1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * CDDL HEADER START 4 * 5 * The contents of this file are subject to the terms of the 6 * Common Development and Distribution License (the "License"). 7 * You may not use this file except in compliance with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or https://opensource.org/licenses/CDDL-1.0. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2012, 2018 by Delphix. All rights reserved. 25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26 */ 27 28 #ifndef _SYS_DNODE_H 29 #define _SYS_DNODE_H 30 31 #include <sys/zfs_context.h> 32 #include <sys/avl.h> 33 #include <sys/spa.h> 34 #include <sys/txg.h> 35 #include <sys/zio.h> 36 #include <sys/zfs_refcount.h> 37 #include <sys/dmu_zfetch.h> 38 #include <sys/zrlock.h> 39 #include <sys/multilist.h> 40 #include <sys/wmsum.h> 41 42 #ifdef __cplusplus 43 extern "C" { 44 #endif 45 46 /* 47 * dnode_hold() flags. 48 */ 49 #define DNODE_MUST_BE_ALLOCATED 1 50 #define DNODE_MUST_BE_FREE 2 51 #define DNODE_DRY_RUN 4 52 53 /* 54 * dnode_next_offset() flags. 55 */ 56 #define DNODE_FIND_HOLE 1 57 #define DNODE_FIND_BACKWARDS 2 58 #define DNODE_FIND_HAVELOCK 4 59 60 /* 61 * Fixed constants. 62 */ 63 #define DNODE_SHIFT 9 /* 512 bytes */ 64 #define DN_MIN_INDBLKSHIFT 12 /* 4k */ 65 /* 66 * If we ever increase this value beyond 20, we need to revisit all logic that 67 * does x << level * ebps to handle overflow. With a 1M indirect block size, 68 * 4 levels of indirect blocks would not be able to guarantee addressing an 69 * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65. 70 */ 71 #define DN_MAX_INDBLKSHIFT 17 /* 128k */ 72 #define DNODE_BLOCK_SHIFT 14 /* 16k */ 73 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */ 74 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */ 75 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */ 76 77 /* 78 * dnode id flags 79 * 80 * Note: a file will never ever have its ids moved from bonus->spill 81 */ 82 #define DN_ID_CHKED_BONUS 0x1 83 #define DN_ID_CHKED_SPILL 0x2 84 #define DN_ID_OLD_EXIST 0x4 85 #define DN_ID_NEW_EXIST 0x8 86 87 /* 88 * Derived constants. 89 */ 90 #define DNODE_MIN_SIZE (1 << DNODE_SHIFT) 91 #define DNODE_MAX_SIZE (1 << DNODE_BLOCK_SHIFT) 92 #define DNODE_BLOCK_SIZE (1 << DNODE_BLOCK_SHIFT) 93 #define DNODE_MIN_SLOTS (DNODE_MIN_SIZE >> DNODE_SHIFT) 94 #define DNODE_MAX_SLOTS (DNODE_MAX_SIZE >> DNODE_SHIFT) 95 #define DN_BONUS_SIZE(dnsize) ((dnsize) - DNODE_CORE_SIZE - \ 96 (1 << SPA_BLKPTRSHIFT)) 97 #define DN_SLOTS_TO_BONUSLEN(slots) DN_BONUS_SIZE((slots) << DNODE_SHIFT) 98 #define DN_OLD_MAX_BONUSLEN (DN_BONUS_SIZE(DNODE_MIN_SIZE)) 99 #define DN_MAX_NBLKPTR ((DNODE_MIN_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT) 100 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT) 101 #define DN_ZERO_BONUSLEN (DN_BONUS_SIZE(DNODE_MAX_SIZE) + 1) 102 #define DN_KILL_SPILLBLK (1) 103 104 #define DN_SLOT_UNINIT ((void *)NULL) /* Uninitialized */ 105 #define DN_SLOT_FREE ((void *)1UL) /* Free slot */ 106 #define DN_SLOT_ALLOCATED ((void *)2UL) /* Allocated slot */ 107 #define DN_SLOT_INTERIOR ((void *)3UL) /* Interior allocated slot */ 108 #define DN_SLOT_IS_PTR(dn) ((void *)dn > DN_SLOT_INTERIOR) 109 #define DN_SLOT_IS_VALID(dn) ((void *)dn != NULL) 110 111 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT) 112 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT) 113 114 /* 115 * This is inaccurate if the indblkshift of the particular object is not the 116 * max. But it's only used by userland to calculate the zvol reservation. 117 */ 118 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT) 119 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT) 120 121 #define DN_MAX_LEVELS (DIV_ROUND_UP(DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT, \ 122 DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT) + 1) 123 124 /* 125 * Use the flexible array instead of the fixed length one dn_bonus 126 * to address memcpy/memmove fortify error 127 */ 128 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus_flexible + \ 129 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) 130 #define DN_MAX_BONUS_LEN(dnp) \ 131 ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ? \ 132 (uint8_t *)DN_SPILL_BLKPTR(dnp) - (uint8_t *)DN_BONUS(dnp) : \ 133 (uint8_t *)(dnp + (dnp->dn_extra_slots + 1)) - (uint8_t *)DN_BONUS(dnp)) 134 135 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ 136 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) 137 138 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) 139 140 struct dmu_buf_impl; 141 struct objset; 142 struct zio; 143 144 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ 145 #define DNODE_FLAG_USED_BYTES (1 << 0) 146 #define DNODE_FLAG_USERUSED_ACCOUNTED (1 << 1) 147 148 /* Does dnode have a SA spill blkptr in bonus? */ 149 #define DNODE_FLAG_SPILL_BLKPTR (1 << 2) 150 151 /* User/Group/Project dnode accounting */ 152 #define DNODE_FLAG_USEROBJUSED_ACCOUNTED (1 << 3) 153 154 /* 155 * This mask defines the set of flags which are "portable", meaning 156 * that they can be preserved when doing a raw encrypted zfs send. 157 * Flags included in this mask will be protected by AAD when the block 158 * of dnodes is encrypted. 159 */ 160 #define DNODE_CRYPT_PORTABLE_FLAGS_MASK (DNODE_FLAG_SPILL_BLKPTR) 161 162 /* 163 * VARIABLE-LENGTH (LARGE) DNODES 164 * 165 * The motivation for variable-length dnodes is to eliminate the overhead 166 * associated with using spill blocks. Spill blocks are used to store 167 * system attribute data (i.e. file metadata) that does not fit in the 168 * dnode's bonus buffer. By allowing a larger bonus buffer area the use of 169 * a spill block can be avoided. Spill blocks potentially incur an 170 * additional read I/O for every dnode in a dnode block. As a worst case 171 * example, reading 32 dnodes from a 16k dnode block and all of the spill 172 * blocks could issue 33 separate reads. Now suppose those dnodes have size 173 * 1024 and therefore don't need spill blocks. Then the worst case number 174 * of blocks read is reduced from 33 to two--one per dnode block. 175 * 176 * ZFS-on-Linux systems that make heavy use of extended attributes benefit 177 * from this feature. In particular, ZFS-on-Linux supports the xattr=sa 178 * dataset property which allows file extended attribute data to be stored 179 * in the dnode bonus buffer as an alternative to the traditional 180 * directory-based format. Workloads such as SELinux and the Lustre 181 * distributed filesystem often store enough xattr data to force spill 182 * blocks when xattr=sa is in effect. Large dnodes may therefore provide a 183 * performance benefit to such systems. Other use cases that benefit from 184 * this feature include files with large ACLs and symbolic links with long 185 * target names. 186 * 187 * The size of a dnode may be a multiple of 512 bytes up to the size of a 188 * dnode block (currently 16384 bytes). The dn_extra_slots field of the 189 * on-disk dnode_phys_t structure describes the size of the physical dnode 190 * on disk. The field represents how many "extra" dnode_phys_t slots a 191 * dnode consumes in its dnode block. This convention results in a value of 192 * 0 for 512 byte dnodes which preserves on-disk format compatibility with 193 * older software which doesn't support large dnodes. 194 * 195 * Similarly, the in-memory dnode_t structure has a dn_num_slots field 196 * to represent the total number of dnode_phys_t slots consumed on disk. 197 * Thus dn->dn_num_slots is 1 greater than the corresponding 198 * dnp->dn_extra_slots. This difference in convention was adopted 199 * because, unlike on-disk structures, backward compatibility is not a 200 * concern for in-memory objects, so we used a more natural way to 201 * represent size for a dnode_t. 202 * 203 * The default size for newly created dnodes is determined by the value of 204 * the "dnodesize" dataset property. By default the property is set to 205 * "legacy" which is compatible with older software. Setting the property 206 * to "auto" will allow the filesystem to choose the most suitable dnode 207 * size. Currently this just sets the default dnode size to 1k, but future 208 * code improvements could dynamically choose a size based on observed 209 * workload patterns. Dnodes of varying sizes can coexist within the same 210 * dataset and even within the same dnode block. 211 */ 212 213 typedef struct dnode_phys { 214 uint8_t dn_type; /* dmu_object_type_t */ 215 uint8_t dn_indblkshift; /* ln2(indirect block size) */ 216 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ 217 uint8_t dn_nblkptr; /* length of dn_blkptr */ 218 uint8_t dn_bonustype; /* type of data in bonus buffer */ 219 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 220 uint8_t dn_compress; /* ZIO_COMPRESS type */ 221 uint8_t dn_flags; /* DNODE_FLAG_* */ 222 uint16_t dn_datablkszsec; /* data block size in 512b sectors */ 223 uint16_t dn_bonuslen; /* length of dn_bonus */ 224 uint8_t dn_extra_slots; /* # of subsequent slots consumed */ 225 uint8_t dn_pad2[3]; 226 227 /* accounting is protected by dn_dirty_mtx */ 228 uint64_t dn_maxblkid; /* largest allocated block ID */ 229 uint64_t dn_used; /* bytes (or sectors) of disk space */ 230 231 /* 232 * Both dn_pad2 and dn_pad3 are protected by the block's MAC. This 233 * allows us to protect any fields that might be added here in the 234 * future. In either case, developers will want to check 235 * zio_crypt_init_uios_dnode() and zio_crypt_do_dnode_hmac_updates() 236 * to ensure the new field is being protected and updated properly. 237 */ 238 uint64_t dn_pad3[4]; 239 240 /* 241 * The tail region is 448 bytes for a 512 byte dnode, and 242 * correspondingly larger for larger dnode sizes. The spill 243 * block pointer, when present, is always at the end of the tail 244 * region. There are three ways this space may be used, using 245 * a 512 byte dnode for this diagram: 246 * 247 * 0 64 128 192 256 320 384 448 (offset) 248 * +---------------+---------------+---------------+-------+ 249 * | dn_blkptr[0] | dn_blkptr[1] | dn_blkptr[2] | / | 250 * +---------------+---------------+---------------+-------+ 251 * | dn_blkptr[0] | dn_bonus[0..319] | 252 * +---------------+-----------------------+---------------+ 253 * | dn_blkptr[0] | dn_bonus[0..191] | dn_spill | 254 * +---------------+-----------------------+---------------+ 255 */ 256 union { 257 blkptr_t dn_blkptr[1+DN_OLD_MAX_BONUSLEN/sizeof (blkptr_t)]; 258 struct { 259 blkptr_t __dn_ignore1; 260 uint8_t dn_bonus[DN_OLD_MAX_BONUSLEN]; 261 }; 262 struct { 263 blkptr_t __dn_ignore2; 264 uint8_t __dn_ignore3[DN_OLD_MAX_BONUSLEN - 265 sizeof (blkptr_t)]; 266 blkptr_t dn_spill; 267 }; 268 struct { 269 blkptr_t __dn_ignore4; 270 uint8_t dn_bonus_flexible[]; 271 }; 272 }; 273 } dnode_phys_t; 274 275 #define DN_SPILL_BLKPTR(dnp) ((blkptr_t *)((char *)(dnp) + \ 276 (((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT))) 277 278 struct dnode { 279 /* 280 * Protects the structure of the dnode, including the number of levels 281 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_* 282 */ 283 krwlock_t dn_struct_rwlock; 284 285 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ 286 list_node_t dn_link; 287 288 /* immutable: */ 289 struct objset *dn_objset; 290 uint64_t dn_object; 291 struct dmu_buf_impl *dn_dbuf; 292 struct dnode_handle *dn_handle; 293 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ 294 295 /* 296 * Copies of stuff in dn_phys. They're valid in the open 297 * context (eg. even before the dnode is first synced). 298 * Where necessary, these are protected by dn_struct_rwlock. 299 */ 300 dmu_object_type_t dn_type; /* object type */ 301 uint16_t dn_bonuslen; /* bonus length */ 302 uint8_t dn_bonustype; /* bonus type */ 303 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ 304 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 305 uint8_t dn_compress; /* ZIO_COMPRESS type */ 306 uint8_t dn_nlevels; 307 uint8_t dn_indblkshift; 308 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ 309 uint8_t dn_moved; /* Has this dnode been moved? */ 310 uint16_t dn_datablkszsec; /* in 512b sectors */ 311 uint32_t dn_datablksz; /* in bytes */ 312 uint64_t dn_maxblkid; 313 uint8_t dn_next_type[TXG_SIZE]; 314 uint8_t dn_num_slots; /* metadnode slots consumed on disk */ 315 uint8_t dn_next_nblkptr[TXG_SIZE]; 316 uint8_t dn_next_nlevels[TXG_SIZE]; 317 uint8_t dn_next_indblkshift[TXG_SIZE]; 318 uint8_t dn_next_bonustype[TXG_SIZE]; 319 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ 320 uint16_t dn_next_bonuslen[TXG_SIZE]; 321 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ 322 uint64_t dn_next_maxblkid[TXG_SIZE]; /* next maxblkid in bytes */ 323 324 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */ 325 uint32_t dn_dbufs_count; /* count of dn_dbufs */ 326 327 /* protected by os_lock: */ 328 multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ 329 330 /* protected by dn_mtx: */ 331 kmutex_t dn_mtx; 332 list_t dn_dirty_records[TXG_SIZE]; 333 struct zfs_range_tree *dn_free_ranges[TXG_SIZE]; 334 uint64_t dn_allocated_txg; 335 uint64_t dn_free_txg; 336 uint64_t dn_assigned_txg; 337 uint8_t dn_dirtycnt; 338 kcondvar_t dn_notxholds; 339 kcondvar_t dn_nodnholds; 340 341 /* protected by own devices */ 342 zfs_refcount_t dn_tx_holds; 343 zfs_refcount_t dn_holds; 344 345 kmutex_t dn_dbufs_mtx; 346 /* 347 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs 348 * can contain multiple dbufs of the same (level, blkid) when a 349 * dbuf is marked DB_EVICTING without being removed from 350 * dn_dbufs. To maintain the avl invariant that there cannot be 351 * duplicate entries, we order the dbufs by an arbitrary value - 352 * their address in memory. This means that dn_dbufs cannot be used to 353 * directly look up a dbuf. Instead, callers must use avl_walk, have 354 * a reference to the dbuf, or look up a non-existent node with 355 * db_state = DB_SEARCH (see dbuf_free_range for an example). 356 */ 357 avl_tree_t dn_dbufs; 358 359 /* protected by dn_struct_rwlock */ 360 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ 361 362 boolean_t dn_have_spill; /* have spill or are spilling */ 363 364 /* parent IO for current sync write */ 365 zio_t *dn_zio; 366 367 /* used in syncing context */ 368 uint64_t dn_oldused; /* old phys used bytes */ 369 uint64_t dn_oldflags; /* old phys dn_flags */ 370 uint64_t dn_olduid, dn_oldgid, dn_oldprojid; 371 uint64_t dn_newuid, dn_newgid, dn_newprojid; 372 int dn_id_flags; 373 374 /* holds prefetch structure */ 375 struct zfetch dn_zfetch; 376 377 /* Not in dn_phys, but should be. set it after taking a hold */ 378 dmu_object_type_t dn_storage_type; /* type for storage class */ 379 }; 380 381 /* 382 * Since AVL already has embedded element counter, use dn_dbufs_count 383 * only for dbufs not counted there (bonus buffers) and just add them. 384 */ 385 #define DN_DBUFS_COUNT(dn) ((dn)->dn_dbufs_count + \ 386 avl_numnodes(&(dn)->dn_dbufs)) 387 388 /* 389 * We use this (otherwise unused) bit to indicate if the value of 390 * dn_next_maxblkid[txgoff] is valid to use in dnode_sync(). 391 */ 392 #define DMU_NEXT_MAXBLKID_SET (1ULL << 63) 393 394 /* 395 * Adds a level of indirection between the dbuf and the dnode to avoid 396 * iterating descendent dbufs in dnode_move(). Handles are not allocated 397 * individually, but as an array of child dnodes in dnode_hold_impl(). 398 */ 399 typedef struct dnode_handle { 400 /* Protects dnh_dnode from modification by dnode_move(). */ 401 zrlock_t dnh_zrlock; 402 dnode_t *dnh_dnode; 403 } dnode_handle_t; 404 405 typedef struct dnode_children { 406 dmu_buf_user_t dnc_dbu; /* User evict data */ 407 size_t dnc_count; /* number of children */ 408 dnode_handle_t dnc_children[]; /* sized dynamically */ 409 } dnode_children_t; 410 411 typedef struct free_range { 412 avl_node_t fr_node; 413 uint64_t fr_blkid; 414 uint64_t fr_nblks; 415 } free_range_t; 416 417 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp, 418 uint64_t object, dnode_handle_t *dnh); 419 void dnode_special_close(dnode_handle_t *dnh); 420 421 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); 422 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); 423 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); 424 425 int dnode_hold(struct objset *dd, uint64_t object, 426 const void *ref, dnode_t **dnp); 427 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, int dn_slots, 428 const void *ref, dnode_t **dnp); 429 boolean_t dnode_add_ref(dnode_t *dn, const void *ref); 430 void dnode_rele(dnode_t *dn, const void *ref); 431 void dnode_rele_and_unlock(dnode_t *dn, const void *tag, boolean_t evicting); 432 int dnode_try_claim(objset_t *os, uint64_t object, int slots); 433 boolean_t dnode_is_dirty(dnode_t *dn); 434 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); 435 void dnode_sync(dnode_t *dn, dmu_tx_t *tx); 436 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 437 dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx); 438 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 439 dmu_object_type_t bonustype, int bonuslen, int dn_slots, 440 boolean_t keep_spill, dmu_tx_t *tx); 441 void dnode_free(dnode_t *dn, dmu_tx_t *tx); 442 void dnode_byteswap(dnode_phys_t *dnp); 443 void dnode_buf_byteswap(void *buf, size_t size); 444 void dnode_verify(dnode_t *dn); 445 int dnode_set_nlevels(dnode_t *dn, int nlevels, dmu_tx_t *tx); 446 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); 447 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); 448 void dnode_diduse_space(dnode_t *dn, int64_t space); 449 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, 450 boolean_t have_read, boolean_t force); 451 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); 452 void dnode_init(void); 453 void dnode_fini(void); 454 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, 455 int minlvl, uint64_t blkfill, uint64_t txg); 456 void dnode_evict_dbufs(dnode_t *dn); 457 void dnode_evict_bonus(dnode_t *dn); 458 void dnode_free_interior_slots(dnode_t *dn); 459 460 void dnode_set_storage_type(dnode_t *dn, dmu_object_type_t type); 461 462 #define DNODE_LEVEL_IS_CACHEABLE(_dn, _level) \ 463 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ 464 (((_level) > 0 || DMU_OT_IS_METADATA((_dn)->dn_type)) && \ 465 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)) 466 467 /* 468 * Used for dnodestats kstat. 469 */ 470 typedef struct dnode_stats { 471 /* 472 * Number of failed attempts to hold a meta dnode dbuf. 473 */ 474 kstat_named_t dnode_hold_dbuf_hold; 475 /* 476 * Number of failed attempts to read a meta dnode dbuf. 477 */ 478 kstat_named_t dnode_hold_dbuf_read; 479 /* 480 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was able 481 * to hold the requested object number which was allocated. This is 482 * the common case when looking up any allocated object number. 483 */ 484 kstat_named_t dnode_hold_alloc_hits; 485 /* 486 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not 487 * able to hold the request object number because it was not allocated. 488 */ 489 kstat_named_t dnode_hold_alloc_misses; 490 /* 491 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not 492 * able to hold the request object number because the object number 493 * refers to an interior large dnode slot. 494 */ 495 kstat_named_t dnode_hold_alloc_interior; 496 /* 497 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) needed 498 * to retry acquiring slot zrl locks due to contention. 499 */ 500 kstat_named_t dnode_hold_alloc_lock_retry; 501 /* 502 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) did not 503 * need to create the dnode because another thread did so after 504 * dropping the read lock but before acquiring the write lock. 505 */ 506 kstat_named_t dnode_hold_alloc_lock_misses; 507 /* 508 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) found 509 * a free dnode instantiated by dnode_create() but not yet allocated 510 * by dnode_allocate(). 511 */ 512 kstat_named_t dnode_hold_alloc_type_none; 513 /* 514 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was able 515 * to hold the requested range of free dnode slots. 516 */ 517 kstat_named_t dnode_hold_free_hits; 518 /* 519 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not 520 * able to hold the requested range of free dnode slots because 521 * at least one slot was allocated. 522 */ 523 kstat_named_t dnode_hold_free_misses; 524 /* 525 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not 526 * able to hold the requested range of free dnode slots because 527 * after acquiring the zrl lock at least one slot was allocated. 528 */ 529 kstat_named_t dnode_hold_free_lock_misses; 530 /* 531 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) needed 532 * to retry acquiring slot zrl locks due to contention. 533 */ 534 kstat_named_t dnode_hold_free_lock_retry; 535 /* 536 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested 537 * a range of dnode slots which were held by another thread. 538 */ 539 kstat_named_t dnode_hold_free_refcount; 540 /* 541 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested 542 * a range of dnode slots which would overflow the dnode_phys_t. 543 */ 544 kstat_named_t dnode_hold_free_overflow; 545 /* 546 * Number of times dnode_free_interior_slots() needed to retry 547 * acquiring a slot zrl lock due to contention. 548 */ 549 kstat_named_t dnode_free_interior_lock_retry; 550 /* 551 * Number of new dnodes allocated by dnode_allocate(). 552 */ 553 kstat_named_t dnode_allocate; 554 /* 555 * Number of dnodes re-allocated by dnode_reallocate(). 556 */ 557 kstat_named_t dnode_reallocate; 558 /* 559 * Number of meta dnode dbufs evicted. 560 */ 561 kstat_named_t dnode_buf_evict; 562 /* 563 * Number of times dmu_object_alloc*() reached the end of the existing 564 * object ID chunk and advanced to a new one. 565 */ 566 kstat_named_t dnode_alloc_next_chunk; 567 /* 568 * Number of times multiple threads attempted to allocate a dnode 569 * from the same block of free dnodes. 570 */ 571 kstat_named_t dnode_alloc_race; 572 /* 573 * Number of times dmu_object_alloc*() was forced to advance to the 574 * next meta dnode dbuf due to an error from dmu_object_next(). 575 */ 576 kstat_named_t dnode_alloc_next_block; 577 /* 578 * Statistics for tracking dnodes which have been moved. 579 */ 580 kstat_named_t dnode_move_invalid; 581 kstat_named_t dnode_move_recheck1; 582 kstat_named_t dnode_move_recheck2; 583 kstat_named_t dnode_move_special; 584 kstat_named_t dnode_move_handle; 585 kstat_named_t dnode_move_rwlock; 586 kstat_named_t dnode_move_active; 587 } dnode_stats_t; 588 589 typedef struct dnode_sums { 590 wmsum_t dnode_hold_dbuf_hold; 591 wmsum_t dnode_hold_dbuf_read; 592 wmsum_t dnode_hold_alloc_hits; 593 wmsum_t dnode_hold_alloc_misses; 594 wmsum_t dnode_hold_alloc_interior; 595 wmsum_t dnode_hold_alloc_lock_retry; 596 wmsum_t dnode_hold_alloc_lock_misses; 597 wmsum_t dnode_hold_alloc_type_none; 598 wmsum_t dnode_hold_free_hits; 599 wmsum_t dnode_hold_free_misses; 600 wmsum_t dnode_hold_free_lock_misses; 601 wmsum_t dnode_hold_free_lock_retry; 602 wmsum_t dnode_hold_free_refcount; 603 wmsum_t dnode_hold_free_overflow; 604 wmsum_t dnode_free_interior_lock_retry; 605 wmsum_t dnode_allocate; 606 wmsum_t dnode_reallocate; 607 wmsum_t dnode_buf_evict; 608 wmsum_t dnode_alloc_next_chunk; 609 wmsum_t dnode_alloc_race; 610 wmsum_t dnode_alloc_next_block; 611 wmsum_t dnode_move_invalid; 612 wmsum_t dnode_move_recheck1; 613 wmsum_t dnode_move_recheck2; 614 wmsum_t dnode_move_special; 615 wmsum_t dnode_move_handle; 616 wmsum_t dnode_move_rwlock; 617 wmsum_t dnode_move_active; 618 } dnode_sums_t; 619 620 extern dnode_stats_t dnode_stats; 621 extern dnode_sums_t dnode_sums; 622 623 #define DNODE_STAT_INCR(stat, val) \ 624 wmsum_add(&dnode_sums.stat, (val)) 625 #define DNODE_STAT_BUMP(stat) \ 626 DNODE_STAT_INCR(stat, 1); 627 628 #ifdef ZFS_DEBUG 629 630 #define dprintf_dnode(dn, fmt, ...) do { \ 631 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 632 char __db_buf[32]; \ 633 uint64_t __db_obj = (dn)->dn_object; \ 634 if (__db_obj == DMU_META_DNODE_OBJECT) \ 635 (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \ 636 else \ 637 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 638 (u_longlong_t)__db_obj);\ 639 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ 640 __db_buf, __VA_ARGS__); \ 641 } \ 642 } while (0) 643 644 #define DNODE_VERIFY(dn) dnode_verify(dn) 645 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) 646 647 #else 648 649 #define dprintf_dnode(db, fmt, ...) 650 #define DNODE_VERIFY(dn) ((void) sizeof ((uintptr_t)(dn))) 651 #define FREE_VERIFY(db, start, end, tx) 652 653 #endif 654 655 #ifdef __cplusplus 656 } 657 #endif 658 659 #endif /* _SYS_DNODE_H */ 660