xref: /freebsd-src/sys/contrib/openzfs/module/os/linux/spl/spl-generic.c (revision 16d6b3b3da62aa5baaf3c66c8d4e6f8c8f70aeb7)
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  *  For details, see <http://zfsonlinux.org/>.
10  *
11  *  The SPL is free software; you can redistribute it and/or modify it
12  *  under the terms of the GNU General Public License as published by the
13  *  Free Software Foundation; either version 2 of the License, or (at your
14  *  option) any later version.
15  *
16  *  The SPL is distributed in the hope that it will be useful, but WITHOUT
17  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
18  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
19  *  for more details.
20  *
21  *  You should have received a copy of the GNU General Public License along
22  *  with the SPL.  If not, see <http://www.gnu.org/licenses/>.
23  *
24  *  Solaris Porting Layer (SPL) Generic Implementation.
25  */
26 
27 #include <sys/sysmacros.h>
28 #include <sys/systeminfo.h>
29 #include <sys/vmsystm.h>
30 #include <sys/kmem.h>
31 #include <sys/kmem_cache.h>
32 #include <sys/vmem.h>
33 #include <sys/mutex.h>
34 #include <sys/rwlock.h>
35 #include <sys/taskq.h>
36 #include <sys/tsd.h>
37 #include <sys/zmod.h>
38 #include <sys/debug.h>
39 #include <sys/proc.h>
40 #include <sys/kstat.h>
41 #include <sys/file.h>
42 #include <sys/sunddi.h>
43 #include <linux/ctype.h>
44 #include <sys/disp.h>
45 #include <sys/random.h>
46 #include <sys/strings.h>
47 #include <linux/kmod.h>
48 #include "zfs_gitrev.h"
49 #include <linux/mod_compat.h>
50 #include <sys/cred.h>
51 #include <sys/vnode.h>
52 
53 char spl_gitrev[64] = ZFS_META_GITREV;
54 
55 /* BEGIN CSTYLED */
56 unsigned long spl_hostid = 0;
57 EXPORT_SYMBOL(spl_hostid);
58 /* BEGIN CSTYLED */
59 module_param(spl_hostid, ulong, 0644);
60 MODULE_PARM_DESC(spl_hostid, "The system hostid.");
61 /* END CSTYLED */
62 
63 proc_t p0;
64 EXPORT_SYMBOL(p0);
65 
66 /*
67  * Xorshift Pseudo Random Number Generator based on work by Sebastiano Vigna
68  *
69  * "Further scramblings of Marsaglia's xorshift generators"
70  * http://vigna.di.unimi.it/ftp/papers/xorshiftplus.pdf
71  *
72  * random_get_pseudo_bytes() is an API function on Illumos whose sole purpose
73  * is to provide bytes containing random numbers. It is mapped to /dev/urandom
74  * on Illumos, which uses a "FIPS 186-2 algorithm". No user of the SPL's
75  * random_get_pseudo_bytes() needs bytes that are of cryptographic quality, so
76  * we can implement it using a fast PRNG that we seed using Linux' actual
77  * equivalent to random_get_pseudo_bytes(). We do this by providing each CPU
78  * with an independent seed so that all calls to random_get_pseudo_bytes() are
79  * free of atomic instructions.
80  *
81  * A consequence of using a fast PRNG is that using random_get_pseudo_bytes()
82  * to generate words larger than 128 bits will paradoxically be limited to
83  * `2^128 - 1` possibilities. This is because we have a sequence of `2^128 - 1`
84  * 128-bit words and selecting the first will implicitly select the second. If
85  * a caller finds this behavior undesirable, random_get_bytes() should be used
86  * instead.
87  *
88  * XXX: Linux interrupt handlers that trigger within the critical section
89  * formed by `s[1] = xp[1];` and `xp[0] = s[0];` and call this function will
90  * see the same numbers. Nothing in the code currently calls this in an
91  * interrupt handler, so this is considered to be okay. If that becomes a
92  * problem, we could create a set of per-cpu variables for interrupt handlers
93  * and use them when in_interrupt() from linux/preempt_mask.h evaluates to
94  * true.
95  */
96 void __percpu *spl_pseudo_entropy;
97 
98 /*
99  * spl_rand_next()/spl_rand_jump() are copied from the following CC-0 licensed
100  * file:
101  *
102  * http://xorshift.di.unimi.it/xorshift128plus.c
103  */
104 
105 static inline uint64_t
106 spl_rand_next(uint64_t *s)
107 {
108 	uint64_t s1 = s[0];
109 	const uint64_t s0 = s[1];
110 	s[0] = s0;
111 	s1 ^= s1 << 23; // a
112 	s[1] = s1 ^ s0 ^ (s1 >> 18) ^ (s0 >> 5); // b, c
113 	return (s[1] + s0);
114 }
115 
116 static inline void
117 spl_rand_jump(uint64_t *s)
118 {
119 	static const uint64_t JUMP[] =
120 	    { 0x8a5cd789635d2dff, 0x121fd2155c472f96 };
121 
122 	uint64_t s0 = 0;
123 	uint64_t s1 = 0;
124 	int i, b;
125 	for (i = 0; i < sizeof (JUMP) / sizeof (*JUMP); i++)
126 		for (b = 0; b < 64; b++) {
127 			if (JUMP[i] & 1ULL << b) {
128 				s0 ^= s[0];
129 				s1 ^= s[1];
130 			}
131 			(void) spl_rand_next(s);
132 		}
133 
134 	s[0] = s0;
135 	s[1] = s1;
136 }
137 
138 int
139 random_get_pseudo_bytes(uint8_t *ptr, size_t len)
140 {
141 	uint64_t *xp, s[2];
142 
143 	ASSERT(ptr);
144 
145 	xp = get_cpu_ptr(spl_pseudo_entropy);
146 
147 	s[0] = xp[0];
148 	s[1] = xp[1];
149 
150 	while (len) {
151 		union {
152 			uint64_t ui64;
153 			uint8_t byte[sizeof (uint64_t)];
154 		}entropy;
155 		int i = MIN(len, sizeof (uint64_t));
156 
157 		len -= i;
158 		entropy.ui64 = spl_rand_next(s);
159 
160 		while (i--)
161 			*ptr++ = entropy.byte[i];
162 	}
163 
164 	xp[0] = s[0];
165 	xp[1] = s[1];
166 
167 	put_cpu_ptr(spl_pseudo_entropy);
168 
169 	return (0);
170 }
171 
172 
173 EXPORT_SYMBOL(random_get_pseudo_bytes);
174 
175 #if BITS_PER_LONG == 32
176 
177 /*
178  * Support 64/64 => 64 division on a 32-bit platform.  While the kernel
179  * provides a div64_u64() function for this we do not use it because the
180  * implementation is flawed.  There are cases which return incorrect
181  * results as late as linux-2.6.35.  Until this is fixed upstream the
182  * spl must provide its own implementation.
183  *
184  * This implementation is a slightly modified version of the algorithm
185  * proposed by the book 'Hacker's Delight'.  The original source can be
186  * found here and is available for use without restriction.
187  *
188  * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
189  */
190 
191 /*
192  * Calculate number of leading of zeros for a 64-bit value.
193  */
194 static int
195 nlz64(uint64_t x)
196 {
197 	register int n = 0;
198 
199 	if (x == 0)
200 		return (64);
201 
202 	if (x <= 0x00000000FFFFFFFFULL) { n = n + 32; x = x << 32; }
203 	if (x <= 0x0000FFFFFFFFFFFFULL) { n = n + 16; x = x << 16; }
204 	if (x <= 0x00FFFFFFFFFFFFFFULL) { n = n +  8; x = x <<  8; }
205 	if (x <= 0x0FFFFFFFFFFFFFFFULL) { n = n +  4; x = x <<  4; }
206 	if (x <= 0x3FFFFFFFFFFFFFFFULL) { n = n +  2; x = x <<  2; }
207 	if (x <= 0x7FFFFFFFFFFFFFFFULL) { n = n +  1; }
208 
209 	return (n);
210 }
211 
212 /*
213  * Newer kernels have a div_u64() function but we define our own
214  * to simplify portability between kernel versions.
215  */
216 static inline uint64_t
217 __div_u64(uint64_t u, uint32_t v)
218 {
219 	(void) do_div(u, v);
220 	return (u);
221 }
222 
223 /*
224  * Turn off missing prototypes warning for these functions. They are
225  * replacements for libgcc-provided functions and will never be called
226  * directly.
227  */
228 #pragma GCC diagnostic push
229 #pragma GCC diagnostic ignored "-Wmissing-prototypes"
230 
231 /*
232  * Implementation of 64-bit unsigned division for 32-bit machines.
233  *
234  * First the procedure takes care of the case in which the divisor is a
235  * 32-bit quantity. There are two subcases: (1) If the left half of the
236  * dividend is less than the divisor, one execution of do_div() is all that
237  * is required (overflow is not possible). (2) Otherwise it does two
238  * divisions, using the grade school method.
239  */
240 uint64_t
241 __udivdi3(uint64_t u, uint64_t v)
242 {
243 	uint64_t u0, u1, v1, q0, q1, k;
244 	int n;
245 
246 	if (v >> 32 == 0) {			// If v < 2**32:
247 		if (u >> 32 < v) {		// If u/v cannot overflow,
248 			return (__div_u64(u, v)); // just do one division.
249 		} else {			// If u/v would overflow:
250 			u1 = u >> 32;		// Break u into two halves.
251 			u0 = u & 0xFFFFFFFF;
252 			q1 = __div_u64(u1, v);	// First quotient digit.
253 			k  = u1 - q1 * v;	// First remainder, < v.
254 			u0 += (k << 32);
255 			q0 = __div_u64(u0, v);	// Seconds quotient digit.
256 			return ((q1 << 32) + q0);
257 		}
258 	} else {				// If v >= 2**32:
259 		n = nlz64(v);			// 0 <= n <= 31.
260 		v1 = (v << n) >> 32;		// Normalize divisor, MSB is 1.
261 		u1 = u >> 1;			// To ensure no overflow.
262 		q1 = __div_u64(u1, v1);		// Get quotient from
263 		q0 = (q1 << n) >> 31;		// Undo normalization and
264 						// division of u by 2.
265 		if (q0 != 0)			// Make q0 correct or
266 			q0 = q0 - 1;		// too small by 1.
267 		if ((u - q0 * v) >= v)
268 			q0 = q0 + 1;		// Now q0 is correct.
269 
270 		return (q0);
271 	}
272 }
273 EXPORT_SYMBOL(__udivdi3);
274 
275 /* BEGIN CSTYLED */
276 #ifndef abs64
277 #define	abs64(x)	({ uint64_t t = (x) >> 63; ((x) ^ t) - t; })
278 #endif
279 /* END CSTYLED */
280 
281 /*
282  * Implementation of 64-bit signed division for 32-bit machines.
283  */
284 int64_t
285 __divdi3(int64_t u, int64_t v)
286 {
287 	int64_t q, t;
288 	// cppcheck-suppress shiftTooManyBitsSigned
289 	q = __udivdi3(abs64(u), abs64(v));
290 	// cppcheck-suppress shiftTooManyBitsSigned
291 	t = (u ^ v) >> 63;	// If u, v have different
292 	return ((q ^ t) - t);	// signs, negate q.
293 }
294 EXPORT_SYMBOL(__divdi3);
295 
296 /*
297  * Implementation of 64-bit unsigned modulo for 32-bit machines.
298  */
299 uint64_t
300 __umoddi3(uint64_t dividend, uint64_t divisor)
301 {
302 	return (dividend - (divisor * __udivdi3(dividend, divisor)));
303 }
304 EXPORT_SYMBOL(__umoddi3);
305 
306 /* 64-bit signed modulo for 32-bit machines. */
307 int64_t
308 __moddi3(int64_t n, int64_t d)
309 {
310 	int64_t q;
311 	boolean_t nn = B_FALSE;
312 
313 	if (n < 0) {
314 		nn = B_TRUE;
315 		n = -n;
316 	}
317 	if (d < 0)
318 		d = -d;
319 
320 	q = __umoddi3(n, d);
321 
322 	return (nn ? -q : q);
323 }
324 EXPORT_SYMBOL(__moddi3);
325 
326 /*
327  * Implementation of 64-bit unsigned division/modulo for 32-bit machines.
328  */
329 uint64_t
330 __udivmoddi4(uint64_t n, uint64_t d, uint64_t *r)
331 {
332 	uint64_t q = __udivdi3(n, d);
333 	if (r)
334 		*r = n - d * q;
335 	return (q);
336 }
337 EXPORT_SYMBOL(__udivmoddi4);
338 
339 /*
340  * Implementation of 64-bit signed division/modulo for 32-bit machines.
341  */
342 int64_t
343 __divmoddi4(int64_t n, int64_t d, int64_t *r)
344 {
345 	int64_t q, rr;
346 	boolean_t nn = B_FALSE;
347 	boolean_t nd = B_FALSE;
348 	if (n < 0) {
349 		nn = B_TRUE;
350 		n = -n;
351 	}
352 	if (d < 0) {
353 		nd = B_TRUE;
354 		d = -d;
355 	}
356 
357 	q = __udivmoddi4(n, d, (uint64_t *)&rr);
358 
359 	if (nn != nd)
360 		q = -q;
361 	if (nn)
362 		rr = -rr;
363 	if (r)
364 		*r = rr;
365 	return (q);
366 }
367 EXPORT_SYMBOL(__divmoddi4);
368 
369 #if defined(__arm) || defined(__arm__)
370 /*
371  * Implementation of 64-bit (un)signed division for 32-bit arm machines.
372  *
373  * Run-time ABI for the ARM Architecture (page 20).  A pair of (unsigned)
374  * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
375  * and the remainder in {r2, r3}.  The return type is specifically left
376  * set to 'void' to ensure the compiler does not overwrite these registers
377  * during the return.  All results are in registers as per ABI
378  */
379 void
380 __aeabi_uldivmod(uint64_t u, uint64_t v)
381 {
382 	uint64_t res;
383 	uint64_t mod;
384 
385 	res = __udivdi3(u, v);
386 	mod = __umoddi3(u, v);
387 	{
388 		register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
389 		register uint32_t r1 asm("r1") = (res >> 32);
390 		register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
391 		register uint32_t r3 asm("r3") = (mod >> 32);
392 
393 		/* BEGIN CSTYLED */
394 		asm volatile(""
395 		    : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3)  /* output */
396 		    : "r"(r0), "r"(r1), "r"(r2), "r"(r3));   /* input */
397 		/* END CSTYLED */
398 
399 		return; /* r0; */
400 	}
401 }
402 EXPORT_SYMBOL(__aeabi_uldivmod);
403 
404 void
405 __aeabi_ldivmod(int64_t u, int64_t v)
406 {
407 	int64_t res;
408 	uint64_t mod;
409 
410 	res =  __divdi3(u, v);
411 	mod = __umoddi3(u, v);
412 	{
413 		register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
414 		register uint32_t r1 asm("r1") = (res >> 32);
415 		register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
416 		register uint32_t r3 asm("r3") = (mod >> 32);
417 
418 		/* BEGIN CSTYLED */
419 		asm volatile(""
420 		    : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3)  /* output */
421 		    : "r"(r0), "r"(r1), "r"(r2), "r"(r3));   /* input */
422 		/* END CSTYLED */
423 
424 		return; /* r0; */
425 	}
426 }
427 EXPORT_SYMBOL(__aeabi_ldivmod);
428 #endif /* __arm || __arm__ */
429 
430 #pragma GCC diagnostic pop
431 
432 #endif /* BITS_PER_LONG */
433 
434 /*
435  * NOTE: The strtoxx behavior is solely based on my reading of the Solaris
436  * ddi_strtol(9F) man page.  I have not verified the behavior of these
437  * functions against their Solaris counterparts.  It is possible that I
438  * may have misinterpreted the man page or the man page is incorrect.
439  */
440 int ddi_strtoul(const char *, char **, int, unsigned long *);
441 int ddi_strtol(const char *, char **, int, long *);
442 int ddi_strtoull(const char *, char **, int, unsigned long long *);
443 int ddi_strtoll(const char *, char **, int, long long *);
444 
445 #define	define_ddi_strtoux(type, valtype)				\
446 int ddi_strtou##type(const char *str, char **endptr,			\
447     int base, valtype *result)						\
448 {									\
449 	valtype last_value, value = 0;					\
450 	char *ptr = (char *)str;					\
451 	int flag = 1, digit;						\
452 									\
453 	if (strlen(ptr) == 0)						\
454 		return (EINVAL);					\
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 		flag = 1;						\
490 		ptr++;							\
491 	}								\
492 									\
493 	if (flag)							\
494 		*result = value;					\
495 									\
496 	if (endptr)							\
497 		*endptr = (char *)(flag ? ptr : str);			\
498 									\
499 	return (0);							\
500 }									\
501 
502 #define	define_ddi_strtox(type, valtype)				\
503 int ddi_strto##type(const char *str, char **endptr,			\
504     int base, valtype *result)						\
505 {									\
506 	int rc;								\
507 									\
508 	if (*str == '-') {						\
509 		rc = ddi_strtou##type(str + 1, endptr, base, result);	\
510 		if (!rc) {						\
511 			if (*endptr == str + 1)				\
512 				*endptr = (char *)str;			\
513 			else						\
514 				*result = -*result;			\
515 		}							\
516 	} else {							\
517 		rc = ddi_strtou##type(str, endptr, base, result);	\
518 	}								\
519 									\
520 	return (rc);							\
521 }
522 
523 define_ddi_strtoux(l, unsigned long)
524 define_ddi_strtox(l, long)
525 define_ddi_strtoux(ll, unsigned long long)
526 define_ddi_strtox(ll, long long)
527 
528 EXPORT_SYMBOL(ddi_strtoul);
529 EXPORT_SYMBOL(ddi_strtol);
530 EXPORT_SYMBOL(ddi_strtoll);
531 EXPORT_SYMBOL(ddi_strtoull);
532 
533 int
534 ddi_copyin(const void *from, void *to, size_t len, int flags)
535 {
536 	/* Fake ioctl() issued by kernel, 'from' is a kernel address */
537 	if (flags & FKIOCTL) {
538 		memcpy(to, from, len);
539 		return (0);
540 	}
541 
542 	return (copyin(from, to, len));
543 }
544 EXPORT_SYMBOL(ddi_copyin);
545 
546 int
547 ddi_copyout(const void *from, void *to, size_t len, int flags)
548 {
549 	/* Fake ioctl() issued by kernel, 'from' is a kernel address */
550 	if (flags & FKIOCTL) {
551 		memcpy(to, from, len);
552 		return (0);
553 	}
554 
555 	return (copyout(from, to, len));
556 }
557 EXPORT_SYMBOL(ddi_copyout);
558 
559 static ssize_t
560 spl_kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
561 {
562 #if defined(HAVE_KERNEL_READ_PPOS)
563 	return (kernel_read(file, buf, count, pos));
564 #else
565 	mm_segment_t saved_fs;
566 	ssize_t ret;
567 
568 	saved_fs = get_fs();
569 	set_fs(KERNEL_DS);
570 
571 	ret = vfs_read(file, (void __user *)buf, count, pos);
572 
573 	set_fs(saved_fs);
574 
575 	return (ret);
576 #endif
577 }
578 
579 static int
580 spl_getattr(struct file *filp, struct kstat *stat)
581 {
582 	int rc;
583 
584 	ASSERT(filp);
585 	ASSERT(stat);
586 
587 #if defined(HAVE_4ARGS_VFS_GETATTR)
588 	rc = vfs_getattr(&filp->f_path, stat, STATX_BASIC_STATS,
589 	    AT_STATX_SYNC_AS_STAT);
590 #elif defined(HAVE_2ARGS_VFS_GETATTR)
591 	rc = vfs_getattr(&filp->f_path, stat);
592 #else
593 	rc = vfs_getattr(filp->f_path.mnt, filp->f_dentry, stat);
594 #endif
595 	if (rc)
596 		return (-rc);
597 
598 	return (0);
599 }
600 
601 /*
602  * Read the unique system identifier from the /etc/hostid file.
603  *
604  * The behavior of /usr/bin/hostid on Linux systems with the
605  * regular eglibc and coreutils is:
606  *
607  *   1. Generate the value if the /etc/hostid file does not exist
608  *      or if the /etc/hostid file is less than four bytes in size.
609  *
610  *   2. If the /etc/hostid file is at least 4 bytes, then return
611  *      the first four bytes [0..3] in native endian order.
612  *
613  *   3. Always ignore bytes [4..] if they exist in the file.
614  *
615  * Only the first four bytes are significant, even on systems that
616  * have a 64-bit word size.
617  *
618  * See:
619  *
620  *   eglibc: sysdeps/unix/sysv/linux/gethostid.c
621  *   coreutils: src/hostid.c
622  *
623  * Notes:
624  *
625  * The /etc/hostid file on Solaris is a text file that often reads:
626  *
627  *   # DO NOT EDIT
628  *   "0123456789"
629  *
630  * Directly copying this file to Linux results in a constant
631  * hostid of 4f442023 because the default comment constitutes
632  * the first four bytes of the file.
633  *
634  */
635 
636 char *spl_hostid_path = HW_HOSTID_PATH;
637 module_param(spl_hostid_path, charp, 0444);
638 MODULE_PARM_DESC(spl_hostid_path, "The system hostid file (/etc/hostid)");
639 
640 static int
641 hostid_read(uint32_t *hostid)
642 {
643 	uint64_t size;
644 	uint32_t value = 0;
645 	int error;
646 	loff_t off;
647 	struct file *filp;
648 	struct kstat stat;
649 
650 	filp = filp_open(spl_hostid_path, 0, 0);
651 
652 	if (IS_ERR(filp))
653 		return (ENOENT);
654 
655 	error = spl_getattr(filp, &stat);
656 	if (error) {
657 		filp_close(filp, 0);
658 		return (error);
659 	}
660 	size = stat.size;
661 	if (size < sizeof (HW_HOSTID_MASK)) {
662 		filp_close(filp, 0);
663 		return (EINVAL);
664 	}
665 
666 	off = 0;
667 	/*
668 	 * Read directly into the variable like eglibc does.
669 	 * Short reads are okay; native behavior is preserved.
670 	 */
671 	error = spl_kernel_read(filp, &value, sizeof (value), &off);
672 	if (error < 0) {
673 		filp_close(filp, 0);
674 		return (EIO);
675 	}
676 
677 	/* Mask down to 32 bits like coreutils does. */
678 	*hostid = (value & HW_HOSTID_MASK);
679 	filp_close(filp, 0);
680 
681 	return (0);
682 }
683 
684 /*
685  * Return the system hostid.  Preferentially use the spl_hostid module option
686  * when set, otherwise use the value in the /etc/hostid file.
687  */
688 uint32_t
689 zone_get_hostid(void *zone)
690 {
691 	uint32_t hostid;
692 
693 	ASSERT3P(zone, ==, NULL);
694 
695 	if (spl_hostid != 0)
696 		return ((uint32_t)(spl_hostid & HW_HOSTID_MASK));
697 
698 	if (hostid_read(&hostid) == 0)
699 		return (hostid);
700 
701 	return (0);
702 }
703 EXPORT_SYMBOL(zone_get_hostid);
704 
705 static int
706 spl_kvmem_init(void)
707 {
708 	int rc = 0;
709 
710 	rc = spl_kmem_init();
711 	if (rc)
712 		return (rc);
713 
714 	rc = spl_vmem_init();
715 	if (rc) {
716 		spl_kmem_fini();
717 		return (rc);
718 	}
719 
720 	return (rc);
721 }
722 
723 /*
724  * We initialize the random number generator with 128 bits of entropy from the
725  * system random number generator. In the improbable case that we have a zero
726  * seed, we fallback to the system jiffies, unless it is also zero, in which
727  * situation we use a preprogrammed seed. We step forward by 2^64 iterations to
728  * initialize each of the per-cpu seeds so that the sequences generated on each
729  * CPU are guaranteed to never overlap in practice.
730  */
731 static void __init
732 spl_random_init(void)
733 {
734 	uint64_t s[2];
735 	int i = 0;
736 
737 	spl_pseudo_entropy = __alloc_percpu(2 * sizeof (uint64_t),
738 	    sizeof (uint64_t));
739 
740 	get_random_bytes(s, sizeof (s));
741 
742 	if (s[0] == 0 && s[1] == 0) {
743 		if (jiffies != 0) {
744 			s[0] = jiffies;
745 			s[1] = ~0 - jiffies;
746 		} else {
747 			(void) memcpy(s, "improbable seed", sizeof (s));
748 		}
749 		printk("SPL: get_random_bytes() returned 0 "
750 		    "when generating random seed. Setting initial seed to "
751 		    "0x%016llx%016llx.\n", cpu_to_be64(s[0]),
752 		    cpu_to_be64(s[1]));
753 	}
754 
755 	for_each_possible_cpu(i) {
756 		uint64_t *wordp = per_cpu_ptr(spl_pseudo_entropy, i);
757 
758 		spl_rand_jump(s);
759 
760 		wordp[0] = s[0];
761 		wordp[1] = s[1];
762 	}
763 }
764 
765 static void
766 spl_random_fini(void)
767 {
768 	free_percpu(spl_pseudo_entropy);
769 }
770 
771 static void
772 spl_kvmem_fini(void)
773 {
774 	spl_vmem_fini();
775 	spl_kmem_fini();
776 }
777 
778 static int __init
779 spl_init(void)
780 {
781 	int rc = 0;
782 
783 	bzero(&p0, sizeof (proc_t));
784 	spl_random_init();
785 
786 	if ((rc = spl_kvmem_init()))
787 		goto out1;
788 
789 	if ((rc = spl_tsd_init()))
790 		goto out2;
791 
792 	if ((rc = spl_taskq_init()))
793 		goto out3;
794 
795 	if ((rc = spl_kmem_cache_init()))
796 		goto out4;
797 
798 	if ((rc = spl_proc_init()))
799 		goto out5;
800 
801 	if ((rc = spl_kstat_init()))
802 		goto out6;
803 
804 	if ((rc = spl_zlib_init()))
805 		goto out7;
806 
807 	return (rc);
808 
809 out7:
810 	spl_kstat_fini();
811 out6:
812 	spl_proc_fini();
813 out5:
814 	spl_kmem_cache_fini();
815 out4:
816 	spl_taskq_fini();
817 out3:
818 	spl_tsd_fini();
819 out2:
820 	spl_kvmem_fini();
821 out1:
822 	return (rc);
823 }
824 
825 static void __exit
826 spl_fini(void)
827 {
828 	spl_zlib_fini();
829 	spl_kstat_fini();
830 	spl_proc_fini();
831 	spl_kmem_cache_fini();
832 	spl_taskq_fini();
833 	spl_tsd_fini();
834 	spl_kvmem_fini();
835 	spl_random_fini();
836 }
837 
838 module_init(spl_init);
839 module_exit(spl_fini);
840 
841 ZFS_MODULE_DESCRIPTION("Solaris Porting Layer");
842 ZFS_MODULE_AUTHOR(ZFS_META_AUTHOR);
843 ZFS_MODULE_LICENSE("GPL");
844 ZFS_MODULE_VERSION(ZFS_META_VERSION "-" ZFS_META_RELEASE);
845