lib/libzpool/kernel.c

   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) 2016 Actifio, Inc. All rights reserved.
  24  */
  25
  26 #include <assert.h>
  27 #include <fcntl.h>
  28 #include <poll.h>
  29 #include <stdio.h>
  30 #include <stdlib.h>
  31 #include <string.h>
  32 #include <zlib.h>
  33 #include <libgen.h>
  34 #include <sys/signal.h>
  35 #include <sys/spa.h>
  36 #include <sys/stat.h>
  37 #include <sys/processor.h>
  38 #include <sys/zfs_context.h>
  39 #include <sys/rrwlock.h>
  40 #include <sys/utsname.h>
  41 #include <sys/time.h>
  42 #include <sys/systeminfo.h>
  43 #include <zfs_fletcher.h>
  44 #include <sys/crypto/icp.h>
  45
  46 /*
  47  * Emulation of kernel services in userland.
  48  */
  49
  50 int aok;
  51 uint64_t physmem;
  52 vnode_t *rootdir = (vnode_t *)0xabcd1234;
  53 char hw_serial[HW_HOSTID_LEN];
  54 struct utsname hw_utsname;
  55 vmem_t *zio_arena = NULL;
  56
  57 /* If set, all blocks read will be copied to the specified directory. */
  58 char *vn_dumpdir = NULL;
  59
  60 /* this only exists to have its address taken */
  61 struct proc p0;
  62
  63 /*
  64  * =========================================================================
  65  * threads
  66  * =========================================================================
  67  */
  68
  69 pthread_cond_t kthread_cond = PTHREAD_COND_INITIALIZER;
  70 pthread_mutex_t kthread_lock = PTHREAD_MUTEX_INITIALIZER;
  71 pthread_key_t kthread_key;
  72 int kthread_nr = 0;
  73
  74 void
  75 thread_init(void)
  76 {
  77         kthread_t *kt;
  78
  79         VERIFY3S(pthread_key_create(&kthread_key, NULL), ==, 0);
  80
  81         /* Create entry for primary kthread */
  82         kt = umem_zalloc(sizeof (kthread_t), UMEM_NOFAIL);
  83         kt->t_tid = pthread_self();
  84         kt->t_func = NULL;
  85
  86         VERIFY3S(pthread_setspecific(kthread_key, kt), ==, 0);
  87
  88         /* Only the main thread should be running at the moment */
  89         ASSERT3S(kthread_nr, ==, 0);
  90         kthread_nr = 1;
  91 }
  92
  93 void
  94 thread_fini(void)
  95 {
  96         kthread_t *kt = curthread;
  97
  98         ASSERT(pthread_equal(kt->t_tid, pthread_self()));
  99         ASSERT3P(kt->t_func, ==, NULL);
 100
 101         umem_free(kt, sizeof (kthread_t));
 102
 103         /* Wait for all threads to exit via thread_exit() */
 104         VERIFY3S(pthread_mutex_lock(&kthread_lock), ==, 0);
 105
 106         kthread_nr--; /* Main thread is exiting */
 107
 108         while (kthread_nr > 0)
 109                 VERIFY0(pthread_cond_wait(&kthread_cond, &kthread_lock));
 110
 111         ASSERT3S(kthread_nr, ==, 0);
 112         VERIFY3S(pthread_mutex_unlock(&kthread_lock), ==, 0);
 113
 114         VERIFY3S(pthread_key_delete(kthread_key), ==, 0);
 115 }
 116
 117 kthread_t *
 118 zk_thread_current(void)
 119 {
 120         kthread_t *kt = pthread_getspecific(kthread_key);
 121
 122         ASSERT3P(kt, !=, NULL);
 123
 124         return (kt);
 125 }
 126
 127 void *
 128 zk_thread_helper(void *arg)
 129 {
 130         kthread_t *kt = (kthread_t *) arg;
 131
 132         VERIFY3S(pthread_setspecific(kthread_key, kt), ==, 0);
 133
 134         VERIFY3S(pthread_mutex_lock(&kthread_lock), ==, 0);
 135         kthread_nr++;
 136         VERIFY3S(pthread_mutex_unlock(&kthread_lock), ==, 0);
 137         (void) setpriority(PRIO_PROCESS, 0, kt->t_pri);
 138
 139         kt->t_tid = pthread_self();
 140         ((thread_func_arg_t) kt->t_func)(kt->t_arg);
 141
 142         /* Unreachable, thread must exit with thread_exit() */
 143         abort();
 144
 145         return (NULL);
 146 }
 147
 148 kthread_t *
 149 zk_thread_create(caddr_t stk, size_t stksize, thread_func_t func, void *arg,
 150     size_t len, proc_t *pp, int state, pri_t pri, int detachstate)
 151 {
 152         kthread_t *kt;
 153         pthread_attr_t attr;
 154         char *stkstr;
 155
 156         ASSERT0(state & ~TS_RUN);
 157
 158         kt = umem_zalloc(sizeof (kthread_t), UMEM_NOFAIL);
 159         kt->t_func = func;
 160         kt->t_arg = arg;
 161         kt->t_pri = pri;
 162
 163         VERIFY0(pthread_attr_init(&attr));
 164         VERIFY0(pthread_attr_setdetachstate(&attr, detachstate));
 165
 166         /*
 167          * We allow the default stack size in user space to be specified by
 168          * setting the ZFS_STACK_SIZE environment variable.  This allows us
 169          * the convenience of observing and debugging stack overruns in
 170          * user space.  Explicitly specified stack sizes will be honored.
 171          * The usage of ZFS_STACK_SIZE is discussed further in the
 172          * ENVIRONMENT VARIABLES sections of the ztest(1) man page.
 173          */
 174         if (stksize == 0) {
 175                 stkstr = getenv("ZFS_STACK_SIZE");
 176
 177                 if (stkstr == NULL)
 178                         stksize = TS_STACK_MAX;
 179                 else
 180                         stksize = MAX(atoi(stkstr), TS_STACK_MIN);
 181         }
 182
 183         VERIFY3S(stksize, >, 0);
 184         stksize = P2ROUNDUP(MAX(stksize, TS_STACK_MIN), PAGESIZE);
 185         /*
 186          * If this ever fails, it may be because the stack size is not a
 187          * multiple of system page size.
 188          */
 189         VERIFY0(pthread_attr_setstacksize(&attr, stksize));
 190         VERIFY0(pthread_attr_setguardsize(&attr, PAGESIZE));
 191
 192         VERIFY0(pthread_create(&kt->t_tid, &attr, &zk_thread_helper, kt));
 193         VERIFY0(pthread_attr_destroy(&attr));
 194
 195         return (kt);
 196 }
 197
 198 void
 199 zk_thread_exit(void)
 200 {
 201         kthread_t *kt = curthread;
 202
 203         ASSERT(pthread_equal(kt->t_tid, pthread_self()));
 204
 205         umem_free(kt, sizeof (kthread_t));
 206
 207         VERIFY0(pthread_mutex_lock(&kthread_lock));
 208         kthread_nr--;
 209         VERIFY0(pthread_mutex_unlock(&kthread_lock));
 210
 211         VERIFY0(pthread_cond_broadcast(&kthread_cond));
 212         pthread_exit((void *)TS_MAGIC);
 213 }
 214
 215 void
 216 zk_thread_join(kt_did_t tid)
 217 {
 218         void *ret;
 219
 220         pthread_join((pthread_t)tid, &ret);
 221         VERIFY3P(ret, ==, (void *)TS_MAGIC);
 222 }
 223
 224 /*
 225  * =========================================================================
 226  * kstats
 227  * =========================================================================
 228  */
 229 /*ARGSUSED*/
 230 kstat_t *
 231 kstat_create(const char *module, int instance, const char *name,
 232     const char *class, uchar_t type, ulong_t ndata, uchar_t ks_flag)
 233 {
 234         return (NULL);
 235 }
 236
 237 /*ARGSUSED*/
 238 void
 239 kstat_install(kstat_t *ksp)
 240 {}
 241
 242 /*ARGSUSED*/
 243 void
 244 kstat_delete(kstat_t *ksp)
 245 {}
 246
 247 /*ARGSUSED*/
 248 void
 249 kstat_waitq_enter(kstat_io_t *kiop)
 250 {}
 251
 252 /*ARGSUSED*/
 253 void
 254 kstat_waitq_exit(kstat_io_t *kiop)
 255 {}
 256
 257 /*ARGSUSED*/
 258 void
 259 kstat_runq_enter(kstat_io_t *kiop)
 260 {}
 261
 262 /*ARGSUSED*/
 263 void
 264 kstat_runq_exit(kstat_io_t *kiop)
 265 {}
 266
 267 /*ARGSUSED*/
 268 void
 269 kstat_waitq_to_runq(kstat_io_t *kiop)
 270 {}
 271
 272 /*ARGSUSED*/
 273 void
 274 kstat_runq_back_to_waitq(kstat_io_t *kiop)
 275 {}
 276
 277 void
 278 kstat_set_raw_ops(kstat_t *ksp,
 279     int (*headers)(char *buf, size_t size),
 280     int (*data)(char *buf, size_t size, void *data),
 281     void *(*addr)(kstat_t *ksp, loff_t index))
 282 {}
 283
 284 /*
 285  * =========================================================================
 286  * mutexes
 287  * =========================================================================
 288  */
 289
 290 void
 291 mutex_init(kmutex_t *mp, char *name, int type, void *cookie)
 292 {
 293         ASSERT3S(type, ==, MUTEX_DEFAULT);
 294         ASSERT3P(cookie, ==, NULL);
 295         mp->m_owner = MTX_INIT;
 296         mp->m_magic = MTX_MAGIC;
 297         VERIFY3S(pthread_mutex_init(&mp->m_lock, NULL), ==, 0);
 298 }
 299
 300 void
 301 mutex_destroy(kmutex_t *mp)
 302 {
 303         ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
 304         ASSERT3P(mp->m_owner, ==, MTX_INIT);
 305         ASSERT0(pthread_mutex_destroy(&(mp)->m_lock));
 306         mp->m_owner = MTX_DEST;
 307         mp->m_magic = 0;
 308 }
 309
 310 void
 311 mutex_enter(kmutex_t *mp)
 312 {
 313         ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
 314         ASSERT3P(mp->m_owner, !=, MTX_DEST);
 315         ASSERT3P(mp->m_owner, !=, curthread);
 316         VERIFY3S(pthread_mutex_lock(&mp->m_lock), ==, 0);
 317         ASSERT3P(mp->m_owner, ==, MTX_INIT);
 318         mp->m_owner = curthread;
 319 }
 320
 321 int
 322 mutex_tryenter(kmutex_t *mp)
 323 {
 324         int err;
 325         ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
 326         ASSERT3P(mp->m_owner, !=, MTX_DEST);
 327         if (0 == (err = pthread_mutex_trylock(&mp->m_lock))) {
 328                 ASSERT3P(mp->m_owner, ==, MTX_INIT);
 329                 mp->m_owner = curthread;
 330                 return (1);
 331         } else {
 332                 VERIFY3S(err, ==, EBUSY);
 333                 return (0);
 334         }
 335 }
 336
 337 void
 338 mutex_exit(kmutex_t *mp)
 339 {
 340         ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
 341         ASSERT3P(mutex_owner(mp), ==, curthread);
 342         mp->m_owner = MTX_INIT;
 343         VERIFY3S(pthread_mutex_unlock(&mp->m_lock), ==, 0);
 344 }
 345
 346 void *
 347 mutex_owner(kmutex_t *mp)
 348 {
 349         ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
 350         return (mp->m_owner);
 351 }
 352
 353 int
 354 mutex_held(kmutex_t *mp)
 355 {
 356         return (mp->m_owner == curthread);
 357 }
 358
 359 /*
 360  * =========================================================================
 361  * rwlocks
 362  * =========================================================================
 363  */
 364
 365 void
 366 rw_init(krwlock_t *rwlp, char *name, int type, void *arg)
 367 {
 368         ASSERT3S(type, ==, RW_DEFAULT);
 369         ASSERT3P(arg, ==, NULL);
 370         VERIFY3S(pthread_rwlock_init(&rwlp->rw_lock, NULL), ==, 0);
 371         rwlp->rw_owner = RW_INIT;
 372         rwlp->rw_wr_owner = RW_INIT;
 373         rwlp->rw_readers = 0;
 374         rwlp->rw_magic = RW_MAGIC;
 375 }
 376
 377 void
 378 rw_destroy(krwlock_t *rwlp)
 379 {
 380         ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
 381         ASSERT(rwlp->rw_readers == 0 && rwlp->rw_wr_owner == RW_INIT);
 382         VERIFY3S(pthread_rwlock_destroy(&rwlp->rw_lock), ==, 0);
 383         rwlp->rw_magic = 0;
 384 }
 385
 386 void
 387 rw_enter(krwlock_t *rwlp, krw_t rw)
 388 {
 389         ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
 390         ASSERT3P(rwlp->rw_owner, !=, curthread);
 391         ASSERT3P(rwlp->rw_wr_owner, !=, curthread);
 392
 393         if (rw == RW_READER) {
 394                 VERIFY3S(pthread_rwlock_rdlock(&rwlp->rw_lock), ==, 0);
 395                 ASSERT3P(rwlp->rw_wr_owner, ==, RW_INIT);
 396
 397                 atomic_inc_uint(&rwlp->rw_readers);
 398         } else {
 399                 VERIFY3S(pthread_rwlock_wrlock(&rwlp->rw_lock), ==, 0);
 400                 ASSERT3P(rwlp->rw_wr_owner, ==, RW_INIT);
 401                 ASSERT3U(rwlp->rw_readers, ==, 0);
 402
 403                 rwlp->rw_wr_owner = curthread;
 404         }
 405
 406         rwlp->rw_owner = curthread;
 407 }
 408
 409 void
 410 rw_exit(krwlock_t *rwlp)
 411 {
 412         ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
 413         ASSERT(RW_LOCK_HELD(rwlp));
 414
 415         if (RW_READ_HELD(rwlp))
 416                 atomic_dec_uint(&rwlp->rw_readers);
 417         else
 418                 rwlp->rw_wr_owner = RW_INIT;
 419
 420         rwlp->rw_owner = RW_INIT;
 421         VERIFY3S(pthread_rwlock_unlock(&rwlp->rw_lock), ==, 0);
 422 }
 423
 424 int
 425 rw_tryenter(krwlock_t *rwlp, krw_t rw)
 426 {
 427         int rv;
 428
 429         ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
 430
 431         if (rw == RW_READER)
 432                 rv = pthread_rwlock_tryrdlock(&rwlp->rw_lock);
 433         else
 434                 rv = pthread_rwlock_trywrlock(&rwlp->rw_lock);
 435
 436         if (rv == 0) {
 437                 ASSERT3P(rwlp->rw_wr_owner, ==, RW_INIT);
 438
 439                 if (rw == RW_READER)
 440                         atomic_inc_uint(&rwlp->rw_readers);
 441                 else {
 442                         ASSERT3U(rwlp->rw_readers, ==, 0);
 443                         rwlp->rw_wr_owner = curthread;
 444                 }
 445
 446                 rwlp->rw_owner = curthread;
 447                 return (1);
 448         }
 449
 450         VERIFY3S(rv, ==, EBUSY);
 451
 452         return (0);
 453 }
 454
 455 int
 456 rw_tryupgrade(krwlock_t *rwlp)
 457 {
 458         ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
 459
 460         return (0);
 461 }
 462
 463 /*
 464  * =========================================================================
 465  * condition variables
 466  * =========================================================================
 467  */
 468
 469 void
 470 cv_init(kcondvar_t *cv, char *name, int type, void *arg)
 471 {
 472         ASSERT3S(type, ==, CV_DEFAULT);
 473         cv->cv_magic = CV_MAGIC;
 474         VERIFY0(pthread_cond_init(&cv->cv, NULL));
 475 }
 476
 477 void
 478 cv_destroy(kcondvar_t *cv)
 479 {
 480         ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
 481         VERIFY0(pthread_cond_destroy(&cv->cv));
 482         cv->cv_magic = 0;
 483 }
 484
 485 void
 486 cv_wait(kcondvar_t *cv, kmutex_t *mp)
 487 {
 488         ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
 489         ASSERT3P(mutex_owner(mp), ==, curthread);
 490         mp->m_owner = MTX_INIT;
 491         VERIFY0(pthread_cond_wait(&cv->cv, &mp->m_lock));
 492         mp->m_owner = curthread;
 493 }
 494
 495 clock_t
 496 cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime)
 497 {
 498         int error;
 499         struct timeval tv;
 500         timestruc_t ts;
 501         clock_t delta;
 502
 503         ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
 504
 505         delta = abstime - ddi_get_lbolt();
 506         if (delta <= 0)
 507                 return (-1);
 508
 509         VERIFY(gettimeofday(&tv, NULL) == 0);
 510
 511         ts.tv_sec = tv.tv_sec + delta / hz;
 512         ts.tv_nsec = tv.tv_usec * NSEC_PER_USEC + (delta % hz) * (NANOSEC / hz);
 513         if (ts.tv_nsec >= NANOSEC) {
 514                 ts.tv_sec++;
 515                 ts.tv_nsec -= NANOSEC;
 516         }
 517
 518         ASSERT3P(mutex_owner(mp), ==, curthread);
 519         mp->m_owner = MTX_INIT;
 520         error = pthread_cond_timedwait(&cv->cv, &mp->m_lock, &ts);
 521         mp->m_owner = curthread;
 522
 523         if (error == ETIMEDOUT)
 524                 return (-1);
 525
 526         VERIFY0(error);
 527
 528         return (1);
 529 }
 530
 531 /*ARGSUSED*/
 532 clock_t
 533 cv_timedwait_hires(kcondvar_t *cv, kmutex_t *mp, hrtime_t tim, hrtime_t res,
 534     int flag)
 535 {
 536         int error;
 537         struct timeval tv;
 538         timestruc_t ts;
 539         hrtime_t delta;
 540
 541         ASSERT(flag == 0 || flag == CALLOUT_FLAG_ABSOLUTE);
 542
 543         delta = tim;
 544         if (flag & CALLOUT_FLAG_ABSOLUTE)
 545                 delta -= gethrtime();
 546
 547         if (delta <= 0)
 548                 return (-1);
 549
 550         VERIFY(gettimeofday(&tv, NULL) == 0);
 551
 552         ts.tv_sec = tv.tv_sec + delta / NANOSEC;
 553         ts.tv_nsec = tv.tv_usec * NSEC_PER_USEC + (delta % NANOSEC);
 554         if (ts.tv_nsec >= NANOSEC) {
 555                 ts.tv_sec++;
 556                 ts.tv_nsec -= NANOSEC;
 557         }
 558
 559         ASSERT(mutex_owner(mp) == curthread);
 560         mp->m_owner = MTX_INIT;
 561         error = pthread_cond_timedwait(&cv->cv, &mp->m_lock, &ts);
 562         mp->m_owner = curthread;
 563
 564         if (error == ETIMEDOUT)
 565                 return (-1);
 566
 567         VERIFY0(error);
 568
 569         return (1);
 570 }
 571
 572 void
 573 cv_signal(kcondvar_t *cv)
 574 {
 575         ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
 576         VERIFY0(pthread_cond_signal(&cv->cv));
 577 }
 578
 579 void
 580 cv_broadcast(kcondvar_t *cv)
 581 {
 582         ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
 583         VERIFY0(pthread_cond_broadcast(&cv->cv));
 584 }
 585
 586 /*
 587  * =========================================================================
 588  * vnode operations
 589  * =========================================================================
 590  */
 591 /*
 592  * Note: for the xxxat() versions of these functions, we assume that the
 593  * starting vp is always rootdir (which is true for spa_directory.c, the only
 594  * ZFS consumer of these interfaces).  We assert this is true, and then emulate
 595  * them by adding '/' in front of the path.
 596  */
 597
 598 /*ARGSUSED*/
 599 int
 600 vn_open(char *path, int x1, int flags, int mode, vnode_t **vpp, int x2, int x3)
 601 {
 602         int fd = -1;
 603         int dump_fd = -1;
 604         vnode_t *vp;
 605         int old_umask = 0;
 606         char *realpath;
 607         struct stat64 st;
 608         int err;
 609
 610         realpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
 611
 612         /*
 613          * If we're accessing a real disk from userland, we need to use
 614          * the character interface to avoid caching.  This is particularly
 615          * important if we're trying to look at a real in-kernel storage
 616          * pool from userland, e.g. via zdb, because otherwise we won't
 617          * see the changes occurring under the segmap cache.
 618          * On the other hand, the stupid character device returns zero
 619          * for its size.  So -- gag -- we open the block device to get
 620          * its size, and remember it for subsequent VOP_GETATTR().
 621          */
 622 #if defined(__sun__) || defined(__sun)
 623         if (strncmp(path, "/dev/", 5) == 0) {
 624 #else
 625         if (0) {
 626 #endif
 627                 char *dsk;
 628                 fd = open64(path, O_RDONLY);
 629                 if (fd == -1) {
 630                         err = errno;
 631                         free(realpath);
 632                         return (err);
 633                 }
 634                 if (fstat64(fd, &st) == -1) {
 635                         err = errno;
 636                         close(fd);
 637                         free(realpath);
 638                         return (err);
 639                 }
 640                 close(fd);
 641                 (void) sprintf(realpath, "%s", path);
 642                 dsk = strstr(path, "/dsk/");
 643                 if (dsk != NULL)
 644                         (void) sprintf(realpath + (dsk - path) + 1, "r%s",
 645                             dsk + 1);
 646         } else {
 647                 (void) sprintf(realpath, "%s", path);
 648                 if (!(flags & FCREAT) && stat64(realpath, &st) == -1) {
 649                         err = errno;
 650                         free(realpath);
 651                         return (err);
 652                 }
 653         }
 654
 655         if (!(flags & FCREAT) && S_ISBLK(st.st_mode)) {
 656 #ifdef __linux__
 657                 flags |= O_DIRECT;
 658 #endif
 659                 /* We shouldn't be writing to block devices in userspace */
 660                 VERIFY(!(flags & FWRITE));
 661         }
 662
 663         if (flags & FCREAT)
 664                 old_umask = umask(0);
 665
 666         /*
 667          * The construct 'flags - FREAD' conveniently maps combinations of
 668          * FREAD and FWRITE to the corresponding O_RDONLY, O_WRONLY, and O_RDWR.
 669          */
 670         fd = open64(realpath, flags - FREAD, mode);
 671         if (fd == -1) {
 672                 err = errno;
 673                 free(realpath);
 674                 return (err);
 675         }
 676
 677         if (flags & FCREAT)
 678                 (void) umask(old_umask);
 679
 680         if (vn_dumpdir != NULL) {
 681                 char *dumppath = umem_zalloc(MAXPATHLEN, UMEM_NOFAIL);
 682                 (void) snprintf(dumppath, MAXPATHLEN,
 683                     "%s/%s", vn_dumpdir, basename(realpath));
 684                 dump_fd = open64(dumppath, O_CREAT | O_WRONLY, 0666);
 685                 umem_free(dumppath, MAXPATHLEN);
 686                 if (dump_fd == -1) {
 687                         err = errno;
 688                         free(realpath);
 689                         close(fd);
 690                         return (err);
 691                 }
 692         } else {
 693                 dump_fd = -1;
 694         }
 695
 696         free(realpath);
 697
 698         if (fstat64_blk(fd, &st) == -1) {
 699                 err = errno;
 700                 close(fd);
 701                 if (dump_fd != -1)
 702                         close(dump_fd);
 703                 return (err);
 704         }
 705
 706         (void) fcntl(fd, F_SETFD, FD_CLOEXEC);
 707
 708         *vpp = vp = umem_zalloc(sizeof (vnode_t), UMEM_NOFAIL);
 709
 710         vp->v_fd = fd;
 711         vp->v_size = st.st_size;
 712         vp->v_path = spa_strdup(path);
 713         vp->v_dump_fd = dump_fd;
 714
 715         return (0);
 716 }
 717
 718 /*ARGSUSED*/
 719 int
 720 vn_openat(char *path, int x1, int flags, int mode, vnode_t **vpp, int x2,
 721     int x3, vnode_t *startvp, int fd)
 722 {
 723         char *realpath = umem_alloc(strlen(path) + 2, UMEM_NOFAIL);
 724         int ret;
 725
 726         ASSERT(startvp == rootdir);
 727         (void) sprintf(realpath, "/%s", path);
 728
 729         /* fd ignored for now, need if want to simulate nbmand support */
 730         ret = vn_open(realpath, x1, flags, mode, vpp, x2, x3);
 731
 732         umem_free(realpath, strlen(path) + 2);
 733
 734         return (ret);
 735 }
 736
 737 /*ARGSUSED*/
 738 int
 739 vn_rdwr(int uio, vnode_t *vp, void *addr, ssize_t len, offset_t offset,
 740         int x1, int x2, rlim64_t x3, void *x4, ssize_t *residp)
 741 {
 742         ssize_t rc, done = 0, split;
 743
 744         if (uio == UIO_READ) {
 745                 rc = pread64(vp->v_fd, addr, len, offset);
 746                 if (vp->v_dump_fd != -1 && rc != -1) {
 747                         int status;
 748                         status = pwrite64(vp->v_dump_fd, addr, rc, offset);
 749                         ASSERT(status != -1);
 750                 }
 751         } else {
 752                 /*
 753                  * To simulate partial disk writes, we split writes into two
 754                  * system calls so that the process can be killed in between.
 755                  */
 756                 int sectors = len >> SPA_MINBLOCKSHIFT;
 757                 split = (sectors > 0 ? rand() % sectors : 0) <<
 758                     SPA_MINBLOCKSHIFT;
 759                 rc = pwrite64(vp->v_fd, addr, split, offset);
 760                 if (rc != -1) {
 761                         done = rc;
 762                         rc = pwrite64(vp->v_fd, (char *)addr + split,
 763                             len - split, offset + split);
 764                 }
 765         }
 766
 767 #ifdef __linux__
 768         if (rc == -1 && errno == EINVAL) {
 769                 /*
 770                  * Under Linux, this most likely means an alignment issue
 771                  * (memory or disk) due to O_DIRECT, so we abort() in order to
 772                  * catch the offender.
 773                  */
 774                 abort();
 775         }
 776 #endif
 777         if (rc == -1)
 778                 return (errno);
 779
 780         done += rc;
 781
 782         if (residp)
 783                 *residp = len - done;
 784         else if (done != len)
 785                 return (EIO);
 786         return (0);
 787 }
 788
 789 void
 790 vn_close(vnode_t *vp)
 791 {
 792         close(vp->v_fd);
 793         if (vp->v_dump_fd != -1)
 794                 close(vp->v_dump_fd);
 795         spa_strfree(vp->v_path);
 796         umem_free(vp, sizeof (vnode_t));
 797 }
 798
 799 /*
 800  * At a minimum we need to update the size since vdev_reopen()
 801  * will no longer call vn_openat().
 802  */
 803 int
 804 fop_getattr(vnode_t *vp, vattr_t *vap)
 805 {
 806         struct stat64 st;
 807         int err;
 808
 809         if (fstat64_blk(vp->v_fd, &st) == -1) {
 810                 err = errno;
 811                 close(vp->v_fd);
 812                 return (err);
 813         }
 814
 815         vap->va_size = st.st_size;
 816         return (0);
 817 }
 818
 819 /*
 820  * =========================================================================
 821  * Figure out which debugging statements to print
 822  * =========================================================================
 823  */
 824
 825 static char *dprintf_string;
 826 static int dprintf_print_all;
 827
 828 int
 829 dprintf_find_string(const char *string)
 830 {
 831         char *tmp_str = dprintf_string;
 832         int len = strlen(string);
 833
 834         /*
 835          * Find out if this is a string we want to print.
 836          * String format: file1.c,function_name1,file2.c,file3.c
 837          */
 838
 839         while (tmp_str != NULL) {
 840                 if (strncmp(tmp_str, string, len) == 0 &&
 841                     (tmp_str[len] == ',' || tmp_str[len] == '\0'))
 842                         return (1);
 843                 tmp_str = strchr(tmp_str, ',');
 844                 if (tmp_str != NULL)
 845                         tmp_str++; /* Get rid of , */
 846         }
 847         return (0);
 848 }
 849
 850 void
 851 dprintf_setup(int *argc, char **argv)
 852 {
 853         int i, j;
 854
 855         /*
 856          * Debugging can be specified two ways: by setting the
 857          * environment variable ZFS_DEBUG, or by including a
 858          * "debug=..."  argument on the command line.  The command
 859          * line setting overrides the environment variable.
 860          */
 861
 862         for (i = 1; i < *argc; i++) {
 863                 int len = strlen("debug=");
 864                 /* First look for a command line argument */
 865                 if (strncmp("debug=", argv[i], len) == 0) {
 866                         dprintf_string = argv[i] + len;
 867                         /* Remove from args */
 868                         for (j = i; j < *argc; j++)
 869                                 argv[j] = argv[j+1];
 870                         argv[j] = NULL;
 871                         (*argc)--;
 872                 }
 873         }
 874
 875         if (dprintf_string == NULL) {
 876                 /* Look for ZFS_DEBUG environment variable */
 877                 dprintf_string = getenv("ZFS_DEBUG");
 878         }
 879
 880         /*
 881          * Are we just turning on all debugging?
 882          */
 883         if (dprintf_find_string("on"))
 884                 dprintf_print_all = 1;
 885
 886         if (dprintf_string != NULL)
 887                 zfs_flags |= ZFS_DEBUG_DPRINTF;
 888 }
 889
 890 /*
 891  * =========================================================================
 892  * debug printfs
 893  * =========================================================================
 894  */
 895 void
 896 __dprintf(const char *file, const char *func, int line, const char *fmt, ...)
 897 {
 898         const char *newfile;
 899         va_list adx;
 900
 901         /*
 902          * Get rid of annoying "../common/" prefix to filename.
 903          */
 904         newfile = strrchr(file, '/');
 905         if (newfile != NULL) {
 906                 newfile = newfile + 1; /* Get rid of leading / */
 907         } else {
 908                 newfile = file;
 909         }
 910
 911         if (dprintf_print_all ||
 912             dprintf_find_string(newfile) ||
 913             dprintf_find_string(func)) {
 914                 /* Print out just the function name if requested */
 915                 flockfile(stdout);
 916                 if (dprintf_find_string("pid"))
 917                         (void) printf("%d ", getpid());
 918                 if (dprintf_find_string("tid"))
 919                         (void) printf("%u ", (uint_t) pthread_self());
 920                 if (dprintf_find_string("cpu"))
 921                         (void) printf("%u ", getcpuid());
 922                 if (dprintf_find_string("time"))
 923                         (void) printf("%llu ", gethrtime());
 924                 if (dprintf_find_string("long"))
 925                         (void) printf("%s, line %d: ", newfile, line);
 926                 (void) printf("%s: ", func);
 927                 va_start(adx, fmt);
 928                 (void) vprintf(fmt, adx);
 929                 va_end(adx);
 930                 funlockfile(stdout);
 931         }
 932 }
 933
 934 /*
 935  * =========================================================================
 936  * cmn_err() and panic()
 937  * =========================================================================
 938  */
 939 static char ce_prefix[CE_IGNORE][10] = { "", "NOTICE: ", "WARNING: ", "" };
 940 static char ce_suffix[CE_IGNORE][2] = { "", "\n", "\n", "" };
 941
 942 void
 943 vpanic(const char *fmt, va_list adx)
 944 {
 945         (void) fprintf(stderr, "error: ");
 946         (void) vfprintf(stderr, fmt, adx);
 947         (void) fprintf(stderr, "\n");
 948
 949         abort();        /* think of it as a "user-level crash dump" */
 950 }
 951
 952 void
 953 panic(const char *fmt, ...)
 954 {
 955         va_list adx;
 956
 957         va_start(adx, fmt);
 958         vpanic(fmt, adx);
 959         va_end(adx);
 960 }
 961
 962 void
 963 vcmn_err(int ce, const char *fmt, va_list adx)
 964 {
 965         if (ce == CE_PANIC)
 966                 vpanic(fmt, adx);
 967         if (ce != CE_NOTE) {    /* suppress noise in userland stress testing */
 968                 (void) fprintf(stderr, "%s", ce_prefix[ce]);
 969                 (void) vfprintf(stderr, fmt, adx);
 970                 (void) fprintf(stderr, "%s", ce_suffix[ce]);
 971         }
 972 }
 973
 974 /*PRINTFLIKE2*/
 975 void
 976 cmn_err(int ce, const char *fmt, ...)
 977 {
 978         va_list adx;
 979
 980         va_start(adx, fmt);
 981         vcmn_err(ce, fmt, adx);
 982         va_end(adx);
 983 }
 984
 985 /*
 986  * =========================================================================
 987  * kobj interfaces
 988  * =========================================================================
 989  */
 990 struct _buf *
 991 kobj_open_file(char *name)
 992 {
 993         struct _buf *file;
 994         vnode_t *vp;
 995
 996         /* set vp as the _fd field of the file */
 997         if (vn_openat(name, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0, rootdir,
 998             -1) != 0)
 999                 return ((void *)-1UL);
1000
1001         file = umem_zalloc(sizeof (struct _buf), UMEM_NOFAIL);
1002         file->_fd = (intptr_t)vp;
1003         return (file);
1004 }
1005
1006 int
1007 kobj_read_file(struct _buf *file, char *buf, unsigned size, unsigned off)
1008 {
1009         ssize_t resid = 0;
1010
1011         if (vn_rdwr(UIO_READ, (vnode_t *)file->_fd, buf, size, (offset_t)off,
1012             UIO_SYSSPACE, 0, 0, 0, &resid) != 0)
1013                 return (-1);
1014
1015         return (size - resid);
1016 }
1017
1018 void
1019 kobj_close_file(struct _buf *file)
1020 {
1021         vn_close((vnode_t *)file->_fd);
1022         umem_free(file, sizeof (struct _buf));
1023 }
1024
1025 int
1026 kobj_get_filesize(struct _buf *file, uint64_t *size)
1027 {
1028         struct stat64 st;
1029         vnode_t *vp = (vnode_t *)file->_fd;
1030
1031         if (fstat64(vp->v_fd, &st) == -1) {
1032                 vn_close(vp);
1033                 return (errno);
1034         }
1035         *size = st.st_size;
1036         return (0);
1037 }
1038
1039 /*
1040  * =========================================================================
1041  * misc routines
1042  * =========================================================================
1043  */
1044
1045 void
1046 delay(clock_t ticks)
1047 {
1048         (void) poll(0, 0, ticks * (1000 / hz));
1049 }
1050
1051 /*
1052  * Find highest one bit set.
1053  *      Returns bit number + 1 of highest bit that is set, otherwise returns 0.
1054  * High order bit is 31 (or 63 in _LP64 kernel).
1055  */
1056 int
1057 highbit64(uint64_t i)
1058 {
1059         register int h = 1;
1060
1061         if (i == 0)
1062                 return (0);
1063         if (i & 0xffffffff00000000ULL) {
1064                 h += 32; i >>= 32;
1065         }
1066         if (i & 0xffff0000) {
1067                 h += 16; i >>= 16;
1068         }
1069         if (i & 0xff00) {
1070                 h += 8; i >>= 8;
1071         }
1072         if (i & 0xf0) {
1073                 h += 4; i >>= 4;
1074         }
1075         if (i & 0xc) {
1076                 h += 2; i >>= 2;
1077         }
1078         if (i & 0x2) {
1079                 h += 1;
1080         }
1081         return (h);
1082 }
1083
1084 /*
1085  * Find lowest one bit set.
1086  * Returns bit number + 1 of lowest bit that is set, otherwise returns 0.
1087  * This is basically a reimplementation of ffsll(), which is GNU specific.
1088  */
1089 int
1090 lowbit64(uint64_t i)
1091 {
1092         register int h = 64;
1093         if (i == 0)
1094                 return (0);
1095
1096         if (i & 0x00000000ffffffffULL)
1097                 h -= 32;
1098         else
1099                 i >>= 32;
1100
1101         if (i & 0x0000ffff)
1102                 h -= 16;
1103         else
1104                 i >>= 16;
1105
1106         if (i & 0x00ff)
1107                 h -= 8;
1108         else
1109                 i >>= 8;
1110
1111         if (i & 0x0f)
1112                 h -= 4;
1113         else
1114                 i >>= 4;
1115
1116         if (i & 0x3)
1117                 h -= 2;
1118         else
1119                 i >>= 2;
1120
1121         if (i & 0x1)
1122                 h -= 1;
1123
1124         return (h);
1125 }
1126
1127 /*
1128  * Find highest one bit set.
1129  * Returns bit number + 1 of highest bit that is set, otherwise returns 0.
1130  * High order bit is 31 (or 63 in _LP64 kernel).
1131  */
1132 int
1133 highbit(ulong_t i)
1134 {
1135 register int h = 1;
1136
1137         if (i == 0)
1138                 return (0);
1139 #ifdef _LP64
1140         if (i & 0xffffffff00000000ul) {
1141                 h += 32; i >>= 32;
1142         }
1143 #endif
1144         if (i & 0xffff0000) {
1145                 h += 16; i >>= 16;
1146         }
1147         if (i & 0xff00) {
1148                 h += 8; i >>= 8;
1149         }
1150         if (i & 0xf0) {
1151                 h += 4; i >>= 4;
1152         }
1153         if (i & 0xc) {
1154                 h += 2; i >>= 2;
1155         }
1156         if (i & 0x2) {
1157                 h += 1;
1158         }
1159         return (h);
1160 }
1161
1162 /*
1163  * Find lowest one bit set.
1164  *     Returns bit number + 1 of lowest bit that is set, otherwise returns 0.
1165  * Low order bit is 0.
1166  */
1167 int
1168 lowbit(ulong_t i)
1169 {
1170         register int h = 1;
1171
1172         if (i == 0)
1173                 return (0);
1174
1175 #ifdef _LP64
1176         if (!(i & 0xffffffff)) {
1177                 h += 32; i >>= 32;
1178         }
1179 #endif
1180         if (!(i & 0xffff)) {
1181                 h += 16; i >>= 16;
1182         }
1183         if (!(i & 0xff)) {
1184                 h += 8; i >>= 8;
1185         }
1186         if (!(i & 0xf)) {
1187                 h += 4; i >>= 4;
1188         }
1189         if (!(i & 0x3)) {
1190                 h += 2; i >>= 2;
1191         }
1192         if (!(i & 0x1)) {
1193                 h += 1;
1194         }
1195         return (h);
1196 }
1197
1198 static int random_fd = -1, urandom_fd = -1;
1199
1200 void
1201 random_init(void)
1202 {
1203         VERIFY((random_fd = open("/dev/random", O_RDONLY)) != -1);
1204         VERIFY((urandom_fd = open("/dev/urandom", O_RDONLY)) != -1);
1205 }
1206
1207 void
1208 random_fini(void)
1209 {
1210         close(random_fd);
1211         close(urandom_fd);
1212
1213         random_fd = -1;
1214         urandom_fd = -1;
1215 }
1216
1217 static int
1218 random_get_bytes_common(uint8_t *ptr, size_t len, int fd)
1219 {
1220         size_t resid = len;
1221         ssize_t bytes;
1222
1223         ASSERT(fd != -1);
1224
1225         while (resid != 0) {
1226                 bytes = read(fd, ptr, resid);
1227                 ASSERT3S(bytes, >=, 0);
1228                 ptr += bytes;
1229                 resid -= bytes;
1230         }
1231
1232         return (0);
1233 }
1234
1235 int
1236 random_get_bytes(uint8_t *ptr, size_t len)
1237 {
1238         return (random_get_bytes_common(ptr, len, random_fd));
1239 }
1240
1241 int
1242 random_get_pseudo_bytes(uint8_t *ptr, size_t len)
1243 {
1244         return (random_get_bytes_common(ptr, len, urandom_fd));
1245 }
1246
1247 int
1248 ddi_strtoul(const char *hw_serial, char **nptr, int base, unsigned long *result)
1249 {
1250         char *end;
1251
1252         *result = strtoul(hw_serial, &end, base);
1253         if (*result == 0)
1254                 return (errno);
1255         return (0);
1256 }
1257
1258 int
1259 ddi_strtoull(const char *str, char **nptr, int base, u_longlong_t *result)
1260 {
1261         char *end;
1262
1263         *result = strtoull(str, &end, base);
1264         if (*result == 0)
1265                 return (errno);
1266         return (0);
1267 }
1268
1269 utsname_t *
1270 utsname(void)
1271 {
1272         return (&hw_utsname);
1273 }
1274
1275 /*
1276  * =========================================================================
1277  * kernel emulation setup & teardown
1278  * =========================================================================
1279  */
1280 static int
1281 umem_out_of_memory(void)
1282 {
1283         char errmsg[] = "out of memory -- generating core dump\n";
1284
1285         (void) fprintf(stderr, "%s", errmsg);
1286         abort();
1287         return (0);
1288 }
1289
1290 static unsigned long
1291 get_spl_hostid(void)
1292 {
1293         FILE *f;
1294         unsigned long hostid;
1295
1296         f = fopen("/sys/module/spl/parameters/spl_hostid", "r");
1297         if (!f)
1298                 return (0);
1299         if (fscanf(f, "%lu", &hostid) != 1)
1300                 hostid = 0;
1301         fclose(f);
1302         return (hostid & 0xffffffff);
1303 }
1304
1305 unsigned long
1306 get_system_hostid(void)
1307 {
1308         unsigned long system_hostid = get_spl_hostid();
1309         if (system_hostid == 0)
1310                 system_hostid = gethostid() & 0xffffffff;
1311         return (system_hostid);
1312 }
1313
1314 void
1315 kernel_init(int mode)
1316 {
1317         extern uint_t rrw_tsd_key;
1318
1319         umem_nofail_callback(umem_out_of_memory);
1320
1321         physmem = sysconf(_SC_PHYS_PAGES);
1322
1323         dprintf("physmem = %llu pages (%.2f GB)\n", physmem,
1324             (double)physmem * sysconf(_SC_PAGE_SIZE) / (1ULL << 30));
1325
1326         (void) snprintf(hw_serial, sizeof (hw_serial), "%ld",
1327             (mode & FWRITE) ? get_system_hostid() : 0);
1328
1329         random_init();
1330
1331         VERIFY0(uname(&hw_utsname));
1332
1333         thread_init();
1334         system_taskq_init();
1335         icp_init();
1336
1337         spa_init(mode);
1338
1339         fletcher_4_init();
1340
1341         tsd_create(&rrw_tsd_key, rrw_tsd_destroy);
1342 }
1343
1344 void
1345 kernel_fini(void)
1346 {
1347         fletcher_4_fini();
1348         spa_fini();
1349
1350         icp_fini();
1351         system_taskq_fini();
1352         thread_fini();
1353
1354         random_fini();
1355 }
1356
1357 uid_t
1358 crgetuid(cred_t *cr)
1359 {
1360         return (0);
1361 }
1362
1363 uid_t
1364 crgetruid(cred_t *cr)
1365 {
1366         return (0);
1367 }
1368
1369 gid_t
1370 crgetgid(cred_t *cr)
1371 {
1372         return (0);
1373 }
1374
1375 int
1376 crgetngroups(cred_t *cr)
1377 {
1378         return (0);
1379 }
1380
1381 gid_t *
1382 crgetgroups(cred_t *cr)
1383 {
1384         return (NULL);
1385 }
1386
1387 int
1388 zfs_secpolicy_snapshot_perms(const char *name, cred_t *cr)
1389 {
1390         return (0);
1391 }
1392
1393 int
1394 zfs_secpolicy_rename_perms(const char *from, const char *to, cred_t *cr)
1395 {
1396         return (0);
1397 }
1398
1399 int
1400 zfs_secpolicy_destroy_perms(const char *name, cred_t *cr)
1401 {
1402         return (0);
1403 }
1404
1405 int
1406 secpolicy_zfs(const cred_t *cr)
1407 {
1408         return (0);
1409 }
1410
1411 ksiddomain_t *
1412 ksid_lookupdomain(const char *dom)
1413 {
1414         ksiddomain_t *kd;
1415
1416         kd = umem_zalloc(sizeof (ksiddomain_t), UMEM_NOFAIL);
1417         kd->kd_name = spa_strdup(dom);
1418         return (kd);
1419 }
1420
1421 void
1422 ksiddomain_rele(ksiddomain_t *ksid)
1423 {
1424         spa_strfree(ksid->kd_name);
1425         umem_free(ksid, sizeof (ksiddomain_t));
1426 }
1427
1428 char *
1429 kmem_vasprintf(const char *fmt, va_list adx)
1430 {
1431         char *buf = NULL;
1432         va_list adx_copy;
1433
1434         va_copy(adx_copy, adx);
1435         VERIFY(vasprintf(&buf, fmt, adx_copy) != -1);
1436         va_end(adx_copy);
1437
1438         return (buf);
1439 }
1440
1441 char *
1442 kmem_asprintf(const char *fmt, ...)
1443 {
1444         char *buf = NULL;
1445         va_list adx;
1446
1447         va_start(adx, fmt);
1448         VERIFY(vasprintf(&buf, fmt, adx) != -1);
1449         va_end(adx);
1450
1451         return (buf);
1452 }
1453
1454 /* ARGSUSED */
1455 int
1456 zfs_onexit_fd_hold(int fd, minor_t *minorp)
1457 {
1458         *minorp = 0;
1459         return (0);
1460 }
1461
1462 /* ARGSUSED */
1463 void
1464 zfs_onexit_fd_rele(int fd)
1465 {
1466 }
1467
1468 /* ARGSUSED */
1469 int
1470 zfs_onexit_add_cb(minor_t minor, void (*func)(void *), void *data,
1471     uint64_t *action_handle)
1472 {
1473         return (0);
1474 }
1475
1476 /* ARGSUSED */
1477 int
1478 zfs_onexit_del_cb(minor_t minor, uint64_t action_handle, boolean_t fire)
1479 {
1480         return (0);
1481 }
1482
1483 /* ARGSUSED */
1484 int
1485 zfs_onexit_cb_data(minor_t minor, uint64_t action_handle, void **data)
1486 {
1487         return (0);
1488 }
1489
1490 fstrans_cookie_t
1491 spl_fstrans_mark(void)
1492 {
1493         return ((fstrans_cookie_t) 0);
1494 }
1495
1496 void
1497 spl_fstrans_unmark(fstrans_cookie_t cookie)
1498 {
1499 }
1500
1501 int
1502 spl_fstrans_check(void)
1503 {
1504         return (0);
1505 }
1506
1507 void *zvol_tag = "zvol_tag";
1508
1509 void
1510 zvol_create_minors(spa_t *spa, const char *name, boolean_t async)
1511 {
1512 }
1513
1514 void
1515 zvol_remove_minor(spa_t *spa, const char *name, boolean_t async)
1516 {
1517 }
1518
1519 void
1520 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async)
1521 {
1522 }
1523
1524 void
1525 zvol_rename_minors(spa_t *spa, const char *oldname, const char *newname,
1526     boolean_t async)
1527 {
1528 }