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 uint64_t dn_pad3[4]; 223 union { 224 blkptr_t dn_blkptr[1+DN_OLD_MAX_BONUSLEN/sizeof (blkptr_t)]; 225 struct { 226 blkptr_t __dn_ignore1; 227 uint8_t dn_bonus[DN_OLD_MAX_BONUSLEN]; 228 }; 229 struct { 230 blkptr_t __dn_ignore2; 231 uint8_t __dn_ignore3[DN_OLD_MAX_BONUSLEN - 232 sizeof (blkptr_t)]; 233 blkptr_t dn_spill; 234 }; 235 }; 236 } dnode_phys_t; 237 238 #define DN_SPILL_BLKPTR(dnp) (blkptr_t *)((char *)(dnp) + \ 239 (((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT)) 240 241 struct dnode { 242 /* 243 * Protects the structure of the dnode, including the number of levels 244 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_* 245 */ 246 krwlock_t dn_struct_rwlock; 247 248 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ 249 list_node_t dn_link; 250 251 /* immutable: */ 252 struct objset *dn_objset; 253 uint64_t dn_object; 254 struct dmu_buf_impl *dn_dbuf; 255 struct dnode_handle *dn_handle; 256 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ 257 258 /* 259 * Copies of stuff in dn_phys. They're valid in the open 260 * context (eg. even before the dnode is first synced). 261 * Where necessary, these are protected by dn_struct_rwlock. 262 */ 263 dmu_object_type_t dn_type; /* object type */ 264 uint16_t dn_bonuslen; /* bonus length */ 265 uint8_t dn_bonustype; /* bonus type */ 266 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ 267 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 268 uint8_t dn_compress; /* ZIO_COMPRESS type */ 269 uint8_t dn_nlevels; 270 uint8_t dn_indblkshift; 271 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ 272 uint8_t dn_moved; /* Has this dnode been moved? */ 273 uint16_t dn_datablkszsec; /* in 512b sectors */ 274 uint32_t dn_datablksz; /* in bytes */ 275 uint64_t dn_maxblkid; 276 uint8_t dn_next_type[TXG_SIZE]; 277 uint8_t dn_num_slots; /* metadnode slots consumed on disk */ 278 uint8_t dn_next_nblkptr[TXG_SIZE]; 279 uint8_t dn_next_nlevels[TXG_SIZE]; 280 uint8_t dn_next_indblkshift[TXG_SIZE]; 281 uint8_t dn_next_bonustype[TXG_SIZE]; 282 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ 283 uint16_t dn_next_bonuslen[TXG_SIZE]; 284 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ 285 286 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */ 287 uint32_t dn_dbufs_count; /* count of dn_dbufs */ 288 289 /* protected by os_lock: */ 290 multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ 291 292 /* protected by dn_mtx: */ 293 kmutex_t dn_mtx; 294 list_t dn_dirty_records[TXG_SIZE]; 295 struct range_tree *dn_free_ranges[TXG_SIZE]; 296 uint64_t dn_allocated_txg; 297 uint64_t dn_free_txg; 298 uint64_t dn_assigned_txg; 299 kcondvar_t dn_notxholds; 300 enum dnode_dirtycontext dn_dirtyctx; 301 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ 302 303 /* protected by own devices */ 304 refcount_t dn_tx_holds; 305 refcount_t dn_holds; 306 307 kmutex_t dn_dbufs_mtx; 308 /* 309 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs 310 * can contain multiple dbufs of the same (level, blkid) when a 311 * dbuf is marked DB_EVICTING without being removed from 312 * dn_dbufs. To maintain the avl invariant that there cannot be 313 * duplicate entries, we order the dbufs by an arbitrary value - 314 * their address in memory. This means that dn_dbufs cannot be used to 315 * directly look up a dbuf. Instead, callers must use avl_walk, have 316 * a reference to the dbuf, or look up a non-existant node with 317 * db_state = DB_SEARCH (see dbuf_free_range for an example). 318 */ 319 avl_tree_t dn_dbufs; 320 321 /* protected by dn_struct_rwlock */ 322 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ 323 324 boolean_t dn_have_spill; /* have spill or are spilling */ 325 326 /* parent IO for current sync write */ 327 zio_t *dn_zio; 328 329 /* used in syncing context */ 330 uint64_t dn_oldused; /* old phys used bytes */ 331 uint64_t dn_oldflags; /* old phys dn_flags */ 332 uint64_t dn_olduid, dn_oldgid; 333 uint64_t dn_newuid, dn_newgid; 334 int dn_id_flags; 335 336 /* holds prefetch structure */ 337 struct zfetch dn_zfetch; 338 }; 339 340 /* 341 * Adds a level of indirection between the dbuf and the dnode to avoid 342 * iterating descendent dbufs in dnode_move(). Handles are not allocated 343 * individually, but as an array of child dnodes in dnode_hold_impl(). 344 */ 345 typedef struct dnode_handle { 346 /* Protects dnh_dnode from modification by dnode_move(). */ 347 zrlock_t dnh_zrlock; 348 dnode_t *dnh_dnode; 349 } dnode_handle_t; 350 351 typedef struct dnode_children { 352 dmu_buf_user_t dnc_dbu; /* User evict data */ 353 size_t dnc_count; /* number of children */ 354 dnode_handle_t dnc_children[]; /* sized dynamically */ 355 } dnode_children_t; 356 357 typedef struct free_range { 358 avl_node_t fr_node; 359 uint64_t fr_blkid; 360 uint64_t fr_nblks; 361 } free_range_t; 362 363 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp, 364 uint64_t object, dnode_handle_t *dnh); 365 void dnode_special_close(dnode_handle_t *dnh); 366 367 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); 368 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); 369 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); 370 371 int dnode_hold(struct objset *dd, uint64_t object, 372 void *ref, dnode_t **dnp); 373 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, int dn_slots, 374 void *ref, dnode_t **dnp); 375 boolean_t dnode_add_ref(dnode_t *dn, void *ref); 376 void dnode_rele(dnode_t *dn, void *ref); 377 void dnode_rele_and_unlock(dnode_t *dn, void *tag, boolean_t evicting); 378 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); 379 void dnode_sync(dnode_t *dn, dmu_tx_t *tx); 380 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 381 dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx); 382 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 383 dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx); 384 void dnode_free(dnode_t *dn, dmu_tx_t *tx); 385 void dnode_byteswap(dnode_phys_t *dnp); 386 void dnode_buf_byteswap(void *buf, size_t size); 387 void dnode_verify(dnode_t *dn); 388 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); 389 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); 390 void dnode_diduse_space(dnode_t *dn, int64_t space); 391 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); 392 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); 393 void dnode_init(void); 394 void dnode_fini(void); 395 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, 396 int minlvl, uint64_t blkfill, uint64_t txg); 397 void dnode_evict_dbufs(dnode_t *dn); 398 void dnode_evict_bonus(dnode_t *dn); 399 void dnode_free_interior_slots(dnode_t *dn); 400 boolean_t dnode_needs_remap(const dnode_t *dn); 401 402 #define DNODE_IS_CACHEABLE(_dn) \ 403 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ 404 (DMU_OT_IS_METADATA((_dn)->dn_type) && \ 405 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)) 406 407 #define DNODE_META_IS_CACHEABLE(_dn) \ 408 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ 409 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA) 410 411 /* 412 * Used for dnodestats kstat. 413 */ 414 typedef struct dnode_stats { 415 /* 416 * Number of failed attempts to hold a meta dnode dbuf. 417 */ 418 kstat_named_t dnode_hold_dbuf_hold; 419 /* 420 * Number of failed attempts to read a meta dnode dbuf. 421 */ 422 kstat_named_t dnode_hold_dbuf_read; 423 /* 424 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was able 425 * to hold the requested object number which was allocated. This is 426 * the common case when looking up any allocated object number. 427 */ 428 kstat_named_t dnode_hold_alloc_hits; 429 /* 430 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not 431 * able to hold the request object number because it was not allocated. 432 */ 433 kstat_named_t dnode_hold_alloc_misses; 434 /* 435 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not 436 * able to hold the request object number because the object number 437 * refers to an interior large dnode slot. 438 */ 439 kstat_named_t dnode_hold_alloc_interior; 440 /* 441 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) needed 442 * to retry acquiring slot zrl locks due to contention. 443 */ 444 kstat_named_t dnode_hold_alloc_lock_retry; 445 /* 446 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) did not 447 * need to create the dnode because another thread did so after 448 * dropping the read lock but before acquiring the write lock. 449 */ 450 kstat_named_t dnode_hold_alloc_lock_misses; 451 /* 452 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) found 453 * a free dnode instantiated by dnode_create() but not yet allocated 454 * by dnode_allocate(). 455 */ 456 kstat_named_t dnode_hold_alloc_type_none; 457 /* 458 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was able 459 * to hold the requested range of free dnode slots. 460 */ 461 kstat_named_t dnode_hold_free_hits; 462 /* 463 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not 464 * able to hold the requested range of free dnode slots because 465 * at least one slot was allocated. 466 */ 467 kstat_named_t dnode_hold_free_misses; 468 /* 469 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not 470 * able to hold the requested range of free dnode slots because 471 * after acquiring the zrl lock at least one slot was allocated. 472 */ 473 kstat_named_t dnode_hold_free_lock_misses; 474 /* 475 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) needed 476 * to retry acquiring slot zrl locks due to contention. 477 */ 478 kstat_named_t dnode_hold_free_lock_retry; 479 /* 480 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested 481 * a range of dnode slots which were held by another thread. 482 */ 483 kstat_named_t dnode_hold_free_refcount; 484 /* 485 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested 486 * a range of dnode slots which would overflow the dnode_phys_t. 487 */ 488 kstat_named_t dnode_hold_free_overflow; 489 /* 490 * Number of times a dnode_hold(...) was attempted on a dnode 491 * which had already been unlinked in an earlier txg. 492 */ 493 kstat_named_t dnode_hold_free_txg; 494 /* 495 * Number of times dnode_free_interior_slots() needed to retry 496 * acquiring a slot zrl lock due to contention. 497 */ 498 kstat_named_t dnode_free_interior_lock_retry; 499 /* 500 * Number of new dnodes allocated by dnode_allocate(). 501 */ 502 kstat_named_t dnode_allocate; 503 /* 504 * Number of dnodes re-allocated by dnode_reallocate(). 505 */ 506 kstat_named_t dnode_reallocate; 507 /* 508 * Number of meta dnode dbufs evicted. 509 */ 510 kstat_named_t dnode_buf_evict; 511 /* 512 * Number of times dmu_object_alloc*() reached the end of the existing 513 * object ID chunk and advanced to a new one. 514 */ 515 kstat_named_t dnode_alloc_next_chunk; 516 /* 517 * Number of times multiple threads attempted to allocate a dnode 518 * from the same block of free dnodes. 519 */ 520 kstat_named_t dnode_alloc_race; 521 /* 522 * Number of times dmu_object_alloc*() was forced to advance to the 523 * next meta dnode dbuf due to an error from dmu_object_next(). 524 */ 525 kstat_named_t dnode_alloc_next_block; 526 /* 527 * Statistics for tracking dnodes which have been moved. 528 */ 529 kstat_named_t dnode_move_invalid; 530 kstat_named_t dnode_move_recheck1; 531 kstat_named_t dnode_move_recheck2; 532 kstat_named_t dnode_move_special; 533 kstat_named_t dnode_move_handle; 534 kstat_named_t dnode_move_rwlock; 535 kstat_named_t dnode_move_active; 536 } dnode_stats_t; 537 538 extern dnode_stats_t dnode_stats; 539 540 #define DNODE_STAT_INCR(stat, val) \ 541 atomic_add_64(&dnode_stats.stat.value.ui64, (val)); 542 #define DNODE_STAT_BUMP(stat) \ 543 DNODE_STAT_INCR(stat, 1); 544 545 #ifdef ZFS_DEBUG 546 547 /* 548 * There should be a ## between the string literal and fmt, to make it 549 * clear that we're joining two strings together, but that piece of shit 550 * gcc doesn't support that preprocessor token. 551 */ 552 #define dprintf_dnode(dn, fmt, ...) do { \ 553 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 554 char __db_buf[32]; \ 555 uint64_t __db_obj = (dn)->dn_object; \ 556 if (__db_obj == DMU_META_DNODE_OBJECT) \ 557 (void) strcpy(__db_buf, "mdn"); \ 558 else \ 559 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 560 (u_longlong_t)__db_obj);\ 561 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ 562 __db_buf, __VA_ARGS__); \ 563 } \ 564 _NOTE(CONSTCOND) } while (0) 565 566 #define DNODE_VERIFY(dn) dnode_verify(dn) 567 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) 568 569 #else 570 571 #define dprintf_dnode(db, fmt, ...) 572 #define DNODE_VERIFY(dn) 573 #define FREE_VERIFY(db, start, end, tx) 574 575 #endif 576 577 #ifdef __cplusplus 578 } 579 #endif 580 581 #endif /* _SYS_DNODE_H */ 582