]>
git.proxmox.com Git - mirror_spl.git/blob - module/spl/spl-generic.c
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 <linux/kmod.h>
47 #include <linux/math64_compat.h>
48 #include <linux/proc_compat.h>
50 char spl_version
[32] = "SPL v" SPL_META_VERSION
"-" SPL_META_RELEASE
;
51 EXPORT_SYMBOL(spl_version
);
53 unsigned long spl_hostid
= 0;
54 EXPORT_SYMBOL(spl_hostid
);
55 module_param(spl_hostid
, ulong
, 0644);
56 MODULE_PARM_DESC(spl_hostid
, "The system hostid.");
62 * Xorshift Pseudo Random Number Generator based on work by Sebastiano Vigna
64 * "Further scramblings of Marsaglia's xorshift generators"
65 * http://vigna.di.unimi.it/ftp/papers/xorshiftplus.pdf
67 * random_get_pseudo_bytes() is an API function on Illumos whose sole purpose
68 * is to provide bytes containing random numbers. It is mapped to /dev/urandom
69 * on Illumos, which uses a "FIPS 186-2 algorithm". No user of the SPL's
70 * random_get_pseudo_bytes() needs bytes that are of cryptographic quality, so
71 * we can implement it using a fast PRNG that we seed using Linux' actual
72 * equivalent to random_get_pseudo_bytes(). We do this by providing each CPU
73 * with an independent seed so that all calls to random_get_pseudo_bytes() are
74 * free of atomic instructions.
76 * A consequence of using a fast PRNG is that using random_get_pseudo_bytes()
77 * to generate words larger than 128 bits will paradoxically be limited to
78 * `2^128 - 1` possibilities. This is because we have a sequence of `2^128 - 1`
79 * 128-bit words and selecting the first will implicitly select the second. If
80 * a caller finds this behavior undesireable, random_get_bytes() should be used
83 * XXX: Linux interrupt handlers that trigger within the critical section
84 * formed by `s[1] = xp[1];` and `xp[0] = s[0];` and call this function will
85 * see the same numbers. Nothing in the code currently calls this in an
86 * interrupt handler, so this is considered to be okay. If that becomes a
87 * problem, we could create a set of per-cpu variables for interrupt handlers
88 * and use them when in_interrupt() from linux/preempt_mask.h evaluates to
91 static DEFINE_PER_CPU(uint64_t[2], spl_pseudo_entropy
);
94 * spl_rand_next()/spl_rand_jump() are copied from the following CC-0 licensed
97 * http://xorshift.di.unimi.it/xorshift128plus.c
100 static inline uint64_t
101 spl_rand_next(uint64_t *s
) {
103 const uint64_t s0
= s
[1];
106 s
[1] = s1
^ s0
^ (s1
>> 18) ^ (s0
>> 5); // b, c
111 spl_rand_jump(uint64_t *s
) {
112 static const uint64_t JUMP
[] =
113 { 0x8a5cd789635d2dff, 0x121fd2155c472f96 };
118 for (i
= 0; i
< sizeof (JUMP
) / sizeof (*JUMP
); i
++)
119 for (b
= 0; b
< 64; b
++) {
120 if (JUMP
[i
] & 1ULL << b
) {
124 (void) spl_rand_next(s
);
132 random_get_pseudo_bytes(uint8_t *ptr
, size_t len
)
138 xp
= get_cpu_var(spl_pseudo_entropy
);
146 uint8_t byte
[sizeof (uint64_t)];
148 int i
= MIN(len
, sizeof (uint64_t));
151 entropy
.ui64
= spl_rand_next(s
);
154 *ptr
++ = entropy
.byte
[i
];
160 put_cpu_var(spl_pseudo_entropy
);
166 EXPORT_SYMBOL(random_get_pseudo_bytes
);
168 #if BITS_PER_LONG == 32
170 * Support 64/64 => 64 division on a 32-bit platform. While the kernel
171 * provides a div64_u64() function for this we do not use it because the
172 * implementation is flawed. There are cases which return incorrect
173 * results as late as linux-2.6.35. Until this is fixed upstream the
174 * spl must provide its own implementation.
176 * This implementation is a slightly modified version of the algorithm
177 * proposed by the book 'Hacker's Delight'. The original source can be
178 * found here and is available for use without restriction.
180 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
184 * Calculate number of leading of zeros for a 64-bit value.
193 if (x
<= 0x00000000FFFFFFFFULL
) { n
= n
+ 32; x
= x
<< 32; }
194 if (x
<= 0x0000FFFFFFFFFFFFULL
) { n
= n
+ 16; x
= x
<< 16; }
195 if (x
<= 0x00FFFFFFFFFFFFFFULL
) { n
= n
+ 8; x
= x
<< 8; }
196 if (x
<= 0x0FFFFFFFFFFFFFFFULL
) { n
= n
+ 4; x
= x
<< 4; }
197 if (x
<= 0x3FFFFFFFFFFFFFFFULL
) { n
= n
+ 2; x
= x
<< 2; }
198 if (x
<= 0x7FFFFFFFFFFFFFFFULL
) { n
= n
+ 1; }
204 * Newer kernels have a div_u64() function but we define our own
205 * to simplify portibility between kernel versions.
207 static inline uint64_t
208 __div_u64(uint64_t u
, uint32_t v
)
215 * Implementation of 64-bit unsigned division for 32-bit machines.
217 * First the procedure takes care of the case in which the divisor is a
218 * 32-bit quantity. There are two subcases: (1) If the left half of the
219 * dividend is less than the divisor, one execution of do_div() is all that
220 * is required (overflow is not possible). (2) Otherwise it does two
221 * divisions, using the grade school method.
224 __udivdi3(uint64_t u
, uint64_t v
)
226 uint64_t u0
, u1
, v1
, q0
, q1
, k
;
229 if (v
>> 32 == 0) { // If v < 2**32:
230 if (u
>> 32 < v
) { // If u/v cannot overflow,
231 return (__div_u64(u
, v
)); // just do one division.
232 } else { // If u/v would overflow:
233 u1
= u
>> 32; // Break u into two halves.
235 q1
= __div_u64(u1
, v
); // First quotient digit.
236 k
= u1
- q1
* v
; // First remainder, < v.
238 q0
= __div_u64(u0
, v
); // Seconds quotient digit.
239 return ((q1
<< 32) + q0
);
241 } else { // If v >= 2**32:
242 n
= nlz64(v
); // 0 <= n <= 31.
243 v1
= (v
<< n
) >> 32; // Normalize divisor, MSB is 1.
244 u1
= u
>> 1; // To ensure no overflow.
245 q1
= __div_u64(u1
, v1
); // Get quotient from
246 q0
= (q1
<< n
) >> 31; // Undo normalization and
247 // division of u by 2.
248 if (q0
!= 0) // Make q0 correct or
249 q0
= q0
- 1; // too small by 1.
250 if ((u
- q0
* v
) >= v
)
251 q0
= q0
+ 1; // Now q0 is correct.
256 EXPORT_SYMBOL(__udivdi3
);
259 * Implementation of 64-bit signed division for 32-bit machines.
262 __divdi3(int64_t u
, int64_t v
)
265 q
= __udivdi3(abs64(u
), abs64(v
));
266 t
= (u
^ v
) >> 63; // If u, v have different
267 return ((q
^ t
) - t
); // signs, negate q.
269 EXPORT_SYMBOL(__divdi3
);
272 * Implementation of 64-bit unsigned modulo for 32-bit machines.
275 __umoddi3(uint64_t dividend
, uint64_t divisor
)
277 return (dividend
- (divisor
* __udivdi3(dividend
, divisor
)));
279 EXPORT_SYMBOL(__umoddi3
);
282 * Implementation of 64-bit unsigned division/modulo for 32-bit machines.
285 __udivmoddi4(uint64_t n
, uint64_t d
, uint64_t *r
)
287 uint64_t q
= __udivdi3(n
, d
);
292 EXPORT_SYMBOL(__udivmoddi4
);
295 * Implementation of 64-bit signed division/modulo for 32-bit machines.
298 __divmoddi4(int64_t n
, int64_t d
, int64_t *r
)
301 boolean_t nn
= B_FALSE
;
302 boolean_t nd
= B_FALSE
;
312 q
= __udivmoddi4(n
, d
, (uint64_t *)&rr
);
322 EXPORT_SYMBOL(__divmoddi4
);
324 #if defined(__arm) || defined(__arm__)
326 * Implementation of 64-bit (un)signed division for 32-bit arm machines.
328 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned)
329 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
330 * and the remainder in {r2, r3}. The return type is specifically left
331 * set to 'void' to ensure the compiler does not overwrite these registers
332 * during the return. All results are in registers as per ABI
335 __aeabi_uldivmod(uint64_t u
, uint64_t v
)
340 res
= __udivdi3(u
, v
);
341 mod
= __umoddi3(u
, v
);
343 register uint32_t r0
asm("r0") = (res
& 0xFFFFFFFF);
344 register uint32_t r1
asm("r1") = (res
>> 32);
345 register uint32_t r2
asm("r2") = (mod
& 0xFFFFFFFF);
346 register uint32_t r3
asm("r3") = (mod
>> 32);
350 : "+r"(r0
), "+r"(r1
), "+r"(r2
),"+r"(r3
) /* output */
351 : "r"(r0
), "r"(r1
), "r"(r2
), "r"(r3
)); /* input */
357 EXPORT_SYMBOL(__aeabi_uldivmod
);
360 __aeabi_ldivmod(int64_t u
, int64_t v
)
365 res
= __divdi3(u
, v
);
366 mod
= __umoddi3(u
, v
);
368 register uint32_t r0
asm("r0") = (res
& 0xFFFFFFFF);
369 register uint32_t r1
asm("r1") = (res
>> 32);
370 register uint32_t r2
asm("r2") = (mod
& 0xFFFFFFFF);
371 register uint32_t r3
asm("r3") = (mod
>> 32);
375 : "+r"(r0
), "+r"(r1
), "+r"(r2
),"+r"(r3
) /* output */
376 : "r"(r0
), "r"(r1
), "r"(r2
), "r"(r3
)); /* input */
382 EXPORT_SYMBOL(__aeabi_ldivmod
);
383 #endif /* __arm || __arm__ */
384 #endif /* BITS_PER_LONG */
387 * NOTE: The strtoxx behavior is solely based on my reading of the Solaris
388 * ddi_strtol(9F) man page. I have not verified the behavior of these
389 * functions against their Solaris counterparts. It is possible that I
390 * may have misinterpreted the man page or the man page is incorrect.
392 int ddi_strtoul(const char *, char **, int, unsigned long *);
393 int ddi_strtol(const char *, char **, int, long *);
394 int ddi_strtoull(const char *, char **, int, unsigned long long *);
395 int ddi_strtoll(const char *, char **, int, long long *);
397 #define define_ddi_strtoux(type, valtype) \
398 int ddi_strtou##type(const char *str, char **endptr, \
399 int base, valtype *result) \
401 valtype last_value, value = 0; \
402 char *ptr = (char *)str; \
403 int flag = 1, digit; \
405 if (strlen(ptr) == 0) \
408 /* Auto-detect base based on prefix */ \
410 if (str[0] == '0') { \
411 if (tolower(str[1]) == 'x' && isxdigit(str[2])) { \
412 base = 16; /* hex */ \
414 } else if (str[1] >= '0' && str[1] < 8) { \
415 base = 8; /* octal */ \
421 base = 10; /* decimal */ \
427 digit = *ptr - '0'; \
428 else if (isalpha(*ptr)) \
429 digit = tolower(*ptr) - 'a' + 10; \
436 last_value = value; \
437 value = value * base + digit; \
438 if (last_value > value) /* Overflow */ \
449 *endptr = (char *)(flag ? ptr : str); \
454 #define define_ddi_strtox(type, valtype) \
455 int ddi_strto##type(const char *str, char **endptr, \
456 int base, valtype *result) \
461 rc = ddi_strtou##type(str + 1, endptr, base, result); \
463 if (*endptr == str + 1) \
464 *endptr = (char *)str; \
466 *result = -*result; \
469 rc = ddi_strtou##type(str, endptr, base, result); \
475 define_ddi_strtoux(l
, unsigned long)
476 define_ddi_strtox(l
, long)
477 define_ddi_strtoux(ll
, unsigned long long)
478 define_ddi_strtox(ll
, long long)
480 EXPORT_SYMBOL(ddi_strtoul
);
481 EXPORT_SYMBOL(ddi_strtol
);
482 EXPORT_SYMBOL(ddi_strtoll
);
483 EXPORT_SYMBOL(ddi_strtoull
);
486 ddi_copyin(const void *from
, void *to
, size_t len
, int flags
)
488 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
489 if (flags
& FKIOCTL
) {
490 memcpy(to
, from
, len
);
494 return (copyin(from
, to
, len
));
496 EXPORT_SYMBOL(ddi_copyin
);
499 ddi_copyout(const void *from
, void *to
, size_t len
, int flags
)
501 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
502 if (flags
& FKIOCTL
) {
503 memcpy(to
, from
, len
);
507 return (copyout(from
, to
, len
));
509 EXPORT_SYMBOL(ddi_copyout
);
512 * Read the unique system identifier from the /etc/hostid file.
514 * The behavior of /usr/bin/hostid on Linux systems with the
515 * regular eglibc and coreutils is:
517 * 1. Generate the value if the /etc/hostid file does not exist
518 * or if the /etc/hostid file is less than four bytes in size.
520 * 2. If the /etc/hostid file is at least 4 bytes, then return
521 * the first four bytes [0..3] in native endian order.
523 * 3. Always ignore bytes [4..] if they exist in the file.
525 * Only the first four bytes are significant, even on systems that
526 * have a 64-bit word size.
530 * eglibc: sysdeps/unix/sysv/linux/gethostid.c
531 * coreutils: src/hostid.c
535 * The /etc/hostid file on Solaris is a text file that often reads:
540 * Directly copying this file to Linux results in a constant
541 * hostid of 4f442023 because the default comment constitutes
542 * the first four bytes of the file.
546 char *spl_hostid_path
= HW_HOSTID_PATH
;
547 module_param(spl_hostid_path
, charp
, 0444);
548 MODULE_PARM_DESC(spl_hostid_path
, "The system hostid file (/etc/hostid)");
551 hostid_read(uint32_t *hostid
)
558 file
= kobj_open_file(spl_hostid_path
);
559 if (file
== (struct _buf
*)-1)
562 error
= kobj_get_filesize(file
, &size
);
564 kobj_close_file(file
);
568 if (size
< sizeof (HW_HOSTID_MASK
)) {
569 kobj_close_file(file
);
574 * Read directly into the variable like eglibc does.
575 * Short reads are okay; native behavior is preserved.
577 error
= kobj_read_file(file
, (char *)&value
, sizeof (value
), 0);
579 kobj_close_file(file
);
583 /* Mask down to 32 bits like coreutils does. */
584 *hostid
= (value
& HW_HOSTID_MASK
);
585 kobj_close_file(file
);
591 * Return the system hostid. Preferentially use the spl_hostid module option
592 * when set, otherwise use the value in the /etc/hostid file.
595 zone_get_hostid(void *zone
)
599 ASSERT3P(zone
, ==, NULL
);
602 return ((uint32_t)(spl_hostid
& HW_HOSTID_MASK
));
604 if (hostid_read(&hostid
) == 0)
609 EXPORT_SYMBOL(zone_get_hostid
);
616 rc
= spl_kmem_init();
620 rc
= spl_vmem_init();
630 * We initialize the random number generator with 128 bits of entropy from the
631 * system random number generator. In the improbable case that we have a zero
632 * seed, we fallback to the system jiffies, unless it is also zero, in which
633 * situation we use a preprogrammed seed. We step forward by 2^64 iterations to
634 * initialize each of the per-cpu seeds so that the sequences generated on each
635 * CPU are guaranteed to never overlap in practice.
638 spl_random_init(void)
643 get_random_bytes(s
, sizeof (s
));
645 if (s
[0] == 0 && s
[1] == 0) {
650 (void) memcpy(s
, "improbable seed", sizeof (s
));
652 printk("SPL: get_random_bytes() returned 0 "
653 "when generating random seed. Setting initial seed to "
654 "0x%016llx%016llx.", cpu_to_be64(s
[0]), cpu_to_be64(s
[1]));
657 for_each_possible_cpu(i
) {
658 uint64_t *wordp
= per_cpu(spl_pseudo_entropy
, i
);
679 bzero(&p0
, sizeof (proc_t
));
682 if ((rc
= spl_kvmem_init()))
685 if ((rc
= spl_mutex_init()))
688 if ((rc
= spl_rw_init()))
691 if ((rc
= spl_tsd_init()))
694 if ((rc
= spl_taskq_init()))
697 if ((rc
= spl_kmem_cache_init()))
700 if ((rc
= spl_vn_init()))
703 if ((rc
= spl_proc_init()))
706 if ((rc
= spl_kstat_init()))
709 if ((rc
= spl_zlib_init()))
712 printk(KERN_NOTICE
"SPL: Loaded module v%s-%s%s\n", SPL_META_VERSION
,
713 SPL_META_RELEASE
, SPL_DEBUG_STR
);
723 spl_kmem_cache_fini();
735 printk(KERN_NOTICE
"SPL: Failed to Load Solaris Porting Layer "
736 "v%s-%s%s, rc = %d\n", SPL_META_VERSION
, SPL_META_RELEASE
,
745 printk(KERN_NOTICE
"SPL: Unloaded module v%s-%s%s\n",
746 SPL_META_VERSION
, SPL_META_RELEASE
, SPL_DEBUG_STR
);
751 spl_kmem_cache_fini();
759 module_init(spl_init
);
760 module_exit(spl_fini
);
762 MODULE_DESCRIPTION("Solaris Porting Layer");
763 MODULE_AUTHOR(SPL_META_AUTHOR
);
764 MODULE_LICENSE(SPL_META_LICENSE
);
765 MODULE_VERSION(SPL_META_VERSION
"-" SPL_META_RELEASE
);