]>
git.proxmox.com Git - mirror_spl.git/blob - module/spl/spl-generic.c
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>.
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/>.
23 *****************************************************************************
24 * Solaris Porting Layer (SPL) Generic Implementation.
25 \*****************************************************************************/
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
[] = { 0x8a5cd789635d2dff, 0x121fd2155c472f96 };
117 for(i
= 0; i
< sizeof JUMP
/ sizeof *JUMP
; i
++)
118 for(b
= 0; b
< 64; b
++) {
119 if (JUMP
[i
] & 1ULL << b
) {
123 (void) spl_rand_next(s
);
131 random_get_pseudo_bytes(uint8_t *ptr
, size_t len
)
137 xp
= get_cpu_var(spl_pseudo_entropy
);
145 uint8_t byte
[sizeof (uint64_t)];
147 int i
= MIN(len
, sizeof (uint64_t));
150 entropy
.ui64
= spl_rand_next(s
);
153 *ptr
++ = entropy
.byte
[i
];
159 put_cpu_var(spl_pseudo_entropy
);
165 EXPORT_SYMBOL(random_get_pseudo_bytes
);
167 #if BITS_PER_LONG == 32
169 * Support 64/64 => 64 division on a 32-bit platform. While the kernel
170 * provides a div64_u64() function for this we do not use it because the
171 * implementation is flawed. There are cases which return incorrect
172 * results as late as linux-2.6.35. Until this is fixed upstream the
173 * spl must provide its own implementation.
175 * This implementation is a slightly modified version of the algorithm
176 * proposed by the book 'Hacker's Delight'. The original source can be
177 * found here and is available for use without restriction.
179 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
183 * Calculate number of leading of zeros for a 64-bit value.
192 if (x
<= 0x00000000FFFFFFFFULL
) {n
= n
+ 32; x
= x
<< 32;}
193 if (x
<= 0x0000FFFFFFFFFFFFULL
) {n
= n
+ 16; x
= x
<< 16;}
194 if (x
<= 0x00FFFFFFFFFFFFFFULL
) {n
= n
+ 8; x
= x
<< 8;}
195 if (x
<= 0x0FFFFFFFFFFFFFFFULL
) {n
= n
+ 4; x
= x
<< 4;}
196 if (x
<= 0x3FFFFFFFFFFFFFFFULL
) {n
= n
+ 2; x
= x
<< 2;}
197 if (x
<= 0x7FFFFFFFFFFFFFFFULL
) {n
= n
+ 1;}
203 * Newer kernels have a div_u64() function but we define our own
204 * to simplify portibility between kernel versions.
206 static inline uint64_t
207 __div_u64(uint64_t u
, uint32_t v
)
214 * Implementation of 64-bit unsigned division for 32-bit machines.
216 * First the procedure takes care of the case in which the divisor is a
217 * 32-bit quantity. There are two subcases: (1) If the left half of the
218 * dividend is less than the divisor, one execution of do_div() is all that
219 * is required (overflow is not possible). (2) Otherwise it does two
220 * divisions, using the grade school method.
223 __udivdi3(uint64_t u
, uint64_t v
)
225 uint64_t u0
, u1
, v1
, q0
, q1
, k
;
228 if (v
>> 32 == 0) { // If v < 2**32:
229 if (u
>> 32 < v
) { // If u/v cannot overflow,
230 return __div_u64(u
, v
); // just do one division.
231 } else { // If u/v would overflow:
232 u1
= u
>> 32; // Break u into two halves.
234 q1
= __div_u64(u1
, v
); // First quotient digit.
235 k
= u1
- q1
* v
; // First remainder, < v.
237 q0
= __div_u64(u0
, v
); // Seconds quotient digit.
238 return (q1
<< 32) + q0
;
240 } else { // If v >= 2**32:
241 n
= nlz64(v
); // 0 <= n <= 31.
242 v1
= (v
<< n
) >> 32; // Normalize divisor, MSB is 1.
243 u1
= u
>> 1; // To ensure no overflow.
244 q1
= __div_u64(u1
, v1
); // Get quotient from
245 q0
= (q1
<< n
) >> 31; // Undo normalization and
246 // division of u by 2.
247 if (q0
!= 0) // Make q0 correct or
248 q0
= q0
- 1; // too small by 1.
249 if ((u
- q0
* v
) >= v
)
250 q0
= q0
+ 1; // Now q0 is correct.
255 EXPORT_SYMBOL(__udivdi3
);
258 * Implementation of 64-bit signed division for 32-bit machines.
261 __divdi3(int64_t u
, int64_t v
)
264 q
= __udivdi3(abs64(u
), abs64(v
));
265 t
= (u
^ v
) >> 63; // If u, v have different
266 return (q
^ t
) - t
; // signs, negate q.
268 EXPORT_SYMBOL(__divdi3
);
271 * Implementation of 64-bit unsigned modulo for 32-bit machines.
274 __umoddi3(uint64_t dividend
, uint64_t divisor
)
276 return (dividend
- (divisor
* __udivdi3(dividend
, divisor
)));
278 EXPORT_SYMBOL(__umoddi3
);
281 * Implementation of 64-bit unsigned division/modulo for 32-bit machines.
284 __udivmoddi4(uint64_t n
, uint64_t d
, uint64_t *r
)
286 uint64_t q
= __udivdi3(n
, d
);
291 EXPORT_SYMBOL(__udivmoddi4
);
294 * Implementation of 64-bit signed division/modulo for 32-bit machines.
297 __divmoddi4(int64_t n
, int64_t d
, int64_t *r
)
300 boolean_t nn
= B_FALSE
;
301 boolean_t nd
= B_FALSE
;
311 q
= __udivmoddi4(n
, d
, (uint64_t *)&rr
);
321 EXPORT_SYMBOL(__divmoddi4
);
323 #if defined(__arm) || defined(__arm__)
325 * Implementation of 64-bit (un)signed division for 32-bit arm machines.
327 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned)
328 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
329 * and the remainder in {r2, r3}. The return type is specifically left
330 * set to 'void' to ensure the compiler does not overwrite these registers
331 * during the return. All results are in registers as per ABI
334 __aeabi_uldivmod(uint64_t u
, uint64_t v
)
339 res
= __udivdi3(u
, v
);
340 mod
= __umoddi3(u
, v
);
342 register uint32_t r0
asm("r0") = (res
& 0xFFFFFFFF);
343 register uint32_t r1
asm("r1") = (res
>> 32);
344 register uint32_t r2
asm("r2") = (mod
& 0xFFFFFFFF);
345 register uint32_t r3
asm("r3") = (mod
>> 32);
348 : "+r"(r0
), "+r"(r1
), "+r"(r2
),"+r"(r3
) /* output */
349 : "r"(r0
), "r"(r1
), "r"(r2
), "r"(r3
)); /* input */
354 EXPORT_SYMBOL(__aeabi_uldivmod
);
357 __aeabi_ldivmod(int64_t u
, int64_t v
)
362 res
= __divdi3(u
, v
);
363 mod
= __umoddi3(u
, v
);
365 register uint32_t r0
asm("r0") = (res
& 0xFFFFFFFF);
366 register uint32_t r1
asm("r1") = (res
>> 32);
367 register uint32_t r2
asm("r2") = (mod
& 0xFFFFFFFF);
368 register uint32_t r3
asm("r3") = (mod
>> 32);
371 : "+r"(r0
), "+r"(r1
), "+r"(r2
),"+r"(r3
) /* output */
372 : "r"(r0
), "r"(r1
), "r"(r2
), "r"(r3
)); /* input */
377 EXPORT_SYMBOL(__aeabi_ldivmod
);
378 #endif /* __arm || __arm__ */
379 #endif /* BITS_PER_LONG */
381 /* NOTE: The strtoxx behavior is solely based on my reading of the Solaris
382 * ddi_strtol(9F) man page. I have not verified the behavior of these
383 * functions against their Solaris counterparts. It is possible that I
384 * may have misinterpreted the man page or the man page is incorrect.
386 int ddi_strtoul(const char *, char **, int, unsigned long *);
387 int ddi_strtol(const char *, char **, int, long *);
388 int ddi_strtoull(const char *, char **, int, unsigned long long *);
389 int ddi_strtoll(const char *, char **, int, long long *);
391 #define define_ddi_strtoux(type, valtype) \
392 int ddi_strtou##type(const char *str, char **endptr, \
393 int base, valtype *result) \
395 valtype last_value, value = 0; \
396 char *ptr = (char *)str; \
397 int flag = 1, digit; \
399 if (strlen(ptr) == 0) \
402 /* Auto-detect base based on prefix */ \
404 if (str[0] == '0') { \
405 if (tolower(str[1])=='x' && isxdigit(str[2])) { \
406 base = 16; /* hex */ \
408 } else if (str[1] >= '0' && str[1] < 8) { \
409 base = 8; /* octal */ \
415 base = 10; /* decimal */ \
421 digit = *ptr - '0'; \
422 else if (isalpha(*ptr)) \
423 digit = tolower(*ptr) - 'a' + 10; \
430 last_value = value; \
431 value = value * base + digit; \
432 if (last_value > value) /* Overflow */ \
443 *endptr = (char *)(flag ? ptr : str); \
448 #define define_ddi_strtox(type, valtype) \
449 int ddi_strto##type(const char *str, char **endptr, \
450 int base, valtype *result) \
455 rc = ddi_strtou##type(str + 1, endptr, base, result); \
457 if (*endptr == str + 1) \
458 *endptr = (char *)str; \
460 *result = -*result; \
463 rc = ddi_strtou##type(str, endptr, base, result); \
469 define_ddi_strtoux(l
, unsigned long)
470 define_ddi_strtox(l
, long)
471 define_ddi_strtoux(ll
, unsigned long long)
472 define_ddi_strtox(ll
, long long)
474 EXPORT_SYMBOL(ddi_strtoul
);
475 EXPORT_SYMBOL(ddi_strtol
);
476 EXPORT_SYMBOL(ddi_strtoll
);
477 EXPORT_SYMBOL(ddi_strtoull
);
480 ddi_copyin(const void *from
, void *to
, size_t len
, int flags
)
482 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
483 if (flags
& FKIOCTL
) {
484 memcpy(to
, from
, len
);
488 return copyin(from
, to
, len
);
490 EXPORT_SYMBOL(ddi_copyin
);
493 ddi_copyout(const void *from
, void *to
, size_t len
, int flags
)
495 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
496 if (flags
& FKIOCTL
) {
497 memcpy(to
, from
, len
);
501 return copyout(from
, to
, len
);
503 EXPORT_SYMBOL(ddi_copyout
);
506 * Read the unique system identifier from the /etc/hostid file.
508 * The behavior of /usr/bin/hostid on Linux systems with the
509 * regular eglibc and coreutils is:
511 * 1. Generate the value if the /etc/hostid file does not exist
512 * or if the /etc/hostid file is less than four bytes in size.
514 * 2. If the /etc/hostid file is at least 4 bytes, then return
515 * the first four bytes [0..3] in native endian order.
517 * 3. Always ignore bytes [4..] if they exist in the file.
519 * Only the first four bytes are significant, even on systems that
520 * have a 64-bit word size.
524 * eglibc: sysdeps/unix/sysv/linux/gethostid.c
525 * coreutils: src/hostid.c
529 * The /etc/hostid file on Solaris is a text file that often reads:
534 * Directly copying this file to Linux results in a constant
535 * hostid of 4f442023 because the default comment constitutes
536 * the first four bytes of the file.
540 char *spl_hostid_path
= HW_HOSTID_PATH
;
541 module_param(spl_hostid_path
, charp
, 0444);
542 MODULE_PARM_DESC(spl_hostid_path
, "The system hostid file (/etc/hostid)");
545 hostid_read(uint32_t *hostid
)
552 file
= kobj_open_file(spl_hostid_path
);
553 if (file
== (struct _buf
*)-1)
556 error
= kobj_get_filesize(file
, &size
);
558 kobj_close_file(file
);
562 if (size
< sizeof(HW_HOSTID_MASK
)) {
563 kobj_close_file(file
);
568 * Read directly into the variable like eglibc does.
569 * Short reads are okay; native behavior is preserved.
571 error
= kobj_read_file(file
, (char *)&value
, sizeof(value
), 0);
573 kobj_close_file(file
);
577 /* Mask down to 32 bits like coreutils does. */
578 *hostid
= (value
& HW_HOSTID_MASK
);
579 kobj_close_file(file
);
585 * Return the system hostid. Preferentially use the spl_hostid module option
586 * when set, otherwise use the value in the /etc/hostid file.
589 zone_get_hostid(void *zone
)
593 ASSERT3P(zone
, ==, NULL
);
596 return ((uint32_t)(spl_hostid
& HW_HOSTID_MASK
));
598 if (hostid_read(&hostid
) == 0)
603 EXPORT_SYMBOL(zone_get_hostid
);
610 rc
= spl_kmem_init();
614 rc
= spl_vmem_init();
624 * We initialize the random number generator with 128 bits of entropy from the
625 * system random number generator. In the improbable case that we have a zero
626 * seed, we fallback to the system jiffies, unless it is also zero, in which
627 * situation we use a preprogrammed seed. We step forward by 2^64 iterations to
628 * initialize each of the per-cpu seeds so that the sequences generated on each
629 * CPU are guaranteed to never overlap in practice.
632 spl_random_init(void)
637 get_random_bytes(s
, sizeof (s
));
639 if (s
[0] == 0 && s
[1] == 0) {
644 (void) memcpy(s
, "improbable seed", sizeof (s
));
646 printk("SPL: get_random_bytes() returned 0 "
647 "when generating random seed. Setting initial seed to "
648 "0x%016llx%016llx.", cpu_to_be64(s
[0]), cpu_to_be64(s
[1]));
651 for_each_possible_cpu(i
) {
652 uint64_t *wordp
= per_cpu(spl_pseudo_entropy
, i
);
673 bzero(&p0
, sizeof (proc_t
));
676 if ((rc
= spl_kvmem_init()))
679 if ((rc
= spl_mutex_init()))
682 if ((rc
= spl_rw_init()))
685 if ((rc
= spl_tsd_init()))
688 if ((rc
= spl_taskq_init()))
691 if ((rc
= spl_kmem_cache_init()))
694 if ((rc
= spl_vn_init()))
697 if ((rc
= spl_proc_init()))
700 if ((rc
= spl_kstat_init()))
703 if ((rc
= spl_zlib_init()))
706 printk(KERN_NOTICE
"SPL: Loaded module v%s-%s%s\n", SPL_META_VERSION
,
707 SPL_META_RELEASE
, SPL_DEBUG_STR
);
717 spl_kmem_cache_fini();
729 printk(KERN_NOTICE
"SPL: Failed to Load Solaris Porting Layer "
730 "v%s-%s%s, rc = %d\n", SPL_META_VERSION
, SPL_META_RELEASE
,
739 printk(KERN_NOTICE
"SPL: Unloaded module v%s-%s%s\n",
740 SPL_META_VERSION
, SPL_META_RELEASE
, SPL_DEBUG_STR
);
745 spl_kmem_cache_fini();
753 module_init(spl_init
);
754 module_exit(spl_fini
);
756 MODULE_DESCRIPTION("Solaris Porting Layer");
757 MODULE_AUTHOR(SPL_META_AUTHOR
);
758 MODULE_LICENSE(SPL_META_LICENSE
);
759 MODULE_VERSION(SPL_META_VERSION
"-" SPL_META_RELEASE
);