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