subr_hash.c revision 111739 
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 111739 2003-03-02 15:29:13Z des $
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/sched.h>
54#include <sys/sysctl.h>
55#include <sys/vnode.h>
56
57#include <vm/vm.h>
58#include <vm/vm_page.h>
59#include <vm/vm_map.h>
60#ifdef ZERO_COPY_SOCKETS
61#include <vm/vm_param.h>
62#endif
63#if defined(ZERO_COPY_SOCKETS) || defined(ENABLE_VFS_IOOPT)
64#include <vm/vm_object.h>
65#endif
66
67SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, NULL, UIO_MAXIOV,
68	"Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)");
69
70#ifdef ZERO_COPY_SOCKETS
71/* Declared in uipc_socket.c */
72extern int so_zero_copy_receive;
73
74static int
75vm_pgmoveco(vm_map_t mapa, vm_object_t srcobj, vm_offset_t kaddr,
76    vm_offset_t uaddr)
77{
78	vm_map_t map = mapa;
79	vm_page_t kern_pg, user_pg;
80	vm_object_t uobject;
81	vm_map_entry_t entry;
82	vm_pindex_t upindex, kpindex;
83	vm_prot_t prot;
84	boolean_t wired;
85
86	/*
87	 * First lookup the kernel page.
88	 */
89	kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
90
91	if ((vm_map_lookup(&map, uaddr,
92			   VM_PROT_READ, &entry, &uobject,
93			   &upindex, &prot, &wired)) != KERN_SUCCESS) {
94		return(EFAULT);
95	}
96	if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
97		do
98			vm_page_lock_queues();
99		while (vm_page_sleep_if_busy(user_pg, 1, "vm_pgmoveco"));
100		vm_page_busy(user_pg);
101		pmap_remove_all(user_pg);
102		vm_page_free(user_pg);
103	} else
104		vm_page_lock_queues();
105	if (kern_pg->busy || ((kern_pg->queue - kern_pg->pc) == PQ_FREE) ||
106	    (kern_pg->hold_count != 0)|| (kern_pg->flags & PG_BUSY)) {
107		printf("vm_pgmoveco: pindex(%lu), busy(%d), PG_BUSY(%d), "
108		       "hold(%d) paddr(0x%lx)\n", (u_long)kern_pg->pindex,
109			kern_pg->busy, (kern_pg->flags & PG_BUSY) ? 1 : 0,
110			kern_pg->hold_count, (u_long)kern_pg->phys_addr);
111		if ((kern_pg->queue - kern_pg->pc) == PQ_FREE)
112			panic("vm_pgmoveco: renaming free page");
113		else
114			panic("vm_pgmoveco: renaming busy page");
115	}
116	kpindex = kern_pg->pindex;
117	vm_page_busy(kern_pg);
118	vm_page_rename(kern_pg, uobject, upindex);
119	vm_page_flag_clear(kern_pg, PG_BUSY);
120	kern_pg->valid = VM_PAGE_BITS_ALL;
121	vm_page_unlock_queues();
122
123	vm_map_lookup_done(map, entry);
124	return(KERN_SUCCESS);
125}
126#endif /* ZERO_COPY_SOCKETS */
127
128int
129uiomove(void *cp, int n, struct uio *uio)
130{
131	struct thread *td = curthread;
132	struct iovec *iov;
133	u_int cnt;
134	int error = 0;
135	int save = 0;
136
137	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
138	    ("uiomove: mode"));
139	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
140	    ("uiomove proc"));
141
142	if (td) {
143		mtx_lock_spin(&sched_lock);
144		save = td->td_flags & TDF_DEADLKTREAT;
145		td->td_flags |= TDF_DEADLKTREAT;
146		mtx_unlock_spin(&sched_lock);
147	}
148
149	while (n > 0 && uio->uio_resid) {
150		iov = uio->uio_iov;
151		cnt = iov->iov_len;
152		if (cnt == 0) {
153			uio->uio_iov++;
154			uio->uio_iovcnt--;
155			continue;
156		}
157		if (cnt > n)
158			cnt = n;
159
160		switch (uio->uio_segflg) {
161
162		case UIO_USERSPACE:
163			if (ticks - PCPU_GET(switchticks) >= hogticks)
164				uio_yield();
165			if (uio->uio_rw == UIO_READ)
166				error = copyout(cp, iov->iov_base, cnt);
167			else
168				error = copyin(iov->iov_base, cp, cnt);
169			if (error)
170				goto out;
171			break;
172
173		case UIO_SYSSPACE:
174			if (uio->uio_rw == UIO_READ)
175				bcopy(cp, iov->iov_base, cnt);
176			else
177				bcopy(iov->iov_base, cp, cnt);
178			break;
179		case UIO_NOCOPY:
180			break;
181		}
182		iov->iov_base = (char *)iov->iov_base + cnt;
183		iov->iov_len -= cnt;
184		uio->uio_resid -= cnt;
185		uio->uio_offset += cnt;
186		cp = (char *)cp + cnt;
187		n -= cnt;
188	}
189out:
190	if (td != curthread) printf("uiomove: IT CHANGED!");
191	td = curthread;	/* Might things have changed in copyin/copyout? */
192	if (td) {
193		mtx_lock_spin(&sched_lock);
194		td->td_flags = (td->td_flags & ~TDF_DEADLKTREAT) | save;
195		mtx_unlock_spin(&sched_lock);
196	}
197	return (error);
198}
199
200#if defined(ENABLE_VFS_IOOPT) || defined(ZERO_COPY_SOCKETS)
201/*
202 * Experimental support for zero-copy I/O
203 */
204static int
205userspaceco(void *cp, u_int cnt, struct uio *uio, struct vm_object *obj,
206    int disposable)
207{
208	struct iovec *iov;
209	int error;
210
211	iov = uio->uio_iov;
212
213#ifdef ZERO_COPY_SOCKETS
214
215	if (uio->uio_rw == UIO_READ) {
216		if ((so_zero_copy_receive != 0)
217		 && (obj != NULL)
218		 && ((cnt & PAGE_MASK) == 0)
219		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
220		 && ((uio->uio_offset & PAGE_MASK) == 0)
221		 && ((((intptr_t) cp) & PAGE_MASK) == 0)
222		 && (obj->type == OBJT_DEFAULT)
223		 && (disposable != 0)) {
224			/* SOCKET: use page-trading */
225			/*
226			 * We only want to call vm_pgmoveco() on
227			 * disposeable pages, since it gives the
228			 * kernel page to the userland process.
229			 */
230			error =	vm_pgmoveco(&curproc->p_vmspace->vm_map,
231					    obj, (vm_offset_t)cp,
232					    (vm_offset_t)iov->iov_base);
233
234			/*
235			 * If we get an error back, attempt
236			 * to use copyout() instead.  The
237			 * disposable page should be freed
238			 * automatically if we weren't able to move
239			 * it into userland.
240			 */
241			if (error != 0)
242				error = copyout(cp, iov->iov_base, cnt);
243#ifdef ENABLE_VFS_IOOPT
244		} else if ((vfs_ioopt != 0)
245		 && ((cnt & PAGE_MASK) == 0)
246		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
247		 && ((uio->uio_offset & PAGE_MASK) == 0)
248		 && ((((intptr_t) cp) & PAGE_MASK) == 0)) {
249			error = vm_uiomove(&curproc->p_vmspace->vm_map, obj,
250					   uio->uio_offset, cnt,
251					   (vm_offset_t) iov->iov_base, NULL);
252#endif /* ENABLE_VFS_IOOPT */
253		} else {
254			error = copyout(cp, iov->iov_base, cnt);
255		}
256	} else {
257		error = copyin(iov->iov_base, cp, cnt);
258	}
259#else /* ZERO_COPY_SOCKETS */
260	if (uio->uio_rw == UIO_READ) {
261#ifdef ENABLE_VFS_IOOPT
262		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		} else
271#endif /* ENABLE_VFS_IOOPT */
272		{
273			error = copyout(cp, iov->iov_base, cnt);
274		}
275	} else {
276		error = copyin(iov->iov_base, cp, cnt);
277	}
278#endif /* ZERO_COPY_SOCKETS */
279
280	return (error);
281}
282
283int
284uiomoveco(void *cp, int n, struct uio *uio, struct vm_object *obj,
285    int disposable)
286{
287	struct iovec *iov;
288	u_int cnt;
289	int error;
290
291	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
292	    ("uiomoveco: mode"));
293	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
294	    ("uiomoveco proc"));
295
296	while (n > 0 && uio->uio_resid) {
297		iov = uio->uio_iov;
298		cnt = iov->iov_len;
299		if (cnt == 0) {
300			uio->uio_iov++;
301			uio->uio_iovcnt--;
302			continue;
303		}
304		if (cnt > n)
305			cnt = n;
306
307		switch (uio->uio_segflg) {
308
309		case UIO_USERSPACE:
310			if (ticks - PCPU_GET(switchticks) >= hogticks)
311				uio_yield();
312
313			error = userspaceco(cp, cnt, uio, obj, disposable);
314
315			if (error)
316				return (error);
317			break;
318
319		case UIO_SYSSPACE:
320			if (uio->uio_rw == UIO_READ)
321				bcopy(cp, iov->iov_base, cnt);
322			else
323				bcopy(iov->iov_base, cp, cnt);
324			break;
325		case UIO_NOCOPY:
326			break;
327		}
328		iov->iov_base = (char *)iov->iov_base + cnt;
329		iov->iov_len -= cnt;
330		uio->uio_resid -= cnt;
331		uio->uio_offset += cnt;
332		cp = (char *)cp + cnt;
333		n -= cnt;
334	}
335	return (0);
336}
337#endif /* ENABLE_VFS_IOOPT || ZERO_COPY_SOCKETS */
338
339#ifdef ENABLE_VFS_IOOPT
340
341/*
342 * Experimental support for zero-copy I/O
343 */
344int
345uioread(int n, struct uio *uio, struct vm_object *obj, int *nread)
346{
347	int npagesmoved;
348	struct iovec *iov;
349	u_int cnt, tcnt;
350	int error;
351
352	*nread = 0;
353	if (vfs_ioopt < 2)
354		return 0;
355
356	error = 0;
357
358	while (n > 0 && uio->uio_resid) {
359		iov = uio->uio_iov;
360		cnt = iov->iov_len;
361		if (cnt == 0) {
362			uio->uio_iov++;
363			uio->uio_iovcnt--;
364			continue;
365		}
366		if (cnt > n)
367			cnt = n;
368
369		if ((uio->uio_segflg == UIO_USERSPACE) &&
370			((((intptr_t) iov->iov_base) & PAGE_MASK) == 0) &&
371				 ((uio->uio_offset & PAGE_MASK) == 0) ) {
372
373			if (cnt < PAGE_SIZE)
374				break;
375
376			cnt &= ~PAGE_MASK;
377
378			if (ticks - PCPU_GET(switchticks) >= hogticks)
379				uio_yield();
380			error = vm_uiomove(&curproc->p_vmspace->vm_map, obj,
381						uio->uio_offset, cnt,
382						(vm_offset_t) iov->iov_base, &npagesmoved);
383
384			if (npagesmoved == 0)
385				break;
386
387			tcnt = npagesmoved * PAGE_SIZE;
388			cnt = tcnt;
389
390			if (error)
391				break;
392
393			iov->iov_base = (char *)iov->iov_base + cnt;
394			iov->iov_len -= cnt;
395			uio->uio_resid -= cnt;
396			uio->uio_offset += cnt;
397			*nread += cnt;
398			n -= cnt;
399		} else {
400			break;
401		}
402	}
403	return error;
404}
405#endif /* ENABLE_VFS_IOOPT */
406
407/*
408 * Give next character to user as result of read.
409 */
410int
411ureadc(int c, struct uio *uio)
412{
413	struct iovec *iov;
414	char *iov_base;
415
416again:
417	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
418		panic("ureadc");
419	iov = uio->uio_iov;
420	if (iov->iov_len == 0) {
421		uio->uio_iovcnt--;
422		uio->uio_iov++;
423		goto again;
424	}
425	switch (uio->uio_segflg) {
426
427	case UIO_USERSPACE:
428		if (subyte(iov->iov_base, c) < 0)
429			return (EFAULT);
430		break;
431
432	case UIO_SYSSPACE:
433		iov_base = iov->iov_base;
434		*iov_base = c;
435		iov->iov_base = iov_base;
436		break;
437
438	case UIO_NOCOPY:
439		break;
440	}
441	iov->iov_base = (char *)iov->iov_base + 1;
442	iov->iov_len--;
443	uio->uio_resid--;
444	uio->uio_offset++;
445	return (0);
446}
447
448/*
449 * General routine to allocate a hash table.
450 */
451void *
452hashinit(int elements, struct malloc_type *type, u_long *hashmask)
453{
454	long hashsize;
455	LIST_HEAD(generic, generic) *hashtbl;
456	int i;
457
458	if (elements <= 0)
459		panic("hashinit: bad elements");
460	for (hashsize = 1; hashsize <= elements; hashsize <<= 1)
461		continue;
462	hashsize >>= 1;
463	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
464	for (i = 0; i < hashsize; i++)
465		LIST_INIT(&hashtbl[i]);
466	*hashmask = hashsize - 1;
467	return (hashtbl);
468}
469
470void
471hashdestroy(void *vhashtbl, struct malloc_type *type, u_long hashmask)
472{
473	LIST_HEAD(generic, generic) *hashtbl, *hp;
474
475	hashtbl = vhashtbl;
476	for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++)
477		if (!LIST_EMPTY(hp))
478			panic("hashdestroy: hash not empty");
479	free(hashtbl, type);
480}
481
482static int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039,
483			2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653,
484			7159, 7673, 8191, 12281, 16381, 24571, 32749 };
485#define NPRIMES (sizeof(primes) / sizeof(primes[0]))
486
487/*
488 * General routine to allocate a prime number sized hash table.
489 */
490void *
491phashinit(int elements, struct malloc_type *type, u_long *nentries)
492{
493	long hashsize;
494	LIST_HEAD(generic, generic) *hashtbl;
495	int i;
496
497	if (elements <= 0)
498		panic("phashinit: bad elements");
499	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
500		i++;
501		if (i == NPRIMES)
502			break;
503		hashsize = primes[i];
504	}
505	hashsize = primes[i - 1];
506	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
507	for (i = 0; i < hashsize; i++)
508		LIST_INIT(&hashtbl[i]);
509	*nentries = hashsize;
510	return (hashtbl);
511}
512
513void
514uio_yield(void)
515{
516	struct thread *td;
517
518	td = curthread;
519	mtx_lock_spin(&sched_lock);
520	DROP_GIANT();
521	sched_prio(td, td->td_ksegrp->kg_user_pri); /* XXXKSE */
522	td->td_proc->p_stats->p_ru.ru_nivcsw++;
523	mi_switch();
524	mtx_unlock_spin(&sched_lock);
525	PICKUP_GIANT();
526}
527
528int
529copyinfrom(const void *src, void *dst, size_t len, int seg)
530{
531	int error = 0;
532
533	switch (seg) {
534	case UIO_USERSPACE:
535		error = copyin(src, dst, len);
536		break;
537	case UIO_SYSSPACE:
538		bcopy(src, dst, len);
539		break;
540	default:
541		panic("copyinfrom: bad seg %d\n", seg);
542	}
543	return (error);
544}
545
546int
547copyinstrfrom(const void *src, void *dst, size_t len, size_t *copied, int seg)
548{
549	int error = 0;
550
551	switch (seg) {
552	case UIO_USERSPACE:
553		error = copyinstr(src, dst, len, copied);
554		break;
555	case UIO_SYSSPACE:
556		error = copystr(src, dst, len, copied);
557		break;
558	default:
559		panic("copyinstrfrom: bad seg %d\n", seg);
560	}
561	return (error);
562}
563