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);
441 * struct jail_set_args {
442 * struct iovec *iovp;
443 * unsigned int iovcnt;
448 sys_jail_set(struct thread *td, struct jail_set_args *uap)
453 /* Check that we have an even number of iovecs. */
457 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
460 error = kern_jail_set(td, auio, uap->flags);
466 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
473 struct in6_addr *ip6;
476 struct vfsoptlist *opts;
477 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
479 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
480 char *g_path, *osrelstr;
481 struct bool_flags *bf;
482 struct jailsys_flags *jsf;
483 #if defined(INET) || defined(INET6)
488 size_t namelen, onamelen, pnamelen;
489 int born, created, cuflags, descend, enforce;
490 int error, errmsg_len, errmsg_pos;
491 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
492 int jid, jsys, len, level;
493 int childmax, osreldt, rsnum, slevel;
494 int fullpath_disabled;
495 #if defined(INET) || defined(INET6)
504 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
505 uint64_t pr_allow_diff;
509 error = priv_check(td, PRIV_JAIL_SET);
510 if (!error && (flags & JAIL_ATTACH))
511 error = priv_check(td, PRIV_JAIL_ATTACH);
514 mypr = td->td_ucred->cr_prison;
515 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
517 if (flags & ~JAIL_SET_MASK)
521 * Check all the parameters before committing to anything. Not all
522 * errors can be caught early, but we may as well try. Also, this
523 * takes care of some expensive stuff (path lookup) before getting
524 * the allprison lock.
526 * XXX Jails are not filesystems, and jail parameters are not mount
527 * options. But it makes more sense to re-use the vfsopt code
528 * than duplicate it under a different name.
530 error = vfs_buildopts(optuio, &opts);
541 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
544 vfs_opterror(opts, "no valid operation (create or update)");
548 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
554 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
563 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
571 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
576 else if (enforce < 0 || enforce > 2) {
582 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
590 pr_flags = ch_flags = 0;
591 for (bf = pr_flag_bool;
592 bf < pr_flag_bool + nitems(pr_flag_bool);
594 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
595 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
597 ch_flags |= pr_flags;
598 for (jsf = pr_flag_jailsys;
599 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
601 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
607 case JAIL_SYS_DISABLE:
612 pr_flags |= jsf->disable;
615 pr_flags |= jsf->new;
617 case JAIL_SYS_INHERIT:
623 ch_flags |= jsf->new | jsf->disable;
625 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
626 && !(pr_flags & PR_PERSIST)) {
628 vfs_opterror(opts, "new jail must persist or attach");
632 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
634 vfs_opterror(opts, "vnet cannot be changed after creation");
639 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
641 vfs_opterror(opts, "ip4 cannot be changed after creation");
646 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
648 vfs_opterror(opts, "ip6 cannot be changed after creation");
653 pr_allow = ch_allow = 0;
654 for (bf = pr_flag_allow;
655 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
657 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
658 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
660 ch_allow |= pr_allow;
662 error = vfs_getopt(opts, "name", (void **)&name, &len);
668 if (len == 0 || name[len - 1] != '\0') {
672 if (len > MAXHOSTNAMELEN) {
673 error = ENAMETOOLONG;
678 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
686 if (len == 0 || host[len - 1] != '\0') {
690 if (len > MAXHOSTNAMELEN) {
691 error = ENAMETOOLONG;
696 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
704 if (len == 0 || domain[len - 1] != '\0') {
708 if (len > MAXHOSTNAMELEN) {
709 error = ENAMETOOLONG;
714 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
722 if (len == 0 || uuid[len - 1] != '\0') {
726 if (len > HOSTUUIDLEN) {
727 error = ENAMETOOLONG;
732 #ifdef COMPAT_FREEBSD32
733 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
736 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
740 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
752 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
757 else if (ip4s & (sizeof(*ip4) - 1)) {
761 ch_flags |= PR_IP4_USER;
762 pr_flags |= PR_IP4_USER;
764 ip4s /= sizeof(*ip4);
765 if (ip4s > jail_max_af_ips) {
767 vfs_opterror(opts, "too many IPv4 addresses");
770 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
771 bcopy(op, ip4, ip4s * sizeof(*ip4));
773 * IP addresses are all sorted but ip[0] to preserve
774 * the primary IP address as given from userland.
775 * This special IP is used for unbound outgoing
776 * connections as well for "loopback" traffic in case
777 * source address selection cannot find any more fitting
778 * address to connect from.
781 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
784 * Check for duplicate addresses and do some simple
785 * zero and broadcast checks. If users give other bogus
786 * addresses it is their problem.
788 * We do not have to care about byte order for these
789 * checks so we will do them in NBO.
791 for (ii = 0; ii < ip4s; ii++) {
792 if (ip4[ii].s_addr == INADDR_ANY ||
793 ip4[ii].s_addr == INADDR_BROADCAST) {
798 (ip4[0].s_addr == ip4[ii+1].s_addr ||
799 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
809 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
814 else if (ip6s & (sizeof(*ip6) - 1)) {
818 ch_flags |= PR_IP6_USER;
819 pr_flags |= PR_IP6_USER;
821 ip6s /= sizeof(*ip6);
822 if (ip6s > jail_max_af_ips) {
824 vfs_opterror(opts, "too many IPv6 addresses");
827 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
828 bcopy(op, ip6, ip6s * sizeof(*ip6));
830 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
832 for (ii = 0; ii < ip6s; ii++) {
833 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
838 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
839 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
849 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
850 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
853 "vnet jails cannot have IP address restrictions");
858 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
864 if (flags & JAIL_UPDATE) {
867 "osrelease cannot be changed after creation");
870 if (len == 0 || len >= OSRELEASELEN) {
873 "osrelease string must be 1-%d bytes long",
879 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
885 if (flags & JAIL_UPDATE) {
888 "osreldate cannot be changed after creation");
893 vfs_opterror(opts, "osreldate cannot be 0");
898 fullpath_disabled = 0;
900 error = vfs_getopt(opts, "path", (void **)&path, &len);
906 if (flags & JAIL_UPDATE) {
909 "path cannot be changed after creation");
912 if (len == 0 || path[len - 1] != '\0') {
916 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
922 NDFREE(&nd, NDF_ONLY_PNBUF);
923 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
924 strlcpy(g_path, path, MAXPATHLEN);
925 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
928 else if (error == ENODEV) {
929 /* proceed if sysctl debug.disablefullpath == 1 */
930 fullpath_disabled = 1;
931 if (len < 2 || (len == 2 && path[0] == '/'))
934 /* exit on other errors */
937 if (root->v_type != VDIR) {
943 if (fullpath_disabled) {
944 /* Leave room for a real-root full pathname. */
945 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
946 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
947 error = ENAMETOOLONG;
955 * Find the specified jail, or at least its parent.
956 * This abuses the file error codes ENOENT and EEXIST.
960 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
961 namelc = strrchr(name, '.');
962 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
966 sx_xlock(&allprison_lock);
969 * See if a requested jid already exists. There is an
970 * information leak here if the jid exists but is not within
971 * the caller's jail hierarchy. Jail creators will get EEXIST
972 * even though they cannot see the jail, and CREATE | UPDATE
973 * will return ENOENT which is not normally a valid error.
977 vfs_opterror(opts, "negative jid");
978 goto done_unlock_list;
980 pr = prison_find(jid);
983 /* Create: jid must not exist. */
984 if (cuflags == JAIL_CREATE) {
985 mtx_unlock(&pr->pr_mtx);
987 vfs_opterror(opts, "jail %d already exists",
989 goto done_unlock_list;
991 if (!prison_ischild(mypr, pr)) {
992 mtx_unlock(&pr->pr_mtx);
994 } else if (pr->pr_uref == 0) {
995 if (!(flags & JAIL_DYING)) {
996 mtx_unlock(&pr->pr_mtx);
998 vfs_opterror(opts, "jail %d is dying",
1000 goto done_unlock_list;
1001 } else if ((flags & JAIL_ATTACH) ||
1002 (pr_flags & PR_PERSIST)) {
1004 * A dying jail might be resurrected
1005 * (via attach or persist), but first
1006 * it must determine if another jail
1007 * has claimed its name. Accomplish
1008 * this by implicitly re-setting the
1012 name = prison_name(mypr, pr);
1017 /* Update: jid must exist. */
1018 if (cuflags == JAIL_UPDATE) {
1020 vfs_opterror(opts, "jail %d not found", jid);
1021 goto done_unlock_list;
1026 * If the caller provided a name, look for a jail by that name.
1027 * This has different semantics for creates and updates keyed by jid
1028 * (where the name must not already exist in a different jail),
1029 * and updates keyed by the name itself (where the name must exist
1030 * because that is the jail being updated).
1034 namelc = strrchr(name, '.');
1039 * This is a hierarchical name. Split it into the
1040 * parent and child names, and make sure the parent
1041 * exists or matches an already found jail.
1044 if (strncmp(name, ppr->pr_name, namelc - name)
1045 || ppr->pr_name[namelc - name] != '\0') {
1046 mtx_unlock(&pr->pr_mtx);
1049 "cannot change jail's parent");
1050 goto done_unlock_list;
1054 ppr = prison_find_name(mypr, name);
1058 "jail \"%s\" not found", name);
1059 goto done_unlock_list;
1061 mtx_unlock(&ppr->pr_mtx);
1066 if (namelc[0] != '\0') {
1068 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1071 FOREACH_PRISON_CHILD(ppr, tpr) {
1072 if (tpr != pr && tpr->pr_ref > 0 &&
1073 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1075 cuflags != JAIL_CREATE) {
1076 mtx_lock(&tpr->pr_mtx);
1077 if (tpr->pr_ref > 0) {
1082 if (tpr->pr_uref > 0) {
1088 mtx_unlock(&tpr->pr_mtx);
1089 } else if (tpr->pr_uref > 0) {
1091 * Create, or update(jid):
1092 * name must not exist in an
1093 * active sibling jail.
1097 mtx_unlock(&pr->pr_mtx);
1099 "jail \"%s\" already exists",
1101 goto done_unlock_list;
1105 /* If no active jail is found, use a dying one. */
1106 if (deadpr != NULL && pr == NULL) {
1107 if (flags & JAIL_DYING) {
1108 mtx_lock(&deadpr->pr_mtx);
1109 if (deadpr->pr_ref == 0) {
1110 mtx_unlock(&deadpr->pr_mtx);
1114 } else if (cuflags == JAIL_UPDATE) {
1117 "jail \"%s\" is dying", name);
1118 goto done_unlock_list;
1121 /* Update: name must exist if no jid. */
1122 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1124 vfs_opterror(opts, "jail \"%s\" not found",
1126 goto done_unlock_list;
1130 /* Update: must provide a jid or name. */
1131 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1133 vfs_opterror(opts, "update specified no jail");
1134 goto done_unlock_list;
1137 /* If there's no prison to update, create a new one and link it in. */
1139 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1140 if (tpr->pr_childcount >= tpr->pr_childmax) {
1142 vfs_opterror(opts, "prison limit exceeded");
1143 goto done_unlock_list;
1146 mtx_lock(&ppr->pr_mtx);
1147 if (ppr->pr_ref == 0) {
1148 mtx_unlock(&ppr->pr_mtx);
1150 vfs_opterror(opts, "jail \"%s\" not found",
1151 prison_name(mypr, ppr));
1152 goto done_unlock_list;
1156 mtx_unlock(&ppr->pr_mtx);
1157 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1159 /* Find the next free jid. */
1162 if (jid == JAIL_MAX)
1164 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1165 if (tpr->pr_id < jid)
1167 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1168 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1171 if (jid == lastprid) {
1174 "no available jail IDs");
1176 prison_deref(ppr, PD_DEREF |
1177 PD_DEUREF | PD_LIST_XLOCKED);
1186 * The jail already has a jid (that did not yet exist),
1187 * so just find where to insert it.
1189 TAILQ_FOREACH(tpr, &allprison, pr_list)
1190 if (tpr->pr_id >= jid) {
1191 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1196 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1197 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1198 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1199 tpr->pr_childcount++;
1201 pr->pr_parent = ppr;
1204 /* Set some default values, and inherit some from the parent. */
1209 root = mypr->pr_root;
1212 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1213 pr->pr_flags |= PR_HOST;
1214 #if defined(INET) || defined(INET6)
1216 if (!(pr_flags & PR_VNET))
1220 if (!(ch_flags & PR_IP4_USER))
1221 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1222 else if (!(pr_flags & PR_IP4_USER)) {
1223 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1224 if (ppr->pr_ip4 != NULL) {
1225 pr->pr_ip4s = ppr->pr_ip4s;
1226 pr->pr_ip4 = malloc(pr->pr_ip4s *
1227 sizeof(struct in_addr), M_PRISON,
1229 bcopy(ppr->pr_ip4, pr->pr_ip4,
1230 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1235 if (!(ch_flags & PR_IP6_USER))
1236 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1237 else if (!(pr_flags & PR_IP6_USER)) {
1238 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1239 if (ppr->pr_ip6 != NULL) {
1240 pr->pr_ip6s = ppr->pr_ip6s;
1241 pr->pr_ip6 = malloc(pr->pr_ip6s *
1242 sizeof(struct in6_addr), M_PRISON,
1244 bcopy(ppr->pr_ip6, pr->pr_ip6,
1245 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1251 /* Source address selection is always on by default. */
1252 pr->pr_flags |= _PR_IP_SADDRSEL;
1254 pr->pr_securelevel = ppr->pr_securelevel;
1255 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1256 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1257 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1259 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1260 if (osrelstr == NULL)
1261 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1263 strcpy(pr->pr_osrelease, osrelstr);
1265 LIST_INIT(&pr->pr_children);
1266 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1267 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1270 /* Allocate a new vnet if specified. */
1271 pr->pr_vnet = (pr_flags & PR_VNET)
1272 ? vnet_alloc() : ppr->pr_vnet;
1275 * Allocate a dedicated cpuset for each jail.
1276 * Unlike other initial settings, this may return an erorr.
1278 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1280 prison_deref(pr, PD_LIST_XLOCKED);
1284 mtx_lock(&pr->pr_mtx);
1286 * New prisons do not yet have a reference, because we do not
1287 * want others to see the incomplete prison once the
1288 * allprison_lock is downgraded.
1293 * Grab a reference for existing prisons, to ensure they
1294 * continue to exist for the duration of the call.
1297 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1298 if ((pr->pr_flags & PR_VNET) &&
1299 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1302 "vnet jails cannot have IP address restrictions");
1303 goto done_deref_locked;
1307 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1310 "ip4 cannot be changed after creation");
1311 goto done_deref_locked;
1315 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1318 "ip6 cannot be changed after creation");
1319 goto done_deref_locked;
1324 /* Do final error checking before setting anything. */
1326 if (slevel < ppr->pr_securelevel) {
1328 goto done_deref_locked;
1332 if (childmax >= ppr->pr_childmax) {
1334 goto done_deref_locked;
1338 if (enforce < ppr->pr_enforce_statfs) {
1340 goto done_deref_locked;
1345 * devfs_rsnum is a uint16_t
1347 if (rsnum < 0 || rsnum > 65535) {
1349 goto done_deref_locked;
1352 * Nested jails always inherit parent's devfs ruleset
1354 if (jailed(td->td_ucred)) {
1355 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1357 goto done_deref_locked;
1359 rsnum = ppr->pr_devfs_rsnum;
1364 if (ppr->pr_flags & PR_IP4) {
1366 * Make sure the new set of IP addresses is a
1367 * subset of the parent's list. Don't worry
1368 * about the parent being unlocked, as any
1369 * setting is done with allprison_lock held.
1371 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1372 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1374 if (ij == ppr->pr_ip4s) {
1376 goto done_deref_locked;
1379 for (ii = ij = 1; ii < ip4s; ii++) {
1380 if (ip4[ii].s_addr ==
1381 ppr->pr_ip4[0].s_addr)
1383 for (; ij < ppr->pr_ip4s; ij++)
1384 if (ip4[ii].s_addr ==
1385 ppr->pr_ip4[ij].s_addr)
1387 if (ij == ppr->pr_ip4s)
1390 if (ij == ppr->pr_ip4s) {
1392 goto done_deref_locked;
1397 * Check for conflicting IP addresses. We permit them
1398 * if there is no more than one IP on each jail. If
1399 * there is a duplicate on a jail with more than one
1400 * IP stop checking and return error.
1403 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1404 if (tppr->pr_flags & PR_VNET)
1409 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1412 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1414 tpr->pr_uref == 0) {
1418 if (!(tpr->pr_flags & PR_IP4_USER))
1421 if (tpr->pr_ip4 == NULL ||
1422 (ip4s == 1 && tpr->pr_ip4s == 1))
1424 for (ii = 0; ii < ip4s; ii++) {
1425 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1429 "IPv4 addresses clash");
1430 goto done_deref_locked;
1438 if (ppr->pr_flags & PR_IP6) {
1440 * Make sure the new set of IP addresses is a
1441 * subset of the parent's list.
1443 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1444 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1447 if (ij == ppr->pr_ip6s) {
1449 goto done_deref_locked;
1452 for (ii = ij = 1; ii < ip6s; ii++) {
1453 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1456 for (; ij < ppr->pr_ip6s; ij++)
1457 if (IN6_ARE_ADDR_EQUAL(
1458 &ip6[ii], &ppr->pr_ip6[ij]))
1460 if (ij == ppr->pr_ip6s)
1463 if (ij == ppr->pr_ip6s) {
1465 goto done_deref_locked;
1469 /* Check for conflicting IP addresses. */
1471 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1472 if (tppr->pr_flags & PR_VNET)
1477 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1480 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1482 tpr->pr_uref == 0) {
1486 if (!(tpr->pr_flags & PR_IP6_USER))
1489 if (tpr->pr_ip6 == NULL ||
1490 (ip6s == 1 && tpr->pr_ip6s == 1))
1492 for (ii = 0; ii < ip6s; ii++) {
1493 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1497 "IPv6 addresses clash");
1498 goto done_deref_locked;
1504 onamelen = namelen = 0;
1505 if (namelc != NULL) {
1506 /* Give a default name of the jid. Also allow the name to be
1507 * explicitly the jid - but not any other number, and only in
1508 * normal form (no leading zero/etc).
1510 if (namelc[0] == '\0')
1511 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1512 else if ((strtoul(namelc, &p, 10) != jid ||
1513 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1516 "name cannot be numeric (unless it is the jid)");
1517 goto done_deref_locked;
1520 * Make sure the name isn't too long for the prison or its
1523 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1524 onamelen = strlen(pr->pr_name + pnamelen);
1525 namelen = strlen(namelc);
1526 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1527 error = ENAMETOOLONG;
1528 goto done_deref_locked;
1530 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1531 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1532 sizeof(pr->pr_name)) {
1533 error = ENAMETOOLONG;
1534 goto done_deref_locked;
1538 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1539 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1541 goto done_deref_locked;
1545 * Let modules check their parameters. This requires unlocking and
1546 * then re-locking the prison, but this is still a valid state as long
1547 * as allprison_lock remains xlocked.
1549 mtx_unlock(&pr->pr_mtx);
1550 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1552 prison_deref(pr, created
1554 : PD_DEREF | PD_LIST_XLOCKED);
1557 mtx_lock(&pr->pr_mtx);
1559 /* At this point, all valid parameters should have been noted. */
1560 TAILQ_FOREACH(opt, opts, link) {
1561 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1563 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1564 goto done_deref_locked;
1568 /* Set the parameters of the prison. */
1571 if (pr_flags & PR_IP4_USER) {
1572 pr->pr_flags |= PR_IP4;
1573 free(pr->pr_ip4, M_PRISON);
1577 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1579 if (tpr->pr_flags & PR_VNET) {
1584 if (prison_restrict_ip4(tpr, NULL)) {
1593 if (pr_flags & PR_IP6_USER) {
1594 pr->pr_flags |= PR_IP6;
1595 free(pr->pr_ip6, M_PRISON);
1599 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1601 if (tpr->pr_flags & PR_VNET) {
1606 if (prison_restrict_ip6(tpr, NULL)) {
1614 pr->pr_securelevel = slevel;
1615 /* Set all child jails to be at least this level. */
1616 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1617 if (tpr->pr_securelevel < slevel)
1618 tpr->pr_securelevel = slevel;
1621 pr->pr_childmax = childmax;
1622 /* Set all child jails to under this limit. */
1623 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1624 if (tpr->pr_childmax > childmax - level)
1625 tpr->pr_childmax = childmax > level
1626 ? childmax - level : 0;
1629 pr->pr_enforce_statfs = enforce;
1630 /* Pass this restriction on to the children. */
1631 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1632 if (tpr->pr_enforce_statfs < enforce)
1633 tpr->pr_enforce_statfs = enforce;
1636 pr->pr_devfs_rsnum = rsnum;
1637 /* Pass this restriction on to the children. */
1638 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1639 tpr->pr_devfs_rsnum = rsnum;
1641 if (namelc != NULL) {
1642 if (ppr == &prison0)
1643 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1645 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1646 ppr->pr_name, namelc);
1647 /* Change this component of child names. */
1648 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1649 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1650 strlen(tpr->pr_name + onamelen) + 1);
1651 bcopy(pr->pr_name, tpr->pr_name, namelen);
1655 /* Try to keep a real-rooted full pathname. */
1656 if (fullpath_disabled && path[0] == '/' &&
1657 strcmp(mypr->pr_path, "/"))
1658 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1659 mypr->pr_path, path);
1661 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1664 if (PR_HOST & ch_flags & ~pr_flags) {
1665 if (pr->pr_flags & PR_HOST) {
1667 * Copy the parent's host info. As with pr_ip4 above,
1668 * the lack of a lock on the parent is not a problem;
1669 * it is always set with allprison_lock at least
1670 * shared, and is held exclusively here.
1672 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1673 sizeof(pr->pr_hostname));
1674 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1675 sizeof(pr->pr_domainname));
1676 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1677 sizeof(pr->pr_hostuuid));
1678 pr->pr_hostid = pr->pr_parent->pr_hostid;
1680 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1681 /* Set this prison, and any descendants without PR_HOST. */
1683 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1685 strlcpy(pr->pr_domainname, domain,
1686 sizeof(pr->pr_domainname));
1688 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1690 pr->pr_hostid = hid;
1691 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1692 if (tpr->pr_flags & PR_HOST)
1696 strlcpy(tpr->pr_hostname,
1698 sizeof(tpr->pr_hostname));
1700 strlcpy(tpr->pr_domainname,
1702 sizeof(tpr->pr_domainname));
1704 strlcpy(tpr->pr_hostuuid,
1706 sizeof(tpr->pr_hostuuid));
1708 tpr->pr_hostid = hid;
1712 if ((tallow = ch_allow & ~pr_allow)) {
1713 /* Clear allow bits in all children. */
1714 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1715 tpr->pr_allow &= ~tallow;
1717 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1719 * Persistent prisons get an extra reference, and prisons losing their
1720 * persist flag lose that reference. Only do this for existing prisons
1721 * for now, so new ones will remain unseen until after the module
1722 * handlers have completed.
1724 born = pr->pr_uref == 0;
1725 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1726 if (pr_flags & PR_PERSIST) {
1734 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1735 mtx_unlock(&pr->pr_mtx);
1738 if (racct_enable && created)
1739 prison_racct_attach(pr);
1742 /* Locks may have prevented a complete restriction of child IP
1743 * addresses. If so, allocate some more memory and try again.
1748 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1749 mtx_lock(&pr->pr_mtx);
1751 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1753 if (tpr->pr_flags & PR_VNET) {
1758 if (prison_restrict_ip4(tpr, ip4)) {
1765 mtx_unlock(&pr->pr_mtx);
1771 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1772 mtx_lock(&pr->pr_mtx);
1774 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1776 if (tpr->pr_flags & PR_VNET) {
1781 if (prison_restrict_ip6(tpr, ip6)) {
1788 mtx_unlock(&pr->pr_mtx);
1792 /* Let the modules do their work. */
1793 sx_downgrade(&allprison_lock);
1795 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1797 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1798 prison_deref(pr, created
1800 : PD_DEREF | PD_LIST_SLOCKED);
1804 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1807 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1808 prison_deref(pr, created
1810 : PD_DEREF | PD_LIST_SLOCKED);
1814 /* Attach this process to the prison if requested. */
1815 if (flags & JAIL_ATTACH) {
1816 mtx_lock(&pr->pr_mtx);
1817 error = do_jail_attach(td, pr);
1819 vfs_opterror(opts, "attach failed");
1821 prison_deref(pr, PD_DEREF);
1827 if (racct_enable && !created) {
1828 if (!(flags & JAIL_ATTACH))
1829 sx_sunlock(&allprison_lock);
1830 prison_racct_modify(pr);
1831 if (!(flags & JAIL_ATTACH))
1832 sx_slock(&allprison_lock);
1836 td->td_retval[0] = pr->pr_id;
1839 * Now that it is all there, drop the temporary reference from existing
1840 * prisons. Or add a reference to newly created persistent prisons
1841 * (which was not done earlier so that the prison would not be publicly
1845 prison_deref(pr, (flags & JAIL_ATTACH)
1847 : PD_DEREF | PD_LIST_SLOCKED);
1849 if (pr_flags & PR_PERSIST) {
1850 mtx_lock(&pr->pr_mtx);
1853 mtx_unlock(&pr->pr_mtx);
1855 if (!(flags & JAIL_ATTACH))
1856 sx_sunlock(&allprison_lock);
1862 prison_deref(pr, created
1863 ? PD_LOCKED | PD_LIST_XLOCKED
1864 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1867 sx_xunlock(&allprison_lock);
1873 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1874 &errmsg_len) == 0 && errmsg_len > 0) {
1875 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1876 if (optuio->uio_segflg == UIO_SYSSPACE)
1878 optuio->uio_iov[errmsg_pos].iov_base,
1882 optuio->uio_iov[errmsg_pos].iov_base,
1888 free(ip4, M_PRISON);
1891 free(ip6, M_PRISON);
1894 free(g_path, M_TEMP);
1901 * struct jail_get_args {
1902 * struct iovec *iovp;
1903 * unsigned int iovcnt;
1908 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1913 /* Check that we have an even number of iovecs. */
1914 if (uap->iovcnt & 1)
1917 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1920 error = kern_jail_get(td, auio, uap->flags);
1922 error = copyout(auio->uio_iov, uap->iovp,
1923 uap->iovcnt * sizeof (struct iovec));
1929 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1931 struct bool_flags *bf;
1932 struct jailsys_flags *jsf;
1933 struct prison *pr, *mypr;
1935 struct vfsoptlist *opts;
1936 char *errmsg, *name;
1937 int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1940 if (flags & ~JAIL_GET_MASK)
1943 /* Get the parameter list. */
1944 error = vfs_buildopts(optuio, &opts);
1947 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1948 mypr = td->td_ucred->cr_prison;
1951 * Find the prison specified by one of: lastjid, jid, name.
1953 sx_slock(&allprison_lock);
1954 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1956 TAILQ_FOREACH(pr, &allprison, pr_list) {
1957 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1958 mtx_lock(&pr->pr_mtx);
1959 if (pr->pr_ref > 0 &&
1960 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1962 mtx_unlock(&pr->pr_mtx);
1968 vfs_opterror(opts, "no jail after %d", jid);
1969 goto done_unlock_list;
1970 } else if (error != ENOENT)
1971 goto done_unlock_list;
1973 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1976 pr = prison_find_child(mypr, jid);
1978 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1979 mtx_unlock(&pr->pr_mtx);
1981 vfs_opterror(opts, "jail %d is dying",
1983 goto done_unlock_list;
1988 vfs_opterror(opts, "jail %d not found", jid);
1989 goto done_unlock_list;
1991 } else if (error != ENOENT)
1992 goto done_unlock_list;
1994 error = vfs_getopt(opts, "name", (void **)&name, &len);
1996 if (len == 0 || name[len - 1] != '\0') {
1998 goto done_unlock_list;
2000 pr = prison_find_name(mypr, name);
2002 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2003 mtx_unlock(&pr->pr_mtx);
2005 vfs_opterror(opts, "jail \"%s\" is dying",
2007 goto done_unlock_list;
2012 vfs_opterror(opts, "jail \"%s\" not found", name);
2013 goto done_unlock_list;
2014 } else if (error != ENOENT)
2015 goto done_unlock_list;
2017 vfs_opterror(opts, "no jail specified");
2019 goto done_unlock_list;
2022 /* Get the parameters of the prison. */
2025 td->td_retval[0] = pr->pr_id;
2026 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2027 if (error != 0 && error != ENOENT)
2029 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2030 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2031 if (error != 0 && error != ENOENT)
2033 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2034 if (error != 0 && error != ENOENT)
2036 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2037 sizeof(pr->pr_cpuset->cs_id));
2038 if (error != 0 && error != ENOENT)
2040 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2041 if (error != 0 && error != ENOENT)
2044 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2045 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2046 if (error != 0 && error != ENOENT)
2050 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2051 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2052 if (error != 0 && error != ENOENT)
2055 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2056 sizeof(pr->pr_securelevel));
2057 if (error != 0 && error != ENOENT)
2059 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2060 sizeof(pr->pr_childcount));
2061 if (error != 0 && error != ENOENT)
2063 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2064 sizeof(pr->pr_childmax));
2065 if (error != 0 && error != ENOENT)
2067 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2068 if (error != 0 && error != ENOENT)
2070 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2071 if (error != 0 && error != ENOENT)
2073 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2074 if (error != 0 && error != ENOENT)
2076 #ifdef COMPAT_FREEBSD32
2077 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2078 uint32_t hid32 = pr->pr_hostid;
2080 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2083 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2084 sizeof(pr->pr_hostid));
2085 if (error != 0 && error != ENOENT)
2087 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2088 sizeof(pr->pr_enforce_statfs));
2089 if (error != 0 && error != ENOENT)
2091 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2092 sizeof(pr->pr_devfs_rsnum));
2093 if (error != 0 && error != ENOENT)
2095 for (bf = pr_flag_bool;
2096 bf < pr_flag_bool + nitems(pr_flag_bool);
2098 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2099 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2100 if (error != 0 && error != ENOENT)
2103 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2104 if (error != 0 && error != ENOENT)
2107 for (jsf = pr_flag_jailsys;
2108 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2110 f = pr->pr_flags & (jsf->disable | jsf->new);
2111 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2112 : (f == jsf->new) ? JAIL_SYS_NEW
2114 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2115 if (error != 0 && error != ENOENT)
2118 for (bf = pr_flag_allow;
2119 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2121 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2122 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2123 if (error != 0 && error != ENOENT)
2126 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2127 if (error != 0 && error != ENOENT)
2130 i = (pr->pr_uref == 0);
2131 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2132 if (error != 0 && error != ENOENT)
2135 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2136 if (error != 0 && error != ENOENT)
2138 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2139 sizeof(pr->pr_osreldate));
2140 if (error != 0 && error != ENOENT)
2142 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2143 if (error != 0 && error != ENOENT)
2146 /* Get the module parameters. */
2147 mtx_unlock(&pr->pr_mtx);
2149 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2152 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2154 /* By now, all parameters should have been noted. */
2155 TAILQ_FOREACH(opt, opts, link) {
2156 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2158 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2163 /* Write the fetched parameters back to userspace. */
2165 TAILQ_FOREACH(opt, opts, link) {
2166 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2167 pos = 2 * opt->pos + 1;
2168 optuio->uio_iov[pos].iov_len = opt->len;
2169 if (opt->value != NULL) {
2170 if (optuio->uio_segflg == UIO_SYSSPACE) {
2172 optuio->uio_iov[pos].iov_base,
2175 error = copyout(opt->value,
2176 optuio->uio_iov[pos].iov_base,
2187 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2191 sx_sunlock(&allprison_lock);
2193 if (error && errmsg_pos >= 0) {
2194 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2195 errmsg_pos = 2 * errmsg_pos + 1;
2196 if (errmsg_len > 0) {
2197 if (optuio->uio_segflg == UIO_SYSSPACE)
2199 optuio->uio_iov[errmsg_pos].iov_base,
2203 optuio->uio_iov[errmsg_pos].iov_base,
2213 * struct jail_remove_args {
2218 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2220 struct prison *pr, *cpr, *lpr, *tpr;
2223 error = priv_check(td, PRIV_JAIL_REMOVE);
2227 sx_xlock(&allprison_lock);
2228 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2230 sx_xunlock(&allprison_lock);
2234 /* Remove all descendants of this prison, then remove this prison. */
2236 if (!LIST_EMPTY(&pr->pr_children)) {
2237 mtx_unlock(&pr->pr_mtx);
2239 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2240 mtx_lock(&cpr->pr_mtx);
2241 if (cpr->pr_ref > 0) {
2245 /* Already removed - do not do it again. */
2248 mtx_unlock(&cpr->pr_mtx);
2250 mtx_lock(&lpr->pr_mtx);
2251 prison_remove_one(lpr);
2252 sx_xlock(&allprison_lock);
2257 mtx_lock(&lpr->pr_mtx);
2258 prison_remove_one(lpr);
2259 sx_xlock(&allprison_lock);
2261 mtx_lock(&pr->pr_mtx);
2263 prison_remove_one(pr);
2268 prison_remove_one(struct prison *pr)
2273 /* If the prison was persistent, it is not anymore. */
2275 if (pr->pr_flags & PR_PERSIST) {
2278 pr->pr_flags &= ~PR_PERSIST;
2282 * jail_remove added a reference. If that's the only one, remove
2285 KASSERT(pr->pr_ref > 0,
2286 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2287 if (pr->pr_ref == 1) {
2289 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2293 mtx_unlock(&pr->pr_mtx);
2294 sx_xunlock(&allprison_lock);
2296 * Kill all processes unfortunate enough to be attached to this prison.
2298 sx_slock(&allproc_lock);
2299 FOREACH_PROC_IN_SYSTEM(p) {
2301 if (p->p_state != PRS_NEW && p->p_ucred &&
2302 p->p_ucred->cr_prison == pr)
2303 kern_psignal(p, SIGKILL);
2306 sx_sunlock(&allproc_lock);
2307 /* Remove the temporary reference added by jail_remove. */
2308 prison_deref(pr, deuref | PD_DEREF);
2313 * struct jail_attach_args {
2318 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2323 error = priv_check(td, PRIV_JAIL_ATTACH);
2328 * Start with exclusive hold on allprison_lock to ensure that a possible
2329 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2330 * But then immediately downgrade it since we don't need to stop
2333 sx_xlock(&allprison_lock);
2334 sx_downgrade(&allprison_lock);
2335 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2337 sx_sunlock(&allprison_lock);
2342 * Do not allow a process to attach to a prison that is not
2343 * considered to be "alive".
2345 if (pr->pr_uref == 0) {
2346 mtx_unlock(&pr->pr_mtx);
2347 sx_sunlock(&allprison_lock);
2351 return (do_jail_attach(td, pr));
2355 do_jail_attach(struct thread *td, struct prison *pr)
2358 struct ucred *newcred, *oldcred;
2362 * XXX: Note that there is a slight race here if two threads
2363 * in the same privileged process attempt to attach to two
2364 * different jails at the same time. It is important for
2365 * user processes not to do this, or they might end up with
2366 * a process root from one prison, but attached to the jail
2371 mtx_unlock(&pr->pr_mtx);
2373 /* Let modules do whatever they need to prepare for attaching. */
2374 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2376 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2379 sx_sunlock(&allprison_lock);
2382 * Reparent the newly attached process to this jail.
2385 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2389 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2390 if ((error = change_dir(pr->pr_root, td)) != 0)
2393 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2396 VOP_UNLOCK(pr->pr_root, 0);
2397 if ((error = pwd_chroot(td, pr->pr_root)))
2402 oldcred = crcopysafe(p, newcred);
2403 newcred->cr_prison = pr;
2404 proc_set_cred(p, newcred);
2407 racct_proc_ucred_changed(p, oldcred, newcred);
2412 rctl_proc_ucred_changed(p, newcred);
2415 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2420 VOP_UNLOCK(pr->pr_root, 0);
2422 /* Tell modules this thread is still in its old jail after all. */
2423 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2424 prison_deref(pr, PD_DEREF | PD_DEUREF);
2430 * Returns a locked prison instance, or NULL on failure.
2433 prison_find(int prid)
2437 sx_assert(&allprison_lock, SX_LOCKED);
2438 TAILQ_FOREACH(pr, &allprison, pr_list) {
2439 if (pr->pr_id == prid) {
2440 mtx_lock(&pr->pr_mtx);
2443 mtx_unlock(&pr->pr_mtx);
2450 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2453 prison_find_child(struct prison *mypr, int prid)
2458 sx_assert(&allprison_lock, SX_LOCKED);
2459 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2460 if (pr->pr_id == prid) {
2461 mtx_lock(&pr->pr_mtx);
2464 mtx_unlock(&pr->pr_mtx);
2471 * Look for the name relative to mypr. Returns a locked prison or NULL.
2474 prison_find_name(struct prison *mypr, const char *name)
2476 struct prison *pr, *deadpr;
2480 sx_assert(&allprison_lock, SX_LOCKED);
2481 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2484 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2485 if (!strcmp(pr->pr_name + mylen, name)) {
2486 mtx_lock(&pr->pr_mtx);
2487 if (pr->pr_ref > 0) {
2488 if (pr->pr_uref > 0)
2492 mtx_unlock(&pr->pr_mtx);
2495 /* There was no valid prison - perhaps there was a dying one. */
2496 if (deadpr != NULL) {
2497 mtx_lock(&deadpr->pr_mtx);
2498 if (deadpr->pr_ref == 0) {
2499 mtx_unlock(&deadpr->pr_mtx);
2507 * See if a prison has the specific flag set.
2510 prison_flag(struct ucred *cred, unsigned flag)
2513 /* This is an atomic read, so no locking is necessary. */
2514 return (cred->cr_prison->pr_flags & flag);
2518 prison_allow(struct ucred *cred, unsigned flag)
2521 /* This is an atomic read, so no locking is necessary. */
2522 return (cred->cr_prison->pr_allow & flag);
2526 * Remove a prison reference. If that was the last reference, remove the
2527 * prison itself - but not in this context in case there are locks held.
2530 prison_free_locked(struct prison *pr)
2534 mtx_assert(&pr->pr_mtx, MA_OWNED);
2536 mtx_unlock(&pr->pr_mtx);
2538 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2542 prison_free(struct prison *pr)
2545 mtx_lock(&pr->pr_mtx);
2546 prison_free_locked(pr);
2550 * Complete a call to either prison_free or prison_proc_free.
2553 prison_complete(void *context, int pending)
2555 struct prison *pr = context;
2557 sx_xlock(&allprison_lock);
2558 mtx_lock(&pr->pr_mtx);
2559 prison_deref(pr, pr->pr_uref
2560 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2561 : PD_LOCKED | PD_LIST_XLOCKED);
2565 * Remove a prison reference (usually). This internal version assumes no
2566 * mutexes are held, except perhaps the prison itself. If there are no more
2567 * references, release and delist the prison. On completion, the prison lock
2568 * and the allprison lock are both unlocked.
2571 prison_deref(struct prison *pr, int flags)
2573 struct prison *ppr, *tpr;
2576 if (!(flags & PD_LOCKED))
2577 mtx_lock(&pr->pr_mtx);
2579 if (flags & PD_DEUREF) {
2580 KASSERT(pr->pr_uref > 0,
2581 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2584 lasturef = pr->pr_uref == 0;
2587 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2590 if (flags & PD_DEREF) {
2591 KASSERT(pr->pr_ref > 0,
2592 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2597 mtx_unlock(&pr->pr_mtx);
2600 * Tell the modules if the last user reference was removed
2601 * (even it sticks around in dying state).
2604 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2605 sx_xlock(&allprison_lock);
2606 flags |= PD_LIST_XLOCKED;
2608 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2609 mtx_lock(&pr->pr_mtx);
2611 mtx_unlock(&pr->pr_mtx);
2614 /* If the prison still has references, nothing else to do. */
2616 if (flags & PD_LIST_SLOCKED)
2617 sx_sunlock(&allprison_lock);
2618 else if (flags & PD_LIST_XLOCKED)
2619 sx_xunlock(&allprison_lock);
2623 if (flags & PD_LIST_SLOCKED) {
2624 if (!sx_try_upgrade(&allprison_lock)) {
2625 sx_sunlock(&allprison_lock);
2626 sx_xlock(&allprison_lock);
2628 } else if (!(flags & PD_LIST_XLOCKED))
2629 sx_xlock(&allprison_lock);
2631 TAILQ_REMOVE(&allprison, pr, pr_list);
2632 LIST_REMOVE(pr, pr_sibling);
2633 ppr = pr->pr_parent;
2634 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2635 tpr->pr_childcount--;
2636 sx_xunlock(&allprison_lock);
2639 if (pr->pr_vnet != ppr->pr_vnet)
2640 vnet_destroy(pr->pr_vnet);
2642 if (pr->pr_root != NULL)
2644 mtx_destroy(&pr->pr_mtx);
2646 free(pr->pr_ip4, M_PRISON);
2649 free(pr->pr_ip6, M_PRISON);
2651 if (pr->pr_cpuset != NULL)
2652 cpuset_rel(pr->pr_cpuset);
2656 prison_racct_detach(pr);
2660 /* Removing a prison frees a reference on its parent. */
2662 mtx_lock(&pr->pr_mtx);
2663 flags = PD_DEREF | PD_DEUREF;
2668 prison_hold_locked(struct prison *pr)
2671 mtx_assert(&pr->pr_mtx, MA_OWNED);
2672 KASSERT(pr->pr_ref > 0,
2673 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2678 prison_hold(struct prison *pr)
2681 mtx_lock(&pr->pr_mtx);
2682 prison_hold_locked(pr);
2683 mtx_unlock(&pr->pr_mtx);
2687 prison_proc_hold(struct prison *pr)
2690 mtx_lock(&pr->pr_mtx);
2691 KASSERT(pr->pr_uref > 0,
2692 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2694 mtx_unlock(&pr->pr_mtx);
2698 prison_proc_free(struct prison *pr)
2701 mtx_lock(&pr->pr_mtx);
2702 KASSERT(pr->pr_uref > 0,
2703 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2704 if (pr->pr_uref > 1)
2708 * Don't remove the last user reference in this context, which
2709 * is expected to be a process that is not only locked, but
2713 mtx_unlock(&pr->pr_mtx);
2714 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2717 mtx_unlock(&pr->pr_mtx);
2721 * Check if a jail supports the given address family.
2723 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2727 prison_check_af(struct ucred *cred, int af)
2732 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2734 pr = cred->cr_prison;
2736 /* Prisons with their own network stack are not limited. */
2737 if (prison_owns_vnet(cred))
2746 if (pr->pr_flags & PR_IP4)
2748 mtx_lock(&pr->pr_mtx);
2749 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2750 error = EAFNOSUPPORT;
2751 mtx_unlock(&pr->pr_mtx);
2757 if (pr->pr_flags & PR_IP6)
2759 mtx_lock(&pr->pr_mtx);
2760 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2761 error = EAFNOSUPPORT;
2762 mtx_unlock(&pr->pr_mtx);
2770 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2771 error = EAFNOSUPPORT;
2777 * Check if given address belongs to the jail referenced by cred (wrapper to
2778 * prison_check_ip[46]).
2780 * Returns 0 if jail doesn't restrict the address family or if address belongs
2781 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2782 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
2785 prison_if(struct ucred *cred, struct sockaddr *sa)
2788 struct sockaddr_in *sai;
2791 struct sockaddr_in6 *sai6;
2795 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2796 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2799 if (prison_owns_vnet(cred))
2804 switch (sa->sa_family)
2808 sai = (struct sockaddr_in *)sa;
2809 error = prison_check_ip4(cred, &sai->sin_addr);
2814 sai6 = (struct sockaddr_in6 *)sa;
2815 error = prison_check_ip6(cred, &sai6->sin6_addr);
2819 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2820 error = EAFNOSUPPORT;
2826 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2829 prison_check(struct ucred *cred1, struct ucred *cred2)
2832 return ((cred1->cr_prison == cred2->cr_prison ||
2833 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2837 * Return 1 if p2 is a child of p1, otherwise 0.
2840 prison_ischild(struct prison *pr1, struct prison *pr2)
2843 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2850 * Return 1 if the passed credential is in a jail, otherwise 0.
2853 jailed(struct ucred *cred)
2856 return (cred->cr_prison != &prison0);
2860 * Return 1 if the passed credential is in a jail and that jail does not
2861 * have its own virtual network stack, otherwise 0.
2864 jailed_without_vnet(struct ucred *cred)
2870 if (prison_owns_vnet(cred))
2878 * Return the correct hostname (domainname, et al) for the passed credential.
2881 getcredhostname(struct ucred *cred, char *buf, size_t size)
2886 * A NULL credential can be used to shortcut to the physical
2887 * system's hostname.
2889 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2890 mtx_lock(&pr->pr_mtx);
2891 strlcpy(buf, pr->pr_hostname, size);
2892 mtx_unlock(&pr->pr_mtx);
2896 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2899 mtx_lock(&cred->cr_prison->pr_mtx);
2900 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2901 mtx_unlock(&cred->cr_prison->pr_mtx);
2905 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2908 mtx_lock(&cred->cr_prison->pr_mtx);
2909 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2910 mtx_unlock(&cred->cr_prison->pr_mtx);
2914 getcredhostid(struct ucred *cred, unsigned long *hostid)
2917 mtx_lock(&cred->cr_prison->pr_mtx);
2918 *hostid = cred->cr_prison->pr_hostid;
2919 mtx_unlock(&cred->cr_prison->pr_mtx);
2924 * Determine whether the prison represented by cred owns
2925 * its vnet rather than having it inherited.
2927 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2930 prison_owns_vnet(struct ucred *cred)
2934 * vnets cannot be added/removed after jail creation,
2935 * so no need to lock here.
2937 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2942 * Determine whether the subject represented by cred can "see"
2943 * status of a mount point.
2944 * Returns: 0 for permitted, ENOENT otherwise.
2945 * XXX: This function should be called cr_canseemount() and should be
2946 * placed in kern_prot.c.
2949 prison_canseemount(struct ucred *cred, struct mount *mp)
2955 pr = cred->cr_prison;
2956 if (pr->pr_enforce_statfs == 0)
2958 if (pr->pr_root->v_mount == mp)
2960 if (pr->pr_enforce_statfs == 2)
2963 * If jail's chroot directory is set to "/" we should be able to see
2964 * all mount-points from inside a jail.
2965 * This is ugly check, but this is the only situation when jail's
2966 * directory ends with '/'.
2968 if (strcmp(pr->pr_path, "/") == 0)
2970 len = strlen(pr->pr_path);
2972 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2975 * Be sure that we don't have situation where jail's root directory
2976 * is "/some/path" and mount point is "/some/pathpath".
2978 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2984 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2986 char jpath[MAXPATHLEN];
2990 pr = cred->cr_prison;
2991 if (pr->pr_enforce_statfs == 0)
2993 if (prison_canseemount(cred, mp) != 0) {
2994 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2995 strlcpy(sp->f_mntonname, "[restricted]",
2996 sizeof(sp->f_mntonname));
2999 if (pr->pr_root->v_mount == mp) {
3001 * Clear current buffer data, so we are sure nothing from
3002 * the valid path left there.
3004 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3005 *sp->f_mntonname = '/';
3009 * If jail's chroot directory is set to "/" we should be able to see
3010 * all mount-points from inside a jail.
3012 if (strcmp(pr->pr_path, "/") == 0)
3014 len = strlen(pr->pr_path);
3015 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3017 * Clear current buffer data, so we are sure nothing from
3018 * the valid path left there.
3020 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3021 if (*jpath == '\0') {
3022 /* Should never happen. */
3023 *sp->f_mntonname = '/';
3025 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3030 * Check with permission for a specific privilege is granted within jail. We
3031 * have a specific list of accepted privileges; the rest are denied.
3034 prison_priv_check(struct ucred *cred, int priv)
3042 * Privileges specific to prisons with a virtual network stack.
3043 * There might be a duplicate entry here in case the privilege
3044 * is only granted conditionally in the legacy jail case.
3049 * NFS-specific privileges.
3051 case PRIV_NFS_DAEMON:
3052 case PRIV_NFS_LOCKD:
3055 * Network stack privileges.
3057 case PRIV_NET_BRIDGE:
3060 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3061 case PRIV_NET_ROUTE:
3063 case PRIV_NET_SETIFMTU:
3064 case PRIV_NET_SETIFFLAGS:
3065 case PRIV_NET_SETIFCAP:
3066 case PRIV_NET_SETIFDESCR:
3067 case PRIV_NET_SETIFNAME :
3068 case PRIV_NET_SETIFMETRIC:
3069 case PRIV_NET_SETIFPHYS:
3070 case PRIV_NET_SETIFMAC:
3071 case PRIV_NET_SETLANPCP:
3072 case PRIV_NET_ADDMULTI:
3073 case PRIV_NET_DELMULTI:
3074 case PRIV_NET_HWIOCTL:
3075 case PRIV_NET_SETLLADDR:
3076 case PRIV_NET_ADDIFGROUP:
3077 case PRIV_NET_DELIFGROUP:
3078 case PRIV_NET_IFCREATE:
3079 case PRIV_NET_IFDESTROY:
3080 case PRIV_NET_ADDIFADDR:
3081 case PRIV_NET_DELIFADDR:
3084 case PRIV_NET_SETIFVNET:
3085 case PRIV_NET_SETIFFIB:
3088 * 802.11-related privileges.
3090 case PRIV_NET80211_GETKEY:
3092 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3099 case PRIV_NETATM_CFG:
3100 case PRIV_NETATM_ADD:
3101 case PRIV_NETATM_DEL:
3102 case PRIV_NETATM_SET:
3105 * Bluetooth privileges.
3107 case PRIV_NETBLUETOOTH_RAW:
3111 * Netgraph and netgraph module privileges.
3113 case PRIV_NETGRAPH_CONTROL:
3115 case PRIV_NETGRAPH_TTY:
3119 * IPv4 and IPv6 privileges.
3121 case PRIV_NETINET_IPFW:
3122 case PRIV_NETINET_DIVERT:
3123 case PRIV_NETINET_PF:
3124 case PRIV_NETINET_DUMMYNET:
3125 case PRIV_NETINET_CARP:
3126 case PRIV_NETINET_MROUTE:
3127 case PRIV_NETINET_RAW:
3128 case PRIV_NETINET_ADDRCTRL6:
3129 case PRIV_NETINET_ND6:
3130 case PRIV_NETINET_SCOPE6:
3131 case PRIV_NETINET_ALIFETIME6:
3132 case PRIV_NETINET_IPSEC:
3133 case PRIV_NETINET_BINDANY:
3148 * No default: or deny here.
3149 * In case of no permit fall through to next switch().
3151 if (cred->cr_prison->pr_flags & PR_VNET)
3159 * Allow ktrace privileges for root in jail.
3165 * Allow jailed processes to configure audit identity and
3166 * submit audit records (login, etc). In the future we may
3167 * want to further refine the relationship between audit and
3170 case PRIV_AUDIT_GETAUDIT:
3171 case PRIV_AUDIT_SETAUDIT:
3172 case PRIV_AUDIT_SUBMIT:
3176 * Allow jailed processes to manipulate process UNIX
3177 * credentials in any way they see fit.
3179 case PRIV_CRED_SETUID:
3180 case PRIV_CRED_SETEUID:
3181 case PRIV_CRED_SETGID:
3182 case PRIV_CRED_SETEGID:
3183 case PRIV_CRED_SETGROUPS:
3184 case PRIV_CRED_SETREUID:
3185 case PRIV_CRED_SETREGID:
3186 case PRIV_CRED_SETRESUID:
3187 case PRIV_CRED_SETRESGID:
3190 * Jail implements visibility constraints already, so allow
3191 * jailed root to override uid/gid-based constraints.
3193 case PRIV_SEEOTHERGIDS:
3194 case PRIV_SEEOTHERUIDS:
3197 * Jail implements inter-process debugging limits already, so
3198 * allow jailed root various debugging privileges.
3200 case PRIV_DEBUG_DIFFCRED:
3201 case PRIV_DEBUG_SUGID:
3202 case PRIV_DEBUG_UNPRIV:
3205 * Allow jail to set various resource limits and login
3206 * properties, and for now, exceed process resource limits.
3208 case PRIV_PROC_LIMIT:
3209 case PRIV_PROC_SETLOGIN:
3210 case PRIV_PROC_SETRLIMIT:
3213 * System V and POSIX IPC privileges are granted in jail.
3216 case PRIV_IPC_WRITE:
3217 case PRIV_IPC_ADMIN:
3218 case PRIV_IPC_MSGSIZE:
3222 * Jail operations within a jail work on child jails.
3224 case PRIV_JAIL_ATTACH:
3226 case PRIV_JAIL_REMOVE:
3229 * Jail implements its own inter-process limits, so allow
3230 * root processes in jail to change scheduling on other
3231 * processes in the same jail. Likewise for signalling.
3233 case PRIV_SCHED_DIFFCRED:
3234 case PRIV_SCHED_CPUSET:
3235 case PRIV_SIGNAL_DIFFCRED:
3236 case PRIV_SIGNAL_SUGID:
3239 * Allow jailed processes to write to sysctls marked as jail
3242 case PRIV_SYSCTL_WRITEJAIL:
3245 * Allow root in jail to manage a variety of quota
3246 * properties. These should likely be conditional on a
3247 * configuration option.
3249 case PRIV_VFS_GETQUOTA:
3250 case PRIV_VFS_SETQUOTA:
3253 * Since Jail relies on chroot() to implement file system
3254 * protections, grant many VFS privileges to root in jail.
3255 * Be careful to exclude mount-related and NFS-related
3259 case PRIV_VFS_WRITE:
3260 case PRIV_VFS_ADMIN:
3262 case PRIV_VFS_LOOKUP:
3263 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3264 case PRIV_VFS_CHFLAGS_DEV:
3265 case PRIV_VFS_CHOWN:
3266 case PRIV_VFS_CHROOT:
3267 case PRIV_VFS_RETAINSUGID:
3268 case PRIV_VFS_FCHROOT:
3270 case PRIV_VFS_SETGID:
3272 case PRIV_VFS_STICKYFILE:
3275 * As in the non-jail case, non-root users are expected to be
3276 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3277 * exists in the jail and they have permission to access it).
3279 case PRIV_KMEM_READ:
3283 * Depending on the global setting, allow privilege of
3284 * setting system flags.
3286 case PRIV_VFS_SYSFLAGS:
3287 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3293 * Depending on the global setting, allow privilege of
3294 * mounting/unmounting file systems.
3296 case PRIV_VFS_MOUNT:
3297 case PRIV_VFS_UNMOUNT:
3298 case PRIV_VFS_MOUNT_NONUSER:
3299 case PRIV_VFS_MOUNT_OWNER:
3300 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3301 cred->cr_prison->pr_enforce_statfs < 2)
3307 * Conditionnaly allow locking (unlocking) physical pages
3311 case PRIV_VM_MUNLOCK:
3312 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3318 * Conditionally allow jailed root to bind reserved ports.
3320 case PRIV_NETINET_RESERVEDPORT:
3321 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3327 * Allow jailed root to reuse in-use ports.
3329 case PRIV_NETINET_REUSEPORT:
3333 * Allow jailed root to set certain IPv4/6 (option) headers.
3335 case PRIV_NETINET_SETHDROPTS:
3339 * Conditionally allow creating raw sockets in jail.
3341 case PRIV_NETINET_RAW:
3342 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3348 * Since jail implements its own visibility limits on netstat
3349 * sysctls, allow getcred. This allows identd to work in
3352 case PRIV_NETINET_GETCRED:
3356 * Allow jailed root to set loginclass.
3358 case PRIV_PROC_SETLOGINCLASS:
3362 * Do not allow a process inside a jail to read the kernel
3363 * message buffer unless explicitly permitted.
3366 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3372 * In all remaining cases, deny the privilege request. This
3373 * includes almost all network privileges, many system
3374 * configuration privileges.
3381 * Return the part of pr2's name that is relative to pr1, or the whole name
3382 * if it does not directly follow.
3386 prison_name(struct prison *pr1, struct prison *pr2)
3390 /* Jails see themselves as "0" (if they see themselves at all). */
3393 name = pr2->pr_name;
3394 if (prison_ischild(pr1, pr2)) {
3396 * pr1 isn't locked (and allprison_lock may not be either)
3397 * so its length can't be counted on. But the number of dots
3398 * can be counted on - and counted.
3400 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3401 name = strchr(name, '.') + 1;
3407 * Return the part of pr2's path that is relative to pr1, or the whole path
3408 * if it does not directly follow.
3411 prison_path(struct prison *pr1, struct prison *pr2)
3413 char *path1, *path2;
3416 path1 = pr1->pr_path;
3417 path2 = pr2->pr_path;
3418 if (!strcmp(path1, "/"))
3420 len1 = strlen(path1);
3421 if (strncmp(path1, path2, len1))
3423 if (path2[len1] == '\0')
3425 if (path2[len1] == '/')
3426 return (path2 + len1);
3432 * Jail-related sysctls.
3434 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3438 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3441 struct prison *pr, *cpr;
3443 struct in_addr *ip4 = NULL;
3447 struct in6_addr *ip6 = NULL;
3452 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3453 pr = req->td->td_ucred->cr_prison;
3455 sx_slock(&allprison_lock);
3456 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3457 #if defined(INET) || defined(INET6)
3460 mtx_lock(&cpr->pr_mtx);
3462 if (cpr->pr_ip4s > 0) {
3463 if (ip4s < cpr->pr_ip4s) {
3464 ip4s = cpr->pr_ip4s;
3465 mtx_unlock(&cpr->pr_mtx);
3466 ip4 = realloc(ip4, ip4s *
3467 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3470 bcopy(cpr->pr_ip4, ip4,
3471 cpr->pr_ip4s * sizeof(struct in_addr));
3475 if (cpr->pr_ip6s > 0) {
3476 if (ip6s < cpr->pr_ip6s) {
3477 ip6s = cpr->pr_ip6s;
3478 mtx_unlock(&cpr->pr_mtx);
3479 ip6 = realloc(ip6, ip6s *
3480 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3483 bcopy(cpr->pr_ip6, ip6,
3484 cpr->pr_ip6s * sizeof(struct in6_addr));
3487 if (cpr->pr_ref == 0) {
3488 mtx_unlock(&cpr->pr_mtx);
3491 bzero(xp, sizeof(*xp));
3492 xp->pr_version = XPRISON_VERSION;
3493 xp->pr_id = cpr->pr_id;
3494 xp->pr_state = cpr->pr_uref > 0
3495 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3496 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3497 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3498 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3500 xp->pr_ip4s = cpr->pr_ip4s;
3503 xp->pr_ip6s = cpr->pr_ip6s;
3505 mtx_unlock(&cpr->pr_mtx);
3506 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3510 if (xp->pr_ip4s > 0) {
3511 error = SYSCTL_OUT(req, ip4,
3512 xp->pr_ip4s * sizeof(struct in_addr));
3518 if (xp->pr_ip6s > 0) {
3519 error = SYSCTL_OUT(req, ip6,
3520 xp->pr_ip6s * sizeof(struct in6_addr));
3526 sx_sunlock(&allprison_lock);
3537 SYSCTL_OID(_security_jail, OID_AUTO, list,
3538 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3539 sysctl_jail_list, "S", "List of active jails");
3542 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3546 injail = jailed(req->td->td_ucred);
3547 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3552 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3553 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3554 sysctl_jail_jailed, "I", "Process in jail?");
3557 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3559 int error, havevnet;
3561 struct ucred *cred = req->td->td_ucred;
3563 havevnet = jailed(cred) && prison_owns_vnet(cred);
3567 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3572 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3573 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3574 sysctl_jail_vnet, "I", "Jail owns vnet?");
3576 #if defined(INET) || defined(INET6)
3577 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3578 &jail_max_af_ips, 0,
3579 "Number of IP addresses a jail may have at most per address family (deprecated)");
3583 * Default parameters for jail(2) compatibility. For historical reasons,
3584 * the sysctl names have varying similarity to the parameter names. Prisons
3585 * just see their own parameters, and can't change them.
3588 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3591 int allow, error, i;
3593 pr = req->td->td_ucred->cr_prison;
3594 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3596 /* Get the current flag value, and convert it to a boolean. */
3597 i = (allow & arg2) ? 1 : 0;
3600 error = sysctl_handle_int(oidp, &i, 0, req);
3601 if (error || !req->newptr)
3607 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3610 mtx_lock(&prison0.pr_mtx);
3611 jail_default_allow = (jail_default_allow & ~arg2) | i;
3612 mtx_unlock(&prison0.pr_mtx);
3616 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3617 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3618 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3619 "Processes in jail can set their hostnames (deprecated)");
3620 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3621 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3622 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3623 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3624 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3625 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3626 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3627 "Processes in jail can use System V IPC primitives (deprecated)");
3628 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3629 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3630 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3631 "Prison root can create raw sockets (deprecated)");
3632 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3633 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3634 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3635 "Processes in jail can alter system file flags (deprecated)");
3636 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3637 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3638 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3639 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3642 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3647 pr = req->td->td_ucred->cr_prison;
3648 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3649 error = sysctl_handle_int(oidp, &level, 0, req);
3650 if (error || !req->newptr)
3652 *(int *)arg1 = level;
3656 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3657 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3658 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3659 sysctl_jail_default_level, "I",
3660 "Processes in jail cannot see all mounted file systems (deprecated)");
3662 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3663 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3664 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3665 sysctl_jail_default_level, "I",
3666 "Ruleset for the devfs filesystem in jail (deprecated)");
3669 * Nodes to describe jail parameters. Maximum length of string parameters
3670 * is returned in the string itself, and the other parameters exist merely
3671 * to make themselves and their types known.
3673 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3677 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3684 switch (oidp->oid_kind & CTLTYPE)
3690 if (!(req->flags & SCTL_MASK32))
3692 return (SYSCTL_OUT(req, &l, sizeof(l)));
3696 return (SYSCTL_OUT(req, &i, sizeof(i)));
3697 case CTLTYPE_STRING:
3698 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3700 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3701 case CTLTYPE_STRUCT:
3703 return (SYSCTL_OUT(req, &s, sizeof(s)));
3709 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3710 * jail creation time but cannot be changed in an existing jail.
3712 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3713 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3714 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3715 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3716 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3717 "I", "Jail secure level");
3718 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3719 "Jail value for kern.osreldate and uname -K");
3720 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3721 "Jail value for kern.osrelease and uname -r");
3722 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3723 "I", "Jail cannot see all mounted file systems");
3724 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3725 "I", "Ruleset for in-jail devfs mounts");
3726 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3727 "B", "Jail persistence");
3729 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3730 "E,jailsys", "Virtual network stack");
3732 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3733 "B", "Jail is in the process of shutting down");
3735 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3736 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3737 "I", "Current number of child jails");
3738 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3739 "I", "Maximum number of child jails");
3741 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3742 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3744 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3745 "Jail NIS domainname");
3746 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3748 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3749 "LU", "Jail host ID");
3751 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3752 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3755 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3756 "Jail IPv4 address virtualization");
3757 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3758 "S,in_addr,a", "Jail IPv4 addresses");
3759 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3760 "B", "Do (not) use IPv4 source address selection rather than the "
3761 "primary jail IPv4 address.");
3764 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3765 "Jail IPv6 address virtualization");
3766 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3767 "S,in6_addr,a", "Jail IPv6 addresses");
3768 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3769 "B", "Do (not) use IPv6 source address selection rather than the "
3770 "primary jail IPv6 address.");
3773 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3774 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3775 "B", "Jail may set hostname");
3776 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3777 "B", "Jail may use SYSV IPC");
3778 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3779 "B", "Jail may create raw sockets");
3780 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3781 "B", "Jail may alter system file flags");
3782 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3783 "B", "Jail may set file quotas");
3784 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3785 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3786 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3787 "B", "Jail may lock (unlock) physical pages in memory");
3788 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3789 "B", "Jail may bind sockets to reserved ports");
3790 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3791 "B", "Jail may read the kernel message buffer");
3792 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
3793 "B", "Unprivileged processes may use process debugging facilities");
3795 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3796 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3797 "B", "Jail may mount/unmount jail-friendly file systems in general");
3800 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
3801 * its associated bit in the pr_allow bitmask, or zero if the parameter was
3805 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3808 struct bool_flags *bf;
3809 struct sysctl_oid *parent;
3810 char *allow_name, *allow_noname, *allowed;
3811 #ifndef NO_SYSCTL_DESCR
3812 char *descr_deprecated;
3814 unsigned allow_flag;
3817 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3819 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3821 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3822 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3823 free(allow_name, M_PRISON);
3828 * See if this parameter has already beed added, i.e. a module was
3829 * previously loaded/unloaded.
3831 mtx_lock(&prison0.pr_mtx);
3832 for (bf = pr_flag_allow;
3833 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3835 if (strcmp(bf->name, allow_name) == 0) {
3836 allow_flag = bf->flag;
3842 * Find a free bit in prison0's pr_allow, failing if there are none
3843 * (which shouldn't happen as long as we keep track of how many
3844 * potential dynamic flags exist).
3846 * Due to per-jail unprivileged process debugging support
3847 * using pr_allow, also verify against PR_ALLOW_ALL_STATIC.
3848 * prison0 may have unprivileged process debugging unset.
3850 for (allow_flag = 1;; allow_flag <<= 1) {
3851 if (allow_flag == 0)
3853 if (allow_flag & PR_ALLOW_ALL_STATIC)
3855 if ((prison0.pr_allow & allow_flag) == 0)
3860 * Note the parameter in the next open slot in pr_flag_allow.
3861 * Set the flag last so code that checks pr_flag_allow can do so
3864 for (bf = pr_flag_allow; bf->flag != 0; bf++)
3865 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3866 /* This should never happen, but is not fatal. */
3870 prison0.pr_allow |= allow_flag;
3871 bf->name = allow_name;
3872 bf->noname = allow_noname;
3873 bf->flag = allow_flag;
3874 mtx_unlock(&prison0.pr_mtx);
3877 * Create sysctls for the paramter, and the back-compat global
3881 ? SYSCTL_ADD_NODE(NULL,
3882 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3883 OID_AUTO, prefix, 0, 0, prefix_descr)
3884 : &sysctl___security_jail_param_allow;
3885 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3886 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3887 NULL, 0, sysctl_jail_param, "B", descr);
3889 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3890 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3891 #ifndef NO_SYSCTL_DESCR
3892 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3895 (void)SYSCTL_ADD_PROC(NULL,
3896 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3897 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3898 sysctl_jail_default_allow, "I", descr_deprecated);
3899 #ifndef NO_SYSCTL_DESCR
3900 free(descr_deprecated, M_TEMP);
3902 free(allowed, M_TEMP);
3907 mtx_unlock(&prison0.pr_mtx);
3908 free(allow_name, M_PRISON);
3909 free(allow_noname, M_PRISON);
3914 * The VFS system will register jail-aware filesystems here. They each get
3915 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3916 * attempts to mount.
3919 prison_add_vfs(struct vfsconf *vfsp)
3921 #ifdef NO_SYSCTL_DESCR
3923 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3928 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3930 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3932 free(descr, M_TEMP);
3938 prison_racct_foreach(void (*callback)(struct racct *racct,
3939 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3940 void *arg2, void *arg3)
3942 struct prison_racct *prr;
3944 ASSERT_RACCT_ENABLED();
3946 sx_slock(&allprison_lock);
3949 LIST_FOREACH(prr, &allprison_racct, prr_next)
3950 (callback)(prr->prr_racct, arg2, arg3);
3953 sx_sunlock(&allprison_lock);
3956 static struct prison_racct *
3957 prison_racct_find_locked(const char *name)
3959 struct prison_racct *prr;
3961 ASSERT_RACCT_ENABLED();
3962 sx_assert(&allprison_lock, SA_XLOCKED);
3964 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3967 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3968 if (strcmp(name, prr->prr_name) != 0)
3971 /* Found prison_racct with a matching name? */
3972 prison_racct_hold(prr);
3976 /* Add new prison_racct. */
3977 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3978 racct_create(&prr->prr_racct);
3980 strcpy(prr->prr_name, name);
3981 refcount_init(&prr->prr_refcount, 1);
3982 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3987 struct prison_racct *
3988 prison_racct_find(const char *name)
3990 struct prison_racct *prr;
3992 ASSERT_RACCT_ENABLED();
3994 sx_xlock(&allprison_lock);
3995 prr = prison_racct_find_locked(name);
3996 sx_xunlock(&allprison_lock);
4001 prison_racct_hold(struct prison_racct *prr)
4004 ASSERT_RACCT_ENABLED();
4006 refcount_acquire(&prr->prr_refcount);
4010 prison_racct_free_locked(struct prison_racct *prr)
4013 ASSERT_RACCT_ENABLED();
4014 sx_assert(&allprison_lock, SA_XLOCKED);
4016 if (refcount_release(&prr->prr_refcount)) {
4017 racct_destroy(&prr->prr_racct);
4018 LIST_REMOVE(prr, prr_next);
4019 free(prr, M_PRISON_RACCT);
4024 prison_racct_free(struct prison_racct *prr)
4027 ASSERT_RACCT_ENABLED();
4028 sx_assert(&allprison_lock, SA_UNLOCKED);
4030 if (refcount_release_if_not_last(&prr->prr_refcount))
4033 sx_xlock(&allprison_lock);
4034 prison_racct_free_locked(prr);
4035 sx_xunlock(&allprison_lock);
4039 prison_racct_attach(struct prison *pr)
4041 struct prison_racct *prr;
4043 ASSERT_RACCT_ENABLED();
4044 sx_assert(&allprison_lock, SA_XLOCKED);
4046 prr = prison_racct_find_locked(pr->pr_name);
4047 KASSERT(prr != NULL, ("cannot find prison_racct"));
4049 pr->pr_prison_racct = prr;
4053 * Handle jail renaming. From the racct point of view, renaming means
4054 * moving from one prison_racct to another.
4057 prison_racct_modify(struct prison *pr)
4063 struct prison_racct *oldprr;
4065 ASSERT_RACCT_ENABLED();
4067 sx_slock(&allproc_lock);
4068 sx_xlock(&allprison_lock);
4070 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4071 sx_xunlock(&allprison_lock);
4072 sx_sunlock(&allproc_lock);
4076 oldprr = pr->pr_prison_racct;
4077 pr->pr_prison_racct = NULL;
4079 prison_racct_attach(pr);
4082 * Move resource utilisation records.
4084 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4088 * Force rctl to reattach rules to processes.
4090 FOREACH_PROC_IN_SYSTEM(p) {
4092 cred = crhold(p->p_ucred);
4094 rctl_proc_ucred_changed(p, cred);
4099 sx_sunlock(&allproc_lock);
4100 prison_racct_free_locked(oldprr);
4101 sx_xunlock(&allprison_lock);
4105 prison_racct_detach(struct prison *pr)
4108 ASSERT_RACCT_ENABLED();
4109 sx_assert(&allprison_lock, SA_UNLOCKED);
4111 if (pr->pr_prison_racct == NULL)
4113 prison_racct_free(pr->pr_prison_racct);
4114 pr->pr_prison_racct = NULL;
4121 db_show_prison(struct prison *pr)
4123 struct bool_flags *bf;
4124 struct jailsys_flags *jsf;
4125 #if defined(INET) || defined(INET6)
4130 char ip4buf[INET_ADDRSTRLEN];
4133 char ip6buf[INET6_ADDRSTRLEN];
4136 db_printf("prison %p:\n", pr);
4137 db_printf(" jid = %d\n", pr->pr_id);
4138 db_printf(" name = %s\n", pr->pr_name);
4139 db_printf(" parent = %p\n", pr->pr_parent);
4140 db_printf(" ref = %d\n", pr->pr_ref);
4141 db_printf(" uref = %d\n", pr->pr_uref);
4142 db_printf(" path = %s\n", pr->pr_path);
4143 db_printf(" cpuset = %d\n", pr->pr_cpuset
4144 ? pr->pr_cpuset->cs_id : -1);
4146 db_printf(" vnet = %p\n", pr->pr_vnet);
4148 db_printf(" root = %p\n", pr->pr_root);
4149 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4150 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4151 db_printf(" children.max = %d\n", pr->pr_childmax);
4152 db_printf(" children.cur = %d\n", pr->pr_childcount);
4153 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4154 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4155 db_printf(" flags = 0x%x", pr->pr_flags);
4156 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4157 if (pr->pr_flags & bf->flag)
4158 db_printf(" %s", bf->name);
4159 for (jsf = pr_flag_jailsys;
4160 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4162 f = pr->pr_flags & (jsf->disable | jsf->new);
4163 db_printf(" %-16s= %s\n", jsf->name,
4164 (f != 0 && f == jsf->disable) ? "disable"
4165 : (f == jsf->new) ? "new"
4168 db_printf(" allow = 0x%x", pr->pr_allow);
4169 for (bf = pr_flag_allow;
4170 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4172 if (pr->pr_allow & bf->flag)
4173 db_printf(" %s", bf->name);
4175 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4176 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4177 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4178 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4179 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4181 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4182 for (ii = 0; ii < pr->pr_ip4s; ii++)
4183 db_printf(" %s %s\n",
4184 ii == 0 ? "ip4.addr =" : " ",
4185 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4188 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4189 for (ii = 0; ii < pr->pr_ip6s; ii++)
4190 db_printf(" %s %s\n",
4191 ii == 0 ? "ip6.addr =" : " ",
4192 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4196 DB_SHOW_COMMAND(prison, db_show_prison_command)
4202 * Show all prisons in the list, and prison0 which is not
4205 db_show_prison(&prison0);
4206 if (!db_pager_quit) {
4207 TAILQ_FOREACH(pr, &allprison, pr_list) {
4219 /* Look for a prison with the ID and with references. */
4220 TAILQ_FOREACH(pr, &allprison, pr_list)
4221 if (pr->pr_id == addr && pr->pr_ref > 0)
4224 /* Look again, without requiring a reference. */
4225 TAILQ_FOREACH(pr, &allprison, pr_list)
4226 if (pr->pr_id == addr)
4229 /* Assume address points to a valid prison. */
4230 pr = (struct prison *)addr;