subr_hash.c revision 130023 
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 130023 2004-06-03 01:47:37Z tjr $");
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_object_t srcobj, vm_offset_t kaddr,
73    vm_offset_t uaddr)
74{
75	vm_map_t map = mapa;
76	vm_page_t kern_pg, user_pg;
77	vm_object_t uobject;
78	vm_map_entry_t entry;
79	vm_pindex_t upindex, kpindex;
80	vm_prot_t prot;
81	boolean_t wired;
82
83	/*
84	 * First lookup the kernel page.
85	 */
86	kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
87	/*
88	 * XXX The vm object containing kern_pg needs locking.
89	 */
90	if ((vm_map_lookup(&map, uaddr,
91			   VM_PROT_WRITE, &entry, &uobject,
92			   &upindex, &prot, &wired)) != KERN_SUCCESS) {
93		return(EFAULT);
94	}
95	VM_OBJECT_LOCK(uobject);
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	VM_OBJECT_UNLOCK(uobject);
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		save = td->td_pflags & TDP_DEADLKTREAT;
144		td->td_pflags |= TDP_DEADLKTREAT;
145	}
146
147	while (n > 0 && uio->uio_resid) {
148		iov = uio->uio_iov;
149		cnt = iov->iov_len;
150		if (cnt == 0) {
151			uio->uio_iov++;
152			uio->uio_iovcnt--;
153			continue;
154		}
155		if (cnt > n)
156			cnt = n;
157
158		switch (uio->uio_segflg) {
159
160		case UIO_USERSPACE:
161			if (ticks - PCPU_GET(switchticks) >= hogticks)
162				uio_yield();
163			if (uio->uio_rw == UIO_READ)
164				error = copyout(cp, iov->iov_base, cnt);
165			else
166				error = copyin(iov->iov_base, cp, cnt);
167			if (error)
168				goto out;
169			break;
170
171		case UIO_SYSSPACE:
172			if (uio->uio_rw == UIO_READ)
173				bcopy(cp, iov->iov_base, cnt);
174			else
175				bcopy(iov->iov_base, cp, cnt);
176			break;
177		case UIO_NOCOPY:
178			break;
179		}
180		iov->iov_base = (char *)iov->iov_base + cnt;
181		iov->iov_len -= cnt;
182		uio->uio_resid -= cnt;
183		uio->uio_offset += cnt;
184		cp = (char *)cp + cnt;
185		n -= cnt;
186	}
187out:
188	if (td && save == 0)
189		td->td_pflags &= ~TDP_DEADLKTREAT;
190	return (error);
191}
192
193/*
194 * Wrapper for uiomove() that validates the arguments against a known-good
195 * kernel buffer.  Currently, uiomove accepts a signed (n) argument, which
196 * is almost definitely a bad thing, so we catch that here as well.  We
197 * return a runtime failure, but it might be desirable to generate a runtime
198 * assertion failure instead.
199 */
200int
201uiomove_frombuf(void *buf, int buflen, struct uio *uio)
202{
203	unsigned int offset, n;
204
205	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
206	    (offset = uio->uio_offset) != uio->uio_offset)
207		return (EINVAL);
208	if (buflen <= 0 || offset >= buflen)
209		return (0);
210	if ((n = buflen - offset) > INT_MAX)
211		return (EINVAL);
212	return (uiomove((char *)buf + offset, n, uio));
213}
214
215#ifdef ZERO_COPY_SOCKETS
216/*
217 * Experimental support for zero-copy I/O
218 */
219static int
220userspaceco(void *cp, u_int cnt, struct uio *uio, struct vm_object *obj,
221    int disposable)
222{
223	struct iovec *iov;
224	int error;
225
226	iov = uio->uio_iov;
227	if (uio->uio_rw == UIO_READ) {
228		if ((so_zero_copy_receive != 0)
229		 && (obj != NULL)
230		 && ((cnt & PAGE_MASK) == 0)
231		 && ((((intptr_t) iov->iov_base) & PAGE_MASK) == 0)
232		 && ((uio->uio_offset & PAGE_MASK) == 0)
233		 && ((((intptr_t) cp) & PAGE_MASK) == 0)
234		 && (obj->type == OBJT_DEFAULT)
235		 && (disposable != 0)) {
236			/* SOCKET: use page-trading */
237			/*
238			 * We only want to call vm_pgmoveco() on
239			 * disposeable pages, since it gives the
240			 * kernel page to the userland process.
241			 */
242			error =	vm_pgmoveco(&curproc->p_vmspace->vm_map,
243					    obj, (vm_offset_t)cp,
244					    (vm_offset_t)iov->iov_base);
245
246			/*
247			 * If we get an error back, attempt
248			 * to use copyout() instead.  The
249			 * disposable page should be freed
250			 * automatically if we weren't able to move
251			 * it into userland.
252			 */
253			if (error != 0)
254				error = copyout(cp, iov->iov_base, cnt);
255		} else {
256			error = copyout(cp, iov->iov_base, cnt);
257		}
258	} else {
259		error = copyin(iov->iov_base, cp, cnt);
260	}
261	return (error);
262}
263
264int
265uiomoveco(void *cp, int n, struct uio *uio, struct vm_object *obj,
266    int disposable)
267{
268	struct iovec *iov;
269	u_int cnt;
270	int error;
271
272	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
273	    ("uiomoveco: mode"));
274	KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
275	    ("uiomoveco proc"));
276
277	while (n > 0 && uio->uio_resid) {
278		iov = uio->uio_iov;
279		cnt = iov->iov_len;
280		if (cnt == 0) {
281			uio->uio_iov++;
282			uio->uio_iovcnt--;
283			continue;
284		}
285		if (cnt > n)
286			cnt = n;
287
288		switch (uio->uio_segflg) {
289
290		case UIO_USERSPACE:
291			if (ticks - PCPU_GET(switchticks) >= hogticks)
292				uio_yield();
293
294			error = userspaceco(cp, cnt, uio, obj, disposable);
295
296			if (error)
297				return (error);
298			break;
299
300		case UIO_SYSSPACE:
301			if (uio->uio_rw == UIO_READ)
302				bcopy(cp, iov->iov_base, cnt);
303			else
304				bcopy(iov->iov_base, cp, cnt);
305			break;
306		case UIO_NOCOPY:
307			break;
308		}
309		iov->iov_base = (char *)iov->iov_base + cnt;
310		iov->iov_len -= cnt;
311		uio->uio_resid -= cnt;
312		uio->uio_offset += cnt;
313		cp = (char *)cp + cnt;
314		n -= cnt;
315	}
316	return (0);
317}
318#endif /* ZERO_COPY_SOCKETS */
319
320/*
321 * Give next character to user as result of read.
322 */
323int
324ureadc(int c, struct uio *uio)
325{
326	struct iovec *iov;
327	char *iov_base;
328
329again:
330	if (uio->uio_iovcnt == 0 || uio->uio_resid == 0)
331		panic("ureadc");
332	iov = uio->uio_iov;
333	if (iov->iov_len == 0) {
334		uio->uio_iovcnt--;
335		uio->uio_iov++;
336		goto again;
337	}
338	switch (uio->uio_segflg) {
339
340	case UIO_USERSPACE:
341		if (subyte(iov->iov_base, c) < 0)
342			return (EFAULT);
343		break;
344
345	case UIO_SYSSPACE:
346		iov_base = iov->iov_base;
347		*iov_base = c;
348		iov->iov_base = iov_base;
349		break;
350
351	case UIO_NOCOPY:
352		break;
353	}
354	iov->iov_base = (char *)iov->iov_base + 1;
355	iov->iov_len--;
356	uio->uio_resid--;
357	uio->uio_offset++;
358	return (0);
359}
360
361/*
362 * General routine to allocate a hash table.
363 */
364void *
365hashinit(int elements, struct malloc_type *type, u_long *hashmask)
366{
367	long hashsize;
368	LIST_HEAD(generic, generic) *hashtbl;
369	int i;
370
371	if (elements <= 0)
372		panic("hashinit: bad elements");
373	for (hashsize = 1; hashsize <= elements; hashsize <<= 1)
374		continue;
375	hashsize >>= 1;
376	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
377	for (i = 0; i < hashsize; i++)
378		LIST_INIT(&hashtbl[i]);
379	*hashmask = hashsize - 1;
380	return (hashtbl);
381}
382
383void
384hashdestroy(void *vhashtbl, struct malloc_type *type, u_long hashmask)
385{
386	LIST_HEAD(generic, generic) *hashtbl, *hp;
387
388	hashtbl = vhashtbl;
389	for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++)
390		if (!LIST_EMPTY(hp))
391			panic("hashdestroy: hash not empty");
392	free(hashtbl, type);
393}
394
395static int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039,
396			2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653,
397			7159, 7673, 8191, 12281, 16381, 24571, 32749 };
398#define NPRIMES (sizeof(primes) / sizeof(primes[0]))
399
400/*
401 * General routine to allocate a prime number sized hash table.
402 */
403void *
404phashinit(int elements, struct malloc_type *type, u_long *nentries)
405{
406	long hashsize;
407	LIST_HEAD(generic, generic) *hashtbl;
408	int i;
409
410	if (elements <= 0)
411		panic("phashinit: bad elements");
412	for (i = 1, hashsize = primes[1]; hashsize <= elements;) {
413		i++;
414		if (i == NPRIMES)
415			break;
416		hashsize = primes[i];
417	}
418	hashsize = primes[i - 1];
419	hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, M_WAITOK);
420	for (i = 0; i < hashsize; i++)
421		LIST_INIT(&hashtbl[i]);
422	*nentries = hashsize;
423	return (hashtbl);
424}
425
426void
427uio_yield(void)
428{
429	struct thread *td;
430
431	td = curthread;
432	mtx_lock_spin(&sched_lock);
433	DROP_GIANT();
434	sched_prio(td, td->td_ksegrp->kg_user_pri); /* XXXKSE */
435	mi_switch(SW_INVOL);
436	mtx_unlock_spin(&sched_lock);
437	PICKUP_GIANT();
438}
439
440int
441copyinfrom(const void * __restrict src, void * __restrict dst, size_t len,
442    int seg)
443{
444	int error = 0;
445
446	switch (seg) {
447	case UIO_USERSPACE:
448		error = copyin(src, dst, len);
449		break;
450	case UIO_SYSSPACE:
451		bcopy(src, dst, len);
452		break;
453	default:
454		panic("copyinfrom: bad seg %d\n", seg);
455	}
456	return (error);
457}
458
459int
460copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len,
461    size_t * __restrict copied, int seg)
462{
463	int error = 0;
464
465	switch (seg) {
466	case UIO_USERSPACE:
467		error = copyinstr(src, dst, len, copied);
468		break;
469	case UIO_SYSSPACE:
470		error = copystr(src, dst, len, copied);
471		break;
472	default:
473		panic("copyinstrfrom: bad seg %d\n", seg);
474	}
475	return (error);
476}
477
478int
479uiofromiov(struct iovec *iovp, u_int iovcnt, struct uio *uio)
480{
481	struct iovec *iov;
482	u_int iovlen;
483	int error, i;
484
485	/* note: can't use iovlen until iovcnt is validated */
486	iovlen = iovcnt * sizeof (struct iovec);
487	if (iovcnt > UIO_MAXIOV) {
488		error = EINVAL;
489		goto done;
490	}
491	MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
492	uio->uio_iov = iov;
493	uio->uio_iovcnt = iovcnt;
494	uio->uio_segflg = UIO_USERSPACE;
495	uio->uio_offset = -1;
496	if ((error = copyin(iovp, iov, iovlen)))
497		goto done;
498	uio->uio_resid = 0;
499	for (i = 0; i < iovcnt; i++) {
500		if (iov->iov_len > INT_MAX - uio->uio_resid) {
501			error = EINVAL;
502			goto done;
503		}
504		uio->uio_resid += iov->iov_len;
505		iov++;
506	}
507
508done:
509	if (error && uio->uio_iov) {
510		FREE(uio->uio_iov, M_IOV);
511		uio->uio_iov = NULL;
512	}
513	return (error);
514
515}
516