]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/sa.c
Linux 4.19-rc3+ compat: Remove refcount_t compat
[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 #ifdef HAVE_EFFICIENT_UNALIGNED_ACCESS
154 #define SA_COPY_DATA(f, s, t, l) \
155 do { \
156 if (f == NULL) { \
157 if (l == 8) { \
158 *(uint64_t *)t = *(uint64_t *)s; \
159 } else if (l == 16) { \
160 *(uint64_t *)t = *(uint64_t *)s; \
161 *(uint64_t *)((uintptr_t)t + 8) = \
162 *(uint64_t *)((uintptr_t)s + 8); \
163 } else { \
164 bcopy(s, t, l); \
165 } \
166 } else { \
167 sa_copy_data(f, s, t, l); \
168 } \
169 } while (0)
170 #else
171 #define SA_COPY_DATA(f, s, t, l) sa_copy_data(f, s, t, l)
172 #endif
173
174 /*
175 * This table is fixed and cannot be changed. Its purpose is to
176 * allow the SA code to work with both old/new ZPL file systems.
177 * It contains the list of legacy attributes. These attributes aren't
178 * stored in the "attribute" registry zap objects, since older ZPL file systems
179 * won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
180 * use this static table.
181 */
182 sa_attr_reg_t sa_legacy_attrs[] = {
183 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
184 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
185 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
186 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
187 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
188 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
189 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
190 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
191 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
192 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
193 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
194 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
195 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
196 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
197 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
198 {"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
199 };
200
201 /*
202 * This is only used for objects of type DMU_OT_ZNODE
203 */
204 sa_attr_type_t sa_legacy_zpl_layout[] = {
205 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
206 };
207
208 /*
209 * Special dummy layout used for buffers with no attributes.
210 */
211 sa_attr_type_t sa_dummy_zpl_layout[] = { 0 };
212
213 static int sa_legacy_attr_count = ARRAY_SIZE(sa_legacy_attrs);
214 static kmem_cache_t *sa_cache = NULL;
215
216 /*ARGSUSED*/
217 static int
218 sa_cache_constructor(void *buf, void *unused, int kmflag)
219 {
220 sa_handle_t *hdl = buf;
221
222 mutex_init(&hdl->sa_lock, NULL, MUTEX_DEFAULT, NULL);
223 return (0);
224 }
225
226 /*ARGSUSED*/
227 static void
228 sa_cache_destructor(void *buf, void *unused)
229 {
230 sa_handle_t *hdl = buf;
231 mutex_destroy(&hdl->sa_lock);
232 }
233
234 void
235 sa_cache_init(void)
236 {
237 sa_cache = kmem_cache_create("sa_cache",
238 sizeof (sa_handle_t), 0, sa_cache_constructor,
239 sa_cache_destructor, NULL, NULL, NULL, 0);
240 }
241
242 void
243 sa_cache_fini(void)
244 {
245 if (sa_cache)
246 kmem_cache_destroy(sa_cache);
247 }
248
249 static int
250 layout_num_compare(const void *arg1, const void *arg2)
251 {
252 const sa_lot_t *node1 = (const sa_lot_t *)arg1;
253 const sa_lot_t *node2 = (const sa_lot_t *)arg2;
254
255 return (AVL_CMP(node1->lot_num, node2->lot_num));
256 }
257
258 static int
259 layout_hash_compare(const void *arg1, const void *arg2)
260 {
261 const sa_lot_t *node1 = (const sa_lot_t *)arg1;
262 const sa_lot_t *node2 = (const sa_lot_t *)arg2;
263
264 int cmp = AVL_CMP(node1->lot_hash, node2->lot_hash);
265 if (likely(cmp))
266 return (cmp);
267
268 return (AVL_CMP(node1->lot_instance, node2->lot_instance));
269 }
270
271 boolean_t
272 sa_layout_equal(sa_lot_t *tbf, sa_attr_type_t *attrs, int count)
273 {
274 int i;
275
276 if (count != tbf->lot_attr_count)
277 return (1);
278
279 for (i = 0; i != count; i++) {
280 if (attrs[i] != tbf->lot_attrs[i])
281 return (1);
282 }
283 return (0);
284 }
285
286 #define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
287
288 static uint64_t
289 sa_layout_info_hash(sa_attr_type_t *attrs, int attr_count)
290 {
291 int i;
292 uint64_t crc = -1ULL;
293
294 for (i = 0; i != attr_count; i++)
295 crc ^= SA_ATTR_HASH(attrs[i]);
296
297 return (crc);
298 }
299
300 static int
301 sa_get_spill(sa_handle_t *hdl)
302 {
303 int rc;
304 if (hdl->sa_spill == NULL) {
305 if ((rc = dmu_spill_hold_existing(hdl->sa_bonus, NULL,
306 &hdl->sa_spill)) == 0)
307 VERIFY(0 == sa_build_index(hdl, SA_SPILL));
308 } else {
309 rc = 0;
310 }
311
312 return (rc);
313 }
314
315 /*
316 * Main attribute lookup/update function
317 * returns 0 for success or non zero for failures
318 *
319 * Operates on bulk array, first failure will abort further processing
320 */
321 int
322 sa_attr_op(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
323 sa_data_op_t data_op, dmu_tx_t *tx)
324 {
325 sa_os_t *sa = hdl->sa_os->os_sa;
326 int i;
327 int error = 0;
328 sa_buf_type_t buftypes;
329
330 buftypes = 0;
331
332 ASSERT(count > 0);
333 for (i = 0; i != count; i++) {
334 ASSERT(bulk[i].sa_attr <= hdl->sa_os->os_sa->sa_num_attrs);
335
336 bulk[i].sa_addr = NULL;
337 /* First check the bonus buffer */
338
339 if (hdl->sa_bonus_tab && TOC_ATTR_PRESENT(
340 hdl->sa_bonus_tab->sa_idx_tab[bulk[i].sa_attr])) {
341 SA_ATTR_INFO(sa, hdl->sa_bonus_tab,
342 SA_GET_HDR(hdl, SA_BONUS),
343 bulk[i].sa_attr, bulk[i], SA_BONUS, hdl);
344 if (tx && !(buftypes & SA_BONUS)) {
345 dmu_buf_will_dirty(hdl->sa_bonus, tx);
346 buftypes |= SA_BONUS;
347 }
348 }
349 if (bulk[i].sa_addr == NULL &&
350 ((error = sa_get_spill(hdl)) == 0)) {
351 if (TOC_ATTR_PRESENT(
352 hdl->sa_spill_tab->sa_idx_tab[bulk[i].sa_attr])) {
353 SA_ATTR_INFO(sa, hdl->sa_spill_tab,
354 SA_GET_HDR(hdl, SA_SPILL),
355 bulk[i].sa_attr, bulk[i], SA_SPILL, hdl);
356 if (tx && !(buftypes & SA_SPILL) &&
357 bulk[i].sa_size == bulk[i].sa_length) {
358 dmu_buf_will_dirty(hdl->sa_spill, tx);
359 buftypes |= SA_SPILL;
360 }
361 }
362 }
363 if (error && error != ENOENT) {
364 return ((error == ECKSUM) ? EIO : error);
365 }
366
367 switch (data_op) {
368 case SA_LOOKUP:
369 if (bulk[i].sa_addr == NULL)
370 return (SET_ERROR(ENOENT));
371 if (bulk[i].sa_data) {
372 SA_COPY_DATA(bulk[i].sa_data_func,
373 bulk[i].sa_addr, bulk[i].sa_data,
374 bulk[i].sa_size);
375 }
376 continue;
377
378 case SA_UPDATE:
379 /* existing rewrite of attr */
380 if (bulk[i].sa_addr &&
381 bulk[i].sa_size == bulk[i].sa_length) {
382 SA_COPY_DATA(bulk[i].sa_data_func,
383 bulk[i].sa_data, bulk[i].sa_addr,
384 bulk[i].sa_length);
385 continue;
386 } else if (bulk[i].sa_addr) { /* attr size change */
387 error = sa_modify_attrs(hdl, bulk[i].sa_attr,
388 SA_REPLACE, bulk[i].sa_data_func,
389 bulk[i].sa_data, bulk[i].sa_length, tx);
390 } else { /* adding new attribute */
391 error = sa_modify_attrs(hdl, bulk[i].sa_attr,
392 SA_ADD, bulk[i].sa_data_func,
393 bulk[i].sa_data, bulk[i].sa_length, tx);
394 }
395 if (error)
396 return (error);
397 break;
398 default:
399 break;
400 }
401 }
402 return (error);
403 }
404
405 static sa_lot_t *
406 sa_add_layout_entry(objset_t *os, sa_attr_type_t *attrs, int attr_count,
407 uint64_t lot_num, uint64_t hash, boolean_t zapadd, dmu_tx_t *tx)
408 {
409 sa_os_t *sa = os->os_sa;
410 sa_lot_t *tb, *findtb;
411 int i;
412 avl_index_t loc;
413
414 ASSERT(MUTEX_HELD(&sa->sa_lock));
415 tb = kmem_zalloc(sizeof (sa_lot_t), KM_SLEEP);
416 tb->lot_attr_count = attr_count;
417 tb->lot_attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
418 KM_SLEEP);
419 bcopy(attrs, tb->lot_attrs, sizeof (sa_attr_type_t) * attr_count);
420 tb->lot_num = lot_num;
421 tb->lot_hash = hash;
422 tb->lot_instance = 0;
423
424 if (zapadd) {
425 char attr_name[8];
426
427 if (sa->sa_layout_attr_obj == 0) {
428 sa->sa_layout_attr_obj = zap_create_link(os,
429 DMU_OT_SA_ATTR_LAYOUTS,
430 sa->sa_master_obj, SA_LAYOUTS, tx);
431 }
432
433 (void) snprintf(attr_name, sizeof (attr_name),
434 "%d", (int)lot_num);
435 VERIFY(0 == zap_update(os, os->os_sa->sa_layout_attr_obj,
436 attr_name, 2, attr_count, attrs, tx));
437 }
438
439 list_create(&tb->lot_idx_tab, sizeof (sa_idx_tab_t),
440 offsetof(sa_idx_tab_t, sa_next));
441
442 for (i = 0; i != attr_count; i++) {
443 if (sa->sa_attr_table[tb->lot_attrs[i]].sa_length == 0)
444 tb->lot_var_sizes++;
445 }
446
447 avl_add(&sa->sa_layout_num_tree, tb);
448
449 /* verify we don't have a hash collision */
450 if ((findtb = avl_find(&sa->sa_layout_hash_tree, tb, &loc)) != NULL) {
451 for (; findtb && findtb->lot_hash == hash;
452 findtb = AVL_NEXT(&sa->sa_layout_hash_tree, findtb)) {
453 if (findtb->lot_instance != tb->lot_instance)
454 break;
455 tb->lot_instance++;
456 }
457 }
458 avl_add(&sa->sa_layout_hash_tree, tb);
459 return (tb);
460 }
461
462 static void
463 sa_find_layout(objset_t *os, uint64_t hash, sa_attr_type_t *attrs,
464 int count, dmu_tx_t *tx, sa_lot_t **lot)
465 {
466 sa_lot_t *tb, tbsearch;
467 avl_index_t loc;
468 sa_os_t *sa = os->os_sa;
469 boolean_t found = B_FALSE;
470
471 mutex_enter(&sa->sa_lock);
472 tbsearch.lot_hash = hash;
473 tbsearch.lot_instance = 0;
474 tb = avl_find(&sa->sa_layout_hash_tree, &tbsearch, &loc);
475 if (tb) {
476 for (; tb && tb->lot_hash == hash;
477 tb = AVL_NEXT(&sa->sa_layout_hash_tree, tb)) {
478 if (sa_layout_equal(tb, attrs, count) == 0) {
479 found = B_TRUE;
480 break;
481 }
482 }
483 }
484 if (!found) {
485 tb = sa_add_layout_entry(os, attrs, count,
486 avl_numnodes(&sa->sa_layout_num_tree), hash, B_TRUE, tx);
487 }
488 mutex_exit(&sa->sa_lock);
489 *lot = tb;
490 }
491
492 static int
493 sa_resize_spill(sa_handle_t *hdl, uint32_t size, dmu_tx_t *tx)
494 {
495 int error;
496 uint32_t blocksize;
497
498 if (size == 0) {
499 blocksize = SPA_MINBLOCKSIZE;
500 } else if (size > SPA_OLD_MAXBLOCKSIZE) {
501 ASSERT(0);
502 return (SET_ERROR(EFBIG));
503 } else {
504 blocksize = P2ROUNDUP_TYPED(size, SPA_MINBLOCKSIZE, uint32_t);
505 }
506
507 error = dbuf_spill_set_blksz(hdl->sa_spill, blocksize, tx);
508 ASSERT(error == 0);
509 return (error);
510 }
511
512 static void
513 sa_copy_data(sa_data_locator_t *func, void *datastart, void *target, int buflen)
514 {
515 if (func == NULL) {
516 bcopy(datastart, target, buflen);
517 } else {
518 boolean_t start;
519 int bytes;
520 void *dataptr;
521 void *saptr = target;
522 uint32_t length;
523
524 start = B_TRUE;
525 bytes = 0;
526 while (bytes < buflen) {
527 func(&dataptr, &length, buflen, start, datastart);
528 bcopy(dataptr, saptr, length);
529 saptr = (void *)((caddr_t)saptr + length);
530 bytes += length;
531 start = B_FALSE;
532 }
533 }
534 }
535
536 /*
537 * Determine several different values pertaining to system attribute
538 * buffers.
539 *
540 * Return the size of the sa_hdr_phys_t header for the buffer. Each
541 * variable length attribute except the first contributes two bytes to
542 * the header size, which is then rounded up to an 8-byte boundary.
543 *
544 * The following output parameters are also computed.
545 *
546 * index - The index of the first attribute in attr_desc that will
547 * spill over. Only valid if will_spill is set.
548 *
549 * total - The total number of bytes of all system attributes described
550 * in attr_desc.
551 *
552 * will_spill - Set when spilling is necessary. It is only set when
553 * the buftype is SA_BONUS.
554 */
555 static int
556 sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count,
557 dmu_buf_t *db, sa_buf_type_t buftype, int full_space, int *index,
558 int *total, boolean_t *will_spill)
559 {
560 int var_size_count = 0;
561 int i;
562 int hdrsize;
563 int extra_hdrsize;
564
565 if (buftype == SA_BONUS && sa->sa_force_spill) {
566 *total = 0;
567 *index = 0;
568 *will_spill = B_TRUE;
569 return (0);
570 }
571
572 *index = -1;
573 *total = 0;
574 *will_spill = B_FALSE;
575
576 extra_hdrsize = 0;
577 hdrsize = (SA_BONUSTYPE_FROM_DB(db) == DMU_OT_ZNODE) ? 0 :
578 sizeof (sa_hdr_phys_t);
579
580 ASSERT(IS_P2ALIGNED(full_space, 8));
581
582 for (i = 0; i != attr_count; i++) {
583 boolean_t is_var_sz, might_spill_here;
584 int tmp_hdrsize;
585
586 *total = P2ROUNDUP(*total, 8);
587 *total += attr_desc[i].sa_length;
588 if (*will_spill)
589 continue;
590
591 is_var_sz = (SA_REGISTERED_LEN(sa, attr_desc[i].sa_attr) == 0);
592 if (is_var_sz)
593 var_size_count++;
594
595 /*
596 * Calculate what the SA header size would be if this
597 * attribute doesn't spill.
598 */
599 tmp_hdrsize = hdrsize + ((is_var_sz && var_size_count > 1) ?
600 sizeof (uint16_t) : 0);
601
602 /*
603 * Check whether this attribute spans into the space
604 * that would be used by the spill block pointer should
605 * a spill block be needed.
606 */
607 might_spill_here =
608 buftype == SA_BONUS && *index == -1 &&
609 (*total + P2ROUNDUP(tmp_hdrsize, 8)) >
610 (full_space - sizeof (blkptr_t));
611
612 if (is_var_sz && var_size_count > 1) {
613 if (buftype == SA_SPILL ||
614 tmp_hdrsize + *total < full_space) {
615 /*
616 * Record the extra header size in case this
617 * increase needs to be reversed due to
618 * spill-over.
619 */
620 hdrsize = tmp_hdrsize;
621 if (*index != -1 || might_spill_here)
622 extra_hdrsize += sizeof (uint16_t);
623 } else {
624 ASSERT(buftype == SA_BONUS);
625 if (*index == -1)
626 *index = i;
627 *will_spill = B_TRUE;
628 continue;
629 }
630 }
631
632 /*
633 * Store index of where spill *could* occur. Then
634 * continue to count the remaining attribute sizes. The
635 * sum is used later for sizing bonus and spill buffer.
636 */
637 if (might_spill_here)
638 *index = i;
639
640 if ((*total + P2ROUNDUP(hdrsize, 8)) > full_space &&
641 buftype == SA_BONUS)
642 *will_spill = B_TRUE;
643 }
644
645 if (*will_spill)
646 hdrsize -= extra_hdrsize;
647
648 hdrsize = P2ROUNDUP(hdrsize, 8);
649 return (hdrsize);
650 }
651
652 #define BUF_SPACE_NEEDED(total, header) (total + header)
653
654 /*
655 * Find layout that corresponds to ordering of attributes
656 * If not found a new layout number is created and added to
657 * persistent layout tables.
658 */
659 static int
660 sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
661 dmu_tx_t *tx)
662 {
663 sa_os_t *sa = hdl->sa_os->os_sa;
664 uint64_t hash;
665 sa_buf_type_t buftype;
666 sa_hdr_phys_t *sahdr;
667 void *data_start;
668 sa_attr_type_t *attrs, *attrs_start;
669 int i, lot_count;
670 int dnodesize;
671 int spill_idx;
672 int hdrsize;
673 int spillhdrsize = 0;
674 int used;
675 dmu_object_type_t bonustype;
676 sa_lot_t *lot;
677 int len_idx;
678 int spill_used;
679 int bonuslen;
680 boolean_t spilling;
681
682 dmu_buf_will_dirty(hdl->sa_bonus, tx);
683 bonustype = SA_BONUSTYPE_FROM_DB(hdl->sa_bonus);
684 dmu_object_dnsize_from_db(hdl->sa_bonus, &dnodesize);
685 bonuslen = DN_BONUS_SIZE(dnodesize);
686
687 /* first determine bonus header size and sum of all attributes */
688 hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus,
689 SA_BONUS, bonuslen, &spill_idx, &used, &spilling);
690
691 if (used > SPA_OLD_MAXBLOCKSIZE)
692 return (SET_ERROR(EFBIG));
693
694 VERIFY0(dmu_set_bonus(hdl->sa_bonus, spilling ?
695 MIN(bonuslen - sizeof (blkptr_t), used + hdrsize) :
696 used + hdrsize, tx));
697
698 ASSERT((bonustype == DMU_OT_ZNODE && spilling == 0) ||
699 bonustype == DMU_OT_SA);
700
701 /* setup and size spill buffer when needed */
702 if (spilling) {
703 boolean_t dummy;
704
705 if (hdl->sa_spill == NULL) {
706 VERIFY(dmu_spill_hold_by_bonus(hdl->sa_bonus, 0, NULL,
707 &hdl->sa_spill) == 0);
708 }
709 dmu_buf_will_dirty(hdl->sa_spill, tx);
710
711 spillhdrsize = sa_find_sizes(sa, &attr_desc[spill_idx],
712 attr_count - spill_idx, hdl->sa_spill, SA_SPILL,
713 hdl->sa_spill->db_size, &i, &spill_used, &dummy);
714
715 if (spill_used > SPA_OLD_MAXBLOCKSIZE)
716 return (SET_ERROR(EFBIG));
717
718 if (BUF_SPACE_NEEDED(spill_used, spillhdrsize) >
719 hdl->sa_spill->db_size)
720 VERIFY(0 == sa_resize_spill(hdl,
721 BUF_SPACE_NEEDED(spill_used, spillhdrsize), tx));
722 }
723
724 /* setup starting pointers to lay down data */
725 data_start = (void *)((uintptr_t)hdl->sa_bonus->db_data + hdrsize);
726 sahdr = (sa_hdr_phys_t *)hdl->sa_bonus->db_data;
727 buftype = SA_BONUS;
728
729 attrs_start = attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
730 KM_SLEEP);
731 lot_count = 0;
732
733 for (i = 0, len_idx = 0, hash = -1ULL; i != attr_count; i++) {
734 uint16_t length;
735
736 ASSERT(IS_P2ALIGNED(data_start, 8));
737 attrs[i] = attr_desc[i].sa_attr;
738 length = SA_REGISTERED_LEN(sa, attrs[i]);
739 if (length == 0)
740 length = attr_desc[i].sa_length;
741
742 if (spilling && i == spill_idx) { /* switch to spill buffer */
743 VERIFY(bonustype == DMU_OT_SA);
744 if (buftype == SA_BONUS && !sa->sa_force_spill) {
745 sa_find_layout(hdl->sa_os, hash, attrs_start,
746 lot_count, tx, &lot);
747 SA_SET_HDR(sahdr, lot->lot_num, hdrsize);
748 }
749
750 buftype = SA_SPILL;
751 hash = -1ULL;
752 len_idx = 0;
753
754 sahdr = (sa_hdr_phys_t *)hdl->sa_spill->db_data;
755 sahdr->sa_magic = SA_MAGIC;
756 data_start = (void *)((uintptr_t)sahdr +
757 spillhdrsize);
758 attrs_start = &attrs[i];
759 lot_count = 0;
760 }
761 hash ^= SA_ATTR_HASH(attrs[i]);
762 attr_desc[i].sa_addr = data_start;
763 attr_desc[i].sa_size = length;
764 SA_COPY_DATA(attr_desc[i].sa_data_func, attr_desc[i].sa_data,
765 data_start, length);
766 if (sa->sa_attr_table[attrs[i]].sa_length == 0) {
767 sahdr->sa_lengths[len_idx++] = length;
768 }
769 data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
770 length), 8);
771 lot_count++;
772 }
773
774 sa_find_layout(hdl->sa_os, hash, attrs_start, lot_count, tx, &lot);
775
776 /*
777 * Verify that old znodes always have layout number 0.
778 * Must be DMU_OT_SA for arbitrary layouts
779 */
780 VERIFY((bonustype == DMU_OT_ZNODE && lot->lot_num == 0) ||
781 (bonustype == DMU_OT_SA && lot->lot_num > 1));
782
783 if (bonustype == DMU_OT_SA) {
784 SA_SET_HDR(sahdr, lot->lot_num,
785 buftype == SA_BONUS ? hdrsize : spillhdrsize);
786 }
787
788 kmem_free(attrs, sizeof (sa_attr_type_t) * attr_count);
789 if (hdl->sa_bonus_tab) {
790 sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
791 hdl->sa_bonus_tab = NULL;
792 }
793 if (!sa->sa_force_spill)
794 VERIFY(0 == sa_build_index(hdl, SA_BONUS));
795 if (hdl->sa_spill) {
796 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
797 if (!spilling) {
798 /*
799 * remove spill block that is no longer needed.
800 */
801 dmu_buf_rele(hdl->sa_spill, NULL);
802 hdl->sa_spill = NULL;
803 hdl->sa_spill_tab = NULL;
804 VERIFY(0 == dmu_rm_spill(hdl->sa_os,
805 sa_handle_object(hdl), tx));
806 } else {
807 VERIFY(0 == sa_build_index(hdl, SA_SPILL));
808 }
809 }
810
811 return (0);
812 }
813
814 static void
815 sa_free_attr_table(sa_os_t *sa)
816 {
817 int i;
818
819 if (sa->sa_attr_table == NULL)
820 return;
821
822 for (i = 0; i != sa->sa_num_attrs; i++) {
823 if (sa->sa_attr_table[i].sa_name)
824 kmem_free(sa->sa_attr_table[i].sa_name,
825 strlen(sa->sa_attr_table[i].sa_name) + 1);
826 }
827
828 kmem_free(sa->sa_attr_table,
829 sizeof (sa_attr_table_t) * sa->sa_num_attrs);
830
831 sa->sa_attr_table = NULL;
832 }
833
834 static int
835 sa_attr_table_setup(objset_t *os, sa_attr_reg_t *reg_attrs, int count)
836 {
837 sa_os_t *sa = os->os_sa;
838 uint64_t sa_attr_count = 0;
839 uint64_t sa_reg_count = 0;
840 int error = 0;
841 uint64_t attr_value;
842 sa_attr_table_t *tb;
843 zap_cursor_t zc;
844 zap_attribute_t za;
845 int registered_count = 0;
846 int i;
847 dmu_objset_type_t ostype = dmu_objset_type(os);
848
849 sa->sa_user_table =
850 kmem_zalloc(count * sizeof (sa_attr_type_t), KM_SLEEP);
851 sa->sa_user_table_sz = count * sizeof (sa_attr_type_t);
852
853 if (sa->sa_reg_attr_obj != 0) {
854 error = zap_count(os, sa->sa_reg_attr_obj,
855 &sa_attr_count);
856
857 /*
858 * Make sure we retrieved a count and that it isn't zero
859 */
860 if (error || (error == 0 && sa_attr_count == 0)) {
861 if (error == 0)
862 error = SET_ERROR(EINVAL);
863 goto bail;
864 }
865 sa_reg_count = sa_attr_count;
866 }
867
868 if (ostype == DMU_OST_ZFS && sa_attr_count == 0)
869 sa_attr_count += sa_legacy_attr_count;
870
871 /* Allocate attribute numbers for attributes that aren't registered */
872 for (i = 0; i != count; i++) {
873 boolean_t found = B_FALSE;
874 int j;
875
876 if (ostype == DMU_OST_ZFS) {
877 for (j = 0; j != sa_legacy_attr_count; j++) {
878 if (strcmp(reg_attrs[i].sa_name,
879 sa_legacy_attrs[j].sa_name) == 0) {
880 sa->sa_user_table[i] =
881 sa_legacy_attrs[j].sa_attr;
882 found = B_TRUE;
883 }
884 }
885 }
886 if (found)
887 continue;
888
889 if (sa->sa_reg_attr_obj)
890 error = zap_lookup(os, sa->sa_reg_attr_obj,
891 reg_attrs[i].sa_name, 8, 1, &attr_value);
892 else
893 error = SET_ERROR(ENOENT);
894 switch (error) {
895 case ENOENT:
896 sa->sa_user_table[i] = (sa_attr_type_t)sa_attr_count;
897 sa_attr_count++;
898 break;
899 case 0:
900 sa->sa_user_table[i] = ATTR_NUM(attr_value);
901 break;
902 default:
903 goto bail;
904 }
905 }
906
907 sa->sa_num_attrs = sa_attr_count;
908 tb = sa->sa_attr_table =
909 kmem_zalloc(sizeof (sa_attr_table_t) * sa_attr_count, KM_SLEEP);
910
911 /*
912 * Attribute table is constructed from requested attribute list,
913 * previously foreign registered attributes, and also the legacy
914 * ZPL set of attributes.
915 */
916
917 if (sa->sa_reg_attr_obj) {
918 for (zap_cursor_init(&zc, os, sa->sa_reg_attr_obj);
919 (error = zap_cursor_retrieve(&zc, &za)) == 0;
920 zap_cursor_advance(&zc)) {
921 uint64_t value;
922 value = za.za_first_integer;
923
924 registered_count++;
925 tb[ATTR_NUM(value)].sa_attr = ATTR_NUM(value);
926 tb[ATTR_NUM(value)].sa_length = ATTR_LENGTH(value);
927 tb[ATTR_NUM(value)].sa_byteswap = ATTR_BSWAP(value);
928 tb[ATTR_NUM(value)].sa_registered = B_TRUE;
929
930 if (tb[ATTR_NUM(value)].sa_name) {
931 continue;
932 }
933 tb[ATTR_NUM(value)].sa_name =
934 kmem_zalloc(strlen(za.za_name) +1, KM_SLEEP);
935 (void) strlcpy(tb[ATTR_NUM(value)].sa_name, za.za_name,
936 strlen(za.za_name) +1);
937 }
938 zap_cursor_fini(&zc);
939 /*
940 * Make sure we processed the correct number of registered
941 * attributes
942 */
943 if (registered_count != sa_reg_count) {
944 ASSERT(error != 0);
945 goto bail;
946 }
947
948 }
949
950 if (ostype == DMU_OST_ZFS) {
951 for (i = 0; i != sa_legacy_attr_count; i++) {
952 if (tb[i].sa_name)
953 continue;
954 tb[i].sa_attr = sa_legacy_attrs[i].sa_attr;
955 tb[i].sa_length = sa_legacy_attrs[i].sa_length;
956 tb[i].sa_byteswap = sa_legacy_attrs[i].sa_byteswap;
957 tb[i].sa_registered = B_FALSE;
958 tb[i].sa_name =
959 kmem_zalloc(strlen(sa_legacy_attrs[i].sa_name) +1,
960 KM_SLEEP);
961 (void) strlcpy(tb[i].sa_name,
962 sa_legacy_attrs[i].sa_name,
963 strlen(sa_legacy_attrs[i].sa_name) + 1);
964 }
965 }
966
967 for (i = 0; i != count; i++) {
968 sa_attr_type_t attr_id;
969
970 attr_id = sa->sa_user_table[i];
971 if (tb[attr_id].sa_name)
972 continue;
973
974 tb[attr_id].sa_length = reg_attrs[i].sa_length;
975 tb[attr_id].sa_byteswap = reg_attrs[i].sa_byteswap;
976 tb[attr_id].sa_attr = attr_id;
977 tb[attr_id].sa_name =
978 kmem_zalloc(strlen(reg_attrs[i].sa_name) + 1, KM_SLEEP);
979 (void) strlcpy(tb[attr_id].sa_name, reg_attrs[i].sa_name,
980 strlen(reg_attrs[i].sa_name) + 1);
981 }
982
983 sa->sa_need_attr_registration =
984 (sa_attr_count != registered_count);
985
986 return (0);
987 bail:
988 kmem_free(sa->sa_user_table, count * sizeof (sa_attr_type_t));
989 sa->sa_user_table = NULL;
990 sa_free_attr_table(sa);
991 ASSERT(error != 0);
992 return (error);
993 }
994
995 int
996 sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count,
997 sa_attr_type_t **user_table)
998 {
999 zap_cursor_t zc;
1000 zap_attribute_t za;
1001 sa_os_t *sa;
1002 dmu_objset_type_t ostype = dmu_objset_type(os);
1003 sa_attr_type_t *tb;
1004 int error;
1005
1006 mutex_enter(&os->os_user_ptr_lock);
1007 if (os->os_sa) {
1008 mutex_enter(&os->os_sa->sa_lock);
1009 mutex_exit(&os->os_user_ptr_lock);
1010 tb = os->os_sa->sa_user_table;
1011 mutex_exit(&os->os_sa->sa_lock);
1012 *user_table = tb;
1013 return (0);
1014 }
1015
1016 sa = kmem_zalloc(sizeof (sa_os_t), KM_SLEEP);
1017 mutex_init(&sa->sa_lock, NULL, MUTEX_DEFAULT, NULL);
1018 sa->sa_master_obj = sa_obj;
1019
1020 os->os_sa = sa;
1021 mutex_enter(&sa->sa_lock);
1022 mutex_exit(&os->os_user_ptr_lock);
1023 avl_create(&sa->sa_layout_num_tree, layout_num_compare,
1024 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node));
1025 avl_create(&sa->sa_layout_hash_tree, layout_hash_compare,
1026 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_hash_node));
1027
1028 if (sa_obj) {
1029 error = zap_lookup(os, sa_obj, SA_LAYOUTS,
1030 8, 1, &sa->sa_layout_attr_obj);
1031 if (error != 0 && error != ENOENT)
1032 goto fail;
1033 error = zap_lookup(os, sa_obj, SA_REGISTRY,
1034 8, 1, &sa->sa_reg_attr_obj);
1035 if (error != 0 && error != ENOENT)
1036 goto fail;
1037 }
1038
1039 if ((error = sa_attr_table_setup(os, reg_attrs, count)) != 0)
1040 goto fail;
1041
1042 if (sa->sa_layout_attr_obj != 0) {
1043 uint64_t layout_count;
1044
1045 error = zap_count(os, sa->sa_layout_attr_obj,
1046 &layout_count);
1047
1048 /*
1049 * Layout number count should be > 0
1050 */
1051 if (error || (error == 0 && layout_count == 0)) {
1052 if (error == 0)
1053 error = SET_ERROR(EINVAL);
1054 goto fail;
1055 }
1056
1057 for (zap_cursor_init(&zc, os, sa->sa_layout_attr_obj);
1058 (error = zap_cursor_retrieve(&zc, &za)) == 0;
1059 zap_cursor_advance(&zc)) {
1060 sa_attr_type_t *lot_attrs;
1061 uint64_t lot_num;
1062
1063 lot_attrs = kmem_zalloc(sizeof (sa_attr_type_t) *
1064 za.za_num_integers, KM_SLEEP);
1065
1066 if ((error = (zap_lookup(os, sa->sa_layout_attr_obj,
1067 za.za_name, 2, za.za_num_integers,
1068 lot_attrs))) != 0) {
1069 kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
1070 za.za_num_integers);
1071 break;
1072 }
1073 VERIFY(ddi_strtoull(za.za_name, NULL, 10,
1074 (unsigned long long *)&lot_num) == 0);
1075
1076 (void) sa_add_layout_entry(os, lot_attrs,
1077 za.za_num_integers, lot_num,
1078 sa_layout_info_hash(lot_attrs,
1079 za.za_num_integers), B_FALSE, NULL);
1080 kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
1081 za.za_num_integers);
1082 }
1083 zap_cursor_fini(&zc);
1084
1085 /*
1086 * Make sure layout count matches number of entries added
1087 * to AVL tree
1088 */
1089 if (avl_numnodes(&sa->sa_layout_num_tree) != layout_count) {
1090 ASSERT(error != 0);
1091 goto fail;
1092 }
1093 }
1094
1095 /* Add special layout number for old ZNODES */
1096 if (ostype == DMU_OST_ZFS) {
1097 (void) sa_add_layout_entry(os, sa_legacy_zpl_layout,
1098 sa_legacy_attr_count, 0,
1099 sa_layout_info_hash(sa_legacy_zpl_layout,
1100 sa_legacy_attr_count), B_FALSE, NULL);
1101
1102 (void) sa_add_layout_entry(os, sa_dummy_zpl_layout, 0, 1,
1103 0, B_FALSE, NULL);
1104 }
1105 *user_table = os->os_sa->sa_user_table;
1106 mutex_exit(&sa->sa_lock);
1107 return (0);
1108 fail:
1109 os->os_sa = NULL;
1110 sa_free_attr_table(sa);
1111 if (sa->sa_user_table)
1112 kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
1113 mutex_exit(&sa->sa_lock);
1114 avl_destroy(&sa->sa_layout_hash_tree);
1115 avl_destroy(&sa->sa_layout_num_tree);
1116 mutex_destroy(&sa->sa_lock);
1117 kmem_free(sa, sizeof (sa_os_t));
1118 return ((error == ECKSUM) ? EIO : error);
1119 }
1120
1121 void
1122 sa_tear_down(objset_t *os)
1123 {
1124 sa_os_t *sa = os->os_sa;
1125 sa_lot_t *layout;
1126 void *cookie;
1127
1128 kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
1129
1130 /* Free up attr table */
1131
1132 sa_free_attr_table(sa);
1133
1134 cookie = NULL;
1135 while ((layout =
1136 avl_destroy_nodes(&sa->sa_layout_hash_tree, &cookie))) {
1137 sa_idx_tab_t *tab;
1138 while ((tab = list_head(&layout->lot_idx_tab))) {
1139 ASSERT(refcount_count(&tab->sa_refcount));
1140 sa_idx_tab_rele(os, tab);
1141 }
1142 }
1143
1144 cookie = NULL;
1145 while ((layout = avl_destroy_nodes(&sa->sa_layout_num_tree, &cookie))) {
1146 kmem_free(layout->lot_attrs,
1147 sizeof (sa_attr_type_t) * layout->lot_attr_count);
1148 kmem_free(layout, sizeof (sa_lot_t));
1149 }
1150
1151 avl_destroy(&sa->sa_layout_hash_tree);
1152 avl_destroy(&sa->sa_layout_num_tree);
1153 mutex_destroy(&sa->sa_lock);
1154
1155 kmem_free(sa, sizeof (sa_os_t));
1156 os->os_sa = NULL;
1157 }
1158
1159 void
1160 sa_build_idx_tab(void *hdr, void *attr_addr, sa_attr_type_t attr,
1161 uint16_t length, int length_idx, boolean_t var_length, void *userp)
1162 {
1163 sa_idx_tab_t *idx_tab = userp;
1164
1165 if (var_length) {
1166 ASSERT(idx_tab->sa_variable_lengths);
1167 idx_tab->sa_variable_lengths[length_idx] = length;
1168 }
1169 TOC_ATTR_ENCODE(idx_tab->sa_idx_tab[attr], length_idx,
1170 (uint32_t)((uintptr_t)attr_addr - (uintptr_t)hdr));
1171 }
1172
1173 static void
1174 sa_attr_iter(objset_t *os, sa_hdr_phys_t *hdr, dmu_object_type_t type,
1175 sa_iterfunc_t func, sa_lot_t *tab, void *userp)
1176 {
1177 void *data_start;
1178 sa_lot_t *tb = tab;
1179 sa_lot_t search;
1180 avl_index_t loc;
1181 sa_os_t *sa = os->os_sa;
1182 int i;
1183 uint16_t *length_start = NULL;
1184 uint8_t length_idx = 0;
1185
1186 if (tab == NULL) {
1187 search.lot_num = SA_LAYOUT_NUM(hdr, type);
1188 tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
1189 ASSERT(tb);
1190 }
1191
1192 if (IS_SA_BONUSTYPE(type)) {
1193 data_start = (void *)P2ROUNDUP(((uintptr_t)hdr +
1194 offsetof(sa_hdr_phys_t, sa_lengths) +
1195 (sizeof (uint16_t) * tb->lot_var_sizes)), 8);
1196 length_start = hdr->sa_lengths;
1197 } else {
1198 data_start = hdr;
1199 }
1200
1201 for (i = 0; i != tb->lot_attr_count; i++) {
1202 int attr_length, reg_length;
1203 uint8_t idx_len;
1204
1205 reg_length = sa->sa_attr_table[tb->lot_attrs[i]].sa_length;
1206 if (reg_length) {
1207 attr_length = reg_length;
1208 idx_len = 0;
1209 } else {
1210 attr_length = length_start[length_idx];
1211 idx_len = length_idx++;
1212 }
1213
1214 func(hdr, data_start, tb->lot_attrs[i], attr_length,
1215 idx_len, reg_length == 0 ? B_TRUE : B_FALSE, userp);
1216
1217 data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
1218 attr_length), 8);
1219 }
1220 }
1221
1222 /*ARGSUSED*/
1223 void
1224 sa_byteswap_cb(void *hdr, void *attr_addr, sa_attr_type_t attr,
1225 uint16_t length, int length_idx, boolean_t variable_length, void *userp)
1226 {
1227 sa_handle_t *hdl = userp;
1228 sa_os_t *sa = hdl->sa_os->os_sa;
1229
1230 sa_bswap_table[sa->sa_attr_table[attr].sa_byteswap](attr_addr, length);
1231 }
1232
1233 void
1234 sa_byteswap(sa_handle_t *hdl, sa_buf_type_t buftype)
1235 {
1236 sa_hdr_phys_t *sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1237 dmu_buf_impl_t *db;
1238 int num_lengths = 1;
1239 int i;
1240 ASSERTV(sa_os_t *sa = hdl->sa_os->os_sa);
1241
1242 ASSERT(MUTEX_HELD(&sa->sa_lock));
1243 if (sa_hdr_phys->sa_magic == SA_MAGIC)
1244 return;
1245
1246 db = SA_GET_DB(hdl, buftype);
1247
1248 if (buftype == SA_SPILL) {
1249 arc_release(db->db_buf, NULL);
1250 arc_buf_thaw(db->db_buf);
1251 }
1252
1253 sa_hdr_phys->sa_magic = BSWAP_32(sa_hdr_phys->sa_magic);
1254 sa_hdr_phys->sa_layout_info = BSWAP_16(sa_hdr_phys->sa_layout_info);
1255
1256 /*
1257 * Determine number of variable lengths in header
1258 * The standard 8 byte header has one for free and a
1259 * 16 byte header would have 4 + 1;
1260 */
1261 if (SA_HDR_SIZE(sa_hdr_phys) > 8)
1262 num_lengths += (SA_HDR_SIZE(sa_hdr_phys) - 8) >> 1;
1263 for (i = 0; i != num_lengths; i++)
1264 sa_hdr_phys->sa_lengths[i] =
1265 BSWAP_16(sa_hdr_phys->sa_lengths[i]);
1266
1267 sa_attr_iter(hdl->sa_os, sa_hdr_phys, DMU_OT_SA,
1268 sa_byteswap_cb, NULL, hdl);
1269
1270 if (buftype == SA_SPILL)
1271 arc_buf_freeze(((dmu_buf_impl_t *)hdl->sa_spill)->db_buf);
1272 }
1273
1274 static int
1275 sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype)
1276 {
1277 sa_hdr_phys_t *sa_hdr_phys;
1278 dmu_buf_impl_t *db = SA_GET_DB(hdl, buftype);
1279 dmu_object_type_t bonustype = SA_BONUSTYPE_FROM_DB(db);
1280 sa_os_t *sa = hdl->sa_os->os_sa;
1281 sa_idx_tab_t *idx_tab;
1282
1283 sa_hdr_phys = SA_GET_HDR(hdl, buftype);
1284
1285 mutex_enter(&sa->sa_lock);
1286
1287 /* Do we need to byteswap? */
1288
1289 /* only check if not old znode */
1290 if (IS_SA_BONUSTYPE(bonustype) && sa_hdr_phys->sa_magic != SA_MAGIC &&
1291 sa_hdr_phys->sa_magic != 0) {
1292 if (BSWAP_32(sa_hdr_phys->sa_magic) != SA_MAGIC) {
1293 mutex_exit(&sa->sa_lock);
1294 zfs_dbgmsg("Buffer Header: %x != SA_MAGIC:%x "
1295 "object=%#llx\n", sa_hdr_phys->sa_magic, SA_MAGIC,
1296 db->db.db_object);
1297 return (SET_ERROR(EIO));
1298 }
1299 sa_byteswap(hdl, buftype);
1300 }
1301
1302 idx_tab = sa_find_idx_tab(hdl->sa_os, bonustype, sa_hdr_phys);
1303
1304 if (buftype == SA_BONUS)
1305 hdl->sa_bonus_tab = idx_tab;
1306 else
1307 hdl->sa_spill_tab = idx_tab;
1308
1309 mutex_exit(&sa->sa_lock);
1310 return (0);
1311 }
1312
1313 /*ARGSUSED*/
1314 static void
1315 sa_evict_sync(void *dbu)
1316 {
1317 panic("evicting sa dbuf\n");
1318 }
1319
1320 static void
1321 sa_idx_tab_rele(objset_t *os, void *arg)
1322 {
1323 sa_os_t *sa = os->os_sa;
1324 sa_idx_tab_t *idx_tab = arg;
1325
1326 if (idx_tab == NULL)
1327 return;
1328
1329 mutex_enter(&sa->sa_lock);
1330 if (refcount_remove(&idx_tab->sa_refcount, NULL) == 0) {
1331 list_remove(&idx_tab->sa_layout->lot_idx_tab, idx_tab);
1332 if (idx_tab->sa_variable_lengths)
1333 kmem_free(idx_tab->sa_variable_lengths,
1334 sizeof (uint16_t) *
1335 idx_tab->sa_layout->lot_var_sizes);
1336 refcount_destroy(&idx_tab->sa_refcount);
1337 kmem_free(idx_tab->sa_idx_tab,
1338 sizeof (uint32_t) * sa->sa_num_attrs);
1339 kmem_free(idx_tab, sizeof (sa_idx_tab_t));
1340 }
1341 mutex_exit(&sa->sa_lock);
1342 }
1343
1344 static void
1345 sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab)
1346 {
1347 ASSERTV(sa_os_t *sa = os->os_sa);
1348
1349 ASSERT(MUTEX_HELD(&sa->sa_lock));
1350 (void) zfs_refcount_add(&idx_tab->sa_refcount, NULL);
1351 }
1352
1353 void
1354 sa_spill_rele(sa_handle_t *hdl)
1355 {
1356 mutex_enter(&hdl->sa_lock);
1357 if (hdl->sa_spill) {
1358 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
1359 dmu_buf_rele(hdl->sa_spill, NULL);
1360 hdl->sa_spill = NULL;
1361 hdl->sa_spill_tab = NULL;
1362 }
1363 mutex_exit(&hdl->sa_lock);
1364 }
1365
1366 void
1367 sa_handle_destroy(sa_handle_t *hdl)
1368 {
1369 dmu_buf_t *db = hdl->sa_bonus;
1370
1371 mutex_enter(&hdl->sa_lock);
1372 (void) dmu_buf_remove_user(db, &hdl->sa_dbu);
1373
1374 if (hdl->sa_bonus_tab)
1375 sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
1376
1377 if (hdl->sa_spill_tab)
1378 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
1379
1380 dmu_buf_rele(hdl->sa_bonus, NULL);
1381
1382 if (hdl->sa_spill)
1383 dmu_buf_rele((dmu_buf_t *)hdl->sa_spill, NULL);
1384 mutex_exit(&hdl->sa_lock);
1385
1386 kmem_cache_free(sa_cache, hdl);
1387 }
1388
1389 int
1390 sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
1391 sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1392 {
1393 int error = 0;
1394 sa_handle_t *handle = NULL;
1395 #ifdef ZFS_DEBUG
1396 dmu_object_info_t doi;
1397
1398 dmu_object_info_from_db(db, &doi);
1399 ASSERT(doi.doi_bonus_type == DMU_OT_SA ||
1400 doi.doi_bonus_type == DMU_OT_ZNODE);
1401 #endif
1402 /* find handle, if it exists */
1403 /* if one doesn't exist then create a new one, and initialize it */
1404
1405 if (hdl_type == SA_HDL_SHARED)
1406 handle = dmu_buf_get_user(db);
1407
1408 if (handle == NULL) {
1409 sa_handle_t *winner = NULL;
1410
1411 handle = kmem_cache_alloc(sa_cache, KM_SLEEP);
1412 handle->sa_dbu.dbu_evict_func_sync = NULL;
1413 handle->sa_dbu.dbu_evict_func_async = NULL;
1414 handle->sa_userp = userp;
1415 handle->sa_bonus = db;
1416 handle->sa_os = os;
1417 handle->sa_spill = NULL;
1418 handle->sa_bonus_tab = NULL;
1419 handle->sa_spill_tab = NULL;
1420
1421 error = sa_build_index(handle, SA_BONUS);
1422
1423 if (hdl_type == SA_HDL_SHARED) {
1424 dmu_buf_init_user(&handle->sa_dbu, sa_evict_sync, NULL,
1425 NULL);
1426 winner = dmu_buf_set_user_ie(db, &handle->sa_dbu);
1427 }
1428
1429 if (winner != NULL) {
1430 kmem_cache_free(sa_cache, handle);
1431 handle = winner;
1432 }
1433 }
1434 *handlepp = handle;
1435
1436 return (error);
1437 }
1438
1439 int
1440 sa_handle_get(objset_t *objset, uint64_t objid, void *userp,
1441 sa_handle_type_t hdl_type, sa_handle_t **handlepp)
1442 {
1443 dmu_buf_t *db;
1444 int error;
1445
1446 if ((error = dmu_bonus_hold(objset, objid, NULL, &db)))
1447 return (error);
1448
1449 return (sa_handle_get_from_db(objset, db, userp, hdl_type,
1450 handlepp));
1451 }
1452
1453 int
1454 sa_buf_hold(objset_t *objset, uint64_t obj_num, void *tag, dmu_buf_t **db)
1455 {
1456 return (dmu_bonus_hold(objset, obj_num, tag, db));
1457 }
1458
1459 void
1460 sa_buf_rele(dmu_buf_t *db, void *tag)
1461 {
1462 dmu_buf_rele(db, tag);
1463 }
1464
1465 int
1466 sa_lookup_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count)
1467 {
1468 ASSERT(hdl);
1469 ASSERT(MUTEX_HELD(&hdl->sa_lock));
1470 return (sa_attr_op(hdl, bulk, count, SA_LOOKUP, NULL));
1471 }
1472
1473 static int
1474 sa_lookup_locked(sa_handle_t *hdl, sa_attr_type_t attr, void *buf,
1475 uint32_t buflen)
1476 {
1477 int error;
1478 sa_bulk_attr_t bulk;
1479
1480 VERIFY3U(buflen, <=, SA_ATTR_MAX_LEN);
1481
1482 bulk.sa_attr = attr;
1483 bulk.sa_data = buf;
1484 bulk.sa_length = buflen;
1485 bulk.sa_data_func = NULL;
1486
1487 ASSERT(hdl);
1488 error = sa_lookup_impl(hdl, &bulk, 1);
1489 return (error);
1490 }
1491
1492 int
1493 sa_lookup(sa_handle_t *hdl, sa_attr_type_t attr, void *buf, uint32_t buflen)
1494 {
1495 int error;
1496
1497 mutex_enter(&hdl->sa_lock);
1498 error = sa_lookup_locked(hdl, attr, buf, buflen);
1499 mutex_exit(&hdl->sa_lock);
1500
1501 return (error);
1502 }
1503
1504 #ifdef _KERNEL
1505 int
1506 sa_lookup_uio(sa_handle_t *hdl, sa_attr_type_t attr, uio_t *uio)
1507 {
1508 int error;
1509 sa_bulk_attr_t bulk;
1510
1511 bulk.sa_data = NULL;
1512 bulk.sa_attr = attr;
1513 bulk.sa_data_func = NULL;
1514
1515 ASSERT(hdl);
1516
1517 mutex_enter(&hdl->sa_lock);
1518 if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) == 0) {
1519 error = uiomove((void *)bulk.sa_addr, MIN(bulk.sa_size,
1520 uio->uio_resid), UIO_READ, uio);
1521 }
1522 mutex_exit(&hdl->sa_lock);
1523 return (error);
1524 }
1525
1526 /*
1527 * For the existed object that is upgraded from old system, its ondisk layout
1528 * has no slot for the project ID attribute. But quota accounting logic needs
1529 * to access related slots by offset directly. So we need to adjust these old
1530 * objects' layout to make the project ID to some unified and fixed offset.
1531 */
1532 int
1533 sa_add_projid(sa_handle_t *hdl, dmu_tx_t *tx, uint64_t projid)
1534 {
1535 znode_t *zp = sa_get_userdata(hdl);
1536 dmu_buf_t *db = sa_get_db(hdl);
1537 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1538 int count = 0, err = 0;
1539 sa_bulk_attr_t *bulk, *attrs;
1540 zfs_acl_locator_cb_t locate = { 0 };
1541 uint64_t uid, gid, mode, rdev, xattr = 0, parent, gen, links;
1542 uint64_t crtime[2], mtime[2], ctime[2], atime[2];
1543 zfs_acl_phys_t znode_acl = { 0 };
1544 char scanstamp[AV_SCANSTAMP_SZ];
1545
1546 if (zp->z_acl_cached == NULL) {
1547 zfs_acl_t *aclp;
1548
1549 mutex_enter(&zp->z_acl_lock);
1550 err = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1551 mutex_exit(&zp->z_acl_lock);
1552 if (err != 0 && err != ENOENT)
1553 return (err);
1554 }
1555
1556 bulk = kmem_zalloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
1557 attrs = kmem_zalloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
1558 mutex_enter(&hdl->sa_lock);
1559 mutex_enter(&zp->z_lock);
1560
1561 err = sa_lookup_locked(hdl, SA_ZPL_PROJID(zfsvfs), &projid,
1562 sizeof (uint64_t));
1563 if (unlikely(err == 0))
1564 /* Someone has added project ID attr by race. */
1565 err = EEXIST;
1566 if (err != ENOENT)
1567 goto out;
1568
1569 /* First do a bulk query of the attributes that aren't cached */
1570 if (zp->z_is_sa) {
1571 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1572 &mode, 8);
1573 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1574 &gen, 8);
1575 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1576 &uid, 8);
1577 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1578 &gid, 8);
1579 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
1580 &parent, 8);
1581 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1582 &atime, 16);
1583 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1584 &mtime, 16);
1585 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1586 &ctime, 16);
1587 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1588 &crtime, 16);
1589 if (S_ISBLK(ZTOI(zp)->i_mode) || S_ISCHR(ZTOI(zp)->i_mode))
1590 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1591 &rdev, 8);
1592 } else {
1593 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1594 &atime, 16);
1595 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1596 &mtime, 16);
1597 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1598 &ctime, 16);
1599 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1600 &crtime, 16);
1601 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1602 &gen, 8);
1603 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1604 &mode, 8);
1605 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
1606 &parent, 8);
1607 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL,
1608 &xattr, 8);
1609 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1610 &rdev, 8);
1611 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1612 &uid, 8);
1613 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1614 &gid, 8);
1615 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1616 &znode_acl, 88);
1617 }
1618 err = sa_bulk_lookup_locked(hdl, bulk, count);
1619 if (err != 0)
1620 goto out;
1621
1622 err = sa_lookup_locked(hdl, SA_ZPL_XATTR(zfsvfs), &xattr, 8);
1623 if (err != 0 && err != ENOENT)
1624 goto out;
1625
1626 zp->z_projid = projid;
1627 zp->z_pflags |= ZFS_PROJID;
1628 links = ZTOI(zp)->i_nlink;
1629 count = 0;
1630 err = 0;
1631
1632 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
1633 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
1634 &zp->z_size, 8);
1635 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_GEN(zfsvfs), NULL, &gen, 8);
1636 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
1637 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
1638 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
1639 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1640 &zp->z_pflags, 8);
1641 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
1642 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
1643 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
1644 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
1645 &crtime, 16);
1646 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
1647 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_PROJID(zfsvfs), NULL, &projid, 8);
1648
1649 if (S_ISBLK(ZTOI(zp)->i_mode) || S_ISCHR(ZTOI(zp)->i_mode))
1650 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
1651 &rdev, 8);
1652
1653 if (zp->z_acl_cached != NULL) {
1654 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
1655 &zp->z_acl_cached->z_acl_count, 8);
1656 if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
1657 zfs_acl_xform(zp, zp->z_acl_cached, CRED());
1658 locate.cb_aclp = zp->z_acl_cached;
1659 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
1660 zfs_acl_data_locator, &locate,
1661 zp->z_acl_cached->z_acl_bytes);
1662 }
1663
1664 if (xattr)
1665 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_XATTR(zfsvfs), NULL,
1666 &xattr, 8);
1667
1668 if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
1669 bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
1670 scanstamp, 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 *
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 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
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
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
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
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
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 bcopy(hdl->sa_bonus->db_data, old_data[0],
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 bcopy(hdl->sa_spill->db_data, old_data[1],
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 != 2; 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
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((dmu_buf_t *)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
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
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
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
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
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
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
2132 sa_object_info(sa_handle_t *hdl, dmu_object_info_t *doi)
2133 {
2134 dmu_object_info_from_db((dmu_buf_t *)hdl->sa_bonus, doi);
2135 }
2136
2137 void
2138 sa_object_size(sa_handle_t *hdl, uint32_t *blksize, u_longlong_t *nblocks)
2139 {
2140 dmu_object_size_from_db((dmu_buf_t *)hdl->sa_bonus,
2141 blksize, nblocks);
2142 }
2143
2144 void
2145 sa_set_userp(sa_handle_t *hdl, void *ptr)
2146 {
2147 hdl->sa_userp = ptr;
2148 }
2149
2150 dmu_buf_t *
2151 sa_get_db(sa_handle_t *hdl)
2152 {
2153 return ((dmu_buf_t *)hdl->sa_bonus);
2154 }
2155
2156 void *
2157 sa_get_userdata(sa_handle_t *hdl)
2158 {
2159 return (hdl->sa_userp);
2160 }
2161
2162 void
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
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
2179 sa_handle_object(sa_handle_t *hdl)
2180 {
2181 return (hdl->sa_bonus->db_object);
2182 }
2183
2184 boolean_t
2185 sa_enabled(objset_t *os)
2186 {
2187 return (os->os_sa == NULL);
2188 }
2189
2190 int
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
2204 sa_hdrsize(void *arg)
2205 {
2206 sa_hdr_phys_t *hdr = arg;
2207
2208 return (SA_HDR_SIZE(hdr));
2209 }
2210
2211 void
2212 sa_handle_lock(sa_handle_t *hdl)
2213 {
2214 ASSERT(hdl);
2215 mutex_enter(&hdl->sa_lock);
2216 }
2217
2218 void
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 */