xref: /freebsd/sys/kern/subr_uio.c (revision b2d48be1bc7df45ddd13b143a160d0acb5a383c5)
1 /*-
2  * Copyright (c) 1982, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Copyright (c) 2014 The FreeBSD Foundation
11  *
12  * Portions of this software were developed by Konstantin Belousov
13  * under sponsorship from the FreeBSD Foundation.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *	@(#)kern_subr.c	8.3 (Berkeley) 1/21/94
40  */
41 
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/limits.h>
49 #include <sys/lock.h>
50 #include <sys/mman.h>
51 #include <sys/proc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/rwlock.h>
54 #include <sys/sched.h>
55 #include <sys/sysctl.h>
56 #include <sys/vnode.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_param.h>
60 #include <vm/vm_extern.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pageout.h>
63 #include <vm/vm_map.h>
64 
65 SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, UIO_MAXIOV,
66 	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
67 
68 static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault);
69 
70 int
71 copyin_nofault(const void *udaddr, void *kaddr, size_t len)
72 {
73 	int error, save;
74 
75 	save = vm_fault_disable_pagefaults();
76 	error = copyin(udaddr, kaddr, len);
77 	vm_fault_enable_pagefaults(save);
78 	return (error);
79 }
80 
81 int
82 copyout_nofault(const void *kaddr, void *udaddr, size_t len)
83 {
84 	int error, save;
85 
86 	save = vm_fault_disable_pagefaults();
87 	error = copyout(kaddr, udaddr, len);
88 	vm_fault_enable_pagefaults(save);
89 	return (error);
90 }
91 
92 #define	PHYS_PAGE_COUNT(len)	(howmany(len, PAGE_SIZE) + 1)
93 
94 int
95 physcopyin(void *src, vm_paddr_t dst, size_t len)
96 {
97 	vm_page_t m[PHYS_PAGE_COUNT(len)];
98 	struct iovec iov[1];
99 	struct uio uio;
100 	int i;
101 
102 	iov[0].iov_base = src;
103 	iov[0].iov_len = len;
104 	uio.uio_iov = iov;
105 	uio.uio_iovcnt = 1;
106 	uio.uio_offset = 0;
107 	uio.uio_resid = len;
108 	uio.uio_segflg = UIO_SYSSPACE;
109 	uio.uio_rw = UIO_WRITE;
110 	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, dst += PAGE_SIZE)
111 		m[i] = PHYS_TO_VM_PAGE(dst);
112 	return (uiomove_fromphys(m, dst & PAGE_MASK, len, &uio));
113 }
114 
115 int
116 physcopyout(vm_paddr_t src, void *dst, size_t len)
117 {
118 	vm_page_t m[PHYS_PAGE_COUNT(len)];
119 	struct iovec iov[1];
120 	struct uio uio;
121 	int i;
122 
123 	iov[0].iov_base = dst;
124 	iov[0].iov_len = len;
125 	uio.uio_iov = iov;
126 	uio.uio_iovcnt = 1;
127 	uio.uio_offset = 0;
128 	uio.uio_resid = len;
129 	uio.uio_segflg = UIO_SYSSPACE;
130 	uio.uio_rw = UIO_READ;
131 	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, src += PAGE_SIZE)
132 		m[i] = PHYS_TO_VM_PAGE(src);
133 	return (uiomove_fromphys(m, src & PAGE_MASK, len, &uio));
134 }
135 
136 #undef PHYS_PAGE_COUNT
137 
138 int
139 uiomove(void *cp, int n, struct uio *uio)
140 {
141 
142 	return (uiomove_faultflag(cp, n, uio, 0));
143 }
144 
145 int
146 uiomove_nofault(void *cp, int n, struct uio *uio)
147 {
148 
149 	return (uiomove_faultflag(cp, n, uio, 1));
150 }
151 
152 static int
153 uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault)
154 {
155 	struct thread *td;
156 	struct iovec *iov;
157 	size_t cnt;
158 	int error, newflags, save;
159 
160 	td = curthread;
161 	error = 0;
162 
163 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
164 	    ("uiomove: mode"));
165 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == td,
166 	    ("uiomove proc"));
167 	if (!nofault)
168 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
169 		    "Calling uiomove()");
170 
171 	/* XXX does it make a sense to set TDP_DEADLKTREAT for UIO_SYSSPACE ? */
172 	newflags = TDP_DEADLKTREAT;
173 	if (uio->uio_segflg == UIO_USERSPACE && nofault) {
174 		/*
175 		 * Fail if a non-spurious page fault occurs.
176 		 */
177 		newflags |= TDP_NOFAULTING | TDP_RESETSPUR;
178 	}
179 	save = curthread_pflags_set(newflags);
180 
181 	while (n > 0 && uio->uio_resid) {
182 		iov = uio->uio_iov;
183 		cnt = iov->iov_len;
184 		if (cnt == 0) {
185 			uio->uio_iov++;
186 			uio->uio_iovcnt--;
187 			continue;
188 		}
189 		if (cnt > n)
190 			cnt = n;
191 
192 		switch (uio->uio_segflg) {
193 
194 		case UIO_USERSPACE:
195 			maybe_yield();
196 			if (uio->uio_rw == UIO_READ)
197 				error = copyout(cp, iov->iov_base, cnt);
198 			else
199 				error = copyin(iov->iov_base, cp, cnt);
200 			if (error)
201 				goto out;
202 			break;
203 
204 		case UIO_SYSSPACE:
205 			if (uio->uio_rw == UIO_READ)
206 				bcopy(cp, iov->iov_base, cnt);
207 			else
208 				bcopy(iov->iov_base, cp, cnt);
209 			break;
210 		case UIO_NOCOPY:
211 			break;
212 		}
213 		iov->iov_base = (char *)iov->iov_base + cnt;
214 		iov->iov_len -= cnt;
215 		uio->uio_resid -= cnt;
216 		uio->uio_offset += cnt;
217 		cp = (char *)cp + cnt;
218 		n -= cnt;
219 	}
220 out:
221 	curthread_pflags_restore(save);
222 	return (error);
223 }
224 
225 /*
226  * Wrapper for uiomove() that validates the arguments against a known-good
227  * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
228  * is almost definitely a bad thing, so we catch that here as well.  We
229  * return a runtime failure, but it might be desirable to generate a runtime
230  * assertion failure instead.
231  */
232 int
233 uiomove_frombuf(void *buf, int buflen, struct uio *uio)
234 {
235 	size_t offset, n;
236 
237 	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
238 	    (offset = uio->uio_offset) != uio->uio_offset)
239 		return (EINVAL);
240 	if (buflen <= 0 || offset >= buflen)
241 		return (0);
242 	if ((n = buflen - offset) > IOSIZE_MAX)
243 		return (EINVAL);
244 	return (uiomove((char *)buf + offset, n, uio));
245 }
246 
247 /*
248  * Give next character to user as result of read.
249  */
250 int
251 ureadc(int c, struct uio *uio)
252 {
253 	struct iovec *iov;
254 	char *iov_base;
255 
256 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
257 	    "Calling ureadc()");
258 
259 again:
260 	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
261 		panic("ureadc");
262 	iov = uio->uio_iov;
263 	if (iov->iov_len == 0) {
264 		uio->uio_iovcnt--;
265 		uio->uio_iov++;
266 		goto again;
267 	}
268 	switch (uio->uio_segflg) {
269 
270 	case UIO_USERSPACE:
271 		if (subyte(iov->iov_base, c) < 0)
272 			return (EFAULT);
273 		break;
274 
275 	case UIO_SYSSPACE:
276 		iov_base = iov->iov_base;
277 		*iov_base = c;
278 		break;
279 
280 	case UIO_NOCOPY:
281 		break;
282 	}
283 	iov->iov_base = (char *)iov->iov_base + 1;
284 	iov->iov_len--;
285 	uio->uio_resid--;
286 	uio->uio_offset++;
287 	return (0);
288 }
289 
290 int
291 copyinfrom(const void * __restrict src, void * __restrict dst, size_t len,
292     int seg)
293 {
294 	int error = 0;
295 
296 	switch (seg) {
297 	case UIO_USERSPACE:
298 		error = copyin(src, dst, len);
299 		break;
300 	case UIO_SYSSPACE:
301 		bcopy(src, dst, len);
302 		break;
303 	default:
304 		panic("copyinfrom: bad seg %d\n", seg);
305 	}
306 	return (error);
307 }
308 
309 int
310 copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len,
311     size_t * __restrict copied, int seg)
312 {
313 	int error = 0;
314 
315 	switch (seg) {
316 	case UIO_USERSPACE:
317 		error = copyinstr(src, dst, len, copied);
318 		break;
319 	case UIO_SYSSPACE:
320 		error = copystr(src, dst, len, copied);
321 		break;
322 	default:
323 		panic("copyinstrfrom: bad seg %d\n", seg);
324 	}
325 	return (error);
326 }
327 
328 int
329 copyiniov(const struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
330 {
331 	u_int iovlen;
332 
333 	*iov = NULL;
334 	if (iovcnt > UIO_MAXIOV)
335 		return (error);
336 	iovlen = iovcnt * sizeof (struct iovec);
337 	*iov = malloc(iovlen, M_IOV, M_WAITOK);
338 	error = copyin(iovp, *iov, iovlen);
339 	if (error) {
340 		free(*iov, M_IOV);
341 		*iov = NULL;
342 	}
343 	return (error);
344 }
345 
346 int
347 copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop)
348 {
349 	struct iovec *iov;
350 	struct uio *uio;
351 	u_int iovlen;
352 	int error, i;
353 
354 	*uiop = NULL;
355 	if (iovcnt > UIO_MAXIOV)
356 		return (EINVAL);
357 	iovlen = iovcnt * sizeof (struct iovec);
358 	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
359 	iov = (struct iovec *)(uio + 1);
360 	error = copyin(iovp, iov, iovlen);
361 	if (error) {
362 		free(uio, M_IOV);
363 		return (error);
364 	}
365 	uio->uio_iov = iov;
366 	uio->uio_iovcnt = iovcnt;
367 	uio->uio_segflg = UIO_USERSPACE;
368 	uio->uio_offset = -1;
369 	uio->uio_resid = 0;
370 	for (i = 0; i < iovcnt; i++) {
371 		if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) {
372 			free(uio, M_IOV);
373 			return (EINVAL);
374 		}
375 		uio->uio_resid += iov->iov_len;
376 		iov++;
377 	}
378 	*uiop = uio;
379 	return (0);
380 }
381 
382 struct uio *
383 cloneuio(struct uio *uiop)
384 {
385 	struct uio *uio;
386 	int iovlen;
387 
388 	iovlen = uiop->uio_iovcnt * sizeof (struct iovec);
389 	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
390 	*uio = *uiop;
391 	uio->uio_iov = (struct iovec *)(uio + 1);
392 	bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
393 	return (uio);
394 }
395 
396 /*
397  * Map some anonymous memory in user space of size sz, rounded up to the page
398  * boundary.
399  */
400 int
401 copyout_map(struct thread *td, vm_offset_t *addr, size_t sz)
402 {
403 	struct vmspace *vms;
404 	int error;
405 	vm_size_t size;
406 
407 	vms = td->td_proc->p_vmspace;
408 
409 	/*
410 	 * Map somewhere after heap in process memory.
411 	 */
412 	*addr = round_page((vm_offset_t)vms->vm_daddr +
413 	    lim_max(td, RLIMIT_DATA));
414 
415 	/* round size up to page boundry */
416 	size = (vm_size_t)round_page(sz);
417 
418 	error = vm_mmap(&vms->vm_map, addr, size, VM_PROT_READ | VM_PROT_WRITE,
419 	    VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, OBJT_DEFAULT, NULL, 0);
420 
421 	return (error);
422 }
423 
424 /*
425  * Unmap memory in user space.
426  */
427 int
428 copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz)
429 {
430 	vm_map_t map;
431 	vm_size_t size;
432 
433 	if (sz == 0)
434 		return (0);
435 
436 	map = &td->td_proc->p_vmspace->vm_map;
437 	size = (vm_size_t)round_page(sz);
438 
439 	if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS)
440 		return (EINVAL);
441 
442 	return (0);
443 }
444 
445 #ifdef NO_FUEWORD
446 /*
447  * XXXKIB The temporal implementation of fue*() functions which do not
448  * handle usermode -1 properly, mixing it with the fault code.  Keep
449  * this until MD code is written.  Currently sparc64, mips and arm do
450  * not have proper implementation.
451  */
452 
453 int
454 fueword(volatile const void *base, long *val)
455 {
456 	long res;
457 
458 	res = fuword(base);
459 	if (res == -1)
460 		return (-1);
461 	*val = res;
462 	return (0);
463 }
464 
465 int
466 fueword32(volatile const void *base, int32_t *val)
467 {
468 	int32_t res;
469 
470 	res = fuword32(base);
471 	if (res == -1)
472 		return (-1);
473 	*val = res;
474 	return (0);
475 }
476 
477 #ifdef _LP64
478 int
479 fueword64(volatile const void *base, int64_t *val)
480 {
481 	int32_t res;
482 
483 	res = fuword64(base);
484 	if (res == -1)
485 		return (-1);
486 	*val = res;
487 	return (0);
488 }
489 #endif
490 
491 int
492 casueword32(volatile uint32_t *base, uint32_t oldval, uint32_t *oldvalp,
493     uint32_t newval)
494 {
495 	int32_t ov;
496 
497 	ov = casuword32(base, oldval, newval);
498 	if (ov == -1)
499 		return (-1);
500 	*oldvalp = ov;
501 	return (0);
502 }
503 
504 int
505 casueword(volatile u_long *p, u_long oldval, u_long *oldvalp, u_long newval)
506 {
507 	u_long ov;
508 
509 	ov = casuword(p, oldval, newval);
510 	if (ov == -1)
511 		return (-1);
512 	*oldvalp = ov;
513 	return (0);
514 }
515 #else /* NO_FUEWORD */
516 int32_t
517 fuword32(volatile const void *addr)
518 {
519 	int rv;
520 	int32_t val;
521 
522 	rv = fueword32(addr, &val);
523 	return (rv == -1 ? -1 : val);
524 }
525 
526 #ifdef _LP64
527 int64_t
528 fuword64(volatile const void *addr)
529 {
530 	int rv;
531 	int64_t val;
532 
533 	rv = fueword64(addr, &val);
534 	return (rv == -1 ? -1 : val);
535 }
536 #endif /* _LP64 */
537 
538 long
539 fuword(volatile const void *addr)
540 {
541 	long val;
542 	int rv;
543 
544 	rv = fueword(addr, &val);
545 	return (rv == -1 ? -1 : val);
546 }
547 
548 uint32_t
549 casuword32(volatile uint32_t *addr, uint32_t old, uint32_t new)
550 {
551 	int rv;
552 	uint32_t val;
553 
554 	rv = casueword32(addr, old, &val, new);
555 	return (rv == -1 ? -1 : val);
556 }
557 
558 u_long
559 casuword(volatile u_long *addr, u_long old, u_long new)
560 {
561 	int rv;
562 	u_long val;
563 
564 	rv = casueword(addr, old, &val, new);
565 	return (rv == -1 ? -1 : val);
566 }
567 
568 #endif /* NO_FUEWORD */
569