2 * Copyright (c) 2006, 2011, 2016-2017 Robert N. M. Watson
5 * Portions of this software were developed by BAE Systems, the University of
6 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
7 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
8 * Computing (TC) research program.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * Support for shared swap-backed anonymous memory objects via
34 * shm_open(2) and shm_unlink(2). While most of the implementation is
35 * here, vm_mmap.c contains mapping logic changes.
39 * (1) Need to export data to a userland tool via a sysctl. Should ipcs(1)
40 * and ipcrm(1) be expanded or should new tools to manage both POSIX
41 * kernel semaphores and POSIX shared memory be written?
43 * (2) Add support for this file type to fstat(1).
45 * (3) Resource limits? Does this need its own resource limits or are the
46 * existing limits in mmap(2) sufficient?
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
52 #include "opt_capsicum.h"
53 #include "opt_ktrace.h"
55 #include <sys/param.h>
56 #include <sys/capsicum.h>
58 #include <sys/fcntl.h>
60 #include <sys/filedesc.h>
61 #include <sys/fnv_hash.h>
62 #include <sys/kernel.h>
64 #include <sys/signal.h>
66 #include <sys/ktrace.h>
68 #include <sys/malloc.h>
70 #include <sys/mutex.h>
73 #include <sys/refcount.h>
74 #include <sys/resourcevar.h>
75 #include <sys/rwlock.h>
77 #include <sys/syscallsubr.h>
78 #include <sys/sysctl.h>
79 #include <sys/sysproto.h>
80 #include <sys/systm.h>
83 #include <sys/vnode.h>
84 #include <sys/unistd.h>
87 #include <security/audit/audit.h>
88 #include <security/mac/mac_framework.h>
91 #include <vm/vm_param.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vm_map.h>
95 #include <vm/vm_kern.h>
96 #include <vm/vm_object.h>
97 #include <vm/vm_page.h>
98 #include <vm/vm_pageout.h>
99 #include <vm/vm_pager.h>
100 #include <vm/swap_pager.h>
105 struct shmfd *sm_shmfd;
106 LIST_ENTRY(shm_mapping) sm_link;
109 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
110 static LIST_HEAD(, shm_mapping) *shm_dictionary;
111 static struct sx shm_dict_lock;
112 static struct mtx shm_timestamp_lock;
113 static u_long shm_hash;
114 static struct unrhdr *shm_ino_unr;
115 static dev_t shm_dev_ino;
117 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
119 static void shm_init(void *arg);
120 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
121 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
122 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
124 static fo_rdwr_t shm_read;
125 static fo_rdwr_t shm_write;
126 static fo_truncate_t shm_truncate;
127 static fo_stat_t shm_stat;
128 static fo_close_t shm_close;
129 static fo_chmod_t shm_chmod;
130 static fo_chown_t shm_chown;
131 static fo_seek_t shm_seek;
132 static fo_fill_kinfo_t shm_fill_kinfo;
133 static fo_mmap_t shm_mmap;
135 /* File descriptor operations. */
136 struct fileops shm_ops = {
138 .fo_write = shm_write,
139 .fo_truncate = shm_truncate,
140 .fo_ioctl = invfo_ioctl,
141 .fo_poll = invfo_poll,
142 .fo_kqfilter = invfo_kqfilter,
144 .fo_close = shm_close,
145 .fo_chmod = shm_chmod,
146 .fo_chown = shm_chown,
147 .fo_sendfile = vn_sendfile,
149 .fo_fill_kinfo = shm_fill_kinfo,
151 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
154 FEATURE(posix_shm, "POSIX shared memory");
157 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
162 int error, offset, rv;
164 idx = OFF_TO_IDX(uio->uio_offset);
165 offset = uio->uio_offset & PAGE_MASK;
166 tlen = MIN(PAGE_SIZE - offset, len);
168 VM_OBJECT_WLOCK(obj);
171 * Read I/O without either a corresponding resident page or swap
172 * page: use zero_region. This is intended to avoid instantiating
173 * pages on read from a sparse region.
175 if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL &&
176 !vm_pager_has_page(obj, idx, NULL, NULL)) {
177 VM_OBJECT_WUNLOCK(obj);
178 return (uiomove(__DECONST(void *, zero_region), tlen, uio));
182 * Parallel reads of the page content from disk are prevented
185 * Although the tmpfs vnode lock is held here, it is
186 * nonetheless safe to sleep waiting for a free page. The
187 * pageout daemon does not need to acquire the tmpfs vnode
188 * lock to page out tobj's pages because tobj is a OBJT_SWAP
191 m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
192 if (m->valid != VM_PAGE_BITS_ALL) {
194 if (vm_pager_has_page(obj, idx, NULL, NULL)) {
195 rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL);
196 if (rv != VM_PAGER_OK) {
198 "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
199 obj, idx, m->valid, rv);
203 VM_OBJECT_WUNLOCK(obj);
207 vm_page_zero_invalid(m, TRUE);
212 if (vm_page_active(m))
213 vm_page_reference(m);
217 VM_OBJECT_WUNLOCK(obj);
218 error = uiomove_fromphys(&m, offset, tlen, uio);
219 if (uio->uio_rw == UIO_WRITE && error == 0) {
220 VM_OBJECT_WLOCK(obj);
222 vm_pager_page_unswapped(m);
223 VM_OBJECT_WUNLOCK(obj);
233 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
240 while ((resid = uio->uio_resid) > 0) {
241 if (obj_size <= uio->uio_offset)
243 len = MIN(obj_size - uio->uio_offset, resid);
246 error = uiomove_object_page(obj, len, uio);
247 if (error != 0 || resid == uio->uio_resid)
254 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
261 foffset = foffset_lock(fp, 0);
266 (offset > 0 && foffset > OFF_MAX - offset)) {
273 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
277 offset += shmfd->shm_size;
285 if (offset < 0 || offset > shmfd->shm_size)
288 td->td_uretoff.tdu_off = offset;
290 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
295 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
296 int flags, struct thread *td)
304 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
308 foffset_lock_uio(fp, uio, flags);
309 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
310 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
311 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
312 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
313 foffset_unlock_uio(fp, uio, flags);
318 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
319 int flags, struct thread *td)
327 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
331 foffset_lock_uio(fp, uio, flags);
332 if ((flags & FOF_OFFSET) == 0) {
333 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
336 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
337 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
340 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
341 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
342 foffset_unlock_uio(fp, uio, flags);
347 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
357 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
361 return (shm_dotruncate(shmfd, length));
365 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
376 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
382 * Attempt to return sanish values for fstat() on a memory file
385 bzero(sb, sizeof(*sb));
386 sb->st_blksize = PAGE_SIZE;
387 sb->st_size = shmfd->shm_size;
388 sb->st_blocks = howmany(sb->st_size, sb->st_blksize);
389 mtx_lock(&shm_timestamp_lock);
390 sb->st_atim = shmfd->shm_atime;
391 sb->st_ctim = shmfd->shm_ctime;
392 sb->st_mtim = shmfd->shm_mtime;
393 sb->st_birthtim = shmfd->shm_birthtime;
394 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
395 sb->st_uid = shmfd->shm_uid;
396 sb->st_gid = shmfd->shm_gid;
397 mtx_unlock(&shm_timestamp_lock);
398 sb->st_dev = shm_dev_ino;
399 sb->st_ino = shmfd->shm_ino;
405 shm_close(struct file *fp, struct thread *td)
417 shm_dotruncate(struct shmfd *shmfd, off_t length)
421 vm_pindex_t idx, nobjsize;
425 KASSERT(length >= 0, ("shm_dotruncate: length < 0"));
426 object = shmfd->shm_object;
427 VM_OBJECT_WLOCK(object);
428 if (length == shmfd->shm_size) {
429 VM_OBJECT_WUNLOCK(object);
432 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
434 /* Are we shrinking? If so, trim the end. */
435 if (length < shmfd->shm_size) {
437 * Disallow any requests to shrink the size if this
438 * object is mapped into the kernel.
440 if (shmfd->shm_kmappings > 0) {
441 VM_OBJECT_WUNLOCK(object);
446 * Zero the truncated part of the last page.
448 base = length & PAGE_MASK;
450 idx = OFF_TO_IDX(length);
452 m = vm_page_lookup(object, idx);
454 if (vm_page_sleep_if_busy(m, "shmtrc"))
456 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
457 m = vm_page_alloc(object, idx,
458 VM_ALLOC_NORMAL | VM_ALLOC_WAITFAIL);
461 rv = vm_pager_get_pages(object, &m, 1, NULL,
464 if (rv == VM_PAGER_OK) {
466 * Since the page was not resident,
467 * and therefore not recently
468 * accessed, immediately enqueue it
469 * for asynchronous laundering. The
470 * current operation is not regarded
479 VM_OBJECT_WUNLOCK(object);
484 pmap_zero_page_area(m, base, PAGE_SIZE - base);
485 KASSERT(m->valid == VM_PAGE_BITS_ALL,
486 ("shm_dotruncate: page %p is invalid", m));
488 vm_pager_page_unswapped(m);
491 delta = IDX_TO_OFF(object->size - nobjsize);
493 /* Toss in memory pages. */
494 if (nobjsize < object->size)
495 vm_object_page_remove(object, nobjsize, object->size,
498 /* Toss pages from swap. */
499 if (object->type == OBJT_SWAP)
500 swap_pager_freespace(object, nobjsize, delta);
502 /* Free the swap accounted for shm */
503 swap_release_by_cred(delta, object->cred);
504 object->charge -= delta;
506 /* Try to reserve additional swap space. */
507 delta = IDX_TO_OFF(nobjsize - object->size);
508 if (!swap_reserve_by_cred(delta, object->cred)) {
509 VM_OBJECT_WUNLOCK(object);
512 object->charge += delta;
514 shmfd->shm_size = length;
515 mtx_lock(&shm_timestamp_lock);
516 vfs_timestamp(&shmfd->shm_ctime);
517 shmfd->shm_mtime = shmfd->shm_ctime;
518 mtx_unlock(&shm_timestamp_lock);
519 object->size = nobjsize;
520 VM_OBJECT_WUNLOCK(object);
525 * shmfd object management including creation and reference counting
529 shm_alloc(struct ucred *ucred, mode_t mode)
534 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
536 shmfd->shm_uid = ucred->cr_uid;
537 shmfd->shm_gid = ucred->cr_gid;
538 shmfd->shm_mode = mode;
539 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
540 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
541 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
542 shmfd->shm_object->pg_color = 0;
543 VM_OBJECT_WLOCK(shmfd->shm_object);
544 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
545 vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT);
546 VM_OBJECT_WUNLOCK(shmfd->shm_object);
547 vfs_timestamp(&shmfd->shm_birthtime);
548 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
549 shmfd->shm_birthtime;
550 ino = alloc_unr(shm_ino_unr);
554 shmfd->shm_ino = ino;
555 refcount_init(&shmfd->shm_refs, 1);
556 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
557 rangelock_init(&shmfd->shm_rl);
559 mac_posixshm_init(shmfd);
560 mac_posixshm_create(ucred, shmfd);
567 shm_hold(struct shmfd *shmfd)
570 refcount_acquire(&shmfd->shm_refs);
575 shm_drop(struct shmfd *shmfd)
578 if (refcount_release(&shmfd->shm_refs)) {
580 mac_posixshm_destroy(shmfd);
582 rangelock_destroy(&shmfd->shm_rl);
583 mtx_destroy(&shmfd->shm_mtx);
584 vm_object_deallocate(shmfd->shm_object);
585 if (shmfd->shm_ino != 0)
586 free_unr(shm_ino_unr, shmfd->shm_ino);
587 free(shmfd, M_SHMFD);
592 * Determine if the credentials have sufficient permissions for a
593 * specified combination of FREAD and FWRITE.
596 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
606 mtx_lock(&shm_timestamp_lock);
607 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
608 accmode, ucred, NULL);
609 mtx_unlock(&shm_timestamp_lock);
614 * Dictionary management. We maintain an in-kernel dictionary to map
615 * paths to shmfd objects. We use the FNV hash on the path to store
616 * the mappings in a hash table.
622 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
623 sx_init(&shm_dict_lock, "shm dictionary");
624 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
625 shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
626 KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
627 shm_dev_ino = devfs_alloc_cdp_inode();
628 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
630 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
632 static struct shmfd *
633 shm_lookup(char *path, Fnv32_t fnv)
635 struct shm_mapping *map;
637 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
638 if (map->sm_fnv != fnv)
640 if (strcmp(map->sm_path, path) == 0)
641 return (map->sm_shmfd);
648 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
650 struct shm_mapping *map;
652 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
655 map->sm_shmfd = shm_hold(shmfd);
656 shmfd->shm_path = path;
657 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
661 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
663 struct shm_mapping *map;
666 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
667 if (map->sm_fnv != fnv)
669 if (strcmp(map->sm_path, path) == 0) {
671 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
675 error = shm_access(map->sm_shmfd, ucred,
679 map->sm_shmfd->shm_path = NULL;
680 LIST_REMOVE(map, sm_link);
681 shm_drop(map->sm_shmfd);
682 free(map->sm_path, M_SHMFD);
692 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode,
693 struct filecaps *fcaps)
695 struct filedesc *fdp;
705 #ifdef CAPABILITY_MODE
707 * shm_open(2) is only allowed for anonymous objects.
709 if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON))
713 AUDIT_ARG_FFLAGS(flags);
714 AUDIT_ARG_MODE(mode);
716 if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR)
719 if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0)
722 fdp = td->td_proc->p_fd;
723 cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
725 error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps);
729 /* A SHM_ANON path pointer creates an anonymous object. */
730 if (userpath == SHM_ANON) {
731 /* A read-only anonymous object is pointless. */
732 if ((flags & O_ACCMODE) == O_RDONLY) {
737 shmfd = shm_alloc(td->td_ucred, cmode);
739 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
740 pr_path = td->td_ucred->cr_prison->pr_path;
742 /* Construct a full pathname for jailed callers. */
743 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
744 : strlcpy(path, pr_path, MAXPATHLEN);
745 error = copyinstr(userpath, path + pr_pathlen,
746 MAXPATHLEN - pr_pathlen, NULL);
748 if (error == 0 && KTRPOINT(curthread, KTR_NAMEI))
751 /* Require paths to start with a '/' character. */
752 if (error == 0 && path[pr_pathlen] != '/')
761 AUDIT_ARG_UPATH1_CANON(path);
762 fnv = fnv_32_str(path, FNV1_32_INIT);
763 sx_xlock(&shm_dict_lock);
764 shmfd = shm_lookup(path, fnv);
766 /* Object does not yet exist, create it if requested. */
767 if (flags & O_CREAT) {
769 error = mac_posixshm_check_create(td->td_ucred,
773 shmfd = shm_alloc(td->td_ucred, cmode);
774 shm_insert(path, fnv, shmfd);
784 * Object already exists, obtain a new
785 * reference if requested and permitted.
788 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
792 error = mac_posixshm_check_open(td->td_ucred,
793 shmfd, FFLAGS(flags & O_ACCMODE));
796 error = shm_access(shmfd, td->td_ucred,
797 FFLAGS(flags & O_ACCMODE));
801 * Truncate the file back to zero length if
802 * O_TRUNC was specified and the object was
803 * opened with read/write.
806 (flags & (O_ACCMODE | O_TRUNC)) ==
807 (O_RDWR | O_TRUNC)) {
809 error = mac_posixshm_check_truncate(
810 td->td_ucred, fp->f_cred, shmfd);
813 shm_dotruncate(shmfd, 0);
818 sx_xunlock(&shm_dict_lock);
827 finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
829 td->td_retval[0] = fd;
837 sys_shm_open(struct thread *td, struct shm_open_args *uap)
840 return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL));
844 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
852 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
853 pr_path = td->td_ucred->cr_prison->pr_path;
854 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
855 : strlcpy(path, pr_path, MAXPATHLEN);
856 error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen,
863 if (KTRPOINT(curthread, KTR_NAMEI))
866 AUDIT_ARG_UPATH1_CANON(path);
867 fnv = fnv_32_str(path, FNV1_32_INIT);
868 sx_xlock(&shm_dict_lock);
869 error = shm_remove(path, fnv, td->td_ucred);
870 sx_xunlock(&shm_dict_lock);
877 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize,
878 vm_prot_t prot, vm_prot_t cap_maxprot, int flags,
879 vm_ooffset_t foff, struct thread *td)
886 maxprot = VM_PROT_NONE;
888 /* FREAD should always be set. */
889 if ((fp->f_flag & FREAD) != 0)
890 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
891 if ((fp->f_flag & FWRITE) != 0)
892 maxprot |= VM_PROT_WRITE;
894 /* Don't permit shared writable mappings on read-only descriptors. */
895 if ((flags & MAP_SHARED) != 0 &&
896 (maxprot & VM_PROT_WRITE) == 0 &&
897 (prot & VM_PROT_WRITE) != 0)
899 maxprot &= cap_maxprot;
901 /* See comment in vn_mmap(). */
906 foff < 0 || foff > OFF_MAX - objsize)
910 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags);
915 mtx_lock(&shm_timestamp_lock);
916 vfs_timestamp(&shmfd->shm_atime);
917 mtx_unlock(&shm_timestamp_lock);
918 vm_object_reference(shmfd->shm_object);
920 error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags,
921 shmfd->shm_object, foff, FALSE, td);
923 vm_object_deallocate(shmfd->shm_object);
928 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
936 mtx_lock(&shm_timestamp_lock);
938 * SUSv4 says that x bits of permission need not be affected.
939 * Be consistent with our shm_open there.
942 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
946 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
947 shmfd->shm_gid, VADMIN, active_cred, NULL);
950 shmfd->shm_mode = mode & ACCESSPERMS;
952 mtx_unlock(&shm_timestamp_lock);
957 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
965 mtx_lock(&shm_timestamp_lock);
967 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
971 if (uid == (uid_t)-1)
972 uid = shmfd->shm_uid;
973 if (gid == (gid_t)-1)
974 gid = shmfd->shm_gid;
975 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
976 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
977 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
979 shmfd->shm_uid = uid;
980 shmfd->shm_gid = gid;
982 mtx_unlock(&shm_timestamp_lock);
987 * Helper routines to allow the backing object of a shared memory file
988 * descriptor to be mapped in the kernel.
991 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
994 vm_offset_t kva, ofs;
998 if (fp->f_type != DTYPE_SHM)
1001 obj = shmfd->shm_object;
1002 VM_OBJECT_WLOCK(obj);
1004 * XXXRW: This validation is probably insufficient, and subject to
1005 * sign errors. It should be fixed.
1007 if (offset >= shmfd->shm_size ||
1008 offset + size > round_page(shmfd->shm_size)) {
1009 VM_OBJECT_WUNLOCK(obj);
1013 shmfd->shm_kmappings++;
1014 vm_object_reference_locked(obj);
1015 VM_OBJECT_WUNLOCK(obj);
1017 /* Map the object into the kernel_map and wire it. */
1018 kva = vm_map_min(kernel_map);
1019 ofs = offset & PAGE_MASK;
1020 offset = trunc_page(offset);
1021 size = round_page(size + ofs);
1022 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
1023 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
1024 VM_PROT_READ | VM_PROT_WRITE, 0);
1025 if (rv == KERN_SUCCESS) {
1026 rv = vm_map_wire(kernel_map, kva, kva + size,
1027 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
1028 if (rv == KERN_SUCCESS) {
1029 *memp = (void *)(kva + ofs);
1032 vm_map_remove(kernel_map, kva, kva + size);
1034 vm_object_deallocate(obj);
1036 /* On failure, drop our mapping reference. */
1037 VM_OBJECT_WLOCK(obj);
1038 shmfd->shm_kmappings--;
1039 VM_OBJECT_WUNLOCK(obj);
1041 return (vm_mmap_to_errno(rv));
1045 * We require the caller to unmap the entire entry. This allows us to
1046 * safely decrement shm_kmappings when a mapping is removed.
1049 shm_unmap(struct file *fp, void *mem, size_t size)
1051 struct shmfd *shmfd;
1052 vm_map_entry_t entry;
1053 vm_offset_t kva, ofs;
1061 if (fp->f_type != DTYPE_SHM)
1064 kva = (vm_offset_t)mem;
1065 ofs = kva & PAGE_MASK;
1066 kva = trunc_page(kva);
1067 size = round_page(size + ofs);
1069 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1070 &obj, &pindex, &prot, &wired);
1071 if (rv != KERN_SUCCESS)
1073 if (entry->start != kva || entry->end != kva + size) {
1074 vm_map_lookup_done(map, entry);
1077 vm_map_lookup_done(map, entry);
1078 if (obj != shmfd->shm_object)
1080 vm_map_remove(map, kva, kva + size);
1081 VM_OBJECT_WLOCK(obj);
1082 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1083 shmfd->shm_kmappings--;
1084 VM_OBJECT_WUNLOCK(obj);
1089 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
1091 const char *path, *pr_path;
1092 struct shmfd *shmfd;
1095 kif->kf_type = KF_TYPE_SHM;
1098 mtx_lock(&shm_timestamp_lock);
1099 kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode; /* XXX */
1100 mtx_unlock(&shm_timestamp_lock);
1101 kif->kf_un.kf_file.kf_file_size = shmfd->shm_size;
1102 if (shmfd->shm_path != NULL) {
1103 sx_slock(&shm_dict_lock);
1104 if (shmfd->shm_path != NULL) {
1105 path = shmfd->shm_path;
1106 pr_path = curthread->td_ucred->cr_prison->pr_path;
1107 if (strcmp(pr_path, "/") != 0) {
1108 /* Return the jail-rooted pathname. */
1109 pr_pathlen = strlen(pr_path);
1110 if (strncmp(path, pr_path, pr_pathlen) == 0 &&
1111 path[pr_pathlen] == '/')
1114 strlcpy(kif->kf_path, path, sizeof(kif->kf_path));
1116 sx_sunlock(&shm_dict_lock);