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.
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.
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
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/>.
23 * Solaris Porting Layer (SPL) Generic Implementation.
26 #include <sys/sysmacros.h>
27 #include <sys/systeminfo.h>
28 #include <sys/vmsystm.h>
30 #include <sys/kmem_cache.h>
32 #include <sys/mutex.h>
33 #include <sys/rwlock.h>
34 #include <sys/taskq.h>
37 #include <sys/debug.h>
39 #include <sys/kstat.h>
41 #include <sys/sunddi.h>
42 #include <linux/ctype.h>
44 #include <sys/random.h>
45 #include <sys/strings.h>
46 #include <linux/kmod.h>
47 #include "zfs_gitrev.h"
48 #include <linux/mod_compat.h>
50 #include <sys/vnode.h>
52 char spl_gitrev[64] = ZFS_META_GITREV;
55 unsigned long spl_hostid = 0;
56 EXPORT_SYMBOL(spl_hostid);
58 module_param(spl_hostid, ulong, 0644);
59 MODULE_PARM_DESC(spl_hostid, "The system hostid.");
66 * Xorshift Pseudo Random Number Generator based on work by Sebastiano Vigna
68 * "Further scramblings of Marsaglia's xorshift generators"
69 * http://vigna.di.unimi.it/ftp/papers/xorshiftplus.pdf
71 * random_get_pseudo_bytes() is an API function on Illumos whose sole purpose
72 * is to provide bytes containing random numbers. It is mapped to /dev/urandom
73 * on Illumos, which uses a "FIPS 186-2 algorithm". No user of the SPL's
74 * random_get_pseudo_bytes() needs bytes that are of cryptographic quality, so
75 * we can implement it using a fast PRNG that we seed using Linux' actual
76 * equivalent to random_get_pseudo_bytes(). We do this by providing each CPU
77 * with an independent seed so that all calls to random_get_pseudo_bytes() are
78 * free of atomic instructions.
80 * A consequence of using a fast PRNG is that using random_get_pseudo_bytes()
81 * to generate words larger than 128 bits will paradoxically be limited to
82 * `2^128 - 1` possibilities. This is because we have a sequence of `2^128 - 1`
83 * 128-bit words and selecting the first will implicitly select the second. If
84 * a caller finds this behavior undesirable, random_get_bytes() should be used
87 * XXX: Linux interrupt handlers that trigger within the critical section
88 * formed by `s[1] = xp[1];` and `xp[0] = s[0];` and call this function will
89 * see the same numbers. Nothing in the code currently calls this in an
90 * interrupt handler, so this is considered to be okay. If that becomes a
91 * problem, we could create a set of per-cpu variables for interrupt handlers
92 * and use them when in_interrupt() from linux/preempt_mask.h evaluates to
95 void __percpu *spl_pseudo_entropy;
98 * spl_rand_next()/spl_rand_jump() are copied from the following CC-0 licensed
101 * http://xorshift.di.unimi.it/xorshift128plus.c
104 static inline uint64_t
105 spl_rand_next(uint64_t *s)
108 const uint64_t s0 = s[1];
111 s[1] = s1 ^ s0 ^ (s1 >> 18) ^ (s0 >> 5); // b, c
116 spl_rand_jump(uint64_t *s)
118 static const uint64_t JUMP[] =
119 { 0x8a5cd789635d2dff, 0x121fd2155c472f96 };
124 for (i = 0; i < sizeof (JUMP) / sizeof (*JUMP); i++)
125 for (b = 0; b < 64; b++) {
126 if (JUMP[i] & 1ULL << b) {
130 (void) spl_rand_next(s);
138 random_get_pseudo_bytes(uint8_t *ptr, size_t len)
144 xp = get_cpu_ptr(spl_pseudo_entropy);
152 uint8_t byte[sizeof (uint64_t)];
154 int i = MIN(len, sizeof (uint64_t));
157 entropy.ui64 = spl_rand_next(s);
160 *ptr++ = entropy.byte[i];
166 put_cpu_ptr(spl_pseudo_entropy);
172 EXPORT_SYMBOL(random_get_pseudo_bytes);
174 #if BITS_PER_LONG == 32
177 * Support 64/64 => 64 division on a 32-bit platform. While the kernel
178 * provides a div64_u64() function for this we do not use it because the
179 * implementation is flawed. There are cases which return incorrect
180 * results as late as linux-2.6.35. Until this is fixed upstream the
181 * spl must provide its own implementation.
183 * This implementation is a slightly modified version of the algorithm
184 * proposed by the book 'Hacker's Delight'. The original source can be
185 * found here and is available for use without restriction.
187 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
191 * Calculate number of leading of zeros for a 64-bit value.
201 if (x <= 0x00000000FFFFFFFFULL) { n = n + 32; x = x << 32; }
202 if (x <= 0x0000FFFFFFFFFFFFULL) { n = n + 16; x = x << 16; }
203 if (x <= 0x00FFFFFFFFFFFFFFULL) { n = n + 8; x = x << 8; }
204 if (x <= 0x0FFFFFFFFFFFFFFFULL) { n = n + 4; x = x << 4; }
205 if (x <= 0x3FFFFFFFFFFFFFFFULL) { n = n + 2; x = x << 2; }
206 if (x <= 0x7FFFFFFFFFFFFFFFULL) { n = n + 1; }
212 * Newer kernels have a div_u64() function but we define our own
213 * to simplify portability between kernel versions.
215 static inline uint64_t
216 __div_u64(uint64_t u, uint32_t v)
223 * Turn off missing prototypes warning for these functions. They are
224 * replacements for libgcc-provided functions and will never be called
227 #pragma GCC diagnostic push
228 #pragma GCC diagnostic ignored "-Wmissing-prototypes"
231 * Implementation of 64-bit unsigned division for 32-bit machines.
233 * First the procedure takes care of the case in which the divisor is a
234 * 32-bit quantity. There are two subcases: (1) If the left half of the
235 * dividend is less than the divisor, one execution of do_div() is all that
236 * is required (overflow is not possible). (2) Otherwise it does two
237 * divisions, using the grade school method.
240 __udivdi3(uint64_t u, uint64_t v)
242 uint64_t u0, u1, v1, q0, q1, k;
245 if (v >> 32 == 0) { // If v < 2**32:
246 if (u >> 32 < v) { // If u/v cannot overflow,
247 return (__div_u64(u, v)); // just do one division.
248 } else { // If u/v would overflow:
249 u1 = u >> 32; // Break u into two halves.
251 q1 = __div_u64(u1, v); // First quotient digit.
252 k = u1 - q1 * v; // First remainder, < v.
254 q0 = __div_u64(u0, v); // Seconds quotient digit.
255 return ((q1 << 32) + q0);
257 } else { // If v >= 2**32:
258 n = nlz64(v); // 0 <= n <= 31.
259 v1 = (v << n) >> 32; // Normalize divisor, MSB is 1.
260 u1 = u >> 1; // To ensure no overflow.
261 q1 = __div_u64(u1, v1); // Get quotient from
262 q0 = (q1 << n) >> 31; // Undo normalization and
263 // division of u by 2.
264 if (q0 != 0) // Make q0 correct or
265 q0 = q0 - 1; // too small by 1.
266 if ((u - q0 * v) >= v)
267 q0 = q0 + 1; // Now q0 is correct.
272 EXPORT_SYMBOL(__udivdi3);
276 #define abs64(x) ({ uint64_t t = (x) >> 63; ((x) ^ t) - t; })
281 * Implementation of 64-bit signed division for 32-bit machines.
284 __divdi3(int64_t u, int64_t v)
287 // cppcheck-suppress shiftTooManyBitsSigned
288 q = __udivdi3(abs64(u), abs64(v));
289 // cppcheck-suppress shiftTooManyBitsSigned
290 t = (u ^ v) >> 63; // If u, v have different
291 return ((q ^ t) - t); // signs, negate q.
293 EXPORT_SYMBOL(__divdi3);
296 * Implementation of 64-bit unsigned modulo for 32-bit machines.
299 __umoddi3(uint64_t dividend, uint64_t divisor)
301 return (dividend - (divisor * __udivdi3(dividend, divisor)));
303 EXPORT_SYMBOL(__umoddi3);
305 /* 64-bit signed modulo for 32-bit machines. */
307 __moddi3(int64_t n, int64_t d)
310 boolean_t nn = B_FALSE;
321 return (nn ? -q : q);
323 EXPORT_SYMBOL(__moddi3);
326 * Implementation of 64-bit unsigned division/modulo for 32-bit machines.
329 __udivmoddi4(uint64_t n, uint64_t d, uint64_t *r)
331 uint64_t q = __udivdi3(n, d);
336 EXPORT_SYMBOL(__udivmoddi4);
339 * Implementation of 64-bit signed division/modulo for 32-bit machines.
342 __divmoddi4(int64_t n, int64_t d, int64_t *r)
345 boolean_t nn = B_FALSE;
346 boolean_t nd = B_FALSE;
356 q = __udivmoddi4(n, d, (uint64_t *)&rr);
366 EXPORT_SYMBOL(__divmoddi4);
368 #if defined(__arm) || defined(__arm__)
370 * Implementation of 64-bit (un)signed division for 32-bit arm machines.
372 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned)
373 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
374 * and the remainder in {r2, r3}. The return type is specifically left
375 * set to 'void' to ensure the compiler does not overwrite these registers
376 * during the return. All results are in registers as per ABI
379 __aeabi_uldivmod(uint64_t u, uint64_t v)
384 res = __udivdi3(u, v);
385 mod = __umoddi3(u, v);
387 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
388 register uint32_t r1 asm("r1") = (res >> 32);
389 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
390 register uint32_t r3 asm("r3") = (mod >> 32);
394 : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3) /* output */
395 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
401 EXPORT_SYMBOL(__aeabi_uldivmod);
404 __aeabi_ldivmod(int64_t u, int64_t v)
409 res = __divdi3(u, v);
410 mod = __umoddi3(u, v);
412 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
413 register uint32_t r1 asm("r1") = (res >> 32);
414 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
415 register uint32_t r3 asm("r3") = (mod >> 32);
419 : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3) /* output */
420 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
426 EXPORT_SYMBOL(__aeabi_ldivmod);
427 #endif /* __arm || __arm__ */
429 #pragma GCC diagnostic pop
431 #endif /* BITS_PER_LONG */
434 * NOTE: The strtoxx behavior is solely based on my reading of the Solaris
435 * ddi_strtol(9F) man page. I have not verified the behavior of these
436 * functions against their Solaris counterparts. It is possible that I
437 * may have misinterpreted the man page or the man page is incorrect.
439 int ddi_strtoul(const char *, char **, int, unsigned long *);
440 int ddi_strtol(const char *, char **, int, long *);
441 int ddi_strtoull(const char *, char **, int, unsigned long long *);
442 int ddi_strtoll(const char *, char **, int, long long *);
444 #define define_ddi_strtoux(type, valtype) \
445 int ddi_strtou##type(const char *str, char **endptr, \
446 int base, valtype *result) \
448 valtype last_value, value = 0; \
449 char *ptr = (char *)str; \
450 int flag = 1, digit; \
452 if (strlen(ptr) == 0) \
455 /* Auto-detect base based on prefix */ \
457 if (str[0] == '0') { \
458 if (tolower(str[1]) == 'x' && isxdigit(str[2])) { \
459 base = 16; /* hex */ \
461 } else if (str[1] >= '0' && str[1] < 8) { \
462 base = 8; /* octal */ \
468 base = 10; /* decimal */ \
474 digit = *ptr - '0'; \
475 else if (isalpha(*ptr)) \
476 digit = tolower(*ptr) - 'a' + 10; \
483 last_value = value; \
484 value = value * base + digit; \
485 if (last_value > value) /* Overflow */ \
496 *endptr = (char *)(flag ? ptr : str); \
501 #define define_ddi_strtox(type, valtype) \
502 int ddi_strto##type(const char *str, char **endptr, \
503 int base, valtype *result) \
508 rc = ddi_strtou##type(str + 1, endptr, base, result); \
510 if (*endptr == str + 1) \
511 *endptr = (char *)str; \
513 *result = -*result; \
516 rc = ddi_strtou##type(str, endptr, base, result); \
522 define_ddi_strtoux(l, unsigned long)
523 define_ddi_strtox(l, long)
524 define_ddi_strtoux(ll, unsigned long long)
525 define_ddi_strtox(ll, long long)
527 EXPORT_SYMBOL(ddi_strtoul);
528 EXPORT_SYMBOL(ddi_strtol);
529 EXPORT_SYMBOL(ddi_strtoll);
530 EXPORT_SYMBOL(ddi_strtoull);
533 ddi_copyin(const void *from, void *to, size_t len, int flags)
535 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
536 if (flags & FKIOCTL) {
537 memcpy(to, from, len);
541 return (copyin(from, to, len));
543 EXPORT_SYMBOL(ddi_copyin);
546 ddi_copyout(const void *from, void *to, size_t len, int flags)
548 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
549 if (flags & FKIOCTL) {
550 memcpy(to, from, len);
554 return (copyout(from, to, len));
556 EXPORT_SYMBOL(ddi_copyout);
559 spl_kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
561 #if defined(HAVE_KERNEL_READ_PPOS)
562 return (kernel_read(file, buf, count, pos));
564 mm_segment_t saved_fs;
570 ret = vfs_read(file, (void __user *)buf, count, pos);
579 spl_getattr(struct file *filp, struct kstat *stat)
586 #if defined(HAVE_4ARGS_VFS_GETATTR)
587 rc = vfs_getattr(&filp->f_path, stat, STATX_BASIC_STATS,
588 AT_STATX_SYNC_AS_STAT);
589 #elif defined(HAVE_2ARGS_VFS_GETATTR)
590 rc = vfs_getattr(&filp->f_path, stat);
592 rc = vfs_getattr(filp->f_path.mnt, filp->f_dentry, stat);
601 * Read the unique system identifier from the /etc/hostid file.
603 * The behavior of /usr/bin/hostid on Linux systems with the
604 * regular eglibc and coreutils is:
606 * 1. Generate the value if the /etc/hostid file does not exist
607 * or if the /etc/hostid file is less than four bytes in size.
609 * 2. If the /etc/hostid file is at least 4 bytes, then return
610 * the first four bytes [0..3] in native endian order.
612 * 3. Always ignore bytes [4..] if they exist in the file.
614 * Only the first four bytes are significant, even on systems that
615 * have a 64-bit word size.
619 * eglibc: sysdeps/unix/sysv/linux/gethostid.c
620 * coreutils: src/hostid.c
624 * The /etc/hostid file on Solaris is a text file that often reads:
629 * Directly copying this file to Linux results in a constant
630 * hostid of 4f442023 because the default comment constitutes
631 * the first four bytes of the file.
635 char *spl_hostid_path = HW_HOSTID_PATH;
636 module_param(spl_hostid_path, charp, 0444);
637 MODULE_PARM_DESC(spl_hostid_path, "The system hostid file (/etc/hostid)");
640 hostid_read(uint32_t *hostid)
649 filp = filp_open(spl_hostid_path, 0, 0);
654 error = spl_getattr(filp, &stat);
660 if (size < sizeof (HW_HOSTID_MASK)) {
667 * Read directly into the variable like eglibc does.
668 * Short reads are okay; native behavior is preserved.
670 error = spl_kernel_read(filp, &value, sizeof (value), &off);
676 /* Mask down to 32 bits like coreutils does. */
677 *hostid = (value & HW_HOSTID_MASK);
684 * Return the system hostid. Preferentially use the spl_hostid module option
685 * when set, otherwise use the value in the /etc/hostid file.
688 zone_get_hostid(void *zone)
692 ASSERT3P(zone, ==, NULL);
695 return ((uint32_t)(spl_hostid & HW_HOSTID_MASK));
697 if (hostid_read(&hostid) == 0)
702 EXPORT_SYMBOL(zone_get_hostid);
709 rc = spl_kmem_init();
713 rc = spl_vmem_init();
723 * We initialize the random number generator with 128 bits of entropy from the
724 * system random number generator. In the improbable case that we have a zero
725 * seed, we fallback to the system jiffies, unless it is also zero, in which
726 * situation we use a preprogrammed seed. We step forward by 2^64 iterations to
727 * initialize each of the per-cpu seeds so that the sequences generated on each
728 * CPU are guaranteed to never overlap in practice.
731 spl_random_init(void)
736 spl_pseudo_entropy = __alloc_percpu(2 * sizeof (uint64_t),
739 get_random_bytes(s, sizeof (s));
741 if (s[0] == 0 && s[1] == 0) {
746 (void) memcpy(s, "improbable seed", sizeof (s));
748 printk("SPL: get_random_bytes() returned 0 "
749 "when generating random seed. Setting initial seed to "
750 "0x%016llx%016llx.\n", cpu_to_be64(s[0]),
754 for_each_possible_cpu(i) {
755 uint64_t *wordp = per_cpu_ptr(spl_pseudo_entropy, i);
765 spl_random_fini(void)
767 free_percpu(spl_pseudo_entropy);
782 bzero(&p0, sizeof (proc_t));
785 if ((rc = spl_kvmem_init()))
788 if ((rc = spl_tsd_init()))
791 if ((rc = spl_taskq_init()))
794 if ((rc = spl_kmem_cache_init()))
797 if ((rc = spl_proc_init()))
800 if ((rc = spl_kstat_init()))
803 if ((rc = spl_zlib_init()))
813 spl_kmem_cache_fini();
830 spl_kmem_cache_fini();
837 module_init(spl_init);
838 module_exit(spl_fini);
840 ZFS_MODULE_DESCRIPTION("Solaris Porting Layer");
841 ZFS_MODULE_AUTHOR(ZFS_META_AUTHOR);
842 ZFS_MODULE_LICENSE("GPL");
843 ZFS_MODULE_VERSION(ZFS_META_VERSION "-" ZFS_META_RELEASE);