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