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>.
8 * This file is part of the SPL, Solaris Porting Layer.
9 * For details, see <http://zfsonlinux.org/>.
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.
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
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/>.
24 * Solaris Porting Layer (SPL) Generic Implementation.
27 #include <sys/sysmacros.h>
28 #include <sys/systeminfo.h>
29 #include <sys/vmsystm.h>
32 #include <sys/kmem_cache.h>
34 #include <sys/mutex.h>
35 #include <sys/rwlock.h>
36 #include <sys/taskq.h>
39 #include <sys/debug.h>
41 #include <sys/kstat.h>
43 #include <linux/ctype.h>
45 #include <sys/random.h>
46 #include <sys/strings.h>
47 #include <linux/kmod.h>
48 #include "zfs_gitrev.h"
50 char spl_gitrev[64] = ZFS_META_GITREV;
53 unsigned long spl_hostid = 0;
54 EXPORT_SYMBOL(spl_hostid);
56 module_param(spl_hostid, ulong, 0644);
57 MODULE_PARM_DESC(spl_hostid, "The system hostid.");
64 * Xorshift Pseudo Random Number Generator based on work by Sebastiano Vigna
66 * "Further scramblings of Marsaglia's xorshift generators"
67 * http://vigna.di.unimi.it/ftp/papers/xorshiftplus.pdf
69 * random_get_pseudo_bytes() is an API function on Illumos whose sole purpose
70 * is to provide bytes containing random numbers. It is mapped to /dev/urandom
71 * on Illumos, which uses a "FIPS 186-2 algorithm". No user of the SPL's
72 * random_get_pseudo_bytes() needs bytes that are of cryptographic quality, so
73 * we can implement it using a fast PRNG that we seed using Linux' actual
74 * equivalent to random_get_pseudo_bytes(). We do this by providing each CPU
75 * with an independent seed so that all calls to random_get_pseudo_bytes() are
76 * free of atomic instructions.
78 * A consequence of using a fast PRNG is that using random_get_pseudo_bytes()
79 * to generate words larger than 128 bits will paradoxically be limited to
80 * `2^128 - 1` possibilities. This is because we have a sequence of `2^128 - 1`
81 * 128-bit words and selecting the first will implicitly select the second. If
82 * a caller finds this behavior undesireable, random_get_bytes() should be used
85 * XXX: Linux interrupt handlers that trigger within the critical section
86 * formed by `s[1] = xp[1];` and `xp[0] = s[0];` and call this function will
87 * see the same numbers. Nothing in the code currently calls this in an
88 * interrupt handler, so this is considered to be okay. If that becomes a
89 * problem, we could create a set of per-cpu variables for interrupt handlers
90 * and use them when in_interrupt() from linux/preempt_mask.h evaluates to
93 static DEFINE_PER_CPU(uint64_t[2], spl_pseudo_entropy);
96 * spl_rand_next()/spl_rand_jump() are copied from the following CC-0 licensed
99 * http://xorshift.di.unimi.it/xorshift128plus.c
102 static inline uint64_t
103 spl_rand_next(uint64_t *s)
106 const uint64_t s0 = s[1];
109 s[1] = s1 ^ s0 ^ (s1 >> 18) ^ (s0 >> 5); // b, c
114 spl_rand_jump(uint64_t *s)
116 static const uint64_t JUMP[] =
117 { 0x8a5cd789635d2dff, 0x121fd2155c472f96 };
122 for (i = 0; i < sizeof (JUMP) / sizeof (*JUMP); i++)
123 for (b = 0; b < 64; b++) {
124 if (JUMP[i] & 1ULL << b) {
128 (void) spl_rand_next(s);
136 random_get_pseudo_bytes(uint8_t *ptr, size_t len)
142 xp = get_cpu_var(spl_pseudo_entropy);
150 uint8_t byte[sizeof (uint64_t)];
152 int i = MIN(len, sizeof (uint64_t));
155 entropy.ui64 = spl_rand_next(s);
158 *ptr++ = entropy.byte[i];
164 put_cpu_var(spl_pseudo_entropy);
170 EXPORT_SYMBOL(random_get_pseudo_bytes);
172 #if BITS_PER_LONG == 32
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.
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.
184 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
188 * Calculate number of leading of zeros for a 64-bit value.
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; }
209 * Newer kernels have a div_u64() function but we define our own
210 * to simplify portibility between kernel versions.
212 static inline uint64_t
213 __div_u64(uint64_t u, uint32_t v)
220 * Implementation of 64-bit unsigned division for 32-bit machines.
222 * First the procedure takes care of the case in which the divisor is a
223 * 32-bit quantity. There are two subcases: (1) If the left half of the
224 * dividend is less than the divisor, one execution of do_div() is all that
225 * is required (overflow is not possible). (2) Otherwise it does two
226 * divisions, using the grade school method.
229 __udivdi3(uint64_t u, uint64_t v)
231 uint64_t u0, u1, v1, q0, q1, k;
234 if (v >> 32 == 0) { // If v < 2**32:
235 if (u >> 32 < v) { // If u/v cannot overflow,
236 return (__div_u64(u, v)); // just do one division.
237 } else { // If u/v would overflow:
238 u1 = u >> 32; // Break u into two halves.
240 q1 = __div_u64(u1, v); // First quotient digit.
241 k = u1 - q1 * v; // First remainder, < v.
243 q0 = __div_u64(u0, v); // Seconds quotient digit.
244 return ((q1 << 32) + q0);
246 } else { // If v >= 2**32:
247 n = nlz64(v); // 0 <= n <= 31.
248 v1 = (v << n) >> 32; // Normalize divisor, MSB is 1.
249 u1 = u >> 1; // To ensure no overflow.
250 q1 = __div_u64(u1, v1); // Get quotient from
251 q0 = (q1 << n) >> 31; // Undo normalization and
252 // division of u by 2.
253 if (q0 != 0) // Make q0 correct or
254 q0 = q0 - 1; // too small by 1.
255 if ((u - q0 * v) >= v)
256 q0 = q0 + 1; // Now q0 is correct.
261 EXPORT_SYMBOL(__udivdi3);
265 #define abs64(x) ({ uint64_t t = (x) >> 63; ((x) ^ t) - t; })
270 * Implementation of 64-bit signed division for 32-bit machines.
273 __divdi3(int64_t u, int64_t v)
276 q = __udivdi3(abs64(u), abs64(v));
277 t = (u ^ v) >> 63; // If u, v have different
278 return ((q ^ t) - t); // signs, negate q.
280 EXPORT_SYMBOL(__divdi3);
283 * Implementation of 64-bit unsigned modulo for 32-bit machines.
286 __umoddi3(uint64_t dividend, uint64_t divisor)
288 return (dividend - (divisor * __udivdi3(dividend, divisor)));
290 EXPORT_SYMBOL(__umoddi3);
293 * Implementation of 64-bit unsigned division/modulo for 32-bit machines.
296 __udivmoddi4(uint64_t n, uint64_t d, uint64_t *r)
298 uint64_t q = __udivdi3(n, d);
303 EXPORT_SYMBOL(__udivmoddi4);
306 * Implementation of 64-bit signed division/modulo for 32-bit machines.
309 __divmoddi4(int64_t n, int64_t d, int64_t *r)
312 boolean_t nn = B_FALSE;
313 boolean_t nd = B_FALSE;
323 q = __udivmoddi4(n, d, (uint64_t *)&rr);
333 EXPORT_SYMBOL(__divmoddi4);
335 #if defined(__arm) || defined(__arm__)
337 * Implementation of 64-bit (un)signed division for 32-bit arm machines.
339 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned)
340 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
341 * and the remainder in {r2, r3}. The return type is specifically left
342 * set to 'void' to ensure the compiler does not overwrite these registers
343 * during the return. All results are in registers as per ABI
346 __aeabi_uldivmod(uint64_t u, uint64_t v)
351 res = __udivdi3(u, v);
352 mod = __umoddi3(u, v);
354 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
355 register uint32_t r1 asm("r1") = (res >> 32);
356 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
357 register uint32_t r3 asm("r3") = (mod >> 32);
361 : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3) /* output */
362 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
368 EXPORT_SYMBOL(__aeabi_uldivmod);
371 __aeabi_ldivmod(int64_t u, int64_t v)
376 res = __divdi3(u, v);
377 mod = __umoddi3(u, v);
379 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
380 register uint32_t r1 asm("r1") = (res >> 32);
381 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
382 register uint32_t r3 asm("r3") = (mod >> 32);
386 : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3) /* output */
387 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
393 EXPORT_SYMBOL(__aeabi_ldivmod);
394 #endif /* __arm || __arm__ */
395 #endif /* BITS_PER_LONG */
398 * NOTE: The strtoxx behavior is solely based on my reading of the Solaris
399 * ddi_strtol(9F) man page. I have not verified the behavior of these
400 * functions against their Solaris counterparts. It is possible that I
401 * may have misinterpreted the man page or the man page is incorrect.
403 int ddi_strtoul(const char *, char **, int, unsigned long *);
404 int ddi_strtol(const char *, char **, int, long *);
405 int ddi_strtoull(const char *, char **, int, unsigned long long *);
406 int ddi_strtoll(const char *, char **, int, long long *);
408 #define define_ddi_strtoux(type, valtype) \
409 int ddi_strtou##type(const char *str, char **endptr, \
410 int base, valtype *result) \
412 valtype last_value, value = 0; \
413 char *ptr = (char *)str; \
414 int flag = 1, digit; \
416 if (strlen(ptr) == 0) \
419 /* Auto-detect base based on prefix */ \
421 if (str[0] == '0') { \
422 if (tolower(str[1]) == 'x' && isxdigit(str[2])) { \
423 base = 16; /* hex */ \
425 } else if (str[1] >= '0' && str[1] < 8) { \
426 base = 8; /* octal */ \
432 base = 10; /* decimal */ \
438 digit = *ptr - '0'; \
439 else if (isalpha(*ptr)) \
440 digit = tolower(*ptr) - 'a' + 10; \
447 last_value = value; \
448 value = value * base + digit; \
449 if (last_value > value) /* Overflow */ \
460 *endptr = (char *)(flag ? ptr : str); \
465 #define define_ddi_strtox(type, valtype) \
466 int ddi_strto##type(const char *str, char **endptr, \
467 int base, valtype *result) \
472 rc = ddi_strtou##type(str + 1, endptr, base, result); \
474 if (*endptr == str + 1) \
475 *endptr = (char *)str; \
477 *result = -*result; \
480 rc = ddi_strtou##type(str, endptr, base, result); \
486 define_ddi_strtoux(l, unsigned long)
487 define_ddi_strtox(l, long)
488 define_ddi_strtoux(ll, unsigned long long)
489 define_ddi_strtox(ll, long long)
491 EXPORT_SYMBOL(ddi_strtoul);
492 EXPORT_SYMBOL(ddi_strtol);
493 EXPORT_SYMBOL(ddi_strtoll);
494 EXPORT_SYMBOL(ddi_strtoull);
497 ddi_copyin(const void *from, void *to, size_t len, int flags)
499 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
500 if (flags & FKIOCTL) {
501 memcpy(to, from, len);
505 return (copyin(from, to, len));
507 EXPORT_SYMBOL(ddi_copyin);
510 ddi_copyout(const void *from, void *to, size_t len, int flags)
512 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
513 if (flags & FKIOCTL) {
514 memcpy(to, from, len);
518 return (copyout(from, to, len));
520 EXPORT_SYMBOL(ddi_copyout);
523 * Read the unique system identifier from the /etc/hostid file.
525 * The behavior of /usr/bin/hostid on Linux systems with the
526 * regular eglibc and coreutils is:
528 * 1. Generate the value if the /etc/hostid file does not exist
529 * or if the /etc/hostid file is less than four bytes in size.
531 * 2. If the /etc/hostid file is at least 4 bytes, then return
532 * the first four bytes [0..3] in native endian order.
534 * 3. Always ignore bytes [4..] if they exist in the file.
536 * Only the first four bytes are significant, even on systems that
537 * have a 64-bit word size.
541 * eglibc: sysdeps/unix/sysv/linux/gethostid.c
542 * coreutils: src/hostid.c
546 * The /etc/hostid file on Solaris is a text file that often reads:
551 * Directly copying this file to Linux results in a constant
552 * hostid of 4f442023 because the default comment constitutes
553 * the first four bytes of the file.
557 char *spl_hostid_path = HW_HOSTID_PATH;
558 module_param(spl_hostid_path, charp, 0444);
559 MODULE_PARM_DESC(spl_hostid_path, "The system hostid file (/etc/hostid)");
562 hostid_read(uint32_t *hostid)
569 file = kobj_open_file(spl_hostid_path);
570 if (file == (struct _buf *)-1)
573 error = kobj_get_filesize(file, &size);
575 kobj_close_file(file);
579 if (size < sizeof (HW_HOSTID_MASK)) {
580 kobj_close_file(file);
585 * Read directly into the variable like eglibc does.
586 * Short reads are okay; native behavior is preserved.
588 error = kobj_read_file(file, (char *)&value, sizeof (value), 0);
590 kobj_close_file(file);
594 /* Mask down to 32 bits like coreutils does. */
595 *hostid = (value & HW_HOSTID_MASK);
596 kobj_close_file(file);
602 * Return the system hostid. Preferentially use the spl_hostid module option
603 * when set, otherwise use the value in the /etc/hostid file.
606 zone_get_hostid(void *zone)
610 ASSERT3P(zone, ==, NULL);
613 return ((uint32_t)(spl_hostid & HW_HOSTID_MASK));
615 if (hostid_read(&hostid) == 0)
620 EXPORT_SYMBOL(zone_get_hostid);
627 rc = spl_kmem_init();
631 rc = spl_vmem_init();
641 * We initialize the random number generator with 128 bits of entropy from the
642 * system random number generator. In the improbable case that we have a zero
643 * seed, we fallback to the system jiffies, unless it is also zero, in which
644 * situation we use a preprogrammed seed. We step forward by 2^64 iterations to
645 * initialize each of the per-cpu seeds so that the sequences generated on each
646 * CPU are guaranteed to never overlap in practice.
649 spl_random_init(void)
654 get_random_bytes(s, sizeof (s));
656 if (s[0] == 0 && s[1] == 0) {
661 (void) memcpy(s, "improbable seed", sizeof (s));
663 printk("SPL: get_random_bytes() returned 0 "
664 "when generating random seed. Setting initial seed to "
665 "0x%016llx%016llx.", cpu_to_be64(s[0]), cpu_to_be64(s[1]));
668 for_each_possible_cpu(i) {
669 uint64_t *wordp = per_cpu(spl_pseudo_entropy, i);
690 bzero(&p0, sizeof (proc_t));
693 if ((rc = spl_kvmem_init()))
696 if ((rc = spl_mutex_init()))
699 if ((rc = spl_rw_init()))
702 if ((rc = spl_tsd_init()))
705 if ((rc = spl_taskq_init()))
708 if ((rc = spl_kmem_cache_init()))
711 if ((rc = spl_vn_init()))
714 if ((rc = spl_proc_init()))
717 if ((rc = spl_kstat_init()))
720 if ((rc = spl_zlib_init()))
732 spl_kmem_cache_fini();
754 spl_kmem_cache_fini();
762 module_init(spl_init);
763 module_exit(spl_fini);
765 MODULE_DESCRIPTION("Solaris Porting Layer");
766 MODULE_AUTHOR(ZFS_META_AUTHOR);
767 MODULE_LICENSE("GPL");
768 MODULE_VERSION(ZFS_META_VERSION "-" ZFS_META_RELEASE);