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},
196 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
197 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
198 PR_ALLOW_UNPRIV_DEBUG},
200 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
202 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
203 PR_ALLOW_RESERVED_PORTS | \
204 PR_ALLOW_UNPRIV_DEBUG)
205 #define JAIL_DEFAULT_ENFORCE_STATFS 2
206 #define JAIL_DEFAULT_DEVFS_RSNUM 0
207 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
208 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
209 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
210 #if defined(INET) || defined(INET6)
211 static unsigned jail_max_af_ips = 255;
215 * Initialize the parts of prison0 that can't be static-initialized with
216 * constants. This is called from proc0_init() after creating thread0 cpuset.
222 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
223 prison0.pr_osreldate = osreldate;
224 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
233 sys_jail(struct thread *td, struct jail_args *uap)
239 error = copyin(uap->jail, &version, sizeof(uint32_t));
248 /* FreeBSD single IPv4 jails. */
249 bzero(&j, sizeof(struct jail));
250 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
253 j.version = j0.version;
255 j.hostname = j0.hostname;
256 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
262 * Version 1 was used by multi-IPv4 jail implementations
263 * that never made it into the official kernel.
267 case 2: /* JAIL_API_VERSION */
268 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
269 error = copyin(uap->jail, &j, sizeof(struct jail));
275 /* Sci-Fi jails are not supported, sorry. */
278 return (kern_jail(td, &j));
282 kern_jail(struct thread *td, struct jail *j)
284 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
293 char *u_path, *u_hostname, *u_name;
294 struct bool_flags *bf;
297 struct in_addr *u_ip4;
300 struct in6_addr *u_ip6;
303 int error, enforce_statfs;
305 bzero(&optiov, sizeof(optiov));
306 opt.uio_iov = optiov;
310 opt.uio_segflg = UIO_SYSSPACE;
311 opt.uio_rw = UIO_READ;
314 /* Set permissions for top-level jails from sysctls. */
315 if (!jailed(td->td_ucred)) {
316 for (bf = pr_flag_allow;
317 bf < pr_flag_allow + nitems(pr_flag_allow) &&
320 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
321 (jail_default_allow & bf->flag)
322 ? bf->name : bf->noname);
323 optiov[opt.uio_iovcnt].iov_len =
324 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
327 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
328 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
330 enforce_statfs = jail_default_enforce_statfs;
331 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
332 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
336 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
338 ip4s = (j->version == 0) ? 1 : j->ip4s;
339 if (ip4s > jail_max_af_ips)
341 tmplen += ip4s * sizeof(struct in_addr);
347 if (j->ip6s > jail_max_af_ips)
349 tmplen += j->ip6s * sizeof(struct in6_addr);
354 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
355 u_hostname = u_path + MAXPATHLEN;
356 u_name = u_hostname + MAXHOSTNAMELEN;
358 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
362 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
364 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
367 optiov[opt.uio_iovcnt].iov_base = "path";
368 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
370 optiov[opt.uio_iovcnt].iov_base = u_path;
371 error = copyinstr(j->path, u_path, MAXPATHLEN,
372 &optiov[opt.uio_iovcnt].iov_len);
374 free(u_path, M_TEMP);
378 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
379 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
381 optiov[opt.uio_iovcnt].iov_base = u_hostname;
382 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
383 &optiov[opt.uio_iovcnt].iov_len);
385 free(u_path, M_TEMP);
389 if (j->jailname != NULL) {
390 optiov[opt.uio_iovcnt].iov_base = "name";
391 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
393 optiov[opt.uio_iovcnt].iov_base = u_name;
394 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
395 &optiov[opt.uio_iovcnt].iov_len);
397 free(u_path, M_TEMP);
403 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
404 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
406 optiov[opt.uio_iovcnt].iov_base = u_ip4;
407 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
409 u_ip4->s_addr = j->ip4s;
411 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
413 free(u_path, M_TEMP);
420 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
421 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
423 optiov[opt.uio_iovcnt].iov_base = u_ip6;
424 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
425 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
427 free(u_path, M_TEMP);
432 KASSERT(opt.uio_iovcnt <= nitems(optiov),
433 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
434 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
435 free(u_path, M_TEMP);
440 * struct jail_set_args {
441 * struct iovec *iovp;
442 * unsigned int iovcnt;
447 sys_jail_set(struct thread *td, struct jail_set_args *uap)
452 /* Check that we have an even number of iovecs. */
456 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
459 error = kern_jail_set(td, auio, uap->flags);
465 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
472 struct in6_addr *ip6;
475 struct vfsoptlist *opts;
476 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
478 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
479 char *g_path, *osrelstr;
480 struct bool_flags *bf;
481 struct jailsys_flags *jsf;
482 #if defined(INET) || defined(INET6)
487 size_t namelen, onamelen, pnamelen;
488 int born, created, cuflags, descend, enforce;
489 int error, errmsg_len, errmsg_pos;
490 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
491 int jid, jsys, len, level;
492 int childmax, osreldt, rsnum, slevel;
493 #if defined(INET) || defined(INET6)
502 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
503 uint64_t pr_allow_diff;
507 error = priv_check(td, PRIV_JAIL_SET);
508 if (!error && (flags & JAIL_ATTACH))
509 error = priv_check(td, PRIV_JAIL_ATTACH);
512 mypr = td->td_ucred->cr_prison;
513 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
515 if (flags & ~JAIL_SET_MASK)
519 * Check all the parameters before committing to anything. Not all
520 * errors can be caught early, but we may as well try. Also, this
521 * takes care of some expensive stuff (path lookup) before getting
522 * the allprison lock.
524 * XXX Jails are not filesystems, and jail parameters are not mount
525 * options. But it makes more sense to re-use the vfsopt code
526 * than duplicate it under a different name.
528 error = vfs_buildopts(optuio, &opts);
539 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
542 vfs_opterror(opts, "no valid operation (create or update)");
546 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
552 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
561 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
569 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
574 else if (enforce < 0 || enforce > 2) {
580 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
588 pr_flags = ch_flags = 0;
589 for (bf = pr_flag_bool;
590 bf < pr_flag_bool + nitems(pr_flag_bool);
592 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
593 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
595 ch_flags |= pr_flags;
596 for (jsf = pr_flag_jailsys;
597 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
599 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
605 case JAIL_SYS_DISABLE:
610 pr_flags |= jsf->disable;
613 pr_flags |= jsf->new;
615 case JAIL_SYS_INHERIT:
621 ch_flags |= jsf->new | jsf->disable;
623 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
624 && !(pr_flags & PR_PERSIST)) {
626 vfs_opterror(opts, "new jail must persist or attach");
630 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
632 vfs_opterror(opts, "vnet cannot be changed after creation");
637 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
639 vfs_opterror(opts, "ip4 cannot be changed after creation");
644 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
646 vfs_opterror(opts, "ip6 cannot be changed after creation");
651 pr_allow = ch_allow = 0;
652 for (bf = pr_flag_allow;
653 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
655 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
656 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
658 ch_allow |= pr_allow;
660 error = vfs_getopt(opts, "name", (void **)&name, &len);
666 if (len == 0 || name[len - 1] != '\0') {
670 if (len > MAXHOSTNAMELEN) {
671 error = ENAMETOOLONG;
676 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
684 if (len == 0 || host[len - 1] != '\0') {
688 if (len > MAXHOSTNAMELEN) {
689 error = ENAMETOOLONG;
694 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
702 if (len == 0 || domain[len - 1] != '\0') {
706 if (len > MAXHOSTNAMELEN) {
707 error = ENAMETOOLONG;
712 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
720 if (len == 0 || uuid[len - 1] != '\0') {
724 if (len > HOSTUUIDLEN) {
725 error = ENAMETOOLONG;
730 #ifdef COMPAT_FREEBSD32
731 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
734 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
738 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
750 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
755 else if (ip4s & (sizeof(*ip4) - 1)) {
759 ch_flags |= PR_IP4_USER;
760 pr_flags |= PR_IP4_USER;
762 ip4s /= sizeof(*ip4);
763 if (ip4s > jail_max_af_ips) {
765 vfs_opterror(opts, "too many IPv4 addresses");
768 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
769 bcopy(op, ip4, ip4s * sizeof(*ip4));
771 * IP addresses are all sorted but ip[0] to preserve
772 * the primary IP address as given from userland.
773 * This special IP is used for unbound outgoing
774 * connections as well for "loopback" traffic in case
775 * source address selection cannot find any more fitting
776 * address to connect from.
779 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
782 * Check for duplicate addresses and do some simple
783 * zero and broadcast checks. If users give other bogus
784 * addresses it is their problem.
786 * We do not have to care about byte order for these
787 * checks so we will do them in NBO.
789 for (ii = 0; ii < ip4s; ii++) {
790 if (ip4[ii].s_addr == INADDR_ANY ||
791 ip4[ii].s_addr == INADDR_BROADCAST) {
796 (ip4[0].s_addr == ip4[ii+1].s_addr ||
797 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
807 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
812 else if (ip6s & (sizeof(*ip6) - 1)) {
816 ch_flags |= PR_IP6_USER;
817 pr_flags |= PR_IP6_USER;
819 ip6s /= sizeof(*ip6);
820 if (ip6s > jail_max_af_ips) {
822 vfs_opterror(opts, "too many IPv6 addresses");
825 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
826 bcopy(op, ip6, ip6s * sizeof(*ip6));
828 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
830 for (ii = 0; ii < ip6s; ii++) {
831 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
836 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
837 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
847 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
848 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
851 "vnet jails cannot have IP address restrictions");
856 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
862 if (flags & JAIL_UPDATE) {
865 "osrelease cannot be changed after creation");
868 if (len == 0 || osrelstr[len - 1] != '\0') {
872 if (len >= OSRELEASELEN) {
873 error = ENAMETOOLONG;
875 "osrelease string must be 1-%d bytes long",
881 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
887 if (flags & JAIL_UPDATE) {
890 "osreldate cannot be changed after creation");
895 vfs_opterror(opts, "osreldate cannot be 0");
901 error = vfs_getopt(opts, "path", (void **)&path, &len);
907 if (flags & JAIL_UPDATE) {
910 "path cannot be changed after creation");
913 if (len == 0 || path[len - 1] != '\0') {
917 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
923 NDFREE(&nd, NDF_ONLY_PNBUF);
924 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
925 strlcpy(g_path, path, MAXPATHLEN);
926 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
930 /* exit on other errors */
933 if (root->v_type != VDIR) {
942 * Find the specified jail, or at least its parent.
943 * This abuses the file error codes ENOENT and EEXIST.
947 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
948 namelc = strrchr(name, '.');
949 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
953 sx_xlock(&allprison_lock);
956 * See if a requested jid already exists. There is an
957 * information leak here if the jid exists but is not within
958 * the caller's jail hierarchy. Jail creators will get EEXIST
959 * even though they cannot see the jail, and CREATE | UPDATE
960 * will return ENOENT which is not normally a valid error.
964 vfs_opterror(opts, "negative jid");
965 goto done_unlock_list;
967 pr = prison_find(jid);
970 /* Create: jid must not exist. */
971 if (cuflags == JAIL_CREATE) {
972 mtx_unlock(&pr->pr_mtx);
974 vfs_opterror(opts, "jail %d already exists",
976 goto done_unlock_list;
978 if (!prison_ischild(mypr, pr)) {
979 mtx_unlock(&pr->pr_mtx);
981 } else if (pr->pr_uref == 0) {
982 if (!(flags & JAIL_DYING)) {
983 mtx_unlock(&pr->pr_mtx);
985 vfs_opterror(opts, "jail %d is dying",
987 goto done_unlock_list;
988 } else if ((flags & JAIL_ATTACH) ||
989 (pr_flags & PR_PERSIST)) {
991 * A dying jail might be resurrected
992 * (via attach or persist), but first
993 * it must determine if another jail
994 * has claimed its name. Accomplish
995 * this by implicitly re-setting the
999 name = prison_name(mypr, pr);
1004 /* Update: jid must exist. */
1005 if (cuflags == JAIL_UPDATE) {
1007 vfs_opterror(opts, "jail %d not found", jid);
1008 goto done_unlock_list;
1013 * If the caller provided a name, look for a jail by that name.
1014 * This has different semantics for creates and updates keyed by jid
1015 * (where the name must not already exist in a different jail),
1016 * and updates keyed by the name itself (where the name must exist
1017 * because that is the jail being updated).
1021 namelc = strrchr(name, '.');
1026 * This is a hierarchical name. Split it into the
1027 * parent and child names, and make sure the parent
1028 * exists or matches an already found jail.
1031 if (strncmp(name, ppr->pr_name, namelc - name)
1032 || ppr->pr_name[namelc - name] != '\0') {
1033 mtx_unlock(&pr->pr_mtx);
1036 "cannot change jail's parent");
1037 goto done_unlock_list;
1041 ppr = prison_find_name(mypr, name);
1045 "jail \"%s\" not found", name);
1046 goto done_unlock_list;
1048 mtx_unlock(&ppr->pr_mtx);
1053 if (namelc[0] != '\0') {
1055 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1058 FOREACH_PRISON_CHILD(ppr, tpr) {
1059 if (tpr != pr && tpr->pr_ref > 0 &&
1060 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1062 cuflags != JAIL_CREATE) {
1063 mtx_lock(&tpr->pr_mtx);
1064 if (tpr->pr_ref > 0) {
1069 if (tpr->pr_uref > 0) {
1075 mtx_unlock(&tpr->pr_mtx);
1076 } else if (tpr->pr_uref > 0) {
1078 * Create, or update(jid):
1079 * name must not exist in an
1080 * active sibling jail.
1084 mtx_unlock(&pr->pr_mtx);
1086 "jail \"%s\" already exists",
1088 goto done_unlock_list;
1092 /* If no active jail is found, use a dying one. */
1093 if (deadpr != NULL && pr == NULL) {
1094 if (flags & JAIL_DYING) {
1095 mtx_lock(&deadpr->pr_mtx);
1096 if (deadpr->pr_ref == 0) {
1097 mtx_unlock(&deadpr->pr_mtx);
1101 } else if (cuflags == JAIL_UPDATE) {
1104 "jail \"%s\" is dying", name);
1105 goto done_unlock_list;
1108 /* Update: name must exist if no jid. */
1109 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1111 vfs_opterror(opts, "jail \"%s\" not found",
1113 goto done_unlock_list;
1117 /* Update: must provide a jid or name. */
1118 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1120 vfs_opterror(opts, "update specified no jail");
1121 goto done_unlock_list;
1124 /* If there's no prison to update, create a new one and link it in. */
1126 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1127 if (tpr->pr_childcount >= tpr->pr_childmax) {
1129 vfs_opterror(opts, "prison limit exceeded");
1130 goto done_unlock_list;
1133 mtx_lock(&ppr->pr_mtx);
1134 if (ppr->pr_ref == 0) {
1135 mtx_unlock(&ppr->pr_mtx);
1137 vfs_opterror(opts, "jail \"%s\" not found",
1138 prison_name(mypr, ppr));
1139 goto done_unlock_list;
1143 mtx_unlock(&ppr->pr_mtx);
1144 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1146 /* Find the next free jid. */
1149 if (jid == JAIL_MAX)
1151 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1152 if (tpr->pr_id < jid)
1154 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1155 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1158 if (jid == lastprid) {
1161 "no available jail IDs");
1163 prison_deref(ppr, PD_DEREF |
1164 PD_DEUREF | PD_LIST_XLOCKED);
1173 * The jail already has a jid (that did not yet exist),
1174 * so just find where to insert it.
1176 TAILQ_FOREACH(tpr, &allprison, pr_list)
1177 if (tpr->pr_id >= jid) {
1178 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1183 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1184 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1185 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1186 tpr->pr_childcount++;
1188 pr->pr_parent = ppr;
1191 /* Set some default values, and inherit some from the parent. */
1196 root = mypr->pr_root;
1199 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1200 pr->pr_flags |= PR_HOST;
1201 #if defined(INET) || defined(INET6)
1203 if (!(pr_flags & PR_VNET))
1207 if (!(ch_flags & PR_IP4_USER))
1208 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1209 else if (!(pr_flags & PR_IP4_USER)) {
1210 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1211 if (ppr->pr_ip4 != NULL) {
1212 pr->pr_ip4s = ppr->pr_ip4s;
1213 pr->pr_ip4 = malloc(pr->pr_ip4s *
1214 sizeof(struct in_addr), M_PRISON,
1216 bcopy(ppr->pr_ip4, pr->pr_ip4,
1217 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1222 if (!(ch_flags & PR_IP6_USER))
1223 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1224 else if (!(pr_flags & PR_IP6_USER)) {
1225 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1226 if (ppr->pr_ip6 != NULL) {
1227 pr->pr_ip6s = ppr->pr_ip6s;
1228 pr->pr_ip6 = malloc(pr->pr_ip6s *
1229 sizeof(struct in6_addr), M_PRISON,
1231 bcopy(ppr->pr_ip6, pr->pr_ip6,
1232 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1238 /* Source address selection is always on by default. */
1239 pr->pr_flags |= _PR_IP_SADDRSEL;
1241 pr->pr_securelevel = ppr->pr_securelevel;
1242 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1243 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1244 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1246 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1247 if (osrelstr == NULL)
1248 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1249 sizeof(pr->pr_osrelease));
1251 strlcpy(pr->pr_osrelease, osrelstr,
1252 sizeof(pr->pr_osrelease));
1254 LIST_INIT(&pr->pr_children);
1255 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1256 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1259 /* Allocate a new vnet if specified. */
1260 pr->pr_vnet = (pr_flags & PR_VNET)
1261 ? vnet_alloc() : ppr->pr_vnet;
1264 * Allocate a dedicated cpuset for each jail.
1265 * Unlike other initial settings, this may return an erorr.
1267 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1269 prison_deref(pr, PD_LIST_XLOCKED);
1273 mtx_lock(&pr->pr_mtx);
1275 * New prisons do not yet have a reference, because we do not
1276 * want others to see the incomplete prison once the
1277 * allprison_lock is downgraded.
1282 * Grab a reference for existing prisons, to ensure they
1283 * continue to exist for the duration of the call.
1286 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1287 if ((pr->pr_flags & PR_VNET) &&
1288 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1291 "vnet jails cannot have IP address restrictions");
1292 goto done_deref_locked;
1296 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1299 "ip4 cannot be changed after creation");
1300 goto done_deref_locked;
1304 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1307 "ip6 cannot be changed after creation");
1308 goto done_deref_locked;
1313 /* Do final error checking before setting anything. */
1315 if (slevel < ppr->pr_securelevel) {
1317 goto done_deref_locked;
1321 if (childmax >= ppr->pr_childmax) {
1323 goto done_deref_locked;
1327 if (enforce < ppr->pr_enforce_statfs) {
1329 goto done_deref_locked;
1334 * devfs_rsnum is a uint16_t
1336 if (rsnum < 0 || rsnum > 65535) {
1338 goto done_deref_locked;
1341 * Nested jails always inherit parent's devfs ruleset
1343 if (jailed(td->td_ucred)) {
1344 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1346 goto done_deref_locked;
1348 rsnum = ppr->pr_devfs_rsnum;
1353 if (ppr->pr_flags & PR_IP4) {
1355 * Make sure the new set of IP addresses is a
1356 * subset of the parent's list. Don't worry
1357 * about the parent being unlocked, as any
1358 * setting is done with allprison_lock held.
1360 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1361 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1363 if (ij == ppr->pr_ip4s) {
1365 goto done_deref_locked;
1368 for (ii = ij = 1; ii < ip4s; ii++) {
1369 if (ip4[ii].s_addr ==
1370 ppr->pr_ip4[0].s_addr)
1372 for (; ij < ppr->pr_ip4s; ij++)
1373 if (ip4[ii].s_addr ==
1374 ppr->pr_ip4[ij].s_addr)
1376 if (ij == ppr->pr_ip4s)
1379 if (ij == ppr->pr_ip4s) {
1381 goto done_deref_locked;
1386 * Check for conflicting IP addresses. We permit them
1387 * if there is no more than one IP on each jail. If
1388 * there is a duplicate on a jail with more than one
1389 * IP stop checking and return error.
1392 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1393 if (tppr->pr_flags & PR_VNET)
1398 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1401 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1403 tpr->pr_uref == 0) {
1407 if (!(tpr->pr_flags & PR_IP4_USER))
1410 if (tpr->pr_ip4 == NULL ||
1411 (ip4s == 1 && tpr->pr_ip4s == 1))
1413 for (ii = 0; ii < ip4s; ii++) {
1414 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1418 "IPv4 addresses clash");
1419 goto done_deref_locked;
1427 if (ppr->pr_flags & PR_IP6) {
1429 * Make sure the new set of IP addresses is a
1430 * subset of the parent's list.
1432 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1433 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1436 if (ij == ppr->pr_ip6s) {
1438 goto done_deref_locked;
1441 for (ii = ij = 1; ii < ip6s; ii++) {
1442 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1445 for (; ij < ppr->pr_ip6s; ij++)
1446 if (IN6_ARE_ADDR_EQUAL(
1447 &ip6[ii], &ppr->pr_ip6[ij]))
1449 if (ij == ppr->pr_ip6s)
1452 if (ij == ppr->pr_ip6s) {
1454 goto done_deref_locked;
1458 /* Check for conflicting IP addresses. */
1460 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1461 if (tppr->pr_flags & PR_VNET)
1466 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1469 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1471 tpr->pr_uref == 0) {
1475 if (!(tpr->pr_flags & PR_IP6_USER))
1478 if (tpr->pr_ip6 == NULL ||
1479 (ip6s == 1 && tpr->pr_ip6s == 1))
1481 for (ii = 0; ii < ip6s; ii++) {
1482 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1486 "IPv6 addresses clash");
1487 goto done_deref_locked;
1493 onamelen = namelen = 0;
1494 if (namelc != NULL) {
1495 /* Give a default name of the jid. Also allow the name to be
1496 * explicitly the jid - but not any other number, and only in
1497 * normal form (no leading zero/etc).
1499 if (namelc[0] == '\0')
1500 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1501 else if ((strtoul(namelc, &p, 10) != jid ||
1502 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1505 "name cannot be numeric (unless it is the jid)");
1506 goto done_deref_locked;
1509 * Make sure the name isn't too long for the prison or its
1512 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1513 onamelen = strlen(pr->pr_name + pnamelen);
1514 namelen = strlen(namelc);
1515 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1516 error = ENAMETOOLONG;
1517 goto done_deref_locked;
1519 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1520 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1521 sizeof(pr->pr_name)) {
1522 error = ENAMETOOLONG;
1523 goto done_deref_locked;
1527 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1528 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1530 goto done_deref_locked;
1534 * Let modules check their parameters. This requires unlocking and
1535 * then re-locking the prison, but this is still a valid state as long
1536 * as allprison_lock remains xlocked.
1538 mtx_unlock(&pr->pr_mtx);
1539 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1541 prison_deref(pr, created
1543 : PD_DEREF | PD_LIST_XLOCKED);
1546 mtx_lock(&pr->pr_mtx);
1548 /* At this point, all valid parameters should have been noted. */
1549 TAILQ_FOREACH(opt, opts, link) {
1550 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1552 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1553 goto done_deref_locked;
1557 /* Set the parameters of the prison. */
1560 if (pr_flags & PR_IP4_USER) {
1561 pr->pr_flags |= PR_IP4;
1562 free(pr->pr_ip4, M_PRISON);
1566 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1568 if (tpr->pr_flags & PR_VNET) {
1573 if (prison_restrict_ip4(tpr, NULL)) {
1582 if (pr_flags & PR_IP6_USER) {
1583 pr->pr_flags |= PR_IP6;
1584 free(pr->pr_ip6, M_PRISON);
1588 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1590 if (tpr->pr_flags & PR_VNET) {
1595 if (prison_restrict_ip6(tpr, NULL)) {
1603 pr->pr_securelevel = slevel;
1604 /* Set all child jails to be at least this level. */
1605 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1606 if (tpr->pr_securelevel < slevel)
1607 tpr->pr_securelevel = slevel;
1610 pr->pr_childmax = childmax;
1611 /* Set all child jails to under this limit. */
1612 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1613 if (tpr->pr_childmax > childmax - level)
1614 tpr->pr_childmax = childmax > level
1615 ? childmax - level : 0;
1618 pr->pr_enforce_statfs = enforce;
1619 /* Pass this restriction on to the children. */
1620 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1621 if (tpr->pr_enforce_statfs < enforce)
1622 tpr->pr_enforce_statfs = enforce;
1625 pr->pr_devfs_rsnum = rsnum;
1626 /* Pass this restriction on to the children. */
1627 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1628 tpr->pr_devfs_rsnum = rsnum;
1630 if (namelc != NULL) {
1631 if (ppr == &prison0)
1632 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1634 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1635 ppr->pr_name, namelc);
1636 /* Change this component of child names. */
1637 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1638 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1639 strlen(tpr->pr_name + onamelen) + 1);
1640 bcopy(pr->pr_name, tpr->pr_name, namelen);
1644 /* Try to keep a real-rooted full pathname. */
1645 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1648 if (PR_HOST & ch_flags & ~pr_flags) {
1649 if (pr->pr_flags & PR_HOST) {
1651 * Copy the parent's host info. As with pr_ip4 above,
1652 * the lack of a lock on the parent is not a problem;
1653 * it is always set with allprison_lock at least
1654 * shared, and is held exclusively here.
1656 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1657 sizeof(pr->pr_hostname));
1658 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1659 sizeof(pr->pr_domainname));
1660 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1661 sizeof(pr->pr_hostuuid));
1662 pr->pr_hostid = pr->pr_parent->pr_hostid;
1664 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1665 /* Set this prison, and any descendants without PR_HOST. */
1667 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1669 strlcpy(pr->pr_domainname, domain,
1670 sizeof(pr->pr_domainname));
1672 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1674 pr->pr_hostid = hid;
1675 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1676 if (tpr->pr_flags & PR_HOST)
1680 strlcpy(tpr->pr_hostname,
1682 sizeof(tpr->pr_hostname));
1684 strlcpy(tpr->pr_domainname,
1686 sizeof(tpr->pr_domainname));
1688 strlcpy(tpr->pr_hostuuid,
1690 sizeof(tpr->pr_hostuuid));
1692 tpr->pr_hostid = hid;
1696 if ((tallow = ch_allow & ~pr_allow)) {
1697 /* Clear allow bits in all children. */
1698 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1699 tpr->pr_allow &= ~tallow;
1701 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1703 * Persistent prisons get an extra reference, and prisons losing their
1704 * persist flag lose that reference. Only do this for existing prisons
1705 * for now, so new ones will remain unseen until after the module
1706 * handlers have completed.
1708 born = pr->pr_uref == 0;
1709 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1710 if (pr_flags & PR_PERSIST) {
1718 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1719 mtx_unlock(&pr->pr_mtx);
1722 if (racct_enable && created)
1723 prison_racct_attach(pr);
1726 /* Locks may have prevented a complete restriction of child IP
1727 * addresses. If so, allocate some more memory and try again.
1732 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1733 mtx_lock(&pr->pr_mtx);
1735 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1737 if (tpr->pr_flags & PR_VNET) {
1742 if (prison_restrict_ip4(tpr, ip4)) {
1749 mtx_unlock(&pr->pr_mtx);
1755 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1756 mtx_lock(&pr->pr_mtx);
1758 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1760 if (tpr->pr_flags & PR_VNET) {
1765 if (prison_restrict_ip6(tpr, ip6)) {
1772 mtx_unlock(&pr->pr_mtx);
1776 /* Let the modules do their work. */
1777 sx_downgrade(&allprison_lock);
1779 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1781 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1782 prison_deref(pr, created
1784 : PD_DEREF | PD_LIST_SLOCKED);
1788 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1791 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1792 prison_deref(pr, created
1794 : PD_DEREF | PD_LIST_SLOCKED);
1798 /* Attach this process to the prison if requested. */
1799 if (flags & JAIL_ATTACH) {
1800 mtx_lock(&pr->pr_mtx);
1801 error = do_jail_attach(td, pr);
1803 vfs_opterror(opts, "attach failed");
1805 prison_deref(pr, PD_DEREF);
1811 if (racct_enable && !created) {
1812 if (!(flags & JAIL_ATTACH))
1813 sx_sunlock(&allprison_lock);
1814 prison_racct_modify(pr);
1815 if (!(flags & JAIL_ATTACH))
1816 sx_slock(&allprison_lock);
1820 td->td_retval[0] = pr->pr_id;
1823 * Now that it is all there, drop the temporary reference from existing
1824 * prisons. Or add a reference to newly created persistent prisons
1825 * (which was not done earlier so that the prison would not be publicly
1829 prison_deref(pr, (flags & JAIL_ATTACH)
1831 : PD_DEREF | PD_LIST_SLOCKED);
1833 if (pr_flags & PR_PERSIST) {
1834 mtx_lock(&pr->pr_mtx);
1837 mtx_unlock(&pr->pr_mtx);
1839 if (!(flags & JAIL_ATTACH))
1840 sx_sunlock(&allprison_lock);
1846 prison_deref(pr, created
1847 ? PD_LOCKED | PD_LIST_XLOCKED
1848 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1851 sx_xunlock(&allprison_lock);
1857 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1858 &errmsg_len) == 0 && errmsg_len > 0) {
1859 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1860 if (optuio->uio_segflg == UIO_SYSSPACE)
1862 optuio->uio_iov[errmsg_pos].iov_base,
1866 optuio->uio_iov[errmsg_pos].iov_base,
1872 free(ip4, M_PRISON);
1875 free(ip6, M_PRISON);
1878 free(g_path, M_TEMP);
1884 * struct jail_get_args {
1885 * struct iovec *iovp;
1886 * unsigned int iovcnt;
1891 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1896 /* Check that we have an even number of iovecs. */
1897 if (uap->iovcnt & 1)
1900 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1903 error = kern_jail_get(td, auio, uap->flags);
1905 error = copyout(auio->uio_iov, uap->iovp,
1906 uap->iovcnt * sizeof (struct iovec));
1912 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1914 struct bool_flags *bf;
1915 struct jailsys_flags *jsf;
1916 struct prison *pr, *mypr;
1918 struct vfsoptlist *opts;
1919 char *errmsg, *name;
1920 int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1923 if (flags & ~JAIL_GET_MASK)
1926 /* Get the parameter list. */
1927 error = vfs_buildopts(optuio, &opts);
1930 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1931 mypr = td->td_ucred->cr_prison;
1934 * Find the prison specified by one of: lastjid, jid, name.
1936 sx_slock(&allprison_lock);
1937 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1939 TAILQ_FOREACH(pr, &allprison, pr_list) {
1940 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1941 mtx_lock(&pr->pr_mtx);
1942 if (pr->pr_ref > 0 &&
1943 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1945 mtx_unlock(&pr->pr_mtx);
1951 vfs_opterror(opts, "no jail after %d", jid);
1952 goto done_unlock_list;
1953 } else if (error != ENOENT)
1954 goto done_unlock_list;
1956 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1959 pr = prison_find_child(mypr, jid);
1961 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1962 mtx_unlock(&pr->pr_mtx);
1964 vfs_opterror(opts, "jail %d is dying",
1966 goto done_unlock_list;
1971 vfs_opterror(opts, "jail %d not found", jid);
1972 goto done_unlock_list;
1974 } else if (error != ENOENT)
1975 goto done_unlock_list;
1977 error = vfs_getopt(opts, "name", (void **)&name, &len);
1979 if (len == 0 || name[len - 1] != '\0') {
1981 goto done_unlock_list;
1983 pr = prison_find_name(mypr, name);
1985 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1986 mtx_unlock(&pr->pr_mtx);
1988 vfs_opterror(opts, "jail \"%s\" is dying",
1990 goto done_unlock_list;
1995 vfs_opterror(opts, "jail \"%s\" not found", name);
1996 goto done_unlock_list;
1997 } else if (error != ENOENT)
1998 goto done_unlock_list;
2000 vfs_opterror(opts, "no jail specified");
2002 goto done_unlock_list;
2005 /* Get the parameters of the prison. */
2008 td->td_retval[0] = pr->pr_id;
2009 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2010 if (error != 0 && error != ENOENT)
2012 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2013 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2014 if (error != 0 && error != ENOENT)
2016 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2017 if (error != 0 && error != ENOENT)
2019 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2020 sizeof(pr->pr_cpuset->cs_id));
2021 if (error != 0 && error != ENOENT)
2023 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2024 if (error != 0 && error != ENOENT)
2027 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2028 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2029 if (error != 0 && error != ENOENT)
2033 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2034 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2035 if (error != 0 && error != ENOENT)
2038 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2039 sizeof(pr->pr_securelevel));
2040 if (error != 0 && error != ENOENT)
2042 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2043 sizeof(pr->pr_childcount));
2044 if (error != 0 && error != ENOENT)
2046 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2047 sizeof(pr->pr_childmax));
2048 if (error != 0 && error != ENOENT)
2050 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2051 if (error != 0 && error != ENOENT)
2053 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2054 if (error != 0 && error != ENOENT)
2056 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2057 if (error != 0 && error != ENOENT)
2059 #ifdef COMPAT_FREEBSD32
2060 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2061 uint32_t hid32 = pr->pr_hostid;
2063 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2066 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2067 sizeof(pr->pr_hostid));
2068 if (error != 0 && error != ENOENT)
2070 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2071 sizeof(pr->pr_enforce_statfs));
2072 if (error != 0 && error != ENOENT)
2074 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2075 sizeof(pr->pr_devfs_rsnum));
2076 if (error != 0 && error != ENOENT)
2078 for (bf = pr_flag_bool;
2079 bf < pr_flag_bool + nitems(pr_flag_bool);
2081 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2082 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2083 if (error != 0 && error != ENOENT)
2086 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2087 if (error != 0 && error != ENOENT)
2090 for (jsf = pr_flag_jailsys;
2091 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2093 f = pr->pr_flags & (jsf->disable | jsf->new);
2094 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2095 : (f == jsf->new) ? JAIL_SYS_NEW
2097 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2098 if (error != 0 && error != ENOENT)
2101 for (bf = pr_flag_allow;
2102 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2104 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2105 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2106 if (error != 0 && error != ENOENT)
2109 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2110 if (error != 0 && error != ENOENT)
2113 i = (pr->pr_uref == 0);
2114 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2115 if (error != 0 && error != ENOENT)
2118 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2119 if (error != 0 && error != ENOENT)
2121 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2122 sizeof(pr->pr_osreldate));
2123 if (error != 0 && error != ENOENT)
2125 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2126 if (error != 0 && error != ENOENT)
2129 /* Get the module parameters. */
2130 mtx_unlock(&pr->pr_mtx);
2132 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2135 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2137 /* By now, all parameters should have been noted. */
2138 TAILQ_FOREACH(opt, opts, link) {
2139 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2141 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2146 /* Write the fetched parameters back to userspace. */
2148 TAILQ_FOREACH(opt, opts, link) {
2149 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2150 pos = 2 * opt->pos + 1;
2151 optuio->uio_iov[pos].iov_len = opt->len;
2152 if (opt->value != NULL) {
2153 if (optuio->uio_segflg == UIO_SYSSPACE) {
2155 optuio->uio_iov[pos].iov_base,
2158 error = copyout(opt->value,
2159 optuio->uio_iov[pos].iov_base,
2170 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2174 sx_sunlock(&allprison_lock);
2176 if (error && errmsg_pos >= 0) {
2177 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2178 errmsg_pos = 2 * errmsg_pos + 1;
2179 if (errmsg_len > 0) {
2180 if (optuio->uio_segflg == UIO_SYSSPACE)
2182 optuio->uio_iov[errmsg_pos].iov_base,
2186 optuio->uio_iov[errmsg_pos].iov_base,
2195 * struct jail_remove_args {
2200 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2202 struct prison *pr, *cpr, *lpr, *tpr;
2205 error = priv_check(td, PRIV_JAIL_REMOVE);
2209 sx_xlock(&allprison_lock);
2210 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2212 sx_xunlock(&allprison_lock);
2216 /* Remove all descendants of this prison, then remove this prison. */
2218 if (!LIST_EMPTY(&pr->pr_children)) {
2219 mtx_unlock(&pr->pr_mtx);
2221 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2222 mtx_lock(&cpr->pr_mtx);
2223 if (cpr->pr_ref > 0) {
2227 /* Already removed - do not do it again. */
2230 mtx_unlock(&cpr->pr_mtx);
2232 mtx_lock(&lpr->pr_mtx);
2233 prison_remove_one(lpr);
2234 sx_xlock(&allprison_lock);
2239 mtx_lock(&lpr->pr_mtx);
2240 prison_remove_one(lpr);
2241 sx_xlock(&allprison_lock);
2243 mtx_lock(&pr->pr_mtx);
2245 prison_remove_one(pr);
2250 prison_remove_one(struct prison *pr)
2255 /* If the prison was persistent, it is not anymore. */
2257 if (pr->pr_flags & PR_PERSIST) {
2260 pr->pr_flags &= ~PR_PERSIST;
2264 * jail_remove added a reference. If that's the only one, remove
2267 KASSERT(pr->pr_ref > 0,
2268 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2269 if (pr->pr_ref == 1) {
2271 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2275 mtx_unlock(&pr->pr_mtx);
2276 sx_xunlock(&allprison_lock);
2278 * Kill all processes unfortunate enough to be attached to this prison.
2280 sx_slock(&allproc_lock);
2281 FOREACH_PROC_IN_SYSTEM(p) {
2283 if (p->p_state != PRS_NEW && p->p_ucred &&
2284 p->p_ucred->cr_prison == pr)
2285 kern_psignal(p, SIGKILL);
2288 sx_sunlock(&allproc_lock);
2289 /* Remove the temporary reference added by jail_remove. */
2290 prison_deref(pr, deuref | PD_DEREF);
2294 * struct jail_attach_args {
2299 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2304 error = priv_check(td, PRIV_JAIL_ATTACH);
2309 * Start with exclusive hold on allprison_lock to ensure that a possible
2310 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2311 * But then immediately downgrade it since we don't need to stop
2314 sx_xlock(&allprison_lock);
2315 sx_downgrade(&allprison_lock);
2316 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2318 sx_sunlock(&allprison_lock);
2323 * Do not allow a process to attach to a prison that is not
2324 * considered to be "alive".
2326 if (pr->pr_uref == 0) {
2327 mtx_unlock(&pr->pr_mtx);
2328 sx_sunlock(&allprison_lock);
2332 return (do_jail_attach(td, pr));
2336 do_jail_attach(struct thread *td, struct prison *pr)
2339 struct ucred *newcred, *oldcred;
2343 * XXX: Note that there is a slight race here if two threads
2344 * in the same privileged process attempt to attach to two
2345 * different jails at the same time. It is important for
2346 * user processes not to do this, or they might end up with
2347 * a process root from one prison, but attached to the jail
2352 mtx_unlock(&pr->pr_mtx);
2354 /* Let modules do whatever they need to prepare for attaching. */
2355 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2357 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2360 sx_sunlock(&allprison_lock);
2363 * Reparent the newly attached process to this jail.
2366 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2370 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2371 if ((error = change_dir(pr->pr_root, td)) != 0)
2374 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2377 VOP_UNLOCK(pr->pr_root);
2378 if ((error = pwd_chroot(td, pr->pr_root)))
2383 oldcred = crcopysafe(p, newcred);
2384 newcred->cr_prison = pr;
2385 proc_set_cred(p, newcred);
2388 racct_proc_ucred_changed(p, oldcred, newcred);
2393 rctl_proc_ucred_changed(p, newcred);
2396 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2401 VOP_UNLOCK(pr->pr_root);
2403 /* Tell modules this thread is still in its old jail after all. */
2404 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2405 prison_deref(pr, PD_DEREF | PD_DEUREF);
2410 * Returns a locked prison instance, or NULL on failure.
2413 prison_find(int prid)
2417 sx_assert(&allprison_lock, SX_LOCKED);
2418 TAILQ_FOREACH(pr, &allprison, pr_list) {
2419 if (pr->pr_id == prid) {
2420 mtx_lock(&pr->pr_mtx);
2423 mtx_unlock(&pr->pr_mtx);
2430 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2433 prison_find_child(struct prison *mypr, int prid)
2438 sx_assert(&allprison_lock, SX_LOCKED);
2439 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2440 if (pr->pr_id == prid) {
2441 mtx_lock(&pr->pr_mtx);
2444 mtx_unlock(&pr->pr_mtx);
2451 * Look for the name relative to mypr. Returns a locked prison or NULL.
2454 prison_find_name(struct prison *mypr, const char *name)
2456 struct prison *pr, *deadpr;
2460 sx_assert(&allprison_lock, SX_LOCKED);
2461 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2464 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2465 if (!strcmp(pr->pr_name + mylen, name)) {
2466 mtx_lock(&pr->pr_mtx);
2467 if (pr->pr_ref > 0) {
2468 if (pr->pr_uref > 0)
2472 mtx_unlock(&pr->pr_mtx);
2475 /* There was no valid prison - perhaps there was a dying one. */
2476 if (deadpr != NULL) {
2477 mtx_lock(&deadpr->pr_mtx);
2478 if (deadpr->pr_ref == 0) {
2479 mtx_unlock(&deadpr->pr_mtx);
2487 * See if a prison has the specific flag set.
2490 prison_flag(struct ucred *cred, unsigned flag)
2493 /* This is an atomic read, so no locking is necessary. */
2494 return (cred->cr_prison->pr_flags & flag);
2498 prison_allow(struct ucred *cred, unsigned flag)
2501 /* This is an atomic read, so no locking is necessary. */
2502 return (cred->cr_prison->pr_allow & flag);
2506 * Remove a prison reference. If that was the last reference, remove the
2507 * prison itself - but not in this context in case there are locks held.
2510 prison_free_locked(struct prison *pr)
2514 mtx_assert(&pr->pr_mtx, MA_OWNED);
2516 mtx_unlock(&pr->pr_mtx);
2518 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2522 prison_free(struct prison *pr)
2525 mtx_lock(&pr->pr_mtx);
2526 prison_free_locked(pr);
2530 * Complete a call to either prison_free or prison_proc_free.
2533 prison_complete(void *context, int pending)
2535 struct prison *pr = context;
2537 sx_xlock(&allprison_lock);
2538 mtx_lock(&pr->pr_mtx);
2539 prison_deref(pr, pr->pr_uref
2540 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2541 : PD_LOCKED | PD_LIST_XLOCKED);
2545 * Remove a prison reference (usually). This internal version assumes no
2546 * mutexes are held, except perhaps the prison itself. If there are no more
2547 * references, release and delist the prison. On completion, the prison lock
2548 * and the allprison lock are both unlocked.
2551 prison_deref(struct prison *pr, int flags)
2553 struct prison *ppr, *tpr;
2556 if (!(flags & PD_LOCKED))
2557 mtx_lock(&pr->pr_mtx);
2559 if (flags & PD_DEUREF) {
2560 KASSERT(pr->pr_uref > 0,
2561 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2564 lasturef = pr->pr_uref == 0;
2567 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2570 if (flags & PD_DEREF) {
2571 KASSERT(pr->pr_ref > 0,
2572 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2577 mtx_unlock(&pr->pr_mtx);
2580 * Tell the modules if the last user reference was removed
2581 * (even it sticks around in dying state).
2584 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2585 sx_xlock(&allprison_lock);
2586 flags |= PD_LIST_XLOCKED;
2588 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2589 mtx_lock(&pr->pr_mtx);
2591 mtx_unlock(&pr->pr_mtx);
2594 /* If the prison still has references, nothing else to do. */
2596 if (flags & PD_LIST_SLOCKED)
2597 sx_sunlock(&allprison_lock);
2598 else if (flags & PD_LIST_XLOCKED)
2599 sx_xunlock(&allprison_lock);
2603 if (flags & PD_LIST_SLOCKED) {
2604 if (!sx_try_upgrade(&allprison_lock)) {
2605 sx_sunlock(&allprison_lock);
2606 sx_xlock(&allprison_lock);
2608 } else if (!(flags & PD_LIST_XLOCKED))
2609 sx_xlock(&allprison_lock);
2611 TAILQ_REMOVE(&allprison, pr, pr_list);
2612 LIST_REMOVE(pr, pr_sibling);
2613 ppr = pr->pr_parent;
2614 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2615 tpr->pr_childcount--;
2616 sx_xunlock(&allprison_lock);
2619 if (pr->pr_vnet != ppr->pr_vnet)
2620 vnet_destroy(pr->pr_vnet);
2622 if (pr->pr_root != NULL)
2624 mtx_destroy(&pr->pr_mtx);
2626 free(pr->pr_ip4, M_PRISON);
2629 free(pr->pr_ip6, M_PRISON);
2631 if (pr->pr_cpuset != NULL)
2632 cpuset_rel(pr->pr_cpuset);
2636 prison_racct_detach(pr);
2640 /* Removing a prison frees a reference on its parent. */
2642 mtx_lock(&pr->pr_mtx);
2643 flags = PD_DEREF | PD_DEUREF;
2648 prison_hold_locked(struct prison *pr)
2651 mtx_assert(&pr->pr_mtx, MA_OWNED);
2652 KASSERT(pr->pr_ref > 0,
2653 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2658 prison_hold(struct prison *pr)
2661 mtx_lock(&pr->pr_mtx);
2662 prison_hold_locked(pr);
2663 mtx_unlock(&pr->pr_mtx);
2667 prison_proc_hold(struct prison *pr)
2670 mtx_lock(&pr->pr_mtx);
2671 KASSERT(pr->pr_uref > 0,
2672 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2674 mtx_unlock(&pr->pr_mtx);
2678 prison_proc_free(struct prison *pr)
2681 mtx_lock(&pr->pr_mtx);
2682 KASSERT(pr->pr_uref > 0,
2683 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2684 if (pr->pr_uref > 1)
2688 * Don't remove the last user reference in this context, which
2689 * is expected to be a process that is not only locked, but
2693 mtx_unlock(&pr->pr_mtx);
2694 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2697 mtx_unlock(&pr->pr_mtx);
2701 * Check if a jail supports the given address family.
2703 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2707 prison_check_af(struct ucred *cred, int af)
2712 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2714 pr = cred->cr_prison;
2716 /* Prisons with their own network stack are not limited. */
2717 if (prison_owns_vnet(cred))
2726 if (pr->pr_flags & PR_IP4)
2728 mtx_lock(&pr->pr_mtx);
2729 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2730 error = EAFNOSUPPORT;
2731 mtx_unlock(&pr->pr_mtx);
2737 if (pr->pr_flags & PR_IP6)
2739 mtx_lock(&pr->pr_mtx);
2740 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2741 error = EAFNOSUPPORT;
2742 mtx_unlock(&pr->pr_mtx);
2750 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2751 error = EAFNOSUPPORT;
2757 * Check if given address belongs to the jail referenced by cred (wrapper to
2758 * prison_check_ip[46]).
2760 * Returns 0 if jail doesn't restrict the address family or if address belongs
2761 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2762 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
2765 prison_if(struct ucred *cred, const struct sockaddr *sa)
2768 const struct sockaddr_in *sai;
2771 const struct sockaddr_in6 *sai6;
2775 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2776 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2779 if (prison_owns_vnet(cred))
2784 switch (sa->sa_family)
2788 sai = (const struct sockaddr_in *)sa;
2789 error = prison_check_ip4(cred, &sai->sin_addr);
2794 sai6 = (const struct sockaddr_in6 *)sa;
2795 error = prison_check_ip6(cred, &sai6->sin6_addr);
2799 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2800 error = EAFNOSUPPORT;
2806 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2809 prison_check(struct ucred *cred1, struct ucred *cred2)
2812 return ((cred1->cr_prison == cred2->cr_prison ||
2813 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2817 * Return 1 if p2 is a child of p1, otherwise 0.
2820 prison_ischild(struct prison *pr1, struct prison *pr2)
2823 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2830 * Return 1 if the passed credential is in a jail and that jail does not
2831 * have its own virtual network stack, otherwise 0.
2834 jailed_without_vnet(struct ucred *cred)
2840 if (prison_owns_vnet(cred))
2848 * Return the correct hostname (domainname, et al) for the passed credential.
2851 getcredhostname(struct ucred *cred, char *buf, size_t size)
2856 * A NULL credential can be used to shortcut to the physical
2857 * system's hostname.
2859 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2860 mtx_lock(&pr->pr_mtx);
2861 strlcpy(buf, pr->pr_hostname, size);
2862 mtx_unlock(&pr->pr_mtx);
2866 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2869 mtx_lock(&cred->cr_prison->pr_mtx);
2870 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2871 mtx_unlock(&cred->cr_prison->pr_mtx);
2875 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2878 mtx_lock(&cred->cr_prison->pr_mtx);
2879 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2880 mtx_unlock(&cred->cr_prison->pr_mtx);
2884 getcredhostid(struct ucred *cred, unsigned long *hostid)
2887 mtx_lock(&cred->cr_prison->pr_mtx);
2888 *hostid = cred->cr_prison->pr_hostid;
2889 mtx_unlock(&cred->cr_prison->pr_mtx);
2894 * Determine whether the prison represented by cred owns
2895 * its vnet rather than having it inherited.
2897 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2900 prison_owns_vnet(struct ucred *cred)
2904 * vnets cannot be added/removed after jail creation,
2905 * so no need to lock here.
2907 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2912 * Determine whether the subject represented by cred can "see"
2913 * status of a mount point.
2914 * Returns: 0 for permitted, ENOENT otherwise.
2915 * XXX: This function should be called cr_canseemount() and should be
2916 * placed in kern_prot.c.
2919 prison_canseemount(struct ucred *cred, struct mount *mp)
2925 pr = cred->cr_prison;
2926 if (pr->pr_enforce_statfs == 0)
2928 if (pr->pr_root->v_mount == mp)
2930 if (pr->pr_enforce_statfs == 2)
2933 * If jail's chroot directory is set to "/" we should be able to see
2934 * all mount-points from inside a jail.
2935 * This is ugly check, but this is the only situation when jail's
2936 * directory ends with '/'.
2938 if (strcmp(pr->pr_path, "/") == 0)
2940 len = strlen(pr->pr_path);
2942 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2945 * Be sure that we don't have situation where jail's root directory
2946 * is "/some/path" and mount point is "/some/pathpath".
2948 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2954 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2956 char jpath[MAXPATHLEN];
2960 pr = cred->cr_prison;
2961 if (pr->pr_enforce_statfs == 0)
2963 if (prison_canseemount(cred, mp) != 0) {
2964 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2965 strlcpy(sp->f_mntonname, "[restricted]",
2966 sizeof(sp->f_mntonname));
2969 if (pr->pr_root->v_mount == mp) {
2971 * Clear current buffer data, so we are sure nothing from
2972 * the valid path left there.
2974 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2975 *sp->f_mntonname = '/';
2979 * If jail's chroot directory is set to "/" we should be able to see
2980 * all mount-points from inside a jail.
2982 if (strcmp(pr->pr_path, "/") == 0)
2984 len = strlen(pr->pr_path);
2985 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
2987 * Clear current buffer data, so we are sure nothing from
2988 * the valid path left there.
2990 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2991 if (*jpath == '\0') {
2992 /* Should never happen. */
2993 *sp->f_mntonname = '/';
2995 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3000 * Check with permission for a specific privilege is granted within jail. We
3001 * have a specific list of accepted privileges; the rest are denied.
3004 prison_priv_check(struct ucred *cred, int priv)
3008 * Some policies have custom handlers. This routine should not be
3009 * called for them. See priv_check_cred().
3012 case PRIV_VFS_GENERATION:
3013 KASSERT(0, ("prison_priv_check instead of a custom handler "
3014 "called for %d\n", priv));
3022 * Privileges specific to prisons with a virtual network stack.
3023 * There might be a duplicate entry here in case the privilege
3024 * is only granted conditionally in the legacy jail case.
3029 * NFS-specific privileges.
3031 case PRIV_NFS_DAEMON:
3032 case PRIV_NFS_LOCKD:
3035 * Network stack privileges.
3037 case PRIV_NET_BRIDGE:
3040 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3041 case PRIV_NET_ROUTE:
3043 case PRIV_NET_SETIFMTU:
3044 case PRIV_NET_SETIFFLAGS:
3045 case PRIV_NET_SETIFCAP:
3046 case PRIV_NET_SETIFDESCR:
3047 case PRIV_NET_SETIFNAME :
3048 case PRIV_NET_SETIFMETRIC:
3049 case PRIV_NET_SETIFPHYS:
3050 case PRIV_NET_SETIFMAC:
3051 case PRIV_NET_SETLANPCP:
3052 case PRIV_NET_ADDMULTI:
3053 case PRIV_NET_DELMULTI:
3054 case PRIV_NET_HWIOCTL:
3055 case PRIV_NET_SETLLADDR:
3056 case PRIV_NET_ADDIFGROUP:
3057 case PRIV_NET_DELIFGROUP:
3058 case PRIV_NET_IFCREATE:
3059 case PRIV_NET_IFDESTROY:
3060 case PRIV_NET_ADDIFADDR:
3061 case PRIV_NET_DELIFADDR:
3064 case PRIV_NET_SETIFVNET:
3065 case PRIV_NET_SETIFFIB:
3068 * 802.11-related privileges.
3070 case PRIV_NET80211_GETKEY:
3072 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3079 case PRIV_NETATM_CFG:
3080 case PRIV_NETATM_ADD:
3081 case PRIV_NETATM_DEL:
3082 case PRIV_NETATM_SET:
3085 * Bluetooth privileges.
3087 case PRIV_NETBLUETOOTH_RAW:
3091 * Netgraph and netgraph module privileges.
3093 case PRIV_NETGRAPH_CONTROL:
3095 case PRIV_NETGRAPH_TTY:
3099 * IPv4 and IPv6 privileges.
3101 case PRIV_NETINET_IPFW:
3102 case PRIV_NETINET_DIVERT:
3103 case PRIV_NETINET_PF:
3104 case PRIV_NETINET_DUMMYNET:
3105 case PRIV_NETINET_CARP:
3106 case PRIV_NETINET_MROUTE:
3107 case PRIV_NETINET_RAW:
3108 case PRIV_NETINET_ADDRCTRL6:
3109 case PRIV_NETINET_ND6:
3110 case PRIV_NETINET_SCOPE6:
3111 case PRIV_NETINET_ALIFETIME6:
3112 case PRIV_NETINET_IPSEC:
3113 case PRIV_NETINET_BINDANY:
3128 * No default: or deny here.
3129 * In case of no permit fall through to next switch().
3131 if (cred->cr_prison->pr_flags & PR_VNET)
3139 * Allow ktrace privileges for root in jail.
3145 * Allow jailed processes to configure audit identity and
3146 * submit audit records (login, etc). In the future we may
3147 * want to further refine the relationship between audit and
3150 case PRIV_AUDIT_GETAUDIT:
3151 case PRIV_AUDIT_SETAUDIT:
3152 case PRIV_AUDIT_SUBMIT:
3156 * Allow jailed processes to manipulate process UNIX
3157 * credentials in any way they see fit.
3159 case PRIV_CRED_SETUID:
3160 case PRIV_CRED_SETEUID:
3161 case PRIV_CRED_SETGID:
3162 case PRIV_CRED_SETEGID:
3163 case PRIV_CRED_SETGROUPS:
3164 case PRIV_CRED_SETREUID:
3165 case PRIV_CRED_SETREGID:
3166 case PRIV_CRED_SETRESUID:
3167 case PRIV_CRED_SETRESGID:
3170 * Jail implements visibility constraints already, so allow
3171 * jailed root to override uid/gid-based constraints.
3173 case PRIV_SEEOTHERGIDS:
3174 case PRIV_SEEOTHERUIDS:
3177 * Jail implements inter-process debugging limits already, so
3178 * allow jailed root various debugging privileges.
3180 case PRIV_DEBUG_DIFFCRED:
3181 case PRIV_DEBUG_SUGID:
3182 case PRIV_DEBUG_UNPRIV:
3185 * Allow jail to set various resource limits and login
3186 * properties, and for now, exceed process resource limits.
3188 case PRIV_PROC_LIMIT:
3189 case PRIV_PROC_SETLOGIN:
3190 case PRIV_PROC_SETRLIMIT:
3193 * System V and POSIX IPC privileges are granted in jail.
3196 case PRIV_IPC_WRITE:
3197 case PRIV_IPC_ADMIN:
3198 case PRIV_IPC_MSGSIZE:
3202 * Jail operations within a jail work on child jails.
3204 case PRIV_JAIL_ATTACH:
3206 case PRIV_JAIL_REMOVE:
3209 * Jail implements its own inter-process limits, so allow
3210 * root processes in jail to change scheduling on other
3211 * processes in the same jail. Likewise for signalling.
3213 case PRIV_SCHED_DIFFCRED:
3214 case PRIV_SCHED_CPUSET:
3215 case PRIV_SIGNAL_DIFFCRED:
3216 case PRIV_SIGNAL_SUGID:
3219 * Allow jailed processes to write to sysctls marked as jail
3222 case PRIV_SYSCTL_WRITEJAIL:
3225 * Allow root in jail to manage a variety of quota
3226 * properties. These should likely be conditional on a
3227 * configuration option.
3229 case PRIV_VFS_GETQUOTA:
3230 case PRIV_VFS_SETQUOTA:
3233 * Since Jail relies on chroot() to implement file system
3234 * protections, grant many VFS privileges to root in jail.
3235 * Be careful to exclude mount-related and NFS-related
3239 case PRIV_VFS_WRITE:
3240 case PRIV_VFS_ADMIN:
3242 case PRIV_VFS_LOOKUP:
3243 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3244 case PRIV_VFS_CHFLAGS_DEV:
3245 case PRIV_VFS_CHOWN:
3246 case PRIV_VFS_CHROOT:
3247 case PRIV_VFS_RETAINSUGID:
3248 case PRIV_VFS_FCHROOT:
3250 case PRIV_VFS_SETGID:
3252 case PRIV_VFS_STICKYFILE:
3255 * As in the non-jail case, non-root users are expected to be
3256 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3257 * exists in the jail and they have permission to access it).
3259 case PRIV_KMEM_READ:
3263 * Depending on the global setting, allow privilege of
3264 * setting system flags.
3266 case PRIV_VFS_SYSFLAGS:
3267 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3273 * Depending on the global setting, allow privilege of
3274 * mounting/unmounting file systems.
3276 case PRIV_VFS_MOUNT:
3277 case PRIV_VFS_UNMOUNT:
3278 case PRIV_VFS_MOUNT_NONUSER:
3279 case PRIV_VFS_MOUNT_OWNER:
3280 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3281 cred->cr_prison->pr_enforce_statfs < 2)
3287 * Conditionnaly allow locking (unlocking) physical pages
3291 case PRIV_VM_MUNLOCK:
3292 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3298 * Conditionally allow jailed root to bind reserved ports.
3300 case PRIV_NETINET_RESERVEDPORT:
3301 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3307 * Allow jailed root to reuse in-use ports.
3309 case PRIV_NETINET_REUSEPORT:
3313 * Allow jailed root to set certain IPv4/6 (option) headers.
3315 case PRIV_NETINET_SETHDROPTS:
3319 * Conditionally allow creating raw sockets in jail.
3321 case PRIV_NETINET_RAW:
3322 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3328 * Since jail implements its own visibility limits on netstat
3329 * sysctls, allow getcred. This allows identd to work in
3332 case PRIV_NETINET_GETCRED:
3336 * Allow jailed root to set loginclass.
3338 case PRIV_PROC_SETLOGINCLASS:
3342 * Do not allow a process inside a jail to read the kernel
3343 * message buffer unless explicitly permitted.
3346 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3352 * In all remaining cases, deny the privilege request. This
3353 * includes almost all network privileges, many system
3354 * configuration privileges.
3361 * Return the part of pr2's name that is relative to pr1, or the whole name
3362 * if it does not directly follow.
3366 prison_name(struct prison *pr1, struct prison *pr2)
3370 /* Jails see themselves as "0" (if they see themselves at all). */
3373 name = pr2->pr_name;
3374 if (prison_ischild(pr1, pr2)) {
3376 * pr1 isn't locked (and allprison_lock may not be either)
3377 * so its length can't be counted on. But the number of dots
3378 * can be counted on - and counted.
3380 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3381 name = strchr(name, '.') + 1;
3387 * Return the part of pr2's path that is relative to pr1, or the whole path
3388 * if it does not directly follow.
3391 prison_path(struct prison *pr1, struct prison *pr2)
3393 char *path1, *path2;
3396 path1 = pr1->pr_path;
3397 path2 = pr2->pr_path;
3398 if (!strcmp(path1, "/"))
3400 len1 = strlen(path1);
3401 if (strncmp(path1, path2, len1))
3403 if (path2[len1] == '\0')
3405 if (path2[len1] == '/')
3406 return (path2 + len1);
3411 * Jail-related sysctls.
3413 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3417 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3420 struct prison *pr, *cpr;
3422 struct in_addr *ip4 = NULL;
3426 struct in6_addr *ip6 = NULL;
3431 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3432 pr = req->td->td_ucred->cr_prison;
3434 sx_slock(&allprison_lock);
3435 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3436 #if defined(INET) || defined(INET6)
3439 mtx_lock(&cpr->pr_mtx);
3441 if (cpr->pr_ip4s > 0) {
3442 if (ip4s < cpr->pr_ip4s) {
3443 ip4s = cpr->pr_ip4s;
3444 mtx_unlock(&cpr->pr_mtx);
3445 ip4 = realloc(ip4, ip4s *
3446 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3449 bcopy(cpr->pr_ip4, ip4,
3450 cpr->pr_ip4s * sizeof(struct in_addr));
3454 if (cpr->pr_ip6s > 0) {
3455 if (ip6s < cpr->pr_ip6s) {
3456 ip6s = cpr->pr_ip6s;
3457 mtx_unlock(&cpr->pr_mtx);
3458 ip6 = realloc(ip6, ip6s *
3459 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3462 bcopy(cpr->pr_ip6, ip6,
3463 cpr->pr_ip6s * sizeof(struct in6_addr));
3466 if (cpr->pr_ref == 0) {
3467 mtx_unlock(&cpr->pr_mtx);
3470 bzero(xp, sizeof(*xp));
3471 xp->pr_version = XPRISON_VERSION;
3472 xp->pr_id = cpr->pr_id;
3473 xp->pr_state = cpr->pr_uref > 0
3474 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3475 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3476 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3477 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3479 xp->pr_ip4s = cpr->pr_ip4s;
3482 xp->pr_ip6s = cpr->pr_ip6s;
3484 mtx_unlock(&cpr->pr_mtx);
3485 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3489 if (xp->pr_ip4s > 0) {
3490 error = SYSCTL_OUT(req, ip4,
3491 xp->pr_ip4s * sizeof(struct in_addr));
3497 if (xp->pr_ip6s > 0) {
3498 error = SYSCTL_OUT(req, ip6,
3499 xp->pr_ip6s * sizeof(struct in6_addr));
3505 sx_sunlock(&allprison_lock);
3516 SYSCTL_OID(_security_jail, OID_AUTO, list,
3517 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3518 sysctl_jail_list, "S", "List of active jails");
3521 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3525 injail = jailed(req->td->td_ucred);
3526 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3531 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3532 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3533 sysctl_jail_jailed, "I", "Process in jail?");
3536 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3538 int error, havevnet;
3540 struct ucred *cred = req->td->td_ucred;
3542 havevnet = jailed(cred) && prison_owns_vnet(cred);
3546 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3551 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3552 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3553 sysctl_jail_vnet, "I", "Jail owns vnet?");
3555 #if defined(INET) || defined(INET6)
3556 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3557 &jail_max_af_ips, 0,
3558 "Number of IP addresses a jail may have at most per address family (deprecated)");
3562 * Default parameters for jail(2) compatibility. For historical reasons,
3563 * the sysctl names have varying similarity to the parameter names. Prisons
3564 * just see their own parameters, and can't change them.
3567 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3570 int allow, error, i;
3572 pr = req->td->td_ucred->cr_prison;
3573 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3575 /* Get the current flag value, and convert it to a boolean. */
3576 i = (allow & arg2) ? 1 : 0;
3579 error = sysctl_handle_int(oidp, &i, 0, req);
3580 if (error || !req->newptr)
3586 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3589 mtx_lock(&prison0.pr_mtx);
3590 jail_default_allow = (jail_default_allow & ~arg2) | i;
3591 mtx_unlock(&prison0.pr_mtx);
3595 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3596 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3597 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3598 "Processes in jail can set their hostnames (deprecated)");
3599 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3600 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3601 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3602 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3603 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3604 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3605 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3606 "Processes in jail can use System V IPC primitives (deprecated)");
3607 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3608 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3609 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3610 "Prison root can create raw sockets (deprecated)");
3611 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3612 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3613 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3614 "Processes in jail can alter system file flags (deprecated)");
3615 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3616 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3617 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3618 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3621 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3626 pr = req->td->td_ucred->cr_prison;
3627 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3628 error = sysctl_handle_int(oidp, &level, 0, req);
3629 if (error || !req->newptr)
3631 *(int *)arg1 = level;
3635 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3636 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3637 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3638 sysctl_jail_default_level, "I",
3639 "Processes in jail cannot see all mounted file systems (deprecated)");
3641 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3642 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3643 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3644 sysctl_jail_default_level, "I",
3645 "Ruleset for the devfs filesystem in jail (deprecated)");
3648 * Nodes to describe jail parameters. Maximum length of string parameters
3649 * is returned in the string itself, and the other parameters exist merely
3650 * to make themselves and their types known.
3652 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3656 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3663 switch (oidp->oid_kind & CTLTYPE)
3669 if (!(req->flags & SCTL_MASK32))
3671 return (SYSCTL_OUT(req, &l, sizeof(l)));
3675 return (SYSCTL_OUT(req, &i, sizeof(i)));
3676 case CTLTYPE_STRING:
3677 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3679 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3680 case CTLTYPE_STRUCT:
3682 return (SYSCTL_OUT(req, &s, sizeof(s)));
3688 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3689 * jail creation time but cannot be changed in an existing jail.
3691 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3692 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3693 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3694 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3695 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3696 "I", "Jail secure level");
3697 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3698 "Jail value for kern.osreldate and uname -K");
3699 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3700 "Jail value for kern.osrelease and uname -r");
3701 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3702 "I", "Jail cannot see all mounted file systems");
3703 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3704 "I", "Ruleset for in-jail devfs mounts");
3705 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3706 "B", "Jail persistence");
3708 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3709 "E,jailsys", "Virtual network stack");
3711 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3712 "B", "Jail is in the process of shutting down");
3714 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3715 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3716 "I", "Current number of child jails");
3717 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3718 "I", "Maximum number of child jails");
3720 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3721 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3723 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3724 "Jail NIS domainname");
3725 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3727 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3728 "LU", "Jail host ID");
3730 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3731 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3734 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3735 "Jail IPv4 address virtualization");
3736 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3737 "S,in_addr,a", "Jail IPv4 addresses");
3738 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3739 "B", "Do (not) use IPv4 source address selection rather than the "
3740 "primary jail IPv4 address.");
3743 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3744 "Jail IPv6 address virtualization");
3745 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3746 "S,in6_addr,a", "Jail IPv6 addresses");
3747 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3748 "B", "Do (not) use IPv6 source address selection rather than the "
3749 "primary jail IPv6 address.");
3752 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3753 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3754 "B", "Jail may set hostname");
3755 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3756 "B", "Jail may use SYSV IPC");
3757 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3758 "B", "Jail may create raw sockets");
3759 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3760 "B", "Jail may alter system file flags");
3761 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3762 "B", "Jail may set file quotas");
3763 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3764 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3765 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3766 "B", "Jail may lock (unlock) physical pages in memory");
3767 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3768 "B", "Jail may bind sockets to reserved ports");
3769 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3770 "B", "Jail may read the kernel message buffer");
3771 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
3772 "B", "Unprivileged processes may use process debugging facilities");
3774 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3775 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3776 "B", "Jail may mount/unmount jail-friendly file systems in general");
3779 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
3780 * its associated bit in the pr_allow bitmask, or zero if the parameter was
3784 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3787 struct bool_flags *bf;
3788 struct sysctl_oid *parent;
3789 char *allow_name, *allow_noname, *allowed;
3790 #ifndef NO_SYSCTL_DESCR
3791 char *descr_deprecated;
3793 unsigned allow_flag;
3796 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3798 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3800 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3801 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3802 free(allow_name, M_PRISON);
3807 * See if this parameter has already beed added, i.e. a module was
3808 * previously loaded/unloaded.
3810 mtx_lock(&prison0.pr_mtx);
3811 for (bf = pr_flag_allow;
3812 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3814 if (strcmp(bf->name, allow_name) == 0) {
3815 allow_flag = bf->flag;
3821 * Find a free bit in prison0's pr_allow, failing if there are none
3822 * (which shouldn't happen as long as we keep track of how many
3823 * potential dynamic flags exist).
3825 * Due to per-jail unprivileged process debugging support
3826 * using pr_allow, also verify against PR_ALLOW_ALL_STATIC.
3827 * prison0 may have unprivileged process debugging unset.
3829 for (allow_flag = 1;; allow_flag <<= 1) {
3830 if (allow_flag == 0)
3832 if (allow_flag & PR_ALLOW_ALL_STATIC)
3834 if ((prison0.pr_allow & allow_flag) == 0)
3839 * Note the parameter in the next open slot in pr_flag_allow.
3840 * Set the flag last so code that checks pr_flag_allow can do so
3843 for (bf = pr_flag_allow; bf->flag != 0; bf++)
3844 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3845 /* This should never happen, but is not fatal. */
3849 prison0.pr_allow |= allow_flag;
3850 bf->name = allow_name;
3851 bf->noname = allow_noname;
3852 bf->flag = allow_flag;
3853 mtx_unlock(&prison0.pr_mtx);
3856 * Create sysctls for the paramter, and the back-compat global
3860 ? SYSCTL_ADD_NODE(NULL,
3861 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3862 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
3863 : &sysctl___security_jail_param_allow;
3864 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3865 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3866 NULL, 0, sysctl_jail_param, "B", descr);
3868 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3869 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3870 #ifndef NO_SYSCTL_DESCR
3871 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3874 (void)SYSCTL_ADD_PROC(NULL,
3875 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3876 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3877 sysctl_jail_default_allow, "I", descr_deprecated);
3878 #ifndef NO_SYSCTL_DESCR
3879 free(descr_deprecated, M_TEMP);
3881 free(allowed, M_TEMP);
3886 mtx_unlock(&prison0.pr_mtx);
3887 free(allow_name, M_PRISON);
3888 free(allow_noname, M_PRISON);
3893 * The VFS system will register jail-aware filesystems here. They each get
3894 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3895 * attempts to mount.
3898 prison_add_vfs(struct vfsconf *vfsp)
3900 #ifdef NO_SYSCTL_DESCR
3902 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3907 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3909 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3911 free(descr, M_TEMP);
3917 prison_racct_foreach(void (*callback)(struct racct *racct,
3918 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3919 void *arg2, void *arg3)
3921 struct prison_racct *prr;
3923 ASSERT_RACCT_ENABLED();
3925 sx_slock(&allprison_lock);
3928 LIST_FOREACH(prr, &allprison_racct, prr_next)
3929 (callback)(prr->prr_racct, arg2, arg3);
3932 sx_sunlock(&allprison_lock);
3935 static struct prison_racct *
3936 prison_racct_find_locked(const char *name)
3938 struct prison_racct *prr;
3940 ASSERT_RACCT_ENABLED();
3941 sx_assert(&allprison_lock, SA_XLOCKED);
3943 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3946 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3947 if (strcmp(name, prr->prr_name) != 0)
3950 /* Found prison_racct with a matching name? */
3951 prison_racct_hold(prr);
3955 /* Add new prison_racct. */
3956 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3957 racct_create(&prr->prr_racct);
3959 strcpy(prr->prr_name, name);
3960 refcount_init(&prr->prr_refcount, 1);
3961 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3966 struct prison_racct *
3967 prison_racct_find(const char *name)
3969 struct prison_racct *prr;
3971 ASSERT_RACCT_ENABLED();
3973 sx_xlock(&allprison_lock);
3974 prr = prison_racct_find_locked(name);
3975 sx_xunlock(&allprison_lock);
3980 prison_racct_hold(struct prison_racct *prr)
3983 ASSERT_RACCT_ENABLED();
3985 refcount_acquire(&prr->prr_refcount);
3989 prison_racct_free_locked(struct prison_racct *prr)
3992 ASSERT_RACCT_ENABLED();
3993 sx_assert(&allprison_lock, SA_XLOCKED);
3995 if (refcount_release(&prr->prr_refcount)) {
3996 racct_destroy(&prr->prr_racct);
3997 LIST_REMOVE(prr, prr_next);
3998 free(prr, M_PRISON_RACCT);
4003 prison_racct_free(struct prison_racct *prr)
4006 ASSERT_RACCT_ENABLED();
4007 sx_assert(&allprison_lock, SA_UNLOCKED);
4009 if (refcount_release_if_not_last(&prr->prr_refcount))
4012 sx_xlock(&allprison_lock);
4013 prison_racct_free_locked(prr);
4014 sx_xunlock(&allprison_lock);
4018 prison_racct_attach(struct prison *pr)
4020 struct prison_racct *prr;
4022 ASSERT_RACCT_ENABLED();
4023 sx_assert(&allprison_lock, SA_XLOCKED);
4025 prr = prison_racct_find_locked(pr->pr_name);
4026 KASSERT(prr != NULL, ("cannot find prison_racct"));
4028 pr->pr_prison_racct = prr;
4032 * Handle jail renaming. From the racct point of view, renaming means
4033 * moving from one prison_racct to another.
4036 prison_racct_modify(struct prison *pr)
4042 struct prison_racct *oldprr;
4044 ASSERT_RACCT_ENABLED();
4046 sx_slock(&allproc_lock);
4047 sx_xlock(&allprison_lock);
4049 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4050 sx_xunlock(&allprison_lock);
4051 sx_sunlock(&allproc_lock);
4055 oldprr = pr->pr_prison_racct;
4056 pr->pr_prison_racct = NULL;
4058 prison_racct_attach(pr);
4061 * Move resource utilisation records.
4063 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4067 * Force rctl to reattach rules to processes.
4069 FOREACH_PROC_IN_SYSTEM(p) {
4071 cred = crhold(p->p_ucred);
4073 rctl_proc_ucred_changed(p, cred);
4078 sx_sunlock(&allproc_lock);
4079 prison_racct_free_locked(oldprr);
4080 sx_xunlock(&allprison_lock);
4084 prison_racct_detach(struct prison *pr)
4087 ASSERT_RACCT_ENABLED();
4088 sx_assert(&allprison_lock, SA_UNLOCKED);
4090 if (pr->pr_prison_racct == NULL)
4092 prison_racct_free(pr->pr_prison_racct);
4093 pr->pr_prison_racct = NULL;
4100 db_show_prison(struct prison *pr)
4102 struct bool_flags *bf;
4103 struct jailsys_flags *jsf;
4104 #if defined(INET) || defined(INET6)
4109 char ip4buf[INET_ADDRSTRLEN];
4112 char ip6buf[INET6_ADDRSTRLEN];
4115 db_printf("prison %p:\n", pr);
4116 db_printf(" jid = %d\n", pr->pr_id);
4117 db_printf(" name = %s\n", pr->pr_name);
4118 db_printf(" parent = %p\n", pr->pr_parent);
4119 db_printf(" ref = %d\n", pr->pr_ref);
4120 db_printf(" uref = %d\n", pr->pr_uref);
4121 db_printf(" path = %s\n", pr->pr_path);
4122 db_printf(" cpuset = %d\n", pr->pr_cpuset
4123 ? pr->pr_cpuset->cs_id : -1);
4125 db_printf(" vnet = %p\n", pr->pr_vnet);
4127 db_printf(" root = %p\n", pr->pr_root);
4128 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4129 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4130 db_printf(" children.max = %d\n", pr->pr_childmax);
4131 db_printf(" children.cur = %d\n", pr->pr_childcount);
4132 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4133 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4134 db_printf(" flags = 0x%x", pr->pr_flags);
4135 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4136 if (pr->pr_flags & bf->flag)
4137 db_printf(" %s", bf->name);
4138 for (jsf = pr_flag_jailsys;
4139 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4141 f = pr->pr_flags & (jsf->disable | jsf->new);
4142 db_printf(" %-16s= %s\n", jsf->name,
4143 (f != 0 && f == jsf->disable) ? "disable"
4144 : (f == jsf->new) ? "new"
4147 db_printf(" allow = 0x%x", pr->pr_allow);
4148 for (bf = pr_flag_allow;
4149 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4151 if (pr->pr_allow & bf->flag)
4152 db_printf(" %s", bf->name);
4154 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4155 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4156 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4157 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4158 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4160 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4161 for (ii = 0; ii < pr->pr_ip4s; ii++)
4162 db_printf(" %s %s\n",
4163 ii == 0 ? "ip4.addr =" : " ",
4164 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4167 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4168 for (ii = 0; ii < pr->pr_ip6s; ii++)
4169 db_printf(" %s %s\n",
4170 ii == 0 ? "ip6.addr =" : " ",
4171 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4175 DB_SHOW_COMMAND(prison, db_show_prison_command)
4181 * Show all prisons in the list, and prison0 which is not
4184 db_show_prison(&prison0);
4185 if (!db_pager_quit) {
4186 TAILQ_FOREACH(pr, &allprison, pr_list) {
4198 /* Look for a prison with the ID and with references. */
4199 TAILQ_FOREACH(pr, &allprison, pr_list)
4200 if (pr->pr_id == addr && pr->pr_ref > 0)
4203 /* Look again, without requiring a reference. */
4204 TAILQ_FOREACH(pr, &allprison, pr_list)
4205 if (pr->pr_id == addr)
4208 /* Assume address points to a valid prison. */
4209 pr = (struct prison *)addr;