1 /* 2 * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC. 3 * Copyright (C) 2007 The Regents of the University of California. 4 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). 5 * Written by Brian Behlendorf <behlendorf1@llnl.gov>. 6 * UCRL-CODE-235197 7 * 8 * This file is part of the SPL, Solaris Porting Layer. 9 * 10 * The SPL is free software; you can redistribute it and/or modify it 11 * under the terms of the GNU General Public License as published by the 12 * Free Software Foundation; either version 2 of the License, or (at your 13 * option) any later version. 14 * 15 * The SPL is distributed in the hope that it will be useful, but WITHOUT 16 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 17 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 18 * for more details. 19 * 20 * You should have received a copy of the GNU General Public License along 21 * with the SPL. If not, see <http://www.gnu.org/licenses/>. 22 * 23 * Solaris Porting Layer (SPL) Generic Implementation. 24 */ 25 26 #include <sys/sysmacros.h> 27 #include <sys/systeminfo.h> 28 #include <sys/vmsystm.h> 29 #include <sys/kmem.h> 30 #include <sys/kmem_cache.h> 31 #include <sys/vmem.h> 32 #include <sys/mutex.h> 33 #include <sys/rwlock.h> 34 #include <sys/taskq.h> 35 #include <sys/tsd.h> 36 #include <sys/zmod.h> 37 #include <sys/debug.h> 38 #include <sys/proc.h> 39 #include <sys/kstat.h> 40 #include <sys/file.h> 41 #include <sys/sunddi.h> 42 #include <linux/ctype.h> 43 #include <sys/disp.h> 44 #include <sys/random.h> 45 #include <sys/string.h> 46 #include <linux/kmod.h> 47 #include <linux/mod_compat.h> 48 #include <sys/cred.h> 49 #include <sys/vnode.h> 50 #include <sys/misc.h> 51 52 unsigned long spl_hostid = 0; 53 EXPORT_SYMBOL(spl_hostid); 54 55 /* CSTYLED */ 56 module_param(spl_hostid, ulong, 0644); 57 MODULE_PARM_DESC(spl_hostid, "The system hostid."); 58 59 proc_t p0; 60 EXPORT_SYMBOL(p0); 61 62 /* 63 * Xorshift Pseudo Random Number Generator based on work by Sebastiano Vigna 64 * 65 * "Further scramblings of Marsaglia's xorshift generators" 66 * http://vigna.di.unimi.it/ftp/papers/xorshiftplus.pdf 67 * 68 * random_get_pseudo_bytes() is an API function on Illumos whose sole purpose 69 * is to provide bytes containing random numbers. It is mapped to /dev/urandom 70 * on Illumos, which uses a "FIPS 186-2 algorithm". No user of the SPL's 71 * random_get_pseudo_bytes() needs bytes that are of cryptographic quality, so 72 * we can implement it using a fast PRNG that we seed using Linux' actual 73 * equivalent to random_get_pseudo_bytes(). We do this by providing each CPU 74 * with an independent seed so that all calls to random_get_pseudo_bytes() are 75 * free of atomic instructions. 76 * 77 * A consequence of using a fast PRNG is that using random_get_pseudo_bytes() 78 * to generate words larger than 128 bits will paradoxically be limited to 79 * `2^128 - 1` possibilities. This is because we have a sequence of `2^128 - 1` 80 * 128-bit words and selecting the first will implicitly select the second. If 81 * a caller finds this behavior undesirable, random_get_bytes() should be used 82 * instead. 83 * 84 * XXX: Linux interrupt handlers that trigger within the critical section 85 * formed by `s[1] = xp[1];` and `xp[0] = s[0];` and call this function will 86 * see the same numbers. Nothing in the code currently calls this in an 87 * interrupt handler, so this is considered to be okay. If that becomes a 88 * problem, we could create a set of per-cpu variables for interrupt handlers 89 * and use them when in_interrupt() from linux/preempt_mask.h evaluates to 90 * true. 91 */ 92 void __percpu *spl_pseudo_entropy; 93 94 /* 95 * spl_rand_next()/spl_rand_jump() are copied from the following CC-0 licensed 96 * file: 97 * 98 * http://xorshift.di.unimi.it/xorshift128plus.c 99 */ 100 101 static inline uint64_t 102 spl_rand_next(uint64_t *s) 103 { 104 uint64_t s1 = s[0]; 105 const uint64_t s0 = s[1]; 106 s[0] = s0; 107 s1 ^= s1 << 23; // a 108 s[1] = s1 ^ s0 ^ (s1 >> 18) ^ (s0 >> 5); // b, c 109 return (s[1] + s0); 110 } 111 112 static inline void 113 spl_rand_jump(uint64_t *s) 114 { 115 static const uint64_t JUMP[] = 116 { 0x8a5cd789635d2dff, 0x121fd2155c472f96 }; 117 118 uint64_t s0 = 0; 119 uint64_t s1 = 0; 120 int i, b; 121 for (i = 0; i < sizeof (JUMP) / sizeof (*JUMP); i++) 122 for (b = 0; b < 64; b++) { 123 if (JUMP[i] & 1ULL << b) { 124 s0 ^= s[0]; 125 s1 ^= s[1]; 126 } 127 (void) spl_rand_next(s); 128 } 129 130 s[0] = s0; 131 s[1] = s1; 132 } 133 134 int 135 random_get_pseudo_bytes(uint8_t *ptr, size_t len) 136 { 137 uint64_t *xp, s[2]; 138 139 ASSERT(ptr); 140 141 xp = get_cpu_ptr(spl_pseudo_entropy); 142 143 s[0] = xp[0]; 144 s[1] = xp[1]; 145 146 while (len) { 147 union { 148 uint64_t ui64; 149 uint8_t byte[sizeof (uint64_t)]; 150 }entropy; 151 int i = MIN(len, sizeof (uint64_t)); 152 153 len -= i; 154 entropy.ui64 = spl_rand_next(s); 155 156 while (i--) 157 *ptr++ = entropy.byte[i]; 158 } 159 160 xp[0] = s[0]; 161 xp[1] = s[1]; 162 163 put_cpu_ptr(spl_pseudo_entropy); 164 165 return (0); 166 } 167 168 169 EXPORT_SYMBOL(random_get_pseudo_bytes); 170 171 #if BITS_PER_LONG == 32 172 173 /* 174 * Support 64/64 => 64 division on a 32-bit platform. While the kernel 175 * provides a div64_u64() function for this we do not use it because the 176 * implementation is flawed. There are cases which return incorrect 177 * results as late as linux-2.6.35. Until this is fixed upstream the 178 * spl must provide its own implementation. 179 * 180 * This implementation is a slightly modified version of the algorithm 181 * proposed by the book 'Hacker's Delight'. The original source can be 182 * found here and is available for use without restriction. 183 * 184 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c 185 */ 186 187 /* 188 * Calculate number of leading of zeros for a 64-bit value. 189 */ 190 static int 191 nlz64(uint64_t x) 192 { 193 register int n = 0; 194 195 if (x == 0) 196 return (64); 197 198 if (x <= 0x00000000FFFFFFFFULL) { n = n + 32; x = x << 32; } 199 if (x <= 0x0000FFFFFFFFFFFFULL) { n = n + 16; x = x << 16; } 200 if (x <= 0x00FFFFFFFFFFFFFFULL) { n = n + 8; x = x << 8; } 201 if (x <= 0x0FFFFFFFFFFFFFFFULL) { n = n + 4; x = x << 4; } 202 if (x <= 0x3FFFFFFFFFFFFFFFULL) { n = n + 2; x = x << 2; } 203 if (x <= 0x7FFFFFFFFFFFFFFFULL) { n = n + 1; } 204 205 return (n); 206 } 207 208 /* 209 * Newer kernels have a div_u64() function but we define our own 210 * to simplify portability between kernel versions. 211 */ 212 static inline uint64_t 213 __div_u64(uint64_t u, uint32_t v) 214 { 215 (void) do_div(u, v); 216 return (u); 217 } 218 219 /* 220 * Turn off missing prototypes warning for these functions. They are 221 * replacements for libgcc-provided functions and will never be called 222 * directly. 223 */ 224 #pragma GCC diagnostic push 225 #pragma GCC diagnostic ignored "-Wmissing-prototypes" 226 227 /* 228 * Implementation of 64-bit unsigned division for 32-bit machines. 229 * 230 * First the procedure takes care of the case in which the divisor is a 231 * 32-bit quantity. There are two subcases: (1) If the left half of the 232 * dividend is less than the divisor, one execution of do_div() is all that 233 * is required (overflow is not possible). (2) Otherwise it does two 234 * divisions, using the grade school method. 235 */ 236 uint64_t 237 __udivdi3(uint64_t u, uint64_t v) 238 { 239 uint64_t u0, u1, v1, q0, q1, k; 240 int n; 241 242 if (v >> 32 == 0) { // If v < 2**32: 243 if (u >> 32 < v) { // If u/v cannot overflow, 244 return (__div_u64(u, v)); // just do one division. 245 } else { // If u/v would overflow: 246 u1 = u >> 32; // Break u into two halves. 247 u0 = u & 0xFFFFFFFF; 248 q1 = __div_u64(u1, v); // First quotient digit. 249 k = u1 - q1 * v; // First remainder, < v. 250 u0 += (k << 32); 251 q0 = __div_u64(u0, v); // Seconds quotient digit. 252 return ((q1 << 32) + q0); 253 } 254 } else { // If v >= 2**32: 255 n = nlz64(v); // 0 <= n <= 31. 256 v1 = (v << n) >> 32; // Normalize divisor, MSB is 1. 257 u1 = u >> 1; // To ensure no overflow. 258 q1 = __div_u64(u1, v1); // Get quotient from 259 q0 = (q1 << n) >> 31; // Undo normalization and 260 // division of u by 2. 261 if (q0 != 0) // Make q0 correct or 262 q0 = q0 - 1; // too small by 1. 263 if ((u - q0 * v) >= v) 264 q0 = q0 + 1; // Now q0 is correct. 265 266 return (q0); 267 } 268 } 269 EXPORT_SYMBOL(__udivdi3); 270 271 #ifndef abs64 272 /* CSTYLED */ 273 #define abs64(x) ({ uint64_t t = (x) >> 63; ((x) ^ t) - t; }) 274 #endif 275 276 /* 277 * Implementation of 64-bit signed division for 32-bit machines. 278 */ 279 int64_t 280 __divdi3(int64_t u, int64_t v) 281 { 282 int64_t q, t; 283 q = __udivdi3(abs64(u), abs64(v)); 284 t = (u ^ v) >> 63; // If u, v have different 285 return ((q ^ t) - t); // signs, negate q. 286 } 287 EXPORT_SYMBOL(__divdi3); 288 289 /* 290 * Implementation of 64-bit unsigned modulo for 32-bit machines. 291 */ 292 uint64_t 293 __umoddi3(uint64_t dividend, uint64_t divisor) 294 { 295 return (dividend - (divisor * __udivdi3(dividend, divisor))); 296 } 297 EXPORT_SYMBOL(__umoddi3); 298 299 /* 64-bit signed modulo for 32-bit machines. */ 300 int64_t 301 __moddi3(int64_t n, int64_t d) 302 { 303 int64_t q; 304 boolean_t nn = B_FALSE; 305 306 if (n < 0) { 307 nn = B_TRUE; 308 n = -n; 309 } 310 if (d < 0) 311 d = -d; 312 313 q = __umoddi3(n, d); 314 315 return (nn ? -q : q); 316 } 317 EXPORT_SYMBOL(__moddi3); 318 319 /* 320 * Implementation of 64-bit unsigned division/modulo for 32-bit machines. 321 */ 322 uint64_t 323 __udivmoddi4(uint64_t n, uint64_t d, uint64_t *r) 324 { 325 uint64_t q = __udivdi3(n, d); 326 if (r) 327 *r = n - d * q; 328 return (q); 329 } 330 EXPORT_SYMBOL(__udivmoddi4); 331 332 /* 333 * Implementation of 64-bit signed division/modulo for 32-bit machines. 334 */ 335 int64_t 336 __divmoddi4(int64_t n, int64_t d, int64_t *r) 337 { 338 int64_t q, rr; 339 boolean_t nn = B_FALSE; 340 boolean_t nd = B_FALSE; 341 if (n < 0) { 342 nn = B_TRUE; 343 n = -n; 344 } 345 if (d < 0) { 346 nd = B_TRUE; 347 d = -d; 348 } 349 350 q = __udivmoddi4(n, d, (uint64_t *)&rr); 351 352 if (nn != nd) 353 q = -q; 354 if (nn) 355 rr = -rr; 356 if (r) 357 *r = rr; 358 return (q); 359 } 360 EXPORT_SYMBOL(__divmoddi4); 361 362 #if defined(__arm) || defined(__arm__) 363 /* 364 * Implementation of 64-bit (un)signed division for 32-bit arm machines. 365 * 366 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned) 367 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1}, 368 * and the remainder in {r2, r3}. The return type is specifically left 369 * set to 'void' to ensure the compiler does not overwrite these registers 370 * during the return. All results are in registers as per ABI 371 */ 372 void 373 __aeabi_uldivmod(uint64_t u, uint64_t v) 374 { 375 uint64_t res; 376 uint64_t mod; 377 378 res = __udivdi3(u, v); 379 mod = __umoddi3(u, v); 380 { 381 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF); 382 register uint32_t r1 asm("r1") = (res >> 32); 383 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF); 384 register uint32_t r3 asm("r3") = (mod >> 32); 385 386 asm volatile("" 387 : "+r"(r0), "+r"(r1), "+r"(r2), "+r"(r3) /* output */ 388 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */ 389 390 return; /* r0; */ 391 } 392 } 393 EXPORT_SYMBOL(__aeabi_uldivmod); 394 395 void 396 __aeabi_ldivmod(int64_t u, int64_t v) 397 { 398 int64_t res; 399 uint64_t mod; 400 401 res = __divdi3(u, v); 402 mod = __umoddi3(u, v); 403 { 404 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF); 405 register uint32_t r1 asm("r1") = (res >> 32); 406 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF); 407 register uint32_t r3 asm("r3") = (mod >> 32); 408 409 asm volatile("" 410 : "+r"(r0), "+r"(r1), "+r"(r2), "+r"(r3) /* output */ 411 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */ 412 413 return; /* r0; */ 414 } 415 } 416 EXPORT_SYMBOL(__aeabi_ldivmod); 417 #endif /* __arm || __arm__ */ 418 419 #pragma GCC diagnostic pop 420 421 #endif /* BITS_PER_LONG */ 422 423 /* 424 * NOTE: The strtoxx behavior is solely based on my reading of the Solaris 425 * ddi_strtol(9F) man page. I have not verified the behavior of these 426 * functions against their Solaris counterparts. It is possible that I 427 * may have misinterpreted the man page or the man page is incorrect. 428 */ 429 int ddi_strtol(const char *, char **, int, long *); 430 int ddi_strtoull(const char *, char **, int, unsigned long long *); 431 int ddi_strtoll(const char *, char **, int, long long *); 432 433 #define define_ddi_strtox(type, valtype) \ 434 int ddi_strto##type(const char *str, char **endptr, \ 435 int base, valtype *result) \ 436 { \ 437 valtype last_value, value = 0; \ 438 char *ptr = (char *)str; \ 439 int digit, minus = 0; \ 440 \ 441 while (strchr(" \t\n\r\f", *ptr)) \ 442 ++ptr; \ 443 \ 444 if (strlen(ptr) == 0) \ 445 return (EINVAL); \ 446 \ 447 switch (*ptr) { \ 448 case '-': \ 449 minus = 1; \ 450 zfs_fallthrough; \ 451 case '+': \ 452 ++ptr; \ 453 break; \ 454 } \ 455 \ 456 /* Auto-detect base based on prefix */ \ 457 if (!base) { \ 458 if (str[0] == '0') { \ 459 if (tolower(str[1]) == 'x' && isxdigit(str[2])) { \ 460 base = 16; /* hex */ \ 461 ptr += 2; \ 462 } else if (str[1] >= '0' && str[1] < 8) { \ 463 base = 8; /* octal */ \ 464 ptr += 1; \ 465 } else { \ 466 return (EINVAL); \ 467 } \ 468 } else { \ 469 base = 10; /* decimal */ \ 470 } \ 471 } \ 472 \ 473 while (1) { \ 474 if (isdigit(*ptr)) \ 475 digit = *ptr - '0'; \ 476 else if (isalpha(*ptr)) \ 477 digit = tolower(*ptr) - 'a' + 10; \ 478 else \ 479 break; \ 480 \ 481 if (digit >= base) \ 482 break; \ 483 \ 484 last_value = value; \ 485 value = value * base + digit; \ 486 if (last_value > value) /* Overflow */ \ 487 return (ERANGE); \ 488 \ 489 ptr++; \ 490 } \ 491 \ 492 *result = minus ? -value : value; \ 493 \ 494 if (endptr) \ 495 *endptr = ptr; \ 496 \ 497 return (0); \ 498 } \ 499 500 define_ddi_strtox(l, long) 501 define_ddi_strtox(ull, unsigned long long) 502 define_ddi_strtox(ll, long long) 503 504 EXPORT_SYMBOL(ddi_strtol); 505 EXPORT_SYMBOL(ddi_strtoll); 506 EXPORT_SYMBOL(ddi_strtoull); 507 508 int 509 ddi_copyin(const void *from, void *to, size_t len, int flags) 510 { 511 /* Fake ioctl() issued by kernel, 'from' is a kernel address */ 512 if (flags & FKIOCTL) { 513 memcpy(to, from, len); 514 return (0); 515 } 516 517 return (copyin(from, to, len)); 518 } 519 EXPORT_SYMBOL(ddi_copyin); 520 521 /* 522 * Post a uevent to userspace whenever a new vdev adds to the pool. It is 523 * necessary to sync blkid information with udev, which zed daemon uses 524 * during device hotplug to identify the vdev. 525 */ 526 void 527 spl_signal_kobj_evt(struct block_device *bdev) 528 { 529 #if defined(HAVE_BDEV_KOBJ) || defined(HAVE_PART_TO_DEV) 530 #ifdef HAVE_BDEV_KOBJ 531 struct kobject *disk_kobj = bdev_kobj(bdev); 532 #else 533 struct kobject *disk_kobj = &part_to_dev(bdev->bd_part)->kobj; 534 #endif 535 if (disk_kobj) { 536 int ret = kobject_uevent(disk_kobj, KOBJ_CHANGE); 537 if (ret) { 538 pr_warn("ZFS: Sending event '%d' to kobject: '%s'" 539 " (%p): failed(ret:%d)\n", KOBJ_CHANGE, 540 kobject_name(disk_kobj), disk_kobj, ret); 541 } 542 } 543 #else 544 /* 545 * This is encountered if neither bdev_kobj() nor part_to_dev() is available 546 * in the kernel - likely due to an API change that needs to be chased down. 547 */ 548 #error "Unsupported kernel: unable to get struct kobj from bdev" 549 #endif 550 } 551 EXPORT_SYMBOL(spl_signal_kobj_evt); 552 553 int 554 ddi_copyout(const void *from, void *to, size_t len, int flags) 555 { 556 /* Fake ioctl() issued by kernel, 'from' is a kernel address */ 557 if (flags & FKIOCTL) { 558 memcpy(to, from, len); 559 return (0); 560 } 561 562 return (copyout(from, to, len)); 563 } 564 EXPORT_SYMBOL(ddi_copyout); 565 566 static ssize_t 567 spl_kernel_read(struct file *file, void *buf, size_t count, loff_t *pos) 568 { 569 #if defined(HAVE_KERNEL_READ_PPOS) 570 return (kernel_read(file, buf, count, pos)); 571 #else 572 mm_segment_t saved_fs; 573 ssize_t ret; 574 575 saved_fs = get_fs(); 576 set_fs(KERNEL_DS); 577 578 ret = vfs_read(file, (void __user *)buf, count, pos); 579 580 set_fs(saved_fs); 581 582 return (ret); 583 #endif 584 } 585 586 static int 587 spl_getattr(struct file *filp, struct kstat *stat) 588 { 589 int rc; 590 591 ASSERT(filp); 592 ASSERT(stat); 593 594 #if defined(HAVE_4ARGS_VFS_GETATTR) 595 rc = vfs_getattr(&filp->f_path, stat, STATX_BASIC_STATS, 596 AT_STATX_SYNC_AS_STAT); 597 #elif defined(HAVE_2ARGS_VFS_GETATTR) 598 rc = vfs_getattr(&filp->f_path, stat); 599 #elif defined(HAVE_3ARGS_VFS_GETATTR) 600 rc = vfs_getattr(filp->f_path.mnt, filp->f_dentry, stat); 601 #else 602 #error "No available vfs_getattr()" 603 #endif 604 if (rc) 605 return (-rc); 606 607 return (0); 608 } 609 610 /* 611 * Read the unique system identifier from the /etc/hostid file. 612 * 613 * The behavior of /usr/bin/hostid on Linux systems with the 614 * regular eglibc and coreutils is: 615 * 616 * 1. Generate the value if the /etc/hostid file does not exist 617 * or if the /etc/hostid file is less than four bytes in size. 618 * 619 * 2. If the /etc/hostid file is at least 4 bytes, then return 620 * the first four bytes [0..3] in native endian order. 621 * 622 * 3. Always ignore bytes [4..] if they exist in the file. 623 * 624 * Only the first four bytes are significant, even on systems that 625 * have a 64-bit word size. 626 * 627 * See: 628 * 629 * eglibc: sysdeps/unix/sysv/linux/gethostid.c 630 * coreutils: src/hostid.c 631 * 632 * Notes: 633 * 634 * The /etc/hostid file on Solaris is a text file that often reads: 635 * 636 * # DO NOT EDIT 637 * "0123456789" 638 * 639 * Directly copying this file to Linux results in a constant 640 * hostid of 4f442023 because the default comment constitutes 641 * the first four bytes of the file. 642 * 643 */ 644 645 static char *spl_hostid_path = HW_HOSTID_PATH; 646 module_param(spl_hostid_path, charp, 0444); 647 MODULE_PARM_DESC(spl_hostid_path, "The system hostid file (/etc/hostid)"); 648 649 static int 650 hostid_read(uint32_t *hostid) 651 { 652 uint64_t size; 653 uint32_t value = 0; 654 int error; 655 loff_t off; 656 struct file *filp; 657 struct kstat stat; 658 659 filp = filp_open(spl_hostid_path, 0, 0); 660 661 if (IS_ERR(filp)) 662 return (ENOENT); 663 664 error = spl_getattr(filp, &stat); 665 if (error) { 666 filp_close(filp, 0); 667 return (error); 668 } 669 size = stat.size; 670 // cppcheck-suppress sizeofwithnumericparameter 671 if (size < sizeof (HW_HOSTID_MASK)) { 672 filp_close(filp, 0); 673 return (EINVAL); 674 } 675 676 off = 0; 677 /* 678 * Read directly into the variable like eglibc does. 679 * Short reads are okay; native behavior is preserved. 680 */ 681 error = spl_kernel_read(filp, &value, sizeof (value), &off); 682 if (error < 0) { 683 filp_close(filp, 0); 684 return (EIO); 685 } 686 687 /* Mask down to 32 bits like coreutils does. */ 688 *hostid = (value & HW_HOSTID_MASK); 689 filp_close(filp, 0); 690 691 return (0); 692 } 693 694 /* 695 * Return the system hostid. Preferentially use the spl_hostid module option 696 * when set, otherwise use the value in the /etc/hostid file. 697 */ 698 uint32_t 699 zone_get_hostid(void *zone) 700 { 701 uint32_t hostid; 702 703 ASSERT3P(zone, ==, NULL); 704 705 if (spl_hostid != 0) 706 return ((uint32_t)(spl_hostid & HW_HOSTID_MASK)); 707 708 if (hostid_read(&hostid) == 0) 709 return (hostid); 710 711 return (0); 712 } 713 EXPORT_SYMBOL(zone_get_hostid); 714 715 static int 716 spl_kvmem_init(void) 717 { 718 int rc = 0; 719 720 rc = spl_kmem_init(); 721 if (rc) 722 return (rc); 723 724 rc = spl_vmem_init(); 725 if (rc) { 726 spl_kmem_fini(); 727 return (rc); 728 } 729 730 return (rc); 731 } 732 733 /* 734 * We initialize the random number generator with 128 bits of entropy from the 735 * system random number generator. In the improbable case that we have a zero 736 * seed, we fallback to the system jiffies, unless it is also zero, in which 737 * situation we use a preprogrammed seed. We step forward by 2^64 iterations to 738 * initialize each of the per-cpu seeds so that the sequences generated on each 739 * CPU are guaranteed to never overlap in practice. 740 */ 741 static int __init 742 spl_random_init(void) 743 { 744 uint64_t s[2]; 745 int i = 0; 746 747 spl_pseudo_entropy = __alloc_percpu(2 * sizeof (uint64_t), 748 sizeof (uint64_t)); 749 750 if (!spl_pseudo_entropy) 751 return (-ENOMEM); 752 753 get_random_bytes(s, sizeof (s)); 754 755 if (s[0] == 0 && s[1] == 0) { 756 if (jiffies != 0) { 757 s[0] = jiffies; 758 s[1] = ~0 - jiffies; 759 } else { 760 (void) memcpy(s, "improbable seed", sizeof (s)); 761 } 762 printk("SPL: get_random_bytes() returned 0 " 763 "when generating random seed. Setting initial seed to " 764 "0x%016llx%016llx.\n", cpu_to_be64(s[0]), 765 cpu_to_be64(s[1])); 766 } 767 768 for_each_possible_cpu(i) { 769 uint64_t *wordp = per_cpu_ptr(spl_pseudo_entropy, i); 770 771 spl_rand_jump(s); 772 773 wordp[0] = s[0]; 774 wordp[1] = s[1]; 775 } 776 777 return (0); 778 } 779 780 static void 781 spl_random_fini(void) 782 { 783 free_percpu(spl_pseudo_entropy); 784 } 785 786 static void 787 spl_kvmem_fini(void) 788 { 789 spl_vmem_fini(); 790 spl_kmem_fini(); 791 } 792 793 static int __init 794 spl_init(void) 795 { 796 int rc = 0; 797 798 if ((rc = spl_random_init())) 799 goto out0; 800 801 if ((rc = spl_kvmem_init())) 802 goto out1; 803 804 if ((rc = spl_tsd_init())) 805 goto out2; 806 807 if ((rc = spl_taskq_init())) 808 goto out3; 809 810 if ((rc = spl_kmem_cache_init())) 811 goto out4; 812 813 if ((rc = spl_proc_init())) 814 goto out5; 815 816 if ((rc = spl_kstat_init())) 817 goto out6; 818 819 if ((rc = spl_zlib_init())) 820 goto out7; 821 822 if ((rc = spl_zone_init())) 823 goto out8; 824 825 return (rc); 826 827 out8: 828 spl_zlib_fini(); 829 out7: 830 spl_kstat_fini(); 831 out6: 832 spl_proc_fini(); 833 out5: 834 spl_kmem_cache_fini(); 835 out4: 836 spl_taskq_fini(); 837 out3: 838 spl_tsd_fini(); 839 out2: 840 spl_kvmem_fini(); 841 out1: 842 spl_random_fini(); 843 out0: 844 return (rc); 845 } 846 847 static void __exit 848 spl_fini(void) 849 { 850 spl_zone_fini(); 851 spl_zlib_fini(); 852 spl_kstat_fini(); 853 spl_proc_fini(); 854 spl_kmem_cache_fini(); 855 spl_taskq_fini(); 856 spl_tsd_fini(); 857 spl_kvmem_fini(); 858 spl_random_fini(); 859 } 860 861 module_init(spl_init); 862 module_exit(spl_fini); 863 864 MODULE_DESCRIPTION("Solaris Porting Layer"); 865 MODULE_AUTHOR(ZFS_META_AUTHOR); 866 MODULE_LICENSE("GPL"); 867 MODULE_VERSION(ZFS_META_VERSION "-" ZFS_META_RELEASE); 868