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 */ 24 25 #ifndef _SYS_ZFS_CONTEXT_H 26 #define _SYS_ZFS_CONTEXT_H 27 28 #ifdef __cplusplus 29 extern "C" { 30 #endif 31 32 #define _SYS_MUTEX_H 33 #define _SYS_RWLOCK_H 34 #define _SYS_CONDVAR_H 35 #define _SYS_SYSTM_H 36 #define _SYS_DEBUG_H 37 #define _SYS_T_LOCK_H 38 #define _SYS_VNODE_H 39 #define _SYS_VFS_H 40 #define _SYS_SUNDDI_H 41 #define _SYS_CALLB_H 42 43 #include <stdio.h> 44 #include <stdlib.h> 45 #include <stddef.h> 46 #include <stdarg.h> 47 #include <fcntl.h> 48 #include <unistd.h> 49 #include <errno.h> 50 #include <string.h> 51 #include <strings.h> 52 #include <synch.h> 53 #include <thread.h> 54 #include <assert.h> 55 #include <alloca.h> 56 #include <umem.h> 57 #include <limits.h> 58 #include <atomic.h> 59 #include <dirent.h> 60 #include <time.h> 61 #include <libsysevent.h> 62 #include <sys/note.h> 63 #include <sys/types.h> 64 #include <sys/cred.h> 65 #include <sys/sysmacros.h> 66 #include <sys/bitmap.h> 67 #include <sys/resource.h> 68 #include <sys/byteorder.h> 69 #include <sys/list.h> 70 #include <sys/uio.h> 71 #include <sys/zfs_debug.h> 72 #include <sys/sdt.h> 73 #include <sys/kstat.h> 74 #include <sys/u8_textprep.h> 75 #include <sys/sysevent/eventdefs.h> 76 #include <sys/sysevent/dev.h> 77 #include <sys/sunddi.h> 78 79 /* 80 * Debugging 81 */ 82 83 /* 84 * Note that we are not using the debugging levels. 85 */ 86 87 #define CE_CONT 0 /* continuation */ 88 #define CE_NOTE 1 /* notice */ 89 #define CE_WARN 2 /* warning */ 90 #define CE_PANIC 3 /* panic */ 91 #define CE_IGNORE 4 /* print nothing */ 92 93 /* 94 * ZFS debugging 95 */ 96 97 #ifdef ZFS_DEBUG 98 extern void dprintf_setup(int *argc, char **argv); 99 #endif /* ZFS_DEBUG */ 100 101 extern void cmn_err(int, const char *, ...); 102 extern void vcmn_err(int, const char *, __va_list); 103 extern void panic(const char *, ...); 104 extern void vpanic(const char *, __va_list); 105 106 #define fm_panic panic 107 108 extern int aok; 109 110 /* This definition is copied from assert.h. */ 111 #if defined(__STDC__) 112 #if __STDC_VERSION__ - 0 >= 199901L 113 #define zverify(EX) (void)((EX) || (aok) || \ 114 (__assert_c99(#EX, __FILE__, __LINE__, __func__), 0)) 115 #else 116 #define zverify(EX) (void)((EX) || (aok) || \ 117 (__assert(#EX, __FILE__, __LINE__), 0)) 118 #endif /* __STDC_VERSION__ - 0 >= 199901L */ 119 #else 120 #define zverify(EX) (void)((EX) || (aok) || \ 121 (_assert("EX", __FILE__, __LINE__), 0)) 122 #endif /* __STDC__ */ 123 124 125 #define VERIFY zverify 126 #define ASSERT zverify 127 #undef assert 128 #define assert zverify 129 130 extern void __assert(const char *, const char *, int); 131 132 #ifdef lint 133 #define VERIFY3_IMPL(x, y, z, t) if (x == z) ((void)0) 134 #else 135 /* BEGIN CSTYLED */ 136 #define VERIFY3_IMPL(LEFT, OP, RIGHT, TYPE) do { \ 137 const TYPE __left = (TYPE)(LEFT); \ 138 const TYPE __right = (TYPE)(RIGHT); \ 139 if (!(__left OP __right) && (!aok)) { \ 140 char *__buf = alloca(256); \ 141 (void) snprintf(__buf, 256, "%s %s %s (0x%llx %s 0x%llx)", \ 142 #LEFT, #OP, #RIGHT, \ 143 (u_longlong_t)__left, #OP, (u_longlong_t)__right); \ 144 __assert(__buf, __FILE__, __LINE__); \ 145 } \ 146 _NOTE(CONSTCOND) } while (0) 147 /* END CSTYLED */ 148 #endif /* lint */ 149 150 #define VERIFY3S(x, y, z) VERIFY3_IMPL(x, y, z, int64_t) 151 #define VERIFY3U(x, y, z) VERIFY3_IMPL(x, y, z, uint64_t) 152 #define VERIFY3P(x, y, z) VERIFY3_IMPL(x, y, z, uintptr_t) 153 154 #ifdef NDEBUG 155 #define ASSERT3S(x, y, z) ((void)0) 156 #define ASSERT3U(x, y, z) ((void)0) 157 #define ASSERT3P(x, y, z) ((void)0) 158 #else 159 #define ASSERT3S(x, y, z) VERIFY3S(x, y, z) 160 #define ASSERT3U(x, y, z) VERIFY3U(x, y, z) 161 #define ASSERT3P(x, y, z) VERIFY3P(x, y, z) 162 #endif 163 164 /* 165 * DTrace SDT probes have different signatures in userland than they do in 166 * kernel. If they're being used in kernel code, re-define them out of 167 * existence for their counterparts in libzpool. 168 */ 169 170 #ifdef DTRACE_PROBE 171 #undef DTRACE_PROBE 172 #define DTRACE_PROBE(a) ((void)0) 173 #endif /* DTRACE_PROBE */ 174 175 #ifdef DTRACE_PROBE1 176 #undef DTRACE_PROBE1 177 #define DTRACE_PROBE1(a, b, c) ((void)0) 178 #endif /* DTRACE_PROBE1 */ 179 180 #ifdef DTRACE_PROBE2 181 #undef DTRACE_PROBE2 182 #define DTRACE_PROBE2(a, b, c, d, e) ((void)0) 183 #endif /* DTRACE_PROBE2 */ 184 185 #ifdef DTRACE_PROBE3 186 #undef DTRACE_PROBE3 187 #define DTRACE_PROBE3(a, b, c, d, e, f, g) ((void)0) 188 #endif /* DTRACE_PROBE3 */ 189 190 #ifdef DTRACE_PROBE4 191 #undef DTRACE_PROBE4 192 #define DTRACE_PROBE4(a, b, c, d, e, f, g, h, i) ((void)0) 193 #endif /* DTRACE_PROBE4 */ 194 195 /* 196 * Threads 197 */ 198 #define curthread ((void *)(uintptr_t)thr_self()) 199 200 typedef struct kthread kthread_t; 201 202 #define thread_create(stk, stksize, func, arg, len, pp, state, pri) \ 203 zk_thread_create(func, arg) 204 #define thread_exit() thr_exit(NULL) 205 #define thread_join(t) panic("libzpool cannot join threads") 206 207 #define newproc(f, a, cid, pri, ctp, pid) (ENOSYS) 208 209 /* in libzpool, p0 exists only to have its address taken */ 210 struct proc { 211 uintptr_t this_is_never_used_dont_dereference_it; 212 }; 213 214 extern struct proc p0; 215 216 #define PS_NONE -1 217 218 extern kthread_t *zk_thread_create(void (*func)(), void *arg); 219 220 #define issig(why) (FALSE) 221 #define ISSIG(thr, why) (FALSE) 222 223 /* 224 * Mutexes 225 */ 226 typedef struct kmutex { 227 void *m_owner; 228 boolean_t initialized; 229 mutex_t m_lock; 230 } kmutex_t; 231 232 #define MUTEX_DEFAULT USYNC_THREAD 233 #undef MUTEX_HELD 234 #undef MUTEX_NOT_HELD 235 #define MUTEX_HELD(m) _mutex_held(&(m)->m_lock) 236 #define MUTEX_NOT_HELD(m) (!MUTEX_HELD(m)) 237 238 /* 239 * Argh -- we have to get cheesy here because the kernel and userland 240 * have different signatures for the same routine. 241 */ 242 extern int _mutex_init(mutex_t *mp, int type, void *arg); 243 extern int _mutex_destroy(mutex_t *mp); 244 245 #define mutex_init(mp, b, c, d) zmutex_init((kmutex_t *)(mp)) 246 #define mutex_destroy(mp) zmutex_destroy((kmutex_t *)(mp)) 247 248 extern void zmutex_init(kmutex_t *mp); 249 extern void zmutex_destroy(kmutex_t *mp); 250 extern void mutex_enter(kmutex_t *mp); 251 extern void mutex_exit(kmutex_t *mp); 252 extern int mutex_tryenter(kmutex_t *mp); 253 extern void *mutex_owner(kmutex_t *mp); 254 255 /* 256 * RW locks 257 */ 258 typedef struct krwlock { 259 void *rw_owner; 260 boolean_t initialized; 261 rwlock_t rw_lock; 262 } krwlock_t; 263 264 typedef int krw_t; 265 266 #define RW_READER 0 267 #define RW_WRITER 1 268 #define RW_DEFAULT USYNC_THREAD 269 270 #undef RW_READ_HELD 271 #define RW_READ_HELD(x) _rw_read_held(&(x)->rw_lock) 272 273 #undef RW_WRITE_HELD 274 #define RW_WRITE_HELD(x) _rw_write_held(&(x)->rw_lock) 275 276 extern void rw_init(krwlock_t *rwlp, char *name, int type, void *arg); 277 extern void rw_destroy(krwlock_t *rwlp); 278 extern void rw_enter(krwlock_t *rwlp, krw_t rw); 279 extern int rw_tryenter(krwlock_t *rwlp, krw_t rw); 280 extern int rw_tryupgrade(krwlock_t *rwlp); 281 extern void rw_exit(krwlock_t *rwlp); 282 #define rw_downgrade(rwlp) do { } while (0) 283 284 extern uid_t crgetuid(cred_t *cr); 285 extern gid_t crgetgid(cred_t *cr); 286 extern int crgetngroups(cred_t *cr); 287 extern gid_t *crgetgroups(cred_t *cr); 288 289 /* 290 * Condition variables 291 */ 292 typedef cond_t kcondvar_t; 293 294 #define CV_DEFAULT USYNC_THREAD 295 296 extern void cv_init(kcondvar_t *cv, char *name, int type, void *arg); 297 extern void cv_destroy(kcondvar_t *cv); 298 extern void cv_wait(kcondvar_t *cv, kmutex_t *mp); 299 extern clock_t cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime); 300 extern void cv_signal(kcondvar_t *cv); 301 extern void cv_broadcast(kcondvar_t *cv); 302 303 /* 304 * kstat creation, installation and deletion 305 */ 306 extern kstat_t *kstat_create(char *, int, 307 char *, char *, uchar_t, ulong_t, uchar_t); 308 extern void kstat_install(kstat_t *); 309 extern void kstat_delete(kstat_t *); 310 311 /* 312 * Kernel memory 313 */ 314 #define KM_SLEEP UMEM_NOFAIL 315 #define KM_PUSHPAGE KM_SLEEP 316 #define KM_NOSLEEP UMEM_DEFAULT 317 #define KMC_NODEBUG UMC_NODEBUG 318 #define KMC_NOTOUCH 0 /* not needed for userland caches */ 319 #define kmem_alloc(_s, _f) umem_alloc(_s, _f) 320 #define kmem_zalloc(_s, _f) umem_zalloc(_s, _f) 321 #define kmem_free(_b, _s) umem_free(_b, _s) 322 #define kmem_cache_create(_a, _b, _c, _d, _e, _f, _g, _h, _i) \ 323 umem_cache_create(_a, _b, _c, _d, _e, _f, _g, _h, _i) 324 #define kmem_cache_destroy(_c) umem_cache_destroy(_c) 325 #define kmem_cache_alloc(_c, _f) umem_cache_alloc(_c, _f) 326 #define kmem_cache_free(_c, _b) umem_cache_free(_c, _b) 327 #define kmem_debugging() 0 328 #define kmem_cache_reap_now(_c) /* nothing */ 329 #define kmem_cache_set_move(_c, _cb) /* nothing */ 330 #define POINTER_INVALIDATE(_pp) /* nothing */ 331 #define POINTER_IS_VALID(_p) 0 332 333 typedef umem_cache_t kmem_cache_t; 334 335 typedef enum kmem_cbrc { 336 KMEM_CBRC_YES, 337 KMEM_CBRC_NO, 338 KMEM_CBRC_LATER, 339 KMEM_CBRC_DONT_NEED, 340 KMEM_CBRC_DONT_KNOW 341 } kmem_cbrc_t; 342 343 /* 344 * Task queues 345 */ 346 typedef struct taskq taskq_t; 347 typedef uintptr_t taskqid_t; 348 typedef void (task_func_t)(void *); 349 350 #define TASKQ_PREPOPULATE 0x0001 351 #define TASKQ_CPR_SAFE 0x0002 /* Use CPR safe protocol */ 352 #define TASKQ_DYNAMIC 0x0004 /* Use dynamic thread scheduling */ 353 #define TASKQ_THREADS_CPU_PCT 0x0008 /* Scale # threads by # cpus */ 354 #define TASKQ_DC_BATCH 0x0010 /* Mark threads as batch */ 355 356 #define TQ_SLEEP KM_SLEEP /* Can block for memory */ 357 #define TQ_NOSLEEP KM_NOSLEEP /* cannot block for memory; may fail */ 358 #define TQ_NOQUEUE 0x02 /* Do not enqueue if can't dispatch */ 359 #define TQ_FRONT 0x08 /* Queue in front */ 360 361 extern taskq_t *system_taskq; 362 363 extern taskq_t *taskq_create(const char *, int, pri_t, int, int, uint_t); 364 #define taskq_create_proc(a, b, c, d, e, p, f) \ 365 (taskq_create(a, b, c, d, e, f)) 366 #define taskq_create_sysdc(a, b, d, e, p, dc, f) \ 367 (taskq_create(a, b, maxclsyspri, d, e, f)) 368 extern taskqid_t taskq_dispatch(taskq_t *, task_func_t, void *, uint_t); 369 extern void taskq_destroy(taskq_t *); 370 extern void taskq_wait(taskq_t *); 371 extern int taskq_member(taskq_t *, void *); 372 extern void system_taskq_init(void); 373 extern void system_taskq_fini(void); 374 375 #define XVA_MAPSIZE 3 376 #define XVA_MAGIC 0x78766174 377 378 /* 379 * vnodes 380 */ 381 typedef struct vnode { 382 uint64_t v_size; 383 int v_fd; 384 char *v_path; 385 } vnode_t; 386 387 #define AV_SCANSTAMP_SZ 32 /* length of anti-virus scanstamp */ 388 389 typedef struct xoptattr { 390 timestruc_t xoa_createtime; /* Create time of file */ 391 uint8_t xoa_archive; 392 uint8_t xoa_system; 393 uint8_t xoa_readonly; 394 uint8_t xoa_hidden; 395 uint8_t xoa_nounlink; 396 uint8_t xoa_immutable; 397 uint8_t xoa_appendonly; 398 uint8_t xoa_nodump; 399 uint8_t xoa_settable; 400 uint8_t xoa_opaque; 401 uint8_t xoa_av_quarantined; 402 uint8_t xoa_av_modified; 403 uint8_t xoa_av_scanstamp[AV_SCANSTAMP_SZ]; 404 uint8_t xoa_reparse; 405 uint8_t xoa_offline; 406 uint8_t xoa_sparse; 407 } xoptattr_t; 408 409 typedef struct vattr { 410 uint_t va_mask; /* bit-mask of attributes */ 411 u_offset_t va_size; /* file size in bytes */ 412 } vattr_t; 413 414 415 typedef struct xvattr { 416 vattr_t xva_vattr; /* Embedded vattr structure */ 417 uint32_t xva_magic; /* Magic Number */ 418 uint32_t xva_mapsize; /* Size of attr bitmap (32-bit words) */ 419 uint32_t *xva_rtnattrmapp; /* Ptr to xva_rtnattrmap[] */ 420 uint32_t xva_reqattrmap[XVA_MAPSIZE]; /* Requested attrs */ 421 uint32_t xva_rtnattrmap[XVA_MAPSIZE]; /* Returned attrs */ 422 xoptattr_t xva_xoptattrs; /* Optional attributes */ 423 } xvattr_t; 424 425 typedef struct vsecattr { 426 uint_t vsa_mask; /* See below */ 427 int vsa_aclcnt; /* ACL entry count */ 428 void *vsa_aclentp; /* pointer to ACL entries */ 429 int vsa_dfaclcnt; /* default ACL entry count */ 430 void *vsa_dfaclentp; /* pointer to default ACL entries */ 431 size_t vsa_aclentsz; /* ACE size in bytes of vsa_aclentp */ 432 } vsecattr_t; 433 434 #define AT_TYPE 0x00001 435 #define AT_MODE 0x00002 436 #define AT_UID 0x00004 437 #define AT_GID 0x00008 438 #define AT_FSID 0x00010 439 #define AT_NODEID 0x00020 440 #define AT_NLINK 0x00040 441 #define AT_SIZE 0x00080 442 #define AT_ATIME 0x00100 443 #define AT_MTIME 0x00200 444 #define AT_CTIME 0x00400 445 #define AT_RDEV 0x00800 446 #define AT_BLKSIZE 0x01000 447 #define AT_NBLOCKS 0x02000 448 #define AT_SEQ 0x08000 449 #define AT_XVATTR 0x10000 450 451 #define CRCREAT 0 452 453 extern int fop_getattr(vnode_t *vp, vattr_t *vap); 454 455 #define VOP_CLOSE(vp, f, c, o, cr, ct) 0 456 #define VOP_PUTPAGE(vp, of, sz, fl, cr, ct) 0 457 #define VOP_GETATTR(vp, vap, fl, cr, ct) fop_getattr((vp), (vap)); 458 459 #define VOP_FSYNC(vp, f, cr, ct) fsync((vp)->v_fd) 460 461 #define VN_RELE(vp) vn_close(vp) 462 463 extern int vn_open(char *path, int x1, int oflags, int mode, vnode_t **vpp, 464 int x2, int x3); 465 extern int vn_openat(char *path, int x1, int oflags, int mode, vnode_t **vpp, 466 int x2, int x3, vnode_t *vp, int fd); 467 extern int vn_rdwr(int uio, vnode_t *vp, void *addr, ssize_t len, 468 offset_t offset, int x1, int x2, rlim64_t x3, void *x4, ssize_t *residp); 469 extern void vn_close(vnode_t *vp); 470 471 #define vn_remove(path, x1, x2) remove(path) 472 #define vn_rename(from, to, seg) rename((from), (to)) 473 #define vn_is_readonly(vp) B_FALSE 474 475 extern vnode_t *rootdir; 476 477 #include <sys/file.h> /* for FREAD, FWRITE, etc */ 478 479 /* 480 * Random stuff 481 */ 482 #define ddi_get_lbolt() (gethrtime() >> 23) 483 #define ddi_get_lbolt64() (gethrtime() >> 23) 484 #define hz 119 /* frequency when using gethrtime() >> 23 for lbolt */ 485 486 extern void delay(clock_t ticks); 487 488 #define gethrestime_sec() time(NULL) 489 #define gethrestime(t) \ 490 do {\ 491 (t)->tv_sec = gethrestime_sec();\ 492 (t)->tv_nsec = 0;\ 493 } while (0); 494 495 #define max_ncpus 64 496 497 #define minclsyspri 60 498 #define maxclsyspri 99 499 500 #define CPU_SEQID (thr_self() & (max_ncpus - 1)) 501 502 #define kcred NULL 503 #define CRED() NULL 504 505 #define ptob(x) ((x) * PAGESIZE) 506 507 extern uint64_t physmem; 508 509 extern int highbit(ulong_t i); 510 extern int random_get_bytes(uint8_t *ptr, size_t len); 511 extern int random_get_pseudo_bytes(uint8_t *ptr, size_t len); 512 513 extern void kernel_init(int); 514 extern void kernel_fini(void); 515 516 struct spa; 517 extern void nicenum(uint64_t num, char *buf); 518 extern void show_pool_stats(struct spa *); 519 520 typedef struct callb_cpr { 521 kmutex_t *cc_lockp; 522 } callb_cpr_t; 523 524 #define CALLB_CPR_INIT(cp, lockp, func, name) { \ 525 (cp)->cc_lockp = lockp; \ 526 } 527 528 #define CALLB_CPR_SAFE_BEGIN(cp) { \ 529 ASSERT(MUTEX_HELD((cp)->cc_lockp)); \ 530 } 531 532 #define CALLB_CPR_SAFE_END(cp, lockp) { \ 533 ASSERT(MUTEX_HELD((cp)->cc_lockp)); \ 534 } 535 536 #define CALLB_CPR_EXIT(cp) { \ 537 ASSERT(MUTEX_HELD((cp)->cc_lockp)); \ 538 mutex_exit((cp)->cc_lockp); \ 539 } 540 541 #define zone_dataset_visible(x, y) (1) 542 #define INGLOBALZONE(z) (1) 543 544 extern char *kmem_asprintf(const char *fmt, ...); 545 #define strfree(str) kmem_free((str), strlen(str)+1) 546 547 /* 548 * Hostname information 549 */ 550 extern char hw_serial[]; /* for userland-emulated hostid access */ 551 extern int ddi_strtoul(const char *str, char **nptr, int base, 552 unsigned long *result); 553 554 extern int ddi_strtoull(const char *str, char **nptr, int base, 555 u_longlong_t *result); 556 557 /* ZFS Boot Related stuff. */ 558 559 struct _buf { 560 intptr_t _fd; 561 }; 562 563 struct bootstat { 564 uint64_t st_size; 565 }; 566 567 typedef struct ace_object { 568 uid_t a_who; 569 uint32_t a_access_mask; 570 uint16_t a_flags; 571 uint16_t a_type; 572 uint8_t a_obj_type[16]; 573 uint8_t a_inherit_obj_type[16]; 574 } ace_object_t; 575 576 577 #define ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE 0x05 578 #define ACE_ACCESS_DENIED_OBJECT_ACE_TYPE 0x06 579 #define ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE 0x07 580 #define ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE 0x08 581 582 extern struct _buf *kobj_open_file(char *name); 583 extern int kobj_read_file(struct _buf *file, char *buf, unsigned size, 584 unsigned off); 585 extern void kobj_close_file(struct _buf *file); 586 extern int kobj_get_filesize(struct _buf *file, uint64_t *size); 587 extern int zfs_secpolicy_snapshot_perms(const char *name, cred_t *cr); 588 extern int zfs_secpolicy_rename_perms(const char *from, const char *to, 589 cred_t *cr); 590 extern int zfs_secpolicy_destroy_perms(const char *name, cred_t *cr); 591 extern zoneid_t getzoneid(void); 592 593 /* SID stuff */ 594 typedef struct ksiddomain { 595 uint_t kd_ref; 596 uint_t kd_len; 597 char *kd_name; 598 } ksiddomain_t; 599 600 ksiddomain_t *ksid_lookupdomain(const char *); 601 void ksiddomain_rele(ksiddomain_t *); 602 603 #define DDI_SLEEP KM_SLEEP 604 #define ddi_log_sysevent(_a, _b, _c, _d, _e, _f, _g) \ 605 sysevent_post_event(_c, _d, _b, "libzpool", _e, _f) 606 607 #ifdef __cplusplus 608 } 609 #endif 610 611 #endif /* _SYS_ZFS_CONTEXT_H */ 612