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