uipc_shm.c revision 185533 
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 * (2) Need to export data to a userland tool via a sysctl.  Should ipcs(1)
35 *     and ipcrm(1) be expanded or should new tools to manage both POSIX
36 *     kernel semaphores and POSIX shared memory be written?
37 *
38 * (3) Add support for this file type to fstat(1).
39 *
40 * (4) Resource limits?  Does this need its own resource limits or are the
41 *     existing limits in mmap(2) sufficient?
42 *
43 * (5) Partial page truncation.  vnode_pager_setsize() will zero any parts
44 *     of a partially mapped page as a result of ftruncate(2)/truncate(2).
45 *     We can do the same (with the same pmap evil), but do we need to
46 *     worry about the bits on disk if the page is swapped out or will the
47 *     swapper zero the parts of a page that are invalid if the page is
48 *     swapped back in for us?
49 *
50 * (6) Add MAC support in mac_biba(4) and mac_mls(4).
51 *
52 * (7) Add a MAC check_create() hook for creating new named objects.
53 */
54
55#include <sys/cdefs.h>
56__FBSDID("$FreeBSD: head/sys/kern/uipc_shm.c 185533 2008-12-01 22:33:50Z kan $");
57
58#include "opt_mac.h"
59
60#include <sys/param.h>
61#include <sys/fcntl.h>
62#include <sys/file.h>
63#include <sys/filedesc.h>
64#include <sys/fnv_hash.h>
65#include <sys/kernel.h>
66#include <sys/lock.h>
67#include <sys/malloc.h>
68#include <sys/mman.h>
69#include <sys/mutex.h>
70#include <sys/proc.h>
71#include <sys/refcount.h>
72#include <sys/resourcevar.h>
73#include <sys/stat.h>
74#include <sys/sysctl.h>
75#include <sys/sysproto.h>
76#include <sys/systm.h>
77#include <sys/sx.h>
78#include <sys/time.h>
79#include <sys/vnode.h>
80
81#include <security/mac/mac_framework.h>
82
83#include <vm/vm.h>
84#include <vm/vm_param.h>
85#include <vm/pmap.h>
86#include <vm/vm_map.h>
87#include <vm/vm_object.h>
88#include <vm/vm_page.h>
89#include <vm/vm_pager.h>
90#include <vm/swap_pager.h>
91
92struct shm_mapping {
93	char		*sm_path;
94	Fnv32_t		sm_fnv;
95	struct shmfd	*sm_shmfd;
96	LIST_ENTRY(shm_mapping) sm_link;
97};
98
99static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
100static LIST_HEAD(, shm_mapping) *shm_dictionary;
101static struct sx shm_dict_lock;
102static struct mtx shm_timestamp_lock;
103static u_long shm_hash;
104
105#define	SHM_HASH(fnv)	(&shm_dictionary[(fnv) & shm_hash])
106
107static int	shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
108static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
109static void	shm_dict_init(void *arg);
110static void	shm_drop(struct shmfd *shmfd);
111static struct shmfd *shm_hold(struct shmfd *shmfd);
112static void	shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
113static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
114static int	shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
115static void	shm_dotruncate(struct shmfd *shmfd, off_t length);
116
117static fo_rdwr_t	shm_read;
118static fo_rdwr_t	shm_write;
119static fo_truncate_t	shm_truncate;
120static fo_ioctl_t	shm_ioctl;
121static fo_poll_t	shm_poll;
122static fo_kqfilter_t	shm_kqfilter;
123static fo_stat_t	shm_stat;
124static fo_close_t	shm_close;
125
126/* File descriptor operations. */
127static struct fileops shm_ops = {
128	.fo_read = shm_read,
129	.fo_write = shm_write,
130	.fo_truncate = shm_truncate,
131	.fo_ioctl = shm_ioctl,
132	.fo_poll = shm_poll,
133	.fo_kqfilter = shm_kqfilter,
134	.fo_stat = shm_stat,
135	.fo_close = shm_close,
136	.fo_flags = DFLAG_PASSABLE
137};
138
139FEATURE(posix_shm, "POSIX shared memory");
140
141static int
142shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
143    int flags, struct thread *td)
144{
145
146	return (EOPNOTSUPP);
147}
148
149static int
150shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
151    int flags, struct thread *td)
152{
153
154	return (EOPNOTSUPP);
155}
156
157static int
158shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
159    struct thread *td)
160{
161	struct shmfd *shmfd;
162#ifdef MAC
163	int error;
164#endif
165
166	shmfd = fp->f_data;
167#ifdef MAC
168	error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
169	if (error)
170		return (error);
171#endif
172	shm_dotruncate(shmfd, length);
173	return (0);
174}
175
176static int
177shm_ioctl(struct file *fp, u_long com, void *data,
178    struct ucred *active_cred, struct thread *td)
179{
180
181	return (EOPNOTSUPP);
182}
183
184static int
185shm_poll(struct file *fp, int events, struct ucred *active_cred,
186    struct thread *td)
187{
188
189	return (EOPNOTSUPP);
190}
191
192static int
193shm_kqfilter(struct file *fp, struct knote *kn)
194{
195
196	return (EOPNOTSUPP);
197}
198
199static int
200shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
201    struct thread *td)
202{
203	struct shmfd *shmfd;
204#ifdef MAC
205	int error;
206#endif
207
208	shmfd = fp->f_data;
209
210#ifdef MAC
211	error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
212	if (error)
213		return (error);
214#endif
215
216	/*
217	 * Attempt to return sanish values for fstat() on a memory file
218	 * descriptor.
219	 */
220	bzero(sb, sizeof(*sb));
221	sb->st_mode = S_IFREG | shmfd->shm_mode;		/* XXX */
222	sb->st_blksize = PAGE_SIZE;
223	sb->st_size = shmfd->shm_size;
224	sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
225	sb->st_atimespec = shmfd->shm_atime;
226	sb->st_ctimespec = shmfd->shm_ctime;
227	sb->st_mtimespec = shmfd->shm_mtime;
228	sb->st_birthtimespec = shmfd->shm_birthtime;
229	sb->st_uid = shmfd->shm_uid;
230	sb->st_gid = shmfd->shm_gid;
231
232	return (0);
233}
234
235static int
236shm_close(struct file *fp, struct thread *td)
237{
238	struct shmfd *shmfd;
239
240	shmfd = fp->f_data;
241	fp->f_data = NULL;
242	shm_drop(shmfd);
243
244	return (0);
245}
246
247static void
248shm_dotruncate(struct shmfd *shmfd, off_t length)
249{
250	vm_object_t object;
251	vm_page_t m;
252	vm_pindex_t nobjsize;
253
254	object = shmfd->shm_object;
255	VM_OBJECT_LOCK(object);
256	if (length == shmfd->shm_size) {
257		VM_OBJECT_UNLOCK(object);
258		return;
259	}
260	nobjsize = OFF_TO_IDX(length + PAGE_MASK);
261
262	/* Are we shrinking?  If so, trim the end. */
263	if (length < shmfd->shm_size) {
264		/* Toss in memory pages. */
265		if (nobjsize < object->size)
266			vm_object_page_remove(object, nobjsize, object->size,
267			    FALSE);
268
269		/* Toss pages from swap. */
270		if (object->type == OBJT_SWAP)
271			swap_pager_freespace(object, nobjsize,
272			    object->size - nobjsize);
273
274		/*
275		 * If the last page is partially mapped, then zero out
276		 * the garbage at the end of the page.  See comments
277		 * in vnode_page_setsize() for more details.
278		 *
279		 * XXXJHB: This handles in memory pages, but what about
280		 * a page swapped out to disk?
281		 */
282		if ((length & PAGE_MASK) &&
283		    (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
284		    m->valid != 0) {
285			int base = (int)length & PAGE_MASK;
286			int size = PAGE_SIZE - base;
287
288			pmap_zero_page_area(m, base, size);
289			vm_page_lock_queues();
290			vm_page_set_validclean(m, base, size);
291			if (m->dirty != 0)
292				m->dirty = VM_PAGE_BITS_ALL;
293			vm_page_unlock_queues();
294		} else if ((length & PAGE_MASK) &&
295		    __predict_false(object->cache != NULL)) {
296			vm_page_cache_free(object, OFF_TO_IDX(length),
297			    nobjsize);
298		}
299	}
300	shmfd->shm_size = length;
301	mtx_lock(&shm_timestamp_lock);
302	vfs_timestamp(&shmfd->shm_ctime);
303	shmfd->shm_mtime = shmfd->shm_ctime;
304	mtx_unlock(&shm_timestamp_lock);
305	object->size = nobjsize;
306	VM_OBJECT_UNLOCK(object);
307}
308
309/*
310 * shmfd object management including creation and reference counting
311 * routines.
312 */
313static struct shmfd *
314shm_alloc(struct ucred *ucred, mode_t mode)
315{
316	struct shmfd *shmfd;
317
318	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
319	shmfd->shm_size = 0;
320	shmfd->shm_uid = ucred->cr_uid;
321	shmfd->shm_gid = ucred->cr_gid;
322	shmfd->shm_mode = mode;
323	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
324	    shmfd->shm_size, VM_PROT_DEFAULT, 0);
325	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
326	VM_OBJECT_LOCK(shmfd->shm_object);
327	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
328	vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
329	VM_OBJECT_UNLOCK(shmfd->shm_object);
330	vfs_timestamp(&shmfd->shm_birthtime);
331	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
332	    shmfd->shm_birthtime;
333	refcount_init(&shmfd->shm_refs, 1);
334#ifdef MAC
335	mac_posixshm_init(shmfd);
336	mac_posixshm_create(ucred, shmfd);
337#endif
338
339	return (shmfd);
340}
341
342static struct shmfd *
343shm_hold(struct shmfd *shmfd)
344{
345
346	refcount_acquire(&shmfd->shm_refs);
347	return (shmfd);
348}
349
350static void
351shm_drop(struct shmfd *shmfd)
352{
353
354	if (refcount_release(&shmfd->shm_refs)) {
355#ifdef MAC
356		mac_posixshm_destroy(shmfd);
357#endif
358		vm_object_deallocate(shmfd->shm_object);
359		free(shmfd, M_SHMFD);
360	}
361}
362
363/*
364 * Determine if the credentials have sufficient permissions for a
365 * specified combination of FREAD and FWRITE.
366 */
367static int
368shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
369{
370	accmode_t accmode;
371
372	accmode = 0;
373	if (flags & FREAD)
374		accmode |= VREAD;
375	if (flags & FWRITE)
376		accmode |= VWRITE;
377	return (vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
378	    accmode, ucred, NULL));
379}
380
381/*
382 * Dictionary management.  We maintain an in-kernel dictionary to map
383 * paths to shmfd objects.  We use the FNV hash on the path to store
384 * the mappings in a hash table.
385 */
386static void
387shm_dict_init(void *arg)
388{
389
390	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
391	sx_init(&shm_dict_lock, "shm dictionary");
392	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
393}
394SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
395
396static struct shmfd *
397shm_lookup(char *path, Fnv32_t fnv)
398{
399	struct shm_mapping *map;
400
401	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
402		if (map->sm_fnv != fnv)
403			continue;
404		if (strcmp(map->sm_path, path) == 0)
405			return (map->sm_shmfd);
406	}
407
408	return (NULL);
409}
410
411static void
412shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
413{
414	struct shm_mapping *map;
415
416	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
417	map->sm_path = path;
418	map->sm_fnv = fnv;
419	map->sm_shmfd = shm_hold(shmfd);
420	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
421}
422
423static int
424shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
425{
426	struct shm_mapping *map;
427	int error;
428
429	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
430		if (map->sm_fnv != fnv)
431			continue;
432		if (strcmp(map->sm_path, path) == 0) {
433#ifdef MAC
434			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
435			if (error)
436				return (error);
437#endif
438			error = shm_access(map->sm_shmfd, ucred,
439			    FREAD | FWRITE);
440			if (error)
441				return (error);
442			LIST_REMOVE(map, sm_link);
443			shm_drop(map->sm_shmfd);
444			free(map->sm_path, M_SHMFD);
445			free(map, M_SHMFD);
446			return (0);
447		}
448	}
449
450	return (ENOENT);
451}
452
453/* System calls. */
454int
455shm_open(struct thread *td, struct shm_open_args *uap)
456{
457	struct filedesc *fdp;
458	struct shmfd *shmfd;
459	struct file *fp;
460	char *path;
461	Fnv32_t fnv;
462	mode_t cmode;
463	int fd, error;
464
465	if ((uap->flags & O_ACCMODE) != O_RDONLY &&
466	    (uap->flags & O_ACCMODE) != O_RDWR)
467		return (EINVAL);
468
469	if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
470		return (EINVAL);
471
472	fdp = td->td_proc->p_fd;
473	cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
474
475	error = falloc(td, &fp, &fd);
476	if (error)
477		return (error);
478
479	/* A SHM_ANON path pointer creates an anonymous object. */
480	if (uap->path == SHM_ANON) {
481		/* A read-only anonymous object is pointless. */
482		if ((uap->flags & O_ACCMODE) == O_RDONLY) {
483			fdclose(fdp, fp, fd, td);
484			fdrop(fp, td);
485			return (EINVAL);
486		}
487		shmfd = shm_alloc(td->td_ucred, cmode);
488	} else {
489		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
490		error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
491
492		/* Require paths to start with a '/' character. */
493		if (error == 0 && path[0] != '/')
494			error = EINVAL;
495		if (error) {
496			fdclose(fdp, fp, fd, td);
497			fdrop(fp, td);
498			free(path, M_SHMFD);
499			return (error);
500		}
501
502		fnv = fnv_32_str(path, FNV1_32_INIT);
503		sx_xlock(&shm_dict_lock);
504		shmfd = shm_lookup(path, fnv);
505		if (shmfd == NULL) {
506			/* Object does not yet exist, create it if requested. */
507			if (uap->flags & O_CREAT) {
508				shmfd = shm_alloc(td->td_ucred, cmode);
509				shm_insert(path, fnv, shmfd);
510			} else {
511				free(path, M_SHMFD);
512				error = ENOENT;
513			}
514		} else {
515			/*
516			 * Object already exists, obtain a new
517			 * reference if requested and permitted.
518			 */
519			free(path, M_SHMFD);
520			if ((uap->flags & (O_CREAT | O_EXCL)) ==
521			    (O_CREAT | O_EXCL))
522				error = EEXIST;
523			else {
524#ifdef MAC
525				error = mac_posixshm_check_open(td->td_ucred,
526				    shmfd);
527				if (error == 0)
528#endif
529				error = shm_access(shmfd, td->td_ucred,
530				    FFLAGS(uap->flags & O_ACCMODE));
531			}
532
533			/*
534			 * Truncate the file back to zero length if
535			 * O_TRUNC was specified and the object was
536			 * opened with read/write.
537			 */
538			if (error == 0 &&
539			    (uap->flags & (O_ACCMODE | O_TRUNC)) ==
540			    (O_RDWR | O_TRUNC)) {
541#ifdef MAC
542				error = mac_posixshm_check_truncate(
543					td->td_ucred, fp->f_cred, shmfd);
544				if (error == 0)
545#endif
546					shm_dotruncate(shmfd, 0);
547			}
548			if (error == 0)
549				shm_hold(shmfd);
550		}
551		sx_xunlock(&shm_dict_lock);
552
553		if (error) {
554			fdclose(fdp, fp, fd, td);
555			fdrop(fp, td);
556			return (error);
557		}
558	}
559
560	finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
561
562	FILEDESC_XLOCK(fdp);
563	if (fdp->fd_ofiles[fd] == fp)
564		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
565	FILEDESC_XUNLOCK(fdp);
566	td->td_retval[0] = fd;
567	fdrop(fp, td);
568
569	return (0);
570}
571
572int
573shm_unlink(struct thread *td, struct shm_unlink_args *uap)
574{
575	char *path;
576	Fnv32_t fnv;
577	int error;
578
579	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
580	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
581	if (error) {
582		free(path, M_TEMP);
583		return (error);
584	}
585
586	fnv = fnv_32_str(path, FNV1_32_INIT);
587	sx_xlock(&shm_dict_lock);
588	error = shm_remove(path, fnv, td->td_ucred);
589	sx_xunlock(&shm_dict_lock);
590	free(path, M_TEMP);
591
592	return (error);
593}
594
595/*
596 * mmap() helper to validate mmap() requests against shm object state
597 * and give mmap() the vm_object to use for the mapping.
598 */
599int
600shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
601    vm_object_t *obj)
602{
603
604	/*
605	 * XXXRW: This validation is probably insufficient, and subject to
606	 * sign errors.  It should be fixed.
607	 */
608	if (foff >= shmfd->shm_size ||
609	    foff + objsize > round_page(shmfd->shm_size))
610		return (EINVAL);
611
612	mtx_lock(&shm_timestamp_lock);
613	vfs_timestamp(&shmfd->shm_atime);
614	mtx_unlock(&shm_timestamp_lock);
615	vm_object_reference(shmfd->shm_object);
616	*obj = shmfd->shm_object;
617	return (0);
618}
619