xref: /freebsd/sys/kern/subr_uio.c (revision 7c43148a974877188a930e4078a164f83da8e652)
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 
42 #include <sys/cdefs.h>
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/limits.h>
47 #include <sys/lock.h>
48 #include <sys/mman.h>
49 #include <sys/proc.h>
50 #include <sys/resourcevar.h>
51 #include <sys/rwlock.h>
52 #include <sys/sched.h>
53 #include <sys/sysctl.h>
54 #include <sys/vnode.h>
55 
56 #include <vm/vm.h>
57 #include <vm/vm_param.h>
58 #include <vm/vm_extern.h>
59 #include <vm/vm_page.h>
60 #include <vm/vm_pageout.h>
61 #include <vm/vm_map.h>
62 
63 #include <machine/bus.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 physcopyin_vlist(bus_dma_segment_t *src, off_t offset, vm_paddr_t dst,
140     size_t len)
141 {
142 	size_t seg_len;
143 	int error;
144 
145 	error = 0;
146 	while (offset >= src->ds_len) {
147 		offset -= src->ds_len;
148 		src++;
149 	}
150 
151 	while (len > 0 && error == 0) {
152 		seg_len = MIN(src->ds_len - offset, len);
153 		error = physcopyin((void *)(uintptr_t)(src->ds_addr + offset),
154 		    dst, seg_len);
155 		offset = 0;
156 		src++;
157 		len -= seg_len;
158 		dst += seg_len;
159 	}
160 
161 	return (error);
162 }
163 
164 int
165 physcopyout_vlist(vm_paddr_t src, bus_dma_segment_t *dst, off_t offset,
166     size_t len)
167 {
168 	size_t seg_len;
169 	int error;
170 
171 	error = 0;
172 	while (offset >= dst->ds_len) {
173 		offset -= dst->ds_len;
174 		dst++;
175 	}
176 
177 	while (len > 0 && error == 0) {
178 		seg_len = MIN(dst->ds_len - offset, len);
179 		error = physcopyout(src, (void *)(uintptr_t)(dst->ds_addr +
180 		    offset), seg_len);
181 		offset = 0;
182 		dst++;
183 		len -= seg_len;
184 		src += seg_len;
185 	}
186 
187 	return (error);
188 }
189 
190 int
191 uiomove(void *cp, int n, struct uio *uio)
192 {
193 
194 	return (uiomove_faultflag(cp, n, uio, 0));
195 }
196 
197 int
198 uiomove_nofault(void *cp, int n, struct uio *uio)
199 {
200 
201 	return (uiomove_faultflag(cp, n, uio, 1));
202 }
203 
204 static int
205 uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault)
206 {
207 	struct iovec *iov;
208 	size_t cnt;
209 	int error, newflags, save;
210 
211 	save = error = 0;
212 
213 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
214 	    ("uiomove: mode"));
215 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
216 	    ("uiomove proc"));
217 	KASSERT(uio->uio_resid >= 0,
218 	    ("%s: uio %p resid underflow", __func__, uio));
219 
220 	if (uio->uio_segflg == UIO_USERSPACE) {
221 		newflags = TDP_DEADLKTREAT;
222 		if (nofault) {
223 			/*
224 			 * Fail if a non-spurious page fault occurs.
225 			 */
226 			newflags |= TDP_NOFAULTING | TDP_RESETSPUR;
227 		} else {
228 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
229 			    "Calling uiomove()");
230 		}
231 		save = curthread_pflags_set(newflags);
232 	} else {
233 		KASSERT(nofault == 0, ("uiomove: nofault"));
234 	}
235 
236 	while (n > 0 && uio->uio_resid) {
237 		KASSERT(uio->uio_iovcnt > 0,
238 		    ("%s: uio %p iovcnt underflow", __func__, uio));
239 
240 		iov = uio->uio_iov;
241 		cnt = iov->iov_len;
242 		if (cnt == 0) {
243 			uio->uio_iov++;
244 			uio->uio_iovcnt--;
245 			continue;
246 		}
247 		if (cnt > n)
248 			cnt = n;
249 
250 		switch (uio->uio_segflg) {
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 	case UIO_USERSPACE:
328 		if (subyte(iov->iov_base, c) < 0)
329 			return (EFAULT);
330 		break;
331 
332 	case UIO_SYSSPACE:
333 		iov_base = iov->iov_base;
334 		*iov_base = c;
335 		break;
336 
337 	case UIO_NOCOPY:
338 		break;
339 	}
340 	iov->iov_base = (char *)iov->iov_base + 1;
341 	iov->iov_len--;
342 	uio->uio_resid--;
343 	uio->uio_offset++;
344 	return (0);
345 }
346 
347 int
348 copyiniov(const struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
349 {
350 	u_int iovlen;
351 
352 	*iov = NULL;
353 	if (iovcnt > UIO_MAXIOV)
354 		return (error);
355 	iovlen = iovcnt * sizeof (struct iovec);
356 	*iov = malloc(iovlen, M_IOV, M_WAITOK);
357 	error = copyin(iovp, *iov, iovlen);
358 	if (error) {
359 		free(*iov, M_IOV);
360 		*iov = NULL;
361 	}
362 	return (error);
363 }
364 
365 int
366 copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop)
367 {
368 	struct iovec *iov;
369 	struct uio *uio;
370 	u_int iovlen;
371 	int error, i;
372 
373 	*uiop = NULL;
374 	if (iovcnt > UIO_MAXIOV)
375 		return (EINVAL);
376 	iovlen = iovcnt * sizeof (struct iovec);
377 	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
378 	iov = (struct iovec *)(uio + 1);
379 	error = copyin(iovp, iov, iovlen);
380 	if (error) {
381 		free(uio, M_IOV);
382 		return (error);
383 	}
384 	uio->uio_iov = iov;
385 	uio->uio_iovcnt = iovcnt;
386 	uio->uio_segflg = UIO_USERSPACE;
387 	uio->uio_offset = -1;
388 	uio->uio_resid = 0;
389 	for (i = 0; i < iovcnt; i++) {
390 		if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) {
391 			free(uio, M_IOV);
392 			return (EINVAL);
393 		}
394 		uio->uio_resid += iov->iov_len;
395 		iov++;
396 	}
397 	*uiop = uio;
398 	return (0);
399 }
400 
401 struct uio *
402 cloneuio(struct uio *uiop)
403 {
404 	struct uio *uio;
405 	int iovlen;
406 
407 	iovlen = uiop->uio_iovcnt * sizeof (struct iovec);
408 	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
409 	*uio = *uiop;
410 	uio->uio_iov = (struct iovec *)(uio + 1);
411 	bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
412 	return (uio);
413 }
414 
415 /*
416  * Map some anonymous memory in user space of size sz, rounded up to the page
417  * boundary.
418  */
419 int
420 copyout_map(struct thread *td, vm_offset_t *addr, size_t sz)
421 {
422 	struct vmspace *vms;
423 	int error;
424 	vm_size_t size;
425 
426 	vms = td->td_proc->p_vmspace;
427 
428 	/*
429 	 * Map somewhere after heap in process memory.
430 	 */
431 	*addr = round_page((vm_offset_t)vms->vm_daddr +
432 	    lim_max(td, RLIMIT_DATA));
433 
434 	/* round size up to page boundary */
435 	size = (vm_size_t)round_page(sz);
436 	if (size == 0)
437 		return (EINVAL);
438 	error = vm_mmap_object(&vms->vm_map, addr, size, VM_PROT_READ |
439 	    VM_PROT_WRITE, VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, NULL, 0,
440 	    FALSE, td);
441 	return (error);
442 }
443 
444 /*
445  * Unmap memory in user space.
446  */
447 int
448 copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz)
449 {
450 	vm_map_t map;
451 	vm_size_t size;
452 
453 	if (sz == 0)
454 		return (0);
455 
456 	map = &td->td_proc->p_vmspace->vm_map;
457 	size = (vm_size_t)round_page(sz);
458 
459 	if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS)
460 		return (EINVAL);
461 
462 	return (0);
463 }
464 
465 int32_t
466 fuword32(volatile const void *addr)
467 {
468 	int rv;
469 	int32_t val;
470 
471 	rv = fueword32(addr, &val);
472 	return (rv == -1 ? -1 : val);
473 }
474 
475 #ifdef _LP64
476 int64_t
477 fuword64(volatile const void *addr)
478 {
479 	int rv;
480 	int64_t val;
481 
482 	rv = fueword64(addr, &val);
483 	return (rv == -1 ? -1 : val);
484 }
485 #endif /* _LP64 */
486 
487 long
488 fuword(volatile const void *addr)
489 {
490 	long val;
491 	int rv;
492 
493 	rv = fueword(addr, &val);
494 	return (rv == -1 ? -1 : val);
495 }
496 
497 uint32_t
498 casuword32(volatile uint32_t *addr, uint32_t old, uint32_t new)
499 {
500 	int rv;
501 	uint32_t val;
502 
503 	rv = casueword32(addr, old, &val, new);
504 	return (rv == -1 ? -1 : val);
505 }
506 
507 u_long
508 casuword(volatile u_long *addr, u_long old, u_long new)
509 {
510 	int rv;
511 	u_long val;
512 
513 	rv = casueword(addr, old, &val, new);
514 	return (rv == -1 ? -1 : val);
515 }
516