dataptr
= userdata
;
1594 sa_attr_register_sync(sa_handle_t
*hdl
, dmu_tx_t
*tx
)
1596 uint64_t attr_value
= 0;
1597 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1598 sa_attr_table_t
*tb
= sa
->sa_attr_table
;
1601 mutex_enter(&sa
->sa_lock
);
1603 if (!sa
->sa_need_attr_registration
|| sa
->sa_master_obj
== 0) {
1604 mutex_exit(&sa
->sa_lock
);
1608 if (sa
->sa_reg_attr_obj
== 0) {
1609 sa
->sa_reg_attr_obj
= zap_create_link(hdl
->sa_os
,
1610 DMU_OT_SA_ATTR_REGISTRATION
,
1611 sa
->sa_master_obj
, SA_REGISTRY
, tx
);
1613 for (i
= 0; i
!= sa
->sa_num_attrs
; i
++) {
1614 if (sa
->sa_attr_table
[i
].sa_registered
)
1616 ATTR_ENCODE(attr_value
, tb
[i
].sa_attr
, tb
[i
].sa_length
,
1618 VERIFY(0 == zap_update(hdl
->sa_os
, sa
->sa_reg_attr_obj
,
1619 tb
[i
].sa_name
, 8, 1, &attr_value
, tx
));
1620 tb
[i
].sa_registered
= B_TRUE
;
1622 sa
->sa_need_attr_registration
= B_FALSE
;
1623 mutex_exit(&sa
->sa_lock
);
1627 * Replace all attributes with attributes specified in template.
1628 * If dnode had a spill buffer then those attributes will be
1629 * also be replaced, possibly with just an empty spill block
1631 * This interface is intended to only be used for bulk adding of
1632 * attributes for a new file. It will also be used by the ZPL
1633 * when converting and old formatted znode to native SA support.
1636 sa_replace_all_by_template_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1637 int attr_count
, dmu_tx_t
*tx
)
1639 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1641 if (sa
->sa_need_attr_registration
)
1642 sa_attr_register_sync(hdl
, tx
);
1643 return (sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
));
1647 sa_replace_all_by_template(sa_handle_t
*hdl
, sa_bulk_attr_t
*attr_desc
,
1648 int attr_count
, dmu_tx_t
*tx
)
1652 mutex_enter(&hdl
->sa_lock
);
1653 error
= sa_replace_all_by_template_locked(hdl
, attr_desc
,
1655 mutex_exit(&hdl
->sa_lock
);
1660 * add/remove/replace a single attribute and then rewrite the entire set
1664 sa_modify_attrs(sa_handle_t
*hdl
, sa_attr_type_t newattr
,
1665 sa_data_op_t action
, sa_data_locator_t
*locator
, void *datastart
,
1666 uint16_t buflen
, dmu_tx_t
*tx
)
1668 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1669 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)hdl
->sa_bonus
;
1671 sa_bulk_attr_t
*attr_desc
;
1673 int bonus_attr_count
= 0;
1674 int bonus_data_size
= 0;
1675 int spill_attr_count
= 0;
1678 int i
, j
, k
, length_idx
;
1680 sa_idx_tab_t
*idx_tab
;
1684 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1686 /* First make of copy of the old data */
1690 if (dn
->dn_bonuslen
!= 0) {
1691 bonus_data_size
= hdl
->sa_bonus
->db_size
;
1692 old_data
[0] = kmem_alloc(bonus_data_size
, KM_SLEEP
);
1693 bcopy(hdl
->sa_bonus
->db_data
, old_data
[0],
1694 hdl
->sa_bonus
->db_size
);
1695 bonus_attr_count
= hdl
->sa_bonus_tab
->sa_layout
->lot_attr_count
;
1701 /* Bring spill buffer online if it isn't currently */
1703 if ((error
= sa_get_spill(hdl
)) == 0) {
1704 ASSERT3U(hdl
->sa_spill
->db_size
, <=, SPA_MAXBLOCKSIZE
);
1705 old_data
[1] = sa_spill_alloc(KM_SLEEP
);
1706 bcopy(hdl
->sa_spill
->db_data
, old_data
[1],
1707 hdl
->sa_spill
->db_size
);
1709 hdl
->sa_spill_tab
->sa_layout
->lot_attr_count
;
1710 } else if (error
&& error
!= ENOENT
) {
1712 kmem_free(old_data
[0], bonus_data_size
);
1718 /* build descriptor of all attributes */
1720 attr_count
= bonus_attr_count
+ spill_attr_count
;
1721 if (action
== SA_ADD
)
1723 else if (action
== SA_REMOVE
)
1726 attr_desc
= kmem_zalloc(sizeof (sa_bulk_attr_t
) * attr_count
, KM_SLEEP
);
1729 * loop through bonus and spill buffer if it exists, and
1730 * build up new attr_descriptor to reset the attributes
1733 count
= bonus_attr_count
;
1734 hdr
= SA_GET_HDR(hdl
, SA_BONUS
);
1735 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_BONUS
);
1736 for (; k
!= 2; k
++) {
1737 /* iterate over each attribute in layout */
1738 for (i
= 0, length_idx
= 0; i
!= count
; i
++) {
1739 sa_attr_type_t attr
;
1741 attr
= idx_tab
->sa_layout
->lot_attrs
[i
];
1742 if (attr
== newattr
) {
1743 if (action
== SA_REMOVE
) {
1747 ASSERT(SA_REGISTERED_LEN(sa
, attr
) == 0);
1748 ASSERT(action
== SA_REPLACE
);
1749 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1750 locator
, datastart
, buflen
);
1752 length
= SA_REGISTERED_LEN(sa
, attr
);
1754 length
= hdr
->sa_lengths
[length_idx
++];
1757 SA_ADD_BULK_ATTR(attr_desc
, j
, attr
,
1759 (TOC_OFF(idx_tab
->sa_idx_tab
[attr
]) +
1760 (uintptr_t)old_data
[k
]), length
);
1763 if (k
== 0 && hdl
->sa_spill
) {
1764 hdr
= SA_GET_HDR(hdl
, SA_SPILL
);
1765 idx_tab
= SA_IDX_TAB_GET(hdl
, SA_SPILL
);
1766 count
= spill_attr_count
;
1771 if (action
== SA_ADD
) {
1772 length
= SA_REGISTERED_LEN(sa
, newattr
);
1776 SA_ADD_BULK_ATTR(attr_desc
, j
, newattr
, locator
,
1780 error
= sa_build_layouts(hdl
, attr_desc
, attr_count
, tx
);
1783 kmem_free(old_data
[0], bonus_data_size
);
1785 sa_spill_free(old_data
[1]);
1786 kmem_free(attr_desc
, sizeof (sa_bulk_attr_t
) * attr_count
);
1792 sa_bulk_update_impl(sa_handle_t
*hdl
, sa_bulk_attr_t
*bulk
, int count
,
1796 sa_os_t
*sa
= hdl
->sa_os
->os_sa
;
1797 dmu_object_type_t bonustype
;
1798 dmu_buf_t
*saved_spill
;
1801 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1803 bonustype
= SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl
, SA_BONUS
));
1804 saved_spill
= hdl
->sa_spill
;
1806 /* sync out registration table if necessary */
1807 if (sa
->sa_need_attr_registration
)
1808 sa_attr_register_sync(hdl
, tx
);
1810 error
= sa_attr_op(hdl
, bulk
, count
, SA_UPDATE
, tx
);
1811 if (error
== 0 && !IS_SA_BONUSTYPE(bonustype
) && sa
->sa_update_cb
)
1812 sa
->sa_update_cb(hdl
, tx
);
1815 * If saved_spill is NULL and current sa_spill is not NULL that
1816 * means we increased the refcount of the spill buffer through
1817 * sa_get_spill() or dmu_spill_hold_by_dnode(). Therefore we
1818 * must release the hold before calling dmu_tx_commit() to avoid
1819 * making a copy of this buffer in dbuf_sync_leaf() due to the
1820 * reference count now being greater than 1.
1822 if (!saved_spill
&& hdl
->sa_spill
) {
1823 if (hdl
->sa_spill_tab
) {
1824 sa_idx_tab_rele(hdl
->sa_os
, hdl
->sa_spill_tab
);
1825 hdl
->sa_spill_tab
= NULL
;
1828 dmu_buf_rele((dmu_buf_t
*)hdl
->sa_spill
, NULL
);
1829 hdl
->sa_spill
= NULL
;
1836 * update or add new attribute
1839 sa_update(sa_handle_t
*hdl
, sa_attr_type_t type
,
1840 void *buf
, uint32_t buflen
, dmu_tx_t
*tx
)
1843 sa_bulk_attr_t bulk
;
1845 bulk
.sa_attr
= type
;
1846 bulk
.sa_data_func
= NULL
;
1847 bulk
.sa_length
= buflen
;
1850 mutex_enter(&hdl
->sa_lock
);
1851 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1852 mutex_exit(&hdl
->sa_lock
);
1857 sa_update_from_cb(sa_handle_t
*hdl
, sa_attr_type_t attr
,
1858 uint32_t buflen
, sa_data_locator_t
*locator
, void *userdata
, dmu_tx_t
*tx
)
1861 sa_bulk_attr_t bulk
;
1863 bulk
.sa_attr
= attr
;
1864 bulk
.sa_data
= userdata
;
1865 bulk
.sa_data_func
= locator
;
1866 bulk
.sa_length
= buflen
;
1868 mutex_enter(&hdl
->sa_lock
);
1869 error
= sa_bulk_update_impl(hdl
, &bulk
, 1, tx
);
1870 mutex_exit(&hdl
->sa_lock
);
1875 * Return size of an attribute
1879 sa_size(sa_handle_t
*hdl
, sa_attr_type_t attr
, int *size
)
1881 sa_bulk_attr_t bulk
;
1884 bulk
.sa_data
= NULL
;
1885 bulk
.sa_attr
= attr
;
1886 bulk
.sa_data_func
= NULL
;
1889 mutex_enter(&hdl
->sa_lock
);
1890 if ((error
= sa_attr_op(hdl
, &bulk
, 1, SA_LOOKUP
, NULL
)) != 0) {
1891 mutex_exit(&hdl
->sa_lock
);
1894 *size
= bulk
.sa_size
;
1896 mutex_exit(&hdl
->sa_lock
);
1901 sa_bulk_lookup_locked(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1904 ASSERT(MUTEX_HELD(&hdl
->sa_lock
));
1905 return (sa_lookup_impl(hdl
, attrs
, count
));
1909 sa_bulk_lookup(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
)
1914 mutex_enter(&hdl
->sa_lock
);
1915 error
= sa_bulk_lookup_locked(hdl
, attrs
, count
);
1916 mutex_exit(&hdl
->sa_lock
);
1921 sa_bulk_update(sa_handle_t
*hdl
, sa_bulk_attr_t
*attrs
, int count
, dmu_tx_t
*tx
)
1926 mutex_enter(&hdl
->sa_lock
);
1927 error
= sa_bulk_update_impl(hdl
, attrs
, count
, tx
);
1928 mutex_exit(&hdl
->sa_lock
);
1933 sa_remove(sa_handle_t
*hdl
, sa_attr_type_t attr
, dmu_tx_t
*tx
)
1937 mutex_enter(&hdl
->sa_lock
);
1938 error
= sa_modify_attrs(hdl
, attr
, SA_REMOVE
, NULL
,
1940 mutex_exit(&hdl
->sa_lock
);
1945 sa_object_info(sa_handle_t
*hdl
, dmu_object_info_t
*doi
)
1947 dmu_object_info_from_db((dmu_buf_t
*)hdl
->sa_bonus
, doi
);
1951 sa_object_size(sa_handle_t
*hdl
, uint32_t *blksize
, u_longlong_t
*nblocks
)
1953 dmu_object_size_from_db((dmu_buf_t
*)hdl
->sa_bonus
,
1958 sa_update_user(sa_handle_t
*newhdl
, sa_handle_t
*oldhdl
)
1960 (void) dmu_buf_update_user((dmu_buf_t
*)newhdl
->sa_bonus
,
1961 oldhdl
, newhdl
, NULL
, sa_evict
);
1962 oldhdl
->sa_bonus
= NULL
;
1966 sa_set_userp(sa_handle_t
*hdl
, void *ptr
)
1968 hdl
->sa_userp
= ptr
;
1972 sa_get_db(sa_handle_t
*hdl
)
1974 return ((dmu_buf_t
*)hdl
->sa_bonus
);
1978 sa_get_userdata(sa_handle_t
*hdl
)
1980 return (hdl
->sa_userp
);
1984 sa_register_update_callback_locked(objset_t
*os
, sa_update_cb_t
*func
)
1986 ASSERT(MUTEX_HELD(&os
->os_sa
->sa_lock
));
1987 os
->os_sa
->sa_update_cb
= func
;
1991 sa_register_update_callback(objset_t
*os
, sa_update_cb_t
*func
)
1994 mutex_enter(&os
->os_sa
->sa_lock
);
1995 sa_register_update_callback_locked(os
, func
);
1996 mutex_exit(&os
->os_sa
->sa_lock
);
2000 sa_handle_object(sa_handle_t
*hdl
)
2002 return (hdl
->sa_bonus
->db_object
);
2006 sa_enabled(objset_t
*os
)
2008 return (os
->os_sa
== NULL
);
2012 sa_set_sa_object(objset_t
*os
, uint64_t sa_object
)
2014 sa_os_t
*sa
= os
->os_sa
;
2016 if (sa
->sa_master_obj
)
2019 sa
->sa_master_obj
= sa_object
;
2025 sa_hdrsize(void *arg
)
2027 sa_hdr_phys_t
*hdr
= arg
;
2029 return (SA_HDR_SIZE(hdr
));
2033 sa_handle_lock(sa_handle_t
*hdl
)
2036 mutex_enter(&hdl
->sa_lock
);
2040 sa_handle_unlock(sa_handle_t
*hdl
)
2043 mutex_exit(&hdl
->sa_lock
);
2047 EXPORT_SYMBOL(sa_handle_get
);
2048 EXPORT_SYMBOL(sa_handle_get_from_db
);
2049 EXPORT_SYMBOL(sa_handle_destroy
);
2050 EXPORT_SYMBOL(sa_buf_hold
);
2051 EXPORT_SYMBOL(sa_buf_rele
);
2052 EXPORT_SYMBOL(sa_spill_rele
);
2053 EXPORT_SYMBOL(sa_lookup
);
2054 EXPORT_SYMBOL(sa_update
);
2055 EXPORT_SYMBOL(sa_remove
);
2056 EXPORT_SYMBOL(sa_bulk_lookup
);
2057 EXPORT_SYMBOL(sa_bulk_lookup_locked
);
2058 EXPORT_SYMBOL(sa_bulk_update
);
2059 EXPORT_SYMBOL(sa_size
);
2060 EXPORT_SYMBOL(sa_update_from_cb
);
2061 EXPORT_SYMBOL(sa_object_info
);
2062 EXPORT_SYMBOL(sa_object_size
);
2063 EXPORT_SYMBOL(sa_update_user
);
2064 EXPORT_SYMBOL(sa_get_userdata
);
2065 EXPORT_SYMBOL(sa_set_userp
);
2066 EXPORT_SYMBOL(sa_get_db
);
2067 EXPORT_SYMBOL(sa_handle_object
);
2068 EXPORT_SYMBOL(sa_register_update_callback
);
2069 EXPORT_SYMBOL(sa_setup
);
2070 EXPORT_SYMBOL(sa_replace_all_by_template
);
2071 EXPORT_SYMBOL(sa_replace_all_by_template_locked
);
2072 EXPORT_SYMBOL(sa_enabled
);
2073 EXPORT_SYMBOL(sa_cache_init
);
2074 EXPORT_SYMBOL(sa_cache_fini
);
2075 EXPORT_SYMBOL(sa_spill_alloc
);
2076 EXPORT_SYMBOL(sa_spill_free
);
2077 EXPORT_SYMBOL(sa_set_sa_object
);
2078 EXPORT_SYMBOL(sa_hdrsize
);
2079 EXPORT_SYMBOL(sa_handle_lock
);
2080 EXPORT_SYMBOL(sa_handle_unlock
);
2081 EXPORT_SYMBOL(sa_lookup_uio
);
2082 #endif /* _KERNEL */