xref: /freebsd/sys/kern/subr_uio.c (revision 190cef3d52236565eb22e18b33e9e865ec634aa3) 
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Copyright (c) 2014 The FreeBSD Foundation
13  *
14  * Portions of this software were developed by Konstantin Belousov
15  * under sponsorship from the FreeBSD Foundation.
16  *
17  * Redistribution and use in source and binary forms, with or without
18  * modification, are permitted provided that the following conditions
19  * are met:
20  * 1. Redistributions of source code must retain the above copyright
21  *    notice, this list of conditions and the following disclaimer.
22  * 2. Redistributions in binary form must reproduce the above copyright
23  *    notice, this list of conditions and the following disclaimer in the
24  *    documentation and/or other materials provided with the distribution.
25  * 3. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)kern_subr.c	8.3 (Berkeley) 1/21/94
42  */
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/limits.h>
51 #include <sys/lock.h>
52 #include <sys/mman.h>
53 #include <sys/proc.h>
54 #include <sys/resourcevar.h>
55 #include <sys/rwlock.h>
56 #include <sys/sched.h>
57 #include <sys/sysctl.h>
58 #include <sys/vnode.h>
59 
60 #include <vm/vm.h>
61 #include <vm/vm_param.h>
62 #include <vm/vm_extern.h>
63 #include <vm/vm_page.h>
64 #include <vm/vm_pageout.h>
65 #include <vm/vm_map.h>
66 
67 #include <machine/bus.h>
68 
69 SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, UIO_MAXIOV,
70 	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
71 
72 static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault);
73 
74 int
75 copyin_nofault(const void *udaddr, void *kaddr, size_t len)
76 {
77 	int error, save;
78 
79 	save = vm_fault_disable_pagefaults();
80 	error = copyin(udaddr, kaddr, len);
81 	vm_fault_enable_pagefaults(save);
82 	return (error);
83 }
84 
85 int
86 copyout_nofault(const void *kaddr, void *udaddr, size_t len)
87 {
88 	int error, save;
89 
90 	save = vm_fault_disable_pagefaults();
91 	error = copyout(kaddr, udaddr, len);
92 	vm_fault_enable_pagefaults(save);
93 	return (error);
94 }
95 
96 #define	PHYS_PAGE_COUNT(len)	(howmany(len, PAGE_SIZE) + 1)
97 
98 int
99 physcopyin(void *src, vm_paddr_t dst, size_t len)
100 {
101 	vm_page_t m[PHYS_PAGE_COUNT(len)];
102 	struct iovec iov[1];
103 	struct uio uio;
104 	int i;
105 
106 	iov[0].iov_base = src;
107 	iov[0].iov_len = len;
108 	uio.uio_iov = iov;
109 	uio.uio_iovcnt = 1;
110 	uio.uio_offset = 0;
111 	uio.uio_resid = len;
112 	uio.uio_segflg = UIO_SYSSPACE;
113 	uio.uio_rw = UIO_WRITE;
114 	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, dst += PAGE_SIZE)
115 		m[i] = PHYS_TO_VM_PAGE(dst);
116 	return (uiomove_fromphys(m, dst & PAGE_MASK, len, &uio));
117 }
118 
119 int
120 physcopyout(vm_paddr_t src, void *dst, size_t len)
121 {
122 	vm_page_t m[PHYS_PAGE_COUNT(len)];
123 	struct iovec iov[1];
124 	struct uio uio;
125 	int i;
126 
127 	iov[0].iov_base = dst;
128 	iov[0].iov_len = len;
129 	uio.uio_iov = iov;
130 	uio.uio_iovcnt = 1;
131 	uio.uio_offset = 0;
132 	uio.uio_resid = len;
133 	uio.uio_segflg = UIO_SYSSPACE;
134 	uio.uio_rw = UIO_READ;
135 	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, src += PAGE_SIZE)
136 		m[i] = PHYS_TO_VM_PAGE(src);
137 	return (uiomove_fromphys(m, src & PAGE_MASK, len, &uio));
138 }
139 
140 #undef PHYS_PAGE_COUNT
141 
142 int
143 physcopyin_vlist(bus_dma_segment_t *src, off_t offset, vm_paddr_t dst,
144     size_t len)
145 {
146 	size_t seg_len;
147 	int error;
148 
149 	error = 0;
150 	while (offset >= src->ds_len) {
151 		offset -= src->ds_len;
152 		src++;
153 	}
154 
155 	while (len > 0 && error == 0) {
156 		seg_len = MIN(src->ds_len - offset, len);
157 		error = physcopyin((void *)(uintptr_t)(src->ds_addr + offset),
158 		    dst, seg_len);
159 		offset = 0;
160 		src++;
161 		len -= seg_len;
162 		dst += seg_len;
163 	}
164 
165 	return (error);
166 }
167 
168 int
169 physcopyout_vlist(vm_paddr_t src, bus_dma_segment_t *dst, off_t offset,
170     size_t len)
171 {
172 	size_t seg_len;
173 	int error;
174 
175 	error = 0;
176 	while (offset >= dst->ds_len) {
177 		offset -= dst->ds_len;
178 		dst++;
179 	}
180 
181 	while (len > 0 && error == 0) {
182 		seg_len = MIN(dst->ds_len - offset, len);
183 		error = physcopyout(src, (void *)(uintptr_t)(dst->ds_addr +
184 		    offset), seg_len);
185 		offset = 0;
186 		dst++;
187 		len -= seg_len;
188 		src += seg_len;
189 	}
190 
191 	return (error);
192 }
193 
194 int
195 uiomove(void *cp, int n, struct uio *uio)
196 {
197 
198 	return (uiomove_faultflag(cp, n, uio, 0));
199 }
200 
201 int
202 uiomove_nofault(void *cp, int n, struct uio *uio)
203 {
204 
205 	return (uiomove_faultflag(cp, n, uio, 1));
206 }
207 
208 static int
209 uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault)
210 {
211 	struct iovec *iov;
212 	size_t cnt;
213 	int error, newflags, save;
214 
215 	save = error = 0;
216 
217 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
218 	    ("uiomove: mode"));
219 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
220 	    ("uiomove proc"));
221 
222 	if (uio->uio_segflg == UIO_USERSPACE) {
223 		newflags = TDP_DEADLKTREAT;
224 		if (nofault) {
225 			/*
226 			 * Fail if a non-spurious page fault occurs.
227 			 */
228 			newflags |= TDP_NOFAULTING | TDP_RESETSPUR;
229 		} else {
230 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
231 			    "Calling uiomove()");
232 		}
233 		save = curthread_pflags_set(newflags);
234 	} else {
235 		KASSERT(nofault == 0, ("uiomove: nofault"));
236 	}
237 
238 	while (n > 0 && uio->uio_resid) {
239 		iov = uio->uio_iov;
240 		cnt = iov->iov_len;
241 		if (cnt == 0) {
242 			uio->uio_iov++;
243 			uio->uio_iovcnt--;
244 			continue;
245 		}
246 		if (cnt > n)
247 			cnt = n;
248 
249 		switch (uio->uio_segflg) {
250 
251 		case UIO_USERSPACE:
252 			maybe_yield();
253 			if (uio->uio_rw == UIO_READ)
254 				error = copyout(cp, iov->iov_base, cnt);
255 			else
256 				error = copyin(iov->iov_base, cp, cnt);
257 			if (error)
258 				goto out;
259 			break;
260 
261 		case UIO_SYSSPACE:
262 			if (uio->uio_rw == UIO_READ)
263 				bcopy(cp, iov->iov_base, cnt);
264 			else
265 				bcopy(iov->iov_base, cp, cnt);
266 			break;
267 		case UIO_NOCOPY:
268 			break;
269 		}
270 		iov->iov_base = (char *)iov->iov_base + cnt;
271 		iov->iov_len -= cnt;
272 		uio->uio_resid -= cnt;
273 		uio->uio_offset += cnt;
274 		cp = (char *)cp + cnt;
275 		n -= cnt;
276 	}
277 out:
278 	if (save)
279 		curthread_pflags_restore(save);
280 	return (error);
281 }
282 
283 /*
284  * Wrapper for uiomove() that validates the arguments against a known-good
285  * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
286  * is almost definitely a bad thing, so we catch that here as well.  We
287  * return a runtime failure, but it might be desirable to generate a runtime
288  * assertion failure instead.
289  */
290 int
291 uiomove_frombuf(void *buf, int buflen, struct uio *uio)
292 {
293 	size_t offset, n;
294 
295 	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
296 	    (offset = uio->uio_offset) != uio->uio_offset)
297 		return (EINVAL);
298 	if (buflen <= 0 || offset >= buflen)
299 		return (0);
300 	if ((n = buflen - offset) > IOSIZE_MAX)
301 		return (EINVAL);
302 	return (uiomove((char *)buf + offset, n, uio));
303 }
304 
305 /*
306  * Give next character to user as result of read.
307  */
308 int
309 ureadc(int c, struct uio *uio)
310 {
311 	struct iovec *iov;
312 	char *iov_base;
313 
314 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
315 	    "Calling ureadc()");
316 
317 again:
318 	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
319 		panic("ureadc");
320 	iov = uio->uio_iov;
321 	if (iov->iov_len == 0) {
322 		uio->uio_iovcnt--;
323 		uio->uio_iov++;
324 		goto again;
325 	}
326 	switch (uio->uio_segflg) {
327 
328 	case UIO_USERSPACE:
329 		if (subyte(iov->iov_base, c) < 0)
330 			return (EFAULT);
331 		break;
332 
333 	case UIO_SYSSPACE:
334 		iov_base = iov->iov_base;
335 		*iov_base = c;
336 		break;
337 
338 	case UIO_NOCOPY:
339 		break;
340 	}
341 	iov->iov_base = (char *)iov->iov_base + 1;
342 	iov->iov_len--;
343 	uio->uio_resid--;
344 	uio->uio_offset++;
345 	return (0);
346 }
347 
348 int
349 copyinfrom(const void * __restrict src, void * __restrict dst, size_t len,
350     int seg)
351 {
352 	int error = 0;
353 
354 	switch (seg) {
355 	case UIO_USERSPACE:
356 		error = copyin(src, dst, len);
357 		break;
358 	case UIO_SYSSPACE:
359 		bcopy(src, dst, len);
360 		break;
361 	default:
362 		panic("copyinfrom: bad seg %d\n", seg);
363 	}
364 	return (error);
365 }
366 
367 int
368 copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len,
369     size_t * __restrict copied, int seg)
370 {
371 	int error = 0;
372 
373 	switch (seg) {
374 	case UIO_USERSPACE:
375 		error = copyinstr(src, dst, len, copied);
376 		break;
377 	case UIO_SYSSPACE:
378 		error = copystr(src, dst, len, copied);
379 		break;
380 	default:
381 		panic("copyinstrfrom: bad seg %d\n", seg);
382 	}
383 	return (error);
384 }
385 
386 int
387 copyiniov(const struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
388 {
389 	u_int iovlen;
390 
391 	*iov = NULL;
392 	if (iovcnt > UIO_MAXIOV)
393 		return (error);
394 	iovlen = iovcnt * sizeof (struct iovec);
395 	*iov = malloc(iovlen, M_IOV, M_WAITOK);
396 	error = copyin(iovp, *iov, iovlen);
397 	if (error) {
398 		free(*iov, M_IOV);
399 		*iov = NULL;
400 	}
401 	return (error);
402 }
403 
404 int
405 copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop)
406 {
407 	struct iovec *iov;
408 	struct uio *uio;
409 	u_int iovlen;
410 	int error, i;
411 
412 	*uiop = NULL;
413 	if (iovcnt > UIO_MAXIOV)
414 		return (EINVAL);
415 	iovlen = iovcnt * sizeof (struct iovec);
416 	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
417 	iov = (struct iovec *)(uio + 1);
418 	error = copyin(iovp, iov, iovlen);
419 	if (error) {
420 		free(uio, M_IOV);
421 		return (error);
422 	}
423 	uio->uio_iov = iov;
424 	uio->uio_iovcnt = iovcnt;
425 	uio->uio_segflg = UIO_USERSPACE;
426 	uio->uio_offset = -1;
427 	uio->uio_resid = 0;
428 	for (i = 0; i < iovcnt; i++) {
429 		if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) {
430 			free(uio, M_IOV);
431 			return (EINVAL);
432 		}
433 		uio->uio_resid += iov->iov_len;
434 		iov++;
435 	}
436 	*uiop = uio;
437 	return (0);
438 }
439 
440 struct uio *
441 cloneuio(struct uio *uiop)
442 {
443 	struct uio *uio;
444 	int iovlen;
445 
446 	iovlen = uiop->uio_iovcnt * sizeof (struct iovec);
447 	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
448 	*uio = *uiop;
449 	uio->uio_iov = (struct iovec *)(uio + 1);
450 	bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
451 	return (uio);
452 }
453 
454 /*
455  * Map some anonymous memory in user space of size sz, rounded up to the page
456  * boundary.
457  */
458 int
459 copyout_map(struct thread *td, vm_offset_t *addr, size_t sz)
460 {
461 	struct vmspace *vms;
462 	int error;
463 	vm_size_t size;
464 
465 	vms = td->td_proc->p_vmspace;
466 
467 	/*
468 	 * Map somewhere after heap in process memory.
469 	 */
470 	*addr = round_page((vm_offset_t)vms->vm_daddr +
471 	    lim_max(td, RLIMIT_DATA));
472 
473 	/* round size up to page boundary */
474 	size = (vm_size_t)round_page(sz);
475 	if (size == 0)
476 		return (EINVAL);
477 	error = vm_mmap_object(&vms->vm_map, addr, size, VM_PROT_READ |
478 	    VM_PROT_WRITE, VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, NULL, 0,
479 	    FALSE, td);
480 	return (error);
481 }
482 
483 /*
484  * Unmap memory in user space.
485  */
486 int
487 copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz)
488 {
489 	vm_map_t map;
490 	vm_size_t size;
491 
492 	if (sz == 0)
493 		return (0);
494 
495 	map = &td->td_proc->p_vmspace->vm_map;
496 	size = (vm_size_t)round_page(sz);
497 
498 	if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS)
499 		return (EINVAL);
500 
501 	return (0);
502 }
503 
504 #ifdef NO_FUEWORD
505 /*
506  * XXXKIB The temporal implementation of fue*() functions which do not
507  * handle usermode -1 properly, mixing it with the fault code.  Keep
508  * this until MD code is written.  Currently sparc64 does not have a
509  * proper implementation.
510  */
511 
512 int
513 fueword(volatile const void *base, long *val)
514 {
515 	long res;
516 
517 	res = fuword(base);
518 	if (res == -1)
519 		return (-1);
520 	*val = res;
521 	return (0);
522 }
523 
524 int
525 fueword32(volatile const void *base, int32_t *val)
526 {
527 	int32_t res;
528 
529 	res = fuword32(base);
530 	if (res == -1)
531 		return (-1);
532 	*val = res;
533 	return (0);
534 }
535 
536 #ifdef _LP64
537 int
538 fueword64(volatile const void *base, int64_t *val)
539 {
540 	int64_t res;
541 
542 	res = fuword64(base);
543 	if (res == -1)
544 		return (-1);
545 	*val = res;
546 	return (0);
547 }
548 #endif
549 
550 int
551 casueword32(volatile uint32_t *base, uint32_t oldval, uint32_t *oldvalp,
552     uint32_t newval)
553 {
554 	int32_t ov;
555 
556 	ov = casuword32(base, oldval, newval);
557 	if (ov == -1)
558 		return (-1);
559 	*oldvalp = ov;
560 	return (0);
561 }
562 
563 int
564 casueword(volatile u_long *p, u_long oldval, u_long *oldvalp, u_long newval)
565 {
566 	u_long ov;
567 
568 	ov = casuword(p, oldval, newval);
569 	if (ov == -1)
570 		return (-1);
571 	*oldvalp = ov;
572 	return (0);
573 }
574 #else /* NO_FUEWORD */
575 int32_t
576 fuword32(volatile const void *addr)
577 {
578 	int rv;
579 	int32_t val;
580 
581 	rv = fueword32(addr, &val);
582 	return (rv == -1 ? -1 : val);
583 }
584 
585 #ifdef _LP64
586 int64_t
587 fuword64(volatile const void *addr)
588 {
589 	int rv;
590 	int64_t val;
591 
592 	rv = fueword64(addr, &val);
593 	return (rv == -1 ? -1 : val);
594 }
595 #endif /* _LP64 */
596 
597 long
598 fuword(volatile const void *addr)
599 {
600 	long val;
601 	int rv;
602 
603 	rv = fueword(addr, &val);
604 	return (rv == -1 ? -1 : val);
605 }
606 
607 uint32_t
608 casuword32(volatile uint32_t *addr, uint32_t old, uint32_t new)
609 {
610 	int rv;
611 	uint32_t val;
612 
613 	rv = casueword32(addr, old, &val, new);
614 	return (rv == -1 ? -1 : val);
615 }
616 
617 u_long
618 casuword(volatile u_long *addr, u_long old, u_long new)
619 {
620 	int rv;
621 	u_long val;
622 
623 	rv = casueword(addr, old, &val, new);
624 	return (rv == -1 ? -1 : val);
625 }
626 
627 #endif /* NO_FUEWORD */
628