subr_hash.c revision 137377 
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 137377 2004-11-08 06:57:31Z 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	 * XXX The vm object containing kern_pg needs locking.
88	 */
89	if ((vm_map_lookup(&map, uaddr,
90			   VM_PROT_WRITE, &entry, &uobject,
91			   &upindex, &prot, &wired)) != KERN_SUCCESS) {
92		return(EFAULT);
93	}
94	VM_OBJECT_LOCK(uobject);
95	if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
96		do
97			vm_page_lock_queues();
98		while (vm_page_sleep_if_busy(user_pg, 1, "vm_pgmoveco"));
99		pmap_remove_all(user_pg);
100		vm_page_free(user_pg);
101	} else
102		vm_page_lock_queues();
103	if (kern_pg->busy || ((kern_pg->queue - kern_pg->pc) == PQ_FREE) ||
104	    (kern_pg->hold_count != 0)|| (kern_pg->flags & PG_BUSY)) {
105		printf("vm_pgmoveco: pindex(%lu), busy(%d), PG_BUSY(%d), "
106		       "hold(%d) paddr(0x%lx)\n", (u_long)kern_pg->pindex,
107			kern_pg->busy, (kern_pg->flags & PG_BUSY) ? 1 : 0,
108			kern_pg->hold_count, (u_long)kern_pg->phys_addr);
109		if ((kern_pg->queue - kern_pg->pc) == PQ_FREE)
110			panic("vm_pgmoveco: renaming free page");
111		else
112			panic("vm_pgmoveco: renaming busy page");
113	}
114	vm_page_rename(kern_pg, uobject, upindex);
115	kern_pg->valid = VM_PAGE_BITS_ALL;
116	vm_page_unlock_queues();
117	VM_OBJECT_UNLOCK(uobject);
118	vm_map_lookup_done(map, entry);
119	return(KERN_SUCCESS);
120}
121#endif /* ZERO_COPY_SOCKETS */
122
123int
124uiomove(void *cp, int n, struct uio *uio)
125{
126	struct thread *td = curthread;
127	struct iovec *iov;
128	u_int cnt;
129	int error = 0;
130	int save = 0;
131
132	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
133	    ("uiomove: mode"));
134	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
135	    ("uiomove proc"));
136	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
137	    "Calling uiomove()");
138
139	save = td->td_pflags & TDP_DEADLKTREAT;
140	td->td_pflags |= TDP_DEADLKTREAT;
141
142	while (n > 0 && uio->uio_resid) {
143		iov = uio->uio_iov;
144		cnt = iov->iov_len;
145		if (cnt == 0) {
146			uio->uio_iov++;
147			uio->uio_iovcnt--;
148			continue;
149		}
150		if (cnt > n)
151			cnt = n;
152
153		switch (uio->uio_segflg) {
154
155		case UIO_USERSPACE:
156			if (ticks - PCPU_GET(switchticks) >= hogticks)
157				uio_yield();
158			if (uio->uio_rw == UIO_READ)
159				error = copyout(cp, iov->iov_base, cnt);
160			else
161				error = copyin(iov->iov_base, cp, cnt);
162			if (error)
163				goto out;
164			break;
165
166		case UIO_SYSSPACE:
167			if (uio->uio_rw == UIO_READ)
168				bcopy(cp, iov->iov_base, cnt);
169			else
170				bcopy(iov->iov_base, cp, cnt);
171			break;
172		case UIO_NOCOPY:
173			break;
174		}
175		iov->iov_base = (char *)iov->iov_base + cnt;
176		iov->iov_len -= cnt;
177		uio->uio_resid -= cnt;
178		uio->uio_offset += cnt;
179		cp = (char *)cp + cnt;
180		n -= cnt;
181	}
182out:
183	if (save == 0)
184		td->td_pflags &= ~TDP_DEADLKTREAT;
185	return (error);
186}
187
188/*
189 * Wrapper for uiomove() that validates the arguments against a known-good
190 * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
191 * is almost definitely a bad thing, so we catch that here as well.  We
192 * return a runtime failure, but it might be desirable to generate a runtime
193 * assertion failure instead.
194 */
195int
196uiomove_frombuf(void *buf, int buflen, struct uio *uio)
197{
198	unsigned int offset, n;
199
200	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
201	    (offset = uio->uio_offset) != uio->uio_offset)
202		return (EINVAL);
203	if (buflen <= 0 || offset >= buflen)
204		return (0);
205	if ((n = buflen - offset) > INT_MAX)
206		return (EINVAL);
207	return (uiomove((char *)buf + offset, n, uio));
208}
209
210#ifdef ZERO_COPY_SOCKETS
211/*
212 * Experimental support for zero-copy I/O
213 */
214static int
215userspaceco(void *cp, u_int cnt, struct uio *uio, struct vm_object *obj,
216    int disposable)
217{
218	struct iovec *iov;
219	int error;
220
221	iov = uio->uio_iov;
222	if (uio->uio_rw == UIO_READ) {
223		if ((so_zero_copy_receive != 0)
224		 && (obj != NULL)
225		 && ((cnt & PAGE_MASK) == 0)
226		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
227		 && ((uio->uio_offset & PAGE_MASK) == 0)
228		 && ((((intptr_t) cp) & PAGE_MASK) == 0)
229		 && (obj->type == OBJT_DEFAULT)
230		 && (disposable != 0)) {
231			/* SOCKET: use page-trading */
232			/*
233			 * We only want to call vm_pgmoveco() on
234			 * disposeable pages, since it gives the
235			 * kernel page to the userland process.
236			 */
237			error =	vm_pgmoveco(&curproc->p_vmspace->vm_map,
238			    (vm_offset_t)cp, (vm_offset_t)iov->iov_base);
239
240			/*
241			 * If we get an error back, attempt
242			 * to use copyout() instead.  The
243			 * disposable page should be freed
244			 * automatically if we weren't able to move
245			 * it into userland.
246			 */
247			if (error != 0)
248				error = copyout(cp, iov->iov_base, cnt);
249		} else {
250			error = copyout(cp, iov->iov_base, cnt);
251		}
252	} else {
253		error = copyin(iov->iov_base, cp, cnt);
254	}
255	return (error);
256}
257
258int
259uiomoveco(void *cp, int n, struct uio *uio, struct vm_object *obj,
260    int disposable)
261{
262	struct iovec *iov;
263	u_int cnt;
264	int error;
265
266	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
267	    ("uiomoveco: mode"));
268	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
269	    ("uiomoveco proc"));
270
271	while (n > 0 && uio->uio_resid) {
272		iov = uio->uio_iov;
273		cnt = iov->iov_len;
274		if (cnt == 0) {
275			uio->uio_iov++;
276			uio->uio_iovcnt--;
277			continue;
278		}
279		if (cnt > n)
280			cnt = n;
281
282		switch (uio->uio_segflg) {
283
284		case UIO_USERSPACE:
285			if (ticks - PCPU_GET(switchticks) >= hogticks)
286				uio_yield();
287
288			error = userspaceco(cp, cnt, uio, obj, disposable);
289
290			if (error)
291				return (error);
292			break;
293
294		case UIO_SYSSPACE:
295			if (uio->uio_rw == UIO_READ)
296				bcopy(cp, iov->iov_base, cnt);
297			else
298				bcopy(iov->iov_base, cp, cnt);
299			break;
300		case UIO_NOCOPY:
301			break;
302		}
303		iov->iov_base = (char *)iov->iov_base + cnt;
304		iov->iov_len -= cnt;
305		uio->uio_resid -= cnt;
306		uio->uio_offset += cnt;
307		cp = (char *)cp + cnt;
308		n -= cnt;
309	}
310	return (0);
311}
312#endif /* ZERO_COPY_SOCKETS */
313
314/*
315 * Give next character to user as result of read.
316 */
317int
318ureadc(int c, struct uio *uio)
319{
320	struct iovec *iov;
321	char *iov_base;
322
323again:
324	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
325		panic("ureadc");
326	iov = uio->uio_iov;
327	if (iov->iov_len == 0) {
328		uio->uio_iovcnt--;
329		uio->uio_iov++;
330		goto again;
331	}
332	switch (uio->uio_segflg) {
333
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		iov->iov_base = iov_base;
343		break;
344
345	case UIO_NOCOPY:
346		break;
347	}
348	iov->iov_base = (char *)iov->iov_base + 1;
349	iov->iov_len--;
350	uio->uio_resid--;
351	uio->uio_offset++;
352	return (0);
353}
354
355/*
356 * General routine to allocate a hash table.
357 */
358void *
359hashinit(int elements, struct malloc_type *type, u_long *hashmask)
360{
361	long hashsize;
362	LIST_HEAD(generic, generic) *hashtbl;
363	int i;
364
365	if (elements <= 0)
366		panic("hashinit: bad elements");
367	for (hashsize = 1; hashsize <= elements; hashsize <<= 1)
368		continue;
369	hashsize >>= 1;
370	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
371	for (i = 0; i < hashsize; i++)
372		LIST_INIT(&hashtbl[i]);
373	*hashmask = hashsize - 1;
374	return (hashtbl);
375}
376
377void
378hashdestroy(void *vhashtbl, struct malloc_type *type, u_long hashmask)
379{
380	LIST_HEAD(generic, generic) *hashtbl, *hp;
381
382	hashtbl = vhashtbl;
383	for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++)
384		if (!LIST_EMPTY(hp))
385			panic("hashdestroy: hash not empty");
386	free(hashtbl, type);
387}
388
389static int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039,
390			2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653,
391			7159, 7673, 8191, 12281, 16381, 24571, 32749 };
392#define NPRIMES (sizeof(primes) / sizeof(primes[0]))
393
394/*
395 * General routine to allocate a prime number sized hash table.
396 */
397void *
398phashinit(int elements, struct malloc_type *type, u_long *nentries)
399{
400	long hashsize;
401	LIST_HEAD(generic, generic) *hashtbl;
402	int i;
403
404	if (elements <= 0)
405		panic("phashinit: bad elements");
406	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
407		i++;
408		if (i == NPRIMES)
409			break;
410		hashsize = primes[i];
411	}
412	hashsize = primes[i - 1];
413	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
414	for (i = 0; i < hashsize; i++)
415		LIST_INIT(&hashtbl[i]);
416	*nentries = hashsize;
417	return (hashtbl);
418}
419
420void
421uio_yield(void)
422{
423	struct thread *td;
424
425	td = curthread;
426	mtx_lock_spin(&sched_lock);
427	DROP_GIANT();
428	sched_prio(td, td->td_ksegrp->kg_user_pri); /* XXXKSE */
429	mi_switch(SW_INVOL, NULL);
430	mtx_unlock_spin(&sched_lock);
431	PICKUP_GIANT();
432}
433
434int
435copyinfrom(const void * __restrict src, void * __restrict dst, size_t len,
436    int seg)
437{
438	int error = 0;
439
440	switch (seg) {
441	case UIO_USERSPACE:
442		error = copyin(src, dst, len);
443		break;
444	case UIO_SYSSPACE:
445		bcopy(src, dst, len);
446		break;
447	default:
448		panic("copyinfrom: bad seg %d\n", seg);
449	}
450	return (error);
451}
452
453int
454copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len,
455    size_t * __restrict copied, int seg)
456{
457	int error = 0;
458
459	switch (seg) {
460	case UIO_USERSPACE:
461		error = copyinstr(src, dst, len, copied);
462		break;
463	case UIO_SYSSPACE:
464		error = copystr(src, dst, len, copied);
465		break;
466	default:
467		panic("copyinstrfrom: bad seg %d\n", seg);
468	}
469	return (error);
470}
471
472int
473copyiniov(struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error)
474{
475	u_int iovlen;
476
477	*iov = NULL;
478	if (iovcnt > UIO_MAXIOV)
479		return (error);
480	iovlen = iovcnt * sizeof (struct iovec);
481	*iov = malloc(iovlen, M_IOV, M_WAITOK);
482	error = copyin(iovp, *iov, iovlen);
483	if (error) {
484		free(*iov, M_IOV);
485		*iov = NULL;
486	}
487	return (error);
488}
489
490int
491copyinuio(struct iovec *iovp, u_int iovcnt, struct uio **uiop)
492{
493	struct iovec *iov;
494	struct uio *uio;
495	u_int iovlen;
496	int error, i;
497
498	*uiop = NULL;
499	if (iovcnt > UIO_MAXIOV)
500		return (EINVAL);
501	iovlen = iovcnt * sizeof (struct iovec);
502	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
503	iov = (struct iovec *)(uio + 1);
504	error = copyin(iovp, iov, iovlen);
505	if (error) {
506		free(uio, M_IOV);
507		return (error);
508	}
509	uio->uio_iov = iov;
510	uio->uio_iovcnt = iovcnt;
511	uio->uio_segflg = UIO_USERSPACE;
512	uio->uio_offset = -1;
513	uio->uio_resid = 0;
514	for (i = 0; i < iovcnt; i++) {
515		if (iov->iov_len > INT_MAX - uio->uio_resid) {
516			free(uio, M_IOV);
517			return (EINVAL);
518		}
519		uio->uio_resid += iov->iov_len;
520		iov++;
521	}
522	*uiop = uio;
523	return (0);
524}
525
526struct uio *
527cloneuio(struct uio *uiop)
528{
529	struct uio *uio;
530	int iovlen;
531
532	iovlen = uiop->uio_iovcnt * sizeof (struct iovec);
533	uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK);
534	*uio = *uiop;
535	uio->uio_iov = (struct iovec *)(uio + 1);
536	bcopy(uiop->uio_iov, uio->uio_iov, iovlen);
537	return (uio);
538}
539