2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1999 Poul-Henning Kamp.
5 * Copyright (c) 2008 Bjoern A. Zeeb.
6 * Copyright (c) 2009 James Gritton.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
52 #include <sys/mutex.h>
53 #include <sys/racct.h>
55 #include <sys/refcount.h>
57 #include <sys/sysent.h>
58 #include <sys/namei.h>
59 #include <sys/mount.h>
60 #include <sys/queue.h>
61 #include <sys/socket.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
69 #include <netinet/in.h>
75 #include <security/mac/mac_framework.h>
77 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000"
79 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
80 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
82 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
85 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
87 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
91 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
93 #define _PR_IP_SADDRSEL 0
97 /* prison0 describes what is "real" about the system. */
98 struct prison prison0 = {
104 .pr_securelevel = -1,
106 .pr_childmax = JAIL_MAX,
107 .pr_hostuuid = DEFAULT_HOSTUUID,
108 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
110 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
112 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
114 .pr_allow = PR_ALLOW_ALL_STATIC,
116 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
123 struct jailsys_flags {
129 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
130 struct sx allprison_lock;
131 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
132 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
133 LIST_HEAD(, prison_racct) allprison_racct;
136 static int do_jail_attach(struct thread *td, struct prison *pr);
137 static void prison_complete(void *context, int pending);
138 static void prison_deref(struct prison *pr, int flags);
139 static char *prison_path(struct prison *pr1, struct prison *pr2);
140 static void prison_remove_one(struct prison *pr);
142 static void prison_racct_attach(struct prison *pr);
143 static void prison_racct_modify(struct prison *pr);
144 static void prison_racct_detach(struct prison *pr);
147 /* Flags for prison_deref */
148 #define PD_DEREF 0x01
149 #define PD_DEUREF 0x02
150 #define PD_LOCKED 0x04
151 #define PD_LIST_SLOCKED 0x08
152 #define PD_LIST_XLOCKED 0x10
155 * Parameter names corresponding to PR_* flag values. Size values are for kvm
156 * as we cannot figure out the size of a sparse array, or an array without a
159 static struct bool_flags pr_flag_bool[] = {
160 {"persist", "nopersist", PR_PERSIST},
162 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
165 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
168 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
170 static struct jailsys_flags pr_flag_jailsys[] = {
171 {"host", 0, PR_HOST},
173 {"vnet", 0, PR_VNET},
176 {"ip4", PR_IP4_USER, PR_IP4_USER},
179 {"ip6", PR_IP6_USER, PR_IP6_USER},
182 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
184 /* Make this array full-size so dynamic parameters can be added. */
185 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
186 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
187 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
188 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
189 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
190 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
191 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
192 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
193 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
194 {"allow.reserved_ports", "allow.noreserved_ports",
195 PR_ALLOW_RESERVED_PORTS},
197 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
199 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | PR_ALLOW_RESERVED_PORTS)
200 #define JAIL_DEFAULT_ENFORCE_STATFS 2
201 #define JAIL_DEFAULT_DEVFS_RSNUM 0
202 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
203 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
204 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
205 #if defined(INET) || defined(INET6)
206 static unsigned jail_max_af_ips = 255;
210 * Initialize the parts of prison0 that can't be static-initialized with
211 * constants. This is called from proc0_init() after creating thread0 cpuset.
217 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
218 prison0.pr_osreldate = osreldate;
219 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
228 sys_jail(struct thread *td, struct jail_args *uap)
234 error = copyin(uap->jail, &version, sizeof(uint32_t));
243 /* FreeBSD single IPv4 jails. */
244 bzero(&j, sizeof(struct jail));
245 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
248 j.version = j0.version;
250 j.hostname = j0.hostname;
251 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
257 * Version 1 was used by multi-IPv4 jail implementations
258 * that never made it into the official kernel.
262 case 2: /* JAIL_API_VERSION */
263 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
264 error = copyin(uap->jail, &j, sizeof(struct jail));
270 /* Sci-Fi jails are not supported, sorry. */
273 return (kern_jail(td, &j));
277 kern_jail(struct thread *td, struct jail *j)
279 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
288 char *u_path, *u_hostname, *u_name;
289 struct bool_flags *bf;
292 struct in_addr *u_ip4;
295 struct in6_addr *u_ip6;
298 int error, enforce_statfs;
300 bzero(&optiov, sizeof(optiov));
301 opt.uio_iov = optiov;
305 opt.uio_segflg = UIO_SYSSPACE;
306 opt.uio_rw = UIO_READ;
309 /* Set permissions for top-level jails from sysctls. */
310 if (!jailed(td->td_ucred)) {
311 for (bf = pr_flag_allow;
312 bf < pr_flag_allow + nitems(pr_flag_allow) &&
315 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
316 (jail_default_allow & bf->flag)
317 ? bf->name : bf->noname);
318 optiov[opt.uio_iovcnt].iov_len =
319 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
322 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
323 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
325 enforce_statfs = jail_default_enforce_statfs;
326 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
327 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
331 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
333 ip4s = (j->version == 0) ? 1 : j->ip4s;
334 if (ip4s > jail_max_af_ips)
336 tmplen += ip4s * sizeof(struct in_addr);
342 if (j->ip6s > jail_max_af_ips)
344 tmplen += j->ip6s * sizeof(struct in6_addr);
349 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
350 u_hostname = u_path + MAXPATHLEN;
351 u_name = u_hostname + MAXHOSTNAMELEN;
353 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
357 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
359 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
362 optiov[opt.uio_iovcnt].iov_base = "path";
363 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
365 optiov[opt.uio_iovcnt].iov_base = u_path;
366 error = copyinstr(j->path, u_path, MAXPATHLEN,
367 &optiov[opt.uio_iovcnt].iov_len);
369 free(u_path, M_TEMP);
373 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
374 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
376 optiov[opt.uio_iovcnt].iov_base = u_hostname;
377 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
378 &optiov[opt.uio_iovcnt].iov_len);
380 free(u_path, M_TEMP);
384 if (j->jailname != NULL) {
385 optiov[opt.uio_iovcnt].iov_base = "name";
386 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
388 optiov[opt.uio_iovcnt].iov_base = u_name;
389 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
390 &optiov[opt.uio_iovcnt].iov_len);
392 free(u_path, M_TEMP);
398 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
399 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
401 optiov[opt.uio_iovcnt].iov_base = u_ip4;
402 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
404 u_ip4->s_addr = j->ip4s;
406 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
408 free(u_path, M_TEMP);
415 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
416 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
418 optiov[opt.uio_iovcnt].iov_base = u_ip6;
419 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
420 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
422 free(u_path, M_TEMP);
427 KASSERT(opt.uio_iovcnt <= nitems(optiov),
428 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
429 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
430 free(u_path, M_TEMP);
436 * struct jail_set_args {
437 * struct iovec *iovp;
438 * unsigned int iovcnt;
443 sys_jail_set(struct thread *td, struct jail_set_args *uap)
448 /* Check that we have an even number of iovecs. */
452 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
455 error = kern_jail_set(td, auio, uap->flags);
461 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
468 struct in6_addr *ip6;
471 struct vfsoptlist *opts;
472 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
474 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
475 char *g_path, *osrelstr;
476 struct bool_flags *bf;
477 struct jailsys_flags *jsf;
478 #if defined(INET) || defined(INET6)
483 size_t namelen, onamelen, pnamelen;
484 int born, created, cuflags, descend, enforce;
485 int error, errmsg_len, errmsg_pos;
486 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
487 int jid, jsys, len, level;
488 int childmax, osreldt, rsnum, slevel;
489 int fullpath_disabled;
490 #if defined(INET) || defined(INET6)
499 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
503 error = priv_check(td, PRIV_JAIL_SET);
504 if (!error && (flags & JAIL_ATTACH))
505 error = priv_check(td, PRIV_JAIL_ATTACH);
508 mypr = td->td_ucred->cr_prison;
509 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
511 if (flags & ~JAIL_SET_MASK)
515 * Check all the parameters before committing to anything. Not all
516 * errors can be caught early, but we may as well try. Also, this
517 * takes care of some expensive stuff (path lookup) before getting
518 * the allprison lock.
520 * XXX Jails are not filesystems, and jail parameters are not mount
521 * options. But it makes more sense to re-use the vfsopt code
522 * than duplicate it under a different name.
524 error = vfs_buildopts(optuio, &opts);
535 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
538 vfs_opterror(opts, "no valid operation (create or update)");
542 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
548 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
557 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
565 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
570 else if (enforce < 0 || enforce > 2) {
576 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
584 pr_flags = ch_flags = 0;
585 for (bf = pr_flag_bool;
586 bf < pr_flag_bool + nitems(pr_flag_bool);
588 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
589 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
591 ch_flags |= pr_flags;
592 for (jsf = pr_flag_jailsys;
593 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
595 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
601 case JAIL_SYS_DISABLE:
606 pr_flags |= jsf->disable;
609 pr_flags |= jsf->new;
611 case JAIL_SYS_INHERIT:
617 ch_flags |= jsf->new | jsf->disable;
619 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
620 && !(pr_flags & PR_PERSIST)) {
622 vfs_opterror(opts, "new jail must persist or attach");
626 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
628 vfs_opterror(opts, "vnet cannot be changed after creation");
633 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
635 vfs_opterror(opts, "ip4 cannot be changed after creation");
640 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
642 vfs_opterror(opts, "ip6 cannot be changed after creation");
647 pr_allow = ch_allow = 0;
648 for (bf = pr_flag_allow;
649 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
651 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
652 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
654 ch_allow |= pr_allow;
656 error = vfs_getopt(opts, "name", (void **)&name, &len);
662 if (len == 0 || name[len - 1] != '\0') {
666 if (len > MAXHOSTNAMELEN) {
667 error = ENAMETOOLONG;
672 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
680 if (len == 0 || host[len - 1] != '\0') {
684 if (len > MAXHOSTNAMELEN) {
685 error = ENAMETOOLONG;
690 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
698 if (len == 0 || domain[len - 1] != '\0') {
702 if (len > MAXHOSTNAMELEN) {
703 error = ENAMETOOLONG;
708 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
716 if (len == 0 || uuid[len - 1] != '\0') {
720 if (len > HOSTUUIDLEN) {
721 error = ENAMETOOLONG;
726 #ifdef COMPAT_FREEBSD32
727 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
730 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
734 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
746 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
751 else if (ip4s & (sizeof(*ip4) - 1)) {
755 ch_flags |= PR_IP4_USER;
756 pr_flags |= PR_IP4_USER;
758 ip4s /= sizeof(*ip4);
759 if (ip4s > jail_max_af_ips) {
761 vfs_opterror(opts, "too many IPv4 addresses");
764 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
765 bcopy(op, ip4, ip4s * sizeof(*ip4));
767 * IP addresses are all sorted but ip[0] to preserve
768 * the primary IP address as given from userland.
769 * This special IP is used for unbound outgoing
770 * connections as well for "loopback" traffic in case
771 * source address selection cannot find any more fitting
772 * address to connect from.
775 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
778 * Check for duplicate addresses and do some simple
779 * zero and broadcast checks. If users give other bogus
780 * addresses it is their problem.
782 * We do not have to care about byte order for these
783 * checks so we will do them in NBO.
785 for (ii = 0; ii < ip4s; ii++) {
786 if (ip4[ii].s_addr == INADDR_ANY ||
787 ip4[ii].s_addr == INADDR_BROADCAST) {
792 (ip4[0].s_addr == ip4[ii+1].s_addr ||
793 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
803 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
808 else if (ip6s & (sizeof(*ip6) - 1)) {
812 ch_flags |= PR_IP6_USER;
813 pr_flags |= PR_IP6_USER;
815 ip6s /= sizeof(*ip6);
816 if (ip6s > jail_max_af_ips) {
818 vfs_opterror(opts, "too many IPv6 addresses");
821 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
822 bcopy(op, ip6, ip6s * sizeof(*ip6));
824 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
826 for (ii = 0; ii < ip6s; ii++) {
827 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
832 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
833 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
843 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
844 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
847 "vnet jails cannot have IP address restrictions");
852 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
858 if (flags & JAIL_UPDATE) {
861 "osrelease cannot be changed after creation");
864 if (len == 0 || len >= OSRELEASELEN) {
867 "osrelease string must be 1-%d bytes long",
873 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
879 if (flags & JAIL_UPDATE) {
882 "osreldate cannot be changed after creation");
887 vfs_opterror(opts, "osreldate cannot be 0");
892 fullpath_disabled = 0;
894 error = vfs_getopt(opts, "path", (void **)&path, &len);
900 if (flags & JAIL_UPDATE) {
903 "path cannot be changed after creation");
906 if (len == 0 || path[len - 1] != '\0') {
910 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
916 NDFREE(&nd, NDF_ONLY_PNBUF);
917 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
918 strlcpy(g_path, path, MAXPATHLEN);
919 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
922 else if (error == ENODEV) {
923 /* proceed if sysctl debug.disablefullpath == 1 */
924 fullpath_disabled = 1;
925 if (len < 2 || (len == 2 && path[0] == '/'))
928 /* exit on other errors */
931 if (root->v_type != VDIR) {
937 if (fullpath_disabled) {
938 /* Leave room for a real-root full pathname. */
939 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
940 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
941 error = ENAMETOOLONG;
949 * Find the specified jail, or at least its parent.
950 * This abuses the file error codes ENOENT and EEXIST.
954 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
955 namelc = strrchr(name, '.');
956 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
960 sx_xlock(&allprison_lock);
963 * See if a requested jid already exists. There is an
964 * information leak here if the jid exists but is not within
965 * the caller's jail hierarchy. Jail creators will get EEXIST
966 * even though they cannot see the jail, and CREATE | UPDATE
967 * will return ENOENT which is not normally a valid error.
971 vfs_opterror(opts, "negative jid");
972 goto done_unlock_list;
974 pr = prison_find(jid);
977 /* Create: jid must not exist. */
978 if (cuflags == JAIL_CREATE) {
979 mtx_unlock(&pr->pr_mtx);
981 vfs_opterror(opts, "jail %d already exists",
983 goto done_unlock_list;
985 if (!prison_ischild(mypr, pr)) {
986 mtx_unlock(&pr->pr_mtx);
988 } else if (pr->pr_uref == 0) {
989 if (!(flags & JAIL_DYING)) {
990 mtx_unlock(&pr->pr_mtx);
992 vfs_opterror(opts, "jail %d is dying",
994 goto done_unlock_list;
995 } else if ((flags & JAIL_ATTACH) ||
996 (pr_flags & PR_PERSIST)) {
998 * A dying jail might be resurrected
999 * (via attach or persist), but first
1000 * it must determine if another jail
1001 * has claimed its name. Accomplish
1002 * this by implicitly re-setting the
1006 name = prison_name(mypr, pr);
1011 /* Update: jid must exist. */
1012 if (cuflags == JAIL_UPDATE) {
1014 vfs_opterror(opts, "jail %d not found", jid);
1015 goto done_unlock_list;
1020 * If the caller provided a name, look for a jail by that name.
1021 * This has different semantics for creates and updates keyed by jid
1022 * (where the name must not already exist in a different jail),
1023 * and updates keyed by the name itself (where the name must exist
1024 * because that is the jail being updated).
1028 namelc = strrchr(name, '.');
1033 * This is a hierarchical name. Split it into the
1034 * parent and child names, and make sure the parent
1035 * exists or matches an already found jail.
1038 if (strncmp(name, ppr->pr_name, namelc - name)
1039 || ppr->pr_name[namelc - name] != '\0') {
1040 mtx_unlock(&pr->pr_mtx);
1043 "cannot change jail's parent");
1044 goto done_unlock_list;
1048 ppr = prison_find_name(mypr, name);
1052 "jail \"%s\" not found", name);
1053 goto done_unlock_list;
1055 mtx_unlock(&ppr->pr_mtx);
1060 if (namelc[0] != '\0') {
1062 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1065 FOREACH_PRISON_CHILD(ppr, tpr) {
1066 if (tpr != pr && tpr->pr_ref > 0 &&
1067 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1069 cuflags != JAIL_CREATE) {
1070 mtx_lock(&tpr->pr_mtx);
1071 if (tpr->pr_ref > 0) {
1076 if (tpr->pr_uref > 0) {
1082 mtx_unlock(&tpr->pr_mtx);
1083 } else if (tpr->pr_uref > 0) {
1085 * Create, or update(jid):
1086 * name must not exist in an
1087 * active sibling jail.
1091 mtx_unlock(&pr->pr_mtx);
1093 "jail \"%s\" already exists",
1095 goto done_unlock_list;
1099 /* If no active jail is found, use a dying one. */
1100 if (deadpr != NULL && pr == NULL) {
1101 if (flags & JAIL_DYING) {
1102 mtx_lock(&deadpr->pr_mtx);
1103 if (deadpr->pr_ref == 0) {
1104 mtx_unlock(&deadpr->pr_mtx);
1108 } else if (cuflags == JAIL_UPDATE) {
1111 "jail \"%s\" is dying", name);
1112 goto done_unlock_list;
1115 /* Update: name must exist if no jid. */
1116 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1118 vfs_opterror(opts, "jail \"%s\" not found",
1120 goto done_unlock_list;
1124 /* Update: must provide a jid or name. */
1125 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1127 vfs_opterror(opts, "update specified no jail");
1128 goto done_unlock_list;
1131 /* If there's no prison to update, create a new one and link it in. */
1133 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1134 if (tpr->pr_childcount >= tpr->pr_childmax) {
1136 vfs_opterror(opts, "prison limit exceeded");
1137 goto done_unlock_list;
1140 mtx_lock(&ppr->pr_mtx);
1141 if (ppr->pr_ref == 0) {
1142 mtx_unlock(&ppr->pr_mtx);
1144 vfs_opterror(opts, "jail \"%s\" not found",
1145 prison_name(mypr, ppr));
1146 goto done_unlock_list;
1150 mtx_unlock(&ppr->pr_mtx);
1151 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1153 /* Find the next free jid. */
1156 if (jid == JAIL_MAX)
1158 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1159 if (tpr->pr_id < jid)
1161 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1162 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1165 if (jid == lastprid) {
1168 "no available jail IDs");
1170 prison_deref(ppr, PD_DEREF |
1171 PD_DEUREF | PD_LIST_XLOCKED);
1180 * The jail already has a jid (that did not yet exist),
1181 * so just find where to insert it.
1183 TAILQ_FOREACH(tpr, &allprison, pr_list)
1184 if (tpr->pr_id >= jid) {
1185 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1190 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1191 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1192 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1193 tpr->pr_childcount++;
1195 pr->pr_parent = ppr;
1198 /* Set some default values, and inherit some from the parent. */
1203 root = mypr->pr_root;
1206 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1207 pr->pr_flags |= PR_HOST;
1208 #if defined(INET) || defined(INET6)
1210 if (!(pr_flags & PR_VNET))
1214 if (!(ch_flags & PR_IP4_USER))
1215 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1216 else if (!(pr_flags & PR_IP4_USER)) {
1217 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1218 if (ppr->pr_ip4 != NULL) {
1219 pr->pr_ip4s = ppr->pr_ip4s;
1220 pr->pr_ip4 = malloc(pr->pr_ip4s *
1221 sizeof(struct in_addr), M_PRISON,
1223 bcopy(ppr->pr_ip4, pr->pr_ip4,
1224 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1229 if (!(ch_flags & PR_IP6_USER))
1230 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1231 else if (!(pr_flags & PR_IP6_USER)) {
1232 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1233 if (ppr->pr_ip6 != NULL) {
1234 pr->pr_ip6s = ppr->pr_ip6s;
1235 pr->pr_ip6 = malloc(pr->pr_ip6s *
1236 sizeof(struct in6_addr), M_PRISON,
1238 bcopy(ppr->pr_ip6, pr->pr_ip6,
1239 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1245 /* Source address selection is always on by default. */
1246 pr->pr_flags |= _PR_IP_SADDRSEL;
1248 pr->pr_securelevel = ppr->pr_securelevel;
1249 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1250 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1251 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1253 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1254 if (osrelstr == NULL)
1255 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1257 strcpy(pr->pr_osrelease, osrelstr);
1259 LIST_INIT(&pr->pr_children);
1260 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1261 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1264 /* Allocate a new vnet if specified. */
1265 pr->pr_vnet = (pr_flags & PR_VNET)
1266 ? vnet_alloc() : ppr->pr_vnet;
1269 * Allocate a dedicated cpuset for each jail.
1270 * Unlike other initial settings, this may return an erorr.
1272 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1274 prison_deref(pr, PD_LIST_XLOCKED);
1278 mtx_lock(&pr->pr_mtx);
1280 * New prisons do not yet have a reference, because we do not
1281 * want others to see the incomplete prison once the
1282 * allprison_lock is downgraded.
1287 * Grab a reference for existing prisons, to ensure they
1288 * continue to exist for the duration of the call.
1291 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1292 if ((pr->pr_flags & PR_VNET) &&
1293 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1296 "vnet jails cannot have IP address restrictions");
1297 goto done_deref_locked;
1301 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1304 "ip4 cannot be changed after creation");
1305 goto done_deref_locked;
1309 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1312 "ip6 cannot be changed after creation");
1313 goto done_deref_locked;
1318 /* Do final error checking before setting anything. */
1320 if (slevel < ppr->pr_securelevel) {
1322 goto done_deref_locked;
1326 if (childmax >= ppr->pr_childmax) {
1328 goto done_deref_locked;
1332 if (enforce < ppr->pr_enforce_statfs) {
1334 goto done_deref_locked;
1339 * devfs_rsnum is a uint16_t
1341 if (rsnum < 0 || rsnum > 65535) {
1343 goto done_deref_locked;
1346 * Nested jails always inherit parent's devfs ruleset
1348 if (jailed(td->td_ucred)) {
1349 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1351 goto done_deref_locked;
1353 rsnum = ppr->pr_devfs_rsnum;
1358 if (ppr->pr_flags & PR_IP4) {
1360 * Make sure the new set of IP addresses is a
1361 * subset of the parent's list. Don't worry
1362 * about the parent being unlocked, as any
1363 * setting is done with allprison_lock held.
1365 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1366 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1368 if (ij == ppr->pr_ip4s) {
1370 goto done_deref_locked;
1373 for (ii = ij = 1; ii < ip4s; ii++) {
1374 if (ip4[ii].s_addr ==
1375 ppr->pr_ip4[0].s_addr)
1377 for (; ij < ppr->pr_ip4s; ij++)
1378 if (ip4[ii].s_addr ==
1379 ppr->pr_ip4[ij].s_addr)
1381 if (ij == ppr->pr_ip4s)
1384 if (ij == ppr->pr_ip4s) {
1386 goto done_deref_locked;
1391 * Check for conflicting IP addresses. We permit them
1392 * if there is no more than one IP on each jail. If
1393 * there is a duplicate on a jail with more than one
1394 * IP stop checking and return error.
1398 for (; tppr != &prison0; tppr = tppr->pr_parent)
1399 if (tppr->pr_flags & PR_VNET)
1402 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1405 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1407 tpr->pr_uref == 0) {
1411 if (!(tpr->pr_flags & PR_IP4_USER))
1414 if (tpr->pr_ip4 == NULL ||
1415 (ip4s == 1 && tpr->pr_ip4s == 1))
1417 for (ii = 0; ii < ip4s; ii++) {
1418 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1422 "IPv4 addresses clash");
1423 goto done_deref_locked;
1431 if (ppr->pr_flags & PR_IP6) {
1433 * Make sure the new set of IP addresses is a
1434 * subset of the parent's list.
1436 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1437 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1440 if (ij == ppr->pr_ip6s) {
1442 goto done_deref_locked;
1445 for (ii = ij = 1; ii < ip6s; ii++) {
1446 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1449 for (; ij < ppr->pr_ip6s; ij++)
1450 if (IN6_ARE_ADDR_EQUAL(
1451 &ip6[ii], &ppr->pr_ip6[ij]))
1453 if (ij == ppr->pr_ip6s)
1456 if (ij == ppr->pr_ip6s) {
1458 goto done_deref_locked;
1462 /* Check for conflicting IP addresses. */
1465 for (; tppr != &prison0; tppr = tppr->pr_parent)
1466 if (tppr->pr_flags & PR_VNET)
1469 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1472 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1474 tpr->pr_uref == 0) {
1478 if (!(tpr->pr_flags & PR_IP6_USER))
1481 if (tpr->pr_ip6 == NULL ||
1482 (ip6s == 1 && tpr->pr_ip6s == 1))
1484 for (ii = 0; ii < ip6s; ii++) {
1485 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1489 "IPv6 addresses clash");
1490 goto done_deref_locked;
1496 onamelen = namelen = 0;
1497 if (namelc != NULL) {
1498 /* Give a default name of the jid. Also allow the name to be
1499 * explicitly the jid - but not any other number, and only in
1500 * normal form (no leading zero/etc).
1502 if (namelc[0] == '\0')
1503 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1504 else if ((strtoul(namelc, &p, 10) != jid ||
1505 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1508 "name cannot be numeric (unless it is the jid)");
1509 goto done_deref_locked;
1512 * Make sure the name isn't too long for the prison or its
1515 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1516 onamelen = strlen(pr->pr_name + pnamelen);
1517 namelen = strlen(namelc);
1518 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1519 error = ENAMETOOLONG;
1520 goto done_deref_locked;
1522 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1523 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1524 sizeof(pr->pr_name)) {
1525 error = ENAMETOOLONG;
1526 goto done_deref_locked;
1530 if (pr_allow & ~ppr->pr_allow) {
1532 goto done_deref_locked;
1536 * Let modules check their parameters. This requires unlocking and
1537 * then re-locking the prison, but this is still a valid state as long
1538 * as allprison_lock remains xlocked.
1540 mtx_unlock(&pr->pr_mtx);
1541 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1543 prison_deref(pr, created
1545 : PD_DEREF | PD_LIST_XLOCKED);
1548 mtx_lock(&pr->pr_mtx);
1550 /* At this point, all valid parameters should have been noted. */
1551 TAILQ_FOREACH(opt, opts, link) {
1552 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1554 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1555 goto done_deref_locked;
1559 /* Set the parameters of the prison. */
1562 if (pr_flags & PR_IP4_USER) {
1563 pr->pr_flags |= PR_IP4;
1564 free(pr->pr_ip4, M_PRISON);
1568 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1570 if (tpr->pr_flags & PR_VNET) {
1575 if (prison_restrict_ip4(tpr, NULL)) {
1584 if (pr_flags & PR_IP6_USER) {
1585 pr->pr_flags |= PR_IP6;
1586 free(pr->pr_ip6, M_PRISON);
1590 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1592 if (tpr->pr_flags & PR_VNET) {
1597 if (prison_restrict_ip6(tpr, NULL)) {
1605 pr->pr_securelevel = slevel;
1606 /* Set all child jails to be at least this level. */
1607 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1608 if (tpr->pr_securelevel < slevel)
1609 tpr->pr_securelevel = slevel;
1612 pr->pr_childmax = childmax;
1613 /* Set all child jails to under this limit. */
1614 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1615 if (tpr->pr_childmax > childmax - level)
1616 tpr->pr_childmax = childmax > level
1617 ? childmax - level : 0;
1620 pr->pr_enforce_statfs = enforce;
1621 /* Pass this restriction on to the children. */
1622 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1623 if (tpr->pr_enforce_statfs < enforce)
1624 tpr->pr_enforce_statfs = enforce;
1627 pr->pr_devfs_rsnum = rsnum;
1628 /* Pass this restriction on to the children. */
1629 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1630 tpr->pr_devfs_rsnum = rsnum;
1632 if (namelc != NULL) {
1633 if (ppr == &prison0)
1634 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1636 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1637 ppr->pr_name, namelc);
1638 /* Change this component of child names. */
1639 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1640 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1641 strlen(tpr->pr_name + onamelen) + 1);
1642 bcopy(pr->pr_name, tpr->pr_name, namelen);
1646 /* Try to keep a real-rooted full pathname. */
1647 if (fullpath_disabled && path[0] == '/' &&
1648 strcmp(mypr->pr_path, "/"))
1649 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1650 mypr->pr_path, path);
1652 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1655 if (PR_HOST & ch_flags & ~pr_flags) {
1656 if (pr->pr_flags & PR_HOST) {
1658 * Copy the parent's host info. As with pr_ip4 above,
1659 * the lack of a lock on the parent is not a problem;
1660 * it is always set with allprison_lock at least
1661 * shared, and is held exclusively here.
1663 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1664 sizeof(pr->pr_hostname));
1665 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1666 sizeof(pr->pr_domainname));
1667 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1668 sizeof(pr->pr_hostuuid));
1669 pr->pr_hostid = pr->pr_parent->pr_hostid;
1671 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1672 /* Set this prison, and any descendants without PR_HOST. */
1674 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1676 strlcpy(pr->pr_domainname, domain,
1677 sizeof(pr->pr_domainname));
1679 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1681 pr->pr_hostid = hid;
1682 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1683 if (tpr->pr_flags & PR_HOST)
1687 strlcpy(tpr->pr_hostname,
1689 sizeof(tpr->pr_hostname));
1691 strlcpy(tpr->pr_domainname,
1693 sizeof(tpr->pr_domainname));
1695 strlcpy(tpr->pr_hostuuid,
1697 sizeof(tpr->pr_hostuuid));
1699 tpr->pr_hostid = hid;
1703 if ((tallow = ch_allow & ~pr_allow)) {
1704 /* Clear allow bits in all children. */
1705 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1706 tpr->pr_allow &= ~tallow;
1708 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1710 * Persistent prisons get an extra reference, and prisons losing their
1711 * persist flag lose that reference. Only do this for existing prisons
1712 * for now, so new ones will remain unseen until after the module
1713 * handlers have completed.
1715 born = pr->pr_uref == 0;
1716 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1717 if (pr_flags & PR_PERSIST) {
1725 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1726 mtx_unlock(&pr->pr_mtx);
1729 if (racct_enable && created)
1730 prison_racct_attach(pr);
1733 /* Locks may have prevented a complete restriction of child IP
1734 * addresses. If so, allocate some more memory and try again.
1739 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1740 mtx_lock(&pr->pr_mtx);
1742 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1744 if (tpr->pr_flags & PR_VNET) {
1749 if (prison_restrict_ip4(tpr, ip4)) {
1756 mtx_unlock(&pr->pr_mtx);
1762 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1763 mtx_lock(&pr->pr_mtx);
1765 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1767 if (tpr->pr_flags & PR_VNET) {
1772 if (prison_restrict_ip6(tpr, ip6)) {
1779 mtx_unlock(&pr->pr_mtx);
1783 /* Let the modules do their work. */
1784 sx_downgrade(&allprison_lock);
1786 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1788 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1789 prison_deref(pr, created
1791 : PD_DEREF | PD_LIST_SLOCKED);
1795 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1798 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1799 prison_deref(pr, created
1801 : PD_DEREF | PD_LIST_SLOCKED);
1805 /* Attach this process to the prison if requested. */
1806 if (flags & JAIL_ATTACH) {
1807 mtx_lock(&pr->pr_mtx);
1808 error = do_jail_attach(td, pr);
1810 vfs_opterror(opts, "attach failed");
1812 prison_deref(pr, PD_DEREF);
1818 if (racct_enable && !created) {
1819 if (!(flags & JAIL_ATTACH))
1820 sx_sunlock(&allprison_lock);
1821 prison_racct_modify(pr);
1822 if (!(flags & JAIL_ATTACH))
1823 sx_slock(&allprison_lock);
1827 td->td_retval[0] = pr->pr_id;
1830 * Now that it is all there, drop the temporary reference from existing
1831 * prisons. Or add a reference to newly created persistent prisons
1832 * (which was not done earlier so that the prison would not be publicly
1836 prison_deref(pr, (flags & JAIL_ATTACH)
1838 : PD_DEREF | PD_LIST_SLOCKED);
1840 if (pr_flags & PR_PERSIST) {
1841 mtx_lock(&pr->pr_mtx);
1844 mtx_unlock(&pr->pr_mtx);
1846 if (!(flags & JAIL_ATTACH))
1847 sx_sunlock(&allprison_lock);
1853 prison_deref(pr, created
1854 ? PD_LOCKED | PD_LIST_XLOCKED
1855 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1858 sx_xunlock(&allprison_lock);
1864 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1865 &errmsg_len) == 0 && errmsg_len > 0) {
1866 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1867 if (optuio->uio_segflg == UIO_SYSSPACE)
1869 optuio->uio_iov[errmsg_pos].iov_base,
1873 optuio->uio_iov[errmsg_pos].iov_base,
1879 free(ip4, M_PRISON);
1882 free(ip6, M_PRISON);
1885 free(g_path, M_TEMP);
1892 * struct jail_get_args {
1893 * struct iovec *iovp;
1894 * unsigned int iovcnt;
1899 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1904 /* Check that we have an even number of iovecs. */
1905 if (uap->iovcnt & 1)
1908 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1911 error = kern_jail_get(td, auio, uap->flags);
1913 error = copyout(auio->uio_iov, uap->iovp,
1914 uap->iovcnt * sizeof (struct iovec));
1920 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1922 struct bool_flags *bf;
1923 struct jailsys_flags *jsf;
1924 struct prison *pr, *mypr;
1926 struct vfsoptlist *opts;
1927 char *errmsg, *name;
1928 int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1931 if (flags & ~JAIL_GET_MASK)
1934 /* Get the parameter list. */
1935 error = vfs_buildopts(optuio, &opts);
1938 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1939 mypr = td->td_ucred->cr_prison;
1942 * Find the prison specified by one of: lastjid, jid, name.
1944 sx_slock(&allprison_lock);
1945 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1947 TAILQ_FOREACH(pr, &allprison, pr_list) {
1948 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1949 mtx_lock(&pr->pr_mtx);
1950 if (pr->pr_ref > 0 &&
1951 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1953 mtx_unlock(&pr->pr_mtx);
1959 vfs_opterror(opts, "no jail after %d", jid);
1960 goto done_unlock_list;
1961 } else if (error != ENOENT)
1962 goto done_unlock_list;
1964 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1967 pr = prison_find_child(mypr, jid);
1969 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1970 mtx_unlock(&pr->pr_mtx);
1972 vfs_opterror(opts, "jail %d is dying",
1974 goto done_unlock_list;
1979 vfs_opterror(opts, "jail %d not found", jid);
1980 goto done_unlock_list;
1982 } else if (error != ENOENT)
1983 goto done_unlock_list;
1985 error = vfs_getopt(opts, "name", (void **)&name, &len);
1987 if (len == 0 || name[len - 1] != '\0') {
1989 goto done_unlock_list;
1991 pr = prison_find_name(mypr, name);
1993 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1994 mtx_unlock(&pr->pr_mtx);
1996 vfs_opterror(opts, "jail \"%s\" is dying",
1998 goto done_unlock_list;
2003 vfs_opterror(opts, "jail \"%s\" not found", name);
2004 goto done_unlock_list;
2005 } else if (error != ENOENT)
2006 goto done_unlock_list;
2008 vfs_opterror(opts, "no jail specified");
2010 goto done_unlock_list;
2013 /* Get the parameters of the prison. */
2016 td->td_retval[0] = pr->pr_id;
2017 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2018 if (error != 0 && error != ENOENT)
2020 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2021 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2022 if (error != 0 && error != ENOENT)
2024 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2025 if (error != 0 && error != ENOENT)
2027 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2028 sizeof(pr->pr_cpuset->cs_id));
2029 if (error != 0 && error != ENOENT)
2031 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2032 if (error != 0 && error != ENOENT)
2035 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2036 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2037 if (error != 0 && error != ENOENT)
2041 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2042 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2043 if (error != 0 && error != ENOENT)
2046 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2047 sizeof(pr->pr_securelevel));
2048 if (error != 0 && error != ENOENT)
2050 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2051 sizeof(pr->pr_childcount));
2052 if (error != 0 && error != ENOENT)
2054 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2055 sizeof(pr->pr_childmax));
2056 if (error != 0 && error != ENOENT)
2058 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2059 if (error != 0 && error != ENOENT)
2061 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2062 if (error != 0 && error != ENOENT)
2064 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2065 if (error != 0 && error != ENOENT)
2067 #ifdef COMPAT_FREEBSD32
2068 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2069 uint32_t hid32 = pr->pr_hostid;
2071 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2074 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2075 sizeof(pr->pr_hostid));
2076 if (error != 0 && error != ENOENT)
2078 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2079 sizeof(pr->pr_enforce_statfs));
2080 if (error != 0 && error != ENOENT)
2082 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2083 sizeof(pr->pr_devfs_rsnum));
2084 if (error != 0 && error != ENOENT)
2086 for (bf = pr_flag_bool;
2087 bf < pr_flag_bool + nitems(pr_flag_bool);
2089 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2090 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2091 if (error != 0 && error != ENOENT)
2094 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2095 if (error != 0 && error != ENOENT)
2098 for (jsf = pr_flag_jailsys;
2099 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2101 f = pr->pr_flags & (jsf->disable | jsf->new);
2102 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2103 : (f == jsf->new) ? JAIL_SYS_NEW
2105 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2106 if (error != 0 && error != ENOENT)
2109 for (bf = pr_flag_allow;
2110 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2112 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2113 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2114 if (error != 0 && error != ENOENT)
2117 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2118 if (error != 0 && error != ENOENT)
2121 i = (pr->pr_uref == 0);
2122 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2123 if (error != 0 && error != ENOENT)
2126 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2127 if (error != 0 && error != ENOENT)
2129 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2130 sizeof(pr->pr_osreldate));
2131 if (error != 0 && error != ENOENT)
2133 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2134 if (error != 0 && error != ENOENT)
2137 /* Get the module parameters. */
2138 mtx_unlock(&pr->pr_mtx);
2140 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2143 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2145 /* By now, all parameters should have been noted. */
2146 TAILQ_FOREACH(opt, opts, link) {
2147 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2149 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2154 /* Write the fetched parameters back to userspace. */
2156 TAILQ_FOREACH(opt, opts, link) {
2157 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2158 pos = 2 * opt->pos + 1;
2159 optuio->uio_iov[pos].iov_len = opt->len;
2160 if (opt->value != NULL) {
2161 if (optuio->uio_segflg == UIO_SYSSPACE) {
2163 optuio->uio_iov[pos].iov_base,
2166 error = copyout(opt->value,
2167 optuio->uio_iov[pos].iov_base,
2178 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2182 sx_sunlock(&allprison_lock);
2184 if (error && errmsg_pos >= 0) {
2185 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2186 errmsg_pos = 2 * errmsg_pos + 1;
2187 if (errmsg_len > 0) {
2188 if (optuio->uio_segflg == UIO_SYSSPACE)
2190 optuio->uio_iov[errmsg_pos].iov_base,
2194 optuio->uio_iov[errmsg_pos].iov_base,
2204 * struct jail_remove_args {
2209 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2211 struct prison *pr, *cpr, *lpr, *tpr;
2214 error = priv_check(td, PRIV_JAIL_REMOVE);
2218 sx_xlock(&allprison_lock);
2219 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2221 sx_xunlock(&allprison_lock);
2225 /* Remove all descendants of this prison, then remove this prison. */
2227 if (!LIST_EMPTY(&pr->pr_children)) {
2228 mtx_unlock(&pr->pr_mtx);
2230 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2231 mtx_lock(&cpr->pr_mtx);
2232 if (cpr->pr_ref > 0) {
2236 /* Already removed - do not do it again. */
2239 mtx_unlock(&cpr->pr_mtx);
2241 mtx_lock(&lpr->pr_mtx);
2242 prison_remove_one(lpr);
2243 sx_xlock(&allprison_lock);
2248 mtx_lock(&lpr->pr_mtx);
2249 prison_remove_one(lpr);
2250 sx_xlock(&allprison_lock);
2252 mtx_lock(&pr->pr_mtx);
2254 prison_remove_one(pr);
2259 prison_remove_one(struct prison *pr)
2264 /* If the prison was persistent, it is not anymore. */
2266 if (pr->pr_flags & PR_PERSIST) {
2269 pr->pr_flags &= ~PR_PERSIST;
2273 * jail_remove added a reference. If that's the only one, remove
2276 KASSERT(pr->pr_ref > 0,
2277 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2278 if (pr->pr_ref == 1) {
2280 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2284 mtx_unlock(&pr->pr_mtx);
2285 sx_xunlock(&allprison_lock);
2287 * Kill all processes unfortunate enough to be attached to this prison.
2289 sx_slock(&allproc_lock);
2290 FOREACH_PROC_IN_SYSTEM(p) {
2292 if (p->p_state != PRS_NEW && p->p_ucred &&
2293 p->p_ucred->cr_prison == pr)
2294 kern_psignal(p, SIGKILL);
2297 sx_sunlock(&allproc_lock);
2298 /* Remove the temporary reference added by jail_remove. */
2299 prison_deref(pr, deuref | PD_DEREF);
2304 * struct jail_attach_args {
2309 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2314 error = priv_check(td, PRIV_JAIL_ATTACH);
2319 * Start with exclusive hold on allprison_lock to ensure that a possible
2320 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2321 * But then immediately downgrade it since we don't need to stop
2324 sx_xlock(&allprison_lock);
2325 sx_downgrade(&allprison_lock);
2326 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2328 sx_sunlock(&allprison_lock);
2333 * Do not allow a process to attach to a prison that is not
2334 * considered to be "alive".
2336 if (pr->pr_uref == 0) {
2337 mtx_unlock(&pr->pr_mtx);
2338 sx_sunlock(&allprison_lock);
2342 return (do_jail_attach(td, pr));
2346 do_jail_attach(struct thread *td, struct prison *pr)
2349 struct ucred *newcred, *oldcred;
2353 * XXX: Note that there is a slight race here if two threads
2354 * in the same privileged process attempt to attach to two
2355 * different jails at the same time. It is important for
2356 * user processes not to do this, or they might end up with
2357 * a process root from one prison, but attached to the jail
2362 mtx_unlock(&pr->pr_mtx);
2364 /* Let modules do whatever they need to prepare for attaching. */
2365 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2367 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2370 sx_sunlock(&allprison_lock);
2373 * Reparent the newly attached process to this jail.
2376 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2380 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2381 if ((error = change_dir(pr->pr_root, td)) != 0)
2384 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2387 VOP_UNLOCK(pr->pr_root, 0);
2388 if ((error = pwd_chroot(td, pr->pr_root)))
2393 oldcred = crcopysafe(p, newcred);
2394 newcred->cr_prison = pr;
2395 proc_set_cred(p, newcred);
2398 racct_proc_ucred_changed(p, oldcred, newcred);
2403 rctl_proc_ucred_changed(p, newcred);
2406 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2411 VOP_UNLOCK(pr->pr_root, 0);
2413 /* Tell modules this thread is still in its old jail after all. */
2414 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2415 prison_deref(pr, PD_DEREF | PD_DEUREF);
2421 * Returns a locked prison instance, or NULL on failure.
2424 prison_find(int prid)
2428 sx_assert(&allprison_lock, SX_LOCKED);
2429 TAILQ_FOREACH(pr, &allprison, pr_list) {
2430 if (pr->pr_id == prid) {
2431 mtx_lock(&pr->pr_mtx);
2434 mtx_unlock(&pr->pr_mtx);
2441 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2444 prison_find_child(struct prison *mypr, int prid)
2449 sx_assert(&allprison_lock, SX_LOCKED);
2450 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2451 if (pr->pr_id == prid) {
2452 mtx_lock(&pr->pr_mtx);
2455 mtx_unlock(&pr->pr_mtx);
2462 * Look for the name relative to mypr. Returns a locked prison or NULL.
2465 prison_find_name(struct prison *mypr, const char *name)
2467 struct prison *pr, *deadpr;
2471 sx_assert(&allprison_lock, SX_LOCKED);
2472 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2475 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2476 if (!strcmp(pr->pr_name + mylen, name)) {
2477 mtx_lock(&pr->pr_mtx);
2478 if (pr->pr_ref > 0) {
2479 if (pr->pr_uref > 0)
2483 mtx_unlock(&pr->pr_mtx);
2486 /* There was no valid prison - perhaps there was a dying one. */
2487 if (deadpr != NULL) {
2488 mtx_lock(&deadpr->pr_mtx);
2489 if (deadpr->pr_ref == 0) {
2490 mtx_unlock(&deadpr->pr_mtx);
2498 * See if a prison has the specific flag set.
2501 prison_flag(struct ucred *cred, unsigned flag)
2504 /* This is an atomic read, so no locking is necessary. */
2505 return (cred->cr_prison->pr_flags & flag);
2509 prison_allow(struct ucred *cred, unsigned flag)
2512 /* This is an atomic read, so no locking is necessary. */
2513 return (cred->cr_prison->pr_allow & flag);
2517 * Remove a prison reference. If that was the last reference, remove the
2518 * prison itself - but not in this context in case there are locks held.
2521 prison_free_locked(struct prison *pr)
2525 mtx_assert(&pr->pr_mtx, MA_OWNED);
2527 mtx_unlock(&pr->pr_mtx);
2529 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2533 prison_free(struct prison *pr)
2536 mtx_lock(&pr->pr_mtx);
2537 prison_free_locked(pr);
2541 * Complete a call to either prison_free or prison_proc_free.
2544 prison_complete(void *context, int pending)
2546 struct prison *pr = context;
2548 sx_xlock(&allprison_lock);
2549 mtx_lock(&pr->pr_mtx);
2550 prison_deref(pr, pr->pr_uref
2551 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2552 : PD_LOCKED | PD_LIST_XLOCKED);
2556 * Remove a prison reference (usually). This internal version assumes no
2557 * mutexes are held, except perhaps the prison itself. If there are no more
2558 * references, release and delist the prison. On completion, the prison lock
2559 * and the allprison lock are both unlocked.
2562 prison_deref(struct prison *pr, int flags)
2564 struct prison *ppr, *tpr;
2567 if (!(flags & PD_LOCKED))
2568 mtx_lock(&pr->pr_mtx);
2570 if (flags & PD_DEUREF) {
2571 KASSERT(pr->pr_uref > 0,
2572 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2575 lasturef = pr->pr_uref == 0;
2578 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2581 if (flags & PD_DEREF) {
2582 KASSERT(pr->pr_ref > 0,
2583 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2588 mtx_unlock(&pr->pr_mtx);
2591 * Tell the modules if the last user reference was removed
2592 * (even it sticks around in dying state).
2595 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2596 sx_xlock(&allprison_lock);
2597 flags |= PD_LIST_XLOCKED;
2599 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2600 mtx_lock(&pr->pr_mtx);
2602 mtx_unlock(&pr->pr_mtx);
2605 /* If the prison still has references, nothing else to do. */
2607 if (flags & PD_LIST_SLOCKED)
2608 sx_sunlock(&allprison_lock);
2609 else if (flags & PD_LIST_XLOCKED)
2610 sx_xunlock(&allprison_lock);
2614 if (flags & PD_LIST_SLOCKED) {
2615 if (!sx_try_upgrade(&allprison_lock)) {
2616 sx_sunlock(&allprison_lock);
2617 sx_xlock(&allprison_lock);
2619 } else if (!(flags & PD_LIST_XLOCKED))
2620 sx_xlock(&allprison_lock);
2622 TAILQ_REMOVE(&allprison, pr, pr_list);
2623 LIST_REMOVE(pr, pr_sibling);
2624 ppr = pr->pr_parent;
2625 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2626 tpr->pr_childcount--;
2627 sx_xunlock(&allprison_lock);
2630 if (pr->pr_vnet != ppr->pr_vnet)
2631 vnet_destroy(pr->pr_vnet);
2633 if (pr->pr_root != NULL)
2635 mtx_destroy(&pr->pr_mtx);
2637 free(pr->pr_ip4, M_PRISON);
2640 free(pr->pr_ip6, M_PRISON);
2642 if (pr->pr_cpuset != NULL)
2643 cpuset_rel(pr->pr_cpuset);
2647 prison_racct_detach(pr);
2651 /* Removing a prison frees a reference on its parent. */
2653 mtx_lock(&pr->pr_mtx);
2654 flags = PD_DEREF | PD_DEUREF;
2659 prison_hold_locked(struct prison *pr)
2662 mtx_assert(&pr->pr_mtx, MA_OWNED);
2663 KASSERT(pr->pr_ref > 0,
2664 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2669 prison_hold(struct prison *pr)
2672 mtx_lock(&pr->pr_mtx);
2673 prison_hold_locked(pr);
2674 mtx_unlock(&pr->pr_mtx);
2678 prison_proc_hold(struct prison *pr)
2681 mtx_lock(&pr->pr_mtx);
2682 KASSERT(pr->pr_uref > 0,
2683 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2685 mtx_unlock(&pr->pr_mtx);
2689 prison_proc_free(struct prison *pr)
2692 mtx_lock(&pr->pr_mtx);
2693 KASSERT(pr->pr_uref > 0,
2694 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2695 if (pr->pr_uref > 1)
2699 * Don't remove the last user reference in this context, which
2700 * is expected to be a process that is not only locked, but
2704 mtx_unlock(&pr->pr_mtx);
2705 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2708 mtx_unlock(&pr->pr_mtx);
2712 * Check if a jail supports the given address family.
2714 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2718 prison_check_af(struct ucred *cred, int af)
2723 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2725 pr = cred->cr_prison;
2727 /* Prisons with their own network stack are not limited. */
2728 if (prison_owns_vnet(cred))
2737 if (pr->pr_flags & PR_IP4)
2739 mtx_lock(&pr->pr_mtx);
2740 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2741 error = EAFNOSUPPORT;
2742 mtx_unlock(&pr->pr_mtx);
2748 if (pr->pr_flags & PR_IP6)
2750 mtx_lock(&pr->pr_mtx);
2751 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2752 error = EAFNOSUPPORT;
2753 mtx_unlock(&pr->pr_mtx);
2761 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2762 error = EAFNOSUPPORT;
2768 * Check if given address belongs to the jail referenced by cred (wrapper to
2769 * prison_check_ip[46]).
2771 * Returns 0 if jail doesn't restrict the address family or if address belongs
2772 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2773 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
2776 prison_if(struct ucred *cred, struct sockaddr *sa)
2779 struct sockaddr_in *sai;
2782 struct sockaddr_in6 *sai6;
2786 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2787 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2790 if (prison_owns_vnet(cred))
2795 switch (sa->sa_family)
2799 sai = (struct sockaddr_in *)sa;
2800 error = prison_check_ip4(cred, &sai->sin_addr);
2805 sai6 = (struct sockaddr_in6 *)sa;
2806 error = prison_check_ip6(cred, &sai6->sin6_addr);
2810 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2811 error = EAFNOSUPPORT;
2817 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2820 prison_check(struct ucred *cred1, struct ucred *cred2)
2823 return ((cred1->cr_prison == cred2->cr_prison ||
2824 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2828 * Return 1 if p2 is a child of p1, otherwise 0.
2831 prison_ischild(struct prison *pr1, struct prison *pr2)
2834 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2841 * Return 1 if the passed credential is in a jail, otherwise 0.
2844 jailed(struct ucred *cred)
2847 return (cred->cr_prison != &prison0);
2851 * Return 1 if the passed credential is in a jail and that jail does not
2852 * have its own virtual network stack, otherwise 0.
2855 jailed_without_vnet(struct ucred *cred)
2861 if (prison_owns_vnet(cred))
2869 * Return the correct hostname (domainname, et al) for the passed credential.
2872 getcredhostname(struct ucred *cred, char *buf, size_t size)
2877 * A NULL credential can be used to shortcut to the physical
2878 * system's hostname.
2880 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2881 mtx_lock(&pr->pr_mtx);
2882 strlcpy(buf, pr->pr_hostname, size);
2883 mtx_unlock(&pr->pr_mtx);
2887 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2890 mtx_lock(&cred->cr_prison->pr_mtx);
2891 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2892 mtx_unlock(&cred->cr_prison->pr_mtx);
2896 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2899 mtx_lock(&cred->cr_prison->pr_mtx);
2900 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2901 mtx_unlock(&cred->cr_prison->pr_mtx);
2905 getcredhostid(struct ucred *cred, unsigned long *hostid)
2908 mtx_lock(&cred->cr_prison->pr_mtx);
2909 *hostid = cred->cr_prison->pr_hostid;
2910 mtx_unlock(&cred->cr_prison->pr_mtx);
2915 * Determine whether the prison represented by cred owns
2916 * its vnet rather than having it inherited.
2918 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2921 prison_owns_vnet(struct ucred *cred)
2925 * vnets cannot be added/removed after jail creation,
2926 * so no need to lock here.
2928 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2933 * Determine whether the subject represented by cred can "see"
2934 * status of a mount point.
2935 * Returns: 0 for permitted, ENOENT otherwise.
2936 * XXX: This function should be called cr_canseemount() and should be
2937 * placed in kern_prot.c.
2940 prison_canseemount(struct ucred *cred, struct mount *mp)
2946 pr = cred->cr_prison;
2947 if (pr->pr_enforce_statfs == 0)
2949 if (pr->pr_root->v_mount == mp)
2951 if (pr->pr_enforce_statfs == 2)
2954 * If jail's chroot directory is set to "/" we should be able to see
2955 * all mount-points from inside a jail.
2956 * This is ugly check, but this is the only situation when jail's
2957 * directory ends with '/'.
2959 if (strcmp(pr->pr_path, "/") == 0)
2961 len = strlen(pr->pr_path);
2963 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2966 * Be sure that we don't have situation where jail's root directory
2967 * is "/some/path" and mount point is "/some/pathpath".
2969 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2975 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2977 char jpath[MAXPATHLEN];
2981 pr = cred->cr_prison;
2982 if (pr->pr_enforce_statfs == 0)
2984 if (prison_canseemount(cred, mp) != 0) {
2985 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2986 strlcpy(sp->f_mntonname, "[restricted]",
2987 sizeof(sp->f_mntonname));
2990 if (pr->pr_root->v_mount == mp) {
2992 * Clear current buffer data, so we are sure nothing from
2993 * the valid path left there.
2995 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2996 *sp->f_mntonname = '/';
3000 * If jail's chroot directory is set to "/" we should be able to see
3001 * all mount-points from inside a jail.
3003 if (strcmp(pr->pr_path, "/") == 0)
3005 len = strlen(pr->pr_path);
3006 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3008 * Clear current buffer data, so we are sure nothing from
3009 * the valid path left there.
3011 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3012 if (*jpath == '\0') {
3013 /* Should never happen. */
3014 *sp->f_mntonname = '/';
3016 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3021 * Check with permission for a specific privilege is granted within jail. We
3022 * have a specific list of accepted privileges; the rest are denied.
3025 prison_priv_check(struct ucred *cred, int priv)
3033 * Privileges specific to prisons with a virtual network stack.
3034 * There might be a duplicate entry here in case the privilege
3035 * is only granted conditionally in the legacy jail case.
3040 * NFS-specific privileges.
3042 case PRIV_NFS_DAEMON:
3043 case PRIV_NFS_LOCKD:
3046 * Network stack privileges.
3048 case PRIV_NET_BRIDGE:
3051 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3052 case PRIV_NET_ROUTE:
3054 case PRIV_NET_SETIFMTU:
3055 case PRIV_NET_SETIFFLAGS:
3056 case PRIV_NET_SETIFCAP:
3057 case PRIV_NET_SETIFDESCR:
3058 case PRIV_NET_SETIFNAME :
3059 case PRIV_NET_SETIFMETRIC:
3060 case PRIV_NET_SETIFPHYS:
3061 case PRIV_NET_SETIFMAC:
3062 case PRIV_NET_ADDMULTI:
3063 case PRIV_NET_DELMULTI:
3064 case PRIV_NET_HWIOCTL:
3065 case PRIV_NET_SETLLADDR:
3066 case PRIV_NET_ADDIFGROUP:
3067 case PRIV_NET_DELIFGROUP:
3068 case PRIV_NET_IFCREATE:
3069 case PRIV_NET_IFDESTROY:
3070 case PRIV_NET_ADDIFADDR:
3071 case PRIV_NET_DELIFADDR:
3074 case PRIV_NET_SETIFVNET:
3075 case PRIV_NET_SETIFFIB:
3078 * 802.11-related privileges.
3080 case PRIV_NET80211_GETKEY:
3082 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3089 case PRIV_NETATM_CFG:
3090 case PRIV_NETATM_ADD:
3091 case PRIV_NETATM_DEL:
3092 case PRIV_NETATM_SET:
3095 * Bluetooth privileges.
3097 case PRIV_NETBLUETOOTH_RAW:
3101 * Netgraph and netgraph module privileges.
3103 case PRIV_NETGRAPH_CONTROL:
3105 case PRIV_NETGRAPH_TTY:
3109 * IPv4 and IPv6 privileges.
3111 case PRIV_NETINET_IPFW:
3112 case PRIV_NETINET_DIVERT:
3113 case PRIV_NETINET_PF:
3114 case PRIV_NETINET_DUMMYNET:
3115 case PRIV_NETINET_CARP:
3116 case PRIV_NETINET_MROUTE:
3117 case PRIV_NETINET_RAW:
3118 case PRIV_NETINET_ADDRCTRL6:
3119 case PRIV_NETINET_ND6:
3120 case PRIV_NETINET_SCOPE6:
3121 case PRIV_NETINET_ALIFETIME6:
3122 case PRIV_NETINET_IPSEC:
3123 case PRIV_NETINET_BINDANY:
3138 * No default: or deny here.
3139 * In case of no permit fall through to next switch().
3141 if (cred->cr_prison->pr_flags & PR_VNET)
3149 * Allow ktrace privileges for root in jail.
3155 * Allow jailed processes to configure audit identity and
3156 * submit audit records (login, etc). In the future we may
3157 * want to further refine the relationship between audit and
3160 case PRIV_AUDIT_GETAUDIT:
3161 case PRIV_AUDIT_SETAUDIT:
3162 case PRIV_AUDIT_SUBMIT:
3166 * Allow jailed processes to manipulate process UNIX
3167 * credentials in any way they see fit.
3169 case PRIV_CRED_SETUID:
3170 case PRIV_CRED_SETEUID:
3171 case PRIV_CRED_SETGID:
3172 case PRIV_CRED_SETEGID:
3173 case PRIV_CRED_SETGROUPS:
3174 case PRIV_CRED_SETREUID:
3175 case PRIV_CRED_SETREGID:
3176 case PRIV_CRED_SETRESUID:
3177 case PRIV_CRED_SETRESGID:
3180 * Jail implements visibility constraints already, so allow
3181 * jailed root to override uid/gid-based constraints.
3183 case PRIV_SEEOTHERGIDS:
3184 case PRIV_SEEOTHERUIDS:
3187 * Jail implements inter-process debugging limits already, so
3188 * allow jailed root various debugging privileges.
3190 case PRIV_DEBUG_DIFFCRED:
3191 case PRIV_DEBUG_SUGID:
3192 case PRIV_DEBUG_UNPRIV:
3195 * Allow jail to set various resource limits and login
3196 * properties, and for now, exceed process resource limits.
3198 case PRIV_PROC_LIMIT:
3199 case PRIV_PROC_SETLOGIN:
3200 case PRIV_PROC_SETRLIMIT:
3203 * System V and POSIX IPC privileges are granted in jail.
3206 case PRIV_IPC_WRITE:
3207 case PRIV_IPC_ADMIN:
3208 case PRIV_IPC_MSGSIZE:
3212 * Jail operations within a jail work on child jails.
3214 case PRIV_JAIL_ATTACH:
3216 case PRIV_JAIL_REMOVE:
3219 * Jail implements its own inter-process limits, so allow
3220 * root processes in jail to change scheduling on other
3221 * processes in the same jail. Likewise for signalling.
3223 case PRIV_SCHED_DIFFCRED:
3224 case PRIV_SCHED_CPUSET:
3225 case PRIV_SIGNAL_DIFFCRED:
3226 case PRIV_SIGNAL_SUGID:
3229 * Allow jailed processes to write to sysctls marked as jail
3232 case PRIV_SYSCTL_WRITEJAIL:
3235 * Allow root in jail to manage a variety of quota
3236 * properties. These should likely be conditional on a
3237 * configuration option.
3239 case PRIV_VFS_GETQUOTA:
3240 case PRIV_VFS_SETQUOTA:
3243 * Since Jail relies on chroot() to implement file system
3244 * protections, grant many VFS privileges to root in jail.
3245 * Be careful to exclude mount-related and NFS-related
3249 case PRIV_VFS_WRITE:
3250 case PRIV_VFS_ADMIN:
3252 case PRIV_VFS_LOOKUP:
3253 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3254 case PRIV_VFS_CHFLAGS_DEV:
3255 case PRIV_VFS_CHOWN:
3256 case PRIV_VFS_CHROOT:
3257 case PRIV_VFS_RETAINSUGID:
3258 case PRIV_VFS_FCHROOT:
3260 case PRIV_VFS_SETGID:
3262 case PRIV_VFS_STICKYFILE:
3265 * As in the non-jail case, non-root users are expected to be
3266 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3267 * exists in the jail and they have permission to access it).
3269 case PRIV_KMEM_READ:
3273 * Depending on the global setting, allow privilege of
3274 * setting system flags.
3276 case PRIV_VFS_SYSFLAGS:
3277 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3283 * Depending on the global setting, allow privilege of
3284 * mounting/unmounting file systems.
3286 case PRIV_VFS_MOUNT:
3287 case PRIV_VFS_UNMOUNT:
3288 case PRIV_VFS_MOUNT_NONUSER:
3289 case PRIV_VFS_MOUNT_OWNER:
3290 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3291 cred->cr_prison->pr_enforce_statfs < 2)
3297 * Conditionnaly allow locking (unlocking) physical pages
3301 case PRIV_VM_MUNLOCK:
3302 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3308 * Conditionally allow jailed root to bind reserved ports.
3310 case PRIV_NETINET_RESERVEDPORT:
3311 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3317 * Allow jailed root to reuse in-use ports.
3319 case PRIV_NETINET_REUSEPORT:
3323 * Allow jailed root to set certain IPv4/6 (option) headers.
3325 case PRIV_NETINET_SETHDROPTS:
3329 * Conditionally allow creating raw sockets in jail.
3331 case PRIV_NETINET_RAW:
3332 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3338 * Since jail implements its own visibility limits on netstat
3339 * sysctls, allow getcred. This allows identd to work in
3342 case PRIV_NETINET_GETCRED:
3346 * Allow jailed root to set loginclass.
3348 case PRIV_PROC_SETLOGINCLASS:
3353 * In all remaining cases, deny the privilege request. This
3354 * includes almost all network privileges, many system
3355 * configuration privileges.
3362 * Return the part of pr2's name that is relative to pr1, or the whole name
3363 * if it does not directly follow.
3367 prison_name(struct prison *pr1, struct prison *pr2)
3371 /* Jails see themselves as "0" (if they see themselves at all). */
3374 name = pr2->pr_name;
3375 if (prison_ischild(pr1, pr2)) {
3377 * pr1 isn't locked (and allprison_lock may not be either)
3378 * so its length can't be counted on. But the number of dots
3379 * can be counted on - and counted.
3381 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3382 name = strchr(name, '.') + 1;
3388 * Return the part of pr2's path that is relative to pr1, or the whole path
3389 * if it does not directly follow.
3392 prison_path(struct prison *pr1, struct prison *pr2)
3394 char *path1, *path2;
3397 path1 = pr1->pr_path;
3398 path2 = pr2->pr_path;
3399 if (!strcmp(path1, "/"))
3401 len1 = strlen(path1);
3402 if (strncmp(path1, path2, len1))
3404 if (path2[len1] == '\0')
3406 if (path2[len1] == '/')
3407 return (path2 + len1);
3413 * Jail-related sysctls.
3415 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3419 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3422 struct prison *pr, *cpr;
3424 struct in_addr *ip4 = NULL;
3428 struct in6_addr *ip6 = NULL;
3433 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3434 pr = req->td->td_ucred->cr_prison;
3436 sx_slock(&allprison_lock);
3437 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3438 #if defined(INET) || defined(INET6)
3441 mtx_lock(&cpr->pr_mtx);
3443 if (cpr->pr_ip4s > 0) {
3444 if (ip4s < cpr->pr_ip4s) {
3445 ip4s = cpr->pr_ip4s;
3446 mtx_unlock(&cpr->pr_mtx);
3447 ip4 = realloc(ip4, ip4s *
3448 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3451 bcopy(cpr->pr_ip4, ip4,
3452 cpr->pr_ip4s * sizeof(struct in_addr));
3456 if (cpr->pr_ip6s > 0) {
3457 if (ip6s < cpr->pr_ip6s) {
3458 ip6s = cpr->pr_ip6s;
3459 mtx_unlock(&cpr->pr_mtx);
3460 ip6 = realloc(ip6, ip6s *
3461 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3464 bcopy(cpr->pr_ip6, ip6,
3465 cpr->pr_ip6s * sizeof(struct in6_addr));
3468 if (cpr->pr_ref == 0) {
3469 mtx_unlock(&cpr->pr_mtx);
3472 bzero(xp, sizeof(*xp));
3473 xp->pr_version = XPRISON_VERSION;
3474 xp->pr_id = cpr->pr_id;
3475 xp->pr_state = cpr->pr_uref > 0
3476 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3477 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3478 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3479 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3481 xp->pr_ip4s = cpr->pr_ip4s;
3484 xp->pr_ip6s = cpr->pr_ip6s;
3486 mtx_unlock(&cpr->pr_mtx);
3487 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3491 if (xp->pr_ip4s > 0) {
3492 error = SYSCTL_OUT(req, ip4,
3493 xp->pr_ip4s * sizeof(struct in_addr));
3499 if (xp->pr_ip6s > 0) {
3500 error = SYSCTL_OUT(req, ip6,
3501 xp->pr_ip6s * sizeof(struct in6_addr));
3507 sx_sunlock(&allprison_lock);
3518 SYSCTL_OID(_security_jail, OID_AUTO, list,
3519 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3520 sysctl_jail_list, "S", "List of active jails");
3523 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3527 injail = jailed(req->td->td_ucred);
3528 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3533 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3534 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3535 sysctl_jail_jailed, "I", "Process in jail?");
3538 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3540 int error, havevnet;
3542 struct ucred *cred = req->td->td_ucred;
3544 havevnet = jailed(cred) && prison_owns_vnet(cred);
3548 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3553 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3554 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3555 sysctl_jail_vnet, "I", "Jail owns vnet?");
3557 #if defined(INET) || defined(INET6)
3558 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3559 &jail_max_af_ips, 0,
3560 "Number of IP addresses a jail may have at most per address family (deprecated)");
3564 * Default parameters for jail(2) compatibility. For historical reasons,
3565 * the sysctl names have varying similarity to the parameter names. Prisons
3566 * just see their own parameters, and can't change them.
3569 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3572 int allow, error, i;
3574 pr = req->td->td_ucred->cr_prison;
3575 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3577 /* Get the current flag value, and convert it to a boolean. */
3578 i = (allow & arg2) ? 1 : 0;
3581 error = sysctl_handle_int(oidp, &i, 0, req);
3582 if (error || !req->newptr)
3588 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3591 mtx_lock(&prison0.pr_mtx);
3592 jail_default_allow = (jail_default_allow & ~arg2) | i;
3593 mtx_unlock(&prison0.pr_mtx);
3597 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3598 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3599 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3600 "Processes in jail can set their hostnames (deprecated)");
3601 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3602 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3603 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3604 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3605 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3606 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3607 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3608 "Processes in jail can use System V IPC primitives (deprecated)");
3609 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3610 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3611 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3612 "Prison root can create raw sockets (deprecated)");
3613 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3614 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3615 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3616 "Processes in jail can alter system file flags (deprecated)");
3617 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3618 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3619 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3620 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3623 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3628 pr = req->td->td_ucred->cr_prison;
3629 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3630 error = sysctl_handle_int(oidp, &level, 0, req);
3631 if (error || !req->newptr)
3633 *(int *)arg1 = level;
3637 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3638 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3639 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3640 sysctl_jail_default_level, "I",
3641 "Processes in jail cannot see all mounted file systems (deprecated)");
3643 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3644 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3645 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3646 sysctl_jail_default_level, "I",
3647 "Ruleset for the devfs filesystem in jail (deprecated)");
3650 * Nodes to describe jail parameters. Maximum length of string parameters
3651 * is returned in the string itself, and the other parameters exist merely
3652 * to make themselves and their types known.
3654 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3658 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3665 switch (oidp->oid_kind & CTLTYPE)
3671 if (!(req->flags & SCTL_MASK32))
3673 return (SYSCTL_OUT(req, &l, sizeof(l)));
3677 return (SYSCTL_OUT(req, &i, sizeof(i)));
3678 case CTLTYPE_STRING:
3679 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3681 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3682 case CTLTYPE_STRUCT:
3684 return (SYSCTL_OUT(req, &s, sizeof(s)));
3690 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3691 * jail creation time but cannot be changed in an existing jail.
3693 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3694 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3695 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3696 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3697 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3698 "I", "Jail secure level");
3699 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3700 "Jail value for kern.osreldate and uname -K");
3701 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3702 "Jail value for kern.osrelease and uname -r");
3703 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3704 "I", "Jail cannot see all mounted file systems");
3705 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3706 "I", "Ruleset for in-jail devfs mounts");
3707 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3708 "B", "Jail persistence");
3710 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3711 "E,jailsys", "Virtual network stack");
3713 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3714 "B", "Jail is in the process of shutting down");
3716 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3717 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3718 "I", "Current number of child jails");
3719 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3720 "I", "Maximum number of child jails");
3722 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3723 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3725 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3726 "Jail NIS domainname");
3727 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3729 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3730 "LU", "Jail host ID");
3732 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3733 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3736 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3737 "Jail IPv4 address virtualization");
3738 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3739 "S,in_addr,a", "Jail IPv4 addresses");
3740 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3741 "B", "Do (not) use IPv4 source address selection rather than the "
3742 "primary jail IPv4 address.");
3745 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3746 "Jail IPv6 address virtualization");
3747 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3748 "S,in6_addr,a", "Jail IPv6 addresses");
3749 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3750 "B", "Do (not) use IPv6 source address selection rather than the "
3751 "primary jail IPv6 address.");
3754 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3755 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3756 "B", "Jail may set hostname");
3757 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3758 "B", "Jail may use SYSV IPC");
3759 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3760 "B", "Jail may create raw sockets");
3761 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3762 "B", "Jail may alter system file flags");
3763 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3764 "B", "Jail may set file quotas");
3765 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3766 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3767 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3768 "B", "Jail may lock (unlock) physical pages in memory");
3769 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3770 "B", "Jail may bind sockets to reserved ports");
3772 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3773 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3774 "B", "Jail may mount/unmount jail-friendly file systems in general");
3777 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
3778 * its associated bit in the pr_allow bitmask, or zero if the parameter was
3782 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3785 struct bool_flags *bf;
3786 struct sysctl_oid *parent;
3787 char *allow_name, *allow_noname, *allowed;
3788 #ifndef NO_SYSCTL_DESCR
3789 char *descr_deprecated;
3791 unsigned allow_flag;
3794 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3796 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3798 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3799 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3800 free(allow_name, M_PRISON);
3805 * See if this parameter has already beed added, i.e. a module was
3806 * previously loaded/unloaded.
3808 mtx_lock(&prison0.pr_mtx);
3809 for (bf = pr_flag_allow;
3810 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3812 if (strcmp(bf->name, allow_name) == 0) {
3813 allow_flag = bf->flag;
3819 * Find a free bit in prison0's pr_allow, failing if there are none
3820 * (which shouldn't happen as long as we keep track of how many
3821 * potential dynamic flags exist).
3823 for (allow_flag = 1;; allow_flag <<= 1) {
3824 if (allow_flag == 0)
3826 if ((prison0.pr_allow & allow_flag) == 0)
3831 * Note the parameter in the next open slot in pr_flag_allow.
3832 * Set the flag last so code that checks pr_flag_allow can do so
3835 for (bf = pr_flag_allow; bf->flag != 0; bf++)
3836 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3837 /* This should never happen, but is not fatal. */
3841 prison0.pr_allow |= allow_flag;
3842 bf->name = allow_name;
3843 bf->noname = allow_noname;
3844 bf->flag = allow_flag;
3845 mtx_unlock(&prison0.pr_mtx);
3848 * Create sysctls for the paramter, and the back-compat global
3852 ? SYSCTL_ADD_NODE(NULL,
3853 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3854 OID_AUTO, prefix, 0, 0, prefix_descr)
3855 : &sysctl___security_jail_param_allow;
3856 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3857 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3858 NULL, 0, sysctl_jail_param, "B", descr);
3860 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3861 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3862 #ifndef NO_SYSCTL_DESCR
3863 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3866 (void)SYSCTL_ADD_PROC(NULL,
3867 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3868 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3869 sysctl_jail_default_allow, "I", descr_deprecated);
3870 #ifndef NO_SYSCTL_DESCR
3871 free(descr_deprecated, M_TEMP);
3873 free(allowed, M_TEMP);
3878 mtx_unlock(&prison0.pr_mtx);
3879 free(allow_name, M_PRISON);
3880 free(allow_noname, M_PRISON);
3885 * The VFS system will register jail-aware filesystems here. They each get
3886 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3887 * attempts to mount.
3890 prison_add_vfs(struct vfsconf *vfsp)
3892 #ifdef NO_SYSCTL_DESCR
3894 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3899 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3901 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3903 free(descr, M_TEMP);
3909 prison_racct_foreach(void (*callback)(struct racct *racct,
3910 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3911 void *arg2, void *arg3)
3913 struct prison_racct *prr;
3915 ASSERT_RACCT_ENABLED();
3917 sx_slock(&allprison_lock);
3920 LIST_FOREACH(prr, &allprison_racct, prr_next)
3921 (callback)(prr->prr_racct, arg2, arg3);
3924 sx_sunlock(&allprison_lock);
3927 static struct prison_racct *
3928 prison_racct_find_locked(const char *name)
3930 struct prison_racct *prr;
3932 ASSERT_RACCT_ENABLED();
3933 sx_assert(&allprison_lock, SA_XLOCKED);
3935 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3938 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3939 if (strcmp(name, prr->prr_name) != 0)
3942 /* Found prison_racct with a matching name? */
3943 prison_racct_hold(prr);
3947 /* Add new prison_racct. */
3948 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3949 racct_create(&prr->prr_racct);
3951 strcpy(prr->prr_name, name);
3952 refcount_init(&prr->prr_refcount, 1);
3953 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3958 struct prison_racct *
3959 prison_racct_find(const char *name)
3961 struct prison_racct *prr;
3963 ASSERT_RACCT_ENABLED();
3965 sx_xlock(&allprison_lock);
3966 prr = prison_racct_find_locked(name);
3967 sx_xunlock(&allprison_lock);
3972 prison_racct_hold(struct prison_racct *prr)
3975 ASSERT_RACCT_ENABLED();
3977 refcount_acquire(&prr->prr_refcount);
3981 prison_racct_free_locked(struct prison_racct *prr)
3984 ASSERT_RACCT_ENABLED();
3985 sx_assert(&allprison_lock, SA_XLOCKED);
3987 if (refcount_release(&prr->prr_refcount)) {
3988 racct_destroy(&prr->prr_racct);
3989 LIST_REMOVE(prr, prr_next);
3990 free(prr, M_PRISON_RACCT);
3995 prison_racct_free(struct prison_racct *prr)
3999 ASSERT_RACCT_ENABLED();
4000 sx_assert(&allprison_lock, SA_UNLOCKED);
4002 old = prr->prr_refcount;
4003 if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4006 sx_xlock(&allprison_lock);
4007 prison_racct_free_locked(prr);
4008 sx_xunlock(&allprison_lock);
4012 prison_racct_attach(struct prison *pr)
4014 struct prison_racct *prr;
4016 ASSERT_RACCT_ENABLED();
4017 sx_assert(&allprison_lock, SA_XLOCKED);
4019 prr = prison_racct_find_locked(pr->pr_name);
4020 KASSERT(prr != NULL, ("cannot find prison_racct"));
4022 pr->pr_prison_racct = prr;
4026 * Handle jail renaming. From the racct point of view, renaming means
4027 * moving from one prison_racct to another.
4030 prison_racct_modify(struct prison *pr)
4036 struct prison_racct *oldprr;
4038 ASSERT_RACCT_ENABLED();
4040 sx_slock(&allproc_lock);
4041 sx_xlock(&allprison_lock);
4043 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4044 sx_xunlock(&allprison_lock);
4045 sx_sunlock(&allproc_lock);
4049 oldprr = pr->pr_prison_racct;
4050 pr->pr_prison_racct = NULL;
4052 prison_racct_attach(pr);
4055 * Move resource utilisation records.
4057 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4061 * Force rctl to reattach rules to processes.
4063 FOREACH_PROC_IN_SYSTEM(p) {
4065 cred = crhold(p->p_ucred);
4067 rctl_proc_ucred_changed(p, cred);
4072 sx_sunlock(&allproc_lock);
4073 prison_racct_free_locked(oldprr);
4074 sx_xunlock(&allprison_lock);
4078 prison_racct_detach(struct prison *pr)
4081 ASSERT_RACCT_ENABLED();
4082 sx_assert(&allprison_lock, SA_UNLOCKED);
4084 if (pr->pr_prison_racct == NULL)
4086 prison_racct_free(pr->pr_prison_racct);
4087 pr->pr_prison_racct = NULL;
4094 db_show_prison(struct prison *pr)
4096 struct bool_flags *bf;
4097 struct jailsys_flags *jsf;
4098 #if defined(INET) || defined(INET6)
4103 char ip4buf[INET_ADDRSTRLEN];
4106 char ip6buf[INET6_ADDRSTRLEN];
4109 db_printf("prison %p:\n", pr);
4110 db_printf(" jid = %d\n", pr->pr_id);
4111 db_printf(" name = %s\n", pr->pr_name);
4112 db_printf(" parent = %p\n", pr->pr_parent);
4113 db_printf(" ref = %d\n", pr->pr_ref);
4114 db_printf(" uref = %d\n", pr->pr_uref);
4115 db_printf(" path = %s\n", pr->pr_path);
4116 db_printf(" cpuset = %d\n", pr->pr_cpuset
4117 ? pr->pr_cpuset->cs_id : -1);
4119 db_printf(" vnet = %p\n", pr->pr_vnet);
4121 db_printf(" root = %p\n", pr->pr_root);
4122 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4123 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4124 db_printf(" children.max = %d\n", pr->pr_childmax);
4125 db_printf(" children.cur = %d\n", pr->pr_childcount);
4126 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4127 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4128 db_printf(" flags = 0x%x", pr->pr_flags);
4129 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4130 if (pr->pr_flags & bf->flag)
4131 db_printf(" %s", bf->name);
4132 for (jsf = pr_flag_jailsys;
4133 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4135 f = pr->pr_flags & (jsf->disable | jsf->new);
4136 db_printf(" %-16s= %s\n", jsf->name,
4137 (f != 0 && f == jsf->disable) ? "disable"
4138 : (f == jsf->new) ? "new"
4141 db_printf(" allow = 0x%x", pr->pr_allow);
4142 for (bf = pr_flag_allow;
4143 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4145 if (pr->pr_allow & bf->flag)
4146 db_printf(" %s", bf->name);
4148 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4149 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4150 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4151 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4152 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4154 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4155 for (ii = 0; ii < pr->pr_ip4s; ii++)
4156 db_printf(" %s %s\n",
4157 ii == 0 ? "ip4.addr =" : " ",
4158 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4161 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4162 for (ii = 0; ii < pr->pr_ip6s; ii++)
4163 db_printf(" %s %s\n",
4164 ii == 0 ? "ip6.addr =" : " ",
4165 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4169 DB_SHOW_COMMAND(prison, db_show_prison_command)
4175 * Show all prisons in the list, and prison0 which is not
4178 db_show_prison(&prison0);
4179 if (!db_pager_quit) {
4180 TAILQ_FOREACH(pr, &allprison, pr_list) {
4192 /* Look for a prison with the ID and with references. */
4193 TAILQ_FOREACH(pr, &allprison, pr_list)
4194 if (pr->pr_id == addr && pr->pr_ref > 0)
4197 /* Look again, without requiring a reference. */
4198 TAILQ_FOREACH(pr, &allprison, pr_list)
4199 if (pr->pr_id == addr)
4202 /* Assume address points to a valid prison. */
4203 pr = (struct prison *)addr;