zfs_context.h revision 185029
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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26#ifndef _SYS_ZFS_CONTEXT_H 27#define _SYS_ZFS_CONTEXT_H 28 29#ifdef __cplusplus 30extern "C" { 31#endif 32 33#define _SYS_MUTEX_H 34#define _SYS_RWLOCK_H 35#define _SYS_CONDVAR_H 36#define _SYS_SYSTM_H 37#define _SYS_DEBUG_H 38#define _SYS_T_LOCK_H 39#define _SYS_VNODE_H 40#define _SYS_VFS_H 41#define _SYS_SUNDDI_H 42#define _SYS_CALLB_H 43#define _SYS_SCHED_H_ 44 45#include <solaris.h> 46#include <stdio.h> 47#include <stdlib.h> 48#include <stddef.h> 49#include <stdarg.h> 50#include <fcntl.h> 51#include <unistd.h> 52#include <errno.h> 53#include <string.h> 54#include <strings.h> 55#include <thread.h> 56#include <assert.h> 57#include <limits.h> 58#include <dirent.h> 59#include <time.h> 60#include <math.h> 61#include <umem.h> 62#include <fsshare.h> 63#include <sys/note.h> 64#include <sys/types.h> 65#include <sys/cred.h> 66#include <sys/atomic.h> 67#include <sys/sysmacros.h> 68#include <sys/bitmap.h> 69#include <sys/resource.h> 70#include <sys/byteorder.h> 71#include <sys/list.h> 72#include <sys/time.h> 73#include <sys/uio.h> 74#include <sys/mntent.h> 75#include <sys/mnttab.h> 76#include <sys/zfs_debug.h> 77#include <sys/debug.h> 78#include <sys/sdt.h> 79#include <sys/kstat.h> 80#include <sys/u8_textprep.h> 81#include <sys/kernel.h> 82#include <sys/disk.h> 83#include <sys/sysevent/eventdefs.h> 84#include <machine/atomic.h> 85 86#define ZFS_EXPORTS_PATH "/etc/zfs/exports" 87 88/* 89 * Debugging 90 */ 91 92/* 93 * Note that we are not using the debugging levels. 94 */ 95 96#define CE_CONT 0 /* continuation */ 97#define CE_NOTE 1 /* notice */ 98#define CE_WARN 2 /* warning */ 99#define CE_PANIC 3 /* panic */ 100#define CE_IGNORE 4 /* print nothing */ 101 102/* 103 * ZFS debugging 104 */ 105 106#define ZFS_LOG(...) do { } while (0) 107 108typedef u_longlong_t rlim64_t; 109#define RLIM64_INFINITY ((rlim64_t)-3) 110 111#ifdef ZFS_DEBUG 112extern void dprintf_setup(int *argc, char **argv); 113#endif /* ZFS_DEBUG */ 114 115extern void cmn_err(int, const char *, ...); 116extern void vcmn_err(int, const char *, __va_list); 117extern void panic(const char *, ...); 118extern void vpanic(const char *, __va_list); 119 120#define fm_panic panic 121 122/* This definition is copied from assert.h. */ 123#if defined(__STDC__) 124#if __STDC_VERSION__ - 0 >= 199901L 125#define verify(EX) (void)((EX) || (__assert(#EX, __FILE__, __LINE__), 0)) 126#else 127#define verify(EX) (void)((EX) || (__assert(#EX, __FILE__, __LINE__), 0)) 128#endif /* __STDC_VERSION__ - 0 >= 199901L */ 129#else 130#define verify(EX) (void)((EX) || (_assert("EX", __FILE__, __LINE__), 0)) 131#endif /* __STDC__ */ 132 133 134#define VERIFY verify 135#define ASSERT assert 136 137extern void __assert(const char *, const char *, int); 138 139#ifdef lint 140#define VERIFY3_IMPL(x, y, z, t) if (x == z) ((void)0) 141#else 142/* BEGIN CSTYLED */ 143#define VERIFY3_IMPL(LEFT, OP, RIGHT, TYPE) do { \ 144 const TYPE __left = (TYPE)(LEFT); \ 145 const TYPE __right = (TYPE)(RIGHT); \ 146 if (!(__left OP __right)) { \ 147 char *__buf = alloca(256); \ 148 (void) snprintf(__buf, 256, "%s %s %s (0x%llx %s 0x%llx)", \ 149 #LEFT, #OP, #RIGHT, \ 150 (u_longlong_t)__left, #OP, (u_longlong_t)__right); \ 151 __assert(__buf, __FILE__, __LINE__); \ 152 } \ 153_NOTE(CONSTCOND) } while (0) 154/* END CSTYLED */ 155#endif /* lint */ 156 157#define VERIFY3S(x, y, z) VERIFY3_IMPL(x, y, z, int64_t) 158#define VERIFY3U(x, y, z) VERIFY3_IMPL(x, y, z, uint64_t) 159#define VERIFY3P(x, y, z) VERIFY3_IMPL(x, y, z, uintptr_t) 160 161#ifdef NDEBUG 162#define ASSERT3S(x, y, z) ((void)0) 163#define ASSERT3U(x, y, z) ((void)0) 164#define ASSERT3P(x, y, z) ((void)0) 165#else 166#define ASSERT3S(x, y, z) VERIFY3S(x, y, z) 167#define ASSERT3U(x, y, z) VERIFY3U(x, y, z) 168#define ASSERT3P(x, y, z) VERIFY3P(x, y, z) 169#endif 170 171/* 172 * DTrace SDT probes have different signatures in userland than they do in 173 * kernel. If they're being used in kernel code, re-define them out of 174 * existence for their counterparts in libzpool. 175 */ 176 177#ifdef DTRACE_PROBE 178#undef DTRACE_PROBE 179#define DTRACE_PROBE(a) ((void)0) 180#endif /* DTRACE_PROBE */ 181 182#ifdef DTRACE_PROBE1 183#undef DTRACE_PROBE1 184#define DTRACE_PROBE1(a, b, c) ((void)0) 185#endif /* DTRACE_PROBE1 */ 186 187#ifdef DTRACE_PROBE2 188#undef DTRACE_PROBE2 189#define DTRACE_PROBE2(a, b, c, d, e) ((void)0) 190#endif /* DTRACE_PROBE2 */ 191 192#ifdef DTRACE_PROBE3 193#undef DTRACE_PROBE3 194#define DTRACE_PROBE3(a, b, c, d, e, f, g) ((void)0) 195#endif /* DTRACE_PROBE3 */ 196 197#ifdef DTRACE_PROBE4 198#undef DTRACE_PROBE4 199#define DTRACE_PROBE4(a, b, c, d, e, f, g, h, i) ((void)0) 200#endif /* DTRACE_PROBE4 */ 201 202/* 203 * Threads 204 */ 205#define curthread ((void *)(uintptr_t)thr_self()) 206 207typedef struct kthread kthread_t; 208 209#define thread_create(stk, stksize, func, arg, len, pp, state, pri) \ 210 zk_thread_create(func, arg) 211#define thread_exit() thr_exit(NULL) 212 213extern kthread_t *zk_thread_create(void (*func)(), void *arg); 214 215#define issig(why) (FALSE) 216#define ISSIG(thr, why) (FALSE) 217 218/* 219 * Mutexes 220 */ 221typedef struct kmutex { 222 void *m_owner; 223 boolean_t initialized; 224 mutex_t m_lock; 225} kmutex_t; 226 227#define MUTEX_DEFAULT USYNC_THREAD 228#undef MUTEX_HELD 229#define MUTEX_HELD(m) ((m)->m_owner == curthread) 230 231/* 232 * Argh -- we have to get cheesy here because the kernel and userland 233 * have different signatures for the same routine. 234 */ 235//extern int _mutex_init(mutex_t *mp, int type, void *arg); 236//extern int _mutex_destroy(mutex_t *mp); 237 238#define mutex_init(mp, b, c, d) zmutex_init((kmutex_t *)(mp)) 239#define mutex_destroy(mp) zmutex_destroy((kmutex_t *)(mp)) 240 241extern void zmutex_init(kmutex_t *mp); 242extern void zmutex_destroy(kmutex_t *mp); 243extern void mutex_enter(kmutex_t *mp); 244extern void mutex_exit(kmutex_t *mp); 245extern int mutex_tryenter(kmutex_t *mp); 246extern void *mutex_owner(kmutex_t *mp); 247 248/* 249 * RW locks 250 */ 251typedef struct krwlock { 252 int rw_count; 253 void *rw_owner; 254 boolean_t initialized; 255 rwlock_t rw_lock; 256} krwlock_t; 257 258typedef int krw_t; 259 260#define RW_READER 0 261#define RW_WRITER 1 262#define RW_DEFAULT USYNC_THREAD 263 264#undef RW_READ_HELD 265#define RW_READ_HELD(x) ((x)->rw_owner == NULL && (x)->rw_count > 0) 266 267#undef RW_WRITE_HELD 268#define RW_WRITE_HELD(x) ((x)->rw_owner == curthread) 269#define RW_LOCK_HELD(x) rw_lock_held(x) 270 271extern void rw_init(krwlock_t *rwlp, char *name, int type, void *arg); 272extern void rw_destroy(krwlock_t *rwlp); 273extern void rw_enter(krwlock_t *rwlp, krw_t rw); 274extern int rw_tryenter(krwlock_t *rwlp, krw_t rw); 275extern int rw_tryupgrade(krwlock_t *rwlp); 276extern void rw_exit(krwlock_t *rwlp); 277extern int rw_lock_held(krwlock_t *rwlp); 278#define rw_downgrade(rwlp) do { } while (0) 279 280extern uid_t crgetuid(cred_t *cr); 281extern gid_t crgetgid(cred_t *cr); 282extern int crgetngroups(cred_t *cr); 283extern gid_t *crgetgroups(cred_t *cr); 284 285/* 286 * Condition variables 287 */ 288typedef cond_t kcondvar_t; 289 290#define CV_DEFAULT USYNC_THREAD 291 292extern void cv_init(kcondvar_t *cv, char *name, int type, void *arg); 293extern void cv_destroy(kcondvar_t *cv); 294extern void cv_wait(kcondvar_t *cv, kmutex_t *mp); 295extern clock_t cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime); 296extern void cv_signal(kcondvar_t *cv); 297extern void cv_broadcast(kcondvar_t *cv); 298 299/* 300 * Kernel memory 301 */ 302#define KM_SLEEP UMEM_NOFAIL 303#define KM_PUSHPAGE KM_SLEEP 304#define KM_NOSLEEP UMEM_DEFAULT 305#define KMC_NODEBUG UMC_NODEBUG 306#define kmem_alloc(_s, _f) umem_alloc(_s, _f) 307#define kmem_zalloc(_s, _f) umem_zalloc(_s, _f) 308#define kmem_free(_b, _s) umem_free(_b, _s) 309#define kmem_size() (physmem * PAGESIZE) 310#define kmem_cache_create(_a, _b, _c, _d, _e, _f, _g, _h, _i) \ 311 umem_cache_create(_a, _b, _c, _d, _e, _f, _g, _h, _i) 312#define kmem_cache_destroy(_c) umem_cache_destroy(_c) 313#define kmem_cache_alloc(_c, _f) umem_cache_alloc(_c, _f) 314#define kmem_cache_free(_c, _b) umem_cache_free(_c, _b) 315#define kmem_debugging() 0 316#define kmem_cache_reap_now(c) 317 318typedef umem_cache_t kmem_cache_t; 319 320/* 321 * Task queues 322 */ 323typedef struct taskq taskq_t; 324typedef uintptr_t taskqid_t; 325typedef void (task_func_t)(void *); 326 327#define TASKQ_PREPOPULATE 0x0001 328#define TASKQ_CPR_SAFE 0x0002 /* Use CPR safe protocol */ 329#define TASKQ_DYNAMIC 0x0004 /* Use dynamic thread scheduling */ 330 331#define TQ_SLEEP KM_SLEEP /* Can block for memory */ 332#define TQ_NOSLEEP KM_NOSLEEP /* cannot block for memory; may fail */ 333#define TQ_NOQUEUE 0x02 /* Do not enqueue if can't dispatch */ 334 335extern taskq_t *taskq_create(const char *, int, pri_t, int, int, uint_t); 336extern taskqid_t taskq_dispatch(taskq_t *, task_func_t, void *, uint_t); 337extern void taskq_destroy(taskq_t *); 338extern void taskq_wait(taskq_t *); 339extern int taskq_member(taskq_t *, void *); 340 341#define XVA_MAPSIZE 3 342#define XVA_MAGIC 0x78766174 343 344/* 345 * vnodes 346 */ 347typedef struct vnode { 348 uint64_t v_size; 349 int v_fd; 350 char *v_path; 351} vnode_t; 352 353 354typedef struct xoptattr { 355 timestruc_t xoa_createtime; /* Create time of file */ 356 uint8_t xoa_archive; 357 uint8_t xoa_system; 358 uint8_t xoa_readonly; 359 uint8_t xoa_hidden; 360 uint8_t xoa_nounlink; 361 uint8_t xoa_immutable; 362 uint8_t xoa_appendonly; 363 uint8_t xoa_nodump; 364 uint8_t xoa_settable; 365 uint8_t xoa_opaque; 366 uint8_t xoa_av_quarantined; 367 uint8_t xoa_av_modified; 368} xoptattr_t; 369 370typedef struct vattr { 371 uint_t va_mask; /* bit-mask of attributes */ 372 u_offset_t va_size; /* file size in bytes */ 373} vattr_t; 374 375 376typedef struct xvattr { 377 vattr_t xva_vattr; /* Embedded vattr structure */ 378 uint32_t xva_magic; /* Magic Number */ 379 uint32_t xva_mapsize; /* Size of attr bitmap (32-bit words) */ 380 uint32_t *xva_rtnattrmapp; /* Ptr to xva_rtnattrmap[] */ 381 uint32_t xva_reqattrmap[XVA_MAPSIZE]; /* Requested attrs */ 382 uint32_t xva_rtnattrmap[XVA_MAPSIZE]; /* Returned attrs */ 383 xoptattr_t xva_xoptattrs; /* Optional attributes */ 384} xvattr_t; 385 386typedef struct vsecattr { 387 uint_t vsa_mask; /* See below */ 388 int vsa_aclcnt; /* ACL entry count */ 389 void *vsa_aclentp; /* pointer to ACL entries */ 390 int vsa_dfaclcnt; /* default ACL entry count */ 391 void *vsa_dfaclentp; /* pointer to default ACL entries */ 392 size_t vsa_aclentsz; /* ACE size in bytes of vsa_aclentp */ 393} vsecattr_t; 394 395#define AT_TYPE 0x00001 396#define AT_MODE 0x00002 397#define AT_UID 0x00004 398#define AT_GID 0x00008 399#define AT_FSID 0x00010 400#define AT_NODEID 0x00020 401#define AT_NLINK 0x00040 402#define AT_SIZE 0x00080 403#define AT_ATIME 0x00100 404#define AT_MTIME 0x00200 405#define AT_CTIME 0x00400 406#define AT_RDEV 0x00800 407#define AT_BLKSIZE 0x01000 408#define AT_NBLOCKS 0x02000 409#define AT_SEQ 0x08000 410#define AT_XVATTR 0x10000 411 412#define CRCREAT 0 413 414#define VOP_CLOSE(vp, f, c, o, cr, ct) 0 415#define VOP_PUTPAGE(vp, of, sz, fl, cr, ct) 0 416#define VOP_GETATTR(vp, vap, cr) ((vap)->va_size = (vp)->v_size, 0) 417 418#define VOP_FSYNC(vp, f, cr, ct) fsync((vp)->v_fd) 419 420#define VN_RELE(vp) vn_close(vp, 0, NULL, NULL) 421 422#define vn_lock(vp, type) 423#define VOP_UNLOCK(vp, type) 424#ifdef VFS_LOCK_GIANT 425#undef VFS_LOCK_GIANT 426#endif 427#define VFS_LOCK_GIANT(mp) 0 428#ifdef VFS_UNLOCK_GIANT 429#undef VFS_UNLOCK_GIANT 430#endif 431#define VFS_UNLOCK_GIANT(vfslocked) 432 433extern int vn_open(char *path, int x1, int oflags, int mode, vnode_t **vpp, 434 int x2, int x3); 435extern int vn_openat(char *path, int x1, int oflags, int mode, vnode_t **vpp, 436 int x2, int x3, vnode_t *vp, int fd); 437extern int vn_rdwr(int uio, vnode_t *vp, void *addr, ssize_t len, 438 offset_t offset, int x1, int x2, rlim64_t x3, void *x4, ssize_t *residp); 439extern void vn_close(vnode_t *vp, int openflag, cred_t *cr, kthread_t *td); 440 441#define vn_remove(path, x1, x2) remove(path) 442#define vn_rename(from, to, seg) rename((from), (to)) 443#define vn_is_readonly(vp) B_FALSE 444 445extern vnode_t *rootdir; 446 447#include <sys/file.h> /* for FREAD, FWRITE, etc */ 448#define FTRUNC O_TRUNC 449 450/* 451 * Random stuff 452 */ 453#define lbolt (gethrtime() >> 23) 454#define lbolt64 (gethrtime() >> 23) 455//#define hz 119 /* frequency when using gethrtime() >> 23 for lbolt */ 456 457extern void delay(clock_t ticks); 458 459#define gethrestime_sec() time(NULL) 460 461#define max_ncpus 64 462 463#define minclsyspri 60 464#define maxclsyspri 99 465 466#define CPU_SEQID (thr_self() & (max_ncpus - 1)) 467 468#ifndef ptob 469#define ptob(x) ((x) * PAGESIZE) 470#endif 471 472extern uint64_t physmem; 473 474extern int highbit(ulong_t i); 475extern int random_get_bytes(uint8_t *ptr, size_t len); 476extern int random_get_pseudo_bytes(uint8_t *ptr, size_t len); 477 478extern void kernel_init(int); 479extern void kernel_fini(void); 480 481struct spa; 482extern void nicenum(uint64_t num, char *buf); 483extern void show_pool_stats(struct spa *); 484 485typedef struct callb_cpr { 486 kmutex_t *cc_lockp; 487} callb_cpr_t; 488 489#define CALLB_CPR_INIT(cp, lockp, func, name) { \ 490 (cp)->cc_lockp = lockp; \ 491} 492 493#define CALLB_CPR_SAFE_BEGIN(cp) { \ 494 ASSERT(MUTEX_HELD((cp)->cc_lockp)); \ 495} 496 497#define CALLB_CPR_SAFE_END(cp, lockp) { \ 498 ASSERT(MUTEX_HELD((cp)->cc_lockp)); \ 499} 500 501#define CALLB_CPR_EXIT(cp) { \ 502 ASSERT(MUTEX_HELD((cp)->cc_lockp)); \ 503 mutex_exit((cp)->cc_lockp); \ 504} 505 506#define zone_dataset_visible(x, y) (1) 507#define INGLOBALZONE(z) (1) 508 509/* 510 * Hostname information 511 */ 512extern struct utsname utsname; 513extern char hw_serial[]; 514extern int ddi_strtoul(const char *str, char **nptr, int base, 515 unsigned long *result); 516 517/* ZFS Boot Related stuff. */ 518 519struct _buf { 520 intptr_t _fd; 521}; 522 523struct bootstat { 524 uint64_t st_size; 525}; 526 527typedef struct ace_object { 528 uid_t a_who; 529 uint32_t a_access_mask; 530 uint16_t a_flags; 531 uint16_t a_type; 532 uint8_t a_obj_type[16]; 533 uint8_t a_inherit_obj_type[16]; 534} ace_object_t; 535 536 537#define ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE 0x05 538#define ACE_ACCESS_DENIED_OBJECT_ACE_TYPE 0x06 539#define ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE 0x07 540#define ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE 0x08 541 542extern struct _buf *kobj_open_file(char *name); 543extern int kobj_read_file(struct _buf *file, char *buf, unsigned size, 544 unsigned off); 545extern void kobj_close_file(struct _buf *file); 546extern int kobj_get_filesize(struct _buf *file, uint64_t *size); 547extern int zfs_secpolicy_snapshot_perms(const char *name, cred_t *cr); 548extern int zfs_secpolicy_rename_perms(const char *from, const char *to, 549 cred_t *cr); 550extern int zfs_secpolicy_destroy_perms(const char *name, cred_t *cr); 551extern zoneid_t getzoneid(void); 552/* Random compatibility stuff. */ 553#define lbolt (gethrtime() >> 23) 554#define lbolt64 (gethrtime() >> 23) 555 556extern int hz; 557extern uint64_t physmem; 558 559#define gethrestime_sec() time(NULL) 560 561#define pwrite64(d, p, n, o) pwrite(d, p, n, o) 562#define readdir64(d) readdir(d) 563#define SIGPENDING(td) (0) 564#define root_mount_wait() do { } while (0) 565#define root_mounted() (1) 566 567struct file { 568 void *dummy; 569}; 570 571#define FCREAT O_CREAT 572#define FOFFMAX 0x0 573 574/* SID stuff */ 575typedef struct ksiddomain { 576 uint_t kd_ref; 577 uint_t kd_len; 578 char *kd_name; 579} ksiddomain_t; 580 581ksiddomain_t *ksid_lookupdomain(const char *); 582void ksiddomain_rele(ksiddomain_t *); 583 584#define SX_SYSINIT(name, lock, desc) 585 586#define SYSCTL_DECL(...) 587#define SYSCTL_NODE(...) 588#define SYSCTL_INT(...) 589#define SYSCTL_UINT(...) 590#define SYSCTL_ULONG(...) 591#define SYSCTL_QUAD(...) 592#ifdef TUNABLE_INT 593#undef TUNABLE_INT 594#undef TUNABLE_ULONG 595#undef TUNABLE_QUAD 596#endif 597#define TUNABLE_INT(...) 598#define TUNABLE_ULONG(...) 599#define TUNABLE_QUAD(...) 600 601/* Errors */ 602 603#ifndef ERESTART 604#define ERESTART (-1) 605#endif 606 607#ifdef __cplusplus 608} 609#endif 610 611#endif /* _SYS_ZFS_CONTEXT_H */ 612