1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or https://opensource.org/licenses/CDDL-1.0. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2013, 2017 by Delphix. All rights reserved. 25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26 * Copyright 2023 RackTop Systems, Inc. 27 */ 28 29#include <sys/zfs_context.h> 30#include <sys/types.h> 31#include <sys/param.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/dmu_tx.h> 37#include <sys/dbuf.h> 38#include <sys/dnode.h> 39#include <sys/zap.h> 40#include <sys/sa.h> 41#include <sys/sunddi.h> 42#include <sys/sa_impl.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 what's 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 * attributes. 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 131typedef void (sa_iterfunc_t)(void *hdr, void *addr, sa_attr_type_t, 132 uint16_t length, int length_idx, boolean_t, void *userp); 133 134static int sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype); 135static void sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab); 136static sa_idx_tab_t *sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype, 137 sa_hdr_phys_t *hdr); 138static void sa_idx_tab_rele(objset_t *os, void *arg); 139static void sa_copy_data(sa_data_locator_t *func, void *start, void *target, 140 int buflen); 141static 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 145static 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) \ 155do { \ 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 memcpy(t, s, 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 */ 182static const 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 */ 204static const 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 */ 211static const sa_attr_type_t sa_dummy_zpl_layout[] = { 0 }; 212 213static const size_t sa_legacy_attr_count = ARRAY_SIZE(sa_legacy_attrs); 214static kmem_cache_t *sa_cache = NULL; 215 216static int 217sa_cache_constructor(void *buf, void *unused, int kmflag) 218{ 219 (void) unused, (void) kmflag; 220 sa_handle_t *hdl = buf; 221 222 mutex_init(&hdl->sa_lock, NULL, MUTEX_DEFAULT, NULL); 223 return (0); 224} 225 226static void 227sa_cache_destructor(void *buf, void *unused) 228{ 229 (void) unused; 230 sa_handle_t *hdl = buf; 231 mutex_destroy(&hdl->sa_lock); 232} 233 234void 235sa_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 242void 243sa_cache_fini(void) 244{ 245 if (sa_cache) 246 kmem_cache_destroy(sa_cache); 247} 248 249static int 250layout_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 (TREE_CMP(node1->lot_num, node2->lot_num)); 256} 257 258static int 259layout_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 = TREE_CMP(node1->lot_hash, node2->lot_hash); 265 if (likely(cmp)) 266 return (cmp); 267 268 return (TREE_CMP(node1->lot_instance, node2->lot_instance)); 269} 270 271static boolean_t 272sa_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 288static uint64_t 289sa_layout_info_hash(const sa_attr_type_t *attrs, int attr_count) 290{ 291 uint64_t crc = -1ULL; 292 293 for (int i = 0; i != attr_count; i++) 294 crc ^= SA_ATTR_HASH(attrs[i]); 295 296 return (crc); 297} 298 299static int 300sa_get_spill(sa_handle_t *hdl) 301{ 302 int rc; 303 if (hdl->sa_spill == NULL) { 304 if ((rc = dmu_spill_hold_existing(hdl->sa_bonus, NULL, 305 &hdl->sa_spill)) == 0) 306 VERIFY(0 == sa_build_index(hdl, SA_SPILL)); 307 } else { 308 rc = 0; 309 } 310 311 return (rc); 312} 313 314/* 315 * Main attribute lookup/update function 316 * returns 0 for success or non zero for failures 317 * 318 * Operates on bulk array, first failure will abort further processing 319 */ 320static int 321sa_attr_op(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count, 322 sa_data_op_t data_op, dmu_tx_t *tx) 323{ 324 sa_os_t *sa = hdl->sa_os->os_sa; 325 int i; 326 int error = 0; 327 sa_buf_type_t buftypes; 328 329 buftypes = 0; 330 331 ASSERT(count > 0); 332 for (i = 0; i != count; i++) { 333 ASSERT(bulk[i].sa_attr <= hdl->sa_os->os_sa->sa_num_attrs); 334 335 bulk[i].sa_addr = NULL; 336 /* First check the bonus buffer */ 337 338 if (hdl->sa_bonus_tab && TOC_ATTR_PRESENT( 339 hdl->sa_bonus_tab->sa_idx_tab[bulk[i].sa_attr])) { 340 SA_ATTR_INFO(sa, hdl->sa_bonus_tab, 341 SA_GET_HDR(hdl, SA_BONUS), 342 bulk[i].sa_attr, bulk[i], SA_BONUS, hdl); 343 if (tx && !(buftypes & SA_BONUS)) { 344 dmu_buf_will_dirty(hdl->sa_bonus, tx); 345 buftypes |= SA_BONUS; 346 } 347 } 348 if (bulk[i].sa_addr == NULL && 349 ((error = sa_get_spill(hdl)) == 0)) { 350 if (TOC_ATTR_PRESENT( 351 hdl->sa_spill_tab->sa_idx_tab[bulk[i].sa_attr])) { 352 SA_ATTR_INFO(sa, hdl->sa_spill_tab, 353 SA_GET_HDR(hdl, SA_SPILL), 354 bulk[i].sa_attr, bulk[i], SA_SPILL, hdl); 355 if (tx && !(buftypes & SA_SPILL) && 356 bulk[i].sa_size == bulk[i].sa_length) { 357 dmu_buf_will_dirty(hdl->sa_spill, tx); 358 buftypes |= SA_SPILL; 359 } 360 } 361 } 362 if (error && error != ENOENT) { 363 return ((error == ECKSUM) ? EIO : error); 364 } 365 366 switch (data_op) { 367 case SA_LOOKUP: 368 if (bulk[i].sa_addr == NULL) 369 return (SET_ERROR(ENOENT)); 370 if (bulk[i].sa_data) { 371 SA_COPY_DATA(bulk[i].sa_data_func, 372 bulk[i].sa_addr, bulk[i].sa_data, 373 MIN(bulk[i].sa_size, bulk[i].sa_length)); 374 } 375 continue; 376 377 case SA_UPDATE: 378 /* existing rewrite of attr */ 379 if (bulk[i].sa_addr && 380 bulk[i].sa_size == bulk[i].sa_length) { 381 SA_COPY_DATA(bulk[i].sa_data_func, 382 bulk[i].sa_data, bulk[i].sa_addr, 383 bulk[i].sa_length); 384 continue; 385 } else if (bulk[i].sa_addr) { /* attr size change */ 386 error = sa_modify_attrs(hdl, bulk[i].sa_attr, 387 SA_REPLACE, bulk[i].sa_data_func, 388 bulk[i].sa_data, bulk[i].sa_length, tx); 389 } else { /* adding new attribute */ 390 error = sa_modify_attrs(hdl, bulk[i].sa_attr, 391 SA_ADD, bulk[i].sa_data_func, 392 bulk[i].sa_data, bulk[i].sa_length, tx); 393 } 394 if (error) 395 return (error); 396 break; 397 default: 398 break; 399 } 400 } 401 return (error); 402} 403 404static sa_lot_t * 405sa_add_layout_entry(objset_t *os, const sa_attr_type_t *attrs, int attr_count, 406 uint64_t lot_num, uint64_t hash, boolean_t zapadd, dmu_tx_t *tx) 407{ 408 sa_os_t *sa = os->os_sa; 409 sa_lot_t *tb, *findtb; 410 int i; 411 avl_index_t loc; 412 413 ASSERT(MUTEX_HELD(&sa->sa_lock)); 414 tb = kmem_zalloc(sizeof (sa_lot_t), KM_SLEEP); 415 tb->lot_attr_count = attr_count; 416 tb->lot_attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count, 417 KM_SLEEP); 418 memcpy(tb->lot_attrs, attrs, sizeof (sa_attr_type_t) * attr_count); 419 tb->lot_num = lot_num; 420 tb->lot_hash = hash; 421 tb->lot_instance = 0; 422 423 if (zapadd) { 424 char attr_name[8]; 425 426 if (sa->sa_layout_attr_obj == 0) { 427 sa->sa_layout_attr_obj = zap_create_link(os, 428 DMU_OT_SA_ATTR_LAYOUTS, 429 sa->sa_master_obj, SA_LAYOUTS, tx); 430 } 431 432 (void) snprintf(attr_name, sizeof (attr_name), 433 "%d", (int)lot_num); 434 VERIFY(0 == zap_update(os, os->os_sa->sa_layout_attr_obj, 435 attr_name, 2, attr_count, attrs, tx)); 436 } 437 438 list_create(&tb->lot_idx_tab, sizeof (sa_idx_tab_t), 439 offsetof(sa_idx_tab_t, sa_next)); 440 441 for (i = 0; i != attr_count; i++) { 442 if (sa->sa_attr_table[tb->lot_attrs[i]].sa_length == 0) 443 tb->lot_var_sizes++; 444 } 445 446 avl_add(&sa->sa_layout_num_tree, tb); 447 448 /* verify we don't have a hash collision */ 449 if ((findtb = avl_find(&sa->sa_layout_hash_tree, tb, &loc)) != NULL) { 450 for (; findtb && findtb->lot_hash == hash; 451 findtb = AVL_NEXT(&sa->sa_layout_hash_tree, findtb)) { 452 if (findtb->lot_instance != tb->lot_instance) 453 break; 454 tb->lot_instance++; 455 } 456 } 457 avl_add(&sa->sa_layout_hash_tree, tb); 458 return (tb); 459} 460 461static void 462sa_find_layout(objset_t *os, uint64_t hash, sa_attr_type_t *attrs, 463 int count, dmu_tx_t *tx, sa_lot_t **lot) 464{ 465 sa_lot_t *tb, tbsearch; 466 avl_index_t loc; 467 sa_os_t *sa = os->os_sa; 468 boolean_t found = B_FALSE; 469 470 mutex_enter(&sa->sa_lock); 471 tbsearch.lot_hash = hash; 472 tbsearch.lot_instance = 0; 473 tb = avl_find(&sa->sa_layout_hash_tree, &tbsearch, &loc); 474 if (tb) { 475 for (; tb && tb->lot_hash == hash; 476 tb = AVL_NEXT(&sa->sa_layout_hash_tree, tb)) { 477 if (sa_layout_equal(tb, attrs, count) == 0) { 478 found = B_TRUE; 479 break; 480 } 481 } 482 } 483 if (!found) { 484 tb = sa_add_layout_entry(os, attrs, count, 485 avl_numnodes(&sa->sa_layout_num_tree), hash, B_TRUE, tx); 486 } 487 mutex_exit(&sa->sa_lock); 488 *lot = tb; 489} 490 491static int 492sa_resize_spill(sa_handle_t *hdl, uint32_t size, dmu_tx_t *tx) 493{ 494 int error; 495 uint32_t blocksize; 496 497 if (size == 0) { 498 blocksize = SPA_MINBLOCKSIZE; 499 } else if (size > SPA_OLD_MAXBLOCKSIZE) { 500 ASSERT(0); 501 return (SET_ERROR(EFBIG)); 502 } else { 503 blocksize = P2ROUNDUP_TYPED(size, SPA_MINBLOCKSIZE, uint32_t); 504 } 505 506 error = dbuf_spill_set_blksz(hdl->sa_spill, blocksize, tx); 507 ASSERT(error == 0); 508 return (error); 509} 510 511static void 512sa_copy_data(sa_data_locator_t *func, void *datastart, void *target, int buflen) 513{ 514 if (func == NULL) { 515 memcpy(target, datastart, buflen); 516 } else { 517 boolean_t start; 518 int bytes; 519 void *dataptr; 520 void *saptr = target; 521 uint32_t length; 522 523 start = B_TRUE; 524 bytes = 0; 525 while (bytes < buflen) { 526 func(&dataptr, &length, buflen, start, datastart); 527 memcpy(saptr, dataptr, length); 528 saptr = (void *)((caddr_t)saptr + length); 529 bytes += length; 530 start = B_FALSE; 531 } 532 } 533} 534 535/* 536 * Determine several different values pertaining to system attribute 537 * buffers. 538 * 539 * Return the size of the sa_hdr_phys_t header for the buffer. Each 540 * variable length attribute except the first contributes two bytes to 541 * the header size, which is then rounded up to an 8-byte boundary. 542 * 543 * The following output parameters are also computed. 544 * 545 * index - The index of the first attribute in attr_desc that will 546 * spill over. Only valid if will_spill is set. 547 * 548 * total - The total number of bytes of all system attributes described 549 * in attr_desc. 550 * 551 * will_spill - Set when spilling is necessary. It is only set when 552 * the buftype is SA_BONUS. 553 */ 554static int 555sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count, 556 dmu_buf_t *db, sa_buf_type_t buftype, int full_space, int *index, 557 int *total, boolean_t *will_spill) 558{ 559 int var_size_count = 0; 560 int i; 561 int hdrsize; 562 int extra_hdrsize; 563 564 if (buftype == SA_BONUS && sa->sa_force_spill) { 565 *total = 0; 566 *index = 0; 567 *will_spill = B_TRUE; 568 return (0); 569 } 570 571 *index = -1; 572 *total = 0; 573 *will_spill = B_FALSE; 574 575 extra_hdrsize = 0; 576 hdrsize = (SA_BONUSTYPE_FROM_DB(db) == DMU_OT_ZNODE) ? 0 : 577 sizeof (sa_hdr_phys_t); 578 579 ASSERT(IS_P2ALIGNED(full_space, 8)); 580 581 for (i = 0; i != attr_count; i++) { 582 boolean_t is_var_sz, might_spill_here; 583 int tmp_hdrsize; 584 585 *total = P2ROUNDUP(*total, 8); 586 *total += attr_desc[i].sa_length; 587 if (*will_spill) 588 continue; 589 590 is_var_sz = (SA_REGISTERED_LEN(sa, attr_desc[i].sa_attr) == 0); 591 if (is_var_sz) 592 var_size_count++; 593 594 /* 595 * Calculate what the SA header size would be if this 596 * attribute doesn't spill. 597 */ 598 tmp_hdrsize = hdrsize + ((is_var_sz && var_size_count > 1) ? 599 sizeof (uint16_t) : 0); 600 601 /* 602 * Check whether this attribute spans into the space 603 * that would be used by the spill block pointer should 604 * a spill block be needed. 605 */ 606 might_spill_here = 607 buftype == SA_BONUS && *index == -1 && 608 (*total + P2ROUNDUP(tmp_hdrsize, 8)) > 609 (full_space - sizeof (blkptr_t)); 610 611 if (is_var_sz && var_size_count > 1) { 612 if (buftype == SA_SPILL || 613 tmp_hdrsize + *total < full_space) { 614 /* 615 * Record the extra header size in case this 616 * increase needs to be reversed due to 617 * spill-over. 618 */ 619 hdrsize = tmp_hdrsize; 620 if (*index != -1 || might_spill_here) 621 extra_hdrsize += sizeof (uint16_t); 622 } else { 623 ASSERT(buftype == SA_BONUS); 624 if (*index == -1) 625 *index = i; 626 *will_spill = B_TRUE; 627 continue; 628 } 629 } 630 631 /* 632 * Store index of where spill *could* occur. Then 633 * continue to count the remaining attribute sizes. The 634 * sum is used later for sizing bonus and spill buffer. 635 */ 636 if (might_spill_here) 637 *index = i; 638 639 if ((*total + P2ROUNDUP(hdrsize, 8)) > full_space && 640 buftype == SA_BONUS) 641 *will_spill = B_TRUE; 642 } 643 644 if (*will_spill) 645 hdrsize -= extra_hdrsize; 646 647 hdrsize = P2ROUNDUP(hdrsize, 8); 648 return (hdrsize); 649} 650 651#define BUF_SPACE_NEEDED(total, header) (total + header) 652 653/* 654 * Find layout that corresponds to ordering of attributes 655 * If not found a new layout number is created and added to 656 * persistent layout tables. 657 */ 658static int 659sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count, 660 dmu_tx_t *tx) 661{ 662 sa_os_t *sa = hdl->sa_os->os_sa; 663 uint64_t hash; 664 sa_buf_type_t buftype; 665 sa_hdr_phys_t *sahdr; 666 void *data_start; 667 sa_attr_type_t *attrs, *attrs_start; 668 int i, lot_count; 669 int dnodesize; 670 int spill_idx; 671 int hdrsize; 672 int spillhdrsize = 0; 673 int used; 674 dmu_object_type_t bonustype; 675 sa_lot_t *lot; 676 int len_idx; 677 int spill_used; 678 int bonuslen; 679 boolean_t spilling; 680 681 dmu_buf_will_dirty(hdl->sa_bonus, tx); 682 bonustype = SA_BONUSTYPE_FROM_DB(hdl->sa_bonus); 683 dmu_object_dnsize_from_db(hdl->sa_bonus, &dnodesize); 684 bonuslen = DN_BONUS_SIZE(dnodesize); 685 686 /* first determine bonus header size and sum of all attributes */ 687 hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus, 688 SA_BONUS, bonuslen, &spill_idx, &used, &spilling); 689 690 if (used > SPA_OLD_MAXBLOCKSIZE) 691 return (SET_ERROR(EFBIG)); 692 693 VERIFY0(dmu_set_bonus(hdl->sa_bonus, spilling ? 694 MIN(bonuslen - sizeof (blkptr_t), used + hdrsize) : 695 used + hdrsize, tx)); 696 697 ASSERT((bonustype == DMU_OT_ZNODE && spilling == 0) || 698 bonustype == DMU_OT_SA); 699 700 /* setup and size spill buffer when needed */ 701 if (spilling) { 702 boolean_t dummy; 703 704 if (hdl->sa_spill == NULL) { 705 VERIFY(dmu_spill_hold_by_bonus(hdl->sa_bonus, 0, NULL, 706 &hdl->sa_spill) == 0); 707 } 708 dmu_buf_will_dirty(hdl->sa_spill, tx); 709 710 spillhdrsize = sa_find_sizes(sa, &attr_desc[spill_idx], 711 attr_count - spill_idx, hdl->sa_spill, SA_SPILL, 712 hdl->sa_spill->db_size, &i, &spill_used, &dummy); 713 714 if (spill_used > SPA_OLD_MAXBLOCKSIZE) 715 return (SET_ERROR(EFBIG)); 716 717 if (BUF_SPACE_NEEDED(spill_used, spillhdrsize) > 718 hdl->sa_spill->db_size) 719 VERIFY(0 == sa_resize_spill(hdl, 720 BUF_SPACE_NEEDED(spill_used, spillhdrsize), tx)); 721 } 722 723 /* setup starting pointers to lay down data */ 724 data_start = (void *)((uintptr_t)hdl->sa_bonus->db_data + hdrsize); 725 sahdr = (sa_hdr_phys_t *)hdl->sa_bonus->db_data; 726 buftype = SA_BONUS; 727 728 attrs_start = attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count, 729 KM_SLEEP); 730 lot_count = 0; 731 732 for (i = 0, len_idx = 0, hash = -1ULL; i != attr_count; i++) { 733 uint16_t length; 734 735 ASSERT(IS_P2ALIGNED(data_start, 8)); 736 attrs[i] = attr_desc[i].sa_attr; 737 length = SA_REGISTERED_LEN(sa, attrs[i]); 738 if (length == 0) 739 length = attr_desc[i].sa_length; 740 741 if (spilling && i == spill_idx) { /* switch to spill buffer */ 742 VERIFY(bonustype == DMU_OT_SA); 743 if (buftype == SA_BONUS && !sa->sa_force_spill) { 744 sa_find_layout(hdl->sa_os, hash, attrs_start, 745 lot_count, tx, &lot); 746 SA_SET_HDR(sahdr, lot->lot_num, hdrsize); 747 } 748 749 buftype = SA_SPILL; 750 hash = -1ULL; 751 len_idx = 0; 752 753 sahdr = (sa_hdr_phys_t *)hdl->sa_spill->db_data; 754 sahdr->sa_magic = SA_MAGIC; 755 data_start = (void *)((uintptr_t)sahdr + 756 spillhdrsize); 757 attrs_start = &attrs[i]; 758 lot_count = 0; 759 } 760 hash ^= SA_ATTR_HASH(attrs[i]); 761 attr_desc[i].sa_addr = data_start; 762 attr_desc[i].sa_size = length; 763 SA_COPY_DATA(attr_desc[i].sa_data_func, attr_desc[i].sa_data, 764 data_start, length); 765 if (sa->sa_attr_table[attrs[i]].sa_length == 0) { 766 sahdr->sa_lengths[len_idx++] = length; 767 } 768 data_start = (void *)P2ROUNDUP(((uintptr_t)data_start + 769 length), 8); 770 lot_count++; 771 } 772 773 sa_find_layout(hdl->sa_os, hash, attrs_start, lot_count, tx, &lot); 774 775 /* 776 * Verify that old znodes always have layout number 0. 777 * Must be DMU_OT_SA for arbitrary layouts 778 */ 779 VERIFY((bonustype == DMU_OT_ZNODE && lot->lot_num == 0) || 780 (bonustype == DMU_OT_SA && lot->lot_num > 1)); 781 782 if (bonustype == DMU_OT_SA) { 783 SA_SET_HDR(sahdr, lot->lot_num, 784 buftype == SA_BONUS ? hdrsize : spillhdrsize); 785 } 786 787 kmem_free(attrs, sizeof (sa_attr_type_t) * attr_count); 788 if (hdl->sa_bonus_tab) { 789 sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab); 790 hdl->sa_bonus_tab = NULL; 791 } 792 if (!sa->sa_force_spill) 793 VERIFY(0 == sa_build_index(hdl, SA_BONUS)); 794 if (hdl->sa_spill) { 795 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab); 796 if (!spilling) { 797 /* 798 * remove spill block that is no longer needed. 799 */ 800 dmu_buf_rele(hdl->sa_spill, NULL); 801 hdl->sa_spill = NULL; 802 hdl->sa_spill_tab = NULL; 803 VERIFY(0 == dmu_rm_spill(hdl->sa_os, 804 sa_handle_object(hdl), tx)); 805 } else { 806 VERIFY(0 == sa_build_index(hdl, SA_SPILL)); 807 } 808 } 809 810 return (0); 811} 812 813static void 814sa_free_attr_table(sa_os_t *sa) 815{ 816 int i; 817 818 if (sa->sa_attr_table == NULL) 819 return; 820 821 for (i = 0; i != sa->sa_num_attrs; i++) { 822 if (sa->sa_attr_table[i].sa_name) 823 kmem_free(sa->sa_attr_table[i].sa_name, 824 strlen(sa->sa_attr_table[i].sa_name) + 1); 825 } 826 827 kmem_free(sa->sa_attr_table, 828 sizeof (sa_attr_table_t) * sa->sa_num_attrs); 829 830 sa->sa_attr_table = NULL; 831} 832 833static int 834sa_attr_table_setup(objset_t *os, const sa_attr_reg_t *reg_attrs, int count) 835{ 836 sa_os_t *sa = os->os_sa; 837 uint64_t sa_attr_count = 0; 838 uint64_t sa_reg_count = 0; 839 int error = 0; 840 uint64_t attr_value; 841 sa_attr_table_t *tb; 842 zap_cursor_t zc; 843 zap_attribute_t za; 844 int registered_count = 0; 845 int i; 846 dmu_objset_type_t ostype = dmu_objset_type(os); 847 848 sa->sa_user_table = 849 kmem_zalloc(count * sizeof (sa_attr_type_t), KM_SLEEP); 850 sa->sa_user_table_sz = count * sizeof (sa_attr_type_t); 851 852 if (sa->sa_reg_attr_obj != 0) { 853 error = zap_count(os, sa->sa_reg_attr_obj, 854 &sa_attr_count); 855 856 /* 857 * Make sure we retrieved a count and that it isn't zero 858 */ 859 if (error || (error == 0 && sa_attr_count == 0)) { 860 if (error == 0) 861 error = SET_ERROR(EINVAL); 862 goto bail; 863 } 864 sa_reg_count = sa_attr_count; 865 } 866 867 if (ostype == DMU_OST_ZFS && sa_attr_count == 0) 868 sa_attr_count += sa_legacy_attr_count; 869 870 /* Allocate attribute numbers for attributes that aren't registered */ 871 for (i = 0; i != count; i++) { 872 boolean_t found = B_FALSE; 873 int j; 874 875 if (ostype == DMU_OST_ZFS) { 876 for (j = 0; j != sa_legacy_attr_count; j++) { 877 if (strcmp(reg_attrs[i].sa_name, 878 sa_legacy_attrs[j].sa_name) == 0) { 879 sa->sa_user_table[i] = 880 sa_legacy_attrs[j].sa_attr; 881 found = B_TRUE; 882 } 883 } 884 } 885 if (found) 886 continue; 887 888 if (sa->sa_reg_attr_obj) 889 error = zap_lookup(os, sa->sa_reg_attr_obj, 890 reg_attrs[i].sa_name, 8, 1, &attr_value); 891 else 892 error = SET_ERROR(ENOENT); 893 switch (error) { 894 case ENOENT: 895 sa->sa_user_table[i] = (sa_attr_type_t)sa_attr_count; 896 sa_attr_count++; 897 break; 898 case 0: 899 sa->sa_user_table[i] = ATTR_NUM(attr_value); 900 break; 901 default: 902 goto bail; 903 } 904 } 905 906 sa->sa_num_attrs = sa_attr_count; 907 tb = sa->sa_attr_table = 908 kmem_zalloc(sizeof (sa_attr_table_t) * sa_attr_count, KM_SLEEP); 909 910 /* 911 * Attribute table is constructed from requested attribute list, 912 * previously foreign registered attributes, and also the legacy 913 * ZPL set of attributes. 914 */ 915 916 if (sa->sa_reg_attr_obj) { 917 for (zap_cursor_init(&zc, os, sa->sa_reg_attr_obj); 918 (error = zap_cursor_retrieve(&zc, &za)) == 0; 919 zap_cursor_advance(&zc)) { 920 uint64_t value; 921 value = za.za_first_integer; 922 923 registered_count++; 924 tb[ATTR_NUM(value)].sa_attr = ATTR_NUM(value); 925 tb[ATTR_NUM(value)].sa_length = ATTR_LENGTH(value); 926 tb[ATTR_NUM(value)].sa_byteswap = ATTR_BSWAP(value); 927 tb[ATTR_NUM(value)].sa_registered = B_TRUE; 928 929 if (tb[ATTR_NUM(value)].sa_name) { 930 continue; 931 } 932 tb[ATTR_NUM(value)].sa_name = 933 kmem_zalloc(strlen(za.za_name) +1, KM_SLEEP); 934 (void) strlcpy(tb[ATTR_NUM(value)].sa_name, za.za_name, 935 strlen(za.za_name) +1); 936 } 937 zap_cursor_fini(&zc); 938 /* 939 * Make sure we processed the correct number of registered 940 * attributes 941 */ 942 if (registered_count != sa_reg_count) { 943 ASSERT(error != 0); 944 goto bail; 945 } 946 947 } 948 949 if (ostype == DMU_OST_ZFS) { 950 for (i = 0; i != sa_legacy_attr_count; i++) { 951 if (tb[i].sa_name) 952 continue; 953 tb[i].sa_attr = sa_legacy_attrs[i].sa_attr; 954 tb[i].sa_length = sa_legacy_attrs[i].sa_length; 955 tb[i].sa_byteswap = sa_legacy_attrs[i].sa_byteswap; 956 tb[i].sa_registered = B_FALSE; 957 tb[i].sa_name = 958 kmem_zalloc(strlen(sa_legacy_attrs[i].sa_name) +1, 959 KM_SLEEP); 960 (void) strlcpy(tb[i].sa_name, 961 sa_legacy_attrs[i].sa_name, 962 strlen(sa_legacy_attrs[i].sa_name) + 1); 963 } 964 } 965 966 for (i = 0; i != count; i++) { 967 sa_attr_type_t attr_id; 968 969 attr_id = sa->sa_user_table[i]; 970 if (tb[attr_id].sa_name) 971 continue; 972 973 tb[attr_id].sa_length = reg_attrs[i].sa_length; 974 tb[attr_id].sa_byteswap = reg_attrs[i].sa_byteswap; 975 tb[attr_id].sa_attr = attr_id; 976 tb[attr_id].sa_name = 977 kmem_zalloc(strlen(reg_attrs[i].sa_name) + 1, KM_SLEEP); 978 (void) strlcpy(tb[attr_id].sa_name, reg_attrs[i].sa_name, 979 strlen(reg_attrs[i].sa_name) + 1); 980 } 981 982 sa->sa_need_attr_registration = 983 (sa_attr_count != registered_count); 984 985 return (0); 986bail: 987 kmem_free(sa->sa_user_table, count * sizeof (sa_attr_type_t)); 988 sa->sa_user_table = NULL; 989 sa_free_attr_table(sa); 990 ASSERT(error != 0); 991 return (error); 992} 993 994int 995sa_setup(objset_t *os, uint64_t sa_obj, const sa_attr_reg_t *reg_attrs, 996 int count, sa_attr_type_t **user_table) 997{ 998 zap_cursor_t zc; 999 zap_attribute_t za; 1000 sa_os_t *sa; 1001 dmu_objset_type_t ostype = dmu_objset_type(os); 1002 sa_attr_type_t *tb; 1003 int error; 1004 1005 mutex_enter(&os->os_user_ptr_lock); 1006 if (os->os_sa) { 1007 mutex_enter(&os->os_sa->sa_lock); 1008 mutex_exit(&os->os_user_ptr_lock); 1009 tb = os->os_sa->sa_user_table; 1010 mutex_exit(&os->os_sa->sa_lock); 1011 *user_table = tb; 1012 return (0); 1013 } 1014 1015 sa = kmem_zalloc(sizeof (sa_os_t), KM_SLEEP); 1016 mutex_init(&sa->sa_lock, NULL, MUTEX_NOLOCKDEP, NULL); 1017 sa->sa_master_obj = sa_obj; 1018 1019 os->os_sa = sa; 1020 mutex_enter(&sa->sa_lock); 1021 mutex_exit(&os->os_user_ptr_lock); 1022 avl_create(&sa->sa_layout_num_tree, layout_num_compare, 1023 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node)); 1024 avl_create(&sa->sa_layout_hash_tree, layout_hash_compare, 1025 sizeof (sa_lot_t), offsetof(sa_lot_t, lot_hash_node)); 1026 1027 if (sa_obj) { 1028 error = zap_lookup(os, sa_obj, SA_LAYOUTS, 1029 8, 1, &sa->sa_layout_attr_obj); 1030 if (error != 0 && error != ENOENT) 1031 goto fail; 1032 error = zap_lookup(os, sa_obj, SA_REGISTRY, 1033 8, 1, &sa->sa_reg_attr_obj); 1034 if (error != 0 && error != ENOENT) 1035 goto fail; 1036 } 1037 1038 if ((error = sa_attr_table_setup(os, reg_attrs, count)) != 0) 1039 goto fail; 1040 1041 if (sa->sa_layout_attr_obj != 0) { 1042 uint64_t layout_count; 1043 1044 error = zap_count(os, sa->sa_layout_attr_obj, 1045 &layout_count); 1046 1047 /* 1048 * Layout number count should be > 0 1049 */ 1050 if (error || (error == 0 && layout_count == 0)) { 1051 if (error == 0) 1052 error = SET_ERROR(EINVAL); 1053 goto fail; 1054 } 1055 1056 for (zap_cursor_init(&zc, os, sa->sa_layout_attr_obj); 1057 (error = zap_cursor_retrieve(&zc, &za)) == 0; 1058 zap_cursor_advance(&zc)) { 1059 sa_attr_type_t *lot_attrs; 1060 uint64_t lot_num; 1061 1062 lot_attrs = kmem_zalloc(sizeof (sa_attr_type_t) * 1063 za.za_num_integers, KM_SLEEP); 1064 1065 if ((error = (zap_lookup(os, sa->sa_layout_attr_obj, 1066 za.za_name, 2, za.za_num_integers, 1067 lot_attrs))) != 0) { 1068 kmem_free(lot_attrs, sizeof (sa_attr_type_t) * 1069 za.za_num_integers); 1070 break; 1071 } 1072 VERIFY0(ddi_strtoull(za.za_name, NULL, 10, 1073 (unsigned long long *)&lot_num)); 1074 1075 (void) sa_add_layout_entry(os, lot_attrs, 1076 za.za_num_integers, lot_num, 1077 sa_layout_info_hash(lot_attrs, 1078 za.za_num_integers), B_FALSE, NULL); 1079 kmem_free(lot_attrs, sizeof (sa_attr_type_t) * 1080 za.za_num_integers); 1081 } 1082 zap_cursor_fini(&zc); 1083 1084 /* 1085 * Make sure layout count matches number of entries added 1086 * to AVL tree 1087 */ 1088 if (avl_numnodes(&sa->sa_layout_num_tree) != layout_count) { 1089 ASSERT(error != 0); 1090 goto fail; 1091 } 1092 } 1093 1094 /* Add special layout number for old ZNODES */ 1095 if (ostype == DMU_OST_ZFS) { 1096 (void) sa_add_layout_entry(os, sa_legacy_zpl_layout, 1097 sa_legacy_attr_count, 0, 1098 sa_layout_info_hash(sa_legacy_zpl_layout, 1099 sa_legacy_attr_count), B_FALSE, NULL); 1100 1101 (void) sa_add_layout_entry(os, sa_dummy_zpl_layout, 0, 1, 1102 0, B_FALSE, NULL); 1103 } 1104 *user_table = os->os_sa->sa_user_table; 1105 mutex_exit(&sa->sa_lock); 1106 return (0); 1107fail: 1108 os->os_sa = NULL; 1109 sa_free_attr_table(sa); 1110 if (sa->sa_user_table) 1111 kmem_free(sa->sa_user_table, sa->sa_user_table_sz); 1112 mutex_exit(&sa->sa_lock); 1113 avl_destroy(&sa->sa_layout_hash_tree); 1114 avl_destroy(&sa->sa_layout_num_tree); 1115 mutex_destroy(&sa->sa_lock); 1116 kmem_free(sa, sizeof (sa_os_t)); 1117 return ((error == ECKSUM) ? EIO : error); 1118} 1119 1120void 1121sa_tear_down(objset_t *os) 1122{ 1123 sa_os_t *sa = os->os_sa; 1124 sa_lot_t *layout; 1125 void *cookie; 1126 1127 kmem_free(sa->sa_user_table, sa->sa_user_table_sz); 1128 1129 /* Free up attr table */ 1130 1131 sa_free_attr_table(sa); 1132 1133 cookie = NULL; 1134 while ((layout = 1135 avl_destroy_nodes(&sa->sa_layout_hash_tree, &cookie))) { 1136 sa_idx_tab_t *tab; 1137 while ((tab = list_head(&layout->lot_idx_tab))) { 1138 ASSERT(zfs_refcount_count(&tab->sa_refcount)); 1139 sa_idx_tab_rele(os, tab); 1140 } 1141 } 1142 1143 cookie = NULL; 1144 while ((layout = avl_destroy_nodes(&sa->sa_layout_num_tree, &cookie))) { 1145 kmem_free(layout->lot_attrs, 1146 sizeof (sa_attr_type_t) * layout->lot_attr_count); 1147 kmem_free(layout, sizeof (sa_lot_t)); 1148 } 1149 1150 avl_destroy(&sa->sa_layout_hash_tree); 1151 avl_destroy(&sa->sa_layout_num_tree); 1152 mutex_destroy(&sa->sa_lock); 1153 1154 kmem_free(sa, sizeof (sa_os_t)); 1155 os->os_sa = NULL; 1156} 1157 1158static void 1159sa_build_idx_tab(void *hdr, void *attr_addr, sa_attr_type_t attr, 1160 uint16_t length, int length_idx, boolean_t var_length, void *userp) 1161{ 1162 sa_idx_tab_t *idx_tab = userp; 1163 1164 if (var_length) { 1165 ASSERT(idx_tab->sa_variable_lengths); 1166 idx_tab->sa_variable_lengths[length_idx] = length; 1167 } 1168 TOC_ATTR_ENCODE(idx_tab->sa_idx_tab[attr], length_idx, 1169 (uint32_t)((uintptr_t)attr_addr - (uintptr_t)hdr)); 1170} 1171 1172static void 1173sa_attr_iter(objset_t *os, sa_hdr_phys_t *hdr, dmu_object_type_t type, 1174 sa_iterfunc_t func, sa_lot_t *tab, void *userp) 1175{ 1176 void *data_start; 1177 sa_lot_t *tb = tab; 1178 sa_lot_t search; 1179 avl_index_t loc; 1180 sa_os_t *sa = os->os_sa; 1181 int i; 1182 uint16_t *length_start = NULL; 1183 uint8_t length_idx = 0; 1184 1185 if (tab == NULL) { 1186 search.lot_num = SA_LAYOUT_NUM(hdr, type); 1187 tb = avl_find(&sa->sa_layout_num_tree, &search, &loc); 1188 ASSERT(tb); 1189 } 1190 1191 if (IS_SA_BONUSTYPE(type)) { 1192 data_start = (void *)P2ROUNDUP(((uintptr_t)hdr + 1193 offsetof(sa_hdr_phys_t, sa_lengths) + 1194 (sizeof (uint16_t) * tb->lot_var_sizes)), 8); 1195 length_start = hdr->sa_lengths; 1196 } else { 1197 data_start = hdr; 1198 } 1199 1200 for (i = 0; i != tb->lot_attr_count; i++) { 1201 int attr_length, reg_length; 1202 uint8_t idx_len; 1203 1204 reg_length = sa->sa_attr_table[tb->lot_attrs[i]].sa_length; 1205 IMPLY(reg_length == 0, IS_SA_BONUSTYPE(type)); 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 1222static void 1223sa_byteswap_cb(void *hdr, void *attr_addr, sa_attr_type_t attr, 1224 uint16_t length, int length_idx, boolean_t variable_length, void *userp) 1225{ 1226 (void) hdr, (void) length_idx, (void) variable_length; 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 1233static void 1234sa_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 sa_os_t *sa __maybe_unused = 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 1274static int 1275sa_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 (u_longlong_t)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 1313static void 1314sa_evict_sync(void *dbu) 1315{ 1316 (void) dbu; 1317 panic("evicting sa dbuf\n"); 1318} 1319 1320static void 1321sa_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 (zfs_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 zfs_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 1344static void 1345sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab) 1346{ 1347 sa_os_t *sa __maybe_unused = os->os_sa; 1348 1349 ASSERT(MUTEX_HELD(&sa->sa_lock)); 1350 (void) zfs_refcount_add(&idx_tab->sa_refcount, NULL); 1351} 1352 1353void 1354sa_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 1366void 1367sa_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(hdl->sa_spill, NULL); 1384 mutex_exit(&hdl->sa_lock); 1385 1386 kmem_cache_free(sa_cache, hdl); 1387} 1388 1389int 1390sa_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 1439int 1440sa_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 1453int 1454sa_buf_hold(objset_t *objset, uint64_t obj_num, const void *tag, dmu_buf_t **db) 1455{ 1456 return (dmu_bonus_hold(objset, obj_num, tag, db)); 1457} 1458 1459void 1460sa_buf_rele(dmu_buf_t *db, const void *tag) 1461{ 1462 dmu_buf_rele(db, tag); 1463} 1464 1465static int 1466sa_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 1473static int 1474sa_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 1492int 1493sa_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 1505int 1506sa_lookup_uio(sa_handle_t *hdl, sa_attr_type_t attr, zfs_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 = zfs_uiomove((void *)bulk.sa_addr, MIN(bulk.sa_size, 1520 zfs_uio_resid(uio)), 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 */ 1532int 1533sa_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 (Z_ISBLK(ZTOTYPE(zp)) || Z_ISCHR(ZTOTYPE(zp))) 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 = ZTONLNK(zp); 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 (Z_ISBLK(ZTOTYPE(zp)) || Z_ISCHR(ZTOTYPE(zp))) 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 memcpy(scanstamp, 1670 (caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE, 1671 AV_SCANSTAMP_SZ); 1672 SA_ADD_BULK_ATTR(attrs, count, SA_ZPL_SCANSTAMP(zfsvfs), NULL, 1673 scanstamp, AV_SCANSTAMP_SZ); 1674 zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP; 1675 } 1676 1677 VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0); 1678 VERIFY(sa_replace_all_by_template_locked(hdl, attrs, count, tx) == 0); 1679 if (znode_acl.z_acl_extern_obj) { 1680 VERIFY(0 == dmu_object_free(zfsvfs->z_os, 1681 znode_acl.z_acl_extern_obj, tx)); 1682 } 1683 1684 zp->z_is_sa = B_TRUE; 1685 1686out: 1687 mutex_exit(&zp->z_lock); 1688 mutex_exit(&hdl->sa_lock); 1689 kmem_free(attrs, sizeof (sa_bulk_attr_t) * ZPL_END); 1690 kmem_free(bulk, sizeof (sa_bulk_attr_t) * ZPL_END); 1691 return (err); 1692} 1693#endif 1694 1695static sa_idx_tab_t * 1696sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype, sa_hdr_phys_t *hdr) 1697{ 1698 sa_idx_tab_t *idx_tab; 1699 sa_os_t *sa = os->os_sa; 1700 sa_lot_t *tb, search; 1701 avl_index_t loc; 1702 1703 /* 1704 * Deterimine layout number. If SA node and header == 0 then 1705 * force the index table to the dummy "1" empty layout. 1706 * 1707 * The layout number would only be zero for a newly created file 1708 * that has not added any attributes yet, or with crypto enabled which 1709 * doesn't write any attributes to the bonus buffer. 1710 */ 1711 1712 search.lot_num = SA_LAYOUT_NUM(hdr, bonustype); 1713 1714 tb = avl_find(&sa->sa_layout_num_tree, &search, &loc); 1715 1716 /* Verify header size is consistent with layout information */ 1717 ASSERT(tb); 1718 ASSERT((IS_SA_BONUSTYPE(bonustype) && 1719 SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb)) || !IS_SA_BONUSTYPE(bonustype) || 1720 (IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0)); 1721 1722 /* 1723 * See if any of the already existing TOC entries can be reused? 1724 */ 1725 1726 for (idx_tab = list_head(&tb->lot_idx_tab); idx_tab; 1727 idx_tab = list_next(&tb->lot_idx_tab, idx_tab)) { 1728 boolean_t valid_idx = B_TRUE; 1729 int i; 1730 1731 if (tb->lot_var_sizes != 0 && 1732 idx_tab->sa_variable_lengths != NULL) { 1733 for (i = 0; i != tb->lot_var_sizes; i++) { 1734 if (hdr->sa_lengths[i] != 1735 idx_tab->sa_variable_lengths[i]) { 1736 valid_idx = B_FALSE; 1737 break; 1738 } 1739 } 1740 } 1741 if (valid_idx) { 1742 sa_idx_tab_hold(os, idx_tab); 1743 return (idx_tab); 1744 } 1745 } 1746 1747 /* No such luck, create a new entry */ 1748 idx_tab = kmem_zalloc(sizeof (sa_idx_tab_t), KM_SLEEP); 1749 idx_tab->sa_idx_tab = 1750 kmem_zalloc(sizeof (uint32_t) * sa->sa_num_attrs, KM_SLEEP); 1751 idx_tab->sa_layout = tb; 1752 zfs_refcount_create(&idx_tab->sa_refcount); 1753 if (tb->lot_var_sizes) 1754 idx_tab->sa_variable_lengths = kmem_alloc(sizeof (uint16_t) * 1755 tb->lot_var_sizes, KM_SLEEP); 1756 1757 sa_attr_iter(os, hdr, bonustype, sa_build_idx_tab, 1758 tb, idx_tab); 1759 sa_idx_tab_hold(os, idx_tab); /* one hold for consumer */ 1760 sa_idx_tab_hold(os, idx_tab); /* one for layout */ 1761 list_insert_tail(&tb->lot_idx_tab, idx_tab); 1762 return (idx_tab); 1763} 1764 1765void 1766sa_default_locator(void **dataptr, uint32_t *len, uint32_t total_len, 1767 boolean_t start, void *userdata) 1768{ 1769 ASSERT(start); 1770 1771 *dataptr = userdata; 1772 *len = total_len; 1773} 1774 1775static void 1776sa_attr_register_sync(sa_handle_t *hdl, dmu_tx_t *tx) 1777{ 1778 uint64_t attr_value = 0; 1779 sa_os_t *sa = hdl->sa_os->os_sa; 1780 sa_attr_table_t *tb = sa->sa_attr_table; 1781 int i; 1782 1783 mutex_enter(&sa->sa_lock); 1784 1785 if (!sa->sa_need_attr_registration || sa->sa_master_obj == 0) { 1786 mutex_exit(&sa->sa_lock); 1787 return; 1788 } 1789 1790 if (sa->sa_reg_attr_obj == 0) { 1791 sa->sa_reg_attr_obj = zap_create_link(hdl->sa_os, 1792 DMU_OT_SA_ATTR_REGISTRATION, 1793 sa->sa_master_obj, SA_REGISTRY, tx); 1794 } 1795 for (i = 0; i != sa->sa_num_attrs; i++) { 1796 if (sa->sa_attr_table[i].sa_registered) 1797 continue; 1798 ATTR_ENCODE(attr_value, tb[i].sa_attr, tb[i].sa_length, 1799 tb[i].sa_byteswap); 1800 VERIFY(0 == zap_update(hdl->sa_os, sa->sa_reg_attr_obj, 1801 tb[i].sa_name, 8, 1, &attr_value, tx)); 1802 tb[i].sa_registered = B_TRUE; 1803 } 1804 sa->sa_need_attr_registration = B_FALSE; 1805 mutex_exit(&sa->sa_lock); 1806} 1807 1808/* 1809 * Replace all attributes with attributes specified in template. 1810 * If dnode had a spill buffer then those attributes will be 1811 * also be replaced, possibly with just an empty spill block 1812 * 1813 * This interface is intended to only be used for bulk adding of 1814 * attributes for a new file. It will also be used by the ZPL 1815 * when converting and old formatted znode to native SA support. 1816 */ 1817int 1818sa_replace_all_by_template_locked(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, 1819 int attr_count, dmu_tx_t *tx) 1820{ 1821 sa_os_t *sa = hdl->sa_os->os_sa; 1822 1823 if (sa->sa_need_attr_registration) 1824 sa_attr_register_sync(hdl, tx); 1825 return (sa_build_layouts(hdl, attr_desc, attr_count, tx)); 1826} 1827 1828int 1829sa_replace_all_by_template(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, 1830 int attr_count, dmu_tx_t *tx) 1831{ 1832 int error; 1833 1834 mutex_enter(&hdl->sa_lock); 1835 error = sa_replace_all_by_template_locked(hdl, attr_desc, 1836 attr_count, tx); 1837 mutex_exit(&hdl->sa_lock); 1838 return (error); 1839} 1840 1841/* 1842 * Add/remove a single attribute or replace a variable-sized attribute value 1843 * with a value of a different size, and then rewrite the entire set 1844 * of attributes. 1845 * Same-length attribute value replacement (including fixed-length attributes) 1846 * is handled more efficiently by the upper layers. 1847 */ 1848static int 1849sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr, 1850 sa_data_op_t action, sa_data_locator_t *locator, void *datastart, 1851 uint16_t buflen, dmu_tx_t *tx) 1852{ 1853 sa_os_t *sa = hdl->sa_os->os_sa; 1854 dmu_buf_impl_t *db = (dmu_buf_impl_t *)hdl->sa_bonus; 1855 dnode_t *dn; 1856 sa_bulk_attr_t *attr_desc; 1857 void *old_data[2]; 1858 int bonus_attr_count = 0; 1859 int bonus_data_size = 0; 1860 int spill_data_size = 0; 1861 int spill_attr_count = 0; 1862 int error; 1863 uint16_t length, reg_length; 1864 int i, j, k, length_idx; 1865 sa_hdr_phys_t *hdr; 1866 sa_idx_tab_t *idx_tab; 1867 int attr_count; 1868 int count; 1869 1870 ASSERT(MUTEX_HELD(&hdl->sa_lock)); 1871 1872 /* First make of copy of the old data */ 1873 1874 DB_DNODE_ENTER(db); 1875 dn = DB_DNODE(db); 1876 if (dn->dn_bonuslen != 0) { 1877 bonus_data_size = hdl->sa_bonus->db_size; 1878 old_data[0] = kmem_alloc(bonus_data_size, KM_SLEEP); 1879 memcpy(old_data[0], hdl->sa_bonus->db_data, 1880 hdl->sa_bonus->db_size); 1881 bonus_attr_count = hdl->sa_bonus_tab->sa_layout->lot_attr_count; 1882 } else { 1883 old_data[0] = NULL; 1884 } 1885 DB_DNODE_EXIT(db); 1886 1887 /* Bring spill buffer online if it isn't currently */ 1888 1889 if ((error = sa_get_spill(hdl)) == 0) { 1890 spill_data_size = hdl->sa_spill->db_size; 1891 old_data[1] = vmem_alloc(spill_data_size, KM_SLEEP); 1892 memcpy(old_data[1], hdl->sa_spill->db_data, 1893 hdl->sa_spill->db_size); 1894 spill_attr_count = 1895 hdl->sa_spill_tab->sa_layout->lot_attr_count; 1896 } else if (error && error != ENOENT) { 1897 if (old_data[0]) 1898 kmem_free(old_data[0], bonus_data_size); 1899 return (error); 1900 } else { 1901 old_data[1] = NULL; 1902 } 1903 1904 /* build descriptor of all attributes */ 1905 1906 attr_count = bonus_attr_count + spill_attr_count; 1907 if (action == SA_ADD) 1908 attr_count++; 1909 else if (action == SA_REMOVE) 1910 attr_count--; 1911 1912 attr_desc = kmem_zalloc(sizeof (sa_bulk_attr_t) * attr_count, KM_SLEEP); 1913 1914 /* 1915 * loop through bonus and spill buffer if it exists, and 1916 * build up new attr_descriptor to reset the attributes 1917 */ 1918 k = j = 0; 1919 count = bonus_attr_count; 1920 hdr = SA_GET_HDR(hdl, SA_BONUS); 1921 idx_tab = SA_IDX_TAB_GET(hdl, SA_BONUS); 1922 for (; ; k++) { 1923 /* 1924 * Iterate over each attribute in layout. Fetch the 1925 * size of variable-length attributes needing rewrite 1926 * from sa_lengths[]. 1927 */ 1928 for (i = 0, length_idx = 0; i != count; i++) { 1929 sa_attr_type_t attr; 1930 1931 attr = idx_tab->sa_layout->lot_attrs[i]; 1932 reg_length = SA_REGISTERED_LEN(sa, attr); 1933 if (reg_length == 0) { 1934 length = hdr->sa_lengths[length_idx]; 1935 length_idx++; 1936 } else { 1937 length = reg_length; 1938 } 1939 if (attr == newattr) { 1940 /* 1941 * There is nothing to do for SA_REMOVE, 1942 * so it is just skipped. 1943 */ 1944 if (action == SA_REMOVE) 1945 continue; 1946 1947 /* 1948 * Duplicate attributes are not allowed, so the 1949 * action can not be SA_ADD here. 1950 */ 1951 ASSERT3S(action, ==, SA_REPLACE); 1952 1953 /* 1954 * Only a variable-sized attribute can be 1955 * replaced here, and its size must be changing. 1956 */ 1957 ASSERT3U(reg_length, ==, 0); 1958 ASSERT3U(length, !=, buflen); 1959 SA_ADD_BULK_ATTR(attr_desc, j, attr, 1960 locator, datastart, buflen); 1961 } else { 1962 SA_ADD_BULK_ATTR(attr_desc, j, attr, 1963 NULL, (void *) 1964 (TOC_OFF(idx_tab->sa_idx_tab[attr]) + 1965 (uintptr_t)old_data[k]), length); 1966 } 1967 } 1968 if (k == 0 && hdl->sa_spill) { 1969 hdr = SA_GET_HDR(hdl, SA_SPILL); 1970 idx_tab = SA_IDX_TAB_GET(hdl, SA_SPILL); 1971 count = spill_attr_count; 1972 } else { 1973 break; 1974 } 1975 } 1976 if (action == SA_ADD) { 1977 reg_length = SA_REGISTERED_LEN(sa, newattr); 1978 IMPLY(reg_length != 0, reg_length == buflen); 1979 SA_ADD_BULK_ATTR(attr_desc, j, newattr, locator, 1980 datastart, buflen); 1981 } 1982 ASSERT3U(j, ==, attr_count); 1983 1984 error = sa_build_layouts(hdl, attr_desc, attr_count, tx); 1985 1986 if (old_data[0]) 1987 kmem_free(old_data[0], bonus_data_size); 1988 if (old_data[1]) 1989 vmem_free(old_data[1], spill_data_size); 1990 kmem_free(attr_desc, sizeof (sa_bulk_attr_t) * attr_count); 1991 1992 return (error); 1993} 1994 1995static int 1996sa_bulk_update_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count, 1997 dmu_tx_t *tx) 1998{ 1999 int error; 2000 sa_os_t *sa = hdl->sa_os->os_sa; 2001 dmu_object_type_t bonustype; 2002 dmu_buf_t *saved_spill; 2003 2004 ASSERT(hdl); 2005 ASSERT(MUTEX_HELD(&hdl->sa_lock)); 2006 2007 bonustype = SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl, SA_BONUS)); 2008 saved_spill = hdl->sa_spill; 2009 2010 /* sync out registration table if necessary */ 2011 if (sa->sa_need_attr_registration) 2012 sa_attr_register_sync(hdl, tx); 2013 2014 error = sa_attr_op(hdl, bulk, count, SA_UPDATE, tx); 2015 if (error == 0 && !IS_SA_BONUSTYPE(bonustype) && sa->sa_update_cb) 2016 sa->sa_update_cb(hdl, tx); 2017 2018 /* 2019 * If saved_spill is NULL and current sa_spill is not NULL that 2020 * means we increased the refcount of the spill buffer through 2021 * sa_get_spill() or dmu_spill_hold_by_dnode(). Therefore we 2022 * must release the hold before calling dmu_tx_commit() to avoid 2023 * making a copy of this buffer in dbuf_sync_leaf() due to the 2024 * reference count now being greater than 1. 2025 */ 2026 if (!saved_spill && hdl->sa_spill) { 2027 if (hdl->sa_spill_tab) { 2028 sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab); 2029 hdl->sa_spill_tab = NULL; 2030 } 2031 2032 dmu_buf_rele(hdl->sa_spill, NULL); 2033 hdl->sa_spill = NULL; 2034 } 2035 2036 return (error); 2037} 2038 2039/* 2040 * update or add new attribute 2041 */ 2042int 2043sa_update(sa_handle_t *hdl, sa_attr_type_t type, 2044 void *buf, uint32_t buflen, dmu_tx_t *tx) 2045{ 2046 int error; 2047 sa_bulk_attr_t bulk; 2048 2049 VERIFY3U(buflen, <=, SA_ATTR_MAX_LEN); 2050 2051 bulk.sa_attr = type; 2052 bulk.sa_data_func = NULL; 2053 bulk.sa_length = buflen; 2054 bulk.sa_data = buf; 2055 2056 mutex_enter(&hdl->sa_lock); 2057 error = sa_bulk_update_impl(hdl, &bulk, 1, tx); 2058 mutex_exit(&hdl->sa_lock); 2059 return (error); 2060} 2061 2062/* 2063 * Return size of an attribute 2064 */ 2065 2066int 2067sa_size(sa_handle_t *hdl, sa_attr_type_t attr, int *size) 2068{ 2069 sa_bulk_attr_t bulk; 2070 int error; 2071 2072 bulk.sa_data = NULL; 2073 bulk.sa_attr = attr; 2074 bulk.sa_data_func = NULL; 2075 2076 ASSERT(hdl); 2077 mutex_enter(&hdl->sa_lock); 2078 if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) != 0) { 2079 mutex_exit(&hdl->sa_lock); 2080 return (error); 2081 } 2082 *size = bulk.sa_size; 2083 2084 mutex_exit(&hdl->sa_lock); 2085 return (0); 2086} 2087 2088int 2089sa_bulk_lookup_locked(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count) 2090{ 2091 ASSERT(hdl); 2092 ASSERT(MUTEX_HELD(&hdl->sa_lock)); 2093 return (sa_lookup_impl(hdl, attrs, count)); 2094} 2095 2096int 2097sa_bulk_lookup(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count) 2098{ 2099 int error; 2100 2101 ASSERT(hdl); 2102 mutex_enter(&hdl->sa_lock); 2103 error = sa_bulk_lookup_locked(hdl, attrs, count); 2104 mutex_exit(&hdl->sa_lock); 2105 return (error); 2106} 2107 2108int 2109sa_bulk_update(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count, dmu_tx_t *tx) 2110{ 2111 int error; 2112 2113 ASSERT(hdl); 2114 mutex_enter(&hdl->sa_lock); 2115 error = sa_bulk_update_impl(hdl, attrs, count, tx); 2116 mutex_exit(&hdl->sa_lock); 2117 return (error); 2118} 2119 2120int 2121sa_remove(sa_handle_t *hdl, sa_attr_type_t attr, dmu_tx_t *tx) 2122{ 2123 int error; 2124 2125 mutex_enter(&hdl->sa_lock); 2126 error = sa_modify_attrs(hdl, attr, SA_REMOVE, NULL, 2127 NULL, 0, tx); 2128 mutex_exit(&hdl->sa_lock); 2129 return (error); 2130} 2131 2132void 2133sa_object_info(sa_handle_t *hdl, dmu_object_info_t *doi) 2134{ 2135 dmu_object_info_from_db(hdl->sa_bonus, doi); 2136} 2137 2138void 2139sa_object_size(sa_handle_t *hdl, uint32_t *blksize, u_longlong_t *nblocks) 2140{ 2141 dmu_object_size_from_db(hdl->sa_bonus, 2142 blksize, nblocks); 2143} 2144 2145void 2146sa_set_userp(sa_handle_t *hdl, void *ptr) 2147{ 2148 hdl->sa_userp = ptr; 2149} 2150 2151dmu_buf_t * 2152sa_get_db(sa_handle_t *hdl) 2153{ 2154 return (hdl->sa_bonus); 2155} 2156 2157void * 2158sa_get_userdata(sa_handle_t *hdl) 2159{ 2160 return (hdl->sa_userp); 2161} 2162 2163void 2164sa_register_update_callback_locked(objset_t *os, sa_update_cb_t *func) 2165{ 2166 ASSERT(MUTEX_HELD(&os->os_sa->sa_lock)); 2167 os->os_sa->sa_update_cb = func; 2168} 2169 2170void 2171sa_register_update_callback(objset_t *os, sa_update_cb_t *func) 2172{ 2173 2174 mutex_enter(&os->os_sa->sa_lock); 2175 sa_register_update_callback_locked(os, func); 2176 mutex_exit(&os->os_sa->sa_lock); 2177} 2178 2179uint64_t 2180sa_handle_object(sa_handle_t *hdl) 2181{ 2182 return (hdl->sa_bonus->db_object); 2183} 2184 2185boolean_t 2186sa_enabled(objset_t *os) 2187{ 2188 return (os->os_sa == NULL); 2189} 2190 2191int 2192sa_set_sa_object(objset_t *os, uint64_t sa_object) 2193{ 2194 sa_os_t *sa = os->os_sa; 2195 2196 if (sa->sa_master_obj) 2197 return (1); 2198 2199 sa->sa_master_obj = sa_object; 2200 2201 return (0); 2202} 2203 2204int 2205sa_hdrsize(void *arg) 2206{ 2207 sa_hdr_phys_t *hdr = arg; 2208 2209 return (SA_HDR_SIZE(hdr)); 2210} 2211 2212void 2213sa_handle_lock(sa_handle_t *hdl) 2214{ 2215 ASSERT(hdl); 2216 mutex_enter(&hdl->sa_lock); 2217} 2218 2219void 2220sa_handle_unlock(sa_handle_t *hdl) 2221{ 2222 ASSERT(hdl); 2223 mutex_exit(&hdl->sa_lock); 2224} 2225 2226#ifdef _KERNEL 2227EXPORT_SYMBOL(sa_handle_get); 2228EXPORT_SYMBOL(sa_handle_get_from_db); 2229EXPORT_SYMBOL(sa_handle_destroy); 2230EXPORT_SYMBOL(sa_buf_hold); 2231EXPORT_SYMBOL(sa_buf_rele); 2232EXPORT_SYMBOL(sa_spill_rele); 2233EXPORT_SYMBOL(sa_lookup); 2234EXPORT_SYMBOL(sa_update); 2235EXPORT_SYMBOL(sa_remove); 2236EXPORT_SYMBOL(sa_bulk_lookup); 2237EXPORT_SYMBOL(sa_bulk_lookup_locked); 2238EXPORT_SYMBOL(sa_bulk_update); 2239EXPORT_SYMBOL(sa_size); 2240EXPORT_SYMBOL(sa_object_info); 2241EXPORT_SYMBOL(sa_object_size); 2242EXPORT_SYMBOL(sa_get_userdata); 2243EXPORT_SYMBOL(sa_set_userp); 2244EXPORT_SYMBOL(sa_get_db); 2245EXPORT_SYMBOL(sa_handle_object); 2246EXPORT_SYMBOL(sa_register_update_callback); 2247EXPORT_SYMBOL(sa_setup); 2248EXPORT_SYMBOL(sa_replace_all_by_template); 2249EXPORT_SYMBOL(sa_replace_all_by_template_locked); 2250EXPORT_SYMBOL(sa_enabled); 2251EXPORT_SYMBOL(sa_cache_init); 2252EXPORT_SYMBOL(sa_cache_fini); 2253EXPORT_SYMBOL(sa_set_sa_object); 2254EXPORT_SYMBOL(sa_hdrsize); 2255EXPORT_SYMBOL(sa_handle_lock); 2256EXPORT_SYMBOL(sa_handle_unlock); 2257EXPORT_SYMBOL(sa_lookup_uio); 2258EXPORT_SYMBOL(sa_add_projid); 2259#endif /* _KERNEL */ 2260