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 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following 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 * 4. Neither the name of the University nor the names of its contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 *	@(#)kern_subr.c	8.3 (Berkeley) 1/21/94
35 */
36
37#include <sys/cdefs.h>
38__FBSDID("$FreeBSD$");
39
40#include "opt_zero.h"
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/mutex.h>
49#include <sys/proc.h>
50#include <sys/resourcevar.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_map.h>
60#ifdef ZERO_COPY_SOCKETS
61#include <vm/vm_object.h>
62#endif
63
64SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, NULL, UIO_MAXIOV,
65	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
66
67static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault);
68
69#ifdef ZERO_COPY_SOCKETS
70/* Declared in uipc_socket.c */
71extern int so_zero_copy_receive;
72
73/*
74 * Identify the physical page mapped at the given kernel virtual
75 * address.  Insert this physical page into the given address space at
76 * the given virtual address, replacing the physical page, if any,
77 * that already exists there.
78 */
79static int
80vm_pgmoveco(vm_map_t mapa, vm_offset_t kaddr, vm_offset_t uaddr)
81{
82	vm_map_t map = mapa;
83	vm_page_t kern_pg, user_pg;
84	vm_object_t uobject;
85	vm_map_entry_t entry;
86	vm_pindex_t upindex;
87	vm_prot_t prot;
88	boolean_t wired;
89
90	KASSERT((uaddr & PAGE_MASK) == 0,
91	    ("vm_pgmoveco: uaddr is not page aligned"));
92
93	/*
94	 * Herein the physical page is validated and dirtied.  It is
95	 * unwired in sf_buf_mext().
96	 */
97	kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
98	kern_pg->valid = VM_PAGE_BITS_ALL;
99	KASSERT(kern_pg->queue == PQ_NONE && kern_pg->wire_count == 1,
100	    ("vm_pgmoveco: kern_pg is not correctly wired"));
101
102	if ((vm_map_lookup(&map, uaddr,
103			   VM_PROT_WRITE, &entry, &uobject,
104			   &upindex, &prot, &wired)) != KERN_SUCCESS) {
105		return(EFAULT);
106	}
107	VM_OBJECT_LOCK(uobject);
108retry:
109	if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
110		if (vm_page_sleep_if_busy(user_pg, TRUE, "vm_pgmoveco"))
111			goto retry;
112		vm_page_lock(user_pg);
113		pmap_remove_all(user_pg);
114		vm_page_free(user_pg);
115		vm_page_unlock(user_pg);
116	} else {
117		/*
118		 * Even if a physical page does not exist in the
119		 * object chain's first object, a physical page from a
120		 * backing object may be mapped read only.
121		 */
122		if (uobject->backing_object != NULL)
123			pmap_remove(map->pmap, uaddr, uaddr + PAGE_SIZE);
124	}
125	vm_page_insert(kern_pg, uobject, upindex);
126	vm_page_dirty(kern_pg);
127	VM_OBJECT_UNLOCK(uobject);
128	vm_map_lookup_done(map, entry);
129	return(KERN_SUCCESS);
130}
131#endif /* ZERO_COPY_SOCKETS */
132
133int
134copyin_nofault(const void *udaddr, void *kaddr, size_t len)
135{
136	int error, save;
137
138	save = vm_fault_disable_pagefaults();
139	error = copyin(udaddr, kaddr, len);
140	vm_fault_enable_pagefaults(save);
141	return (error);
142}
143
144int
145copyout_nofault(const void *kaddr, void *udaddr, size_t len)
146{
147	int error, save;
148
149	save = vm_fault_disable_pagefaults();
150	error = copyout(kaddr, udaddr, len);
151	vm_fault_enable_pagefaults(save);
152	return (error);
153}
154
155#define	PHYS_PAGE_COUNT(len)	(howmany(len, PAGE_SIZE) + 1)
156
157int
158physcopyin(void *src, vm_paddr_t dst, size_t len)
159{
160	vm_page_t m[PHYS_PAGE_COUNT(len)];
161	struct iovec iov[1];
162	struct uio uio;
163	int i;
164
165	iov[0].iov_base = src;
166	iov[0].iov_len = len;
167	uio.uio_iov = iov;
168	uio.uio_iovcnt = 1;
169	uio.uio_offset = 0;
170	uio.uio_resid = len;
171	uio.uio_segflg = UIO_SYSSPACE;
172	uio.uio_rw = UIO_WRITE;
173	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, dst += PAGE_SIZE)
174		m[i] = PHYS_TO_VM_PAGE(dst);
175	return (uiomove_fromphys(m, dst & PAGE_MASK, len, &uio));
176}
177
178int
179physcopyout(vm_paddr_t src, void *dst, size_t len)
180{
181	vm_page_t m[PHYS_PAGE_COUNT(len)];
182	struct iovec iov[1];
183	struct uio uio;
184	int i;
185
186	iov[0].iov_base = dst;
187	iov[0].iov_len = len;
188	uio.uio_iov = iov;
189	uio.uio_iovcnt = 1;
190	uio.uio_offset = 0;
191	uio.uio_resid = len;
192	uio.uio_segflg = UIO_SYSSPACE;
193	uio.uio_rw = UIO_READ;
194	for (i = 0; i < PHYS_PAGE_COUNT(len); i++, src += PAGE_SIZE)
195		m[i] = PHYS_TO_VM_PAGE(src);
196	return (uiomove_fromphys(m, src & PAGE_MASK, len, &uio));
197}
198
199#undef PHYS_PAGE_COUNT
200
201int
202uiomove(void *cp, int n, struct uio *uio)
203{
204
205	return (uiomove_faultflag(cp, n, uio, 0));
206}
207
208int
209uiomove_nofault(void *cp, int n, struct uio *uio)
210{
211
212	return (uiomove_faultflag(cp, n, uio, 1));
213}
214
215static int
216uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault)
217{
218	struct thread *td;
219	struct iovec *iov;
220	size_t cnt;
221	int error, newflags, save;
222
223	td = curthread;
224	error = 0;
225
226	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
227	    ("uiomove: mode"));
228	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == td,
229	    ("uiomove proc"));
230	if (!nofault)
231		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
232		    "Calling uiomove()");
233
234	/* XXX does it make a sense to set TDP_DEADLKTREAT for UIO_SYSSPACE ? */
235	newflags = TDP_DEADLKTREAT;
236	if (uio->uio_segflg == UIO_USERSPACE && nofault) {
237		/*
238		 * Fail if a non-spurious page fault occurs.
239		 */
240		newflags |= TDP_NOFAULTING | TDP_RESETSPUR;
241	}
242	save = curthread_pflags_set(newflags);
243
244	while (n > 0 && uio->uio_resid) {
245		iov = uio->uio_iov;
246		cnt = iov->iov_len;
247		if (cnt == 0) {
248			uio->uio_iov++;
249			uio->uio_iovcnt--;
250			continue;
251		}
252		if (cnt > n)
253			cnt = n;
254
255		switch (uio->uio_segflg) {
256
257		case UIO_USERSPACE:
258			maybe_yield();
259			if (uio->uio_rw == UIO_READ)
260				error = copyout(cp, iov->iov_base, cnt);
261			else
262				error = copyin(iov->iov_base, cp, cnt);
263			if (error)
264				goto out;
265			break;
266
267		case UIO_SYSSPACE:
268			if (uio->uio_rw == UIO_READ)
269				bcopy(cp, iov->iov_base, cnt);
270			else
271				bcopy(iov->iov_base, cp, cnt);
272			break;
273		case UIO_NOCOPY:
274			break;
275		}
276		iov->iov_base = (char *)iov->iov_base + cnt;
277		iov->iov_len -= cnt;
278		uio->uio_resid -= cnt;
279		uio->uio_offset += cnt;
280		cp = (char *)cp + cnt;
281		n -= cnt;
282	}
283out:
284	curthread_pflags_restore(save);
285	return (error);
286}
287
288/*
289 * Wrapper for uiomove() that validates the arguments against a known-good
290 * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
291 * is almost definitely a bad thing, so we catch that here as well.  We
292 * return a runtime failure, but it might be desirable to generate a runtime
293 * assertion failure instead.
294 */
295int
296uiomove_frombuf(void *buf, int buflen, struct uio *uio)
297{
298	size_t offset, n;
299
300	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
301	    (offset = uio->uio_offset) != uio->uio_offset)
302		return (EINVAL);
303	if (buflen <= 0 || offset >= buflen)
304		return (0);
305	if ((n = buflen - offset) > IOSIZE_MAX)
306		return (EINVAL);
307	return (uiomove((char *)buf + offset, n, uio));
308}
309
310#ifdef ZERO_COPY_SOCKETS
311/*
312 * Experimental support for zero-copy I/O
313 */
314static int
315userspaceco(void *cp, u_int cnt, struct uio *uio, int disposable)
316{
317	struct iovec *iov;
318	int error;
319
320	iov = uio->uio_iov;
321	if (uio->uio_rw == UIO_READ) {
322		if ((so_zero_copy_receive != 0)
323		 && ((cnt & PAGE_MASK) == 0)
324		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
325		 && ((uio->uio_offset & PAGE_MASK) == 0)
326		 && ((((intptr_t) cp) & PAGE_MASK) == 0)
327		 && (disposable != 0)) {
328			/* SOCKET: use page-trading */
329			/*
330			 * We only want to call vm_pgmoveco() on
331			 * disposeable pages, since it gives the
332			 * kernel page to the userland process.
333			 */
334			error =	vm_pgmoveco(&curproc->p_vmspace->vm_map,
335			    (vm_offset_t)cp, (vm_offset_t)iov->iov_base);
336
337			/*
338			 * If we get an error back, attempt
339			 * to use copyout() instead.  The
340			 * disposable page should be freed
341			 * automatically if we weren't able to move
342			 * it into userland.
343			 */
344			if (error != 0)
345				error = copyout(cp, iov->iov_base, cnt);
346		} else {
347			error = copyout(cp, iov->iov_base, cnt);
348		}
349	} else {
350		error = copyin(iov->iov_base, cp, cnt);
351	}
352	return (error);
353}
354
355int
356uiomoveco(void *cp, int n, struct uio *uio, int disposable)
357{
358	struct iovec *iov;
359	u_int cnt;
360	int error;
361
362	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
363	    ("uiomoveco: mode"));
364	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
365	    ("uiomoveco proc"));
366
367	while (n > 0 && uio->uio_resid) {
368		iov = uio->uio_iov;
369		cnt = iov->iov_len;
370		if (cnt == 0) {
371			uio->uio_iov++;
372			uio->uio_iovcnt--;
373			continue;
374		}
375		if (cnt > n)
376			cnt = n;
377
378		switch (uio->uio_segflg) {
379
380		case UIO_USERSPACE:
381			maybe_yield();
382			error = userspaceco(cp, cnt, uio, disposable);
383			if (error)
384				return (error);
385			break;
386
387		case UIO_SYSSPACE:
388			if (uio->uio_rw == UIO_READ)
389				bcopy(cp, iov->iov_base, cnt);
390			else
391				bcopy(iov->iov_base, cp, cnt);
392			break;
393		case UIO_NOCOPY:
394			break;
395		}
396		iov->iov_base = (char *)iov->iov_base + cnt;
397		iov->iov_len -= cnt;
398		uio->uio_resid -= cnt;
399		uio->uio_offset += cnt;
400		cp = (char *)cp + cnt;
401		n -= cnt;
402	}
403	return (0);
404}
405#endif /* ZERO_COPY_SOCKETS */
406
407/*
408 * Give next character to user as result of read.
409 */
410int
411ureadc(int c, struct uio *uio)
412{
413	struct iovec *iov;
414	char *iov_base;
415
416	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
417	    "Calling ureadc()");
418
419again:
420	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
421		panic("ureadc");
422	iov = uio->uio_iov;
423	if (iov->iov_len == 0) {
424		uio->uio_iovcnt--;
425		uio->uio_iov++;
426		goto again;
427	}
428	switch (uio->uio_segflg) {
429
430	case UIO_USERSPACE:
431		if (subyte(iov->iov_base, c) < 0)
432			return (EFAULT);
433		break;
434
435	case UIO_SYSSPACE:
436		iov_base = iov->iov_base;
437		*iov_base = c;
438		iov->iov_base = iov_base;
439		break;
440
441	case UIO_NOCOPY:
442		break;
443	}
444	iov->iov_base = (char *)iov->iov_base + 1;
445	iov->iov_len--;
446	uio->uio_resid--;
447	uio->uio_offset++;
448	return (0);
449}
450
451int
452copyinfrom(const void * __restrict src, void * __restrict dst, size_t len,
453    int seg)
454{
455	int error = 0;
456
457	switch (seg) {
458	case UIO_USERSPACE:
459		error = copyin(src, dst, len);
460		break;
461	case UIO_SYSSPACE:
462		bcopy(src, dst, len);
463		break;
464	default:
465		panic("copyinfrom: bad seg %d\n", seg);
466	}
467	return (error);
468}
469
470int
471copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len,
472    size_t * __restrict copied, int seg)
473{
474	int error = 0;
475
476	switch (seg) {
477	case UIO_USERSPACE:
478		error = copyinstr(src, dst, len, copied);
479		break;
480	case UIO_SYSSPACE:
481		error = copystr(src, dst, len, copied);
482		break;
483	default:
484		panic("copyinstrfrom: bad seg %d\n", seg);
485	}
486	return (error);
487}
488
489int
490copyiniov(struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
491{
492	u_int iovlen;
493
494	*iov = NULL;
495	if (iovcnt > UIO_MAXIOV)
496		return (error);
497	iovlen = iovcnt * sizeof (struct iovec);
498	*iov = malloc(iovlen, M_IOV, M_WAITOK);
499	error = copyin(iovp, *iov, iovlen);
500	if (error) {
501		free(*iov, M_IOV);
502		*iov = NULL;
503	}
504	return (error);
505}
506
507int
508copyinuio(struct iovec *iovp, u_int iovcnt, struct uio **uiop)
509{
510	struct iovec *iov;
511	struct uio *uio;
512	u_int iovlen;
513	int error, i;
514
515	*uiop = NULL;
516	if (iovcnt > UIO_MAXIOV)
517		return (EINVAL);
518	iovlen = iovcnt * sizeof (struct iovec);
519	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
520	iov = (struct iovec *)(uio + 1);
521	error = copyin(iovp, iov, iovlen);
522	if (error) {
523		free(uio, M_IOV);
524		return (error);
525	}
526	uio->uio_iov = iov;
527	uio->uio_iovcnt = iovcnt;
528	uio->uio_segflg = UIO_USERSPACE;
529	uio->uio_offset = -1;
530	uio->uio_resid = 0;
531	for (i = 0; i < iovcnt; i++) {
532		if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) {
533			free(uio, M_IOV);
534			return (EINVAL);
535		}
536		uio->uio_resid += iov->iov_len;
537		iov++;
538	}
539	*uiop = uio;
540	return (0);
541}
542
543struct uio *
544cloneuio(struct uio *uiop)
545{
546	struct uio *uio;
547	int iovlen;
548
549	iovlen = uiop->uio_iovcnt * sizeof (struct iovec);
550	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
551	*uio = *uiop;
552	uio->uio_iov = (struct iovec *)(uio + 1);
553	bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
554	return (uio);
555}
556
557/*
558 * Map some anonymous memory in user space of size sz, rounded up to the page
559 * boundary.
560 */
561int
562copyout_map(struct thread *td, vm_offset_t *addr, size_t sz)
563{
564	struct vmspace *vms;
565	int error;
566	vm_size_t size;
567
568	vms = td->td_proc->p_vmspace;
569
570	/*
571	 * Map somewhere after heap in process memory.
572	 */
573	PROC_LOCK(td->td_proc);
574	*addr = round_page((vm_offset_t)vms->vm_daddr +
575	    lim_max(td->td_proc, RLIMIT_DATA));
576	PROC_UNLOCK(td->td_proc);
577
578	/* round size up to page boundry */
579	size = (vm_size_t)round_page(sz);
580
581	error = vm_mmap(&vms->vm_map, addr, size, PROT_READ | PROT_WRITE,
582	    VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, OBJT_DEFAULT, NULL, 0);
583
584	return (error);
585}
586
587/*
588 * Unmap memory in user space.
589 */
590int
591copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz)
592{
593	vm_map_t map;
594	vm_size_t size;
595
596	if (sz == 0)
597		return (0);
598
599	map = &td->td_proc->p_vmspace->vm_map;
600	size = (vm_size_t)round_page(sz);
601
602	if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS)
603		return (EINVAL);
604
605	return (0);
606}
607