1 /*-
2  * Copyright (c) 2010 Isilon Systems, Inc.
3  * Copyright (c) 2010 iX Systems, Inc.
4  * Copyright (c) 2010 Panasas, Inc.
5  * Copyright (c) 2013-2016 Mellanox Technologies, Ltd.
6  * Copyright (c) 2014-2015 François Tigeot
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 #ifndef	_LINUXKPI_LINUX_KERNEL_H_
31 #define	_LINUXKPI_LINUX_KERNEL_H_
32 
33 #include <sys/cdefs.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/param.h>
37 #include <sys/libkern.h>
38 #include <sys/stat.h>
39 #include <sys/smp.h>
40 #include <sys/stddef.h>
41 #include <sys/syslog.h>
42 #include <sys/time.h>
43 
44 #include <linux/bitops.h>
45 #include <linux/build_bug.h>
46 #include <linux/compiler.h>
47 #include <linux/container_of.h>
48 #include <linux/limits.h>
49 #include <linux/minmax.h>
50 #include <linux/stringify.h>
51 #include <linux/errno.h>
52 #include <linux/sched.h>
53 #include <linux/types.h>
54 #include <linux/typecheck.h>
55 #include <linux/jiffies.h>
56 #include <linux/log2.h>
57 #include <linux/kconfig.h>
58 
59 #include <asm/byteorder.h>
60 #include <asm/cpufeature.h>
61 #include <asm/processor.h>
62 #include <asm/uaccess.h>
63 
64 #include <linux/stdarg.h>
65 
66 #define KERN_CONT       ""
67 #define	KERN_EMERG	"<0>"
68 #define	KERN_ALERT	"<1>"
69 #define	KERN_CRIT	"<2>"
70 #define	KERN_ERR	"<3>"
71 #define	KERN_WARNING	"<4>"
72 #define	KERN_NOTICE	"<5>"
73 #define	KERN_INFO	"<6>"
74 #define	KERN_DEBUG	"<7>"
75 
76 #define	S8_C(x)  x
77 #define	U8_C(x)  x ## U
78 #define	S16_C(x) x
79 #define	U16_C(x) x ## U
80 #define	S32_C(x) x
81 #define	U32_C(x) x ## U
82 #define	S64_C(x) x ## LL
83 #define	U64_C(x) x ## ULL
84 
85 #define	BUG()			panic("BUG at %s:%d", __FILE__, __LINE__)
86 #define	BUG_ON(cond)		do {				\
87 	if (cond) {						\
88 		panic("BUG ON %s failed at %s:%d",		\
89 		    __stringify(cond), __FILE__, __LINE__);	\
90 	}							\
91 } while (0)
92 
93 extern int linuxkpi_warn_dump_stack;
94 #define	WARN_ON(cond) ({					\
95 	bool __ret = (cond);					\
96 	if (__ret) {						\
97 		printf("WARNING %s failed at %s:%d\n",		\
98 		    __stringify(cond), __FILE__, __LINE__);	\
99 		if (linuxkpi_warn_dump_stack)				\
100 			linux_dump_stack();				\
101 	}								\
102 	unlikely(__ret);						\
103 })
104 
105 #define	WARN_ON_SMP(cond)	WARN_ON(cond)
106 
107 #define	WARN_ON_ONCE(cond) ({					\
108 	static bool __warn_on_once;				\
109 	bool __ret = (cond);					\
110 	if (__ret && !__warn_on_once) {				\
111 		__warn_on_once = 1;				\
112 		printf("WARNING %s failed at %s:%d\n",		\
113 		    __stringify(cond), __FILE__, __LINE__);	\
114 		if (linuxkpi_warn_dump_stack)				\
115 			linux_dump_stack();				\
116 	}								\
117 	unlikely(__ret);						\
118 })
119 
120 #define	oops_in_progress	SCHEDULER_STOPPED()
121 
122 #undef	ALIGN
123 #define	ALIGN(x, y)		roundup2((x), (y))
124 #define	ALIGN_DOWN(x, y)	rounddown2(x, y)
125 #undef PTR_ALIGN
126 #define	PTR_ALIGN(p, a)		((__typeof(p))ALIGN((uintptr_t)(p), (a)))
127 #define	IS_ALIGNED(x, a)	(((x) & ((__typeof(x))(a) - 1)) == 0)
128 #define	DIV_ROUND_UP(x, n)	howmany(x, n)
129 #define	__KERNEL_DIV_ROUND_UP(x, n)	howmany(x, n)
130 #define	DIV_ROUND_UP_ULL(x, n)	DIV_ROUND_UP((unsigned long long)(x), (n))
131 #define	DIV_ROUND_DOWN_ULL(x, n) (((unsigned long long)(x) / (n)) * (n))
132 #define	FIELD_SIZEOF(t, f)	sizeof(((t *)0)->f)
133 
134 #define	printk(...)		printf(__VA_ARGS__)
135 #define	vprintk(f, a)		vprintf(f, a)
136 
137 #define PTR_IF(x, p)		((x) ? (p) : NULL)
138 
139 #define	asm			__asm
140 
141 extern void linux_dump_stack(void);
142 #define	dump_stack()		linux_dump_stack()
143 
144 struct va_format {
145 	const char *fmt;
146 	va_list *va;
147 };
148 
149 static inline int
150 vscnprintf(char *buf, size_t size, const char *fmt, va_list args)
151 {
152 	ssize_t ssize = size;
153 	int i;
154 
155 	i = vsnprintf(buf, size, fmt, args);
156 
157 	return ((i >= ssize) ? (ssize - 1) : i);
158 }
159 
160 static inline int
161 scnprintf(char *buf, size_t size, const char *fmt, ...)
162 {
163 	va_list args;
164 	int i;
165 
166 	va_start(args, fmt);
167 	i = vscnprintf(buf, size, fmt, args);
168 	va_end(args);
169 
170 	return (i);
171 }
172 
173 /*
174  * The "pr_debug()" and "pr_devel()" macros should produce zero code
175  * unless DEBUG is defined:
176  */
177 #ifdef DEBUG
178 extern int linuxkpi_debug;
179 #define pr_debug(fmt, ...)					\
180 	do {							\
181 		if (linuxkpi_debug)				\
182 			log(LOG_DEBUG, fmt, ##__VA_ARGS__);	\
183 	} while (0)
184 #define pr_devel(fmt, ...) \
185 	log(LOG_DEBUG, pr_fmt(fmt), ##__VA_ARGS__)
186 #else
187 #define pr_debug(fmt, ...) \
188 	({ if (0) log(LOG_DEBUG, fmt, ##__VA_ARGS__); 0; })
189 #define pr_devel(fmt, ...) \
190 	({ if (0) log(LOG_DEBUG, pr_fmt(fmt), ##__VA_ARGS__); 0; })
191 #endif
192 
193 #ifndef pr_fmt
194 #define pr_fmt(fmt) fmt
195 #endif
196 
197 /*
198  * Print a one-time message (analogous to WARN_ONCE() et al):
199  */
200 #define printk_once(...) do {			\
201 	static bool __print_once;		\
202 						\
203 	if (!__print_once) {			\
204 		__print_once = true;		\
205 		printk(__VA_ARGS__);		\
206 	}					\
207 } while (0)
208 
209 /*
210  * Log a one-time message (analogous to WARN_ONCE() et al):
211  */
212 #define log_once(level,...) do {		\
213 	static bool __log_once;			\
214 						\
215 	if (unlikely(!__log_once)) {		\
216 		__log_once = true;		\
217 		log(level, __VA_ARGS__);	\
218 	}					\
219 } while (0)
220 
221 #define pr_emerg(fmt, ...) \
222 	log(LOG_EMERG, pr_fmt(fmt), ##__VA_ARGS__)
223 #define pr_alert(fmt, ...) \
224 	log(LOG_ALERT, pr_fmt(fmt), ##__VA_ARGS__)
225 #define pr_crit(fmt, ...) \
226 	log(LOG_CRIT, pr_fmt(fmt), ##__VA_ARGS__)
227 #define pr_err(fmt, ...) \
228 	log(LOG_ERR, pr_fmt(fmt), ##__VA_ARGS__)
229 #define pr_err_once(fmt, ...) \
230 	log_once(LOG_ERR, pr_fmt(fmt), ##__VA_ARGS__)
231 #define pr_warning(fmt, ...) \
232 	log(LOG_WARNING, pr_fmt(fmt), ##__VA_ARGS__)
233 #define pr_warn(...) \
234 	pr_warning(__VA_ARGS__)
235 #define pr_warn_once(fmt, ...) \
236 	log_once(LOG_WARNING, pr_fmt(fmt), ##__VA_ARGS__)
237 #define pr_notice(fmt, ...) \
238 	log(LOG_NOTICE, pr_fmt(fmt), ##__VA_ARGS__)
239 #define pr_info(fmt, ...) \
240 	log(LOG_INFO, pr_fmt(fmt), ##__VA_ARGS__)
241 #define pr_info_once(fmt, ...) \
242 	log_once(LOG_INFO, pr_fmt(fmt), ##__VA_ARGS__)
243 #define pr_cont(fmt, ...) \
244 	printk(KERN_CONT fmt, ##__VA_ARGS__)
245 #define	pr_warn_ratelimited(...) do {		\
246 	static linux_ratelimit_t __ratelimited;	\
247 	if (linux_ratelimited(&__ratelimited))	\
248 		pr_warning(__VA_ARGS__);	\
249 } while (0)
250 
251 #ifndef WARN
252 #define	WARN(condition, ...) ({			\
253 	bool __ret_warn_on = (condition);	\
254 	if (unlikely(__ret_warn_on))		\
255 		pr_warning(__VA_ARGS__);	\
256 	unlikely(__ret_warn_on);		\
257 })
258 #endif
259 
260 #ifndef WARN_ONCE
261 #define	WARN_ONCE(condition, ...) ({		\
262 	bool __ret_warn_on = (condition);	\
263 	if (unlikely(__ret_warn_on))		\
264 		pr_warn_once(__VA_ARGS__);	\
265 	unlikely(__ret_warn_on);		\
266 })
267 #endif
268 
269 #define	ARRAY_SIZE(x)	(sizeof(x) / sizeof((x)[0]))
270 
271 #define	u64_to_user_ptr(val)	((void *)(uintptr_t)(val))
272 
273 #define _RET_IP_		__builtin_return_address(0)
274 
275 static inline unsigned long long
276 simple_strtoull(const char *cp, char **endp, unsigned int base)
277 {
278 	return (strtouq(cp, endp, base));
279 }
280 
281 static inline long long
282 simple_strtoll(const char *cp, char **endp, unsigned int base)
283 {
284 	return (strtoq(cp, endp, base));
285 }
286 
287 static inline unsigned long
288 simple_strtoul(const char *cp, char **endp, unsigned int base)
289 {
290 	return (strtoul(cp, endp, base));
291 }
292 
293 static inline long
294 simple_strtol(const char *cp, char **endp, unsigned int base)
295 {
296 	return (strtol(cp, endp, base));
297 }
298 
299 static inline int
300 kstrtoul(const char *cp, unsigned int base, unsigned long *res)
301 {
302 	char *end;
303 
304 	*res = strtoul(cp, &end, base);
305 
306 	/* skip newline character, if any */
307 	if (*end == '\n')
308 		end++;
309 	if (*cp == 0 || *end != 0)
310 		return (-EINVAL);
311 	return (0);
312 }
313 
314 static inline int
315 kstrtol(const char *cp, unsigned int base, long *res)
316 {
317 	char *end;
318 
319 	*res = strtol(cp, &end, base);
320 
321 	/* skip newline character, if any */
322 	if (*end == '\n')
323 		end++;
324 	if (*cp == 0 || *end != 0)
325 		return (-EINVAL);
326 	return (0);
327 }
328 
329 static inline int
330 kstrtoint(const char *cp, unsigned int base, int *res)
331 {
332 	char *end;
333 	long temp;
334 
335 	*res = temp = strtol(cp, &end, base);
336 
337 	/* skip newline character, if any */
338 	if (*end == '\n')
339 		end++;
340 	if (*cp == 0 || *end != 0)
341 		return (-EINVAL);
342 	if (temp != (int)temp)
343 		return (-ERANGE);
344 	return (0);
345 }
346 
347 static inline int
348 kstrtouint(const char *cp, unsigned int base, unsigned int *res)
349 {
350 	char *end;
351 	unsigned long temp;
352 
353 	*res = temp = strtoul(cp, &end, base);
354 
355 	/* skip newline character, if any */
356 	if (*end == '\n')
357 		end++;
358 	if (*cp == 0 || *end != 0)
359 		return (-EINVAL);
360 	if (temp != (unsigned int)temp)
361 		return (-ERANGE);
362 	return (0);
363 }
364 
365 static inline int
366 kstrtou8(const char *cp, unsigned int base, u8 *res)
367 {
368 	char *end;
369 	unsigned long temp;
370 
371 	*res = temp = strtoul(cp, &end, base);
372 
373 	/* skip newline character, if any */
374 	if (*end == '\n')
375 		end++;
376 	if (*cp == 0 || *end != 0)
377 		return (-EINVAL);
378 	if (temp != (u8)temp)
379 		return (-ERANGE);
380 	return (0);
381 }
382 
383 static inline int
384 kstrtou16(const char *cp, unsigned int base, u16 *res)
385 {
386 	char *end;
387 	unsigned long temp;
388 
389 	*res = temp = strtoul(cp, &end, base);
390 
391 	/* skip newline character, if any */
392 	if (*end == '\n')
393 		end++;
394 	if (*cp == 0 || *end != 0)
395 		return (-EINVAL);
396 	if (temp != (u16)temp)
397 		return (-ERANGE);
398 	return (0);
399 }
400 
401 static inline int
402 kstrtou32(const char *cp, unsigned int base, u32 *res)
403 {
404 
405 	return (kstrtouint(cp, base, res));
406 }
407 
408 static inline int
409 kstrtou64(const char *cp, unsigned int base, u64 *res)
410 {
411        char *end;
412 
413        *res = strtouq(cp, &end, base);
414 
415        /* skip newline character, if any */
416        if (*end == '\n')
417                end++;
418        if (*cp == 0 || *end != 0)
419                return (-EINVAL);
420        return (0);
421 }
422 
423 static inline int
424 kstrtoull(const char *cp, unsigned int base, unsigned long long *res)
425 {
426 	return (kstrtou64(cp, base, (u64 *)res));
427 }
428 
429 static inline int
430 kstrtobool(const char *s, bool *res)
431 {
432 	int len;
433 
434 	if (s == NULL || (len = strlen(s)) == 0 || res == NULL)
435 		return (-EINVAL);
436 
437 	/* skip newline character, if any */
438 	if (s[len - 1] == '\n')
439 		len--;
440 
441 	if (len == 1 && strchr("yY1", s[0]) != NULL)
442 		*res = true;
443 	else if (len == 1 && strchr("nN0", s[0]) != NULL)
444 		*res = false;
445 	else if (strncasecmp("on", s, len) == 0)
446 		*res = true;
447 	else if (strncasecmp("off", s, len) == 0)
448 		*res = false;
449 	else
450 		return (-EINVAL);
451 
452 	return (0);
453 }
454 
455 static inline int
456 kstrtobool_from_user(const char __user *s, size_t count, bool *res)
457 {
458 	char buf[8] = {};
459 
460 	if (count > (sizeof(buf) - 1))
461 		count = (sizeof(buf) - 1);
462 
463 	if (copy_from_user(buf, s, count))
464 		return (-EFAULT);
465 
466 	return (kstrtobool(buf, res));
467 }
468 
469 static inline int
470 kstrtoint_from_user(const char __user *s, size_t count, unsigned int base,
471     int *p)
472 {
473 	char buf[36] = {};
474 
475 	if (count > (sizeof(buf) - 1))
476 		count = (sizeof(buf) - 1);
477 
478 	if (copy_from_user(buf, s, count))
479 		return (-EFAULT);
480 
481 	return (kstrtoint(buf, base, p));
482 }
483 
484 static inline int
485 kstrtouint_from_user(const char __user *s, size_t count, unsigned int base,
486     unsigned int *p)
487 {
488 	char buf[36] = {};
489 
490 	if (count > (sizeof(buf) - 1))
491 		count = (sizeof(buf) - 1);
492 
493 	if (copy_from_user(buf, s, count))
494 		return (-EFAULT);
495 
496 	return (kstrtouint(buf, base, p));
497 }
498 
499 static inline int
500 kstrtou32_from_user(const char __user *s, size_t count, unsigned int base,
501     unsigned int *p)
502 {
503 
504 	return (kstrtouint_from_user(s, count, base, p));
505 }
506 
507 static inline int
508 kstrtou8_from_user(const char __user *s, size_t count, unsigned int base,
509     u8 *p)
510 {
511 	char buf[8] = {};
512 
513 	if (count > (sizeof(buf) - 1))
514 		count = (sizeof(buf) - 1);
515 
516 	if (copy_from_user(buf, s, count))
517 		return (-EFAULT);
518 
519 	return (kstrtou8(buf, base, p));
520 }
521 
522 #define offsetofend(t, m)	\
523         (offsetof(t, m) + sizeof((((t *)0)->m)))
524 
525 /*
526  * This looks more complex than it should be. But we need to
527  * get the type for the ~ right in round_down (it needs to be
528  * as wide as the result!), and we want to evaluate the macro
529  * arguments just once each.
530  */
531 #define __round_mask(x, y) ((__typeof__(x))((y)-1))
532 #define round_up(x, y) ((((x)-1) | __round_mask(x, y))+1)
533 #define round_down(x, y) ((x) & ~__round_mask(x, y))
534 
535 #define	smp_processor_id()	PCPU_GET(cpuid)
536 #define	num_possible_cpus()	mp_ncpus
537 #define	num_online_cpus()	mp_ncpus
538 
539 #if defined(__i386__) || defined(__amd64__)
540 extern bool linux_cpu_has_clflush;
541 #define	cpu_has_clflush		linux_cpu_has_clflush
542 #endif
543 
544 #define	DIV_ROUND_CLOSEST(x, divisor)	(((x) + ((divisor) / 2)) / (divisor))
545 
546 #define	DIV_ROUND_CLOSEST_ULL(x, divisor) ({		\
547 	__typeof(divisor) __d = (divisor);		\
548 	unsigned long long __ret = (x) + (__d) / 2;	\
549 	__ret /= __d;					\
550 	__ret;						\
551 })
552 
553 static inline uintmax_t
554 mult_frac(uintmax_t x, uintmax_t multiplier, uintmax_t divisor)
555 {
556 	uintmax_t q = (x / divisor);
557 	uintmax_t r = (x % divisor);
558 
559 	return ((q * multiplier) + ((r * multiplier) / divisor));
560 }
561 
562 typedef struct linux_ratelimit {
563 	struct timeval lasttime;
564 	int counter;
565 } linux_ratelimit_t;
566 
567 static inline bool
568 linux_ratelimited(linux_ratelimit_t *rl)
569 {
570 	return (ppsratecheck(&rl->lasttime, &rl->counter, 1));
571 }
572 
573 #define	__is_constexpr(x) \
574 	__builtin_constant_p(x)
575 
576 /*
577  * The is_signed() macro below returns true if the passed data type is
578  * signed. Else false is returned.
579  */
580 #define	is_signed(datatype) (((datatype)-1 / (datatype)2) == (datatype)0)
581 
582 #define	TAINT_WARN	0
583 #define	test_taint(x)	(0)
584 #define	add_taint(x,y)	do {	\
585 	} while (0)
586 
587 static inline int
588 _h2b(const char c)
589 {
590 
591 	if (c >= '0' && c <= '9')
592 		return (c - '0');
593 	if (c >= 'a' && c <= 'f')
594 		return (10 + c - 'a');
595 	if (c >= 'A' && c <= 'F')
596 		return (10 + c - 'A');
597 	return (-EINVAL);
598 }
599 
600 static inline int
601 hex2bin(uint8_t *bindst, const char *hexsrc, size_t binlen)
602 {
603 	int hi4, lo4;
604 
605 	while (binlen > 0) {
606 		hi4 = _h2b(*hexsrc++);
607 		lo4 = _h2b(*hexsrc++);
608 		if (hi4 < 0 || lo4 < 0)
609 			return (-EINVAL);
610 
611 		*bindst++ = (hi4 << 4) | lo4;
612 		binlen--;
613 	}
614 
615 	return (0);
616 }
617 
618 static inline bool
619 mac_pton(const char *macin, uint8_t *macout)
620 {
621 	const char *s, *d;
622 	uint8_t mac[6], hx, lx;;
623 	int i;
624 
625 	if (strlen(macin) < (3 * 6 - 1))
626 		return (false);
627 
628 	i = 0;
629 	s = macin;
630 	do {
631 		/* Should we also support '-'-delimiters? */
632 		d = strchrnul(s, ':');
633 		hx = lx = 0;
634 		while (s < d) {
635 			/* Fail on abc:123:xxx:... */
636 			if ((d - s) > 2)
637 				return (false);
638 			/* We do support non-well-formed strings: 3:45:6:... */
639 			if ((d - s) > 1) {
640 				hx = _h2b(*s);
641 				if (hx < 0)
642 					return (false);
643 				s++;
644 			}
645 			lx = _h2b(*s);
646 			if (lx < 0)
647 				return (false);
648 			s++;
649 		}
650 		mac[i] = (hx << 4) | lx;
651 		i++;
652 		if (i >= 6)
653 			return (false);
654 	} while (d != NULL && *d != '\0');
655 
656 	memcpy(macout, mac, 6);
657 	return (true);
658 }
659 
660 #define	DECLARE_FLEX_ARRAY(_t, _n)					\
661     struct { struct { } __dummy_ ## _n; _t _n[0]; }
662 
663 #endif	/* _LINUXKPI_LINUX_KERNEL_H_ */
664