uipc_shm.c revision 175164 
1/*-
2 * Copyright (c) 2006 Robert N. M. Watson
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27/*
28 * Support for shared swap-backed anonymous memory objects via
29 * shm_open(2) and shm_unlink(2).  While most of the implementation is
30 * here, vm_mmap.c contains mapping logic changes.
31 *
32 * TODO:
33 *
34 * (1) Convert test utilities into regression tests and import them into
35 *     src/tools/regression.
36 *
37 * (2) Need to export data to a userland tool via a sysctl.  Should ipcs(1)
38 *     and ipcrm(1) be expanded or should new tools to manage both POSIX
39 *     kernel semaphores and POSIX shared memory be written?
40 *
41 * (3) Add support for this file type to fstat(1).
42 *
43 * (4) Resource limits?  Does this need its own resource limits or are the
44 *     existing limits in mmap(2) sufficient?
45 *
46 * (5) Partial page truncation.  vnode_pager_setsize() will zero any parts
47 *     of a partially mapped page as a result of ftruncate(2)/truncate(2).
48 *     We can do the same (with the same pmap evil), but do we need to
49 *     worry about the bits on disk if the page is swapped out or will the
50 *     swapper zero the parts of a page that are invalid if the page is
51 *     swapped back in for us?
52 */
53
54#include <sys/cdefs.h>
55__FBSDID("$FreeBSD: head/sys/kern/uipc_shm.c 175164 2008-01-08 21:58:16Z jhb $");
56
57#include "opt_mac.h"
58
59#include <sys/param.h>
60#include <sys/fcntl.h>
61#include <sys/file.h>
62#include <sys/filedesc.h>
63#include <sys/fnv_hash.h>
64#include <sys/kernel.h>
65#include <sys/lock.h>
66#include <sys/malloc.h>
67#include <sys/mman.h>
68#include <sys/mutex.h>
69#include <sys/proc.h>
70#include <sys/refcount.h>
71#include <sys/resourcevar.h>
72#include <sys/stat.h>
73#include <sys/sysctl.h>
74#include <sys/sysproto.h>
75#include <sys/systm.h>
76#include <sys/sx.h>
77#include <sys/time.h>
78#include <sys/vnode.h>
79
80#include <security/mac/mac_framework.h>
81
82#include <vm/vm.h>
83#include <vm/vm_param.h>
84#include <vm/pmap.h>
85#include <vm/vm_map.h>
86#include <vm/vm_object.h>
87#include <vm/vm_page.h>
88#include <vm/vm_pager.h>
89#include <vm/swap_pager.h>
90
91struct shm_mapping {
92	char		*sm_path;
93	Fnv32_t		sm_fnv;
94	struct shmfd	*sm_shmfd;
95	LIST_ENTRY(shm_mapping) sm_link;
96};
97
98static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
99static LIST_HEAD(, shm_mapping) *shm_dictionary;
100static struct sx shm_dict_lock;
101static struct mtx shm_timestamp_lock;
102static u_long shm_hash;
103
104#define	SHM_HASH(fnv)	(&shm_dictionary[(fnv) & shm_hash])
105
106static int	shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
107static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
108static void	shm_dict_init(void *arg);
109static void	shm_drop(struct shmfd *shmfd);
110static struct shmfd *shm_hold(struct shmfd *shmfd);
111static void	shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
112static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
113static int	shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
114static void	shm_dotruncate(struct shmfd *shmfd, off_t length);
115
116static fo_rdwr_t	shm_read;
117static fo_rdwr_t	shm_write;
118static fo_truncate_t	shm_truncate;
119static fo_ioctl_t	shm_ioctl;
120static fo_poll_t	shm_poll;
121static fo_kqfilter_t	shm_kqfilter;
122static fo_stat_t	shm_stat;
123static fo_close_t	shm_close;
124
125/* File descriptor operations. */
126static struct fileops shm_ops = {
127	.fo_read = shm_read,
128	.fo_write = shm_write,
129	.fo_truncate = shm_truncate,
130	.fo_ioctl = shm_ioctl,
131	.fo_poll = shm_poll,
132	.fo_kqfilter = shm_kqfilter,
133	.fo_stat = shm_stat,
134	.fo_close = shm_close,
135	.fo_flags = DFLAG_PASSABLE
136};
137
138FEATURE(posix_shm, "POSIX shared memory");
139
140static int
141shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
142    int flags, struct thread *td)
143{
144
145	return (EOPNOTSUPP);
146}
147
148static int
149shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
150    int flags, struct thread *td)
151{
152
153	return (EOPNOTSUPP);
154}
155
156static int
157shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
158    struct thread *td)
159{
160	struct shmfd *shmfd;
161#ifdef MAC
162	int error;
163#endif
164
165	shmfd = fp->f_data;
166#ifdef MAC
167	error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
168	if (error)
169		return (error);
170#endif
171	shm_dotruncate(shmfd, length);
172	return (0);
173}
174
175static int
176shm_ioctl(struct file *fp, u_long com, void *data,
177    struct ucred *active_cred, struct thread *td)
178{
179
180	return (EOPNOTSUPP);
181}
182
183static int
184shm_poll(struct file *fp, int events, struct ucred *active_cred,
185    struct thread *td)
186{
187
188	return (EOPNOTSUPP);
189}
190
191static int
192shm_kqfilter(struct file *fp, struct knote *kn)
193{
194
195	return (EOPNOTSUPP);
196}
197
198static int
199shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
200    struct thread *td)
201{
202	struct shmfd *shmfd;
203#ifdef MAC
204	int error;
205#endif
206
207	shmfd = fp->f_data;
208
209#ifdef MAC
210	error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
211	if (error)
212		return (error);
213#endif
214
215	/*
216	 * Attempt to return sanish values for fstat() on a memory file
217	 * descriptor.
218	 */
219	bzero(sb, sizeof(*sb));
220	sb->st_mode = S_IFREG | shmfd->shm_mode;		/* XXX */
221	sb->st_blksize = PAGE_SIZE;
222	sb->st_size = shmfd->shm_size;
223	sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
224	sb->st_atimespec = shmfd->shm_atime;
225	sb->st_ctimespec = shmfd->shm_ctime;
226	sb->st_mtimespec = shmfd->shm_mtime;
227	sb->st_birthtimespec = shmfd->shm_birthtime;
228	sb->st_uid = shmfd->shm_uid;
229	sb->st_gid = shmfd->shm_gid;
230
231	return (0);
232}
233
234static int
235shm_close(struct file *fp, struct thread *td)
236{
237	struct shmfd *shmfd;
238
239	shmfd = fp->f_data;
240	fp->f_data = NULL;
241	shm_drop(shmfd);
242
243	return (0);
244}
245
246static void
247shm_dotruncate(struct shmfd *shmfd, off_t length)
248{
249	vm_object_t object;
250	vm_page_t m;
251	vm_pindex_t nobjsize;
252
253	object = shmfd->shm_object;
254	VM_OBJECT_LOCK(object);
255	if (length == shmfd->shm_size) {
256		VM_OBJECT_UNLOCK(object);
257		return;
258	}
259	nobjsize = OFF_TO_IDX(length + PAGE_MASK);
260
261	/* Are we shrinking?  If so, trim the end. */
262	if (length < shmfd->shm_size) {
263		/* Toss in memory pages. */
264		if (nobjsize < object->size)
265			vm_object_page_remove(object, nobjsize, object->size,
266			    FALSE);
267
268		/* Toss pages from swap. */
269		if (object->type == OBJT_SWAP)
270			swap_pager_freespace(object, nobjsize,
271			    object->size - nobjsize);
272
273		/*
274		 * If the last page is partially mapped, then zero out
275		 * the garbage at the end of the page.  See comments
276		 * in vnode_page_setsize() for more details.
277		 *
278		 * XXXJHB: This handles in memory pages, but what about
279		 * a page swapped out to disk?
280		 */
281		if ((length & PAGE_MASK) &&
282		    (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
283		    m->valid != 0) {
284			int base = (int)length & PAGE_MASK;
285			int size = PAGE_SIZE - base;
286
287			pmap_zero_page_area(m, base, size);
288			vm_page_lock_queues();
289			vm_page_set_validclean(m, base, size);
290			if (m->dirty != 0)
291				m->dirty = VM_PAGE_BITS_ALL;
292			vm_page_unlock_queues();
293		}
294	}
295	shmfd->shm_size = length;
296	mtx_lock(&shm_timestamp_lock);
297	vfs_timestamp(&shmfd->shm_ctime);
298	shmfd->shm_mtime = shmfd->shm_ctime;
299	mtx_unlock(&shm_timestamp_lock);
300	object->size = nobjsize;
301	VM_OBJECT_UNLOCK(object);
302}
303
304/*
305 * shmfd object management including creation and reference counting
306 * routines.
307 */
308static struct shmfd *
309shm_alloc(struct ucred *ucred, mode_t mode)
310{
311	struct shmfd *shmfd;
312
313	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
314	shmfd->shm_size = 0;
315	shmfd->shm_uid = ucred->cr_uid;
316	shmfd->shm_gid = ucred->cr_gid;
317	shmfd->shm_mode = mode;
318	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
319	    shmfd->shm_size, VM_PROT_DEFAULT, 0);
320	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
321	vfs_timestamp(&shmfd->shm_birthtime);
322	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
323	    shmfd->shm_birthtime;
324	refcount_init(&shmfd->shm_refs, 1);
325#ifdef MAC
326	mac_posixshm_init(shmfd);
327	mac_posixshm_create(ucred, shmfd);
328#endif
329
330	return (shmfd);
331}
332
333static struct shmfd *
334shm_hold(struct shmfd *shmfd)
335{
336
337	refcount_acquire(&shmfd->shm_refs);
338	return (shmfd);
339}
340
341static void
342shm_drop(struct shmfd *shmfd)
343{
344
345	if (refcount_release(&shmfd->shm_refs)) {
346#ifdef MAC
347		mac_posixshm_destroy(shmfd);
348#endif
349		vm_object_deallocate(shmfd->shm_object);
350		free(shmfd, M_SHMFD);
351	}
352}
353
354/*
355 * Determine if the credentials have sufficient permissions for a
356 * specified combination of FREAD and FWRITE.
357 */
358static int
359shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
360{
361	int acc_mode;
362
363	acc_mode = 0;
364	if (flags & FREAD)
365		acc_mode |= VREAD;
366	if (flags & FWRITE)
367		acc_mode |= VWRITE;
368	return (vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
369	    acc_mode, ucred, NULL));
370}
371
372/*
373 * Dictionary management.  We maintain an in-kernel dictionary to map
374 * paths to shmfd objects.  We use the FNV hash on the path to store
375 * the mappings in a hash table.
376 */
377static void
378shm_dict_init(void *arg)
379{
380
381	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
382	sx_init(&shm_dict_lock, "shm dictionary");
383	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
384}
385SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
386
387static struct shmfd *
388shm_lookup(char *path, Fnv32_t fnv)
389{
390	struct shm_mapping *map;
391
392	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
393		if (map->sm_fnv != fnv)
394			continue;
395		if (strcmp(map->sm_path, path) == 0)
396			return (map->sm_shmfd);
397	}
398
399	return (NULL);
400}
401
402static void
403shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
404{
405	struct shm_mapping *map;
406
407	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
408	map->sm_path = path;
409	map->sm_fnv = fnv;
410	map->sm_shmfd = shm_hold(shmfd);
411	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
412}
413
414static int
415shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
416{
417	struct shm_mapping *map;
418	int error;
419
420	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
421		if (map->sm_fnv != fnv)
422			continue;
423		if (strcmp(map->sm_path, path) == 0) {
424#ifdef MAC
425			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
426			if (error)
427				return (error);
428#endif
429			error = shm_access(map->sm_shmfd, ucred,
430			    FREAD | FWRITE);
431			if (error)
432				return (error);
433			LIST_REMOVE(map, sm_link);
434			shm_drop(map->sm_shmfd);
435			free(map->sm_path, M_SHMFD);
436			free(map, M_SHMFD);
437			return (0);
438		}
439	}
440
441	return (ENOENT);
442}
443
444/* System calls. */
445int
446shm_open(struct thread *td, struct shm_open_args *uap)
447{
448	struct filedesc *fdp;
449	struct shmfd *shmfd;
450	struct file *fp;
451	char *path;
452	Fnv32_t fnv;
453	mode_t cmode;
454	int fd, error;
455
456	if ((uap->flags & O_ACCMODE) != O_RDONLY &&
457	    (uap->flags & O_ACCMODE) != O_RDWR)
458		return (EINVAL);
459
460	if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
461		return (EINVAL);
462
463	fdp = td->td_proc->p_fd;
464	cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
465
466	error = falloc(td, &fp, &fd);
467	if (error)
468		return (error);
469
470	/* A SHM_ANON path pointer creates an anonymous object. */
471	if (uap->path == SHM_ANON) {
472		/* A read-only anonymous object is pointless. */
473		if ((uap->flags & O_ACCMODE) == O_RDONLY) {
474			fdclose(fdp, fp, fd, td);
475			fdrop(fp, td);
476			return (EINVAL);
477		}
478		shmfd = shm_alloc(td->td_ucred, cmode);
479	} else {
480		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
481		error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
482
483		/* Require paths to start with a '/' character. */
484		if (error == 0 && path[0] != '/')
485			error = EINVAL;
486		if (error) {
487			fdclose(fdp, fp, fd, td);
488			fdrop(fp, td);
489			free(path, M_SHMFD);
490			return (error);
491		}
492
493		fnv = fnv_32_str(path, FNV1_32_INIT);
494		sx_xlock(&shm_dict_lock);
495		shmfd = shm_lookup(path, fnv);
496		if (shmfd == NULL) {
497			/* Object does not yet exist, create it if requested. */
498			if (uap->flags & O_CREAT) {
499				shmfd = shm_alloc(td->td_ucred, cmode);
500				shm_insert(path, fnv, shmfd);
501			} else {
502				free(path, M_SHMFD);
503				error = ENOENT;
504			}
505		} else {
506			/*
507			 * Object already exists, obtain a new
508			 * reference if requested and permitted.
509			 */
510			free(path, M_SHMFD);
511			if ((uap->flags & (O_CREAT | O_EXCL)) ==
512			    (O_CREAT | O_EXCL))
513				error = EEXIST;
514			else {
515#ifdef MAC
516				error = mac_posixshm_check_open(td->td_ucred,
517				    shmfd);
518				if (error == 0)
519#endif
520				error = shm_access(shmfd, td->td_ucred,
521				    FFLAGS(uap->flags & O_ACCMODE));
522			}
523
524			/*
525			 * Truncate the file back to zero length if
526			 * O_TRUNC was specified and the object was
527			 * opened with read/write.
528			 */
529			if (error == 0 &&
530			    (uap->flags & (O_ACCMODE | O_TRUNC)) ==
531			    (O_RDWR | O_TRUNC)) {
532#ifdef MAC
533				error = mac_posixshm_check_truncate(
534					td->td_ucred, fp->f_cred, shmfd);
535				if (error == 0)
536#endif
537					shm_dotruncate(shmfd, 0);
538			}
539			if (error == 0)
540				shm_hold(shmfd);
541		}
542		sx_xunlock(&shm_dict_lock);
543
544		if (error) {
545			fdclose(fdp, fp, fd, td);
546			fdrop(fp, td);
547			return (error);
548		}
549	}
550
551	finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
552
553	FILEDESC_XLOCK(fdp);
554	if (fdp->fd_ofiles[fd] == fp)
555		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
556	FILEDESC_XUNLOCK(fdp);
557	td->td_retval[0] = fd;
558	fdrop(fp, td);
559
560	return (0);
561}
562
563int
564shm_unlink(struct thread *td, struct shm_unlink_args *uap)
565{
566	char *path;
567	Fnv32_t fnv;
568	int error;
569
570	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
571	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
572	if (error) {
573		free(path, M_TEMP);
574		return (error);
575	}
576
577	fnv = fnv_32_str(path, FNV1_32_INIT);
578	sx_xlock(&shm_dict_lock);
579	error = shm_remove(path, fnv, td->td_ucred);
580	sx_xunlock(&shm_dict_lock);
581	free(path, M_TEMP);
582
583	return (error);
584}
585
586/*
587 * mmap() helper to validate mmap() requests against shm object state
588 * and give mmap() the vm_object to use for the mapping.
589 */
590int
591shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
592    vm_object_t *obj)
593{
594
595	/*
596	 * XXXRW: This validation is probably insufficient, and subject to
597	 * sign errors.  It should be fixed.
598	 */
599	if (foff >= shmfd->shm_size || foff + objsize > shmfd->shm_size)
600		return (EINVAL);
601
602	mtx_lock(&shm_timestamp_lock);
603	vfs_timestamp(&shmfd->shm_atime);
604	mtx_unlock(&shm_timestamp_lock);
605	vm_object_reference(shmfd->shm_object);
606	*obj = shmfd->shm_object;
607	return (0);
608}
609