subr_hash.c revision 99098 
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: head/sys/kern/kern_subr.c 99098 2002-06-30 02:07:26Z iedowse $
40 */
41
42#include "opt_zero.h"
43
44#include <sys/param.h>
45#include <sys/systm.h>
46#include <sys/kernel.h>
47#include <sys/ktr.h>
48#include <sys/lock.h>
49#include <sys/mutex.h>
50#include <sys/proc.h>
51#include <sys/malloc.h>
52#include <sys/resourcevar.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
60SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, NULL, UIO_MAXIOV,
61	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
62
63#ifdef ZERO_COPY_SOCKETS
64#include <vm/vm.h>
65#include <vm/vm_param.h>
66#include <sys/lock.h>
67#include <vm/pmap.h>
68#include <vm/vm_map.h>
69#include <vm/vm_page.h>
70#include <vm/vm_object.h>
71#include <vm/vm_pager.h>
72#include <vm/vm_kern.h>
73#include <vm/vm_extern.h>
74#include <vm/swap_pager.h>
75#include <sys/mbuf.h>
76#include <machine/cpu.h>
77
78/* Declared in uipc_socket.c */
79extern int so_zero_copy_receive;
80
81static int vm_pgmoveco(vm_map_t mapa, vm_object_t srcobj, vm_offset_t kaddr,
82		       vm_offset_t uaddr);
83static int userspaceco(caddr_t cp, u_int cnt, struct uio *uio,
84			    struct vm_object *obj, int disposable);
85
86static int
87vm_pgmoveco(mapa, srcobj,  kaddr, uaddr)
88        vm_map_t mapa;
89	vm_object_t srcobj;
90	vm_offset_t kaddr, uaddr;
91{
92	vm_map_t map = mapa;
93	vm_page_t kern_pg, user_pg;
94	vm_object_t uobject;
95	vm_map_entry_t entry;
96	vm_pindex_t upindex, kpindex;
97	vm_prot_t prot;
98	boolean_t wired;
99
100	/*
101	 * First lookup the kernel page.
102	 */
103	kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
104
105	if ((vm_map_lookup(&map, uaddr,
106			   VM_PROT_READ, &entry, &uobject,
107			   &upindex, &prot, &wired)) != KERN_SUCCESS) {
108		return(EFAULT);
109	}
110	if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
111		vm_page_sleep_busy(user_pg, 1, "vm_pgmoveco");
112		pmap_remove(map->pmap, uaddr, uaddr+PAGE_SIZE);
113		vm_page_busy(user_pg);
114		vm_page_free(user_pg);
115	}
116
117	if (kern_pg->busy || ((kern_pg->queue - kern_pg->pc) == PQ_FREE) ||
118	    (kern_pg->hold_count != 0)|| (kern_pg->flags & PG_BUSY)) {
119		printf("vm_pgmoveco: pindex(%lu), busy(%d), PG_BUSY(%d), "
120		       "hold(%d) paddr(0x%lx)\n", (u_long)kern_pg->pindex,
121			kern_pg->busy, (kern_pg->flags & PG_BUSY) ? 1 : 0,
122			kern_pg->hold_count, (u_long)kern_pg->phys_addr);
123		if ((kern_pg->queue - kern_pg->pc) == PQ_FREE)
124			panic("vm_pgmoveco: renaming free page");
125		else
126			panic("vm_pgmoveco: renaming busy page");
127	}
128	kpindex = kern_pg->pindex;
129	vm_page_busy(kern_pg);
130	vm_page_rename(kern_pg, uobject, upindex);
131	vm_page_flag_clear(kern_pg, PG_BUSY);
132	kern_pg->valid = VM_PAGE_BITS_ALL;
133
134	vm_map_lookup_done(map, entry);
135	return(KERN_SUCCESS);
136}
137#endif /* ZERO_COPY_SOCKETS */
138
139int
140uiomove(cp, n, uio)
141	register caddr_t cp;
142	register int n;
143	register struct uio *uio;
144{
145	struct thread *td = curthread;
146	register struct iovec *iov;
147	u_int cnt;
148	int error = 0;
149	int save = 0;
150
151	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
152	    ("uiomove: mode"));
153	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
154	    ("uiomove proc"));
155
156	if (td) {
157		mtx_lock_spin(&sched_lock);
158		save = td->td_flags & TDF_DEADLKTREAT;
159		td->td_flags |= TDF_DEADLKTREAT;
160		mtx_unlock_spin(&sched_lock);
161	}
162
163	while (n > 0 && uio->uio_resid) {
164		iov = uio->uio_iov;
165		cnt = iov->iov_len;
166		if (cnt == 0) {
167			uio->uio_iov++;
168			uio->uio_iovcnt--;
169			continue;
170		}
171		if (cnt > n)
172			cnt = n;
173
174		switch (uio->uio_segflg) {
175
176		case UIO_USERSPACE:
177			if (ticks - PCPU_GET(switchticks) >= hogticks)
178				uio_yield();
179			if (uio->uio_rw == UIO_READ)
180				error = copyout(cp, iov->iov_base, cnt);
181			else
182				error = copyin(iov->iov_base, cp, cnt);
183			if (error)
184				goto out;
185			break;
186
187		case UIO_SYSSPACE:
188			if (uio->uio_rw == UIO_READ)
189				bcopy(cp, iov->iov_base, cnt);
190			else
191				bcopy(iov->iov_base, cp, cnt);
192			break;
193		case UIO_NOCOPY:
194			break;
195		}
196		iov->iov_base += cnt;
197		iov->iov_len -= cnt;
198		uio->uio_resid -= cnt;
199		uio->uio_offset += cnt;
200		cp += cnt;
201		n -= cnt;
202	}
203out:
204	if (td != curthread) printf("uiomove: IT CHANGED!");
205	td = curthread;	/* Might things have changed in copyin/copyout? */
206	if (td) {
207		mtx_lock_spin(&sched_lock);
208		td->td_flags = (td->td_flags & ~TDF_DEADLKTREAT) | save;
209		mtx_unlock_spin(&sched_lock);
210	}
211	return (error);
212}
213
214#if defined(ENABLE_VFS_IOOPT) || defined(ZERO_COPY_SOCKETS)
215/*
216 * Experimental support for zero-copy I/O
217 */
218static int
219userspaceco(cp, cnt, uio, obj, disposable)
220	caddr_t cp;
221	u_int cnt;
222	struct uio *uio;
223	struct vm_object *obj;
224	int disposable;
225{
226	struct iovec *iov;
227	int error;
228
229	iov = uio->uio_iov;
230
231#ifdef ZERO_COPY_SOCKETS
232
233	if (uio->uio_rw == UIO_READ) {
234		if ((so_zero_copy_receive != 0)
235		 && (obj != NULL)
236		 && ((cnt & PAGE_MASK) == 0)
237		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
238		 && ((uio->uio_offset & PAGE_MASK) == 0)
239		 && ((((intptr_t) cp) & PAGE_MASK) == 0)
240		 && (obj->type == OBJT_DEFAULT)
241		 && (disposable != 0)) {
242			/* SOCKET: use page-trading */
243			/*
244			 * We only want to call vm_pgmoveco() on
245			 * disposeable pages, since it gives the
246			 * kernel page to the userland process.
247			 */
248			error =	vm_pgmoveco(&curproc->p_vmspace->vm_map,
249					    obj, (vm_offset_t)cp,
250					    (vm_offset_t)iov->iov_base);
251
252			/*
253			 * If we get an error back, attempt
254			 * to use copyout() instead.  The
255			 * disposable page should be freed
256			 * automatically if we weren't able to move
257			 * it into userland.
258			 */
259			if (error != 0)
260				error = copyout(cp, iov->iov_base, cnt);
261#ifdef ENABLE_VFS_IOOPT
262		} else if ((vfs_ioopt != 0)
263		 && ((cnt & PAGE_MASK) == 0)
264		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
265		 && ((uio->uio_offset & PAGE_MASK) == 0)
266		 && ((((intptr_t) cp) & PAGE_MASK) == 0)) {
267			error = vm_uiomove(&curproc->p_vmspace->vm_map, obj,
268					   uio->uio_offset, cnt,
269					   (vm_offset_t) iov->iov_base, NULL);
270#endif /* ENABLE_VFS_IOOPT */
271		} else {
272			error = copyout(cp, iov->iov_base, cnt);
273		}
274	} else {
275		error = copyin(iov->iov_base, cp, cnt);
276	}
277#else /* ZERO_COPY_SOCKETS */
278	if (uio->uio_rw == UIO_READ) {
279#ifdef ENABLE_VFS_IOOPT
280		if ((vfs_ioopt != 0)
281		 && ((cnt & PAGE_MASK) == 0)
282		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
283		 && ((uio->uio_offset & PAGE_MASK) == 0)
284		 && ((((intptr_t) cp) & PAGE_MASK) == 0)) {
285			error = vm_uiomove(&curproc->p_vmspace->vm_map, obj,
286					   uio->uio_offset, cnt,
287					   (vm_offset_t) iov->iov_base, NULL);
288		} else
289#endif /* ENABLE_VFS_IOOPT */
290		{
291			error = copyout(cp, iov->iov_base, cnt);
292		}
293	} else {
294		error = copyin(iov->iov_base, cp, cnt);
295	}
296#endif /* ZERO_COPY_SOCKETS */
297
298	return (error);
299}
300
301int
302uiomoveco(cp, n, uio, obj, disposable)
303	caddr_t cp;
304	int n;
305	struct uio *uio;
306	struct vm_object *obj;
307	int disposable;
308{
309	struct iovec *iov;
310	u_int cnt;
311	int error;
312
313	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
314	    ("uiomoveco: mode"));
315	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
316	    ("uiomoveco proc"));
317
318	while (n > 0 && uio->uio_resid) {
319		iov = uio->uio_iov;
320		cnt = iov->iov_len;
321		if (cnt == 0) {
322			uio->uio_iov++;
323			uio->uio_iovcnt--;
324			continue;
325		}
326		if (cnt > n)
327			cnt = n;
328
329		switch (uio->uio_segflg) {
330
331		case UIO_USERSPACE:
332			if (ticks - PCPU_GET(switchticks) >= hogticks)
333				uio_yield();
334
335			error = userspaceco(cp, cnt, uio, obj, disposable);
336
337			if (error)
338				return (error);
339			break;
340
341		case UIO_SYSSPACE:
342			if (uio->uio_rw == UIO_READ)
343				bcopy(cp, iov->iov_base, cnt);
344			else
345				bcopy(iov->iov_base, cp, cnt);
346			break;
347		case UIO_NOCOPY:
348			break;
349		}
350		iov->iov_base += cnt;
351		iov->iov_len -= cnt;
352		uio->uio_resid -= cnt;
353		uio->uio_offset += cnt;
354		cp += cnt;
355		n -= cnt;
356	}
357	return (0);
358}
359#endif /* ENABLE_VFS_IOOPT || ZERO_COPY_SOCKETS */
360
361#ifdef ENABLE_VFS_IOOPT
362
363/*
364 * Experimental support for zero-copy I/O
365 */
366int
367uioread(n, uio, obj, nread)
368	int n;
369	struct uio *uio;
370	struct vm_object *obj;
371	int *nread;
372{
373	int npagesmoved;
374	struct iovec *iov;
375	u_int cnt, tcnt;
376	int error;
377
378	*nread = 0;
379	if (vfs_ioopt < 2)
380		return 0;
381
382	error = 0;
383
384	while (n > 0 && uio->uio_resid) {
385		iov = uio->uio_iov;
386		cnt = iov->iov_len;
387		if (cnt == 0) {
388			uio->uio_iov++;
389			uio->uio_iovcnt--;
390			continue;
391		}
392		if (cnt > n)
393			cnt = n;
394
395		if ((uio->uio_segflg == UIO_USERSPACE) &&
396			((((intptr_t) iov->iov_base) & PAGE_MASK) == 0) &&
397				 ((uio->uio_offset & PAGE_MASK) == 0) ) {
398
399			if (cnt < PAGE_SIZE)
400				break;
401
402			cnt &= ~PAGE_MASK;
403
404			if (ticks - PCPU_GET(switchticks) >= hogticks)
405				uio_yield();
406			error = vm_uiomove(&curproc->p_vmspace->vm_map, obj,
407						uio->uio_offset, cnt,
408						(vm_offset_t) iov->iov_base, &npagesmoved);
409
410			if (npagesmoved == 0)
411				break;
412
413			tcnt = npagesmoved * PAGE_SIZE;
414			cnt = tcnt;
415
416			if (error)
417				break;
418
419			iov->iov_base += cnt;
420			iov->iov_len -= cnt;
421			uio->uio_resid -= cnt;
422			uio->uio_offset += cnt;
423			*nread += cnt;
424			n -= cnt;
425		} else {
426			break;
427		}
428	}
429	return error;
430}
431#endif /* ENABLE_VFS_IOOPT */
432
433/*
434 * Give next character to user as result of read.
435 */
436int
437ureadc(c, uio)
438	register int c;
439	register struct uio *uio;
440{
441	register struct iovec *iov;
442
443again:
444	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
445		panic("ureadc");
446	iov = uio->uio_iov;
447	if (iov->iov_len == 0) {
448		uio->uio_iovcnt--;
449		uio->uio_iov++;
450		goto again;
451	}
452	switch (uio->uio_segflg) {
453
454	case UIO_USERSPACE:
455		if (subyte(iov->iov_base, c) < 0)
456			return (EFAULT);
457		break;
458
459	case UIO_SYSSPACE:
460		*iov->iov_base = c;
461		break;
462
463	case UIO_NOCOPY:
464		break;
465	}
466	iov->iov_base++;
467	iov->iov_len--;
468	uio->uio_resid--;
469	uio->uio_offset++;
470	return (0);
471}
472
473/*
474 * General routine to allocate a hash table.
475 */
476void *
477hashinit(elements, type, hashmask)
478	int elements;
479	struct malloc_type *type;
480	u_long *hashmask;
481{
482	long hashsize;
483	LIST_HEAD(generic, generic) *hashtbl;
484	int i;
485
486	if (elements <= 0)
487		panic("hashinit: bad elements");
488	for (hashsize = 1; hashsize <= elements; hashsize <<= 1)
489		continue;
490	hashsize >>= 1;
491	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
492	for (i = 0; i < hashsize; i++)
493		LIST_INIT(&hashtbl[i]);
494	*hashmask = hashsize - 1;
495	return (hashtbl);
496}
497
498void
499hashdestroy(vhashtbl, type, hashmask)
500	void *vhashtbl;
501	struct malloc_type *type;
502	u_long hashmask;
503{
504	LIST_HEAD(generic, generic) *hashtbl, *hp;
505
506	hashtbl = vhashtbl;
507	for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++)
508		if (!LIST_EMPTY(hp))
509			panic("hashdestroy: hash not empty");
510	free(hashtbl, type);
511}
512
513static int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039,
514			2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653,
515			7159, 7673, 8191, 12281, 16381, 24571, 32749 };
516#define NPRIMES (sizeof(primes) / sizeof(primes[0]))
517
518/*
519 * General routine to allocate a prime number sized hash table.
520 */
521void *
522phashinit(elements, type, nentries)
523	int elements;
524	struct malloc_type *type;
525	u_long *nentries;
526{
527	long hashsize;
528	LIST_HEAD(generic, generic) *hashtbl;
529	int i;
530
531	if (elements <= 0)
532		panic("phashinit: bad elements");
533	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
534		i++;
535		if (i == NPRIMES)
536			break;
537		hashsize = primes[i];
538	}
539	hashsize = primes[i - 1];
540	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
541	for (i = 0; i < hashsize; i++)
542		LIST_INIT(&hashtbl[i]);
543	*nentries = hashsize;
544	return (hashtbl);
545}
546
547void
548uio_yield()
549{
550	struct thread *td;
551
552	td = curthread;
553	mtx_lock_spin(&sched_lock);
554	DROP_GIANT();
555	td->td_priority = td->td_ksegrp->kg_user_pri; /* XXXKSE */
556	td->td_proc->p_stats->p_ru.ru_nivcsw++;
557	mi_switch();
558	mtx_unlock_spin(&sched_lock);
559	PICKUP_GIANT();
560}
561
562int
563copyinfrom(const void *src, void *dst, size_t len, int seg)
564{
565	int error = 0;
566
567	switch (seg) {
568	case UIO_USERSPACE:
569		error = copyin(src, dst, len);
570		break;
571	case UIO_SYSSPACE:
572		bcopy(src, dst, len);
573		break;
574	default:
575		panic("copyinfrom: bad seg %d\n", seg);
576	}
577	return (error);
578}
579
580int
581copyinstrfrom(const void *src, void *dst, size_t len, size_t *copied, int seg)
582{
583	int error = 0;
584
585	switch (seg) {
586	case UIO_USERSPACE:
587		error = copyinstr(src, dst, len, copied);
588		break;
589	case UIO_SYSSPACE:
590		error = copystr(src, dst, len, copied);
591		break;
592	default:
593		panic("copyinstrfrom: bad seg %d\n", seg);
594	}
595	return (error);
596}
597