xref: /freebsd/sys/kern/subr_uio.c (revision 473c90ac04cec0abbb414978c53e9c259c9129e8)
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/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/mman.h>
48 #include <sys/proc.h>
49 #include <sys/resourcevar.h>
50 #include <sys/rwlock.h>
51 #include <sys/sched.h>
52 #include <sys/sysctl.h>
53 #include <sys/vnode.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_param.h>
57 #include <vm/vm_extern.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pageout.h>
60 #include <vm/vm_map.h>
61 
62 #include <machine/bus.h>
63 
64 SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, UIO_MAXIOV,
65 	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
66 
67 static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault);
68 
69 int
copyin_nofault(const void * udaddr,void * kaddr,size_t len)70 copyin_nofault(const void *udaddr, void *kaddr, size_t len)
71 {
72 	int error, save;
73 
74 	save = vm_fault_disable_pagefaults();
75 	error = copyin(udaddr, kaddr, len);
76 	vm_fault_enable_pagefaults(save);
77 	return (error);
78 }
79 
80 int
copyout_nofault(const void * kaddr,void * udaddr,size_t len)81 copyout_nofault(const void *kaddr, void *udaddr, size_t len)
82 {
83 	int error, save;
84 
85 	save = vm_fault_disable_pagefaults();
86 	error = copyout(kaddr, udaddr, len);
87 	vm_fault_enable_pagefaults(save);
88 	return (error);
89 }
90 
91 #define	PHYS_PAGE_COUNT(len)	(howmany(len, PAGE_SIZE) + 1)
92 
93 int
physcopyin(void * src,vm_paddr_t dst,size_t len)94 physcopyin(void *src, vm_paddr_t dst, size_t len)
95 {
96 	vm_page_t m[PHYS_PAGE_COUNT(len)];
97 	struct iovec iov[1];
98 	struct uio uio;
99 	int i;
100 
101 	iov[0].iov_base = src;
102 	iov[0].iov_len = len;
103 	uio.uio_iov = iov;
104 	uio.uio_iovcnt = 1;
105 	uio.uio_offset = 0;
106 	uio.uio_resid = len;
107 	uio.uio_segflg = UIO_SYSSPACE;
108 	uio.uio_rw = UIO_WRITE;
109 	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, dst += PAGE_SIZE)
110 		m[i] = PHYS_TO_VM_PAGE(dst);
111 	return (uiomove_fromphys(m, dst & PAGE_MASK, len, &uio));
112 }
113 
114 int
physcopyout(vm_paddr_t src,void * dst,size_t len)115 physcopyout(vm_paddr_t src, void *dst, size_t len)
116 {
117 	vm_page_t m[PHYS_PAGE_COUNT(len)];
118 	struct iovec iov[1];
119 	struct uio uio;
120 	int i;
121 
122 	iov[0].iov_base = dst;
123 	iov[0].iov_len = len;
124 	uio.uio_iov = iov;
125 	uio.uio_iovcnt = 1;
126 	uio.uio_offset = 0;
127 	uio.uio_resid = len;
128 	uio.uio_segflg = UIO_SYSSPACE;
129 	uio.uio_rw = UIO_READ;
130 	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, src += PAGE_SIZE)
131 		m[i] = PHYS_TO_VM_PAGE(src);
132 	return (uiomove_fromphys(m, src & PAGE_MASK, len, &uio));
133 }
134 
135 #undef PHYS_PAGE_COUNT
136 
137 int
physcopyin_vlist(bus_dma_segment_t * src,off_t offset,vm_paddr_t dst,size_t len)138 physcopyin_vlist(bus_dma_segment_t *src, off_t offset, vm_paddr_t dst,
139     size_t len)
140 {
141 	size_t seg_len;
142 	int error;
143 
144 	error = 0;
145 	while (offset >= src->ds_len) {
146 		offset -= src->ds_len;
147 		src++;
148 	}
149 
150 	while (len > 0 && error == 0) {
151 		seg_len = MIN(src->ds_len - offset, len);
152 		error = physcopyin((void *)(uintptr_t)(src->ds_addr + offset),
153 		    dst, seg_len);
154 		offset = 0;
155 		src++;
156 		len -= seg_len;
157 		dst += seg_len;
158 	}
159 
160 	return (error);
161 }
162 
163 int
physcopyout_vlist(vm_paddr_t src,bus_dma_segment_t * dst,off_t offset,size_t len)164 physcopyout_vlist(vm_paddr_t src, bus_dma_segment_t *dst, off_t offset,
165     size_t len)
166 {
167 	size_t seg_len;
168 	int error;
169 
170 	error = 0;
171 	while (offset >= dst->ds_len) {
172 		offset -= dst->ds_len;
173 		dst++;
174 	}
175 
176 	while (len > 0 && error == 0) {
177 		seg_len = MIN(dst->ds_len - offset, len);
178 		error = physcopyout(src, (void *)(uintptr_t)(dst->ds_addr +
179 		    offset), seg_len);
180 		offset = 0;
181 		dst++;
182 		len -= seg_len;
183 		src += seg_len;
184 	}
185 
186 	return (error);
187 }
188 
189 int
uiomove(void * cp,int n,struct uio * uio)190 uiomove(void *cp, int n, struct uio *uio)
191 {
192 
193 	return (uiomove_faultflag(cp, n, uio, 0));
194 }
195 
196 int
uiomove_nofault(void * cp,int n,struct uio * uio)197 uiomove_nofault(void *cp, int n, struct uio *uio)
198 {
199 
200 	return (uiomove_faultflag(cp, n, uio, 1));
201 }
202 
203 static int
uiomove_faultflag(void * cp,int n,struct uio * uio,int nofault)204 uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault)
205 {
206 	struct iovec *iov;
207 	size_t cnt;
208 	int error, newflags, save;
209 
210 	save = error = 0;
211 
212 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
213 	    ("uiomove: mode"));
214 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
215 	    ("uiomove proc"));
216 	KASSERT(uio->uio_resid >= 0,
217 	    ("%s: uio %p resid underflow", __func__, uio));
218 
219 	if (uio->uio_segflg == UIO_USERSPACE) {
220 		newflags = TDP_DEADLKTREAT;
221 		if (nofault) {
222 			/*
223 			 * Fail if a non-spurious page fault occurs.
224 			 */
225 			newflags |= TDP_NOFAULTING | TDP_RESETSPUR;
226 		} else {
227 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
228 			    "Calling uiomove()");
229 		}
230 		save = curthread_pflags_set(newflags);
231 	} else {
232 		KASSERT(nofault == 0, ("uiomove: nofault"));
233 	}
234 
235 	while (n > 0 && uio->uio_resid) {
236 		KASSERT(uio->uio_iovcnt > 0,
237 		    ("%s: uio %p iovcnt underflow", __func__, uio));
238 
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 		case UIO_USERSPACE:
251 			maybe_yield();
252 			switch (uio->uio_rw) {
253 			case UIO_READ:
254 				error = copyout(cp, iov->iov_base, cnt);
255 				break;
256 			case UIO_WRITE:
257 				error = copyin(iov->iov_base, cp, cnt);
258 				break;
259 			}
260 			if (error)
261 				goto out;
262 			break;
263 
264 		case UIO_SYSSPACE:
265 			switch (uio->uio_rw) {
266 			case UIO_READ:
267 				bcopy(cp, iov->iov_base, cnt);
268 				break;
269 			case UIO_WRITE:
270 				bcopy(iov->iov_base, cp, cnt);
271 				break;
272 			}
273 			break;
274 		case UIO_NOCOPY:
275 			break;
276 		}
277 		iov->iov_base = (char *)iov->iov_base + cnt;
278 		iov->iov_len -= cnt;
279 		uio->uio_resid -= cnt;
280 		uio->uio_offset += cnt;
281 		cp = (char *)cp + cnt;
282 		n -= cnt;
283 	}
284 out:
285 	if (save)
286 		curthread_pflags_restore(save);
287 	return (error);
288 }
289 
290 /*
291  * Wrapper for uiomove() that validates the arguments against a known-good
292  * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
293  * is almost definitely a bad thing, so we catch that here as well.  We
294  * return a runtime failure, but it might be desirable to generate a runtime
295  * assertion failure instead.
296  */
297 int
uiomove_frombuf(void * buf,int buflen,struct uio * uio)298 uiomove_frombuf(void *buf, int buflen, struct uio *uio)
299 {
300 	size_t offset, n;
301 
302 	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
303 	    (offset = uio->uio_offset) != uio->uio_offset)
304 		return (EINVAL);
305 	if (buflen <= 0 || offset >= buflen)
306 		return (0);
307 	if ((n = buflen - offset) > IOSIZE_MAX)
308 		return (EINVAL);
309 	return (uiomove((char *)buf + offset, n, uio));
310 }
311 
312 /*
313  * Give next character to user as result of read.
314  */
315 int
ureadc(int c,struct uio * uio)316 ureadc(int c, struct uio *uio)
317 {
318 	struct iovec *iov;
319 	char *iov_base;
320 
321 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
322 	    "Calling ureadc()");
323 
324 again:
325 	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
326 		panic("ureadc");
327 	iov = uio->uio_iov;
328 	if (iov->iov_len == 0) {
329 		uio->uio_iovcnt--;
330 		uio->uio_iov++;
331 		goto again;
332 	}
333 	switch (uio->uio_segflg) {
334 	case UIO_USERSPACE:
335 		if (subyte(iov->iov_base, c) < 0)
336 			return (EFAULT);
337 		break;
338 
339 	case UIO_SYSSPACE:
340 		iov_base = iov->iov_base;
341 		*iov_base = c;
342 		break;
343 
344 	case UIO_NOCOPY:
345 		break;
346 	}
347 	iov->iov_base = (char *)iov->iov_base + 1;
348 	iov->iov_len--;
349 	uio->uio_resid--;
350 	uio->uio_offset++;
351 	return (0);
352 }
353 
354 int
copyiniov(const struct iovec * iovp,u_int iovcnt,struct iovec ** iov,int error)355 copyiniov(const struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
356 {
357 	u_int iovlen;
358 
359 	*iov = NULL;
360 	if (iovcnt > UIO_MAXIOV)
361 		return (error);
362 	iovlen = iovcnt * sizeof(struct iovec);
363 	*iov = malloc(iovlen, M_IOV, M_WAITOK);
364 	error = copyin(iovp, *iov, iovlen);
365 	if (error) {
366 		free(*iov, M_IOV);
367 		*iov = NULL;
368 	}
369 	return (error);
370 }
371 
372 int
copyinuio(const struct iovec * iovp,u_int iovcnt,struct uio ** uiop)373 copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop)
374 {
375 	struct iovec *iov;
376 	struct uio *uio;
377 	u_int iovlen;
378 	int error, i;
379 
380 	*uiop = NULL;
381 	if (iovcnt > UIO_MAXIOV)
382 		return (EINVAL);
383 	iovlen = iovcnt * sizeof(struct iovec);
384 	uio = allocuio(iovcnt);
385 	iov = uio->uio_iov;
386 	error = copyin(iovp, iov, iovlen);
387 	if (error != 0) {
388 		freeuio(uio);
389 		return (error);
390 	}
391 	uio->uio_iovcnt = iovcnt;
392 	uio->uio_segflg = UIO_USERSPACE;
393 	uio->uio_offset = -1;
394 	uio->uio_resid = 0;
395 	for (i = 0; i < iovcnt; i++) {
396 		if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) {
397 			freeuio(uio);
398 			return (EINVAL);
399 		}
400 		uio->uio_resid += iov->iov_len;
401 		iov++;
402 	}
403 	*uiop = uio;
404 	return (0);
405 }
406 
407 struct uio *
allocuio(u_int iovcnt)408 allocuio(u_int iovcnt)
409 {
410 	struct uio *uio;
411 	int iovlen;
412 
413 	KASSERT(iovcnt <= UIO_MAXIOV,
414 	    ("Requested %u iovecs exceed UIO_MAXIOV", iovcnt));
415 	iovlen = iovcnt * sizeof(struct iovec);
416 	uio = malloc(iovlen + sizeof(*uio), M_IOV, M_WAITOK);
417 	uio->uio_iov = (struct iovec *)(uio + 1);
418 
419 	return (uio);
420 }
421 
422 void
freeuio(struct uio * uio)423 freeuio(struct uio *uio)
424 {
425 	free(uio, M_IOV);
426 }
427 
428 struct uio *
cloneuio(struct uio * uiop)429 cloneuio(struct uio *uiop)
430 {
431 	struct iovec *iov;
432 	struct uio *uio;
433 	int iovlen;
434 
435 	iovlen = uiop->uio_iovcnt * sizeof(struct iovec);
436 	uio = allocuio(uiop->uio_iovcnt);
437 	iov = uio->uio_iov;
438 	*uio = *uiop;
439 	uio->uio_iov = iov;
440 	bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
441 	return (uio);
442 }
443 
444 /*
445  * Map some anonymous memory in user space of size sz, rounded up to the page
446  * boundary.
447  */
448 int
copyout_map(struct thread * td,vm_offset_t * addr,size_t sz)449 copyout_map(struct thread *td, vm_offset_t *addr, size_t sz)
450 {
451 	struct vmspace *vms;
452 	int error;
453 	vm_size_t size;
454 
455 	vms = td->td_proc->p_vmspace;
456 
457 	/*
458 	 * Map somewhere after heap in process memory.
459 	 */
460 	*addr = round_page((vm_offset_t)vms->vm_daddr +
461 	    lim_max(td, RLIMIT_DATA));
462 
463 	/* round size up to page boundary */
464 	size = (vm_size_t)round_page(sz);
465 	if (size == 0)
466 		return (EINVAL);
467 	error = vm_mmap_object(&vms->vm_map, addr, size, VM_PROT_READ |
468 	    VM_PROT_WRITE, VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, NULL, 0,
469 	    FALSE, td);
470 	return (error);
471 }
472 
473 /*
474  * Unmap memory in user space.
475  */
476 int
copyout_unmap(struct thread * td,vm_offset_t addr,size_t sz)477 copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz)
478 {
479 	vm_map_t map;
480 	vm_size_t size;
481 
482 	if (sz == 0)
483 		return (0);
484 
485 	map = &td->td_proc->p_vmspace->vm_map;
486 	size = (vm_size_t)round_page(sz);
487 
488 	if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS)
489 		return (EINVAL);
490 
491 	return (0);
492 }
493 
494 int32_t
fuword32(volatile const void * addr)495 fuword32(volatile const void *addr)
496 {
497 	int rv;
498 	int32_t val;
499 
500 	rv = fueword32(addr, &val);
501 	return (rv == -1 ? -1 : val);
502 }
503 
504 #ifdef _LP64
505 int64_t
fuword64(volatile const void * addr)506 fuword64(volatile const void *addr)
507 {
508 	int rv;
509 	int64_t val;
510 
511 	rv = fueword64(addr, &val);
512 	return (rv == -1 ? -1 : val);
513 }
514 #endif /* _LP64 */
515 
516 long
fuword(volatile const void * addr)517 fuword(volatile const void *addr)
518 {
519 	long val;
520 	int rv;
521 
522 	rv = fueword(addr, &val);
523 	return (rv == -1 ? -1 : val);
524 }
525 
526 uint32_t
casuword32(volatile uint32_t * addr,uint32_t old,uint32_t new)527 casuword32(volatile uint32_t *addr, uint32_t old, uint32_t new)
528 {
529 	int rv;
530 	uint32_t val;
531 
532 	rv = casueword32(addr, old, &val, new);
533 	return (rv == -1 ? -1 : val);
534 }
535 
536 u_long
casuword(volatile u_long * addr,u_long old,u_long new)537 casuword(volatile u_long *addr, u_long old, u_long new)
538 {
539 	int rv;
540 	u_long val;
541 
542 	rv = casueword(addr, old, &val, new);
543 	return (rv == -1 ? -1 : val);
544 }
545