xref: /dflybsd-src/sys/kern/kern_subr.c (revision 767a43118674e3039d1f3dd7d3f2662a2eccf599)
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_subr.c	8.3 (Berkeley) 1/21/94
39  * $FreeBSD: src/sys/kern/kern_subr.c,v 1.31.2.2 2002/04/21 08:09:37 bde Exp $
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/proc.h>
46 #include <sys/malloc.h>
47 #include <sys/lock.h>
48 #include <sys/resourcevar.h>
49 #include <sys/sysctl.h>
50 #include <sys/uio.h>
51 #include <sys/vnode.h>
52 #include <sys/thread2.h>
53 #include <machine/limits.h>
54 
55 #include <cpu/lwbuf.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_map.h>
60 
61 SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, NULL, UIO_MAXIOV,
62 	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
63 
64 /*
65  * UIO_READ:	copy the kernelspace cp to the user or kernelspace UIO
66  * UIO_WRITE:	copy the user or kernelspace UIO to the kernelspace cp
67  *
68  * For userspace UIO's, uio_td must be the current thread.
69  *
70  * The syscall interface is responsible for limiting the length to
71  * ssize_t for things like read() or write() which return the bytes
72  * read or written as ssize_t.  These functions work with unsigned
73  * lengths.
74  */
75 int
76 uiomove(caddr_t cp, size_t n, struct uio *uio)
77 {
78 	thread_t td = curthread;
79 	struct iovec *iov;
80 	size_t cnt;
81 	int error = 0;
82 	int save = 0;
83 
84 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
85 	    ("uiomove: mode"));
86 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == td,
87 	    ("uiomove proc"));
88 
89 	crit_enter();
90 	save = td->td_flags & TDF_DEADLKTREAT;
91 	td->td_flags |= TDF_DEADLKTREAT;
92 	crit_exit();
93 
94 	while (n > 0 && uio->uio_resid) {
95 		iov = uio->uio_iov;
96 		cnt = iov->iov_len;
97 		if (cnt == 0) {
98 			uio->uio_iov++;
99 			uio->uio_iovcnt--;
100 			continue;
101 		}
102 		if (cnt > n)
103 			cnt = n;
104 
105 		switch (uio->uio_segflg) {
106 
107 		case UIO_USERSPACE:
108 			lwkt_user_yield();
109 			if (uio->uio_rw == UIO_READ)
110 				error = copyout(cp, iov->iov_base, cnt);
111 			else
112 				error = copyin(iov->iov_base, cp, cnt);
113 			if (error)
114 				break;
115 			break;
116 
117 		case UIO_SYSSPACE:
118 			if (uio->uio_rw == UIO_READ)
119 				bcopy(cp, iov->iov_base, cnt);
120 			else
121 				bcopy(iov->iov_base, cp, cnt);
122 			break;
123 		case UIO_NOCOPY:
124 			break;
125 		}
126 		iov->iov_base = (char *)iov->iov_base + cnt;
127 		iov->iov_len -= cnt;
128 		uio->uio_resid -= cnt;
129 		uio->uio_offset += cnt;
130 		cp += cnt;
131 		n -= cnt;
132 	}
133 	crit_enter();
134 	td->td_flags = (td->td_flags & ~TDF_DEADLKTREAT) | save;
135 	crit_exit();
136 	return (error);
137 }
138 
139 /*
140  * This is the same as uiomove() except (cp, n) is within the bounds of
141  * the passed, locked buffer.  Under certain circumstances a VM fault
142  * occuring with a locked buffer held can result in a deadlock or an
143  * attempt to recursively lock the buffer.
144  *
145  * This procedure deals with these cases.
146  *
147  * If the buffer represents a regular file, is B_CACHE, but the last VM page
148  * is not fully valid we fix-up the last VM page.  This should handle the
149  * recursive lock issue.
150  *
151  * Deadlocks are another issue.  We are holding the vp and the bp locked
152  * and could deadlock against a different vp and/or bp if another thread is
153  * trying to access us while we accessing it.  The only solution here is
154  * to release the bp and vnode lock and do the uio to/from a system buffer,
155  * then regain the locks and copyback (if applicable).  XXX TODO.
156  */
157 int
158 uiomovebp(struct buf *bp, caddr_t cp, size_t n, struct uio *uio)
159 {
160 	int count;
161 	vm_page_t m;
162 
163 	if (bp->b_vp && bp->b_vp->v_type == VREG &&
164 	    (bp->b_flags & B_CACHE) &&
165 	    (count = bp->b_xio.xio_npages) != 0 &&
166 	    (m = bp->b_xio.xio_pages[count-1])->valid != VM_PAGE_BITS_ALL) {
167 		vm_page_zero_invalid(m, TRUE);
168 	}
169 	return (uiomove(cp, n, uio));
170 }
171 
172 /*
173  * Like uiomove() but copies zero-fill.  Only allowed for UIO_READ,
174  * for obvious reasons.
175  */
176 int
177 uiomovez(size_t n, struct uio *uio)
178 {
179 	struct iovec *iov;
180 	size_t cnt;
181 	int error = 0;
182 
183 	KASSERT(uio->uio_rw == UIO_READ, ("uiomovez: mode"));
184 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
185 		("uiomove proc"));
186 
187 	while (n > 0 && uio->uio_resid) {
188 		iov = uio->uio_iov;
189 		cnt = iov->iov_len;
190 		if (cnt == 0) {
191 			uio->uio_iov++;
192 			uio->uio_iovcnt--;
193 			continue;
194 		}
195 		if (cnt > n)
196 			cnt = n;
197 
198 		switch (uio->uio_segflg) {
199 		case UIO_USERSPACE:
200 			error = copyout(ZeroPage, iov->iov_base, cnt);
201 			if (error)
202 				break;
203 			break;
204 		case UIO_SYSSPACE:
205 			bzero(iov->iov_base, cnt);
206 			break;
207 		case UIO_NOCOPY:
208 			break;
209 		}
210 		iov->iov_base = (char *)iov->iov_base + cnt;
211 		iov->iov_len -= cnt;
212 		uio->uio_resid -= cnt;
213 		uio->uio_offset += cnt;
214 		n -= cnt;
215 	}
216 	return (error);
217 }
218 
219 /*
220  * Wrapper for uiomove() that validates the arguments against a known-good
221  * kernel buffer.  This function automatically indexes the buffer by
222  * uio_offset and handles all range checking.
223  */
224 int
225 uiomove_frombuf(void *buf, size_t buflen, struct uio *uio)
226 {
227 	size_t offset;
228 
229 	offset = (size_t)uio->uio_offset;
230 	if ((off_t)offset != uio->uio_offset)
231 		return (EINVAL);
232 	if (buflen == 0 || offset >= buflen)
233 		return (0);
234 	return (uiomove((char *)buf + offset, buflen - offset, uio));
235 }
236 
237 /*
238  * Give next character to user as result of read.
239  */
240 int
241 ureadc(int c, struct uio *uio)
242 {
243 	struct iovec *iov;
244 	char *iov_base;
245 
246 again:
247 	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
248 		panic("ureadc");
249 	iov = uio->uio_iov;
250 	if (iov->iov_len == 0) {
251 		uio->uio_iovcnt--;
252 		uio->uio_iov++;
253 		goto again;
254 	}
255 	switch (uio->uio_segflg) {
256 
257 	case UIO_USERSPACE:
258 		if (subyte(iov->iov_base, c) < 0)
259 			return (EFAULT);
260 		break;
261 
262 	case UIO_SYSSPACE:
263 		iov_base = iov->iov_base;
264 		*iov_base = c;
265 		iov->iov_base = iov_base;
266 		break;
267 
268 	case UIO_NOCOPY:
269 		break;
270 	}
271 	iov->iov_base = (char *)iov->iov_base + 1;
272 	iov->iov_len--;
273 	uio->uio_resid--;
274 	uio->uio_offset++;
275 	return (0);
276 }
277 
278 /*
279  * General routine to allocate a hash table.  Make the hash table size a
280  * power of 2 greater or equal to the number of elements requested, and
281  * store the masking value in *hashmask.
282  */
283 void *
284 hashinit(int elements, struct malloc_type *type, u_long *hashmask)
285 {
286 	long hashsize;
287 	LIST_HEAD(generic, generic) *hashtbl;
288 	int i;
289 
290 	if (elements <= 0)
291 		panic("hashinit: bad elements");
292 	for (hashsize = 2; hashsize < elements; hashsize <<= 1)
293 		continue;
294 	hashtbl = kmalloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
295 	for (i = 0; i < hashsize; i++)
296 		LIST_INIT(&hashtbl[i]);
297 	*hashmask = hashsize - 1;
298 	return (hashtbl);
299 }
300 
301 void
302 hashdestroy(void *vhashtbl, struct malloc_type *type, u_long hashmask)
303 {
304 	LIST_HEAD(generic, generic) *hashtbl, *hp;
305 
306 	hashtbl = vhashtbl;
307 	for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++)
308 		KASSERT(LIST_EMPTY(hp), ("%s: hash not empty", __func__));
309 	kfree(hashtbl, type);
310 }
311 
312 /*
313  * This is a newer version which allocates a hash table of structures.
314  *
315  * The returned array will be zero'd.  The caller is responsible for
316  * initializing the structures.
317  */
318 void *
319 hashinit_ext(int elements, size_t size, struct malloc_type *type,
320 	     u_long *hashmask)
321 {
322 	long hashsize;
323 	void *hashtbl;
324 
325 	if (elements <= 0)
326 		panic("hashinit: bad elements");
327 	for (hashsize = 2; hashsize < elements; hashsize <<= 1)
328 		continue;
329 	hashtbl = kmalloc((size_t)hashsize * size, type, M_WAITOK | M_ZERO);
330 	*hashmask = hashsize - 1;
331 	return (hashtbl);
332 }
333 
334 static int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039,
335 			2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653,
336 			7159, 7673, 8191, 12281, 16381, 24571, 32749 };
337 #define NPRIMES NELEM(primes)
338 
339 /*
340  * General routine to allocate a prime number sized hash table.
341  */
342 void *
343 phashinit(int elements, struct malloc_type *type, u_long *nentries)
344 {
345 	long hashsize;
346 	LIST_HEAD(generic, generic) *hashtbl;
347 	int i;
348 
349 	if (elements <= 0)
350 		panic("phashinit: bad elements");
351 	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
352 		i++;
353 		if (i == NPRIMES)
354 			break;
355 		hashsize = primes[i];
356 	}
357 	hashsize = primes[i - 1];
358 	hashtbl = kmalloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
359 	for (i = 0; i < hashsize; i++)
360 		LIST_INIT(&hashtbl[i]);
361 	*nentries = hashsize;
362 	return (hashtbl);
363 }
364 
365 /*
366  * This is a newer version which allocates a hash table of structures
367  * in a prime-number size.
368  *
369  * The returned array will be zero'd.  The caller is responsible for
370  * initializing the structures.
371  */
372 void *
373 phashinit_ext(int elements, size_t size, struct malloc_type *type,
374 	      u_long *nentries)
375 {
376 	long hashsize;
377 	void *hashtbl;
378 	int i;
379 
380 	if (elements <= 0)
381 		panic("phashinit: bad elements");
382 	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
383 		i++;
384 		if (i == NPRIMES)
385 			break;
386 		hashsize = primes[i];
387 	}
388 	hashsize = primes[i - 1];
389 	hashtbl = kmalloc((size_t)hashsize * size, type, M_WAITOK | M_ZERO);
390 	*nentries = hashsize;
391 	return (hashtbl);
392 }
393 
394 /*
395  * Copyin an iovec.  If the iovec array fits, use the preallocated small
396  * iovec structure.  If it is too big, dynamically allocate an iovec array
397  * of sufficient size.
398  *
399  * MPSAFE
400  */
401 int
402 iovec_copyin(struct iovec *uiov, struct iovec **kiov, struct iovec *siov,
403 	     size_t iov_cnt, size_t *iov_len)
404 {
405 	struct iovec *iovp;
406 	int error, i;
407 	size_t len;
408 
409 	if (iov_cnt > UIO_MAXIOV)
410 		return EMSGSIZE;
411 	if (iov_cnt > UIO_SMALLIOV) {
412 		*kiov = kmalloc(sizeof(struct iovec) * iov_cnt, M_IOV,
413 				M_WAITOK);
414 	} else {
415 		*kiov = siov;
416 	}
417 	error = copyin(uiov, *kiov, iov_cnt * sizeof(struct iovec));
418 	if (error == 0) {
419 		*iov_len = 0;
420 		for (i = 0, iovp = *kiov; i < iov_cnt; i++, iovp++) {
421 			/*
422 			 * Check for both *iov_len overflows and out of
423 			 * range iovp->iov_len's.  We limit to the
424 			 * capabilities of signed integers.
425 			 *
426 			 * GCC4 - overflow check opt requires assign/test.
427 			 */
428 			len = *iov_len + iovp->iov_len;
429 			if (len < *iov_len)
430 				error = EINVAL;
431 			*iov_len = len;
432 		}
433 	}
434 
435 	/*
436 	 * From userland disallow iovec's which exceed the sized size
437 	 * limit as the system calls return ssize_t.
438 	 *
439 	 * NOTE: Internal kernel interfaces can handle the unsigned
440 	 *	 limit.
441 	 */
442 	if (error == 0 && (ssize_t)*iov_len < 0)
443 		error = EINVAL;
444 
445 	if (error)
446 		iovec_free(kiov, siov);
447 	return (error);
448 }
449 
450 
451 /*
452  * Copyright (c) 2004 Alan L. Cox <alc@cs.rice.edu>
453  * Copyright (c) 1982, 1986, 1991, 1993
454  *	The Regents of the University of California.  All rights reserved.
455  * (c) UNIX System Laboratories, Inc.
456  * All or some portions of this file are derived from material licensed
457  * to the University of California by American Telephone and Telegraph
458  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
459  * the permission of UNIX System Laboratories, Inc.
460  *
461  * Redistribution and use in source and binary forms, with or without
462  * modification, are permitted provided that the following conditions
463  * are met:
464  * 1. Redistributions of source code must retain the above copyright
465  *    notice, this list of conditions and the following disclaimer.
466  * 2. Redistributions in binary form must reproduce the above copyright
467  *    notice, this list of conditions and the following disclaimer in the
468  *    documentation and/or other materials provided with the distribution.
469  * 4. Neither the name of the University nor the names of its contributors
470  *    may be used to endorse or promote products derived from this software
471  *    without specific prior written permission.
472  *
473  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
474  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
475  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
476  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
477  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
478  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
479  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
480  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
481  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
482  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
483  * SUCH DAMAGE.
484  *
485  * @(#)kern_subr.c	8.3 (Berkeley) 1/21/94
486  * $FreeBSD: src/sys/i386/i386/uio_machdep.c,v 1.1 2004/03/21 20:28:36 alc Exp $
487  */
488 
489 /*
490  * Implement uiomove(9) from physical memory using lwbuf's to reduce
491  * the creation and destruction of ephemeral mappings.
492  */
493 int
494 uiomove_fromphys(vm_page_t *ma, vm_offset_t offset, size_t n, struct uio *uio)
495 {
496 	struct lwbuf lwb_cache;
497 	struct lwbuf *lwb;
498 	struct thread *td = curthread;
499 	struct iovec *iov;
500 	void *cp;
501 	vm_offset_t page_offset;
502 	vm_page_t m;
503 	size_t cnt;
504 	int error = 0;
505 	int save = 0;
506 
507 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
508 	    ("uiomove_fromphys: mode"));
509 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
510 	    ("uiomove_fromphys proc"));
511 
512 	crit_enter();
513 	save = td->td_flags & TDF_DEADLKTREAT;
514 	td->td_flags |= TDF_DEADLKTREAT;
515 	crit_exit();
516 
517 	while (n > 0 && uio->uio_resid) {
518 		iov = uio->uio_iov;
519 		cnt = iov->iov_len;
520 		if (cnt == 0) {
521 			uio->uio_iov++;
522 			uio->uio_iovcnt--;
523 			continue;
524 		}
525 		if (cnt > n)
526 			cnt = n;
527 		page_offset = offset & PAGE_MASK;
528 		cnt = min(cnt, PAGE_SIZE - page_offset);
529 		m = ma[offset >> PAGE_SHIFT];
530 		lwb = lwbuf_alloc(m, &lwb_cache);
531 		cp = (char *)lwbuf_kva(lwb) + page_offset;
532 		switch (uio->uio_segflg) {
533 		case UIO_USERSPACE:
534 			/*
535 			 * note: removed uioyield (it was the wrong place to
536 			 * put it).
537 			 */
538 			if (uio->uio_rw == UIO_READ)
539 				error = copyout(cp, iov->iov_base, cnt);
540 			else
541 				error = copyin(iov->iov_base, cp, cnt);
542 			if (error) {
543 				lwbuf_free(lwb);
544 				goto out;
545 			}
546 			break;
547 		case UIO_SYSSPACE:
548 			if (uio->uio_rw == UIO_READ)
549 				bcopy(cp, iov->iov_base, cnt);
550 			else
551 				bcopy(iov->iov_base, cp, cnt);
552 			break;
553 		case UIO_NOCOPY:
554 			break;
555 		}
556 		lwbuf_free(lwb);
557 		iov->iov_base = (char *)iov->iov_base + cnt;
558 		iov->iov_len -= cnt;
559 		uio->uio_resid -= cnt;
560 		uio->uio_offset += cnt;
561 		offset += cnt;
562 		n -= cnt;
563 	}
564 out:
565 	if (save == 0) {
566 		crit_enter();
567 		td->td_flags &= ~TDF_DEADLKTREAT;
568 		crit_exit();
569 	}
570 	return (error);
571 }
572 
573