2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2006, 2011, 2016-2017 Robert N. M. Watson
7 * Portions of this software were developed by BAE Systems, the University of
8 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
9 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
10 * Computing (TC) research program.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Support for shared swap-backed anonymous memory objects via
36 * shm_open(2) and shm_unlink(2). While most of the implementation is
37 * here, vm_mmap.c contains mapping logic changes.
39 * posixshmcontrol(1) allows users to inspect the state of the memory
40 * objects. Per-uid swap resource limit controls total amount of
41 * memory that user can consume for anonymous objects, including
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
48 #include "opt_capsicum.h"
49 #include "opt_ktrace.h"
51 #include <sys/param.h>
52 #include <sys/capsicum.h>
54 #include <sys/fcntl.h>
56 #include <sys/filedesc.h>
57 #include <sys/filio.h>
58 #include <sys/fnv_hash.h>
59 #include <sys/kernel.h>
61 #include <sys/signal.h>
63 #include <sys/ktrace.h>
65 #include <sys/malloc.h>
67 #include <sys/mutex.h>
70 #include <sys/refcount.h>
71 #include <sys/resourcevar.h>
72 #include <sys/rwlock.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/systm.h>
81 #include <sys/vnode.h>
82 #include <sys/unistd.h>
85 #include <security/audit/audit.h>
86 #include <security/mac/mac_framework.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_extern.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_object.h>
95 #include <vm/vm_page.h>
96 #include <vm/vm_pageout.h>
97 #include <vm/vm_pager.h>
98 #include <vm/swap_pager.h>
103 struct shmfd *sm_shmfd;
104 LIST_ENTRY(shm_mapping) sm_link;
107 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
108 static LIST_HEAD(, shm_mapping) *shm_dictionary;
109 static struct sx shm_dict_lock;
110 static struct mtx shm_timestamp_lock;
111 static u_long shm_hash;
112 static struct unrhdr64 shm_ino_unr;
113 static dev_t shm_dev_ino;
115 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
117 static void shm_init(void *arg);
118 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
119 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
120 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
122 static fo_rdwr_t shm_read;
123 static fo_rdwr_t shm_write;
124 static fo_truncate_t shm_truncate;
125 static fo_ioctl_t shm_ioctl;
126 static fo_stat_t shm_stat;
127 static fo_close_t shm_close;
128 static fo_chmod_t shm_chmod;
129 static fo_chown_t shm_chown;
130 static fo_seek_t shm_seek;
131 static fo_fill_kinfo_t shm_fill_kinfo;
132 static fo_mmap_t shm_mmap;
134 /* File descriptor operations. */
135 struct fileops shm_ops = {
137 .fo_write = shm_write,
138 .fo_truncate = shm_truncate,
139 .fo_ioctl = shm_ioctl,
140 .fo_poll = invfo_poll,
141 .fo_kqfilter = invfo_kqfilter,
143 .fo_close = shm_close,
144 .fo_chmod = shm_chmod,
145 .fo_chown = shm_chown,
146 .fo_sendfile = vn_sendfile,
148 .fo_fill_kinfo = shm_fill_kinfo,
150 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
153 FEATURE(posix_shm, "POSIX shared memory");
156 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
161 int error, offset, rv;
163 idx = OFF_TO_IDX(uio->uio_offset);
164 offset = uio->uio_offset & PAGE_MASK;
165 tlen = MIN(PAGE_SIZE - offset, len);
167 VM_OBJECT_WLOCK(obj);
170 * Read I/O without either a corresponding resident page or swap
171 * page: use zero_region. This is intended to avoid instantiating
172 * pages on read from a sparse region.
174 if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL &&
175 !vm_pager_has_page(obj, idx, NULL, NULL)) {
176 VM_OBJECT_WUNLOCK(obj);
177 return (uiomove(__DECONST(void *, zero_region), tlen, uio));
181 * Parallel reads of the page content from disk are prevented
184 * Although the tmpfs vnode lock is held here, it is
185 * nonetheless safe to sleep waiting for a free page. The
186 * pageout daemon does not need to acquire the tmpfs vnode
187 * lock to page out tobj's pages because tobj is a OBJT_SWAP
190 m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
191 if (m->valid != VM_PAGE_BITS_ALL) {
193 if (vm_pager_has_page(obj, idx, NULL, NULL)) {
194 rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL);
195 if (rv != VM_PAGER_OK) {
197 "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
198 obj, idx, m->valid, rv);
202 VM_OBJECT_WUNLOCK(obj);
206 vm_page_zero_invalid(m, TRUE);
211 if (vm_page_active(m))
212 vm_page_reference(m);
216 VM_OBJECT_WUNLOCK(obj);
217 error = uiomove_fromphys(&m, offset, tlen, uio);
218 if (uio->uio_rw == UIO_WRITE && error == 0) {
219 VM_OBJECT_WLOCK(obj);
221 vm_pager_page_unswapped(m);
222 VM_OBJECT_WUNLOCK(obj);
232 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
239 while ((resid = uio->uio_resid) > 0) {
240 if (obj_size <= uio->uio_offset)
242 len = MIN(obj_size - uio->uio_offset, resid);
245 error = uiomove_object_page(obj, len, uio);
246 if (error != 0 || resid == uio->uio_resid)
253 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
260 foffset = foffset_lock(fp, 0);
265 (offset > 0 && foffset > OFF_MAX - offset)) {
272 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
276 offset += shmfd->shm_size;
284 if (offset < 0 || offset > shmfd->shm_size)
287 td->td_uretoff.tdu_off = offset;
289 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
294 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
295 int flags, struct thread *td)
303 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
307 foffset_lock_uio(fp, uio, flags);
308 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
309 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
310 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
311 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
312 foffset_unlock_uio(fp, uio, flags);
317 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
318 int flags, struct thread *td)
326 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
330 foffset_lock_uio(fp, uio, flags);
331 if ((flags & FOF_OFFSET) == 0) {
332 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
335 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
336 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
339 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
340 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
341 foffset_unlock_uio(fp, uio, flags);
346 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
356 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
360 return (shm_dotruncate(shmfd, length));
364 shm_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
372 * Allow fcntl(fd, F_SETFL, O_NONBLOCK) to work,
373 * just like it would on an unlinked regular file
382 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
393 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
399 * Attempt to return sanish values for fstat() on a memory file
402 bzero(sb, sizeof(*sb));
403 sb->st_blksize = PAGE_SIZE;
404 sb->st_size = shmfd->shm_size;
405 sb->st_blocks = howmany(sb->st_size, sb->st_blksize);
406 mtx_lock(&shm_timestamp_lock);
407 sb->st_atim = shmfd->shm_atime;
408 sb->st_ctim = shmfd->shm_ctime;
409 sb->st_mtim = shmfd->shm_mtime;
410 sb->st_birthtim = shmfd->shm_birthtime;
411 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
412 sb->st_uid = shmfd->shm_uid;
413 sb->st_gid = shmfd->shm_gid;
414 mtx_unlock(&shm_timestamp_lock);
415 sb->st_dev = shm_dev_ino;
416 sb->st_ino = shmfd->shm_ino;
417 sb->st_nlink = shmfd->shm_object->ref_count;
423 shm_close(struct file *fp, struct thread *td)
435 shm_dotruncate(struct shmfd *shmfd, off_t length)
439 vm_pindex_t idx, nobjsize;
443 KASSERT(length >= 0, ("shm_dotruncate: length < 0"));
444 object = shmfd->shm_object;
445 VM_OBJECT_WLOCK(object);
446 if (length == shmfd->shm_size) {
447 VM_OBJECT_WUNLOCK(object);
450 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
452 /* Are we shrinking? If so, trim the end. */
453 if (length < shmfd->shm_size) {
455 * Disallow any requests to shrink the size if this
456 * object is mapped into the kernel.
458 if (shmfd->shm_kmappings > 0) {
459 VM_OBJECT_WUNLOCK(object);
464 * Zero the truncated part of the last page.
466 base = length & PAGE_MASK;
468 idx = OFF_TO_IDX(length);
470 m = vm_page_lookup(object, idx);
472 if (vm_page_sleep_if_busy(m, "shmtrc"))
474 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
475 m = vm_page_alloc(object, idx,
476 VM_ALLOC_NORMAL | VM_ALLOC_WAITFAIL);
479 rv = vm_pager_get_pages(object, &m, 1, NULL,
482 if (rv == VM_PAGER_OK) {
484 * Since the page was not resident,
485 * and therefore not recently
486 * accessed, immediately enqueue it
487 * for asynchronous laundering. The
488 * current operation is not regarded
497 VM_OBJECT_WUNLOCK(object);
502 pmap_zero_page_area(m, base, PAGE_SIZE - base);
503 KASSERT(m->valid == VM_PAGE_BITS_ALL,
504 ("shm_dotruncate: page %p is invalid", m));
506 vm_pager_page_unswapped(m);
509 delta = IDX_TO_OFF(object->size - nobjsize);
511 /* Toss in memory pages. */
512 if (nobjsize < object->size)
513 vm_object_page_remove(object, nobjsize, object->size,
516 /* Toss pages from swap. */
517 if (object->type == OBJT_SWAP)
518 swap_pager_freespace(object, nobjsize, delta);
520 /* Free the swap accounted for shm */
521 swap_release_by_cred(delta, object->cred);
522 object->charge -= delta;
524 /* Try to reserve additional swap space. */
525 delta = IDX_TO_OFF(nobjsize - object->size);
526 if (!swap_reserve_by_cred(delta, object->cred)) {
527 VM_OBJECT_WUNLOCK(object);
530 object->charge += delta;
532 shmfd->shm_size = length;
533 mtx_lock(&shm_timestamp_lock);
534 vfs_timestamp(&shmfd->shm_ctime);
535 shmfd->shm_mtime = shmfd->shm_ctime;
536 mtx_unlock(&shm_timestamp_lock);
537 object->size = nobjsize;
538 VM_OBJECT_WUNLOCK(object);
543 * shmfd object management including creation and reference counting
547 shm_alloc(struct ucred *ucred, mode_t mode)
551 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
553 shmfd->shm_uid = ucred->cr_uid;
554 shmfd->shm_gid = ucred->cr_gid;
555 shmfd->shm_mode = mode;
556 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
557 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
558 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
559 shmfd->shm_object->pg_color = 0;
560 VM_OBJECT_WLOCK(shmfd->shm_object);
561 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
562 vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT);
563 VM_OBJECT_WUNLOCK(shmfd->shm_object);
564 vfs_timestamp(&shmfd->shm_birthtime);
565 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
566 shmfd->shm_birthtime;
567 shmfd->shm_ino = alloc_unr64(&shm_ino_unr);
568 refcount_init(&shmfd->shm_refs, 1);
569 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
570 rangelock_init(&shmfd->shm_rl);
572 mac_posixshm_init(shmfd);
573 mac_posixshm_create(ucred, shmfd);
580 shm_hold(struct shmfd *shmfd)
583 refcount_acquire(&shmfd->shm_refs);
588 shm_drop(struct shmfd *shmfd)
591 if (refcount_release(&shmfd->shm_refs)) {
593 mac_posixshm_destroy(shmfd);
595 rangelock_destroy(&shmfd->shm_rl);
596 mtx_destroy(&shmfd->shm_mtx);
597 vm_object_deallocate(shmfd->shm_object);
598 free(shmfd, M_SHMFD);
603 * Determine if the credentials have sufficient permissions for a
604 * specified combination of FREAD and FWRITE.
607 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
617 mtx_lock(&shm_timestamp_lock);
618 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
619 accmode, ucred, NULL);
620 mtx_unlock(&shm_timestamp_lock);
625 * Dictionary management. We maintain an in-kernel dictionary to map
626 * paths to shmfd objects. We use the FNV hash on the path to store
627 * the mappings in a hash table.
633 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
634 sx_init(&shm_dict_lock, "shm dictionary");
635 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
636 new_unrhdr64(&shm_ino_unr, 1);
637 shm_dev_ino = devfs_alloc_cdp_inode();
638 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
640 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
642 static struct shmfd *
643 shm_lookup(char *path, Fnv32_t fnv)
645 struct shm_mapping *map;
647 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
648 if (map->sm_fnv != fnv)
650 if (strcmp(map->sm_path, path) == 0)
651 return (map->sm_shmfd);
658 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
660 struct shm_mapping *map;
662 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
665 map->sm_shmfd = shm_hold(shmfd);
666 shmfd->shm_path = path;
667 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
671 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
673 struct shm_mapping *map;
676 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
677 if (map->sm_fnv != fnv)
679 if (strcmp(map->sm_path, path) == 0) {
681 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
685 error = shm_access(map->sm_shmfd, ucred,
689 map->sm_shmfd->shm_path = NULL;
690 LIST_REMOVE(map, sm_link);
691 shm_drop(map->sm_shmfd);
692 free(map->sm_path, M_SHMFD);
702 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode,
703 struct filecaps *fcaps)
705 struct filedesc *fdp;
715 #ifdef CAPABILITY_MODE
717 * shm_open(2) is only allowed for anonymous objects.
719 if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON))
723 AUDIT_ARG_FFLAGS(flags);
724 AUDIT_ARG_MODE(mode);
726 if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR)
729 if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0)
732 fdp = td->td_proc->p_fd;
733 cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
735 error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps);
739 /* A SHM_ANON path pointer creates an anonymous object. */
740 if (userpath == SHM_ANON) {
741 /* A read-only anonymous object is pointless. */
742 if ((flags & O_ACCMODE) == O_RDONLY) {
747 shmfd = shm_alloc(td->td_ucred, cmode);
749 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
750 pr_path = td->td_ucred->cr_prison->pr_path;
752 /* Construct a full pathname for jailed callers. */
753 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
754 : strlcpy(path, pr_path, MAXPATHLEN);
755 error = copyinstr(userpath, path + pr_pathlen,
756 MAXPATHLEN - pr_pathlen, NULL);
758 if (error == 0 && KTRPOINT(curthread, KTR_NAMEI))
761 /* Require paths to start with a '/' character. */
762 if (error == 0 && path[pr_pathlen] != '/')
771 AUDIT_ARG_UPATH1_CANON(path);
772 fnv = fnv_32_str(path, FNV1_32_INIT);
773 sx_xlock(&shm_dict_lock);
774 shmfd = shm_lookup(path, fnv);
776 /* Object does not yet exist, create it if requested. */
777 if (flags & O_CREAT) {
779 error = mac_posixshm_check_create(td->td_ucred,
783 shmfd = shm_alloc(td->td_ucred, cmode);
784 shm_insert(path, fnv, shmfd);
794 * Object already exists, obtain a new
795 * reference if requested and permitted.
798 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
802 error = mac_posixshm_check_open(td->td_ucred,
803 shmfd, FFLAGS(flags & O_ACCMODE));
806 error = shm_access(shmfd, td->td_ucred,
807 FFLAGS(flags & O_ACCMODE));
811 * Truncate the file back to zero length if
812 * O_TRUNC was specified and the object was
813 * opened with read/write.
816 (flags & (O_ACCMODE | O_TRUNC)) ==
817 (O_RDWR | O_TRUNC)) {
819 error = mac_posixshm_check_truncate(
820 td->td_ucred, fp->f_cred, shmfd);
823 shm_dotruncate(shmfd, 0);
828 sx_xunlock(&shm_dict_lock);
837 finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
839 td->td_retval[0] = fd;
847 sys_shm_open(struct thread *td, struct shm_open_args *uap)
850 return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL));
854 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
862 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
863 pr_path = td->td_ucred->cr_prison->pr_path;
864 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
865 : strlcpy(path, pr_path, MAXPATHLEN);
866 error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen,
873 if (KTRPOINT(curthread, KTR_NAMEI))
876 AUDIT_ARG_UPATH1_CANON(path);
877 fnv = fnv_32_str(path, FNV1_32_INIT);
878 sx_xlock(&shm_dict_lock);
879 error = shm_remove(path, fnv, td->td_ucred);
880 sx_xunlock(&shm_dict_lock);
887 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize,
888 vm_prot_t prot, vm_prot_t cap_maxprot, int flags,
889 vm_ooffset_t foff, struct thread *td)
896 maxprot = VM_PROT_NONE;
898 /* FREAD should always be set. */
899 if ((fp->f_flag & FREAD) != 0)
900 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
901 if ((fp->f_flag & FWRITE) != 0)
902 maxprot |= VM_PROT_WRITE;
904 /* Don't permit shared writable mappings on read-only descriptors. */
905 if ((flags & MAP_SHARED) != 0 &&
906 (maxprot & VM_PROT_WRITE) == 0 &&
907 (prot & VM_PROT_WRITE) != 0)
909 maxprot &= cap_maxprot;
911 /* See comment in vn_mmap(). */
916 foff < 0 || foff > OFF_MAX - objsize)
920 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags);
925 mtx_lock(&shm_timestamp_lock);
926 vfs_timestamp(&shmfd->shm_atime);
927 mtx_unlock(&shm_timestamp_lock);
928 vm_object_reference(shmfd->shm_object);
930 error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags,
931 shmfd->shm_object, foff, FALSE, td);
933 vm_object_deallocate(shmfd->shm_object);
938 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
946 mtx_lock(&shm_timestamp_lock);
948 * SUSv4 says that x bits of permission need not be affected.
949 * Be consistent with our shm_open there.
952 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
956 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
957 shmfd->shm_gid, VADMIN, active_cred, NULL);
960 shmfd->shm_mode = mode & ACCESSPERMS;
962 mtx_unlock(&shm_timestamp_lock);
967 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
975 mtx_lock(&shm_timestamp_lock);
977 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
981 if (uid == (uid_t)-1)
982 uid = shmfd->shm_uid;
983 if (gid == (gid_t)-1)
984 gid = shmfd->shm_gid;
985 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
986 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
987 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN)))
989 shmfd->shm_uid = uid;
990 shmfd->shm_gid = gid;
992 mtx_unlock(&shm_timestamp_lock);
997 * Helper routines to allow the backing object of a shared memory file
998 * descriptor to be mapped in the kernel.
1001 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
1003 struct shmfd *shmfd;
1004 vm_offset_t kva, ofs;
1008 if (fp->f_type != DTYPE_SHM)
1011 obj = shmfd->shm_object;
1012 VM_OBJECT_WLOCK(obj);
1014 * XXXRW: This validation is probably insufficient, and subject to
1015 * sign errors. It should be fixed.
1017 if (offset >= shmfd->shm_size ||
1018 offset + size > round_page(shmfd->shm_size)) {
1019 VM_OBJECT_WUNLOCK(obj);
1023 shmfd->shm_kmappings++;
1024 vm_object_reference_locked(obj);
1025 VM_OBJECT_WUNLOCK(obj);
1027 /* Map the object into the kernel_map and wire it. */
1028 kva = vm_map_min(kernel_map);
1029 ofs = offset & PAGE_MASK;
1030 offset = trunc_page(offset);
1031 size = round_page(size + ofs);
1032 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
1033 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
1034 VM_PROT_READ | VM_PROT_WRITE, 0);
1035 if (rv == KERN_SUCCESS) {
1036 rv = vm_map_wire(kernel_map, kva, kva + size,
1037 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
1038 if (rv == KERN_SUCCESS) {
1039 *memp = (void *)(kva + ofs);
1042 vm_map_remove(kernel_map, kva, kva + size);
1044 vm_object_deallocate(obj);
1046 /* On failure, drop our mapping reference. */
1047 VM_OBJECT_WLOCK(obj);
1048 shmfd->shm_kmappings--;
1049 VM_OBJECT_WUNLOCK(obj);
1051 return (vm_mmap_to_errno(rv));
1055 * We require the caller to unmap the entire entry. This allows us to
1056 * safely decrement shm_kmappings when a mapping is removed.
1059 shm_unmap(struct file *fp, void *mem, size_t size)
1061 struct shmfd *shmfd;
1062 vm_map_entry_t entry;
1063 vm_offset_t kva, ofs;
1071 if (fp->f_type != DTYPE_SHM)
1074 kva = (vm_offset_t)mem;
1075 ofs = kva & PAGE_MASK;
1076 kva = trunc_page(kva);
1077 size = round_page(size + ofs);
1079 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1080 &obj, &pindex, &prot, &wired);
1081 if (rv != KERN_SUCCESS)
1083 if (entry->start != kva || entry->end != kva + size) {
1084 vm_map_lookup_done(map, entry);
1087 vm_map_lookup_done(map, entry);
1088 if (obj != shmfd->shm_object)
1090 vm_map_remove(map, kva, kva + size);
1091 VM_OBJECT_WLOCK(obj);
1092 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1093 shmfd->shm_kmappings--;
1094 VM_OBJECT_WUNLOCK(obj);
1099 shm_fill_kinfo_locked(struct shmfd *shmfd, struct kinfo_file *kif, bool list)
1101 const char *path, *pr_path;
1105 sx_assert(&shm_dict_lock, SA_LOCKED);
1106 kif->kf_type = KF_TYPE_SHM;
1107 kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode;
1108 kif->kf_un.kf_file.kf_file_size = shmfd->shm_size;
1109 if (shmfd->shm_path != NULL) {
1110 if (shmfd->shm_path != NULL) {
1111 path = shmfd->shm_path;
1112 pr_path = curthread->td_ucred->cr_prison->pr_path;
1113 if (strcmp(pr_path, "/") != 0) {
1114 /* Return the jail-rooted pathname. */
1115 pr_pathlen = strlen(pr_path);
1116 visible = strncmp(path, pr_path, pr_pathlen)
1117 == 0 && path[pr_pathlen] == '/';
1118 if (list && !visible)
1123 strlcpy(kif->kf_path, path, sizeof(kif->kf_path));
1130 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1131 struct filedesc *fdp __unused)
1135 sx_slock(&shm_dict_lock);
1136 res = shm_fill_kinfo_locked(fp->f_data, kif, false);
1137 sx_sunlock(&shm_dict_lock);
1142 sysctl_posix_shm_list(SYSCTL_HANDLER_ARGS)
1144 struct shm_mapping *shmm;
1146 struct kinfo_file kif;
1151 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file) * 5, req);
1152 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
1155 sx_slock(&shm_dict_lock);
1156 for (i = 0; i < shm_hash + 1; i++) {
1157 LIST_FOREACH(shmm, &shm_dictionary[i], sm_link) {
1158 error = shm_fill_kinfo_locked(shmm->sm_shmfd,
1165 if (req->oldptr != NULL &&
1166 kif.kf_structsize + curlen > req->oldlen)
1168 error = sbuf_bcat(&sb, &kif, kif.kf_structsize) == 0 ?
1172 curlen += kif.kf_structsize;
1175 sx_sunlock(&shm_dict_lock);
1176 error2 = sbuf_finish(&sb);
1178 return (error != 0 ? error : error2);
1181 SYSCTL_PROC(_kern_ipc, OID_AUTO, posix_shm_list,
1182 CTLFLAG_RD | CTLFLAG_MPSAFE | CTLTYPE_OPAQUE,
1183 NULL, 0, sysctl_posix_shm_list, "",