1 /*
2 * CDDL HEADER START
3 *
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.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 */
27
28 #include <sys/zfs_context.h>
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/sysmacros.h>
32 #include <sys/dmu.h>
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/dbuf.h>
37 #include <sys/dnode.h>
38 #include <sys/zap.h>
39 #include <sys/sa.h>
40 #include <sys/sunddi.h>
41 #include <sys/sa_impl.h>
42 #include <sys/errno.h>
43 #include <sys/zfs_context.h>
44
45 #ifdef _KERNEL
46 #include <sys/zfs_znode.h>
47 #endif
48
49 /*
50 * ZFS System attributes:
51 *
52 * A generic mechanism to allow for arbitrary attributes
53 * to be stored in a dnode. The data will be stored in the bonus buffer of
54 * the dnode and if necessary a special "spill" block will be used to handle
55 * overflow situations. The spill block will be sized to fit the data
56 * from 512 - 128K. When a spill block is used the BP (blkptr_t) for the
57 * spill block is stored at the end of the current bonus buffer. Any
58 * attributes that would be in the way of the blkptr_t will be relocated
59 * into the spill block.
60 *
61 * Attribute registration:
62 *
63 * Stored persistently on a per dataset basis
64 * a mapping between attribute "string" names and their actual attribute
65 * numeric values, length, and byteswap function. The names are only used
66 * during registration. All attributes are known by their unique attribute
67 * id value. If an attribute can have a variable size then the value
68 * 0 will be used to indicate this.
69 *
70 * Attribute Layout:
71 *
72 * Attribute layouts are a way to compactly store multiple attributes, but
73 * without taking the overhead associated with managing each attribute
74 * individually. Since you will typically have the same set of attributes
75 * stored in the same order a single table will be used to represent that
76 * layout. The ZPL for example will usually have only about 10 different
77 * layouts (regular files, device files, symlinks,
78 * regular files + scanstamp, files/dir with extended attributes, and then
79 * you have the possibility of all of those minus ACL, because it would
80 * be kicked out into the spill block)
81 *
82 * Layouts are simply an array of the attributes and their
83 * ordering i.e. [0, 1, 4, 5, 2]
84 *
85 * Each distinct layout is given a unique layout number and that is what's
86 * stored in the header at the beginning of the SA data buffer.
87 *
88 * A layout only covers a single dbuf (bonus or spill). If a set of
89 * attributes is split up between the bonus buffer and a spill buffer then
90 * two different layouts will be used. This allows us to byteswap the
91 * spill without looking at the bonus buffer and keeps the on disk format of
92 * the bonus and spill buffer the same.
93 *
94 * Adding a single attribute will cause the entire set of attributes to
95 * be rewritten and could result in a new layout number being constructed
96 * as part of the rewrite if no such layout exists for the new set of
97 * attributes. The new attribute will be appended to the end of the already
98 * existing attributes.
99 *
100 * Both the attribute registration and attribute layout information are
101 * stored in normal ZAP attributes. Their should be a small number of
102 * known layouts and the set of attributes is assumed to typically be quite
103 * small.
104 *
105 * The registered attributes and layout "table" information is maintained
106 * in core and a special "sa_os_t" is attached to the objset_t.
107 *
108 * A special interface is provided to allow for quickly applying
109 * a large set of attributes at once. sa_replace_all_by_template() is
110 * used to set an array of attributes. This is used by the ZPL when
111 * creating a brand new file. The template that is passed into the function
112 * specifies the attribute, size for variable length attributes, location of
113 * data and special "data locator" function if the data isn't in a contiguous
114 * location.
115 *
116 * Byteswap implications:
117 *
118 * Since the SA attributes are not entirely self describing we can't do
119 * the normal byteswap processing. The special ZAP layout attribute and
120 * attribute registration attributes define the byteswap function and the
121 * size of the attributes, unless it is variable sized.
122 * The normal ZFS byteswapping infrastructure assumes you don't need
123 * to read any objects in order to do the necessary byteswapping. Whereas
124 * SA attributes can only be properly byteswapped if the dataset is opened
125 * and the layout/attribute ZAP attributes are available. Because of this
126 * the SA attributes will be byteswapped when they are first accessed by
127 * the SA code that will read the SA data.
128 */
129
130 typedef void (sa_iterfunc_t)(void *hdr, void *addr, sa_attr_type_t,
131 uint16_t length, int length_idx, boolean_t, void *userp);
132
133 static int sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype);
134 static void sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab);
135 static sa_idx_tab_t *sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype,
136 sa_hdr_phys_t *hdr);
137 static void sa_idx_tab_rele(objset_t *os, void *arg);
138 static void sa_copy_data(sa_data_locator_t *func, void *start, void *target,
139 int buflen);
140 static int sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
141 sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
142 uint16_t buflen, dmu_tx_t *tx);
143
144 static arc_byteswap_func_t sa_bswap_table[] = {
145 byteswap_uint64_array,
146 byteswap_uint32_array,
147 byteswap_uint16_array,
148 byteswap_uint8_array,
149 zfs_acl_byteswap,
150 };
151
152 #ifdef HAVE_EFFICIENT_UNALIGNED_ACCESS
153 #define SA_COPY_DATA(f, s, t, l) \
154 do { \
155 if (f == NULL) { \
156 if (l == 8) { \
157 *(uint64_t *)t = *(uint64_t *)s; \
158 } else if (l == 16) { \
159 *(uint64_t *)t = *(uint64_t *)s; \
160 *(uint64_t *)((uintptr_t)t + 8) = \
161 *(uint64_t *)((uintptr_t)s + 8); \
162 } else { \
163 memcpy(t, s, l); \
164 } \
165 } else { \
166 sa_copy_data(f, s, t, l); \
167 } \
168 } while (0)
169 #else
170 #define SA_COPY_DATA(f, s, t, l) sa_copy_data(f, s, t, l)
171 #endif
172
173 /*
174 * This table is fixed and cannot be changed. Its purpose is to
175 * allow the SA code to work with both old/new ZPL file systems.
176 * It contains the list of legacy attributes. These attributes aren't
177 * stored in the "attribute" registry zap objects, since older ZPL file systems
178 * won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
179 * use this static table.
180 */
181 static const sa_attr_reg_t sa_legacy_attrs[] = {
182 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
183 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
184 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
185 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
186 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
187 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
188 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
189 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
190 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
191 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
192 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
193 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
194 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
195 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
196 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
197 {"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
198 };
199
200 /*
201 * This is only used for objects of type DMU_OT_ZNODE
202 */
203 static const sa_attr_type_t sa_legacy_zpl_layout[] = {
204 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
205 };
206
207 /*
208 * Special dummy layout used for buffers with no attributes.
209 */
210 static const sa_attr_type_t sa_dummy_zpl_layout[] = { 0 };
211
212 static const size_t sa_legacy_attr_count = ARRAY_SIZE(sa_legacy_attrs);
213 static kmem_cache_t *sa_cache = NULL;
214
215 static int
sa_cache_constructor(void * buf,void * unused,int kmflag)216 sa_cache_constructor(void *buf, void *unused, int kmflag)
217 {
218 (void) unused, (void) kmflag;
219 sa_handle_t *hdl = buf;
220
221 mutex_init(&hdl->sa_lock, NULL, MUTEX_DEFAULT, NULL);
222 return (0);
223 }
224
225 static void
sa_cache_destructor(void * buf,void * unused)226 sa_cache_destructor(void *buf, void *unused)
227 {
228 (void) unused;
229 sa_handle_t *hdl = buf;
230 mutex_destroy(&hdl->sa_lock);
231 }
232
233 void
sa_cache_init(void)234 sa_cache_init(void)
235 {
236 sa_cache = kmem_cache_create("sa_cache",
237 sizeof (sa_handle_t), 0, sa_cache_constructor,
238 sa_cache_destructor, NULL, NULL, NULL, 0);
239 }
240
241 void
sa_cache_fini(void)242 sa_cache_fini(void)
243 {
244 if (sa_cache)
245 kmem_cache_destroy(sa_cache);
246 }
247
248 static int
layout_num_compare(const void * arg1,const void * arg2)249 layout_num_compare(const void *arg1, const void *arg2)
250 {
251 const sa_lot_t *node1 = (const sa_lot_t *)arg1;
252 const sa_lot_t *node2 = (const sa_lot_t *)arg2;
253
254 return (TREE_CMP(node1->lot_num, node2->lot_num));
255 }
256
257 static int
layout_hash_compare(const void * arg1,const void * arg2)258 layout_hash_compare(const void *arg1, const void *arg2)
259 {
260 const sa_lot_t *node1 = (const sa_lot_t *)arg1;
261 const sa_lot_t *node2 = (const sa_lot_t *)arg2;
262
263 int cmp = TREE_CMP(node1->lot_hash, node2->lot_hash);
264 if (likely(cmp))
265 return (cmp);
266
267 return (TREE_CMP(node1->lot_instance, node2->lot_instance));
268 }
269
270 static boolean_t
sa_layout_equal(sa_lot_t * tbf,sa_attr_type_t * attrs,int count)271 sa_layout_equal(sa_lot_t *tbf, sa_attr_type_t *attrs, int count)
272 {
273 int i;
274
275 if (count != tbf->lot_attr_count)
276 return (1);
277
278 for (i = 0; i != count; i++) {
279 if (attrs[i] != tbf->lot_attrs[i])
280 return (1);
281 }
282 return (0);
283 }
284
285 #define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
286
287 static uint64_t
sa_layout_info_hash(const sa_attr_type_t * attrs,int attr_count)288 sa_layout_info_hash(const sa_attr_type_t *attrs, int attr_count)
289 {
290 uint64_t crc = -1ULL;
291
292 for (int i = 0; i != attr_count; i++)
293 crc ^= SA_ATTR_HASH(attrs[i]);
294
295 return (crc);
296 }
297
298 static int
sa_get_spill(sa_handle_t * hdl)299 sa_get_spill(sa_handle_t *hdl)
300 {
301 int rc;
302 if (hdl->sa_spill == NULL) {
303 if ((rc = dmu_spill_hold_existing(hdl->sa_bonus, NULL,
304 &hdl->sa_spill)) == 0)
305 VERIFY(0 == sa_build_index(hdl, SA_SPILL));
306 } else {
307 rc = 0;
308 }
309
310 return (rc);
311 }
312
313 /*
314 * Main attribute lookup/update function
315 * returns 0 for success or non zero for failures
316 *
317 * Operates on bulk array, first failure will abort further processing
318 */
319 static int
sa_attr_op(sa_handle_t * hdl,sa_bulk_attr_t * bulk,int count,sa_data_op_t data_op,dmu_tx_t * tx)320 sa_attr_op(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
321 sa_data_op_t data_op, dmu_tx_t *tx)
322 {
323 sa_os_t *sa = hdl->sa_os->os_sa;
324 int i;
325 int error = 0;
326 sa_buf_type_t buftypes;
327
328 buftypes = 0;
329
330 ASSERT(count > 0);
331 for (i = 0; i != count; i++) {
332 ASSERT(bulk[i].sa_attr <= hdl->sa_os->os_sa->sa_num_attrs);
333
334 bulk[i].sa_addr = NULL;
335 /* First check the bonus buffer */
336
337 if (hdl->sa_bonus_tab && TOC_ATTR_PRESENT(
338 hdl->sa_bonus_tab->sa_idx_tab[bulk[i].sa_attr])) {
339 SA_ATTR_INFO(sa, hdl->sa_bonus_tab,
340 SA_GET_HDR(hdl, SA_BONUS),
341 bulk[i].sa_attr, bulk[i], SA_BONUS, hdl);
342 if (tx && !(buftypes & SA_BONUS)) {
343 dmu_buf_will_dirty(hdl->sa_bonus, tx);
344 buftypes |= SA_BONUS;
345 }
346 }
347 if (bulk[i].sa_addr == NULL &&
348 ((error = sa_get_spill(hdl)) == 0)) {
349 if (TOC_ATTR_PRESENT(
350 hdl->sa_spill_tab->sa_idx_tab[bulk[i].sa_attr])) {
351 SA_ATTR_INFO(sa, hdl->sa_spill_tab,
352 SA_GET_HDR(hdl, SA_SPILL),
353 bulk[i].sa_attr, bulk[i], SA_SPILL, hdl);
354 if (tx && !(buftypes & SA_SPILL) &&
355 bulk[i].sa_size == bulk[i].sa_length) {
356 dmu_buf_will_dirty(hdl->sa_spill, tx);
357 buftypes |= SA_SPILL;
358 }
359 }
360 }
361 if (error && error != ENOENT) {
362 return ((error == ECKSUM) ? EIO : error);
363 }
364
365 switch (data_op) {
366 case SA_LOOKUP:
367 if (bulk[i].sa_addr == NULL)
368 return (SET_ERROR(ENOENT));
369 if (bulk[i].sa_data) {
370 SA_COPY_DATA(bulk[i].sa_data_func,
371 bulk[i].sa_addr, bulk[i].sa_data,
372 bulk[i].sa_size);
373 }
374 continue;
375
376 case SA_UPDATE:
377 /* existing rewrite of attr */
378 if (bulk[i].sa_addr &&
379 bulk[i].sa_size == bulk[i].sa_length) {
380 SA_COPY_DATA(bulk[i].sa_data_func,
381 bulk[i].sa_data, bulk[i].sa_addr,
382 bulk[i].sa_length);
383 continue;
384 } else if (bulk[i].sa_addr) { /* attr size change */
385 error = sa_modify_attrs(hdl, bulk[i].sa_attr,
386 SA_REPLACE, bulk[i].sa_data_func,
387 bulk[i].sa_data, bulk[i].sa_length, tx);
388 } else { /* adding new attribute */
389 error = sa_modify_attrs(hdl, bulk[i].sa_attr,
390 SA_ADD, bulk[i].sa_data_func,
391 bulk[i].sa_data, bulk[i].sa_length, tx);
392 }
393 if (error)
394 return (error);
395 break;
396 default:
397 break;
398 }
399 }
400 return (error);
401 }
402
403 static sa_lot_t *
sa_add_layout_entry(objset_t * os,const sa_attr_type_t * attrs,int attr_count,uint64_t lot_num,uint64_t hash,boolean_t zapadd,dmu_tx_t * tx)404 sa_add_layout_entry(objset_t *os, const sa_attr_type_t *attrs, int attr_count,
405 uint64_t lot_num, uint64_t hash, boolean_t zapadd, dmu_tx_t *tx)
406 {
407 sa_os_t *sa = os->os_sa;
408 sa_lot_t *tb, *findtb;
409 int i;
410 avl_index_t loc;
411
412 ASSERT(MUTEX_HELD(&sa->sa_lock));
413 tb = kmem_zalloc(sizeof (sa_lot_t), KM_SLEEP);
414 tb->lot_attr_count = attr_count;
415 tb->lot_attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
416 KM_SLEEP);
417 memcpy(tb->lot_attrs, attrs, sizeof (sa_attr_type_t) * attr_count);
418 tb->lot_num = lot_num;
419 tb->lot_hash = hash;
420 tb->lot_instance = 0;
421
422 if (zapadd) {
423 char attr_name[8];
424
425 if (sa->sa_layout_attr_obj == 0) {
426 sa->sa_layout_attr_obj = zap_create_link(os,
427 DMU_OT_SA_ATTR_LAYOUTS,
428 sa->sa_master_obj, SA_LAYOUTS, tx);
429 }
430
431 (void) snprintf(attr_name, sizeof (attr_name),
432 "%d", (int)lot_num);
433 VERIFY(0 == zap_update(os, os->os_sa->sa_layout_attr_obj,
434 attr_name, 2, attr_count, attrs, tx));
435 }
436
437 list_create(&tb->lot_idx_tab, sizeof (sa_idx_tab_t),
438 offsetof(sa_idx_tab_t, sa_next));
439
440 for (i = 0; i != attr_count; i++) {
441 if (sa->sa_attr_table[tb->lot_attrs[i]].sa_length == 0)
442 tb->lot_var_sizes++;
443 }
444
445 avl_add(&sa->sa_layout_num_tree, tb);
446
447 /* verify we don't have a hash collision */
448 if ((findtb = avl_find(&sa->sa_layout_hash_tree, tb, &loc)) != NULL) {
449 for (; findtb && findtb->lot_hash == hash;
450 findtb = AVL_NEXT(&sa->sa_layout_hash_tree, findtb)) {
451 if (findtb->lot_instance != tb->lot_instance)
452 break;
453 tb->lot_instance++;
454 }
455 }
456 avl_add(&sa->sa_layout_hash_tree, tb);
457 return (tb);
458 }
459
460 static void
sa_find_layout(objset_t * os,uint64_t hash,sa_attr_type_t * attrs,int count,dmu_tx_t * tx,sa_lot_t ** lot)461 sa_find_layout(objset_t *os, uint64_t hash, sa_attr_type_t *attrs,
462 int count, dmu_tx_t *tx, sa_lot_t **lot)
463 {
464 sa_lot_t *tb, tbsearch;
465 avl_index_t loc;
466 sa_os_t *sa = os->os_sa;
467 boolean_t found = B_FALSE;
468
469 mutex_enter(&sa->sa_lock);
470 tbsearch.lot_hash = hash;
471 tbsearch.lot_instance = 0;
472 tb = avl_find(&sa->sa_layout_hash_tree, &tbsearch, &loc);
473 if (tb) {
474 for (; tb && tb->lot_hash == hash;
475 tb = AVL_NEXT(&sa->sa_layout_hash_tree, tb)) {
476 if (sa_layout_equal(tb, attrs, count) == 0) {
477 found = B_TRUE;
478 break;
479 }
480 }
481 }
482 if (!found) {
483 tb = sa_add_layout_entry(os, attrs, count,
484 avl_numnodes(&sa->sa_layout_num_tree), hash, B_TRUE, tx);
485 }
486 mutex_exit(&sa->sa_lock);
487 *lot = tb;
488 }
489
490 static int
sa_resize_spill(sa_handle_t * hdl,uint32_t size,dmu_tx_t * tx)491 sa_resize_spill(sa_handle_t *hdl, uint32_t size, dmu_tx_t *tx)
492 {
493 int error;
494 uint32_t blocksize;
495
496 if (size == 0) {
497 blocksize = SPA_MINBLOCKSIZE;
498 } else if (size > SPA_OLD_MAXBLOCKSIZE) {
499 ASSERT(0);
500 return (SET_ERROR(EFBIG));
501 } else {
502 blocksize = P2ROUNDUP_TYPED(size, SPA_MINBLOCKSIZE, uint32_t);
503 }
504
505 error = dbuf_spill_set_blksz(hdl->sa_spill, blocksize, tx);
506 ASSERT(error == 0);
507 return (error);
508 }
509
510 static void
sa_copy_data(sa_data_locator_t * func,void * datastart,void * target,int buflen)511 sa_copy_data(sa_data_locator_t *func, void *datastart, void *target, int buflen)
512 {
513 if (func == NULL) {
514 memcpy(target, datastart, buflen);
515 } else {
516 boolean_t start;
517 int bytes;
518 void *dataptr;
519 void *saptr = target;
520 uint32_t length;
521
522 start = B_TRUE;
523 bytes = 0;
524 while (bytes < buflen) {
525 func(&dataptr, &length, buflen, start, datastart);
526 memcpy(saptr, dataptr, length);
527 saptr = (void *)((caddr_t)saptr + length);
528 bytes += length;
529 start = B_FALSE;
530 }
531 }
532 }
533
534 /*
535 * Determine several different values pertaining to system attribute
536 * buffers.
537 *
538 * Return the size of the sa_hdr_phys_t header for the buffer. Each
539 * variable length attribute except the first contributes two bytes to
540 * the header size, which is then rounded up to an 8-byte boundary.
541 *
542 * The following output parameters are also computed.
543 *
544 * index - The index of the first attribute in attr_desc that will
545 * spill over. Only valid if will_spill is set.
546 *
547 * total - The total number of bytes of all system attributes described
548 * in attr_desc.
549 *
550 * will_spill - Set when spilling is necessary. It is only set when
551 * the buftype is SA_BONUS.
552 */
553 static int
sa_find_sizes(sa_os_t * sa,sa_bulk_attr_t * attr_desc,int attr_count,dmu_buf_t * db,sa_buf_type_t buftype,int full_space,int * index,int * total,boolean_t * will_spill)554 sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count,
555 dmu_buf_t *db, sa_buf_type_t buftype, int full_space, int *index,
556 int *total, boolean_t *will_spill)
557 {
558 int var_size_count = 0;
559 int i;
560 int hdrsize;
561 int extra_hdrsize;
562
563 if (buftype == SA_BONUS && sa->sa_force_spill) {
564 *total = 0;
565 *index = 0;
566 *will_spill = B_TRUE;
567 return (0);
568 }
569
570 *index = -1;
571 *total = 0;
572 *will_spill = B_FALSE;
573
574 extra_hdrsize = 0;
575 hdrsize = (SA_BONUSTYPE_FROM_DB(db) == DMU_OT_ZNODE) ? 0 :
576 sizeof (sa_hdr_phys_t);
577
578 ASSERT(IS_P2ALIGNED(full_space, 8));
579
580 for (i = 0; i != attr_count; i++) {
581 boolean_t is_var_sz, might_spill_here;
582 int tmp_hdrsize;
583
584 *total = P2ROUNDUP(*total, 8);
585 *total += attr_desc[i].sa_length;
586 if (*will_spill)
587 continue;
588
589 is_var_sz = (SA_REGISTERED_LEN(sa, attr_desc[i].sa_attr) == 0);
590 if (is_var_sz)
591 var_size_count++;
592
593 /*
594 * Calculate what the SA header size would be if this
595 * attribute doesn't spill.
596 */
597 tmp_hdrsize = hdrsize + ((is_var_sz && var_size_count > 1) ?
598 sizeof (uint16_t) : 0);
599
600 /*
601 * Check whether this attribute spans into the space
602 * that would be used by the spill block pointer should
603 * a spill block be needed.
604 */
605 might_spill_here =
606 buftype == SA_BONUS && *index == -1 &&
607 (*total + P2ROUNDUP(tmp_hdrsize, 8)) >
608 (full_space - sizeof (blkptr_t));
609
610 if (is_var_sz && var_size_count > 1) {
611 if (buftype == SA_SPILL ||
612 tmp_hdrsize + *total < full_space) {
613 /*
614 * Record the extra header size in case this
615 * increase needs to be reversed due to
616 * spill-over.
617 */
618 hdrsize = tmp_hdrsize;
619 if (*index != -1 || might_spill_here)
620 extra_hdrsize += sizeof (uint16_t);
621 } else {
622 ASSERT(buftype == SA_BONUS);
623 if (*index == -1)
624 *index = i;
625 *will_spill = B_TRUE;
626 continue;
627 }
628 }
629
630 /*
631 * Store index of where spill *could* occur. Then
632 * continue to count the remaining attribute sizes. The
633 * sum is used later for sizing bonus and spill buffer.
634 */
635 if (might_spill_here)
636 *index = i;
637
638 if ((*total + P2ROUNDUP(hdrsize, 8)) > full_space &&
639 buftype == SA_BONUS)
640 *will_spill = B_TRUE;
641 }
642
643 if (*will_spill)
644 hdrsize -= extra_hdrsize;
645
646 hdrsize = P2ROUNDUP(hdrsize, 8);
647 return (hdrsize);
648 }
649
650 #define BUF_SPACE_NEEDED(total, header) (total + header)
651
652 /*
653 * Find layout that corresponds to ordering of attributes
654 * If not found a new layout number is created and added to
655 * persistent layout tables.
656 */
657 static int
sa_build_layouts(sa_handle_t * hdl,sa_bulk_attr_t * attr_desc,int attr_count,dmu_tx_t * tx)658 sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
659 dmu_tx_t *tx)
660 {
661 sa_os_t *sa = hdl->sa_os->os_sa;
662 uint64_t hash;
663 sa_buf_type_t buftype;
664 sa_hdr_phys_t *sahdr;
665 void *data_start;
666 sa_attr_type_t *attrs, *attrs_start;
667 int i, lot_count;
668 int dnodesize;
669 int spill_idx;
670 int hdrsize;
671 int spillhdrsize = 0;
672 int used;
673 dmu_object_type_t bonustype;
674 sa_lot_t *lot;
675 int len_idx;
676 int spill_used;
677 int bonuslen;
678 boolean_t spilling;
679
680 dmu_buf_will_dirty(hdl->sa_bonus, tx);
681 bonustype = SA_BONUSTYPE_FROM_DB(hdl->sa_bonus);
682 dmu_object_dnsize_from_db(hdl->sa_bonus, &dnodesize);
683 bonuslen = DN_BONUS_SIZE(dnodesize);
684
685 /* first determine bonus header size and sum of all attributes */
686 hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus,
687 SA_BONUS, bonuslen, &spill_idx, &used, &spilling);
688
689 if (used > SPA_OLD_MAXBLOCKSIZE)
690 return (SET_ERROR(EFBIG));
691
692 VERIFY0(dmu_set_bonus(hdl->sa_bonus, spilling ?
693 MIN(bonuslen - sizeof (blkptr_t), used + hdrsize) :
694 used + hdrsize, tx));
695
696 ASSERT((bonustype == DMU_OT_ZNODE && spilling == 0) ||
697 bonustype == DMU_OT_SA);
698
699 /* setup and size spill buffer when needed */
700 if (spilling) {
701 boolean_t dummy;
702
703 if (hdl->sa_spill == NULL) {
704 VERIFY(dmu_spill_hold_by_bonus(hdl->sa_bonus, 0, NULL,
705 &hdl->sa_spill) == 0);
706 }
707 dmu_buf_will_dirty(hdl->sa_spill, tx);
708
709 spillhdrsize = sa_find_sizes(sa, &attr_desc[spill_idx],
710 attr_count - spill_idx, hdl->sa_spill, SA_SPILL,
711 hdl->sa_spill->db_size, &i, &spill_used, &dummy);
712
713 if (spill_used > SPA_OLD_MAXBLOCKSIZE)
714 return (SET_ERROR(EFBIG));
715
716 if (BUF_SPACE_NEEDED(spill_used, spillhdrsize) >
717 hdl->sa_spill->db_size)
718 VERIFY(0 == sa_resize_spill(hdl,
719 BUF_SPACE_NEEDED(spill_used, spillhdrsize), tx));
720 }
721
722 /* setup starting pointers to lay down data */
723 data_start = (void *)((uintptr_t)hdl->sa_bonus->db_data + hdrsize);
724 sahdr = (sa_hdr_phys_t *)hdl->sa_bonus->db_data;
725 buftype = SA_BONUS;
726
727 attrs_start = attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
728 KM_SLEEP);
729 lot_count = 0;
730
731 for (i = 0, len_idx = 0, hash = -1ULL; i != attr_count; i++) {
732 uint16_t length;
733
734 ASSERT(IS_P2ALIGNED(data_start, 8));
735 attrs[i] = attr_desc[i].sa_attr;
736 length = SA_REGISTERED_LEN(sa, attrs[i]);
737 if (length == 0)
738 length = attr_desc[i].sa_length;
739
740 if (spilling && i == spill_idx) { /* switch to spill buffer */
741 VERIFY(bonustype == DMU_OT_SA);
742 if (buftype == SA_BONUS && !sa->sa_force_spill) {
743 sa_find_layout(hdl->sa_os, hash, attrs_start,
744 lot_count, tx, &lot);
745 SA_SET_HDR(sahdr, lot->lot_num, hdrsize);
746 }
747
748 buftype = SA_SPILL;
749 hash = -1ULL;
750 len_idx = 0;
751
752 sahdr = (sa_hdr_phys_t *)hdl->sa_spill->db_data;
753 sahdr->sa_magic = SA_MAGIC;
754 data_start = (void *)((uintptr_t)sahdr +
755 spillhdrsize);
756 attrs_start = &attrs[i];
757 lot_count = 0;
758 }
759 hash ^= SA_ATTR_HASH(attrs[i]);
760 attr_desc[i].sa_addr = data_start;
761 attr_desc[i].sa_size = length;
762 SA_COPY_DATA(attr_desc[i].sa_data_func, attr_desc[i].sa_data,
763 data_start, length);
764 if (sa->sa_attr_table[attrs[i]].sa_length == 0) {
765 sahdr->sa_lengths[len_idx++] = length;
766 }
767 data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
768 length), 8);
769 lot_count++;
770 }
771
772 sa_find_layout(hdl->sa_os, hash, attrs_start, lot_count, tx, &lot);
773
774 /*
775 * Verify that old znodes always have layout number 0.
776 * Must be DMU_OT_SA for arbitrary layouts
777 */
778 VERIFY((bonustype == DMU_OT_ZNODE && lot->lot_num == 0) ||
779 (bonustype == DMU_OT_SA && lot->lot_num > 1));
780
781 if (bonustype == DMU_OT_SA) {
782 SA_SET_HDR(sahdr, lot->lot_num,
783 buftype == SA_BONUS ? hdrsize : spillhdrsize);
784 }
785
786 kmem_free(attrs, sizeof (sa_attr_type_t) * attr_count);
787 if (hdl->sa_bonus_tab) {
788 sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
789 hdl->sa_bonus_tab = NULL;
790 }
791 if (!sa->sa_force_spill)
792 VERIFY(0 == sa_build_index(hdl, SA_BONUS));
793 if (hdl->sa_spill) {
794 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
795 if (!spilling) {
796 /*
797 * remove spill block that is no longer needed.
798 */
799 dmu_buf_rele(hdl->sa_spill, NULL);
800 hdl->sa_spill = NULL;
801 hdl->sa_spill_tab = NULL;
802 VERIFY(0 == dmu_rm_spill(hdl->sa_os,
803 sa_handle_object(hdl), tx));
804 } else {
805 VERIFY(0 == sa_build_index(hdl, SA_SPILL));
806 }
807 }
808
809 return (0);
810 }
811
812 static void
sa_free_attr_table(sa_os_t * sa)813 sa_free_attr_table(sa_os_t *sa)
814 {
815 int i;
816
817 if (sa->sa_attr_table == NULL)
818 return;
819
820 for (i = 0; i != sa->sa_num_attrs; i++) {
821 if (sa->sa_attr_table[i].sa_name)
822 kmem_free(sa->sa_attr_table[i].sa_name,
823 strlen(sa->sa_attr_table[i].sa_name) + 1);
824 }
825
826 kmem_free(sa->sa_attr_table,
827 sizeof (sa_attr_table_t) * sa->sa_num_attrs);
828
829 sa->sa_attr_table = NULL;
830 }
831
832 static int
sa_attr_table_setup(objset_t * os,const sa_attr_reg_t * reg_attrs,int count)833 sa_attr_table_setup(objset_t *os, const sa_attr_reg_t *reg_attrs, int count)
834 {
835 sa_os_t *sa = os->os_sa;
836 uint64_t sa_attr_count = 0;
837 uint64_t sa_reg_count = 0;
838 int error = 0;
839 uint64_t attr_value;
840 sa_attr_table_t *tb;
841 zap_cursor_t zc;
842 zap_attribute_t za;
843 int registered_count = 0;
844 int i;
845 dmu_objset_type_t ostype = dmu_objset_type(os);
846
847 sa->sa_user_table =
848 kmem_zalloc(count * sizeof (sa_attr_type_t), KM_SLEEP);
849 sa->sa_user_table_sz = count * sizeof (sa_attr_type_t);
850
851 if (sa->sa_reg_attr_obj != 0) {
852 error = zap_count(os, sa->sa_reg_attr_obj,
853 &sa_attr_count);
854
855 /*
856 * Make sure we retrieved a count and that it isn't zero
857 */
858 if (error || (error == 0 && sa_attr_count == 0)) {
859 if (error == 0)
860 error = SET_ERROR(EINVAL);
861 goto bail;
862 }
863 sa_reg_count = sa_attr_count;
864 }
865
866 if (ostype == DMU_OST_ZFS && sa_attr_count == 0)
867 sa_attr_count += sa_legacy_attr_count;
868
869 /* Allocate attribute numbers for attributes that aren't registered */
870 for (i = 0; i != count; i++) {
871 boolean_t found = B_FALSE;
872 int j;
873
874 if (ostype == DMU_OST_ZFS) {
875 for (j = 0; j != sa_legacy_attr_count; j++) {
876 if (strcmp(reg_attrs[i].sa_name,
877 sa_legacy_attrs[j].sa_name) == 0) {
878 sa->sa_user_table[i] =
879 sa_legacy_attrs[j].sa_attr;
880 found = B_TRUE;
881 }
882 }
883 }
884 if (found)
885 continue;
886
887 if (sa->sa_reg_attr_obj)
888 error = zap_lookup(os, sa->sa_reg_attr_obj,
889 reg_attrs[i].sa_name, 8, 1, &attr_value);
890 else
891 error = SET_ERROR(ENOENT);
892 switch (error) {
893 case ENOENT:
894 sa->sa_user_table[i] = (sa_attr_type_t)sa_attr_count;
895 sa_attr_count++;
896 break;
897 case 0:
898 sa->sa_user_table[i] = ATTR_NUM(attr_value);
899 break;
900 default:
901 goto bail;
902 }
903 }
904
905 sa->sa_num_attrs = sa_attr_count;
906 tb = sa->sa_attr_table =
907 kmem_zalloc(sizeof (sa_attr_table_t) * sa_attr_count, KM_SLEEP);
908
909 /*
910 * Attribute table is constructed from requested attribute list,
911 * previously foreign registered attributes, and also the legacy
912 * ZPL set of attributes.
913 */
914
915 if (sa->sa_reg_attr_obj) {
916 for (zap_cursor_init(&zc, os, sa->sa_reg_attr_obj);
917 (error = zap_cursor_retrieve(&zc, &za)) == 0;
918 zap_cursor_advance(&zc)) {
919 uint64_t value;
920 value = za.za_first_integer;
921
922 registered_count++;
923 tb[ATTR_NUM(value)].sa_attr = ATTR_NUM(value);
924 tb[ATTR_NUM(value)].sa_length = ATTR_LENGTH(value);
925 tb[ATTR_NUM(value)].sa_byteswap = ATTR_BSWAP(value);
926 tb[ATTR_NUM(value)].sa_registered = B_TRUE;
927
928 if (tb[ATTR_NUM(value)].sa_name) {
929 continue;
930 }
931 tb[ATTR_NUM(value)].sa_name =
932 kmem_zalloc(strlen(za.za_name) +1, KM_SLEEP);
933 (void) strlcpy(tb[ATTR_NUM(value)].sa_name, za.za_name,
934 strlen(za.za_name) +1);
935 }
936 zap_cursor_fini(&zc);
937 /*
938 * Make sure we processed the correct number of registered
939 * attributes
940 */
941 if (registered_count != sa_reg_count) {
942 ASSERT(error != 0);
943 goto bail;
944 }
945
946 }
947
948 if (ostype == DMU_OST_ZFS) {
949 for (i = 0; i != sa_legacy_attr_count; i++) {
950 if (tb[i].sa_name)
951 continue;
952 tb[i].sa_attr = sa_legacy_attrs[i].sa_attr;
953 tb[i].sa_length = sa_legacy_attrs[i].sa_length;
954 tb[i].sa_byteswap = sa_legacy_attrs[i].sa_byteswap;
955 tb[i].sa_registered = B_FALSE;
956 tb[i].sa_name =
957 kmem_zalloc(strlen(sa_legacy_attrs[i].sa_name) +1,
958 KM_SLEEP);
959 (void) strlcpy(tb[i].sa_name,
960 sa_legacy_attrs[i].sa_name,
961 strlen(sa_legacy_attrs[i].sa_name) + 1);
962 }
963 }
964
965 for (i = 0; i != count; i++) {
966 sa_attr_type_t attr_id;
967
968 attr_id = sa->sa_user_table[i];
969 if (tb[attr_id].sa_name)
970 continue;
971
972 tb[attr_id].sa_length = reg_attrs[i].sa_length;
973 tb[attr_id].sa_byteswap = reg_attrs[i].sa_byteswap;
974 tb[attr_id].sa_attr = attr_id;
975 tb[attr_id].sa_name =
976 kmem_zalloc(strlen(reg_attrs[i].sa_name) + 1, KM_SLEEP);
977 (void) strlcpy(tb[attr_id].sa_name, reg_attrs[i].sa_name,
978 strlen(reg_attrs[i].sa_name) + 1);
979 }
980
981 sa->sa_need_attr_registration =
982 (sa_attr_count != registered_count);
983
984 return (0);
985 bail:
986 kmem_free(sa->sa_user_table, count * sizeof (sa_attr_type_t));
987 sa->sa_user_table = NULL;
988 sa_free_attr_table(sa);
989 ASSERT(error != 0);
990 return (error);
991 }
992
993 int
sa_setup(objset_t * os,uint64_t sa_obj,const sa_attr_reg_t * reg_attrs,int count,sa_attr_type_t ** user_table)994 sa_setup(objset_t *os, uint64_t sa_obj, const sa_attr_reg_t *reg_attrs,
995 int count, sa_attr_type_t **user_table)
996 {
997 zap_cursor_t zc;
998 zap_attribute_t za;
999 sa_os_t *sa;
1000 dmu_objset_type_t ostype = dmu_objset_type(os);
1001 sa_attr_type_t *tb;
1002 int error;
1003
1004 mutex_enter(&os->os_user_ptr_lock);
1005 if (os->os_sa) {
1006 mutex_enter(&os->os_sa->sa_lock);
1007 mutex_exit(&os->os_user_ptr_lock);
1008 tb = os->os_sa->sa_user_table;
1009 mutex_exit(&os->os_sa->sa_lock);
1010 *user_table = tb;
1011 return (0);
1012 }
1013
1014 sa = kmem_zalloc(sizeof (sa_os_t), KM_SLEEP);
1015 mutex_init(&sa->sa_lock, NULL, MUTEX_NOLOCKDEP, NULL);
1016 sa->sa_master_obj = sa_obj;
1017
1018 os->os_sa = sa;
1019 mutex_enter(&sa->sa_lock);
1020 mutex_exit(&os->os_user_ptr_lock);
1021 avl_create(&sa->sa_layout_num_tree, layout_num_compare,
1022 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node));
1023 avl_create(&sa->sa_layout_hash_tree, layout_hash_compare,
1024 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_hash_node));
1025
1026 if (sa_obj) {
1027 error = zap_lookup(os, sa_obj, SA_LAYOUTS,
1028 8, 1, &sa->sa_layout_attr_obj);
1029 if (error != 0 && error != ENOENT)
1030 goto fail;
1031 error = zap_lookup(os, sa_obj, SA_REGISTRY,
1032 8, 1, &sa->sa_reg_attr_obj);
1033 if (error != 0 && error != ENOENT)
1034 goto fail;
1035 }
1036
1037 if ((error = sa_attr_table_setup(os, reg_attrs, count)) != 0)
1038 goto fail;
1039
1040 if (sa->sa_layout_attr_obj != 0) {
1041 uint64_t layout_count;
1042
1043 error = zap_count(os, sa->sa_layout_attr_obj,
1044 &layout_count);
1045
1046 /*
1047 * Layout number count should be > 0
1048 */
1049 if (error || (error == 0 && layout_count == 0)) {
1050 if (error == 0)
1051 error = SET_ERROR(EINVAL);
1052 goto fail;
1053 }
1054
1055 for (zap_cursor_init(&zc, os, sa->sa_layout_attr_obj);
1056 (error = zap_cursor_retrieve(&zc, &za)) == 0;
1057 zap_cursor_advance(&zc)) {
1058 sa_attr_type_t *lot_attrs;
1059 uint64_t lot_num;
1060
1061 lot_attrs = kmem_zalloc(sizeof (sa_attr_type_t) *
1062 za.za_num_integers, KM_SLEEP);
1063
1064 if ((error = (zap_lookup(os, sa->sa_layout_attr_obj,
1065 za.za_name, 2, za.za_num_integers,
1066 lot_attrs))) != 0) {
1067 kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
1068 za.za_num_integers);
1069 break;
1070 }
1071 VERIFY0(ddi_strtoull(za.za_name, NULL, 10,
1072 (unsigned long long *)&lot_num));
1073
1074 (void) sa_add_layout_entry(os, lot_attrs,
1075 za.za_num_integers, lot_num,
1076 sa_layout_info_hash(lot_attrs,
1077 za.za_num_integers), B_FALSE, NULL);
1078 kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
1079 za.za_num_integers);
1080 }
1081 zap_cursor_fini(&zc);
1082
1083 /*
1084 * Make sure layout count matches number of entries added
1085 * to AVL tree
1086 */
1087 if (avl_numnodes(&sa->sa_layout_num_tree) != layout_count) {
1088 ASSERT(error != 0);
1089 goto fail;
1090 }
1091 }
1092
1093 /* Add special layout number for old ZNODES */
1094 if (ostype == DMU_OST_ZFS) {
1095 (void) sa_add_layout_entry(os, sa_legacy_zpl_layout,
1096 sa_legacy_attr_count, 0,
1097 sa_layout_info_hash(sa_legacy_zpl_layout,
1098 sa_legacy_attr_count), B_FALSE, NULL);
1099
1100 (void) sa_add_layout_entry(os, sa_dummy_zpl_layout, 0, 1,
1101 0, B_FALSE, NULL);
1102 }
1103 *user_table = os->os_sa->sa_user_table;
1104 mutex_exit(&sa->sa_lock);
1105 return (0);
1106 fail:
1107 os->os_sa = NULL;
1108 sa_free_attr_table(sa);
1109 if (sa->sa_user_table)
1110 kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
1111 mutex_exit(&sa->sa_lock);
1112 avl_destroy(&sa->sa_layout_hash_tree);
1113 avl_destroy(&sa->sa_layout_num_tree);
1114 mutex_destroy(&sa->sa_lock);
1115 kmem_free(sa, sizeof (sa_os_t));
1116 return ((error == ECKSUM) ? EIO : error);
1117 }
1118
1119 void
sa_tear_down(objset_t * os)1120 sa_tear_down(objset_t *os)
1121 {
1122 sa_os_t *sa = os->os_sa;
1123 sa_lot_t *layout;
1124 void *cookie;
1125
1126 kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
1127
1128 /* Free up attr table */
1129
1130 sa_free_attr_table(sa);
1131
1132 cookie = NULL;
1133 while ((layout =
1134 avl_destroy_nodes(&sa->sa_layout_hash_tree, &cookie))) {
1135 sa_idx_tab_t *tab;
1136 while ((tab = list_head(&layout->lot_idx_tab))) {
1137 ASSERT(zfs_refcount_count(&tab->sa_refcount));
1138 sa_idx_tab_rele(os, tab);
1139 }
1140 }
1141
1142 cookie = NULL;
1143 while ((layout = avl_destroy_nodes(&sa->sa_layout_num_tree, &cookie))) {
1144 kmem_free(layout->lot_attrs,
1145 sizeof (sa_attr_type_t) * layout->lot_attr_count);
1146 kmem_free(layout, sizeof (sa_lot_t));
1147 }
1148
1149 avl_destroy(&sa->sa_layout_hash_tree);
1150 avl_destroy(&sa->sa_layout_num_tree);
1151 mutex_destroy(&sa->sa_lock);
1152
1153 kmem_free(sa, sizeof (sa_os_t));
1154 os->os_sa = NULL;
1155 }
1156
1157 static void
sa_build_idx_tab(void * hdr,void * attr_addr,sa_attr_type_t attr,uint16_t length,int length_idx,boolean_t var_length,void * userp)1158 sa_build_idx_tab(void *hdr, void *attr_addr, sa_attr_type_t attr,
1159 uint16_t length, int length_idx, boolean_t var_length, void *userp)
1160 {
1161 sa_idx_tab_t *idx_tab = userp;
1162
1163 if (var_length) {
1164 ASSERT(idx_tab->sa_variable_lengths);
1165 idx_tab->sa_variable_lengths[length_idx] = length;
1166 }
1167 TOC_ATTR_ENCODE(idx_tab->sa_idx_tab[attr], length_idx,
1168 (uint32_t)((uintptr_t)attr_addr - (uintptr_t)hdr));
1169 }
1170
1171 static void
sa_attr_iter(objset_t * os,sa_hdr_phys_t * hdr,dmu_object_type_t type,sa_iterfunc_t func,sa_lot_t * tab,void * userp)1172 sa_attr_iter(objset_t *os, sa_hdr_phys_t *hdr, dmu_object_type_t type,
1173 sa_iterfunc_t func, sa_lot_t *tab, void *userp)
1174 {
1175 void *data_start;
1176 sa_lot_t *tb = tab;
1177 sa_lot_t search;
1178 avl_index_t loc;
1179 sa_os_t *sa = os->os_sa;
1180 int i;
1181 uint16_t *length_start = NULL;
1182 uint8_t length_idx = 0;
1183
1184 if (tab == NULL) {
1185 search.lot_num = SA_LAYOUT_NUM(hdr, type);
1186 tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
1187 ASSERT(tb);
1188 }
1189
1190 if (IS_SA_BONUSTYPE(type)) {
1191 data_start = (void *)P2ROUNDUP(((uintptr_t)hdr +
1192 offsetof(sa_hdr_phys_t, sa_lengths) +
1193 (sizeof (uint16_t) * tb->lot_var_sizes)), 8);
1194 length_start = hdr->sa_lengths;
1195 } else {
1196 data_start = hdr;
1197 }
1198
1199 for (i = 0; i != tb->lot_attr_count; i++) {
1200 int attr_length, reg_length;
1201 uint8_t idx_len;
1202
1203 reg_length = sa->sa_attr_table[tb->lot_attrs[i]].sa_length;
1204 IMPLY(reg_length == 0, IS_SA_BONUSTYPE(type));
1205 if (reg_length) {
1206 attr_length = reg_length;
1207 idx_len = 0;
1208 } else {
1209 attr_length = length_start[length_idx];
1210 idx_len = length_idx++;
1211 }
1212
1213 func(hdr, data_start, tb->lot_attrs[i], attr_length,
1214 idx_len, reg_length == 0 ? B_TRUE : B_FALSE, userp);
1215
1216 data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
1217 attr_length), 8);
1218 }
1219 }
1220
1221 static void
sa_byteswap_cb(void * hdr,void * attr_addr,sa_attr_type_t attr,uint16_t length,int length_idx,boolean_t variable_length,void * userp)1222 sa_byteswap_cb(void *hdr, void *attr_addr, sa_attr_type_t attr,
1223 uint16_t length, int length_idx, boolean_t variable_length, void *userp)
1224 {
1225 (void) hdr, (void) length_idx, (void) variable_length;
1226 sa_handle_t *hdl = userp;
1227 sa_os_t *sa = hdl->sa_os->os_sa;
1228
1229 sa_bswap_table[sa->sa_attr_table[attr].sa_byteswap](attr_addr, length);
1230 }
1231
1232 static void
sa_byteswap(sa_handle_t * hdl,sa_buf_type_t buftype)1233 sa_byteswap(sa_handle_t *hdl, sa_buf_type_t buftype)
1234 {
1235 sa_hdr_phys_t *sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1236 dmu_buf_impl_t *db;
1237 int num_lengths = 1;
1238 int i;
1239 sa_os_t *sa __maybe_unused = hdl->sa_os->os_sa;
1240
1241 ASSERT(MUTEX_HELD(&sa->sa_lock));
1242 if (sa_hdr_phys->sa_magic == SA_MAGIC)
1243 return;
1244
1245 db = SA_GET_DB(hdl, buftype);
1246
1247 if (buftype == SA_SPILL) {
1248 arc_release(db->db_buf, NULL);
1249 arc_buf_thaw(db->db_buf);
1250 }
1251
1252 sa_hdr_phys->sa_magic = BSWAP_32(sa_hdr_phys->sa_magic);
1253 sa_hdr_phys->sa_layout_info = BSWAP_16(sa_hdr_phys->sa_layout_info);
1254
1255 /*
1256 * Determine number of variable lengths in header
1257 * The standard 8 byte header has one for free and a
1258 * 16 byte header would have 4 + 1;
1259 */
1260 if (SA_HDR_SIZE(sa_hdr_phys) > 8)
1261 num_lengths += (SA_HDR_SIZE(sa_hdr_phys) - 8) >> 1;
1262 for (i = 0; i != num_lengths; i++)
1263 sa_hdr_phys->sa_lengths[i] =
1264 BSWAP_16(sa_hdr_phys->sa_lengths[i]);
1265
1266 sa_attr_iter(hdl->sa_os, sa_hdr_phys, DMU_OT_SA,
1267 sa_byteswap_cb, NULL, hdl);
1268
1269 if (buftype == SA_SPILL)
1270 arc_buf_freeze(((dmu_buf_impl_t *)hdl->sa_spill)->db_buf);
1271 }
1272
1273 static int
sa_build_index(sa_handle_t * hdl,sa_buf_type_t buftype)1274 sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype)
1275 {
1276 sa_hdr_phys_t *sa_hdr_phys;
1277 dmu_buf_impl_t *db = SA_GET_DB(hdl, buftype);
1278 dmu_object_type_t bonustype = SA_BONUSTYPE_FROM_DB(db);
1279 sa_os_t *sa = hdl->sa_os->os_sa;
1280 sa_idx_tab_t *idx_tab;
1281
1282 sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1283
1284 mutex_enter(&sa->sa_lock);
1285
1286 /* Do we need to byteswap? */
1287
1288 /* only check if not old znode */
1289 if (IS_SA_BONUSTYPE(bonustype) && sa_hdr_phys->sa_magic != SA_MAGIC &&
1290 sa_hdr_phys->sa_magic != 0) {
1291 if (BSWAP_32(sa_hdr_phys->sa_magic) != SA_MAGIC) {
1292 mutex_exit(&sa->sa_lock);
1293 zfs_dbgmsg("Buffer Header: %x != SA_MAGIC:%x "
1294 "object=%#llx\n", sa_hdr_phys->sa_magic, SA_MAGIC,
1295 (u_longlong_t)db->db.db_object);
1296 return (SET_ERROR(EIO));
1297 }
1298 sa_byteswap(hdl, buftype);
1299 }
1300
1301 idx_tab = sa_find_idx_tab(hdl->sa_os, bonustype, sa_hdr_phys);
1302
1303 if (buftype == SA_BONUS)
1304 hdl->sa_bonus_tab = idx_tab;
1305 else
1306 hdl->sa_spill_tab = idx_tab;
1307
1308 mutex_exit(&sa->sa_lock);
1309 return (0);
1310 }
1311
1312 static void
sa_evict_sync(void * dbu)1313 sa_evict_sync(void *dbu)
1314 {
1315 (void) dbu;
1316 panic("evicting sa dbuf\n");
1317 }
1318
1319 static void
sa_idx_tab_rele(objset_t * os,void * arg)1320 sa_idx_tab_rele(objset_t *os, void *arg)
1321 {
1322 sa_os_t *sa = os->os_sa;
1323 sa_idx_tab_t *idx_tab = arg;
1324
1325 if (idx_tab == NULL)
1326 return;
1327
1328 mutex_enter(&sa->sa_lock);
1329 if (zfs_refcount_remove(&idx_tab->sa_refcount, NULL) == 0) {
1330 list_remove(&idx_tab->sa_layout->lot_idx_tab, idx_tab);
1331 if (idx_tab->sa_variable_lengths)
1332 kmem_free(idx_tab->sa_variable_lengths,
1333 sizeof (uint16_t) *
1334 idx_tab->sa_layout->lot_var_sizes);
1335 zfs_refcount_destroy(&idx_tab->sa_refcount);
1336 kmem_free(idx_tab->sa_idx_tab,
1337 sizeof (uint32_t) * sa->sa_num_attrs);
1338 kmem_free(idx_tab, sizeof (sa_idx_tab_t));
1339 }
1340 mutex_exit(&sa->sa_lock);
1341 }
1342
1343 static void
sa_idx_tab_hold(objset_t * os,sa_idx_tab_t * idx_tab)1344 sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab)
1345 {
1346 sa_os_t *sa __maybe_unused = os->os_sa;
1347
1348 ASSERT(MUTEX_HELD(&sa->sa_lock));
1349 (void) zfs_refcount_add(&idx_tab->sa_refcount, NULL);
1350 }
1351
1352 void
sa_spill_rele(sa_handle_t * hdl)1353 sa_spill_rele(sa_handle_t *hdl)
1354 {
1355 mutex_enter(&hdl->sa_lock);
1356 if (hdl->sa_spill) {
1357 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
1358 dmu_buf_rele(hdl->sa_spill, NULL);
1359 hdl->sa_spill = NULL;
1360 hdl->sa_spill_tab = NULL;
1361 }
1362 mutex_exit(&hdl->sa_lock);
1363 }
1364
1365 void
sa_handle_destroy(sa_handle_t * hdl)1366 sa_handle_destroy(sa_handle_t *hdl)
1367 {
1368 dmu_buf_t *db = hdl->sa_bonus;
1369
1370 mutex_enter(&hdl->sa_lock);
1371 (void) dmu_buf_remove_user(db, &hdl->sa_dbu);
1372
1373 if (hdl->sa_bonus_tab)
1374 sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
1375
1376 if (hdl->sa_spill_tab)
1377 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
1378
1379 dmu_buf_rele(hdl->sa_bonus, NULL);
1380
1381 if (hdl->sa_spill)
1382 dmu_buf_rele(hdl->sa_spill, NULL);
1383 mutex_exit(&hdl->sa_lock);
1384
1385 kmem_cache_free(sa_cache, hdl);
1386 }
1387
1388 int
sa_handle_get_from_db(objset_t * os,dmu_buf_t * db,void * userp,sa_handle_type_t hdl_type,sa_handle_t ** handlepp)1389 sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
1390 sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1391 {
1392 int error = 0;
1393 sa_handle_t *handle = NULL;
1394 #ifdef ZFS_DEBUG
1395 dmu_object_info_t doi;
1396
1397 dmu_object_info_from_db(db, &doi);
1398 ASSERT(doi.doi_bonus_type == DMU_OT_SA ||
1399 doi.doi_bonus_type == DMU_OT_ZNODE);
1400 #endif
1401 /* find handle, if it exists */
1402 /* if one doesn't exist then create a new one, and initialize it */
1403
1404 if (hdl_type == SA_HDL_SHARED)
1405 handle = dmu_buf_get_user(db);
1406
1407 if (handle == NULL) {
1408 sa_handle_t *winner = NULL;
1409
1410 handle = kmem_cache_alloc(sa_cache, KM_SLEEP);
1411 handle->sa_dbu.dbu_evict_func_sync = NULL;
1412 handle->sa_dbu.dbu_evict_func_async = NULL;
1413 handle->sa_userp = userp;
1414 handle->sa_bonus = db;
1415 handle->sa_os = os;
1416 handle->sa_spill = NULL;
1417 handle->sa_bonus_tab = NULL;
1418 handle->sa_spill_tab = NULL;
1419
1420 error = sa_build_index(handle, SA_BONUS);
1421
1422 if (hdl_type == SA_HDL_SHARED) {
1423 dmu_buf_init_user(&handle->sa_dbu, sa_evict_sync, NULL,
1424 NULL);
1425 winner = dmu_buf_set_user_ie(db, &handle->sa_dbu);
1426 }
1427
1428 if (winner != NULL) {
1429 kmem_cache_free(sa_cache, handle);
1430 handle = winner;
1431 }
1432 }
1433 *handlepp = handle;
1434
1435 return (error);
1436 }
1437
1438 int
sa_handle_get(objset_t * objset,uint64_t objid,void * userp,sa_handle_type_t hdl_type,sa_handle_t ** handlepp)1439 sa_handle_get(objset_t *objset, uint64_t objid, void *userp,
1440 sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1441 {
1442 dmu_buf_t *db;
1443 int error;
1444
1445 if ((error = dmu_bonus_hold(objset, objid, NULL, &db)))
1446 return (error);
1447
1448 return (sa_handle_get_from_db(objset, db, userp, hdl_type,
1449 handlepp));
1450 }
1451
1452 int
sa_buf_hold(objset_t * objset,uint64_t obj_num,const void * tag,dmu_buf_t ** db)1453 sa_buf_hold(objset_t *objset, uint64_t obj_num, const void *tag, dmu_buf_t **db)
1454 {
1455 return (dmu_bonus_hold(objset, obj_num, tag, db));
1456 }
1457
1458 void
sa_buf_rele(dmu_buf_t * db,const void * tag)1459 sa_buf_rele(dmu_buf_t *db, const void *tag)
1460 {
1461 dmu_buf_rele(db, tag);
1462 }
1463
1464 static int
sa_lookup_impl(sa_handle_t * hdl,sa_bulk_attr_t * bulk,int count)1465 sa_lookup_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count)
1466 {
1467 ASSERT(hdl);
1468 ASSERT(MUTEX_HELD(&hdl->sa_lock));
1469 return (sa_attr_op(hdl, bulk, count, SA_LOOKUP, NULL));
1470 }
1471
1472 static int
sa_lookup_locked(sa_handle_t * hdl,sa_attr_type_t attr,void * buf,uint32_t buflen)1473 sa_lookup_locked(sa_handle_t *hdl, sa_attr_type_t attr, void *buf,
1474 uint32_t buflen)
1475 {
1476 int error;
1477 sa_bulk_attr_t bulk;
1478
1479 VERIFY3U(buflen, <=, SA_ATTR_MAX_LEN);
1480
1481 bulk.sa_attr = attr;
1482 bulk.sa_data = buf;
1483 bulk.sa_length = buflen;
1484 bulk.sa_data_func = NULL;
1485
1486 ASSERT(hdl);
1487 error = sa_lookup_impl(hdl, &bulk, 1);
1488 return (error);
1489 }
1490
1491 int
sa_lookup(sa_handle_t * hdl,sa_attr_type_t attr,void * buf,uint32_t buflen)1492 sa_lookup(sa_handle_t *hdl, sa_attr_type_t attr, void *buf, uint32_t buflen)
1493 {
1494 int error;
1495
1496 mutex_enter(&hdl->sa_lock);
1497 error = sa_lookup_locked(hdl, attr, buf, buflen);
1498 mutex_exit(&hdl->sa_lock);
1499
1500 return (error);
1501 }
1502
1503 #ifdef _KERNEL
1504 int
sa_lookup_uio(sa_handle_t * hdl,sa_attr_type_t attr,zfs_uio_t * uio)1505 sa_lookup_uio(sa_handle_t *hdl, sa_attr_type_t attr, zfs_uio_t *uio)
1506 {
1507 int error;
1508 sa_bulk_attr_t bulk;
1509
1510 bulk.sa_data = NULL;
1511 bulk.sa_attr = attr;
1512 bulk.sa_data_func = NULL;
1513
1514 ASSERT(hdl);
1515
1516 mutex_enter(&hdl->sa_lock);
1517 if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) == 0) {
1518 error = zfs_uiomove((void *)bulk.sa_addr, MIN(bulk.sa_size,
1519 zfs_uio_resid(uio)), UIO_READ, uio);
1520 }
1521 mutex_exit(&hdl->sa_lock);
1522 return (error);
1523 }
1524
1525 /*
1526 * For the existed object that is upgraded from old system, its ondisk layout
1527 * has no slot for the project ID attribute. But quota accounting logic needs
1528 * to access related slots by offset directly. So we need to adjust these old
1529 * objects' layout to make the project ID to some unified and fixed offset.
1530 */
1531 int
sa_add_projid(sa_handle_t * hdl,dmu_tx_t * tx,uint64_t projid)1532 sa_add_projid(sa_handle_t *hdl, dmu_tx_t *tx, uint64_t projid)
1533 {
1534 znode_t *zp = sa_get_userdata(hdl);
1535 dmu_buf_t *db = sa_get_db(hdl);
1536 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1537 int count = 0, err = 0;
1538 sa_bulk_attr_t *bulk, *attrs;
1539 zfs_acl_locator_cb_t locate = { 0 };
1540 uint64_t uid, gid, mode, rdev, xattr = 0, parent, gen, links;
1541 uint64_t crtime[2], mtime[2], ctime[2], atime[2];
1542 zfs_acl_phys_t znode_acl = { 0 };
1543 char scanstamp[AV_SCANSTAMP_SZ];
1544
1545 if (zp->z_acl_cached == NULL) {
1546 zfs_acl_t *aclp;
1547
1548 mutex_enter(&zp->z_acl_lock);
1549 err = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1550 mutex_exit(&zp->z_acl_lock);
1551 if (err != 0 && err != ENOENT)
1552 return (err);
1553 }
1554
1555 bulk = kmem_zalloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
1556 attrs = kmem_zalloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
1557 mutex_enter(&hdl->sa_lock);
1558 mutex_enter(&zp->z_lock);
1559
1560 err = sa_lookup_locked(hdl, SA_ZPL_PROJID(zfsvfs), &projid,
1561 sizeof (uint64_t));
1562 if (unlikely(err == 0))
1563 /* Someone has added project ID attr by race. */
1564 err = EEXIST;
1565 if (err != ENOENT)
1566 goto out;
1567
1568 /* First do a bulk query of the attributes that aren't cached */
1569 if (zp->z_is_sa) {
1570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1571 &mode, 8);
1572 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1573 &gen, 8);
1574 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1575 &uid, 8);
1576 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1577 &gid, 8);
1578 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
1579 &parent, 8);
1580 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1581 &atime, 16);
1582 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1583 &mtime, 16);
1584 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1585 &ctime, 16);
1586 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1587 &crtime, 16);
1588 if (Z_ISBLK(ZTOTYPE(zp)) || Z_ISCHR(ZTOTYPE(zp)))
1589 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1590 &rdev, 8);
1591 } else {
1592 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1593 &atime, 16);
1594 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1595 &mtime, 16);
1596 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1597 &ctime, 16);
1598 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1599 &crtime, 16);
1600 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1601 &gen, 8);
1602 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1603 &mode, 8);
1604 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
1605 &parent, 8);
1606 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL,
1607 &xattr, 8);
1608 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1609 &rdev, 8);
1610 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1611 &uid, 8);
1612 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1613 &gid, 8);
1614 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1615 &znode_acl, 88);
1616 }
1617 err = sa_bulk_lookup_locked(hdl, bulk, count);
1618 if (err != 0)
1619 goto out;
1620
1621 err = sa_lookup_locked(hdl, SA_ZPL_XATTR(zfsvfs), &xattr, 8);
1622 if (err != 0 && err != ENOENT)
1623 goto out;
1624
1625 zp->z_projid = projid;
1626 zp->z_pflags |= ZFS_PROJID;
1627 links = ZTONLNK(zp);
1628 count = 0;
1629 err = 0;
1630
1631 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
1632 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
1633 &zp->z_size, 8);
1634 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_GEN(zfsvfs), NULL, &gen, 8);
1635 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
1636 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
1637 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
1638 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1639 &zp->z_pflags, 8);
1640 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
1641 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
1642 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
1643 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1644 &crtime, 16);
1645 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
1646 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_PROJID(zfsvfs), NULL, &projid, 8);
1647
1648 if (Z_ISBLK(ZTOTYPE(zp)) || Z_ISCHR(ZTOTYPE(zp)))
1649 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
1650 &rdev, 8);
1651
1652 if (zp->z_acl_cached != NULL) {
1653 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
1654 &zp->z_acl_cached->z_acl_count, 8);
1655 if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
1656 zfs_acl_xform(zp, zp->z_acl_cached, CRED());
1657 locate.cb_aclp = zp->z_acl_cached;
1658 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
1659 zfs_acl_data_locator, &locate,
1660 zp->z_acl_cached->z_acl_bytes);
1661 }
1662
1663 if (xattr)
1664 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_XATTR(zfsvfs), NULL,
1665 &xattr, 8);
1666
1667 if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
1668 memcpy(scanstamp,
1669 (caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
1670 AV_SCANSTAMP_SZ);
1671 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_SCANSTAMP(zfsvfs), NULL,
1672 scanstamp, AV_SCANSTAMP_SZ);
1673 zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
1674 }
1675
1676 VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
1677 VERIFY(sa_replace_all_by_template_locked(hdl, attrs, count, tx) == 0);
1678 if (znode_acl.z_acl_extern_obj) {
1679 VERIFY(0 == dmu_object_free(zfsvfs->z_os,
1680 znode_acl.z_acl_extern_obj, tx));
1681 }
1682
1683 zp->z_is_sa = B_TRUE;
1684
1685 out:
1686 mutex_exit(&zp->z_lock);
1687 mutex_exit(&hdl->sa_lock);
1688 kmem_free(attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
1689 kmem_free(bulk, sizeof (sa_bulk_attr_t) * ZPL_END);
1690 return (err);
1691 }
1692 #endif
1693
1694 static sa_idx_tab_t *
sa_find_idx_tab(objset_t * os,dmu_object_type_t bonustype,sa_hdr_phys_t * hdr)1695 sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype, sa_hdr_phys_t *hdr)
1696 {
1697 sa_idx_tab_t *idx_tab;
1698 sa_os_t *sa = os->os_sa;
1699 sa_lot_t *tb, search;
1700 avl_index_t loc;
1701
1702 /*
1703 * Deterimine layout number. If SA node and header == 0 then
1704 * force the index table to the dummy "1" empty layout.
1705 *
1706 * The layout number would only be zero for a newly created file
1707 * that has not added any attributes yet, or with crypto enabled which
1708 * doesn't write any attributes to the bonus buffer.
1709 */
1710
1711 search.lot_num = SA_LAYOUT_NUM(hdr, bonustype);
1712
1713 tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
1714
1715 /* Verify header size is consistent with layout information */
1716 ASSERT(tb);
1717 ASSERT((IS_SA_BONUSTYPE(bonustype) &&
1718 SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb)) || !IS_SA_BONUSTYPE(bonustype) ||
1719 (IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0));
1720
1721 /*
1722 * See if any of the already existing TOC entries can be reused?
1723 */
1724
1725 for (idx_tab = list_head(&tb->lot_idx_tab); idx_tab;
1726 idx_tab = list_next(&tb->lot_idx_tab, idx_tab)) {
1727 boolean_t valid_idx = B_TRUE;
1728 int i;
1729
1730 if (tb->lot_var_sizes != 0 &&
1731 idx_tab->sa_variable_lengths != NULL) {
1732 for (i = 0; i != tb->lot_var_sizes; i++) {
1733 if (hdr->sa_lengths[i] !=
1734 idx_tab->sa_variable_lengths[i]) {
1735 valid_idx = B_FALSE;
1736 break;
1737 }
1738 }
1739 }
1740 if (valid_idx) {
1741 sa_idx_tab_hold(os, idx_tab);
1742 return (idx_tab);
1743 }
1744 }
1745
1746 /* No such luck, create a new entry */
1747 idx_tab = kmem_zalloc(sizeof (sa_idx_tab_t), KM_SLEEP);
1748 idx_tab->sa_idx_tab =
1749 kmem_zalloc(sizeof (uint32_t) * sa->sa_num_attrs, KM_SLEEP);
1750 idx_tab->sa_layout = tb;
1751 zfs_refcount_create(&idx_tab->sa_refcount);
1752 if (tb->lot_var_sizes)
1753 idx_tab->sa_variable_lengths = kmem_alloc(sizeof (uint16_t) *
1754 tb->lot_var_sizes, KM_SLEEP);
1755
1756 sa_attr_iter(os, hdr, bonustype, sa_build_idx_tab,
1757 tb, idx_tab);
1758 sa_idx_tab_hold(os, idx_tab); /* one hold for consumer */
1759 sa_idx_tab_hold(os, idx_tab); /* one for layout */
1760 list_insert_tail(&tb->lot_idx_tab, idx_tab);
1761 return (idx_tab);
1762 }
1763
1764 void
sa_default_locator(void ** dataptr,uint32_t * len,uint32_t total_len,boolean_t start,void * userdata)1765 sa_default_locator(void **dataptr, uint32_t *len, uint32_t total_len,
1766 boolean_t start, void *userdata)
1767 {
1768 ASSERT(start);
1769
1770 *dataptr = userdata;
1771 *len = total_len;
1772 }
1773
1774 static void
sa_attr_register_sync(sa_handle_t * hdl,dmu_tx_t * tx)1775 sa_attr_register_sync(sa_handle_t *hdl, dmu_tx_t *tx)
1776 {
1777 uint64_t attr_value = 0;
1778 sa_os_t *sa = hdl->sa_os->os_sa;
1779 sa_attr_table_t *tb = sa->sa_attr_table;
1780 int i;
1781
1782 mutex_enter(&sa->sa_lock);
1783
1784 if (!sa->sa_need_attr_registration || sa->sa_master_obj == 0) {
1785 mutex_exit(&sa->sa_lock);
1786 return;
1787 }
1788
1789 if (sa->sa_reg_attr_obj == 0) {
1790 sa->sa_reg_attr_obj = zap_create_link(hdl->sa_os,
1791 DMU_OT_SA_ATTR_REGISTRATION,
1792 sa->sa_master_obj, SA_REGISTRY, tx);
1793 }
1794 for (i = 0; i != sa->sa_num_attrs; i++) {
1795 if (sa->sa_attr_table[i].sa_registered)
1796 continue;
1797 ATTR_ENCODE(attr_value, tb[i].sa_attr, tb[i].sa_length,
1798 tb[i].sa_byteswap);
1799 VERIFY(0 == zap_update(hdl->sa_os, sa->sa_reg_attr_obj,
1800 tb[i].sa_name, 8, 1, &attr_value, tx));
1801 tb[i].sa_registered = B_TRUE;
1802 }
1803 sa->sa_need_attr_registration = B_FALSE;
1804 mutex_exit(&sa->sa_lock);
1805 }
1806
1807 /*
1808 * Replace all attributes with attributes specified in template.
1809 * If dnode had a spill buffer then those attributes will be
1810 * also be replaced, possibly with just an empty spill block
1811 *
1812 * This interface is intended to only be used for bulk adding of
1813 * attributes for a new file. It will also be used by the ZPL
1814 * when converting and old formatted znode to native SA support.
1815 */
1816 int
sa_replace_all_by_template_locked(sa_handle_t * hdl,sa_bulk_attr_t * attr_desc,int attr_count,dmu_tx_t * tx)1817 sa_replace_all_by_template_locked(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
1818 int attr_count, dmu_tx_t *tx)
1819 {
1820 sa_os_t *sa = hdl->sa_os->os_sa;
1821
1822 if (sa->sa_need_attr_registration)
1823 sa_attr_register_sync(hdl, tx);
1824 return (sa_build_layouts(hdl, attr_desc, attr_count, tx));
1825 }
1826
1827 int
sa_replace_all_by_template(sa_handle_t * hdl,sa_bulk_attr_t * attr_desc,int attr_count,dmu_tx_t * tx)1828 sa_replace_all_by_template(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
1829 int attr_count, dmu_tx_t *tx)
1830 {
1831 int error;
1832
1833 mutex_enter(&hdl->sa_lock);
1834 error = sa_replace_all_by_template_locked(hdl, attr_desc,
1835 attr_count, tx);
1836 mutex_exit(&hdl->sa_lock);
1837 return (error);
1838 }
1839
1840 /*
1841 * Add/remove a single attribute or replace a variable-sized attribute value
1842 * with a value of a different size, and then rewrite the entire set
1843 * of attributes.
1844 * Same-length attribute value replacement (including fixed-length attributes)
1845 * is handled more efficiently by the upper layers.
1846 */
1847 static int
sa_modify_attrs(sa_handle_t * hdl,sa_attr_type_t newattr,sa_data_op_t action,sa_data_locator_t * locator,void * datastart,uint16_t buflen,dmu_tx_t * tx)1848 sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
1849 sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
1850 uint16_t buflen, dmu_tx_t *tx)
1851 {
1852 sa_os_t *sa = hdl->sa_os->os_sa;
1853 dmu_buf_impl_t *db = (dmu_buf_impl_t *)hdl->sa_bonus;
1854 dnode_t *dn;
1855 sa_bulk_attr_t *attr_desc;
1856 void *old_data[2];
1857 int bonus_attr_count = 0;
1858 int bonus_data_size = 0;
1859 int spill_data_size = 0;
1860 int spill_attr_count = 0;
1861 int error;
1862 uint16_t length, reg_length;
1863 int i, j, k, length_idx;
1864 sa_hdr_phys_t *hdr;
1865 sa_idx_tab_t *idx_tab;
1866 int attr_count;
1867 int count;
1868
1869 ASSERT(MUTEX_HELD(&hdl->sa_lock));
1870
1871 /* First make of copy of the old data */
1872
1873 DB_DNODE_ENTER(db);
1874 dn = DB_DNODE(db);
1875 if (dn->dn_bonuslen != 0) {
1876 bonus_data_size = hdl->sa_bonus->db_size;
1877 old_data[0] = kmem_alloc(bonus_data_size, KM_SLEEP);
1878 memcpy(old_data[0], hdl->sa_bonus->db_data,
1879 hdl->sa_bonus->db_size);
1880 bonus_attr_count = hdl->sa_bonus_tab->sa_layout->lot_attr_count;
1881 } else {
1882 old_data[0] = NULL;
1883 }
1884 DB_DNODE_EXIT(db);
1885
1886 /* Bring spill buffer online if it isn't currently */
1887
1888 if ((error = sa_get_spill(hdl)) == 0) {
1889 spill_data_size = hdl->sa_spill->db_size;
1890 old_data[1] = vmem_alloc(spill_data_size, KM_SLEEP);
1891 memcpy(old_data[1], hdl->sa_spill->db_data,
1892 hdl->sa_spill->db_size);
1893 spill_attr_count =
1894 hdl->sa_spill_tab->sa_layout->lot_attr_count;
1895 } else if (error && error != ENOENT) {
1896 if (old_data[0])
1897 kmem_free(old_data[0], bonus_data_size);
1898 return (error);
1899 } else {
1900 old_data[1] = NULL;
1901 }
1902
1903 /* build descriptor of all attributes */
1904
1905 attr_count = bonus_attr_count + spill_attr_count;
1906 if (action == SA_ADD)
1907 attr_count++;
1908 else if (action == SA_REMOVE)
1909 attr_count--;
1910
1911 attr_desc = kmem_zalloc(sizeof (sa_bulk_attr_t) * attr_count, KM_SLEEP);
1912
1913 /*
1914 * loop through bonus and spill buffer if it exists, and
1915 * build up new attr_descriptor to reset the attributes
1916 */
1917 k = j = 0;
1918 count = bonus_attr_count;
1919 hdr = SA_GET_HDR(hdl, SA_BONUS);
1920 idx_tab = SA_IDX_TAB_GET(hdl, SA_BONUS);
1921 for (; ; k++) {
1922 /*
1923 * Iterate over each attribute in layout. Fetch the
1924 * size of variable-length attributes needing rewrite
1925 * from sa_lengths[].
1926 */
1927 for (i = 0, length_idx = 0; i != count; i++) {
1928 sa_attr_type_t attr;
1929
1930 attr = idx_tab->sa_layout->lot_attrs[i];
1931 reg_length = SA_REGISTERED_LEN(sa, attr);
1932 if (reg_length == 0) {
1933 length = hdr->sa_lengths[length_idx];
1934 length_idx++;
1935 } else {
1936 length = reg_length;
1937 }
1938 if (attr == newattr) {
1939 /*
1940 * There is nothing to do for SA_REMOVE,
1941 * so it is just skipped.
1942 */
1943 if (action == SA_REMOVE)
1944 continue;
1945
1946 /*
1947 * Duplicate attributes are not allowed, so the
1948 * action can not be SA_ADD here.
1949 */
1950 ASSERT3S(action, ==, SA_REPLACE);
1951
1952 /*
1953 * Only a variable-sized attribute can be
1954 * replaced here, and its size must be changing.
1955 */
1956 ASSERT3U(reg_length, ==, 0);
1957 ASSERT3U(length, !=, buflen);
1958 SA_ADD_BULK_ATTR(attr_desc, j, attr,
1959 locator, datastart, buflen);
1960 } else {
1961 SA_ADD_BULK_ATTR(attr_desc, j, attr,
1962 NULL, (void *)
1963 (TOC_OFF(idx_tab->sa_idx_tab[attr]) +
1964 (uintptr_t)old_data[k]), length);
1965 }
1966 }
1967 if (k == 0 && hdl->sa_spill) {
1968 hdr = SA_GET_HDR(hdl, SA_SPILL);
1969 idx_tab = SA_IDX_TAB_GET(hdl, SA_SPILL);
1970 count = spill_attr_count;
1971 } else {
1972 break;
1973 }
1974 }
1975 if (action == SA_ADD) {
1976 reg_length = SA_REGISTERED_LEN(sa, newattr);
1977 IMPLY(reg_length != 0, reg_length == buflen);
1978 SA_ADD_BULK_ATTR(attr_desc, j, newattr, locator,
1979 datastart, buflen);
1980 }
1981 ASSERT3U(j, ==, attr_count);
1982
1983 error = sa_build_layouts(hdl, attr_desc, attr_count, tx);
1984
1985 if (old_data[0])
1986 kmem_free(old_data[0], bonus_data_size);
1987 if (old_data[1])
1988 vmem_free(old_data[1], spill_data_size);
1989 kmem_free(attr_desc, sizeof (sa_bulk_attr_t) * attr_count);
1990
1991 return (error);
1992 }
1993
1994 static int
sa_bulk_update_impl(sa_handle_t * hdl,sa_bulk_attr_t * bulk,int count,dmu_tx_t * tx)1995 sa_bulk_update_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
1996 dmu_tx_t *tx)
1997 {
1998 int error;
1999 sa_os_t *sa = hdl->sa_os->os_sa;
2000 dmu_object_type_t bonustype;
2001 dmu_buf_t *saved_spill;
2002
2003 ASSERT(hdl);
2004 ASSERT(MUTEX_HELD(&hdl->sa_lock));
2005
2006 bonustype = SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl, SA_BONUS));
2007 saved_spill = hdl->sa_spill;
2008
2009 /* sync out registration table if necessary */
2010 if (sa->sa_need_attr_registration)
2011 sa_attr_register_sync(hdl, tx);
2012
2013 error = sa_attr_op(hdl, bulk, count, SA_UPDATE, tx);
2014 if (error == 0 && !IS_SA_BONUSTYPE(bonustype) && sa->sa_update_cb)
2015 sa->sa_update_cb(hdl, tx);
2016
2017 /*
2018 * If saved_spill is NULL and current sa_spill is not NULL that
2019 * means we increased the refcount of the spill buffer through
2020 * sa_get_spill() or dmu_spill_hold_by_dnode(). Therefore we
2021 * must release the hold before calling dmu_tx_commit() to avoid
2022 * making a copy of this buffer in dbuf_sync_leaf() due to the
2023 * reference count now being greater than 1.
2024 */
2025 if (!saved_spill && hdl->sa_spill) {
2026 if (hdl->sa_spill_tab) {
2027 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
2028 hdl->sa_spill_tab = NULL;
2029 }
2030
2031 dmu_buf_rele(hdl->sa_spill, NULL);
2032 hdl->sa_spill = NULL;
2033 }
2034
2035 return (error);
2036 }
2037
2038 /*
2039 * update or add new attribute
2040 */
2041 int
sa_update(sa_handle_t * hdl,sa_attr_type_t type,void * buf,uint32_t buflen,dmu_tx_t * tx)2042 sa_update(sa_handle_t *hdl, sa_attr_type_t type,
2043 void *buf, uint32_t buflen, dmu_tx_t *tx)
2044 {
2045 int error;
2046 sa_bulk_attr_t bulk;
2047
2048 VERIFY3U(buflen, <=, SA_ATTR_MAX_LEN);
2049
2050 bulk.sa_attr = type;
2051 bulk.sa_data_func = NULL;
2052 bulk.sa_length = buflen;
2053 bulk.sa_data = buf;
2054
2055 mutex_enter(&hdl->sa_lock);
2056 error = sa_bulk_update_impl(hdl, &bulk, 1, tx);
2057 mutex_exit(&hdl->sa_lock);
2058 return (error);
2059 }
2060
2061 /*
2062 * Return size of an attribute
2063 */
2064
2065 int
sa_size(sa_handle_t * hdl,sa_attr_type_t attr,int * size)2066 sa_size(sa_handle_t *hdl, sa_attr_type_t attr, int *size)
2067 {
2068 sa_bulk_attr_t bulk;
2069 int error;
2070
2071 bulk.sa_data = NULL;
2072 bulk.sa_attr = attr;
2073 bulk.sa_data_func = NULL;
2074
2075 ASSERT(hdl);
2076 mutex_enter(&hdl->sa_lock);
2077 if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) != 0) {
2078 mutex_exit(&hdl->sa_lock);
2079 return (error);
2080 }
2081 *size = bulk.sa_size;
2082
2083 mutex_exit(&hdl->sa_lock);
2084 return (0);
2085 }
2086
2087 int
sa_bulk_lookup_locked(sa_handle_t * hdl,sa_bulk_attr_t * attrs,int count)2088 sa_bulk_lookup_locked(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
2089 {
2090 ASSERT(hdl);
2091 ASSERT(MUTEX_HELD(&hdl->sa_lock));
2092 return (sa_lookup_impl(hdl, attrs, count));
2093 }
2094
2095 int
sa_bulk_lookup(sa_handle_t * hdl,sa_bulk_attr_t * attrs,int count)2096 sa_bulk_lookup(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
2097 {
2098 int error;
2099
2100 ASSERT(hdl);
2101 mutex_enter(&hdl->sa_lock);
2102 error = sa_bulk_lookup_locked(hdl, attrs, count);
2103 mutex_exit(&hdl->sa_lock);
2104 return (error);
2105 }
2106
2107 int
sa_bulk_update(sa_handle_t * hdl,sa_bulk_attr_t * attrs,int count,dmu_tx_t * tx)2108 sa_bulk_update(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count, dmu_tx_t *tx)
2109 {
2110 int error;
2111
2112 ASSERT(hdl);
2113 mutex_enter(&hdl->sa_lock);
2114 error = sa_bulk_update_impl(hdl, attrs, count, tx);
2115 mutex_exit(&hdl->sa_lock);
2116 return (error);
2117 }
2118
2119 int
sa_remove(sa_handle_t * hdl,sa_attr_type_t attr,dmu_tx_t * tx)2120 sa_remove(sa_handle_t *hdl, sa_attr_type_t attr, dmu_tx_t *tx)
2121 {
2122 int error;
2123
2124 mutex_enter(&hdl->sa_lock);
2125 error = sa_modify_attrs(hdl, attr, SA_REMOVE, NULL,
2126 NULL, 0, tx);
2127 mutex_exit(&hdl->sa_lock);
2128 return (error);
2129 }
2130
2131 void
sa_object_info(sa_handle_t * hdl,dmu_object_info_t * doi)2132 sa_object_info(sa_handle_t *hdl, dmu_object_info_t *doi)
2133 {
2134 dmu_object_info_from_db(hdl->sa_bonus, doi);
2135 }
2136
2137 void
sa_object_size(sa_handle_t * hdl,uint32_t * blksize,u_longlong_t * nblocks)2138 sa_object_size(sa_handle_t *hdl, uint32_t *blksize, u_longlong_t *nblocks)
2139 {
2140 dmu_object_size_from_db(hdl->sa_bonus,
2141 blksize, nblocks);
2142 }
2143
2144 void
sa_set_userp(sa_handle_t * hdl,void * ptr)2145 sa_set_userp(sa_handle_t *hdl, void *ptr)
2146 {
2147 hdl->sa_userp = ptr;
2148 }
2149
2150 dmu_buf_t *
sa_get_db(sa_handle_t * hdl)2151 sa_get_db(sa_handle_t *hdl)
2152 {
2153 return (hdl->sa_bonus);
2154 }
2155
2156 void *
sa_get_userdata(sa_handle_t * hdl)2157 sa_get_userdata(sa_handle_t *hdl)
2158 {
2159 return (hdl->sa_userp);
2160 }
2161
2162 void
sa_register_update_callback_locked(objset_t * os,sa_update_cb_t * func)2163 sa_register_update_callback_locked(objset_t *os, sa_update_cb_t *func)
2164 {
2165 ASSERT(MUTEX_HELD(&os->os_sa->sa_lock));
2166 os->os_sa->sa_update_cb = func;
2167 }
2168
2169 void
sa_register_update_callback(objset_t * os,sa_update_cb_t * func)2170 sa_register_update_callback(objset_t *os, sa_update_cb_t *func)
2171 {
2172
2173 mutex_enter(&os->os_sa->sa_lock);
2174 sa_register_update_callback_locked(os, func);
2175 mutex_exit(&os->os_sa->sa_lock);
2176 }
2177
2178 uint64_t
sa_handle_object(sa_handle_t * hdl)2179 sa_handle_object(sa_handle_t *hdl)
2180 {
2181 return (hdl->sa_bonus->db_object);
2182 }
2183
2184 boolean_t
sa_enabled(objset_t * os)2185 sa_enabled(objset_t *os)
2186 {
2187 return (os->os_sa == NULL);
2188 }
2189
2190 int
sa_set_sa_object(objset_t * os,uint64_t sa_object)2191 sa_set_sa_object(objset_t *os, uint64_t sa_object)
2192 {
2193 sa_os_t *sa = os->os_sa;
2194
2195 if (sa->sa_master_obj)
2196 return (1);
2197
2198 sa->sa_master_obj = sa_object;
2199
2200 return (0);
2201 }
2202
2203 int
sa_hdrsize(void * arg)2204 sa_hdrsize(void *arg)
2205 {
2206 sa_hdr_phys_t *hdr = arg;
2207
2208 return (SA_HDR_SIZE(hdr));
2209 }
2210
2211 void
sa_handle_lock(sa_handle_t * hdl)2212 sa_handle_lock(sa_handle_t *hdl)
2213 {
2214 ASSERT(hdl);
2215 mutex_enter(&hdl->sa_lock);
2216 }
2217
2218 void
sa_handle_unlock(sa_handle_t * hdl)2219 sa_handle_unlock(sa_handle_t *hdl)
2220 {
2221 ASSERT(hdl);
2222 mutex_exit(&hdl->sa_lock);
2223 }
2224
2225 #ifdef _KERNEL
2226 EXPORT_SYMBOL(sa_handle_get);
2227 EXPORT_SYMBOL(sa_handle_get_from_db);
2228 EXPORT_SYMBOL(sa_handle_destroy);
2229 EXPORT_SYMBOL(sa_buf_hold);
2230 EXPORT_SYMBOL(sa_buf_rele);
2231 EXPORT_SYMBOL(sa_spill_rele);
2232 EXPORT_SYMBOL(sa_lookup);
2233 EXPORT_SYMBOL(sa_update);
2234 EXPORT_SYMBOL(sa_remove);
2235 EXPORT_SYMBOL(sa_bulk_lookup);
2236 EXPORT_SYMBOL(sa_bulk_lookup_locked);
2237 EXPORT_SYMBOL(sa_bulk_update);
2238 EXPORT_SYMBOL(sa_size);
2239 EXPORT_SYMBOL(sa_object_info);
2240 EXPORT_SYMBOL(sa_object_size);
2241 EXPORT_SYMBOL(sa_get_userdata);
2242 EXPORT_SYMBOL(sa_set_userp);
2243 EXPORT_SYMBOL(sa_get_db);
2244 EXPORT_SYMBOL(sa_handle_object);
2245 EXPORT_SYMBOL(sa_register_update_callback);
2246 EXPORT_SYMBOL(sa_setup);
2247 EXPORT_SYMBOL(sa_replace_all_by_template);
2248 EXPORT_SYMBOL(sa_replace_all_by_template_locked);
2249 EXPORT_SYMBOL(sa_enabled);
2250 EXPORT_SYMBOL(sa_cache_init);
2251 EXPORT_SYMBOL(sa_cache_fini);
2252 EXPORT_SYMBOL(sa_set_sa_object);
2253 EXPORT_SYMBOL(sa_hdrsize);
2254 EXPORT_SYMBOL(sa_handle_lock);
2255 EXPORT_SYMBOL(sa_handle_unlock);
2256 EXPORT_SYMBOL(sa_lookup_uio);
2257 EXPORT_SYMBOL(sa_add_projid);
2258 #endif /* _KERNEL */
2259