xref: /dragonfly/sys/kern/kern_subr.c (revision c3762235)
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. 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  * $FreeBSD: src/sys/kern/kern_subr.c,v 1.31.2.2 2002/04/21 08:09:37 bde Exp $
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/proc.h>
42 #include <sys/malloc.h>
43 #include <sys/lock.h>
44 #include <sys/resourcevar.h>
45 #include <sys/sysctl.h>
46 #include <sys/uio.h>
47 #include <sys/vnode.h>
48 #include <sys/thread2.h>
49 #include <machine/limits.h>
50 
51 #include <cpu/lwbuf.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_page.h>
55 #include <vm/vm_map.h>
56 
57 SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, NULL, UIO_MAXIOV,
58 	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
59 
60 int
61 copyin_nofault(const void *udaddr, void *kaddr, size_t len)
62 {
63 	thread_t td = curthread;
64 	int error;
65 
66 	atomic_set_int(&td->td_flags, TDF_NOFAULT);
67 	error = copyin(udaddr, kaddr, len);
68 	atomic_clear_int(&td->td_flags, TDF_NOFAULT);
69 	return error;
70 }
71 
72 int
73 copyout_nofault(const void *kaddr, void *udaddr, size_t len)
74 {
75 	thread_t td = curthread;
76 	int error;
77 
78 	atomic_set_int(&td->td_flags, TDF_NOFAULT);
79 	error = copyout(kaddr, udaddr, len);
80 	atomic_clear_int(&td->td_flags, TDF_NOFAULT);
81 	return error;
82 }
83 
84 /*
85  * UIO_READ:	copy the kernelspace cp to the user or kernelspace UIO
86  * UIO_WRITE:	copy the user or kernelspace UIO to the kernelspace cp
87  *
88  * For userspace UIO's, uio_td must be the current thread.
89  *
90  * The syscall interface is responsible for limiting the length to
91  * ssize_t for things like read() or write() which return the bytes
92  * read or written as ssize_t.  These functions work with unsigned
93  * lengths.
94  */
95 int
96 uiomove(caddr_t cp, size_t n, struct uio *uio)
97 {
98 	thread_t td = curthread;
99 	struct iovec *iov;
100 	size_t cnt;
101 	size_t tot;
102 	int error = 0;
103 	int save = 0;
104 
105 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
106 	    ("uiomove: mode"));
107 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == td,
108 	    ("uiomove proc"));
109 
110 	crit_enter();
111 	save = td->td_flags & TDF_DEADLKTREAT;
112 	td->td_flags |= TDF_DEADLKTREAT;
113 	crit_exit();
114 
115 	tot = 0;
116 
117 	while (n > 0 && uio->uio_resid) {
118 		iov = uio->uio_iov;
119 		cnt = iov->iov_len;
120 		if (cnt == 0) {
121 			uio->uio_iov++;
122 			uio->uio_iovcnt--;
123 			continue;
124 		}
125 		if (cnt > n)
126 			cnt = n;
127 		tot += cnt;
128 
129 		switch (uio->uio_segflg) {
130 		case UIO_USERSPACE:
131 			if (tot > 1024*1024)
132 				lwkt_user_yield();
133 			if (uio->uio_rw == UIO_READ)
134 				error = copyout(cp, iov->iov_base, cnt);
135 			else
136 				error = copyin(iov->iov_base, cp, cnt);
137 			break;
138 		case UIO_SYSSPACE:
139 			if (uio->uio_rw == UIO_READ)
140 				bcopy(cp, iov->iov_base, cnt);
141 			else
142 				bcopy(iov->iov_base, cp, cnt);
143 			break;
144 		case UIO_NOCOPY:
145 			break;
146 		}
147 
148 		if (error)
149 			break;
150 		iov->iov_base = (char *)iov->iov_base + cnt;
151 		iov->iov_len -= cnt;
152 		uio->uio_resid -= cnt;
153 		uio->uio_offset += cnt;
154 		cp += cnt;
155 		n -= cnt;
156 	}
157 	crit_enter();
158 	td->td_flags = (td->td_flags & ~TDF_DEADLKTREAT) | save;
159 	crit_exit();
160 
161 	return (error);
162 }
163 
164 /*
165  * This is the same as uiomove() except (cp, n) is within the bounds of
166  * the passed, locked buffer.  Under certain circumstances a VM fault
167  * occuring with a locked buffer held can result in a deadlock or an
168  * attempt to recursively lock the buffer.
169  *
170  * This procedure deals with these cases.
171  *
172  * If the buffer represents a regular file, is B_CACHE, but the last VM page
173  * is not fully valid we fix-up the last VM page.  This should handle the
174  * recursive lock issue.
175  *
176  * Deadlocks are another issue.  We are holding the vp and the bp locked
177  * and could deadlock against a different vp and/or bp if another thread is
178  * trying to access us while we accessing it.  The only solution here is
179  * to release the bp and vnode lock and do the uio to/from a system buffer,
180  * then regain the locks and copyback (if applicable).  XXX TODO.
181  */
182 int
183 uiomovebp(struct buf *bp, caddr_t cp, size_t n, struct uio *uio)
184 {
185 	int count;
186 	vm_page_t m;
187 
188 	if (bp->b_vp && bp->b_vp->v_type == VREG &&
189 	    (bp->b_flags & B_CACHE) &&
190 	    (count = bp->b_xio.xio_npages) != 0 &&
191 	    (m = bp->b_xio.xio_pages[count-1])->valid != VM_PAGE_BITS_ALL) {
192 		vm_page_zero_invalid(m, TRUE);
193 	}
194 	return (uiomove(cp, n, uio));
195 }
196 
197 /*
198  * uiomove() but fail for non-trivial VM faults, even if the VM fault is
199  * valid.  Returns EFAULT if a VM fault occurred via the copyin/copyout
200  * onfault code.
201  *
202  * This allows callers to hold e.g. a busy VM page, or a busy VM object,
203  * or a locked vnode through the call and then fall-back to safer code
204  * if we fail.
205  */
206 int
207 uiomove_nofault(caddr_t cp, size_t n, struct uio *uio)
208 {
209 	thread_t td = curthread;
210 	int error;
211 
212 	atomic_set_int(&td->td_flags, TDF_NOFAULT);
213 	error = uiomove(cp, n, uio);
214 	atomic_clear_int(&td->td_flags, TDF_NOFAULT);
215 	return error;
216 }
217 
218 /*
219  * Like uiomove() but copies zero-fill.  Only allowed for UIO_READ,
220  * for obvious reasons.
221  */
222 int
223 uiomovez(size_t n, struct uio *uio)
224 {
225 	struct iovec *iov;
226 	size_t cnt;
227 	int error = 0;
228 
229 	KASSERT(uio->uio_rw == UIO_READ, ("uiomovez: mode"));
230 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
231 		("uiomove proc"));
232 
233 	while (n > 0 && uio->uio_resid) {
234 		iov = uio->uio_iov;
235 		cnt = iov->iov_len;
236 		if (cnt == 0) {
237 			uio->uio_iov++;
238 			uio->uio_iovcnt--;
239 			continue;
240 		}
241 		if (cnt > n)
242 			cnt = n;
243 
244 		switch (uio->uio_segflg) {
245 		case UIO_USERSPACE:
246 			error = copyout(ZeroPage, iov->iov_base, cnt);
247 			break;
248 		case UIO_SYSSPACE:
249 			bzero(iov->iov_base, cnt);
250 			break;
251 		case UIO_NOCOPY:
252 			break;
253 		}
254 
255 		if (error)
256 			break;
257 		iov->iov_base = (char *)iov->iov_base + cnt;
258 		iov->iov_len -= cnt;
259 		uio->uio_resid -= cnt;
260 		uio->uio_offset += cnt;
261 		n -= cnt;
262 	}
263 	return (error);
264 }
265 
266 /*
267  * Wrapper for uiomove() that validates the arguments against a known-good
268  * kernel buffer.  This function automatically indexes the buffer by
269  * uio_offset and handles all range checking.
270  */
271 int
272 uiomove_frombuf(void *buf, size_t buflen, struct uio *uio)
273 {
274 	size_t offset;
275 
276 	offset = (size_t)uio->uio_offset;
277 	if ((off_t)offset != uio->uio_offset)
278 		return (EINVAL);
279 	if (buflen == 0 || offset >= buflen)
280 		return (0);
281 	return (uiomove((char *)buf + offset, buflen - offset, uio));
282 }
283 
284 /*
285  * Give next character to user as result of read.
286  */
287 int
288 ureadc(int c, struct uio *uio)
289 {
290 	struct iovec *iov;
291 	char *iov_base;
292 
293 again:
294 	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
295 		panic("ureadc");
296 	iov = uio->uio_iov;
297 	if (iov->iov_len == 0) {
298 		uio->uio_iovcnt--;
299 		uio->uio_iov++;
300 		goto again;
301 	}
302 
303 	switch (uio->uio_segflg) {
304 	case UIO_USERSPACE:
305 		if (subyte(iov->iov_base, c) < 0)
306 			return (EFAULT);
307 		break;
308 	case UIO_SYSSPACE:
309 		iov_base = iov->iov_base;
310 		*iov_base = c;
311 		iov->iov_base = iov_base;
312 		break;
313 	case UIO_NOCOPY:
314 		break;
315 	}
316 
317 	iov->iov_base = (char *)iov->iov_base + 1;
318 	iov->iov_len--;
319 	uio->uio_resid--;
320 	uio->uio_offset++;
321 	return (0);
322 }
323 
324 /*
325  * General routine to allocate a hash table.  Make the hash table size a
326  * power of 2 greater or equal to the number of elements requested, and
327  * store the masking value in *hashmask.
328  */
329 void *
330 hashinit(int elements, struct malloc_type *type, u_long *hashmask)
331 {
332 	long hashsize;
333 	LIST_HEAD(generic, generic) *hashtbl;
334 	int i;
335 
336 	if (elements <= 0)
337 		panic("hashinit: bad elements");
338 	for (hashsize = 2; hashsize < elements; hashsize <<= 1)
339 		continue;
340 	hashtbl = kmalloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
341 	for (i = 0; i < hashsize; i++)
342 		LIST_INIT(&hashtbl[i]);
343 	*hashmask = hashsize - 1;
344 	return (hashtbl);
345 }
346 
347 void
348 hashdestroy(void *vhashtbl, struct malloc_type *type, u_long hashmask)
349 {
350 	LIST_HEAD(generic, generic) *hashtbl, *hp;
351 
352 	hashtbl = vhashtbl;
353 	for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++)
354 		KASSERT(LIST_EMPTY(hp), ("%s: hash not empty", __func__));
355 	kfree(hashtbl, type);
356 }
357 
358 /*
359  * This is a newer version which allocates a hash table of structures.
360  *
361  * The returned array will be zero'd.  The caller is responsible for
362  * initializing the structures.
363  */
364 void *
365 hashinit_ext(int elements, size_t size, struct malloc_type *type,
366 	     u_long *hashmask)
367 {
368 	long hashsize;
369 	void *hashtbl;
370 
371 	if (elements <= 0)
372 		panic("hashinit: bad elements");
373 	for (hashsize = 2; hashsize < elements; hashsize <<= 1)
374 		continue;
375 	hashtbl = kmalloc((size_t)hashsize * size, type, M_WAITOK | M_ZERO);
376 	*hashmask = hashsize - 1;
377 	return (hashtbl);
378 }
379 
380 static int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039,
381 			2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653,
382 			7159, 7673, 8191, 12281, 16381, 24571, 32749 };
383 #define NPRIMES NELEM(primes)
384 
385 /*
386  * General routine to allocate a prime number sized hash table.
387  */
388 void *
389 phashinit(int elements, struct malloc_type *type, u_long *nentries)
390 {
391 	long hashsize;
392 	LIST_HEAD(generic, generic) *hashtbl;
393 	int i;
394 
395 	if (elements <= 0)
396 		panic("phashinit: bad elements");
397 	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
398 		i++;
399 		if (i == NPRIMES)
400 			break;
401 		hashsize = primes[i];
402 	}
403 	hashsize = primes[i - 1];
404 	hashtbl = kmalloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
405 	for (i = 0; i < hashsize; i++)
406 		LIST_INIT(&hashtbl[i]);
407 	*nentries = hashsize;
408 	return (hashtbl);
409 }
410 
411 /*
412  * This is a newer version which allocates a hash table of structures
413  * in a prime-number size.
414  *
415  * The returned array will be zero'd.  The caller is responsible for
416  * initializing the structures.
417  */
418 void *
419 phashinit_ext(int elements, size_t size, struct malloc_type *type,
420 	      u_long *nentries)
421 {
422 	long hashsize;
423 	void *hashtbl;
424 	int i;
425 
426 	if (elements <= 0)
427 		panic("phashinit: bad elements");
428 	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
429 		i++;
430 		if (i == NPRIMES)
431 			break;
432 		hashsize = primes[i];
433 	}
434 	hashsize = primes[i - 1];
435 	hashtbl = kmalloc((size_t)hashsize * size, type, M_WAITOK | M_ZERO);
436 	*nentries = hashsize;
437 	return (hashtbl);
438 }
439 
440 /*
441  * Copyin an iovec.  If the iovec array fits, use the preallocated small
442  * iovec structure.  If it is too big, dynamically allocate an iovec array
443  * of sufficient size.
444  *
445  * MPSAFE
446  */
447 int
448 iovec_copyin(struct iovec *uiov, struct iovec **kiov, struct iovec *siov,
449 	     size_t iov_cnt, size_t *iov_len)
450 {
451 	struct iovec *iovp;
452 	int error, i;
453 	size_t len;
454 
455 	if (iov_cnt > UIO_MAXIOV)
456 		return EMSGSIZE;
457 	if (iov_cnt > UIO_SMALLIOV) {
458 		*kiov = kmalloc(sizeof(struct iovec) * iov_cnt, M_IOV,
459 				M_WAITOK);
460 	} else {
461 		*kiov = siov;
462 	}
463 	error = copyin(uiov, *kiov, iov_cnt * sizeof(struct iovec));
464 	if (error == 0) {
465 		*iov_len = 0;
466 		for (i = 0, iovp = *kiov; i < iov_cnt; i++, iovp++) {
467 			/*
468 			 * Check for both *iov_len overflows and out of
469 			 * range iovp->iov_len's.  We limit to the
470 			 * capabilities of signed integers.
471 			 *
472 			 * GCC4 - overflow check opt requires assign/test.
473 			 */
474 			len = *iov_len + iovp->iov_len;
475 			if (len < *iov_len)
476 				error = EINVAL;
477 			*iov_len = len;
478 		}
479 	}
480 
481 	/*
482 	 * From userland disallow iovec's which exceed the sized size
483 	 * limit as the system calls return ssize_t.
484 	 *
485 	 * NOTE: Internal kernel interfaces can handle the unsigned
486 	 *	 limit.
487 	 */
488 	if (error == 0 && (ssize_t)*iov_len < 0)
489 		error = EINVAL;
490 
491 	if (error)
492 		iovec_free(kiov, siov);
493 	return (error);
494 }
495 
496 
497 /*
498  * Copyright (c) 2004 Alan L. Cox <alc@cs.rice.edu>
499  * Copyright (c) 1982, 1986, 1991, 1993
500  *	The Regents of the University of California.  All rights reserved.
501  * (c) UNIX System Laboratories, Inc.
502  * All or some portions of this file are derived from material licensed
503  * to the University of California by American Telephone and Telegraph
504  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
505  * the permission of UNIX System Laboratories, Inc.
506  *
507  * Redistribution and use in source and binary forms, with or without
508  * modification, are permitted provided that the following conditions
509  * are met:
510  * 1. Redistributions of source code must retain the above copyright
511  *    notice, this list of conditions and the following disclaimer.
512  * 2. Redistributions in binary form must reproduce the above copyright
513  *    notice, this list of conditions and the following disclaimer in the
514  *    documentation and/or other materials provided with the distribution.
515  * 4. Neither the name of the University nor the names of its contributors
516  *    may be used to endorse or promote products derived from this software
517  *    without specific prior written permission.
518  *
519  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
520  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
521  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
522  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
523  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
524  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
525  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
526  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
527  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
528  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
529  * SUCH DAMAGE.
530  *
531  * @(#)kern_subr.c	8.3 (Berkeley) 1/21/94
532  * $FreeBSD: src/sys/i386/i386/uio_machdep.c,v 1.1 2004/03/21 20:28:36 alc Exp $
533  */
534 
535 /*
536  * Implement uiomove(9) from physical memory using lwbuf's to reduce
537  * the creation and destruction of ephemeral mappings.
538  */
539 int
540 uiomove_fromphys(vm_page_t *ma, vm_offset_t offset, size_t n, struct uio *uio)
541 {
542 	struct lwbuf lwb_cache;
543 	struct lwbuf *lwb;
544 	struct thread *td = curthread;
545 	struct iovec *iov;
546 	void *cp;
547 	vm_offset_t page_offset;
548 	vm_page_t m;
549 	size_t cnt;
550 	int error = 0;
551 	int save = 0;
552 
553 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
554 	    ("uiomove_fromphys: mode"));
555 	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
556 	    ("uiomove_fromphys proc"));
557 
558 	crit_enter();
559 	save = td->td_flags & TDF_DEADLKTREAT;
560 	td->td_flags |= TDF_DEADLKTREAT;
561 	crit_exit();
562 
563 	while (n > 0 && uio->uio_resid) {
564 		iov = uio->uio_iov;
565 		cnt = iov->iov_len;
566 		if (cnt == 0) {
567 			uio->uio_iov++;
568 			uio->uio_iovcnt--;
569 			continue;
570 		}
571 		if (cnt > n)
572 			cnt = n;
573 		page_offset = offset & PAGE_MASK;
574 		cnt = min(cnt, PAGE_SIZE - page_offset);
575 		m = ma[offset >> PAGE_SHIFT];
576 		lwb = lwbuf_alloc(m, &lwb_cache);
577 		cp = (char *)lwbuf_kva(lwb) + page_offset;
578 
579 		switch (uio->uio_segflg) {
580 		case UIO_USERSPACE:
581 			/*
582 			 * note: removed uioyield (it was the wrong place to
583 			 * put it).
584 			 */
585 			if (uio->uio_rw == UIO_READ)
586 				error = copyout(cp, iov->iov_base, cnt);
587 			else
588 				error = copyin(iov->iov_base, cp, cnt);
589 			if (error) {
590 				lwbuf_free(lwb);
591 				goto out;
592 			}
593 			break;
594 		case UIO_SYSSPACE:
595 			if (uio->uio_rw == UIO_READ)
596 				bcopy(cp, iov->iov_base, cnt);
597 			else
598 				bcopy(iov->iov_base, cp, cnt);
599 			break;
600 		case UIO_NOCOPY:
601 			break;
602 		}
603 		lwbuf_free(lwb);
604 		iov->iov_base = (char *)iov->iov_base + cnt;
605 		iov->iov_len -= cnt;
606 		uio->uio_resid -= cnt;
607 		uio->uio_offset += cnt;
608 		offset += cnt;
609 		n -= cnt;
610 	}
611 out:
612 	if (save == 0) {
613 		crit_enter();
614 		td->td_flags &= ~TDF_DEADLKTREAT;
615 		crit_exit();
616 	}
617 	return (error);
618 }
619 
620