subr_hash.c revision 138424
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: head/sys/kern/kern_subr.c 138424 2004-12-06 00:43:40Z alc $");
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/ktr.h>
46#include <sys/limits.h>
47#include <sys/lock.h>
48#include <sys/mutex.h>
49#include <sys/proc.h>
50#include <sys/malloc.h>
51#include <sys/resourcevar.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_page.h>
58#include <vm/vm_map.h>
59#ifdef ZERO_COPY_SOCKETS
60#include <vm/vm_param.h>
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
67#ifdef ZERO_COPY_SOCKETS
68/* Declared in uipc_socket.c */
69extern int so_zero_copy_receive;
70
71static int
72vm_pgmoveco(vm_map_t mapa, vm_offset_t kaddr, vm_offset_t uaddr)
73{
74	vm_map_t map = mapa;
75	vm_page_t kern_pg, user_pg;
76	vm_object_t uobject;
77	vm_map_entry_t entry;
78	vm_pindex_t upindex;
79	vm_prot_t prot;
80	boolean_t wired;
81
82	/*
83	 * First lookup the kernel page.
84	 */
85	kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
86
87	if ((vm_map_lookup(&map, uaddr,
88			   VM_PROT_WRITE, &entry, &uobject,
89			   &upindex, &prot, &wired)) != KERN_SUCCESS) {
90		return(EFAULT);
91	}
92	VM_OBJECT_LOCK(uobject);
93	if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
94		do
95			vm_page_lock_queues();
96		while (vm_page_sleep_if_busy(user_pg, 1, "vm_pgmoveco"));
97		pmap_remove_all(user_pg);
98		vm_page_free(user_pg);
99	} else
100		vm_page_lock_queues();
101	if (kern_pg->busy || ((kern_pg->queue - kern_pg->pc) == PQ_FREE) ||
102	    (kern_pg->hold_count != 0)|| (kern_pg->flags & PG_BUSY)) {
103		printf("vm_pgmoveco: pindex(%lu), busy(%d), PG_BUSY(%d), "
104		       "hold(%d) paddr(0x%lx)\n", (u_long)kern_pg->pindex,
105			kern_pg->busy, (kern_pg->flags & PG_BUSY) ? 1 : 0,
106			kern_pg->hold_count, (u_long)kern_pg->phys_addr);
107		if ((kern_pg->queue - kern_pg->pc) == PQ_FREE)
108			panic("vm_pgmoveco: renaming free page");
109		else
110			panic("vm_pgmoveco: renaming busy page");
111	}
112	vm_page_insert(kern_pg, uobject, upindex);
113	vm_page_dirty(kern_pg);
114	kern_pg->valid = VM_PAGE_BITS_ALL;
115	vm_page_unlock_queues();
116	VM_OBJECT_UNLOCK(uobject);
117	vm_map_lookup_done(map, entry);
118	return(KERN_SUCCESS);
119}
120#endif /* ZERO_COPY_SOCKETS */
121
122int
123uiomove(void *cp, int n, struct uio *uio)
124{
125	struct thread *td = curthread;
126	struct iovec *iov;
127	u_int cnt;
128	int error = 0;
129	int save = 0;
130
131	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
132	    ("uiomove: mode"));
133	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
134	    ("uiomove proc"));
135	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
136	    "Calling uiomove()");
137
138	save = td->td_pflags & TDP_DEADLKTREAT;
139	td->td_pflags |= TDP_DEADLKTREAT;
140
141	while (n > 0 && uio->uio_resid) {
142		iov = uio->uio_iov;
143		cnt = iov->iov_len;
144		if (cnt == 0) {
145			uio->uio_iov++;
146			uio->uio_iovcnt--;
147			continue;
148		}
149		if (cnt > n)
150			cnt = n;
151
152		switch (uio->uio_segflg) {
153
154		case UIO_USERSPACE:
155			if (ticks - PCPU_GET(switchticks) >= hogticks)
156				uio_yield();
157			if (uio->uio_rw == UIO_READ)
158				error = copyout(cp, iov->iov_base, cnt);
159			else
160				error = copyin(iov->iov_base, cp, cnt);
161			if (error)
162				goto out;
163			break;
164
165		case UIO_SYSSPACE:
166			if (uio->uio_rw == UIO_READ)
167				bcopy(cp, iov->iov_base, cnt);
168			else
169				bcopy(iov->iov_base, cp, cnt);
170			break;
171		case UIO_NOCOPY:
172			break;
173		}
174		iov->iov_base = (char *)iov->iov_base + cnt;
175		iov->iov_len -= cnt;
176		uio->uio_resid -= cnt;
177		uio->uio_offset += cnt;
178		cp = (char *)cp + cnt;
179		n -= cnt;
180	}
181out:
182	if (save == 0)
183		td->td_pflags &= ~TDP_DEADLKTREAT;
184	return (error);
185}
186
187/*
188 * Wrapper for uiomove() that validates the arguments against a known-good
189 * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
190 * is almost definitely a bad thing, so we catch that here as well.  We
191 * return a runtime failure, but it might be desirable to generate a runtime
192 * assertion failure instead.
193 */
194int
195uiomove_frombuf(void *buf, int buflen, struct uio *uio)
196{
197	unsigned int offset, n;
198
199	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
200	    (offset = uio->uio_offset) != uio->uio_offset)
201		return (EINVAL);
202	if (buflen <= 0 || offset >= buflen)
203		return (0);
204	if ((n = buflen - offset) > INT_MAX)
205		return (EINVAL);
206	return (uiomove((char *)buf + offset, n, uio));
207}
208
209#ifdef ZERO_COPY_SOCKETS
210/*
211 * Experimental support for zero-copy I/O
212 */
213static int
214userspaceco(void *cp, u_int cnt, struct uio *uio, struct vm_object *obj,
215    int disposable)
216{
217	struct iovec *iov;
218	int error;
219
220	iov = uio->uio_iov;
221	if (uio->uio_rw == UIO_READ) {
222		if ((so_zero_copy_receive != 0)
223		 && (obj == NULL)
224		 && ((cnt & PAGE_MASK) == 0)
225		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
226		 && ((uio->uio_offset & PAGE_MASK) == 0)
227		 && ((((intptr_t) cp) & PAGE_MASK) == 0)
228		 && (disposable != 0)) {
229			/* SOCKET: use page-trading */
230			/*
231			 * We only want to call vm_pgmoveco() on
232			 * disposeable pages, since it gives the
233			 * kernel page to the userland process.
234			 */
235			error =	vm_pgmoveco(&curproc->p_vmspace->vm_map,
236			    (vm_offset_t)cp, (vm_offset_t)iov->iov_base);
237
238			/*
239			 * If we get an error back, attempt
240			 * to use copyout() instead.  The
241			 * disposable page should be freed
242			 * automatically if we weren't able to move
243			 * it into userland.
244			 */
245			if (error != 0)
246				error = copyout(cp, iov->iov_base, cnt);
247		} else {
248			error = copyout(cp, iov->iov_base, cnt);
249		}
250	} else {
251		error = copyin(iov->iov_base, cp, cnt);
252	}
253	return (error);
254}
255
256int
257uiomoveco(void *cp, int n, struct uio *uio, struct vm_object *obj,
258    int disposable)
259{
260	struct iovec *iov;
261	u_int cnt;
262	int error;
263
264	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
265	    ("uiomoveco: mode"));
266	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
267	    ("uiomoveco proc"));
268
269	while (n > 0 && uio->uio_resid) {
270		iov = uio->uio_iov;
271		cnt = iov->iov_len;
272		if (cnt == 0) {
273			uio->uio_iov++;
274			uio->uio_iovcnt--;
275			continue;
276		}
277		if (cnt > n)
278			cnt = n;
279
280		switch (uio->uio_segflg) {
281
282		case UIO_USERSPACE:
283			if (ticks - PCPU_GET(switchticks) >= hogticks)
284				uio_yield();
285
286			error = userspaceco(cp, cnt, uio, obj, disposable);
287
288			if (error)
289				return (error);
290			break;
291
292		case UIO_SYSSPACE:
293			if (uio->uio_rw == UIO_READ)
294				bcopy(cp, iov->iov_base, cnt);
295			else
296				bcopy(iov->iov_base, cp, cnt);
297			break;
298		case UIO_NOCOPY:
299			break;
300		}
301		iov->iov_base = (char *)iov->iov_base + cnt;
302		iov->iov_len -= cnt;
303		uio->uio_resid -= cnt;
304		uio->uio_offset += cnt;
305		cp = (char *)cp + cnt;
306		n -= cnt;
307	}
308	return (0);
309}
310#endif /* ZERO_COPY_SOCKETS */
311
312/*
313 * Give next character to user as result of read.
314 */
315int
316ureadc(int c, struct uio *uio)
317{
318	struct iovec *iov;
319	char *iov_base;
320
321again:
322	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
323		panic("ureadc");
324	iov = uio->uio_iov;
325	if (iov->iov_len == 0) {
326		uio->uio_iovcnt--;
327		uio->uio_iov++;
328		goto again;
329	}
330	switch (uio->uio_segflg) {
331
332	case UIO_USERSPACE:
333		if (subyte(iov->iov_base, c) < 0)
334			return (EFAULT);
335		break;
336
337	case UIO_SYSSPACE:
338		iov_base = iov->iov_base;
339		*iov_base = c;
340		iov->iov_base = iov_base;
341		break;
342
343	case UIO_NOCOPY:
344		break;
345	}
346	iov->iov_base = (char *)iov->iov_base + 1;
347	iov->iov_len--;
348	uio->uio_resid--;
349	uio->uio_offset++;
350	return (0);
351}
352
353/*
354 * General routine to allocate a hash table.
355 */
356void *
357hashinit(int elements, struct malloc_type *type, u_long *hashmask)
358{
359	long hashsize;
360	LIST_HEAD(generic, generic) *hashtbl;
361	int i;
362
363	if (elements <= 0)
364		panic("hashinit: bad elements");
365	for (hashsize = 1; hashsize <= elements; hashsize <<= 1)
366		continue;
367	hashsize >>= 1;
368	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
369	for (i = 0; i < hashsize; i++)
370		LIST_INIT(&hashtbl[i]);
371	*hashmask = hashsize - 1;
372	return (hashtbl);
373}
374
375void
376hashdestroy(void *vhashtbl, struct malloc_type *type, u_long hashmask)
377{
378	LIST_HEAD(generic, generic) *hashtbl, *hp;
379
380	hashtbl = vhashtbl;
381	for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++)
382		if (!LIST_EMPTY(hp))
383			panic("hashdestroy: hash not empty");
384	free(hashtbl, type);
385}
386
387static int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039,
388			2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653,
389			7159, 7673, 8191, 12281, 16381, 24571, 32749 };
390#define NPRIMES (sizeof(primes) / sizeof(primes[0]))
391
392/*
393 * General routine to allocate a prime number sized hash table.
394 */
395void *
396phashinit(int elements, struct malloc_type *type, u_long *nentries)
397{
398	long hashsize;
399	LIST_HEAD(generic, generic) *hashtbl;
400	int i;
401
402	if (elements <= 0)
403		panic("phashinit: bad elements");
404	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
405		i++;
406		if (i == NPRIMES)
407			break;
408		hashsize = primes[i];
409	}
410	hashsize = primes[i - 1];
411	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
412	for (i = 0; i < hashsize; i++)
413		LIST_INIT(&hashtbl[i]);
414	*nentries = hashsize;
415	return (hashtbl);
416}
417
418void
419uio_yield(void)
420{
421	struct thread *td;
422
423	td = curthread;
424	mtx_lock_spin(&sched_lock);
425	DROP_GIANT();
426	sched_prio(td, td->td_ksegrp->kg_user_pri); /* XXXKSE */
427	mi_switch(SW_INVOL, NULL);
428	mtx_unlock_spin(&sched_lock);
429	PICKUP_GIANT();
430}
431
432int
433copyinfrom(const void * __restrict src, void * __restrict dst, size_t len,
434    int seg)
435{
436	int error = 0;
437
438	switch (seg) {
439	case UIO_USERSPACE:
440		error = copyin(src, dst, len);
441		break;
442	case UIO_SYSSPACE:
443		bcopy(src, dst, len);
444		break;
445	default:
446		panic("copyinfrom: bad seg %d\n", seg);
447	}
448	return (error);
449}
450
451int
452copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len,
453    size_t * __restrict copied, int seg)
454{
455	int error = 0;
456
457	switch (seg) {
458	case UIO_USERSPACE:
459		error = copyinstr(src, dst, len, copied);
460		break;
461	case UIO_SYSSPACE:
462		error = copystr(src, dst, len, copied);
463		break;
464	default:
465		panic("copyinstrfrom: bad seg %d\n", seg);
466	}
467	return (error);
468}
469
470int
471copyiniov(struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
472{
473	u_int iovlen;
474
475	*iov = NULL;
476	if (iovcnt > UIO_MAXIOV)
477		return (error);
478	iovlen = iovcnt * sizeof (struct iovec);
479	*iov = malloc(iovlen, M_IOV, M_WAITOK);
480	error = copyin(iovp, *iov, iovlen);
481	if (error) {
482		free(*iov, M_IOV);
483		*iov = NULL;
484	}
485	return (error);
486}
487
488int
489copyinuio(struct iovec *iovp, u_int iovcnt, struct uio **uiop)
490{
491	struct iovec *iov;
492	struct uio *uio;
493	u_int iovlen;
494	int error, i;
495
496	*uiop = NULL;
497	if (iovcnt > UIO_MAXIOV)
498		return (EINVAL);
499	iovlen = iovcnt * sizeof (struct iovec);
500	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
501	iov = (struct iovec *)(uio + 1);
502	error = copyin(iovp, iov, iovlen);
503	if (error) {
504		free(uio, M_IOV);
505		return (error);
506	}
507	uio->uio_iov = iov;
508	uio->uio_iovcnt = iovcnt;
509	uio->uio_segflg = UIO_USERSPACE;
510	uio->uio_offset = -1;
511	uio->uio_resid = 0;
512	for (i = 0; i < iovcnt; i++) {
513		if (iov->iov_len > INT_MAX - uio->uio_resid) {
514			free(uio, M_IOV);
515			return (EINVAL);
516		}
517		uio->uio_resid += iov->iov_len;
518		iov++;
519	}
520	*uiop = uio;
521	return (0);
522}
523
524struct uio *
525cloneuio(struct uio *uiop)
526{
527	struct uio *uio;
528	int iovlen;
529
530	iovlen = uiop->uio_iovcnt * sizeof (struct iovec);
531	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
532	*uio = *uiop;
533	uio->uio_iov = (struct iovec *)(uio + 1);
534	bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
535	return (uio);
536}
537