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://github.com/behlendorf/spl/>.
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/mutex.h>
33 #include <sys/rwlock.h>
34 #include <sys/taskq.h>
37 #include <sys/debug.h>
39 #include <sys/kstat.h>
40 #include <sys/utsname.h>
42 #include <linux/kmod.h>
43 #include <linux/proc_compat.h>
44 #include <spl-debug.h>
46 #ifdef SS_DEBUG_SUBSYS
47 #undef SS_DEBUG_SUBSYS
50 #define SS_DEBUG_SUBSYS SS_GENERIC
52 char spl_version
[32] = "SPL v" SPL_META_VERSION
"-" SPL_META_RELEASE
;
53 EXPORT_SYMBOL(spl_version
);
55 unsigned long spl_hostid
= HW_INVALID_HOSTID
;
56 EXPORT_SYMBOL(spl_hostid
);
57 module_param(spl_hostid
, ulong
, 0644);
58 MODULE_PARM_DESC(spl_hostid
, "The system hostid.");
60 char hw_serial
[HW_HOSTID_LEN
] = "<none>";
61 EXPORT_SYMBOL(hw_serial
);
66 #ifndef HAVE_KALLSYMS_LOOKUP_NAME
67 kallsyms_lookup_name_t spl_kallsyms_lookup_name_fn
= SYMBOL_POISON
;
71 highbit(unsigned long i
)
78 #if BITS_PER_LONG == 64
79 if (i
& 0xffffffff00000000ul
) {
100 EXPORT_SYMBOL(highbit
);
102 #if BITS_PER_LONG == 32
104 * Support 64/64 => 64 division on a 32-bit platform. While the kernel
105 * provides a div64_u64() function for this we do not use it because the
106 * implementation is flawed. There are cases which return incorrect
107 * results as late as linux-2.6.35. Until this is fixed upstream the
108 * spl must provide its own implementation.
110 * This implementation is a slightly modified version of the algorithm
111 * proposed by the book 'Hacker's Delight'. The original source can be
112 * found here and is available for use without restriction.
114 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
118 * Calculate number of leading of zeros for a 64-bit value.
127 if (x
<= 0x00000000FFFFFFFFULL
) {n
= n
+ 32; x
= x
<< 32;}
128 if (x
<= 0x0000FFFFFFFFFFFFULL
) {n
= n
+ 16; x
= x
<< 16;}
129 if (x
<= 0x00FFFFFFFFFFFFFFULL
) {n
= n
+ 8; x
= x
<< 8;}
130 if (x
<= 0x0FFFFFFFFFFFFFFFULL
) {n
= n
+ 4; x
= x
<< 4;}
131 if (x
<= 0x3FFFFFFFFFFFFFFFULL
) {n
= n
+ 2; x
= x
<< 2;}
132 if (x
<= 0x7FFFFFFFFFFFFFFFULL
) {n
= n
+ 1;}
138 * Newer kernels have a div_u64() function but we define our own
139 * to simplify portibility between kernel versions.
141 static inline uint64_t
142 __div_u64(uint64_t u
, uint32_t v
)
149 * Implementation of 64-bit unsigned division for 32-bit machines.
151 * First the procedure takes care of the case in which the divisor is a
152 * 32-bit quantity. There are two subcases: (1) If the left half of the
153 * dividend is less than the divisor, one execution of do_div() is all that
154 * is required (overflow is not possible). (2) Otherwise it does two
155 * divisions, using the grade school method.
158 __udivdi3(uint64_t u
, uint64_t v
)
160 uint64_t u0
, u1
, v1
, q0
, q1
, k
;
163 if (v
>> 32 == 0) { // If v < 2**32:
164 if (u
>> 32 < v
) { // If u/v cannot overflow,
165 return __div_u64(u
, v
); // just do one division.
166 } else { // If u/v would overflow:
167 u1
= u
>> 32; // Break u into two halves.
169 q1
= __div_u64(u1
, v
); // First quotient digit.
170 k
= u1
- q1
* v
; // First remainder, < v.
172 q0
= __div_u64(u0
, v
); // Seconds quotient digit.
173 return (q1
<< 32) + q0
;
175 } else { // If v >= 2**32:
176 n
= nlz64(v
); // 0 <= n <= 31.
177 v1
= (v
<< n
) >> 32; // Normalize divisor, MSB is 1.
178 u1
= u
>> 1; // To ensure no overflow.
179 q1
= __div_u64(u1
, v1
); // Get quotient from
180 q0
= (q1
<< n
) >> 31; // Undo normalization and
181 // division of u by 2.
182 if (q0
!= 0) // Make q0 correct or
183 q0
= q0
- 1; // too small by 1.
184 if ((u
- q0
* v
) >= v
)
185 q0
= q0
+ 1; // Now q0 is correct.
190 EXPORT_SYMBOL(__udivdi3
);
193 * Implementation of 64-bit signed division for 32-bit machines.
196 __divdi3(int64_t u
, int64_t v
)
199 q
= __udivdi3(abs64(u
), abs64(v
));
200 t
= (u
^ v
) >> 63; // If u, v have different
201 return (q
^ t
) - t
; // signs, negate q.
203 EXPORT_SYMBOL(__divdi3
);
206 * Implementation of 64-bit unsigned modulo for 32-bit machines.
209 __umoddi3(uint64_t dividend
, uint64_t divisor
)
211 return (dividend
- (divisor
* __udivdi3(dividend
, divisor
)));
213 EXPORT_SYMBOL(__umoddi3
);
215 #if defined(__arm) || defined(__arm__)
217 * Implementation of 64-bit (un)signed division for 32-bit arm machines.
219 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned)
220 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
221 * and the remainder in {r2, r3}. The return type is specifically left
222 * set to 'void' to ensure the compiler does not overwrite these registers
223 * during the return. All results are in registers as per ABI
226 __aeabi_uldivmod(uint64_t u
, uint64_t v
)
231 res
= __udivdi3(u
, v
);
232 mod
= __umoddi3(u
, v
);
234 register uint32_t r0
asm("r0") = (res
& 0xFFFFFFFF);
235 register uint32_t r1
asm("r1") = (res
>> 32);
236 register uint32_t r2
asm("r2") = (mod
& 0xFFFFFFFF);
237 register uint32_t r3
asm("r3") = (mod
>> 32);
240 : "+r"(r0
), "+r"(r1
), "+r"(r2
),"+r"(r3
) /* output */
241 : "r"(r0
), "r"(r1
), "r"(r2
), "r"(r3
)); /* input */
246 EXPORT_SYMBOL(__aeabi_uldivmod
);
249 __aeabi_ldivmod(int64_t u
, int64_t v
)
254 res
= __divdi3(u
, v
);
255 mod
= __umoddi3(u
, v
);
257 register uint32_t r0
asm("r0") = (res
& 0xFFFFFFFF);
258 register uint32_t r1
asm("r1") = (res
>> 32);
259 register uint32_t r2
asm("r2") = (mod
& 0xFFFFFFFF);
260 register uint32_t r3
asm("r3") = (mod
>> 32);
263 : "+r"(r0
), "+r"(r1
), "+r"(r2
),"+r"(r3
) /* output */
264 : "r"(r0
), "r"(r1
), "r"(r2
), "r"(r3
)); /* input */
269 EXPORT_SYMBOL(__aeabi_ldivmod
);
270 #endif /* __arm || __arm__ */
271 #endif /* BITS_PER_LONG */
273 /* NOTE: The strtoxx behavior is solely based on my reading of the Solaris
274 * ddi_strtol(9F) man page. I have not verified the behavior of these
275 * functions against their Solaris counterparts. It is possible that I
276 * may have misinterpreted the man page or the man page is incorrect.
278 int ddi_strtoul(const char *, char **, int, unsigned long *);
279 int ddi_strtol(const char *, char **, int, long *);
280 int ddi_strtoull(const char *, char **, int, unsigned long long *);
281 int ddi_strtoll(const char *, char **, int, long long *);
283 #define define_ddi_strtoux(type, valtype) \
284 int ddi_strtou##type(const char *str, char **endptr, \
285 int base, valtype *result) \
287 valtype last_value, value = 0; \
288 char *ptr = (char *)str; \
289 int flag = 1, digit; \
291 if (strlen(ptr) == 0) \
294 /* Auto-detect base based on prefix */ \
296 if (str[0] == '0') { \
297 if (tolower(str[1])=='x' && isxdigit(str[2])) { \
298 base = 16; /* hex */ \
300 } else if (str[1] >= '0' && str[1] < 8) { \
301 base = 8; /* octal */ \
307 base = 10; /* decimal */ \
313 digit = *ptr - '0'; \
314 else if (isalpha(*ptr)) \
315 digit = tolower(*ptr) - 'a' + 10; \
322 last_value = value; \
323 value = value * base + digit; \
324 if (last_value > value) /* Overflow */ \
335 *endptr = (char *)(flag ? ptr : str); \
340 #define define_ddi_strtox(type, valtype) \
341 int ddi_strto##type(const char *str, char **endptr, \
342 int base, valtype *result) \
347 rc = ddi_strtou##type(str + 1, endptr, base, result); \
349 if (*endptr == str + 1) \
350 *endptr = (char *)str; \
352 *result = -*result; \
355 rc = ddi_strtou##type(str, endptr, base, result); \
361 define_ddi_strtoux(l
, unsigned long)
362 define_ddi_strtox(l
, long)
363 define_ddi_strtoux(ll
, unsigned long long)
364 define_ddi_strtox(ll
, long long)
366 EXPORT_SYMBOL(ddi_strtoul
);
367 EXPORT_SYMBOL(ddi_strtol
);
368 EXPORT_SYMBOL(ddi_strtoll
);
369 EXPORT_SYMBOL(ddi_strtoull
);
372 ddi_copyin(const void *from
, void *to
, size_t len
, int flags
)
374 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
375 if (flags
& FKIOCTL
) {
376 memcpy(to
, from
, len
);
380 return copyin(from
, to
, len
);
382 EXPORT_SYMBOL(ddi_copyin
);
385 ddi_copyout(const void *from
, void *to
, size_t len
, int flags
)
387 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
388 if (flags
& FKIOCTL
) {
389 memcpy(to
, from
, len
);
393 return copyout(from
, to
, len
);
395 EXPORT_SYMBOL(ddi_copyout
);
397 #ifndef HAVE_PUT_TASK_STRUCT
399 * This is only a stub function which should never be used. The SPL should
400 * never be putting away the last reference on a task structure so this will
401 * not be called. However, we still need to define it so the module does not
402 * have undefined symbol at load time. That all said if this impossible
403 * thing does somehow happen PANIC immediately so we know about it.
406 __put_task_struct(struct task_struct
*t
)
408 PANIC("Unexpectly put last reference on task %d\n", (int)t
->pid
);
410 EXPORT_SYMBOL(__put_task_struct
);
411 #endif /* HAVE_PUT_TASK_STRUCT */
413 struct new_utsname
*__utsname(void)
415 #ifdef HAVE_INIT_UTSNAME
416 return init_utsname();
418 return &system_utsname
;
421 EXPORT_SYMBOL(__utsname
);
425 * Read the unique system identifier from the /etc/hostid file.
427 * The behavior of /usr/bin/hostid on Linux systems with the
428 * regular eglibc and coreutils is:
430 * 1. Generate the value if the /etc/hostid file does not exist
431 * or if the /etc/hostid file is less than four bytes in size.
433 * 2. If the /etc/hostid file is at least 4 bytes, then return
434 * the first four bytes [0..3] in native endian order.
436 * 3. Always ignore bytes [4..] if they exist in the file.
438 * Only the first four bytes are significant, even on systems that
439 * have a 64-bit word size.
443 * eglibc: sysdeps/unix/sysv/linux/gethostid.c
444 * coreutils: src/hostid.c
448 * The /etc/hostid file on Solaris is a text file that often reads:
453 * Directly copying this file to Linux results in a constant
454 * hostid of 4f442023 because the default comment constitutes
455 * the first four bytes of the file.
459 char *spl_hostid_path
= HW_HOSTID_PATH
;
460 module_param(spl_hostid_path
, charp
, 0444);
461 MODULE_PARM_DESC(spl_hostid_path
, "The system hostid file (/etc/hostid)");
469 unsigned long hostid
= 0;
471 file
= kobj_open_file(spl_hostid_path
);
473 if (file
== (struct _buf
*)-1)
476 result
= kobj_get_filesize(file
, &size
);
480 "SPL: kobj_get_filesize returned %i on %s\n",
481 result
, spl_hostid_path
);
482 kobj_close_file(file
);
486 if (size
< sizeof(HW_HOSTID_MASK
)) {
488 "SPL: Ignoring the %s file because it is %llu bytes; "
489 "expecting %lu bytes instead.\n", spl_hostid_path
,
490 size
, (unsigned long)sizeof(HW_HOSTID_MASK
));
491 kobj_close_file(file
);
495 /* Read directly into the variable like eglibc does. */
496 /* Short reads are okay; native behavior is preserved. */
497 result
= kobj_read_file(file
, (char *)&hostid
, sizeof(hostid
), 0);
501 "SPL: kobj_read_file returned %i on %s\n",
502 result
, spl_hostid_path
);
503 kobj_close_file(file
);
507 /* Mask down to 32 bits like coreutils does. */
508 spl_hostid
= hostid
& HW_HOSTID_MASK
;
509 kobj_close_file(file
);
513 #define GET_HOSTID_CMD \
514 "exec 0</dev/null " \
515 " 1>/proc/sys/kernel/spl/hostid " \
522 char *argv
[] = { "/bin/sh",
526 char *envp
[] = { "HOME=/",
528 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
532 /* Doing address resolution in the kernel is tricky and just
533 * not a good idea in general. So to set the proper 'hw_serial'
534 * use the usermodehelper support to ask '/bin/sh' to run
535 * '/usr/bin/hostid' and redirect the result to /proc/sys/spl/hostid
536 * for us to use. It's a horrific solution but it will do for now.
538 rc
= call_usermodehelper(argv
[0], argv
, envp
, 1);
540 printk("SPL: Failed user helper '%s %s %s', rc = %d\n",
541 argv
[0], argv
[1], argv
[2], rc
);
547 zone_get_hostid(void *zone
)
549 unsigned long hostid
;
551 /* Only the global zone is supported */
552 ASSERT(zone
== NULL
);
554 if (ddi_strtoul(hw_serial
, NULL
, HW_HOSTID_LEN
-1, &hostid
) != 0)
555 return HW_INVALID_HOSTID
;
557 return (uint32_t)hostid
;
559 EXPORT_SYMBOL(zone_get_hostid
);
561 #ifndef HAVE_KALLSYMS_LOOKUP_NAME
563 * The kallsyms_lookup_name() kernel function is not an exported symbol in
564 * Linux 2.6.19 through 2.6.32 inclusive.
566 * This function replaces the functionality by performing an upcall to user
567 * space where /proc/kallsyms is consulted for the requested address.
571 #define GET_KALLSYMS_ADDR_CMD \
572 "exec 0</dev/null " \
573 " 1>/proc/sys/kernel/spl/kallsyms_lookup_name " \
575 "awk '{ if ( $3 == \"kallsyms_lookup_name\" ) { print $1 } }' " \
579 set_kallsyms_lookup_name(void)
581 char *argv
[] = { "/bin/sh",
583 GET_KALLSYMS_ADDR_CMD
,
585 char *envp
[] = { "HOME=/",
587 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
591 rc
= call_usermodehelper(argv
[0], argv
, envp
, 1);
593 printk("SPL: Failed user helper '%s %s %s', rc = %d\n",
594 argv
[0], argv
[1], argv
[2], rc
);
601 __init
spl_init(void)
605 if ((rc
= spl_debug_init()))
608 if ((rc
= spl_kmem_init()))
611 if ((rc
= spl_mutex_init()))
614 if ((rc
= spl_rw_init()))
617 if ((rc
= spl_taskq_init()))
620 if ((rc
= spl_vn_init()))
623 if ((rc
= spl_proc_init()))
626 if ((rc
= spl_kstat_init()))
629 if ((rc
= spl_tsd_init()))
632 if ((rc
= spl_zlib_init()))
636 * Get the hostid if it was not passed as a module parameter. Try
637 * reading the /etc/hostid file directly, and then fall back to calling
638 * the /usr/bin/hostid utility.
641 if (spl_hostid
== HW_INVALID_HOSTID
642 && (rc
= hostid_read()) && (rc
= hostid_exec()))
643 SGOTO(out10
, rc
= -EADDRNOTAVAIL
);
645 #ifndef HAVE_KALLSYMS_LOOKUP_NAME
646 if ((rc
= set_kallsyms_lookup_name()))
647 SGOTO(out10
, rc
= -EADDRNOTAVAIL
);
648 #endif /* HAVE_KALLSYMS_LOOKUP_NAME */
650 if ((rc
= spl_kmem_init_kallsyms_lookup()))
653 if ((rc
= spl_vn_init_kallsyms_lookup()))
656 printk(KERN_NOTICE
"SPL: Loaded module v%s-%s%s, using hostid "
657 "0x%08x\n", SPL_META_VERSION
, SPL_META_RELEASE
, SPL_DEBUG_STR
,
658 (unsigned int) spl_hostid
);
681 printk(KERN_NOTICE
"SPL: Failed to Load Solaris Porting Layer "
682 "v%s-%s%s, rc = %d\n", SPL_META_VERSION
, SPL_META_RELEASE
,
692 printk(KERN_NOTICE
"SPL: Unloaded module v%s-%s%s\n",
693 SPL_META_VERSION
, SPL_META_RELEASE
, SPL_DEBUG_STR
);
706 /* Called when a dependent module is loaded */
713 * At module load time the pwd is set to '/' on a Solaris system.
714 * On a Linux system will be set to whatever directory the caller
715 * was in when executing insmod/modprobe.
717 rc
= vn_set_pwd("/");
719 printk("SPL: Warning unable to set pwd to '/': %d\n", rc
);
721 EXPORT_SYMBOL(spl_setup
);
723 /* Called when a dependent module is unloaded */
728 EXPORT_SYMBOL(spl_cleanup
);
730 module_init(spl_init
);
731 module_exit(spl_fini
);
733 MODULE_AUTHOR("Lawrence Livermore National Labs");
734 MODULE_DESCRIPTION("Solaris Porting Layer");
735 MODULE_LICENSE("GPL");