1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * 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/refcount.h> 36 #include <sys/dmu_zfetch.h> 37 #include <sys/zrlock.h> 38 #include <sys/multilist.h> 39 40 #ifdef __cplusplus 41 extern "C" { 42 #endif 43 44 /* 45 * dnode_hold() flags. 46 */ 47 #define DNODE_MUST_BE_ALLOCATED 1 48 #define DNODE_MUST_BE_FREE 2 49 50 /* 51 * dnode_next_offset() flags. 52 */ 53 #define DNODE_FIND_HOLE 1 54 #define DNODE_FIND_BACKWARDS 2 55 #define DNODE_FIND_HAVELOCK 4 56 57 /* 58 * Fixed constants. 59 */ 60 #define DNODE_SHIFT 9 /* 512 bytes */ 61 #define DN_MIN_INDBLKSHIFT 12 /* 4k */ 62 /* 63 * If we ever increase this value beyond 20, we need to revisit all logic that 64 * does x << level * ebps to handle overflow. With a 1M indirect block size, 65 * 4 levels of indirect blocks would not be able to guarantee addressing an 66 * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65. 67 */ 68 #define DN_MAX_INDBLKSHIFT 17 /* 128k */ 69 #define DNODE_BLOCK_SHIFT 14 /* 16k */ 70 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */ 71 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */ 72 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */ 73 74 /* 75 * dnode id flags 76 * 77 * Note: a file will never ever have its 78 * ids moved from bonus->spill 79 * and only in a crypto environment would it be on 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 /* The +2 here is a cheesy way to round up */ 121 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \ 122 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT))) 123 124 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \ 125 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) 126 #define DN_MAX_BONUS_LEN(dnp) \ 127 ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ? \ 128 (uint8_t *)DN_SPILL_BLKPTR(dnp) - (uint8_t *)DN_BONUS(dnp) : \ 129 (uint8_t *)(dnp + (dnp->dn_extra_slots + 1)) - (uint8_t *)DN_BONUS(dnp)) 130 131 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ 132 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) 133 134 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) 135 136 struct dmu_buf_impl; 137 struct objset; 138 struct zio; 139 140 enum dnode_dirtycontext { 141 DN_UNDIRTIED, 142 DN_DIRTY_OPEN, 143 DN_DIRTY_SYNC 144 }; 145 146 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ 147 #define DNODE_FLAG_USED_BYTES (1<<0) 148 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1) 149 150 /* Does dnode have a SA spill blkptr in bonus? */ 151 #define DNODE_FLAG_SPILL_BLKPTR (1<<2) 152 153 /* 154 * VARIABLE-LENGTH (LARGE) DNODES 155 * 156 * The motivation for variable-length dnodes is to eliminate the overhead 157 * associated with using spill blocks. Spill blocks are used to store 158 * system attribute data (i.e. file metadata) that does not fit in the 159 * dnode's bonus buffer. By allowing a larger bonus buffer area the use of 160 * a spill block can be avoided. Spill blocks potentially incur an 161 * additional read I/O for every dnode in a dnode block. As a worst case 162 * example, reading 32 dnodes from a 16k dnode block and all of the spill 163 * blocks could issue 33 separate reads. Now suppose those dnodes have size 164 * 1024 and therefore don't need spill blocks. Then the worst case number 165 * of blocks read is reduced to from 33 to two--one per dnode block. 166 * 167 * ZFS-on-Linux systems that make heavy use of extended attributes benefit 168 * from this feature. In particular, ZFS-on-Linux supports the xattr=sa 169 * dataset property which allows file extended attribute data to be stored 170 * in the dnode bonus buffer as an alternative to the traditional 171 * directory-based format. Workloads such as SELinux and the Lustre 172 * distributed filesystem often store enough xattr data to force spill 173 * blocks when xattr=sa is in effect. Large dnodes may therefore provide a 174 * performance benefit to such systems. Other use cases that benefit from 175 * this feature include files with large ACLs and symbolic links with long 176 * target names. 177 * 178 * The size of a dnode may be a multiple of 512 bytes up to the size of a 179 * dnode block (currently 16384 bytes). The dn_extra_slots field of the 180 * on-disk dnode_phys_t structure describes the size of the physical dnode 181 * on disk. The field represents how many "extra" dnode_phys_t slots a 182 * dnode consumes in its dnode block. This convention results in a value of 183 * 0 for 512 byte dnodes which preserves on-disk format compatibility with 184 * older software which doesn't support large dnodes. 185 * 186 * Similarly, the in-memory dnode_t structure has a dn_num_slots field 187 * to represent the total number of dnode_phys_t slots consumed on disk. 188 * Thus dn->dn_num_slots is 1 greater than the corresponding 189 * dnp->dn_extra_slots. This difference in convention was adopted 190 * because, unlike on-disk structures, backward compatibility is not a 191 * concern for in-memory objects, so we used a more natural way to 192 * represent size for a dnode_t. 193 * 194 * The default size for newly created dnodes is determined by the value of 195 * the "dnodesize" dataset property. By default the property is set to 196 * "legacy" which is compatible with older software. Setting the property 197 * to "auto" will allow the filesystem to choose the most suitable dnode 198 * size. Currently this just sets the default dnode size to 1k, but future 199 * code improvements could dynamically choose a size based on observed 200 * workload patterns. Dnodes of varying sizes can coexist within the same 201 * dataset and even within the same dnode block. 202 */ 203 204 typedef struct dnode_phys { 205 uint8_t dn_type; /* dmu_object_type_t */ 206 uint8_t dn_indblkshift; /* ln2(indirect block size) */ 207 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ 208 uint8_t dn_nblkptr; /* length of dn_blkptr */ 209 uint8_t dn_bonustype; /* type of data in bonus buffer */ 210 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 211 uint8_t dn_compress; /* ZIO_COMPRESS type */ 212 uint8_t dn_flags; /* DNODE_FLAG_* */ 213 uint16_t dn_datablkszsec; /* data block size in 512b sectors */ 214 uint16_t dn_bonuslen; /* length of dn_bonus */ 215 uint8_t dn_extra_slots; /* # of subsequent slots consumed */ 216 uint8_t dn_pad2[3]; 217 218 /* accounting is protected by dn_dirty_mtx */ 219 uint64_t dn_maxblkid; /* largest allocated block ID */ 220 uint64_t dn_used; /* bytes (or sectors) of disk space */ 221 222 /* 223 * Both dn_pad2 and dn_pad3 are protected by the block's MAC. This 224 * allows us to protect any fields that might be added here in the 225 * future. In either case, developers will want to check 226 * zio_crypt_init_uios_dnode() to ensure the new field is being 227 * protected properly. 228 */ 229 uint64_t dn_pad3[4]; 230 union { 231 blkptr_t dn_blkptr[1+DN_OLD_MAX_BONUSLEN/sizeof (blkptr_t)]; 232 struct { 233 blkptr_t __dn_ignore1; 234 uint8_t dn_bonus[DN_OLD_MAX_BONUSLEN]; 235 }; 236 struct { 237 blkptr_t __dn_ignore2; 238 uint8_t __dn_ignore3[DN_OLD_MAX_BONUSLEN - 239 sizeof (blkptr_t)]; 240 blkptr_t dn_spill; 241 }; 242 }; 243 } dnode_phys_t; 244 245 #define DN_SPILL_BLKPTR(dnp) (blkptr_t *)((char *)(dnp) + \ 246 (((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT)) 247 248 struct dnode { 249 /* 250 * Protects the structure of the dnode, including the number of levels 251 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_* 252 */ 253 krwlock_t dn_struct_rwlock; 254 255 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ 256 list_node_t dn_link; 257 258 /* immutable: */ 259 struct objset *dn_objset; 260 uint64_t dn_object; 261 struct dmu_buf_impl *dn_dbuf; 262 struct dnode_handle *dn_handle; 263 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ 264 265 /* 266 * Copies of stuff in dn_phys. They're valid in the open 267 * context (eg. even before the dnode is first synced). 268 * Where necessary, these are protected by dn_struct_rwlock. 269 */ 270 dmu_object_type_t dn_type; /* object type */ 271 uint16_t dn_bonuslen; /* bonus length */ 272 uint8_t dn_bonustype; /* bonus type */ 273 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ 274 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 275 uint8_t dn_compress; /* ZIO_COMPRESS type */ 276 uint8_t dn_nlevels; 277 uint8_t dn_indblkshift; 278 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ 279 uint8_t dn_moved; /* Has this dnode been moved? */ 280 uint16_t dn_datablkszsec; /* in 512b sectors */ 281 uint32_t dn_datablksz; /* in bytes */ 282 uint64_t dn_maxblkid; 283 uint8_t dn_next_type[TXG_SIZE]; 284 uint8_t dn_num_slots; /* metadnode slots consumed on disk */ 285 uint8_t dn_next_nblkptr[TXG_SIZE]; 286 uint8_t dn_next_nlevels[TXG_SIZE]; 287 uint8_t dn_next_indblkshift[TXG_SIZE]; 288 uint8_t dn_next_bonustype[TXG_SIZE]; 289 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ 290 uint16_t dn_next_bonuslen[TXG_SIZE]; 291 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ 292 293 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */ 294 uint32_t dn_dbufs_count; /* count of dn_dbufs */ 295 296 /* protected by os_lock: */ 297 multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ 298 299 /* protected by dn_mtx: */ 300 kmutex_t dn_mtx; 301 list_t dn_dirty_records[TXG_SIZE]; 302 struct range_tree *dn_free_ranges[TXG_SIZE]; 303 uint64_t dn_allocated_txg; 304 uint64_t dn_free_txg; 305 uint64_t dn_assigned_txg; 306 uint64_t dn_dirty_txg; /* txg dnode was last dirtied */ 307 kcondvar_t dn_notxholds; 308 enum dnode_dirtycontext dn_dirtyctx; 309 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ 310 311 /* protected by own devices */ 312 zfs_refcount_t dn_tx_holds; 313 zfs_refcount_t dn_holds; 314 315 kmutex_t dn_dbufs_mtx; 316 /* 317 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs 318 * can contain multiple dbufs of the same (level, blkid) when a 319 * dbuf is marked DB_EVICTING without being removed from 320 * dn_dbufs. To maintain the avl invariant that there cannot be 321 * duplicate entries, we order the dbufs by an arbitrary value - 322 * their address in memory. This means that dn_dbufs cannot be used to 323 * directly look up a dbuf. Instead, callers must use avl_walk, have 324 * a reference to the dbuf, or look up a non-existant node with 325 * db_state = DB_SEARCH (see dbuf_free_range for an example). 326 */ 327 avl_tree_t dn_dbufs; 328 329 /* protected by dn_struct_rwlock */ 330 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ 331 332 boolean_t dn_have_spill; /* have spill or are spilling */ 333 334 /* parent IO for current sync write */ 335 zio_t *dn_zio; 336 337 /* used in syncing context */ 338 uint64_t dn_oldused; /* old phys used bytes */ 339 uint64_t dn_oldflags; /* old phys dn_flags */ 340 uint64_t dn_olduid, dn_oldgid; 341 uint64_t dn_newuid, dn_newgid; 342 int dn_id_flags; 343 344 /* holds prefetch structure */ 345 struct zfetch dn_zfetch; 346 }; 347 348 /* 349 * Since AVL already has embedded element counter, use dn_dbufs_count 350 * only for dbufs not counted there (bonus buffers) and just add them. 351 */ 352 #define DN_DBUFS_COUNT(dn) ((dn)->dn_dbufs_count + \ 353 avl_numnodes(&(dn)->dn_dbufs)) 354 355 /* 356 * Adds a level of indirection between the dbuf and the dnode to avoid 357 * iterating descendent dbufs in dnode_move(). Handles are not allocated 358 * individually, but as an array of child dnodes in dnode_hold_impl(). 359 */ 360 typedef struct dnode_handle { 361 /* Protects dnh_dnode from modification by dnode_move(). */ 362 zrlock_t dnh_zrlock; 363 dnode_t *dnh_dnode; 364 } dnode_handle_t; 365 366 typedef struct dnode_children { 367 dmu_buf_user_t dnc_dbu; /* User evict data */ 368 size_t dnc_count; /* number of children */ 369 dnode_handle_t dnc_children[]; /* sized dynamically */ 370 } dnode_children_t; 371 372 typedef struct free_range { 373 avl_node_t fr_node; 374 uint64_t fr_blkid; 375 uint64_t fr_nblks; 376 } free_range_t; 377 378 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp, 379 uint64_t object, dnode_handle_t *dnh); 380 void dnode_special_close(dnode_handle_t *dnh); 381 382 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); 383 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); 384 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); 385 386 int dnode_hold(struct objset *dd, uint64_t object, 387 void *ref, dnode_t **dnp); 388 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, int dn_slots, 389 void *ref, dnode_t **dnp); 390 boolean_t dnode_add_ref(dnode_t *dn, void *ref); 391 void dnode_rele(dnode_t *dn, void *ref); 392 void dnode_rele_and_unlock(dnode_t *dn, void *tag, boolean_t evicting); 393 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); 394 void dnode_sync(dnode_t *dn, dmu_tx_t *tx); 395 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 396 dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx); 397 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 398 dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx); 399 void dnode_free(dnode_t *dn, dmu_tx_t *tx); 400 void dnode_byteswap(dnode_phys_t *dnp); 401 void dnode_buf_byteswap(void *buf, size_t size); 402 void dnode_verify(dnode_t *dn); 403 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); 404 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); 405 void dnode_diduse_space(dnode_t *dn, int64_t space); 406 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); 407 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); 408 void dnode_init(void); 409 void dnode_fini(void); 410 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, 411 int minlvl, uint64_t blkfill, uint64_t txg); 412 void dnode_evict_dbufs(dnode_t *dn); 413 void dnode_evict_bonus(dnode_t *dn); 414 void dnode_free_interior_slots(dnode_t *dn); 415 boolean_t dnode_needs_remap(const dnode_t *dn); 416 417 #define DNODE_IS_DIRTY(_dn) \ 418 ((_dn)->dn_dirty_txg >= spa_syncing_txg((_dn)->dn_objset->os_spa)) 419 420 #define DNODE_IS_CACHEABLE(_dn) \ 421 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ 422 (DMU_OT_IS_METADATA((_dn)->dn_type) && \ 423 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)) 424 425 #define DNODE_META_IS_CACHEABLE(_dn) \ 426 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ 427 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA) 428 429 /* 430 * Used for dnodestats kstat. 431 */ 432 typedef struct dnode_stats { 433 /* 434 * Number of failed attempts to hold a meta dnode dbuf. 435 */ 436 kstat_named_t dnode_hold_dbuf_hold; 437 /* 438 * Number of failed attempts to read a meta dnode dbuf. 439 */ 440 kstat_named_t dnode_hold_dbuf_read; 441 /* 442 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was able 443 * to hold the requested object number which was allocated. This is 444 * the common case when looking up any allocated object number. 445 */ 446 kstat_named_t dnode_hold_alloc_hits; 447 /* 448 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not 449 * able to hold the request object number because it was not allocated. 450 */ 451 kstat_named_t dnode_hold_alloc_misses; 452 /* 453 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not 454 * able to hold the request object number because the object number 455 * refers to an interior large dnode slot. 456 */ 457 kstat_named_t dnode_hold_alloc_interior; 458 /* 459 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) needed 460 * to retry acquiring slot zrl locks due to contention. 461 */ 462 kstat_named_t dnode_hold_alloc_lock_retry; 463 /* 464 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) did not 465 * need to create the dnode because another thread did so after 466 * dropping the read lock but before acquiring the write lock. 467 */ 468 kstat_named_t dnode_hold_alloc_lock_misses; 469 /* 470 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) found 471 * a free dnode instantiated by dnode_create() but not yet allocated 472 * by dnode_allocate(). 473 */ 474 kstat_named_t dnode_hold_alloc_type_none; 475 /* 476 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was able 477 * to hold the requested range of free dnode slots. 478 */ 479 kstat_named_t dnode_hold_free_hits; 480 /* 481 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not 482 * able to hold the requested range of free dnode slots because 483 * at least one slot was allocated. 484 */ 485 kstat_named_t dnode_hold_free_misses; 486 /* 487 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not 488 * able to hold the requested range of free dnode slots because 489 * after acquiring the zrl lock at least one slot was allocated. 490 */ 491 kstat_named_t dnode_hold_free_lock_misses; 492 /* 493 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) needed 494 * to retry acquiring slot zrl locks due to contention. 495 */ 496 kstat_named_t dnode_hold_free_lock_retry; 497 /* 498 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested 499 * a range of dnode slots which were held by another thread. 500 */ 501 kstat_named_t dnode_hold_free_refcount; 502 /* 503 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested 504 * a range of dnode slots which would overflow the dnode_phys_t. 505 */ 506 kstat_named_t dnode_hold_free_overflow; 507 /* 508 * Number of times a dnode_hold(...) was attempted on a dnode 509 * which had already been unlinked in an earlier txg. 510 */ 511 kstat_named_t dnode_hold_free_txg; 512 /* 513 * Number of times dnode_free_interior_slots() needed to retry 514 * acquiring a slot zrl lock due to contention. 515 */ 516 kstat_named_t dnode_free_interior_lock_retry; 517 /* 518 * Number of new dnodes allocated by dnode_allocate(). 519 */ 520 kstat_named_t dnode_allocate; 521 /* 522 * Number of dnodes re-allocated by dnode_reallocate(). 523 */ 524 kstat_named_t dnode_reallocate; 525 /* 526 * Number of meta dnode dbufs evicted. 527 */ 528 kstat_named_t dnode_buf_evict; 529 /* 530 * Number of times dmu_object_alloc*() reached the end of the existing 531 * object ID chunk and advanced to a new one. 532 */ 533 kstat_named_t dnode_alloc_next_chunk; 534 /* 535 * Number of times multiple threads attempted to allocate a dnode 536 * from the same block of free dnodes. 537 */ 538 kstat_named_t dnode_alloc_race; 539 /* 540 * Number of times dmu_object_alloc*() was forced to advance to the 541 * next meta dnode dbuf due to an error from dmu_object_next(). 542 */ 543 kstat_named_t dnode_alloc_next_block; 544 /* 545 * Statistics for tracking dnodes which have been moved. 546 */ 547 kstat_named_t dnode_move_invalid; 548 kstat_named_t dnode_move_recheck1; 549 kstat_named_t dnode_move_recheck2; 550 kstat_named_t dnode_move_special; 551 kstat_named_t dnode_move_handle; 552 kstat_named_t dnode_move_rwlock; 553 kstat_named_t dnode_move_active; 554 } dnode_stats_t; 555 556 extern dnode_stats_t dnode_stats; 557 558 #define DNODE_STAT_INCR(stat, val) \ 559 atomic_add_64(&dnode_stats.stat.value.ui64, (val)); 560 #define DNODE_STAT_BUMP(stat) \ 561 DNODE_STAT_INCR(stat, 1); 562 563 #ifdef ZFS_DEBUG 564 565 /* 566 * There should be a ## between the string literal and fmt, to make it 567 * clear that we're joining two strings together, but that piece of shit 568 * gcc doesn't support that preprocessor token. 569 */ 570 #define dprintf_dnode(dn, fmt, ...) do { \ 571 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 572 char __db_buf[32]; \ 573 uint64_t __db_obj = (dn)->dn_object; \ 574 if (__db_obj == DMU_META_DNODE_OBJECT) \ 575 (void) strcpy(__db_buf, "mdn"); \ 576 else \ 577 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 578 (u_longlong_t)__db_obj);\ 579 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ 580 __db_buf, __VA_ARGS__); \ 581 } \ 582 _NOTE(CONSTCOND) } while (0) 583 584 #define DNODE_VERIFY(dn) dnode_verify(dn) 585 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) 586 587 #else 588 589 #define dprintf_dnode(db, fmt, ...) 590 #define DNODE_VERIFY(dn) 591 #define FREE_VERIFY(db, start, end, tx) 592 593 #endif 594 595 #ifdef __cplusplus 596 } 597 #endif 598 599 #endif /* _SYS_DNODE_H */ 600