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