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 int fullpath_disabled;
494 #if defined(INET) || defined(INET6)
503 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
504 uint64_t pr_allow_diff;
508 error = priv_check(td, PRIV_JAIL_SET);
509 if (!error && (flags & JAIL_ATTACH))
510 error = priv_check(td, PRIV_JAIL_ATTACH);
513 mypr = td->td_ucred->cr_prison;
514 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
516 if (flags & ~JAIL_SET_MASK)
520 * Check all the parameters before committing to anything. Not all
521 * errors can be caught early, but we may as well try. Also, this
522 * takes care of some expensive stuff (path lookup) before getting
523 * the allprison lock.
525 * XXX Jails are not filesystems, and jail parameters are not mount
526 * options. But it makes more sense to re-use the vfsopt code
527 * than duplicate it under a different name.
529 error = vfs_buildopts(optuio, &opts);
540 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
543 vfs_opterror(opts, "no valid operation (create or update)");
547 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
553 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
562 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
570 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
575 else if (enforce < 0 || enforce > 2) {
581 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
589 pr_flags = ch_flags = 0;
590 for (bf = pr_flag_bool;
591 bf < pr_flag_bool + nitems(pr_flag_bool);
593 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
594 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
596 ch_flags |= pr_flags;
597 for (jsf = pr_flag_jailsys;
598 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
600 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
606 case JAIL_SYS_DISABLE:
611 pr_flags |= jsf->disable;
614 pr_flags |= jsf->new;
616 case JAIL_SYS_INHERIT:
622 ch_flags |= jsf->new | jsf->disable;
624 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
625 && !(pr_flags & PR_PERSIST)) {
627 vfs_opterror(opts, "new jail must persist or attach");
631 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
633 vfs_opterror(opts, "vnet cannot be changed after creation");
638 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
640 vfs_opterror(opts, "ip4 cannot be changed after creation");
645 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
647 vfs_opterror(opts, "ip6 cannot be changed after creation");
652 pr_allow = ch_allow = 0;
653 for (bf = pr_flag_allow;
654 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
656 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
657 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
659 ch_allow |= pr_allow;
661 error = vfs_getopt(opts, "name", (void **)&name, &len);
667 if (len == 0 || name[len - 1] != '\0') {
671 if (len > MAXHOSTNAMELEN) {
672 error = ENAMETOOLONG;
677 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
685 if (len == 0 || host[len - 1] != '\0') {
689 if (len > MAXHOSTNAMELEN) {
690 error = ENAMETOOLONG;
695 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
703 if (len == 0 || domain[len - 1] != '\0') {
707 if (len > MAXHOSTNAMELEN) {
708 error = ENAMETOOLONG;
713 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
721 if (len == 0 || uuid[len - 1] != '\0') {
725 if (len > HOSTUUIDLEN) {
726 error = ENAMETOOLONG;
731 #ifdef COMPAT_FREEBSD32
732 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
735 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
739 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
751 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
756 else if (ip4s & (sizeof(*ip4) - 1)) {
760 ch_flags |= PR_IP4_USER;
761 pr_flags |= PR_IP4_USER;
763 ip4s /= sizeof(*ip4);
764 if (ip4s > jail_max_af_ips) {
766 vfs_opterror(opts, "too many IPv4 addresses");
769 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
770 bcopy(op, ip4, ip4s * sizeof(*ip4));
772 * IP addresses are all sorted but ip[0] to preserve
773 * the primary IP address as given from userland.
774 * This special IP is used for unbound outgoing
775 * connections as well for "loopback" traffic in case
776 * source address selection cannot find any more fitting
777 * address to connect from.
780 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
783 * Check for duplicate addresses and do some simple
784 * zero and broadcast checks. If users give other bogus
785 * addresses it is their problem.
787 * We do not have to care about byte order for these
788 * checks so we will do them in NBO.
790 for (ii = 0; ii < ip4s; ii++) {
791 if (ip4[ii].s_addr == INADDR_ANY ||
792 ip4[ii].s_addr == INADDR_BROADCAST) {
797 (ip4[0].s_addr == ip4[ii+1].s_addr ||
798 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
808 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
813 else if (ip6s & (sizeof(*ip6) - 1)) {
817 ch_flags |= PR_IP6_USER;
818 pr_flags |= PR_IP6_USER;
820 ip6s /= sizeof(*ip6);
821 if (ip6s > jail_max_af_ips) {
823 vfs_opterror(opts, "too many IPv6 addresses");
826 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
827 bcopy(op, ip6, ip6s * sizeof(*ip6));
829 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
831 for (ii = 0; ii < ip6s; ii++) {
832 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
837 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
838 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
848 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
849 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
852 "vnet jails cannot have IP address restrictions");
857 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
863 if (flags & JAIL_UPDATE) {
866 "osrelease cannot be changed after creation");
869 if (len == 0 || len >= OSRELEASELEN) {
872 "osrelease string must be 1-%d bytes long",
878 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
884 if (flags & JAIL_UPDATE) {
887 "osreldate cannot be changed after creation");
892 vfs_opterror(opts, "osreldate cannot be 0");
897 fullpath_disabled = 0;
899 error = vfs_getopt(opts, "path", (void **)&path, &len);
905 if (flags & JAIL_UPDATE) {
908 "path cannot be changed after creation");
911 if (len == 0 || path[len - 1] != '\0') {
915 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
921 NDFREE(&nd, NDF_ONLY_PNBUF);
922 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
923 strlcpy(g_path, path, MAXPATHLEN);
924 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
927 else if (error == ENODEV) {
928 /* proceed if sysctl debug.disablefullpath == 1 */
929 fullpath_disabled = 1;
930 if (len < 2 || (len == 2 && path[0] == '/'))
933 /* exit on other errors */
936 if (root->v_type != VDIR) {
942 if (fullpath_disabled) {
943 /* Leave room for a real-root full pathname. */
944 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
945 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
946 error = ENAMETOOLONG;
954 * Find the specified jail, or at least its parent.
955 * This abuses the file error codes ENOENT and EEXIST.
959 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
960 namelc = strrchr(name, '.');
961 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
965 sx_xlock(&allprison_lock);
968 * See if a requested jid already exists. There is an
969 * information leak here if the jid exists but is not within
970 * the caller's jail hierarchy. Jail creators will get EEXIST
971 * even though they cannot see the jail, and CREATE | UPDATE
972 * will return ENOENT which is not normally a valid error.
976 vfs_opterror(opts, "negative jid");
977 goto done_unlock_list;
979 pr = prison_find(jid);
982 /* Create: jid must not exist. */
983 if (cuflags == JAIL_CREATE) {
984 mtx_unlock(&pr->pr_mtx);
986 vfs_opterror(opts, "jail %d already exists",
988 goto done_unlock_list;
990 if (!prison_ischild(mypr, pr)) {
991 mtx_unlock(&pr->pr_mtx);
993 } else if (pr->pr_uref == 0) {
994 if (!(flags & JAIL_DYING)) {
995 mtx_unlock(&pr->pr_mtx);
997 vfs_opterror(opts, "jail %d is dying",
999 goto done_unlock_list;
1000 } else if ((flags & JAIL_ATTACH) ||
1001 (pr_flags & PR_PERSIST)) {
1003 * A dying jail might be resurrected
1004 * (via attach or persist), but first
1005 * it must determine if another jail
1006 * has claimed its name. Accomplish
1007 * this by implicitly re-setting the
1011 name = prison_name(mypr, pr);
1016 /* Update: jid must exist. */
1017 if (cuflags == JAIL_UPDATE) {
1019 vfs_opterror(opts, "jail %d not found", jid);
1020 goto done_unlock_list;
1025 * If the caller provided a name, look for a jail by that name.
1026 * This has different semantics for creates and updates keyed by jid
1027 * (where the name must not already exist in a different jail),
1028 * and updates keyed by the name itself (where the name must exist
1029 * because that is the jail being updated).
1033 namelc = strrchr(name, '.');
1038 * This is a hierarchical name. Split it into the
1039 * parent and child names, and make sure the parent
1040 * exists or matches an already found jail.
1043 if (strncmp(name, ppr->pr_name, namelc - name)
1044 || ppr->pr_name[namelc - name] != '\0') {
1045 mtx_unlock(&pr->pr_mtx);
1048 "cannot change jail's parent");
1049 goto done_unlock_list;
1053 ppr = prison_find_name(mypr, name);
1057 "jail \"%s\" not found", name);
1058 goto done_unlock_list;
1060 mtx_unlock(&ppr->pr_mtx);
1065 if (namelc[0] != '\0') {
1067 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1070 FOREACH_PRISON_CHILD(ppr, tpr) {
1071 if (tpr != pr && tpr->pr_ref > 0 &&
1072 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1074 cuflags != JAIL_CREATE) {
1075 mtx_lock(&tpr->pr_mtx);
1076 if (tpr->pr_ref > 0) {
1081 if (tpr->pr_uref > 0) {
1087 mtx_unlock(&tpr->pr_mtx);
1088 } else if (tpr->pr_uref > 0) {
1090 * Create, or update(jid):
1091 * name must not exist in an
1092 * active sibling jail.
1096 mtx_unlock(&pr->pr_mtx);
1098 "jail \"%s\" already exists",
1100 goto done_unlock_list;
1104 /* If no active jail is found, use a dying one. */
1105 if (deadpr != NULL && pr == NULL) {
1106 if (flags & JAIL_DYING) {
1107 mtx_lock(&deadpr->pr_mtx);
1108 if (deadpr->pr_ref == 0) {
1109 mtx_unlock(&deadpr->pr_mtx);
1113 } else if (cuflags == JAIL_UPDATE) {
1116 "jail \"%s\" is dying", name);
1117 goto done_unlock_list;
1120 /* Update: name must exist if no jid. */
1121 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1123 vfs_opterror(opts, "jail \"%s\" not found",
1125 goto done_unlock_list;
1129 /* Update: must provide a jid or name. */
1130 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1132 vfs_opterror(opts, "update specified no jail");
1133 goto done_unlock_list;
1136 /* If there's no prison to update, create a new one and link it in. */
1138 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1139 if (tpr->pr_childcount >= tpr->pr_childmax) {
1141 vfs_opterror(opts, "prison limit exceeded");
1142 goto done_unlock_list;
1145 mtx_lock(&ppr->pr_mtx);
1146 if (ppr->pr_ref == 0) {
1147 mtx_unlock(&ppr->pr_mtx);
1149 vfs_opterror(opts, "jail \"%s\" not found",
1150 prison_name(mypr, ppr));
1151 goto done_unlock_list;
1155 mtx_unlock(&ppr->pr_mtx);
1156 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1158 /* Find the next free jid. */
1161 if (jid == JAIL_MAX)
1163 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1164 if (tpr->pr_id < jid)
1166 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1167 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1170 if (jid == lastprid) {
1173 "no available jail IDs");
1175 prison_deref(ppr, PD_DEREF |
1176 PD_DEUREF | PD_LIST_XLOCKED);
1185 * The jail already has a jid (that did not yet exist),
1186 * so just find where to insert it.
1188 TAILQ_FOREACH(tpr, &allprison, pr_list)
1189 if (tpr->pr_id >= jid) {
1190 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1195 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1196 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1197 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1198 tpr->pr_childcount++;
1200 pr->pr_parent = ppr;
1203 /* Set some default values, and inherit some from the parent. */
1208 root = mypr->pr_root;
1211 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1212 pr->pr_flags |= PR_HOST;
1213 #if defined(INET) || defined(INET6)
1215 if (!(pr_flags & PR_VNET))
1219 if (!(ch_flags & PR_IP4_USER))
1220 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1221 else if (!(pr_flags & PR_IP4_USER)) {
1222 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1223 if (ppr->pr_ip4 != NULL) {
1224 pr->pr_ip4s = ppr->pr_ip4s;
1225 pr->pr_ip4 = malloc(pr->pr_ip4s *
1226 sizeof(struct in_addr), M_PRISON,
1228 bcopy(ppr->pr_ip4, pr->pr_ip4,
1229 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1234 if (!(ch_flags & PR_IP6_USER))
1235 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1236 else if (!(pr_flags & PR_IP6_USER)) {
1237 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1238 if (ppr->pr_ip6 != NULL) {
1239 pr->pr_ip6s = ppr->pr_ip6s;
1240 pr->pr_ip6 = malloc(pr->pr_ip6s *
1241 sizeof(struct in6_addr), M_PRISON,
1243 bcopy(ppr->pr_ip6, pr->pr_ip6,
1244 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1250 /* Source address selection is always on by default. */
1251 pr->pr_flags |= _PR_IP_SADDRSEL;
1253 pr->pr_securelevel = ppr->pr_securelevel;
1254 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1255 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1256 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1258 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1259 if (osrelstr == NULL)
1260 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1262 strcpy(pr->pr_osrelease, osrelstr);
1264 LIST_INIT(&pr->pr_children);
1265 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1266 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1269 /* Allocate a new vnet if specified. */
1270 pr->pr_vnet = (pr_flags & PR_VNET)
1271 ? vnet_alloc() : ppr->pr_vnet;
1274 * Allocate a dedicated cpuset for each jail.
1275 * Unlike other initial settings, this may return an erorr.
1277 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1279 prison_deref(pr, PD_LIST_XLOCKED);
1283 mtx_lock(&pr->pr_mtx);
1285 * New prisons do not yet have a reference, because we do not
1286 * want others to see the incomplete prison once the
1287 * allprison_lock is downgraded.
1292 * Grab a reference for existing prisons, to ensure they
1293 * continue to exist for the duration of the call.
1296 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1297 if ((pr->pr_flags & PR_VNET) &&
1298 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1301 "vnet jails cannot have IP address restrictions");
1302 goto done_deref_locked;
1306 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1309 "ip4 cannot be changed after creation");
1310 goto done_deref_locked;
1314 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1317 "ip6 cannot be changed after creation");
1318 goto done_deref_locked;
1323 /* Do final error checking before setting anything. */
1325 if (slevel < ppr->pr_securelevel) {
1327 goto done_deref_locked;
1331 if (childmax >= ppr->pr_childmax) {
1333 goto done_deref_locked;
1337 if (enforce < ppr->pr_enforce_statfs) {
1339 goto done_deref_locked;
1344 * devfs_rsnum is a uint16_t
1346 if (rsnum < 0 || rsnum > 65535) {
1348 goto done_deref_locked;
1351 * Nested jails always inherit parent's devfs ruleset
1353 if (jailed(td->td_ucred)) {
1354 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1356 goto done_deref_locked;
1358 rsnum = ppr->pr_devfs_rsnum;
1363 if (ppr->pr_flags & PR_IP4) {
1365 * Make sure the new set of IP addresses is a
1366 * subset of the parent's list. Don't worry
1367 * about the parent being unlocked, as any
1368 * setting is done with allprison_lock held.
1370 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1371 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1373 if (ij == ppr->pr_ip4s) {
1375 goto done_deref_locked;
1378 for (ii = ij = 1; ii < ip4s; ii++) {
1379 if (ip4[ii].s_addr ==
1380 ppr->pr_ip4[0].s_addr)
1382 for (; ij < ppr->pr_ip4s; ij++)
1383 if (ip4[ii].s_addr ==
1384 ppr->pr_ip4[ij].s_addr)
1386 if (ij == ppr->pr_ip4s)
1389 if (ij == ppr->pr_ip4s) {
1391 goto done_deref_locked;
1396 * Check for conflicting IP addresses. We permit them
1397 * if there is no more than one IP on each jail. If
1398 * there is a duplicate on a jail with more than one
1399 * IP stop checking and return error.
1402 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1403 if (tppr->pr_flags & PR_VNET)
1408 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1411 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1413 tpr->pr_uref == 0) {
1417 if (!(tpr->pr_flags & PR_IP4_USER))
1420 if (tpr->pr_ip4 == NULL ||
1421 (ip4s == 1 && tpr->pr_ip4s == 1))
1423 for (ii = 0; ii < ip4s; ii++) {
1424 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1428 "IPv4 addresses clash");
1429 goto done_deref_locked;
1437 if (ppr->pr_flags & PR_IP6) {
1439 * Make sure the new set of IP addresses is a
1440 * subset of the parent's list.
1442 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1443 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1446 if (ij == ppr->pr_ip6s) {
1448 goto done_deref_locked;
1451 for (ii = ij = 1; ii < ip6s; ii++) {
1452 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1455 for (; ij < ppr->pr_ip6s; ij++)
1456 if (IN6_ARE_ADDR_EQUAL(
1457 &ip6[ii], &ppr->pr_ip6[ij]))
1459 if (ij == ppr->pr_ip6s)
1462 if (ij == ppr->pr_ip6s) {
1464 goto done_deref_locked;
1468 /* Check for conflicting IP addresses. */
1470 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1471 if (tppr->pr_flags & PR_VNET)
1476 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1479 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1481 tpr->pr_uref == 0) {
1485 if (!(tpr->pr_flags & PR_IP6_USER))
1488 if (tpr->pr_ip6 == NULL ||
1489 (ip6s == 1 && tpr->pr_ip6s == 1))
1491 for (ii = 0; ii < ip6s; ii++) {
1492 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1496 "IPv6 addresses clash");
1497 goto done_deref_locked;
1503 onamelen = namelen = 0;
1504 if (namelc != NULL) {
1505 /* Give a default name of the jid. Also allow the name to be
1506 * explicitly the jid - but not any other number, and only in
1507 * normal form (no leading zero/etc).
1509 if (namelc[0] == '\0')
1510 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1511 else if ((strtoul(namelc, &p, 10) != jid ||
1512 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1515 "name cannot be numeric (unless it is the jid)");
1516 goto done_deref_locked;
1519 * Make sure the name isn't too long for the prison or its
1522 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1523 onamelen = strlen(pr->pr_name + pnamelen);
1524 namelen = strlen(namelc);
1525 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1526 error = ENAMETOOLONG;
1527 goto done_deref_locked;
1529 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1530 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1531 sizeof(pr->pr_name)) {
1532 error = ENAMETOOLONG;
1533 goto done_deref_locked;
1537 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1538 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1540 goto done_deref_locked;
1544 * Let modules check their parameters. This requires unlocking and
1545 * then re-locking the prison, but this is still a valid state as long
1546 * as allprison_lock remains xlocked.
1548 mtx_unlock(&pr->pr_mtx);
1549 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1551 prison_deref(pr, created
1553 : PD_DEREF | PD_LIST_XLOCKED);
1556 mtx_lock(&pr->pr_mtx);
1558 /* At this point, all valid parameters should have been noted. */
1559 TAILQ_FOREACH(opt, opts, link) {
1560 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1562 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1563 goto done_deref_locked;
1567 /* Set the parameters of the prison. */
1570 if (pr_flags & PR_IP4_USER) {
1571 pr->pr_flags |= PR_IP4;
1572 free(pr->pr_ip4, M_PRISON);
1576 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1578 if (tpr->pr_flags & PR_VNET) {
1583 if (prison_restrict_ip4(tpr, NULL)) {
1592 if (pr_flags & PR_IP6_USER) {
1593 pr->pr_flags |= PR_IP6;
1594 free(pr->pr_ip6, M_PRISON);
1598 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1600 if (tpr->pr_flags & PR_VNET) {
1605 if (prison_restrict_ip6(tpr, NULL)) {
1613 pr->pr_securelevel = slevel;
1614 /* Set all child jails to be at least this level. */
1615 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1616 if (tpr->pr_securelevel < slevel)
1617 tpr->pr_securelevel = slevel;
1620 pr->pr_childmax = childmax;
1621 /* Set all child jails to under this limit. */
1622 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1623 if (tpr->pr_childmax > childmax - level)
1624 tpr->pr_childmax = childmax > level
1625 ? childmax - level : 0;
1628 pr->pr_enforce_statfs = enforce;
1629 /* Pass this restriction on to the children. */
1630 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1631 if (tpr->pr_enforce_statfs < enforce)
1632 tpr->pr_enforce_statfs = enforce;
1635 pr->pr_devfs_rsnum = rsnum;
1636 /* Pass this restriction on to the children. */
1637 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1638 tpr->pr_devfs_rsnum = rsnum;
1640 if (namelc != NULL) {
1641 if (ppr == &prison0)
1642 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1644 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1645 ppr->pr_name, namelc);
1646 /* Change this component of child names. */
1647 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1648 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1649 strlen(tpr->pr_name + onamelen) + 1);
1650 bcopy(pr->pr_name, tpr->pr_name, namelen);
1654 /* Try to keep a real-rooted full pathname. */
1655 if (fullpath_disabled && path[0] == '/' &&
1656 strcmp(mypr->pr_path, "/"))
1657 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1658 mypr->pr_path, path);
1660 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1663 if (PR_HOST & ch_flags & ~pr_flags) {
1664 if (pr->pr_flags & PR_HOST) {
1666 * Copy the parent's host info. As with pr_ip4 above,
1667 * the lack of a lock on the parent is not a problem;
1668 * it is always set with allprison_lock at least
1669 * shared, and is held exclusively here.
1671 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1672 sizeof(pr->pr_hostname));
1673 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1674 sizeof(pr->pr_domainname));
1675 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1676 sizeof(pr->pr_hostuuid));
1677 pr->pr_hostid = pr->pr_parent->pr_hostid;
1679 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1680 /* Set this prison, and any descendants without PR_HOST. */
1682 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1684 strlcpy(pr->pr_domainname, domain,
1685 sizeof(pr->pr_domainname));
1687 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1689 pr->pr_hostid = hid;
1690 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1691 if (tpr->pr_flags & PR_HOST)
1695 strlcpy(tpr->pr_hostname,
1697 sizeof(tpr->pr_hostname));
1699 strlcpy(tpr->pr_domainname,
1701 sizeof(tpr->pr_domainname));
1703 strlcpy(tpr->pr_hostuuid,
1705 sizeof(tpr->pr_hostuuid));
1707 tpr->pr_hostid = hid;
1711 if ((tallow = ch_allow & ~pr_allow)) {
1712 /* Clear allow bits in all children. */
1713 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1714 tpr->pr_allow &= ~tallow;
1716 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1718 * Persistent prisons get an extra reference, and prisons losing their
1719 * persist flag lose that reference. Only do this for existing prisons
1720 * for now, so new ones will remain unseen until after the module
1721 * handlers have completed.
1723 born = pr->pr_uref == 0;
1724 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1725 if (pr_flags & PR_PERSIST) {
1733 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1734 mtx_unlock(&pr->pr_mtx);
1737 if (racct_enable && created)
1738 prison_racct_attach(pr);
1741 /* Locks may have prevented a complete restriction of child IP
1742 * addresses. If so, allocate some more memory and try again.
1747 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1748 mtx_lock(&pr->pr_mtx);
1750 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1752 if (tpr->pr_flags & PR_VNET) {
1757 if (prison_restrict_ip4(tpr, ip4)) {
1764 mtx_unlock(&pr->pr_mtx);
1770 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1771 mtx_lock(&pr->pr_mtx);
1773 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1775 if (tpr->pr_flags & PR_VNET) {
1780 if (prison_restrict_ip6(tpr, ip6)) {
1787 mtx_unlock(&pr->pr_mtx);
1791 /* Let the modules do their work. */
1792 sx_downgrade(&allprison_lock);
1794 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1796 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1797 prison_deref(pr, created
1799 : PD_DEREF | PD_LIST_SLOCKED);
1803 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1806 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1807 prison_deref(pr, created
1809 : PD_DEREF | PD_LIST_SLOCKED);
1813 /* Attach this process to the prison if requested. */
1814 if (flags & JAIL_ATTACH) {
1815 mtx_lock(&pr->pr_mtx);
1816 error = do_jail_attach(td, pr);
1818 vfs_opterror(opts, "attach failed");
1820 prison_deref(pr, PD_DEREF);
1826 if (racct_enable && !created) {
1827 if (!(flags & JAIL_ATTACH))
1828 sx_sunlock(&allprison_lock);
1829 prison_racct_modify(pr);
1830 if (!(flags & JAIL_ATTACH))
1831 sx_slock(&allprison_lock);
1835 td->td_retval[0] = pr->pr_id;
1838 * Now that it is all there, drop the temporary reference from existing
1839 * prisons. Or add a reference to newly created persistent prisons
1840 * (which was not done earlier so that the prison would not be publicly
1844 prison_deref(pr, (flags & JAIL_ATTACH)
1846 : PD_DEREF | PD_LIST_SLOCKED);
1848 if (pr_flags & PR_PERSIST) {
1849 mtx_lock(&pr->pr_mtx);
1852 mtx_unlock(&pr->pr_mtx);
1854 if (!(flags & JAIL_ATTACH))
1855 sx_sunlock(&allprison_lock);
1861 prison_deref(pr, created
1862 ? PD_LOCKED | PD_LIST_XLOCKED
1863 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1866 sx_xunlock(&allprison_lock);
1872 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1873 &errmsg_len) == 0 && errmsg_len > 0) {
1874 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1875 if (optuio->uio_segflg == UIO_SYSSPACE)
1877 optuio->uio_iov[errmsg_pos].iov_base,
1881 optuio->uio_iov[errmsg_pos].iov_base,
1887 free(ip4, M_PRISON);
1890 free(ip6, M_PRISON);
1893 free(g_path, M_TEMP);
1899 * struct jail_get_args {
1900 * struct iovec *iovp;
1901 * unsigned int iovcnt;
1906 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1911 /* Check that we have an even number of iovecs. */
1912 if (uap->iovcnt & 1)
1915 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1918 error = kern_jail_get(td, auio, uap->flags);
1920 error = copyout(auio->uio_iov, uap->iovp,
1921 uap->iovcnt * sizeof (struct iovec));
1927 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1929 struct bool_flags *bf;
1930 struct jailsys_flags *jsf;
1931 struct prison *pr, *mypr;
1933 struct vfsoptlist *opts;
1934 char *errmsg, *name;
1935 int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1938 if (flags & ~JAIL_GET_MASK)
1941 /* Get the parameter list. */
1942 error = vfs_buildopts(optuio, &opts);
1945 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1946 mypr = td->td_ucred->cr_prison;
1949 * Find the prison specified by one of: lastjid, jid, name.
1951 sx_slock(&allprison_lock);
1952 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1954 TAILQ_FOREACH(pr, &allprison, pr_list) {
1955 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1956 mtx_lock(&pr->pr_mtx);
1957 if (pr->pr_ref > 0 &&
1958 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1960 mtx_unlock(&pr->pr_mtx);
1966 vfs_opterror(opts, "no jail after %d", jid);
1967 goto done_unlock_list;
1968 } else if (error != ENOENT)
1969 goto done_unlock_list;
1971 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1974 pr = prison_find_child(mypr, jid);
1976 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1977 mtx_unlock(&pr->pr_mtx);
1979 vfs_opterror(opts, "jail %d is dying",
1981 goto done_unlock_list;
1986 vfs_opterror(opts, "jail %d not found", jid);
1987 goto done_unlock_list;
1989 } else if (error != ENOENT)
1990 goto done_unlock_list;
1992 error = vfs_getopt(opts, "name", (void **)&name, &len);
1994 if (len == 0 || name[len - 1] != '\0') {
1996 goto done_unlock_list;
1998 pr = prison_find_name(mypr, name);
2000 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2001 mtx_unlock(&pr->pr_mtx);
2003 vfs_opterror(opts, "jail \"%s\" is dying",
2005 goto done_unlock_list;
2010 vfs_opterror(opts, "jail \"%s\" not found", name);
2011 goto done_unlock_list;
2012 } else if (error != ENOENT)
2013 goto done_unlock_list;
2015 vfs_opterror(opts, "no jail specified");
2017 goto done_unlock_list;
2020 /* Get the parameters of the prison. */
2023 td->td_retval[0] = pr->pr_id;
2024 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2025 if (error != 0 && error != ENOENT)
2027 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2028 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2029 if (error != 0 && error != ENOENT)
2031 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2032 if (error != 0 && error != ENOENT)
2034 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2035 sizeof(pr->pr_cpuset->cs_id));
2036 if (error != 0 && error != ENOENT)
2038 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2039 if (error != 0 && error != ENOENT)
2042 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2043 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2044 if (error != 0 && error != ENOENT)
2048 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2049 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2050 if (error != 0 && error != ENOENT)
2053 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2054 sizeof(pr->pr_securelevel));
2055 if (error != 0 && error != ENOENT)
2057 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2058 sizeof(pr->pr_childcount));
2059 if (error != 0 && error != ENOENT)
2061 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2062 sizeof(pr->pr_childmax));
2063 if (error != 0 && error != ENOENT)
2065 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2066 if (error != 0 && error != ENOENT)
2068 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2069 if (error != 0 && error != ENOENT)
2071 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2072 if (error != 0 && error != ENOENT)
2074 #ifdef COMPAT_FREEBSD32
2075 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2076 uint32_t hid32 = pr->pr_hostid;
2078 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2081 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2082 sizeof(pr->pr_hostid));
2083 if (error != 0 && error != ENOENT)
2085 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2086 sizeof(pr->pr_enforce_statfs));
2087 if (error != 0 && error != ENOENT)
2089 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2090 sizeof(pr->pr_devfs_rsnum));
2091 if (error != 0 && error != ENOENT)
2093 for (bf = pr_flag_bool;
2094 bf < pr_flag_bool + nitems(pr_flag_bool);
2096 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2097 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2098 if (error != 0 && error != ENOENT)
2101 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2102 if (error != 0 && error != ENOENT)
2105 for (jsf = pr_flag_jailsys;
2106 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2108 f = pr->pr_flags & (jsf->disable | jsf->new);
2109 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2110 : (f == jsf->new) ? JAIL_SYS_NEW
2112 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2113 if (error != 0 && error != ENOENT)
2116 for (bf = pr_flag_allow;
2117 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2119 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2120 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2121 if (error != 0 && error != ENOENT)
2124 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2125 if (error != 0 && error != ENOENT)
2128 i = (pr->pr_uref == 0);
2129 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2130 if (error != 0 && error != ENOENT)
2133 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2134 if (error != 0 && error != ENOENT)
2136 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2137 sizeof(pr->pr_osreldate));
2138 if (error != 0 && error != ENOENT)
2140 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2141 if (error != 0 && error != ENOENT)
2144 /* Get the module parameters. */
2145 mtx_unlock(&pr->pr_mtx);
2147 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2150 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2152 /* By now, all parameters should have been noted. */
2153 TAILQ_FOREACH(opt, opts, link) {
2154 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2156 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2161 /* Write the fetched parameters back to userspace. */
2163 TAILQ_FOREACH(opt, opts, link) {
2164 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2165 pos = 2 * opt->pos + 1;
2166 optuio->uio_iov[pos].iov_len = opt->len;
2167 if (opt->value != NULL) {
2168 if (optuio->uio_segflg == UIO_SYSSPACE) {
2170 optuio->uio_iov[pos].iov_base,
2173 error = copyout(opt->value,
2174 optuio->uio_iov[pos].iov_base,
2185 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2189 sx_sunlock(&allprison_lock);
2191 if (error && errmsg_pos >= 0) {
2192 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2193 errmsg_pos = 2 * errmsg_pos + 1;
2194 if (errmsg_len > 0) {
2195 if (optuio->uio_segflg == UIO_SYSSPACE)
2197 optuio->uio_iov[errmsg_pos].iov_base,
2201 optuio->uio_iov[errmsg_pos].iov_base,
2210 * struct jail_remove_args {
2215 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2217 struct prison *pr, *cpr, *lpr, *tpr;
2220 error = priv_check(td, PRIV_JAIL_REMOVE);
2224 sx_xlock(&allprison_lock);
2225 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2227 sx_xunlock(&allprison_lock);
2231 /* Remove all descendants of this prison, then remove this prison. */
2233 if (!LIST_EMPTY(&pr->pr_children)) {
2234 mtx_unlock(&pr->pr_mtx);
2236 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2237 mtx_lock(&cpr->pr_mtx);
2238 if (cpr->pr_ref > 0) {
2242 /* Already removed - do not do it again. */
2245 mtx_unlock(&cpr->pr_mtx);
2247 mtx_lock(&lpr->pr_mtx);
2248 prison_remove_one(lpr);
2249 sx_xlock(&allprison_lock);
2254 mtx_lock(&lpr->pr_mtx);
2255 prison_remove_one(lpr);
2256 sx_xlock(&allprison_lock);
2258 mtx_lock(&pr->pr_mtx);
2260 prison_remove_one(pr);
2265 prison_remove_one(struct prison *pr)
2270 /* If the prison was persistent, it is not anymore. */
2272 if (pr->pr_flags & PR_PERSIST) {
2275 pr->pr_flags &= ~PR_PERSIST;
2279 * jail_remove added a reference. If that's the only one, remove
2282 KASSERT(pr->pr_ref > 0,
2283 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2284 if (pr->pr_ref == 1) {
2286 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2290 mtx_unlock(&pr->pr_mtx);
2291 sx_xunlock(&allprison_lock);
2293 * Kill all processes unfortunate enough to be attached to this prison.
2295 sx_slock(&allproc_lock);
2296 FOREACH_PROC_IN_SYSTEM(p) {
2298 if (p->p_state != PRS_NEW && p->p_ucred &&
2299 p->p_ucred->cr_prison == pr)
2300 kern_psignal(p, SIGKILL);
2303 sx_sunlock(&allproc_lock);
2304 /* Remove the temporary reference added by jail_remove. */
2305 prison_deref(pr, deuref | PD_DEREF);
2309 * struct jail_attach_args {
2314 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2319 error = priv_check(td, PRIV_JAIL_ATTACH);
2324 * Start with exclusive hold on allprison_lock to ensure that a possible
2325 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2326 * But then immediately downgrade it since we don't need to stop
2329 sx_xlock(&allprison_lock);
2330 sx_downgrade(&allprison_lock);
2331 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2333 sx_sunlock(&allprison_lock);
2338 * Do not allow a process to attach to a prison that is not
2339 * considered to be "alive".
2341 if (pr->pr_uref == 0) {
2342 mtx_unlock(&pr->pr_mtx);
2343 sx_sunlock(&allprison_lock);
2347 return (do_jail_attach(td, pr));
2351 do_jail_attach(struct thread *td, struct prison *pr)
2354 struct ucred *newcred, *oldcred;
2358 * XXX: Note that there is a slight race here if two threads
2359 * in the same privileged process attempt to attach to two
2360 * different jails at the same time. It is important for
2361 * user processes not to do this, or they might end up with
2362 * a process root from one prison, but attached to the jail
2367 mtx_unlock(&pr->pr_mtx);
2369 /* Let modules do whatever they need to prepare for attaching. */
2370 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2372 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2375 sx_sunlock(&allprison_lock);
2378 * Reparent the newly attached process to this jail.
2381 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2385 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2386 if ((error = change_dir(pr->pr_root, td)) != 0)
2389 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2392 VOP_UNLOCK(pr->pr_root);
2393 if ((error = pwd_chroot(td, pr->pr_root)))
2398 oldcred = crcopysafe(p, newcred);
2399 newcred->cr_prison = pr;
2400 proc_set_cred(p, newcred);
2403 racct_proc_ucred_changed(p, oldcred, newcred);
2408 rctl_proc_ucred_changed(p, newcred);
2411 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2416 VOP_UNLOCK(pr->pr_root);
2418 /* Tell modules this thread is still in its old jail after all. */
2419 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2420 prison_deref(pr, PD_DEREF | PD_DEUREF);
2425 * Returns a locked prison instance, or NULL on failure.
2428 prison_find(int prid)
2432 sx_assert(&allprison_lock, SX_LOCKED);
2433 TAILQ_FOREACH(pr, &allprison, pr_list) {
2434 if (pr->pr_id == prid) {
2435 mtx_lock(&pr->pr_mtx);
2438 mtx_unlock(&pr->pr_mtx);
2445 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2448 prison_find_child(struct prison *mypr, int prid)
2453 sx_assert(&allprison_lock, SX_LOCKED);
2454 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2455 if (pr->pr_id == prid) {
2456 mtx_lock(&pr->pr_mtx);
2459 mtx_unlock(&pr->pr_mtx);
2466 * Look for the name relative to mypr. Returns a locked prison or NULL.
2469 prison_find_name(struct prison *mypr, const char *name)
2471 struct prison *pr, *deadpr;
2475 sx_assert(&allprison_lock, SX_LOCKED);
2476 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2479 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2480 if (!strcmp(pr->pr_name + mylen, name)) {
2481 mtx_lock(&pr->pr_mtx);
2482 if (pr->pr_ref > 0) {
2483 if (pr->pr_uref > 0)
2487 mtx_unlock(&pr->pr_mtx);
2490 /* There was no valid prison - perhaps there was a dying one. */
2491 if (deadpr != NULL) {
2492 mtx_lock(&deadpr->pr_mtx);
2493 if (deadpr->pr_ref == 0) {
2494 mtx_unlock(&deadpr->pr_mtx);
2502 * See if a prison has the specific flag set.
2505 prison_flag(struct ucred *cred, unsigned flag)
2508 /* This is an atomic read, so no locking is necessary. */
2509 return (cred->cr_prison->pr_flags & flag);
2513 prison_allow(struct ucred *cred, unsigned flag)
2516 /* This is an atomic read, so no locking is necessary. */
2517 return (cred->cr_prison->pr_allow & flag);
2521 * Remove a prison reference. If that was the last reference, remove the
2522 * prison itself - but not in this context in case there are locks held.
2525 prison_free_locked(struct prison *pr)
2529 mtx_assert(&pr->pr_mtx, MA_OWNED);
2531 mtx_unlock(&pr->pr_mtx);
2533 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2537 prison_free(struct prison *pr)
2540 mtx_lock(&pr->pr_mtx);
2541 prison_free_locked(pr);
2545 * Complete a call to either prison_free or prison_proc_free.
2548 prison_complete(void *context, int pending)
2550 struct prison *pr = context;
2552 sx_xlock(&allprison_lock);
2553 mtx_lock(&pr->pr_mtx);
2554 prison_deref(pr, pr->pr_uref
2555 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2556 : PD_LOCKED | PD_LIST_XLOCKED);
2560 * Remove a prison reference (usually). This internal version assumes no
2561 * mutexes are held, except perhaps the prison itself. If there are no more
2562 * references, release and delist the prison. On completion, the prison lock
2563 * and the allprison lock are both unlocked.
2566 prison_deref(struct prison *pr, int flags)
2568 struct prison *ppr, *tpr;
2571 if (!(flags & PD_LOCKED))
2572 mtx_lock(&pr->pr_mtx);
2574 if (flags & PD_DEUREF) {
2575 KASSERT(pr->pr_uref > 0,
2576 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2579 lasturef = pr->pr_uref == 0;
2582 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2585 if (flags & PD_DEREF) {
2586 KASSERT(pr->pr_ref > 0,
2587 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2592 mtx_unlock(&pr->pr_mtx);
2595 * Tell the modules if the last user reference was removed
2596 * (even it sticks around in dying state).
2599 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2600 sx_xlock(&allprison_lock);
2601 flags |= PD_LIST_XLOCKED;
2603 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2604 mtx_lock(&pr->pr_mtx);
2606 mtx_unlock(&pr->pr_mtx);
2609 /* If the prison still has references, nothing else to do. */
2611 if (flags & PD_LIST_SLOCKED)
2612 sx_sunlock(&allprison_lock);
2613 else if (flags & PD_LIST_XLOCKED)
2614 sx_xunlock(&allprison_lock);
2618 if (flags & PD_LIST_SLOCKED) {
2619 if (!sx_try_upgrade(&allprison_lock)) {
2620 sx_sunlock(&allprison_lock);
2621 sx_xlock(&allprison_lock);
2623 } else if (!(flags & PD_LIST_XLOCKED))
2624 sx_xlock(&allprison_lock);
2626 TAILQ_REMOVE(&allprison, pr, pr_list);
2627 LIST_REMOVE(pr, pr_sibling);
2628 ppr = pr->pr_parent;
2629 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2630 tpr->pr_childcount--;
2631 sx_xunlock(&allprison_lock);
2634 if (pr->pr_vnet != ppr->pr_vnet)
2635 vnet_destroy(pr->pr_vnet);
2637 if (pr->pr_root != NULL)
2639 mtx_destroy(&pr->pr_mtx);
2641 free(pr->pr_ip4, M_PRISON);
2644 free(pr->pr_ip6, M_PRISON);
2646 if (pr->pr_cpuset != NULL)
2647 cpuset_rel(pr->pr_cpuset);
2651 prison_racct_detach(pr);
2655 /* Removing a prison frees a reference on its parent. */
2657 mtx_lock(&pr->pr_mtx);
2658 flags = PD_DEREF | PD_DEUREF;
2663 prison_hold_locked(struct prison *pr)
2666 mtx_assert(&pr->pr_mtx, MA_OWNED);
2667 KASSERT(pr->pr_ref > 0,
2668 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2673 prison_hold(struct prison *pr)
2676 mtx_lock(&pr->pr_mtx);
2677 prison_hold_locked(pr);
2678 mtx_unlock(&pr->pr_mtx);
2682 prison_proc_hold(struct prison *pr)
2685 mtx_lock(&pr->pr_mtx);
2686 KASSERT(pr->pr_uref > 0,
2687 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2689 mtx_unlock(&pr->pr_mtx);
2693 prison_proc_free(struct prison *pr)
2696 mtx_lock(&pr->pr_mtx);
2697 KASSERT(pr->pr_uref > 0,
2698 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2699 if (pr->pr_uref > 1)
2703 * Don't remove the last user reference in this context, which
2704 * is expected to be a process that is not only locked, but
2708 mtx_unlock(&pr->pr_mtx);
2709 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2712 mtx_unlock(&pr->pr_mtx);
2716 * Check if a jail supports the given address family.
2718 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2722 prison_check_af(struct ucred *cred, int af)
2727 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2729 pr = cred->cr_prison;
2731 /* Prisons with their own network stack are not limited. */
2732 if (prison_owns_vnet(cred))
2741 if (pr->pr_flags & PR_IP4)
2743 mtx_lock(&pr->pr_mtx);
2744 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2745 error = EAFNOSUPPORT;
2746 mtx_unlock(&pr->pr_mtx);
2752 if (pr->pr_flags & PR_IP6)
2754 mtx_lock(&pr->pr_mtx);
2755 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2756 error = EAFNOSUPPORT;
2757 mtx_unlock(&pr->pr_mtx);
2765 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2766 error = EAFNOSUPPORT;
2772 * Check if given address belongs to the jail referenced by cred (wrapper to
2773 * prison_check_ip[46]).
2775 * Returns 0 if jail doesn't restrict the address family or if address belongs
2776 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2777 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
2780 prison_if(struct ucred *cred, const struct sockaddr *sa)
2783 const struct sockaddr_in *sai;
2786 const struct sockaddr_in6 *sai6;
2790 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2791 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2794 if (prison_owns_vnet(cred))
2799 switch (sa->sa_family)
2803 sai = (const struct sockaddr_in *)sa;
2804 error = prison_check_ip4(cred, &sai->sin_addr);
2809 sai6 = (const struct sockaddr_in6 *)sa;
2810 error = prison_check_ip6(cred, &sai6->sin6_addr);
2814 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2815 error = EAFNOSUPPORT;
2821 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2824 prison_check(struct ucred *cred1, struct ucred *cred2)
2827 return ((cred1->cr_prison == cred2->cr_prison ||
2828 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2832 * Return 1 if p2 is a child of p1, otherwise 0.
2835 prison_ischild(struct prison *pr1, struct prison *pr2)
2838 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2845 * Return 1 if the passed credential is in a jail, otherwise 0.
2848 jailed(struct ucred *cred)
2851 return (cred->cr_prison != &prison0);
2855 * Return 1 if the passed credential is in a jail and that jail does not
2856 * have its own virtual network stack, otherwise 0.
2859 jailed_without_vnet(struct ucred *cred)
2865 if (prison_owns_vnet(cred))
2873 * Return the correct hostname (domainname, et al) for the passed credential.
2876 getcredhostname(struct ucred *cred, char *buf, size_t size)
2881 * A NULL credential can be used to shortcut to the physical
2882 * system's hostname.
2884 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2885 mtx_lock(&pr->pr_mtx);
2886 strlcpy(buf, pr->pr_hostname, size);
2887 mtx_unlock(&pr->pr_mtx);
2891 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2894 mtx_lock(&cred->cr_prison->pr_mtx);
2895 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2896 mtx_unlock(&cred->cr_prison->pr_mtx);
2900 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2903 mtx_lock(&cred->cr_prison->pr_mtx);
2904 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2905 mtx_unlock(&cred->cr_prison->pr_mtx);
2909 getcredhostid(struct ucred *cred, unsigned long *hostid)
2912 mtx_lock(&cred->cr_prison->pr_mtx);
2913 *hostid = cred->cr_prison->pr_hostid;
2914 mtx_unlock(&cred->cr_prison->pr_mtx);
2919 * Determine whether the prison represented by cred owns
2920 * its vnet rather than having it inherited.
2922 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2925 prison_owns_vnet(struct ucred *cred)
2929 * vnets cannot be added/removed after jail creation,
2930 * so no need to lock here.
2932 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2937 * Determine whether the subject represented by cred can "see"
2938 * status of a mount point.
2939 * Returns: 0 for permitted, ENOENT otherwise.
2940 * XXX: This function should be called cr_canseemount() and should be
2941 * placed in kern_prot.c.
2944 prison_canseemount(struct ucred *cred, struct mount *mp)
2950 pr = cred->cr_prison;
2951 if (pr->pr_enforce_statfs == 0)
2953 if (pr->pr_root->v_mount == mp)
2955 if (pr->pr_enforce_statfs == 2)
2958 * If jail's chroot directory is set to "/" we should be able to see
2959 * all mount-points from inside a jail.
2960 * This is ugly check, but this is the only situation when jail's
2961 * directory ends with '/'.
2963 if (strcmp(pr->pr_path, "/") == 0)
2965 len = strlen(pr->pr_path);
2967 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2970 * Be sure that we don't have situation where jail's root directory
2971 * is "/some/path" and mount point is "/some/pathpath".
2973 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2979 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2981 char jpath[MAXPATHLEN];
2985 pr = cred->cr_prison;
2986 if (pr->pr_enforce_statfs == 0)
2988 if (prison_canseemount(cred, mp) != 0) {
2989 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2990 strlcpy(sp->f_mntonname, "[restricted]",
2991 sizeof(sp->f_mntonname));
2994 if (pr->pr_root->v_mount == mp) {
2996 * Clear current buffer data, so we are sure nothing from
2997 * the valid path left there.
2999 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3000 *sp->f_mntonname = '/';
3004 * If jail's chroot directory is set to "/" we should be able to see
3005 * all mount-points from inside a jail.
3007 if (strcmp(pr->pr_path, "/") == 0)
3009 len = strlen(pr->pr_path);
3010 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3012 * Clear current buffer data, so we are sure nothing from
3013 * the valid path left there.
3015 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3016 if (*jpath == '\0') {
3017 /* Should never happen. */
3018 *sp->f_mntonname = '/';
3020 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3025 * Check with permission for a specific privilege is granted within jail. We
3026 * have a specific list of accepted privileges; the rest are denied.
3029 prison_priv_check(struct ucred *cred, int priv)
3037 * Privileges specific to prisons with a virtual network stack.
3038 * There might be a duplicate entry here in case the privilege
3039 * is only granted conditionally in the legacy jail case.
3044 * NFS-specific privileges.
3046 case PRIV_NFS_DAEMON:
3047 case PRIV_NFS_LOCKD:
3050 * Network stack privileges.
3052 case PRIV_NET_BRIDGE:
3055 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3056 case PRIV_NET_ROUTE:
3058 case PRIV_NET_SETIFMTU:
3059 case PRIV_NET_SETIFFLAGS:
3060 case PRIV_NET_SETIFCAP:
3061 case PRIV_NET_SETIFDESCR:
3062 case PRIV_NET_SETIFNAME :
3063 case PRIV_NET_SETIFMETRIC:
3064 case PRIV_NET_SETIFPHYS:
3065 case PRIV_NET_SETIFMAC:
3066 case PRIV_NET_SETLANPCP:
3067 case PRIV_NET_ADDMULTI:
3068 case PRIV_NET_DELMULTI:
3069 case PRIV_NET_HWIOCTL:
3070 case PRIV_NET_SETLLADDR:
3071 case PRIV_NET_ADDIFGROUP:
3072 case PRIV_NET_DELIFGROUP:
3073 case PRIV_NET_IFCREATE:
3074 case PRIV_NET_IFDESTROY:
3075 case PRIV_NET_ADDIFADDR:
3076 case PRIV_NET_DELIFADDR:
3079 case PRIV_NET_SETIFVNET:
3080 case PRIV_NET_SETIFFIB:
3083 * 802.11-related privileges.
3085 case PRIV_NET80211_GETKEY:
3087 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3094 case PRIV_NETATM_CFG:
3095 case PRIV_NETATM_ADD:
3096 case PRIV_NETATM_DEL:
3097 case PRIV_NETATM_SET:
3100 * Bluetooth privileges.
3102 case PRIV_NETBLUETOOTH_RAW:
3106 * Netgraph and netgraph module privileges.
3108 case PRIV_NETGRAPH_CONTROL:
3110 case PRIV_NETGRAPH_TTY:
3114 * IPv4 and IPv6 privileges.
3116 case PRIV_NETINET_IPFW:
3117 case PRIV_NETINET_DIVERT:
3118 case PRIV_NETINET_PF:
3119 case PRIV_NETINET_DUMMYNET:
3120 case PRIV_NETINET_CARP:
3121 case PRIV_NETINET_MROUTE:
3122 case PRIV_NETINET_RAW:
3123 case PRIV_NETINET_ADDRCTRL6:
3124 case PRIV_NETINET_ND6:
3125 case PRIV_NETINET_SCOPE6:
3126 case PRIV_NETINET_ALIFETIME6:
3127 case PRIV_NETINET_IPSEC:
3128 case PRIV_NETINET_BINDANY:
3143 * No default: or deny here.
3144 * In case of no permit fall through to next switch().
3146 if (cred->cr_prison->pr_flags & PR_VNET)
3154 * Allow ktrace privileges for root in jail.
3160 * Allow jailed processes to configure audit identity and
3161 * submit audit records (login, etc). In the future we may
3162 * want to further refine the relationship between audit and
3165 case PRIV_AUDIT_GETAUDIT:
3166 case PRIV_AUDIT_SETAUDIT:
3167 case PRIV_AUDIT_SUBMIT:
3171 * Allow jailed processes to manipulate process UNIX
3172 * credentials in any way they see fit.
3174 case PRIV_CRED_SETUID:
3175 case PRIV_CRED_SETEUID:
3176 case PRIV_CRED_SETGID:
3177 case PRIV_CRED_SETEGID:
3178 case PRIV_CRED_SETGROUPS:
3179 case PRIV_CRED_SETREUID:
3180 case PRIV_CRED_SETREGID:
3181 case PRIV_CRED_SETRESUID:
3182 case PRIV_CRED_SETRESGID:
3185 * Jail implements visibility constraints already, so allow
3186 * jailed root to override uid/gid-based constraints.
3188 case PRIV_SEEOTHERGIDS:
3189 case PRIV_SEEOTHERUIDS:
3192 * Jail implements inter-process debugging limits already, so
3193 * allow jailed root various debugging privileges.
3195 case PRIV_DEBUG_DIFFCRED:
3196 case PRIV_DEBUG_SUGID:
3197 case PRIV_DEBUG_UNPRIV:
3200 * Allow jail to set various resource limits and login
3201 * properties, and for now, exceed process resource limits.
3203 case PRIV_PROC_LIMIT:
3204 case PRIV_PROC_SETLOGIN:
3205 case PRIV_PROC_SETRLIMIT:
3208 * System V and POSIX IPC privileges are granted in jail.
3211 case PRIV_IPC_WRITE:
3212 case PRIV_IPC_ADMIN:
3213 case PRIV_IPC_MSGSIZE:
3217 * Jail operations within a jail work on child jails.
3219 case PRIV_JAIL_ATTACH:
3221 case PRIV_JAIL_REMOVE:
3224 * Jail implements its own inter-process limits, so allow
3225 * root processes in jail to change scheduling on other
3226 * processes in the same jail. Likewise for signalling.
3228 case PRIV_SCHED_DIFFCRED:
3229 case PRIV_SCHED_CPUSET:
3230 case PRIV_SIGNAL_DIFFCRED:
3231 case PRIV_SIGNAL_SUGID:
3234 * Allow jailed processes to write to sysctls marked as jail
3237 case PRIV_SYSCTL_WRITEJAIL:
3240 * Allow root in jail to manage a variety of quota
3241 * properties. These should likely be conditional on a
3242 * configuration option.
3244 case PRIV_VFS_GETQUOTA:
3245 case PRIV_VFS_SETQUOTA:
3248 * Since Jail relies on chroot() to implement file system
3249 * protections, grant many VFS privileges to root in jail.
3250 * Be careful to exclude mount-related and NFS-related
3254 case PRIV_VFS_WRITE:
3255 case PRIV_VFS_ADMIN:
3257 case PRIV_VFS_LOOKUP:
3258 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3259 case PRIV_VFS_CHFLAGS_DEV:
3260 case PRIV_VFS_CHOWN:
3261 case PRIV_VFS_CHROOT:
3262 case PRIV_VFS_RETAINSUGID:
3263 case PRIV_VFS_FCHROOT:
3265 case PRIV_VFS_SETGID:
3267 case PRIV_VFS_STICKYFILE:
3270 * As in the non-jail case, non-root users are expected to be
3271 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3272 * exists in the jail and they have permission to access it).
3274 case PRIV_KMEM_READ:
3278 * Depending on the global setting, allow privilege of
3279 * setting system flags.
3281 case PRIV_VFS_SYSFLAGS:
3282 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3288 * Depending on the global setting, allow privilege of
3289 * mounting/unmounting file systems.
3291 case PRIV_VFS_MOUNT:
3292 case PRIV_VFS_UNMOUNT:
3293 case PRIV_VFS_MOUNT_NONUSER:
3294 case PRIV_VFS_MOUNT_OWNER:
3295 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3296 cred->cr_prison->pr_enforce_statfs < 2)
3302 * Conditionnaly allow locking (unlocking) physical pages
3306 case PRIV_VM_MUNLOCK:
3307 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3313 * Conditionally allow jailed root to bind reserved ports.
3315 case PRIV_NETINET_RESERVEDPORT:
3316 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3322 * Allow jailed root to reuse in-use ports.
3324 case PRIV_NETINET_REUSEPORT:
3328 * Allow jailed root to set certain IPv4/6 (option) headers.
3330 case PRIV_NETINET_SETHDROPTS:
3334 * Conditionally allow creating raw sockets in jail.
3336 case PRIV_NETINET_RAW:
3337 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3343 * Since jail implements its own visibility limits on netstat
3344 * sysctls, allow getcred. This allows identd to work in
3347 case PRIV_NETINET_GETCRED:
3351 * Allow jailed root to set loginclass.
3353 case PRIV_PROC_SETLOGINCLASS:
3357 * Do not allow a process inside a jail to read the kernel
3358 * message buffer unless explicitly permitted.
3361 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3367 * In all remaining cases, deny the privilege request. This
3368 * includes almost all network privileges, many system
3369 * configuration privileges.
3376 * Return the part of pr2's name that is relative to pr1, or the whole name
3377 * if it does not directly follow.
3381 prison_name(struct prison *pr1, struct prison *pr2)
3385 /* Jails see themselves as "0" (if they see themselves at all). */
3388 name = pr2->pr_name;
3389 if (prison_ischild(pr1, pr2)) {
3391 * pr1 isn't locked (and allprison_lock may not be either)
3392 * so its length can't be counted on. But the number of dots
3393 * can be counted on - and counted.
3395 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3396 name = strchr(name, '.') + 1;
3402 * Return the part of pr2's path that is relative to pr1, or the whole path
3403 * if it does not directly follow.
3406 prison_path(struct prison *pr1, struct prison *pr2)
3408 char *path1, *path2;
3411 path1 = pr1->pr_path;
3412 path2 = pr2->pr_path;
3413 if (!strcmp(path1, "/"))
3415 len1 = strlen(path1);
3416 if (strncmp(path1, path2, len1))
3418 if (path2[len1] == '\0')
3420 if (path2[len1] == '/')
3421 return (path2 + len1);
3426 * Jail-related sysctls.
3428 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3432 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3435 struct prison *pr, *cpr;
3437 struct in_addr *ip4 = NULL;
3441 struct in6_addr *ip6 = NULL;
3446 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3447 pr = req->td->td_ucred->cr_prison;
3449 sx_slock(&allprison_lock);
3450 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3451 #if defined(INET) || defined(INET6)
3454 mtx_lock(&cpr->pr_mtx);
3456 if (cpr->pr_ip4s > 0) {
3457 if (ip4s < cpr->pr_ip4s) {
3458 ip4s = cpr->pr_ip4s;
3459 mtx_unlock(&cpr->pr_mtx);
3460 ip4 = realloc(ip4, ip4s *
3461 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3464 bcopy(cpr->pr_ip4, ip4,
3465 cpr->pr_ip4s * sizeof(struct in_addr));
3469 if (cpr->pr_ip6s > 0) {
3470 if (ip6s < cpr->pr_ip6s) {
3471 ip6s = cpr->pr_ip6s;
3472 mtx_unlock(&cpr->pr_mtx);
3473 ip6 = realloc(ip6, ip6s *
3474 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3477 bcopy(cpr->pr_ip6, ip6,
3478 cpr->pr_ip6s * sizeof(struct in6_addr));
3481 if (cpr->pr_ref == 0) {
3482 mtx_unlock(&cpr->pr_mtx);
3485 bzero(xp, sizeof(*xp));
3486 xp->pr_version = XPRISON_VERSION;
3487 xp->pr_id = cpr->pr_id;
3488 xp->pr_state = cpr->pr_uref > 0
3489 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3490 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3491 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3492 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3494 xp->pr_ip4s = cpr->pr_ip4s;
3497 xp->pr_ip6s = cpr->pr_ip6s;
3499 mtx_unlock(&cpr->pr_mtx);
3500 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3504 if (xp->pr_ip4s > 0) {
3505 error = SYSCTL_OUT(req, ip4,
3506 xp->pr_ip4s * sizeof(struct in_addr));
3512 if (xp->pr_ip6s > 0) {
3513 error = SYSCTL_OUT(req, ip6,
3514 xp->pr_ip6s * sizeof(struct in6_addr));
3520 sx_sunlock(&allprison_lock);
3531 SYSCTL_OID(_security_jail, OID_AUTO, list,
3532 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3533 sysctl_jail_list, "S", "List of active jails");
3536 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3540 injail = jailed(req->td->td_ucred);
3541 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3546 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3547 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3548 sysctl_jail_jailed, "I", "Process in jail?");
3551 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3553 int error, havevnet;
3555 struct ucred *cred = req->td->td_ucred;
3557 havevnet = jailed(cred) && prison_owns_vnet(cred);
3561 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3566 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3567 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3568 sysctl_jail_vnet, "I", "Jail owns vnet?");
3570 #if defined(INET) || defined(INET6)
3571 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3572 &jail_max_af_ips, 0,
3573 "Number of IP addresses a jail may have at most per address family (deprecated)");
3577 * Default parameters for jail(2) compatibility. For historical reasons,
3578 * the sysctl names have varying similarity to the parameter names. Prisons
3579 * just see their own parameters, and can't change them.
3582 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3585 int allow, error, i;
3587 pr = req->td->td_ucred->cr_prison;
3588 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3590 /* Get the current flag value, and convert it to a boolean. */
3591 i = (allow & arg2) ? 1 : 0;
3594 error = sysctl_handle_int(oidp, &i, 0, req);
3595 if (error || !req->newptr)
3601 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3604 mtx_lock(&prison0.pr_mtx);
3605 jail_default_allow = (jail_default_allow & ~arg2) | i;
3606 mtx_unlock(&prison0.pr_mtx);
3610 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3611 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3612 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3613 "Processes in jail can set their hostnames (deprecated)");
3614 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3615 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3616 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3617 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3618 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3619 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3620 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3621 "Processes in jail can use System V IPC primitives (deprecated)");
3622 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3623 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3624 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3625 "Prison root can create raw sockets (deprecated)");
3626 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3627 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3628 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3629 "Processes in jail can alter system file flags (deprecated)");
3630 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3631 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3632 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3633 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3636 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3641 pr = req->td->td_ucred->cr_prison;
3642 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3643 error = sysctl_handle_int(oidp, &level, 0, req);
3644 if (error || !req->newptr)
3646 *(int *)arg1 = level;
3650 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3651 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3652 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3653 sysctl_jail_default_level, "I",
3654 "Processes in jail cannot see all mounted file systems (deprecated)");
3656 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3657 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3658 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3659 sysctl_jail_default_level, "I",
3660 "Ruleset for the devfs filesystem in jail (deprecated)");
3663 * Nodes to describe jail parameters. Maximum length of string parameters
3664 * is returned in the string itself, and the other parameters exist merely
3665 * to make themselves and their types known.
3667 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3671 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3678 switch (oidp->oid_kind & CTLTYPE)
3684 if (!(req->flags & SCTL_MASK32))
3686 return (SYSCTL_OUT(req, &l, sizeof(l)));
3690 return (SYSCTL_OUT(req, &i, sizeof(i)));
3691 case CTLTYPE_STRING:
3692 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3694 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3695 case CTLTYPE_STRUCT:
3697 return (SYSCTL_OUT(req, &s, sizeof(s)));
3703 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3704 * jail creation time but cannot be changed in an existing jail.
3706 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3707 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3708 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3709 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3710 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3711 "I", "Jail secure level");
3712 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3713 "Jail value for kern.osreldate and uname -K");
3714 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3715 "Jail value for kern.osrelease and uname -r");
3716 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3717 "I", "Jail cannot see all mounted file systems");
3718 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3719 "I", "Ruleset for in-jail devfs mounts");
3720 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3721 "B", "Jail persistence");
3723 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3724 "E,jailsys", "Virtual network stack");
3726 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3727 "B", "Jail is in the process of shutting down");
3729 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3730 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3731 "I", "Current number of child jails");
3732 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3733 "I", "Maximum number of child jails");
3735 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3736 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3738 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3739 "Jail NIS domainname");
3740 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3742 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3743 "LU", "Jail host ID");
3745 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3746 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3749 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3750 "Jail IPv4 address virtualization");
3751 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3752 "S,in_addr,a", "Jail IPv4 addresses");
3753 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3754 "B", "Do (not) use IPv4 source address selection rather than the "
3755 "primary jail IPv4 address.");
3758 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3759 "Jail IPv6 address virtualization");
3760 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3761 "S,in6_addr,a", "Jail IPv6 addresses");
3762 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3763 "B", "Do (not) use IPv6 source address selection rather than the "
3764 "primary jail IPv6 address.");
3767 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3768 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3769 "B", "Jail may set hostname");
3770 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3771 "B", "Jail may use SYSV IPC");
3772 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3773 "B", "Jail may create raw sockets");
3774 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3775 "B", "Jail may alter system file flags");
3776 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3777 "B", "Jail may set file quotas");
3778 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3779 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3780 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3781 "B", "Jail may lock (unlock) physical pages in memory");
3782 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3783 "B", "Jail may bind sockets to reserved ports");
3784 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3785 "B", "Jail may read the kernel message buffer");
3786 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
3787 "B", "Unprivileged processes may use process debugging facilities");
3789 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3790 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3791 "B", "Jail may mount/unmount jail-friendly file systems in general");
3794 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
3795 * its associated bit in the pr_allow bitmask, or zero if the parameter was
3799 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3802 struct bool_flags *bf;
3803 struct sysctl_oid *parent;
3804 char *allow_name, *allow_noname, *allowed;
3805 #ifndef NO_SYSCTL_DESCR
3806 char *descr_deprecated;
3808 unsigned allow_flag;
3811 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3813 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3815 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3816 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3817 free(allow_name, M_PRISON);
3822 * See if this parameter has already beed added, i.e. a module was
3823 * previously loaded/unloaded.
3825 mtx_lock(&prison0.pr_mtx);
3826 for (bf = pr_flag_allow;
3827 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3829 if (strcmp(bf->name, allow_name) == 0) {
3830 allow_flag = bf->flag;
3836 * Find a free bit in prison0's pr_allow, failing if there are none
3837 * (which shouldn't happen as long as we keep track of how many
3838 * potential dynamic flags exist).
3840 * Due to per-jail unprivileged process debugging support
3841 * using pr_allow, also verify against PR_ALLOW_ALL_STATIC.
3842 * prison0 may have unprivileged process debugging unset.
3844 for (allow_flag = 1;; allow_flag <<= 1) {
3845 if (allow_flag == 0)
3847 if (allow_flag & PR_ALLOW_ALL_STATIC)
3849 if ((prison0.pr_allow & allow_flag) == 0)
3854 * Note the parameter in the next open slot in pr_flag_allow.
3855 * Set the flag last so code that checks pr_flag_allow can do so
3858 for (bf = pr_flag_allow; bf->flag != 0; bf++)
3859 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3860 /* This should never happen, but is not fatal. */
3864 prison0.pr_allow |= allow_flag;
3865 bf->name = allow_name;
3866 bf->noname = allow_noname;
3867 bf->flag = allow_flag;
3868 mtx_unlock(&prison0.pr_mtx);
3871 * Create sysctls for the paramter, and the back-compat global
3875 ? SYSCTL_ADD_NODE(NULL,
3876 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3877 OID_AUTO, prefix, 0, 0, prefix_descr)
3878 : &sysctl___security_jail_param_allow;
3879 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3880 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3881 NULL, 0, sysctl_jail_param, "B", descr);
3883 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3884 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3885 #ifndef NO_SYSCTL_DESCR
3886 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3889 (void)SYSCTL_ADD_PROC(NULL,
3890 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3891 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3892 sysctl_jail_default_allow, "I", descr_deprecated);
3893 #ifndef NO_SYSCTL_DESCR
3894 free(descr_deprecated, M_TEMP);
3896 free(allowed, M_TEMP);
3901 mtx_unlock(&prison0.pr_mtx);
3902 free(allow_name, M_PRISON);
3903 free(allow_noname, M_PRISON);
3908 * The VFS system will register jail-aware filesystems here. They each get
3909 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3910 * attempts to mount.
3913 prison_add_vfs(struct vfsconf *vfsp)
3915 #ifdef NO_SYSCTL_DESCR
3917 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3922 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3924 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3926 free(descr, M_TEMP);
3932 prison_racct_foreach(void (*callback)(struct racct *racct,
3933 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3934 void *arg2, void *arg3)
3936 struct prison_racct *prr;
3938 ASSERT_RACCT_ENABLED();
3940 sx_slock(&allprison_lock);
3943 LIST_FOREACH(prr, &allprison_racct, prr_next)
3944 (callback)(prr->prr_racct, arg2, arg3);
3947 sx_sunlock(&allprison_lock);
3950 static struct prison_racct *
3951 prison_racct_find_locked(const char *name)
3953 struct prison_racct *prr;
3955 ASSERT_RACCT_ENABLED();
3956 sx_assert(&allprison_lock, SA_XLOCKED);
3958 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3961 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3962 if (strcmp(name, prr->prr_name) != 0)
3965 /* Found prison_racct with a matching name? */
3966 prison_racct_hold(prr);
3970 /* Add new prison_racct. */
3971 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3972 racct_create(&prr->prr_racct);
3974 strcpy(prr->prr_name, name);
3975 refcount_init(&prr->prr_refcount, 1);
3976 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3981 struct prison_racct *
3982 prison_racct_find(const char *name)
3984 struct prison_racct *prr;
3986 ASSERT_RACCT_ENABLED();
3988 sx_xlock(&allprison_lock);
3989 prr = prison_racct_find_locked(name);
3990 sx_xunlock(&allprison_lock);
3995 prison_racct_hold(struct prison_racct *prr)
3998 ASSERT_RACCT_ENABLED();
4000 refcount_acquire(&prr->prr_refcount);
4004 prison_racct_free_locked(struct prison_racct *prr)
4007 ASSERT_RACCT_ENABLED();
4008 sx_assert(&allprison_lock, SA_XLOCKED);
4010 if (refcount_release(&prr->prr_refcount)) {
4011 racct_destroy(&prr->prr_racct);
4012 LIST_REMOVE(prr, prr_next);
4013 free(prr, M_PRISON_RACCT);
4018 prison_racct_free(struct prison_racct *prr)
4021 ASSERT_RACCT_ENABLED();
4022 sx_assert(&allprison_lock, SA_UNLOCKED);
4024 if (refcount_release_if_not_last(&prr->prr_refcount))
4027 sx_xlock(&allprison_lock);
4028 prison_racct_free_locked(prr);
4029 sx_xunlock(&allprison_lock);
4033 prison_racct_attach(struct prison *pr)
4035 struct prison_racct *prr;
4037 ASSERT_RACCT_ENABLED();
4038 sx_assert(&allprison_lock, SA_XLOCKED);
4040 prr = prison_racct_find_locked(pr->pr_name);
4041 KASSERT(prr != NULL, ("cannot find prison_racct"));
4043 pr->pr_prison_racct = prr;
4047 * Handle jail renaming. From the racct point of view, renaming means
4048 * moving from one prison_racct to another.
4051 prison_racct_modify(struct prison *pr)
4057 struct prison_racct *oldprr;
4059 ASSERT_RACCT_ENABLED();
4061 sx_slock(&allproc_lock);
4062 sx_xlock(&allprison_lock);
4064 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4065 sx_xunlock(&allprison_lock);
4066 sx_sunlock(&allproc_lock);
4070 oldprr = pr->pr_prison_racct;
4071 pr->pr_prison_racct = NULL;
4073 prison_racct_attach(pr);
4076 * Move resource utilisation records.
4078 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4082 * Force rctl to reattach rules to processes.
4084 FOREACH_PROC_IN_SYSTEM(p) {
4086 cred = crhold(p->p_ucred);
4088 rctl_proc_ucred_changed(p, cred);
4093 sx_sunlock(&allproc_lock);
4094 prison_racct_free_locked(oldprr);
4095 sx_xunlock(&allprison_lock);
4099 prison_racct_detach(struct prison *pr)
4102 ASSERT_RACCT_ENABLED();
4103 sx_assert(&allprison_lock, SA_UNLOCKED);
4105 if (pr->pr_prison_racct == NULL)
4107 prison_racct_free(pr->pr_prison_racct);
4108 pr->pr_prison_racct = NULL;
4115 db_show_prison(struct prison *pr)
4117 struct bool_flags *bf;
4118 struct jailsys_flags *jsf;
4119 #if defined(INET) || defined(INET6)
4124 char ip4buf[INET_ADDRSTRLEN];
4127 char ip6buf[INET6_ADDRSTRLEN];
4130 db_printf("prison %p:\n", pr);
4131 db_printf(" jid = %d\n", pr->pr_id);
4132 db_printf(" name = %s\n", pr->pr_name);
4133 db_printf(" parent = %p\n", pr->pr_parent);
4134 db_printf(" ref = %d\n", pr->pr_ref);
4135 db_printf(" uref = %d\n", pr->pr_uref);
4136 db_printf(" path = %s\n", pr->pr_path);
4137 db_printf(" cpuset = %d\n", pr->pr_cpuset
4138 ? pr->pr_cpuset->cs_id : -1);
4140 db_printf(" vnet = %p\n", pr->pr_vnet);
4142 db_printf(" root = %p\n", pr->pr_root);
4143 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4144 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4145 db_printf(" children.max = %d\n", pr->pr_childmax);
4146 db_printf(" children.cur = %d\n", pr->pr_childcount);
4147 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4148 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4149 db_printf(" flags = 0x%x", pr->pr_flags);
4150 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4151 if (pr->pr_flags & bf->flag)
4152 db_printf(" %s", bf->name);
4153 for (jsf = pr_flag_jailsys;
4154 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4156 f = pr->pr_flags & (jsf->disable | jsf->new);
4157 db_printf(" %-16s= %s\n", jsf->name,
4158 (f != 0 && f == jsf->disable) ? "disable"
4159 : (f == jsf->new) ? "new"
4162 db_printf(" allow = 0x%x", pr->pr_allow);
4163 for (bf = pr_flag_allow;
4164 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4166 if (pr->pr_allow & bf->flag)
4167 db_printf(" %s", bf->name);
4169 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4170 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4171 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4172 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4173 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4175 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4176 for (ii = 0; ii < pr->pr_ip4s; ii++)
4177 db_printf(" %s %s\n",
4178 ii == 0 ? "ip4.addr =" : " ",
4179 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4182 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4183 for (ii = 0; ii < pr->pr_ip6s; ii++)
4184 db_printf(" %s %s\n",
4185 ii == 0 ? "ip6.addr =" : " ",
4186 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4190 DB_SHOW_COMMAND(prison, db_show_prison_command)
4196 * Show all prisons in the list, and prison0 which is not
4199 db_show_prison(&prison0);
4200 if (!db_pager_quit) {
4201 TAILQ_FOREACH(pr, &allprison, pr_list) {
4213 /* Look for a prison with the ID and with references. */
4214 TAILQ_FOREACH(pr, &allprison, pr_list)
4215 if (pr->pr_id == addr && pr->pr_ref > 0)
4218 /* Look again, without requiring a reference. */
4219 TAILQ_FOREACH(pr, &allprison, pr_list)
4220 if (pr->pr_id == addr)
4223 /* Assume address points to a valid prison. */
4224 pr = (struct prison *)addr;