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>
51 #include <sys/linker.h>
53 #include <sys/mutex.h>
54 #include <sys/racct.h>
56 #include <sys/refcount.h>
58 #include <sys/sysent.h>
59 #include <sys/namei.h>
60 #include <sys/mount.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
71 #include <netinet/in.h>
77 #include <security/mac/mac_framework.h>
79 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000"
80 #define PRISON0_HOSTUUID_MODULE "hostuuid"
82 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
83 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
85 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
88 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
90 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
94 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
96 #define _PR_IP_SADDRSEL 0
100 /* prison0 describes what is "real" about the system. */
101 struct prison prison0 = {
107 .pr_securelevel = -1,
109 .pr_childmax = JAIL_MAX,
110 .pr_hostuuid = DEFAULT_HOSTUUID,
111 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
113 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
115 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
117 .pr_allow = PR_ALLOW_ALL_STATIC,
119 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
126 struct jailsys_flags {
132 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
133 struct sx allprison_lock;
134 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
135 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
136 LIST_HEAD(, prison_racct) allprison_racct;
139 static int do_jail_attach(struct thread *td, struct prison *pr);
140 static void prison_complete(void *context, int pending);
141 static void prison_deref(struct prison *pr, int flags);
142 static char *prison_path(struct prison *pr1, struct prison *pr2);
143 static void prison_remove_one(struct prison *pr);
145 static void prison_racct_attach(struct prison *pr);
146 static void prison_racct_modify(struct prison *pr);
147 static void prison_racct_detach(struct prison *pr);
150 /* Flags for prison_deref */
151 #define PD_DEREF 0x01
152 #define PD_DEUREF 0x02
153 #define PD_LOCKED 0x04
154 #define PD_LIST_SLOCKED 0x08
155 #define PD_LIST_XLOCKED 0x10
158 * Parameter names corresponding to PR_* flag values. Size values are for kvm
159 * as we cannot figure out the size of a sparse array, or an array without a
162 static struct bool_flags pr_flag_bool[] = {
163 {"persist", "nopersist", PR_PERSIST},
165 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
168 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
171 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
173 static struct jailsys_flags pr_flag_jailsys[] = {
174 {"host", 0, PR_HOST},
176 {"vnet", 0, PR_VNET},
179 {"ip4", PR_IP4_USER, PR_IP4_USER},
182 {"ip6", PR_IP6_USER, PR_IP6_USER},
185 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
187 /* Make this array full-size so dynamic parameters can be added. */
188 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
189 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
190 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
191 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
192 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
193 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
194 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
195 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
196 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
197 {"allow.reserved_ports", "allow.noreserved_ports",
198 PR_ALLOW_RESERVED_PORTS},
199 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
200 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
201 PR_ALLOW_UNPRIV_DEBUG},
203 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
205 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
206 PR_ALLOW_RESERVED_PORTS | \
207 PR_ALLOW_UNPRIV_DEBUG)
208 #define JAIL_DEFAULT_ENFORCE_STATFS 2
209 #define JAIL_DEFAULT_DEVFS_RSNUM 0
210 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
211 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
212 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
213 #if defined(INET) || defined(INET6)
214 static unsigned jail_max_af_ips = 255;
218 * Initialize the parts of prison0 that can't be static-initialized with
219 * constants. This is called from proc0_init() after creating thread0 cpuset.
224 uint8_t *file, *data;
227 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
228 prison0.pr_osreldate = osreldate;
229 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
231 /* If we have a preloaded hostuuid, use it. */
232 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
234 data = preload_fetch_addr(file);
235 size = preload_fetch_size(file);
238 * The preloaded data may include trailing whitespace, almost
239 * certainly a newline; skip over any whitespace or
240 * non-printable characters to be safe.
242 while (size > 0 && data[size - 1] <= 0x20) {
245 if (validate_uuid(data, size, NULL, 0) == 0) {
246 (void)strlcpy(prison0.pr_hostuuid, data,
248 } else if (bootverbose) {
249 printf("hostuuid: preload data malformed: '%s'",
255 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
264 sys_jail(struct thread *td, struct jail_args *uap)
270 error = copyin(uap->jail, &version, sizeof(uint32_t));
279 /* FreeBSD single IPv4 jails. */
280 bzero(&j, sizeof(struct jail));
281 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
284 j.version = j0.version;
286 j.hostname = j0.hostname;
287 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
293 * Version 1 was used by multi-IPv4 jail implementations
294 * that never made it into the official kernel.
298 case 2: /* JAIL_API_VERSION */
299 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
300 error = copyin(uap->jail, &j, sizeof(struct jail));
306 /* Sci-Fi jails are not supported, sorry. */
309 return (kern_jail(td, &j));
313 kern_jail(struct thread *td, struct jail *j)
315 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
324 char *u_path, *u_hostname, *u_name;
325 struct bool_flags *bf;
328 struct in_addr *u_ip4;
331 struct in6_addr *u_ip6;
334 int error, enforce_statfs;
336 bzero(&optiov, sizeof(optiov));
337 opt.uio_iov = optiov;
341 opt.uio_segflg = UIO_SYSSPACE;
342 opt.uio_rw = UIO_READ;
345 /* Set permissions for top-level jails from sysctls. */
346 if (!jailed(td->td_ucred)) {
347 for (bf = pr_flag_allow;
348 bf < pr_flag_allow + nitems(pr_flag_allow) &&
351 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
352 (jail_default_allow & bf->flag)
353 ? bf->name : bf->noname);
354 optiov[opt.uio_iovcnt].iov_len =
355 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
358 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
359 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
361 enforce_statfs = jail_default_enforce_statfs;
362 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
363 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
367 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
369 ip4s = (j->version == 0) ? 1 : j->ip4s;
370 if (ip4s > jail_max_af_ips)
372 tmplen += ip4s * sizeof(struct in_addr);
378 if (j->ip6s > jail_max_af_ips)
380 tmplen += j->ip6s * sizeof(struct in6_addr);
385 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
386 u_hostname = u_path + MAXPATHLEN;
387 u_name = u_hostname + MAXHOSTNAMELEN;
389 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
393 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
395 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
398 optiov[opt.uio_iovcnt].iov_base = "path";
399 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
401 optiov[opt.uio_iovcnt].iov_base = u_path;
402 error = copyinstr(j->path, u_path, MAXPATHLEN,
403 &optiov[opt.uio_iovcnt].iov_len);
405 free(u_path, M_TEMP);
409 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
410 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
412 optiov[opt.uio_iovcnt].iov_base = u_hostname;
413 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
414 &optiov[opt.uio_iovcnt].iov_len);
416 free(u_path, M_TEMP);
420 if (j->jailname != NULL) {
421 optiov[opt.uio_iovcnt].iov_base = "name";
422 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
424 optiov[opt.uio_iovcnt].iov_base = u_name;
425 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
426 &optiov[opt.uio_iovcnt].iov_len);
428 free(u_path, M_TEMP);
434 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
435 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
437 optiov[opt.uio_iovcnt].iov_base = u_ip4;
438 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
440 u_ip4->s_addr = j->ip4s;
442 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
444 free(u_path, M_TEMP);
451 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
452 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
454 optiov[opt.uio_iovcnt].iov_base = u_ip6;
455 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
456 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
458 free(u_path, M_TEMP);
463 KASSERT(opt.uio_iovcnt <= nitems(optiov),
464 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
465 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
466 free(u_path, M_TEMP);
471 * struct jail_set_args {
472 * struct iovec *iovp;
473 * unsigned int iovcnt;
478 sys_jail_set(struct thread *td, struct jail_set_args *uap)
483 /* Check that we have an even number of iovecs. */
487 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
490 error = kern_jail_set(td, auio, uap->flags);
496 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
503 struct in6_addr *ip6;
506 struct vfsoptlist *opts;
507 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
509 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
510 char *g_path, *osrelstr;
511 struct bool_flags *bf;
512 struct jailsys_flags *jsf;
513 #if defined(INET) || defined(INET6)
518 size_t namelen, onamelen, pnamelen;
519 int born, created, cuflags, descend, enforce;
520 int error, errmsg_len, errmsg_pos;
521 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
522 int jid, jsys, len, level;
523 int childmax, osreldt, rsnum, slevel;
524 #if defined(INET) || defined(INET6)
533 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
534 uint64_t pr_allow_diff;
538 error = priv_check(td, PRIV_JAIL_SET);
539 if (!error && (flags & JAIL_ATTACH))
540 error = priv_check(td, PRIV_JAIL_ATTACH);
543 mypr = td->td_ucred->cr_prison;
544 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
546 if (flags & ~JAIL_SET_MASK)
550 * Check all the parameters before committing to anything. Not all
551 * errors can be caught early, but we may as well try. Also, this
552 * takes care of some expensive stuff (path lookup) before getting
553 * the allprison lock.
555 * XXX Jails are not filesystems, and jail parameters are not mount
556 * options. But it makes more sense to re-use the vfsopt code
557 * than duplicate it under a different name.
559 error = vfs_buildopts(optuio, &opts);
570 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
573 vfs_opterror(opts, "no valid operation (create or update)");
577 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
583 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
592 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
600 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
605 else if (enforce < 0 || enforce > 2) {
611 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
619 pr_flags = ch_flags = 0;
620 for (bf = pr_flag_bool;
621 bf < pr_flag_bool + nitems(pr_flag_bool);
623 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
624 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
626 ch_flags |= pr_flags;
627 for (jsf = pr_flag_jailsys;
628 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
630 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
636 case JAIL_SYS_DISABLE:
641 pr_flags |= jsf->disable;
644 pr_flags |= jsf->new;
646 case JAIL_SYS_INHERIT:
652 ch_flags |= jsf->new | jsf->disable;
654 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
655 && !(pr_flags & PR_PERSIST)) {
657 vfs_opterror(opts, "new jail must persist or attach");
661 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
663 vfs_opterror(opts, "vnet cannot be changed after creation");
668 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
670 vfs_opterror(opts, "ip4 cannot be changed after creation");
675 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
677 vfs_opterror(opts, "ip6 cannot be changed after creation");
682 pr_allow = ch_allow = 0;
683 for (bf = pr_flag_allow;
684 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
686 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
687 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
689 ch_allow |= pr_allow;
691 error = vfs_getopt(opts, "name", (void **)&name, &len);
697 if (len == 0 || name[len - 1] != '\0') {
701 if (len > MAXHOSTNAMELEN) {
702 error = ENAMETOOLONG;
707 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
715 if (len == 0 || host[len - 1] != '\0') {
719 if (len > MAXHOSTNAMELEN) {
720 error = ENAMETOOLONG;
725 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
733 if (len == 0 || domain[len - 1] != '\0') {
737 if (len > MAXHOSTNAMELEN) {
738 error = ENAMETOOLONG;
743 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
751 if (len == 0 || uuid[len - 1] != '\0') {
755 if (len > HOSTUUIDLEN) {
756 error = ENAMETOOLONG;
761 #ifdef COMPAT_FREEBSD32
762 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
765 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
769 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
781 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
786 else if (ip4s & (sizeof(*ip4) - 1)) {
790 ch_flags |= PR_IP4_USER;
791 pr_flags |= PR_IP4_USER;
793 ip4s /= sizeof(*ip4);
794 if (ip4s > jail_max_af_ips) {
796 vfs_opterror(opts, "too many IPv4 addresses");
799 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
800 bcopy(op, ip4, ip4s * sizeof(*ip4));
802 * IP addresses are all sorted but ip[0] to preserve
803 * the primary IP address as given from userland.
804 * This special IP is used for unbound outgoing
805 * connections as well for "loopback" traffic in case
806 * source address selection cannot find any more fitting
807 * address to connect from.
810 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
813 * Check for duplicate addresses and do some simple
814 * zero and broadcast checks. If users give other bogus
815 * addresses it is their problem.
817 * We do not have to care about byte order for these
818 * checks so we will do them in NBO.
820 for (ii = 0; ii < ip4s; ii++) {
821 if (ip4[ii].s_addr == INADDR_ANY ||
822 ip4[ii].s_addr == INADDR_BROADCAST) {
827 (ip4[0].s_addr == ip4[ii+1].s_addr ||
828 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
838 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
843 else if (ip6s & (sizeof(*ip6) - 1)) {
847 ch_flags |= PR_IP6_USER;
848 pr_flags |= PR_IP6_USER;
850 ip6s /= sizeof(*ip6);
851 if (ip6s > jail_max_af_ips) {
853 vfs_opterror(opts, "too many IPv6 addresses");
856 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
857 bcopy(op, ip6, ip6s * sizeof(*ip6));
859 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
861 for (ii = 0; ii < ip6s; ii++) {
862 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
867 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
868 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
878 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
879 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
882 "vnet jails cannot have IP address restrictions");
887 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
893 if (flags & JAIL_UPDATE) {
896 "osrelease cannot be changed after creation");
899 if (len == 0 || osrelstr[len - 1] != '\0') {
903 if (len >= OSRELEASELEN) {
904 error = ENAMETOOLONG;
906 "osrelease string must be 1-%d bytes long",
912 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
918 if (flags & JAIL_UPDATE) {
921 "osreldate cannot be changed after creation");
926 vfs_opterror(opts, "osreldate cannot be 0");
932 error = vfs_getopt(opts, "path", (void **)&path, &len);
938 if (flags & JAIL_UPDATE) {
941 "path cannot be changed after creation");
944 if (len == 0 || path[len - 1] != '\0') {
948 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
954 NDFREE(&nd, NDF_ONLY_PNBUF);
955 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
956 strlcpy(g_path, path, MAXPATHLEN);
957 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
961 /* exit on other errors */
964 if (root->v_type != VDIR) {
973 * Find the specified jail, or at least its parent.
974 * This abuses the file error codes ENOENT and EEXIST.
978 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
979 namelc = strrchr(name, '.');
980 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
984 sx_xlock(&allprison_lock);
987 * See if a requested jid already exists. There is an
988 * information leak here if the jid exists but is not within
989 * the caller's jail hierarchy. Jail creators will get EEXIST
990 * even though they cannot see the jail, and CREATE | UPDATE
991 * will return ENOENT which is not normally a valid error.
995 vfs_opterror(opts, "negative jid");
996 goto done_unlock_list;
998 pr = prison_find(jid);
1000 ppr = pr->pr_parent;
1001 /* Create: jid must not exist. */
1002 if (cuflags == JAIL_CREATE) {
1003 mtx_unlock(&pr->pr_mtx);
1005 vfs_opterror(opts, "jail %d already exists",
1007 goto done_unlock_list;
1009 if (!prison_ischild(mypr, pr)) {
1010 mtx_unlock(&pr->pr_mtx);
1012 } else if (pr->pr_uref == 0) {
1013 if (!(flags & JAIL_DYING)) {
1014 mtx_unlock(&pr->pr_mtx);
1016 vfs_opterror(opts, "jail %d is dying",
1018 goto done_unlock_list;
1019 } else if ((flags & JAIL_ATTACH) ||
1020 (pr_flags & PR_PERSIST)) {
1022 * A dying jail might be resurrected
1023 * (via attach or persist), but first
1024 * it must determine if another jail
1025 * has claimed its name. Accomplish
1026 * this by implicitly re-setting the
1030 name = prison_name(mypr, pr);
1035 /* Update: jid must exist. */
1036 if (cuflags == JAIL_UPDATE) {
1038 vfs_opterror(opts, "jail %d not found", jid);
1039 goto done_unlock_list;
1044 * If the caller provided a name, look for a jail by that name.
1045 * This has different semantics for creates and updates keyed by jid
1046 * (where the name must not already exist in a different jail),
1047 * and updates keyed by the name itself (where the name must exist
1048 * because that is the jail being updated).
1052 namelc = strrchr(name, '.');
1057 * This is a hierarchical name. Split it into the
1058 * parent and child names, and make sure the parent
1059 * exists or matches an already found jail.
1062 if (strncmp(name, ppr->pr_name, namelc - name)
1063 || ppr->pr_name[namelc - name] != '\0') {
1064 mtx_unlock(&pr->pr_mtx);
1067 "cannot change jail's parent");
1068 goto done_unlock_list;
1072 ppr = prison_find_name(mypr, name);
1076 "jail \"%s\" not found", name);
1077 goto done_unlock_list;
1079 mtx_unlock(&ppr->pr_mtx);
1084 if (namelc[0] != '\0') {
1086 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1089 FOREACH_PRISON_CHILD(ppr, tpr) {
1090 if (tpr != pr && tpr->pr_ref > 0 &&
1091 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1093 cuflags != JAIL_CREATE) {
1094 mtx_lock(&tpr->pr_mtx);
1095 if (tpr->pr_ref > 0) {
1100 if (tpr->pr_uref > 0) {
1106 mtx_unlock(&tpr->pr_mtx);
1107 } else if (tpr->pr_uref > 0) {
1109 * Create, or update(jid):
1110 * name must not exist in an
1111 * active sibling jail.
1115 mtx_unlock(&pr->pr_mtx);
1117 "jail \"%s\" already exists",
1119 goto done_unlock_list;
1123 /* If no active jail is found, use a dying one. */
1124 if (deadpr != NULL && pr == NULL) {
1125 if (flags & JAIL_DYING) {
1126 mtx_lock(&deadpr->pr_mtx);
1127 if (deadpr->pr_ref == 0) {
1128 mtx_unlock(&deadpr->pr_mtx);
1132 } else if (cuflags == JAIL_UPDATE) {
1135 "jail \"%s\" is dying", name);
1136 goto done_unlock_list;
1139 /* Update: name must exist if no jid. */
1140 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1142 vfs_opterror(opts, "jail \"%s\" not found",
1144 goto done_unlock_list;
1148 /* Update: must provide a jid or name. */
1149 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1151 vfs_opterror(opts, "update specified no jail");
1152 goto done_unlock_list;
1155 /* If there's no prison to update, create a new one and link it in. */
1157 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1158 if (tpr->pr_childcount >= tpr->pr_childmax) {
1160 vfs_opterror(opts, "prison limit exceeded");
1161 goto done_unlock_list;
1164 mtx_lock(&ppr->pr_mtx);
1165 if (ppr->pr_ref == 0) {
1166 mtx_unlock(&ppr->pr_mtx);
1168 vfs_opterror(opts, "jail \"%s\" not found",
1169 prison_name(mypr, ppr));
1170 goto done_unlock_list;
1174 mtx_unlock(&ppr->pr_mtx);
1175 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1177 /* Find the next free jid. */
1180 if (jid == JAIL_MAX)
1182 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1183 if (tpr->pr_id < jid)
1185 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1186 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1189 if (jid == lastprid) {
1192 "no available jail IDs");
1194 prison_deref(ppr, PD_DEREF |
1195 PD_DEUREF | PD_LIST_XLOCKED);
1204 * The jail already has a jid (that did not yet exist),
1205 * so just find where to insert it.
1207 TAILQ_FOREACH(tpr, &allprison, pr_list)
1208 if (tpr->pr_id >= jid) {
1209 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1214 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1215 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1216 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1217 tpr->pr_childcount++;
1219 pr->pr_parent = ppr;
1222 /* Set some default values, and inherit some from the parent. */
1227 root = mypr->pr_root;
1230 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1231 pr->pr_flags |= PR_HOST;
1232 #if defined(INET) || defined(INET6)
1234 if (!(pr_flags & PR_VNET))
1238 if (!(ch_flags & PR_IP4_USER))
1239 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1240 else if (!(pr_flags & PR_IP4_USER)) {
1241 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1242 if (ppr->pr_ip4 != NULL) {
1243 pr->pr_ip4s = ppr->pr_ip4s;
1244 pr->pr_ip4 = malloc(pr->pr_ip4s *
1245 sizeof(struct in_addr), M_PRISON,
1247 bcopy(ppr->pr_ip4, pr->pr_ip4,
1248 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1253 if (!(ch_flags & PR_IP6_USER))
1254 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1255 else if (!(pr_flags & PR_IP6_USER)) {
1256 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1257 if (ppr->pr_ip6 != NULL) {
1258 pr->pr_ip6s = ppr->pr_ip6s;
1259 pr->pr_ip6 = malloc(pr->pr_ip6s *
1260 sizeof(struct in6_addr), M_PRISON,
1262 bcopy(ppr->pr_ip6, pr->pr_ip6,
1263 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1269 /* Source address selection is always on by default. */
1270 pr->pr_flags |= _PR_IP_SADDRSEL;
1272 pr->pr_securelevel = ppr->pr_securelevel;
1273 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1274 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1275 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1277 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1278 if (osrelstr == NULL)
1279 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1280 sizeof(pr->pr_osrelease));
1282 strlcpy(pr->pr_osrelease, osrelstr,
1283 sizeof(pr->pr_osrelease));
1285 LIST_INIT(&pr->pr_children);
1286 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1287 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1290 /* Allocate a new vnet if specified. */
1291 pr->pr_vnet = (pr_flags & PR_VNET)
1292 ? vnet_alloc() : ppr->pr_vnet;
1295 * Allocate a dedicated cpuset for each jail.
1296 * Unlike other initial settings, this may return an erorr.
1298 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1300 prison_deref(pr, PD_LIST_XLOCKED);
1304 mtx_lock(&pr->pr_mtx);
1306 * New prisons do not yet have a reference, because we do not
1307 * want others to see the incomplete prison once the
1308 * allprison_lock is downgraded.
1313 * Grab a reference for existing prisons, to ensure they
1314 * continue to exist for the duration of the call.
1317 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1318 if ((pr->pr_flags & PR_VNET) &&
1319 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1322 "vnet jails cannot have IP address restrictions");
1323 goto done_deref_locked;
1327 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1330 "ip4 cannot be changed after creation");
1331 goto done_deref_locked;
1335 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1338 "ip6 cannot be changed after creation");
1339 goto done_deref_locked;
1344 /* Do final error checking before setting anything. */
1346 if (slevel < ppr->pr_securelevel) {
1348 goto done_deref_locked;
1352 if (childmax >= ppr->pr_childmax) {
1354 goto done_deref_locked;
1358 if (enforce < ppr->pr_enforce_statfs) {
1360 goto done_deref_locked;
1365 * devfs_rsnum is a uint16_t
1367 if (rsnum < 0 || rsnum > 65535) {
1369 goto done_deref_locked;
1372 * Nested jails always inherit parent's devfs ruleset
1374 if (jailed(td->td_ucred)) {
1375 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1377 goto done_deref_locked;
1379 rsnum = ppr->pr_devfs_rsnum;
1384 if (ppr->pr_flags & PR_IP4) {
1386 * Make sure the new set of IP addresses is a
1387 * subset of the parent's list. Don't worry
1388 * about the parent being unlocked, as any
1389 * setting is done with allprison_lock held.
1391 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1392 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1394 if (ij == ppr->pr_ip4s) {
1396 goto done_deref_locked;
1399 for (ii = ij = 1; ii < ip4s; ii++) {
1400 if (ip4[ii].s_addr ==
1401 ppr->pr_ip4[0].s_addr)
1403 for (; ij < ppr->pr_ip4s; ij++)
1404 if (ip4[ii].s_addr ==
1405 ppr->pr_ip4[ij].s_addr)
1407 if (ij == ppr->pr_ip4s)
1410 if (ij == ppr->pr_ip4s) {
1412 goto done_deref_locked;
1417 * Check for conflicting IP addresses. We permit them
1418 * if there is no more than one IP on each jail. If
1419 * there is a duplicate on a jail with more than one
1420 * IP stop checking and return error.
1423 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1424 if (tppr->pr_flags & PR_VNET)
1429 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1432 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1434 tpr->pr_uref == 0) {
1438 if (!(tpr->pr_flags & PR_IP4_USER))
1441 if (tpr->pr_ip4 == NULL ||
1442 (ip4s == 1 && tpr->pr_ip4s == 1))
1444 for (ii = 0; ii < ip4s; ii++) {
1445 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1449 "IPv4 addresses clash");
1450 goto done_deref_locked;
1458 if (ppr->pr_flags & PR_IP6) {
1460 * Make sure the new set of IP addresses is a
1461 * subset of the parent's list.
1463 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1464 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1467 if (ij == ppr->pr_ip6s) {
1469 goto done_deref_locked;
1472 for (ii = ij = 1; ii < ip6s; ii++) {
1473 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1476 for (; ij < ppr->pr_ip6s; ij++)
1477 if (IN6_ARE_ADDR_EQUAL(
1478 &ip6[ii], &ppr->pr_ip6[ij]))
1480 if (ij == ppr->pr_ip6s)
1483 if (ij == ppr->pr_ip6s) {
1485 goto done_deref_locked;
1489 /* Check for conflicting IP addresses. */
1491 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1492 if (tppr->pr_flags & PR_VNET)
1497 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1500 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1502 tpr->pr_uref == 0) {
1506 if (!(tpr->pr_flags & PR_IP6_USER))
1509 if (tpr->pr_ip6 == NULL ||
1510 (ip6s == 1 && tpr->pr_ip6s == 1))
1512 for (ii = 0; ii < ip6s; ii++) {
1513 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1517 "IPv6 addresses clash");
1518 goto done_deref_locked;
1524 onamelen = namelen = 0;
1525 if (namelc != NULL) {
1526 /* Give a default name of the jid. Also allow the name to be
1527 * explicitly the jid - but not any other number, and only in
1528 * normal form (no leading zero/etc).
1530 if (namelc[0] == '\0')
1531 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1532 else if ((strtoul(namelc, &p, 10) != jid ||
1533 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1536 "name cannot be numeric (unless it is the jid)");
1537 goto done_deref_locked;
1540 * Make sure the name isn't too long for the prison or its
1543 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1544 onamelen = strlen(pr->pr_name + pnamelen);
1545 namelen = strlen(namelc);
1546 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1547 error = ENAMETOOLONG;
1548 goto done_deref_locked;
1550 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1551 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1552 sizeof(pr->pr_name)) {
1553 error = ENAMETOOLONG;
1554 goto done_deref_locked;
1558 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1559 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1561 goto done_deref_locked;
1565 * Let modules check their parameters. This requires unlocking and
1566 * then re-locking the prison, but this is still a valid state as long
1567 * as allprison_lock remains xlocked.
1569 mtx_unlock(&pr->pr_mtx);
1570 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1572 prison_deref(pr, created
1574 : PD_DEREF | PD_LIST_XLOCKED);
1577 mtx_lock(&pr->pr_mtx);
1579 /* At this point, all valid parameters should have been noted. */
1580 TAILQ_FOREACH(opt, opts, link) {
1581 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1583 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1584 goto done_deref_locked;
1588 /* Set the parameters of the prison. */
1591 if (pr_flags & PR_IP4_USER) {
1592 pr->pr_flags |= PR_IP4;
1593 free(pr->pr_ip4, M_PRISON);
1597 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1599 if (tpr->pr_flags & PR_VNET) {
1604 if (prison_restrict_ip4(tpr, NULL)) {
1613 if (pr_flags & PR_IP6_USER) {
1614 pr->pr_flags |= PR_IP6;
1615 free(pr->pr_ip6, M_PRISON);
1619 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1621 if (tpr->pr_flags & PR_VNET) {
1626 if (prison_restrict_ip6(tpr, NULL)) {
1634 pr->pr_securelevel = slevel;
1635 /* Set all child jails to be at least this level. */
1636 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1637 if (tpr->pr_securelevel < slevel)
1638 tpr->pr_securelevel = slevel;
1641 pr->pr_childmax = childmax;
1642 /* Set all child jails to under this limit. */
1643 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1644 if (tpr->pr_childmax > childmax - level)
1645 tpr->pr_childmax = childmax > level
1646 ? childmax - level : 0;
1649 pr->pr_enforce_statfs = enforce;
1650 /* Pass this restriction on to the children. */
1651 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1652 if (tpr->pr_enforce_statfs < enforce)
1653 tpr->pr_enforce_statfs = enforce;
1656 pr->pr_devfs_rsnum = rsnum;
1657 /* Pass this restriction on to the children. */
1658 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1659 tpr->pr_devfs_rsnum = rsnum;
1661 if (namelc != NULL) {
1662 if (ppr == &prison0)
1663 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1665 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1666 ppr->pr_name, namelc);
1667 /* Change this component of child names. */
1668 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1669 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1670 strlen(tpr->pr_name + onamelen) + 1);
1671 bcopy(pr->pr_name, tpr->pr_name, namelen);
1675 /* Try to keep a real-rooted full pathname. */
1676 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1679 if (PR_HOST & ch_flags & ~pr_flags) {
1680 if (pr->pr_flags & PR_HOST) {
1682 * Copy the parent's host info. As with pr_ip4 above,
1683 * the lack of a lock on the parent is not a problem;
1684 * it is always set with allprison_lock at least
1685 * shared, and is held exclusively here.
1687 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1688 sizeof(pr->pr_hostname));
1689 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1690 sizeof(pr->pr_domainname));
1691 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1692 sizeof(pr->pr_hostuuid));
1693 pr->pr_hostid = pr->pr_parent->pr_hostid;
1695 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1696 /* Set this prison, and any descendants without PR_HOST. */
1698 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1700 strlcpy(pr->pr_domainname, domain,
1701 sizeof(pr->pr_domainname));
1703 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1705 pr->pr_hostid = hid;
1706 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1707 if (tpr->pr_flags & PR_HOST)
1711 strlcpy(tpr->pr_hostname,
1713 sizeof(tpr->pr_hostname));
1715 strlcpy(tpr->pr_domainname,
1717 sizeof(tpr->pr_domainname));
1719 strlcpy(tpr->pr_hostuuid,
1721 sizeof(tpr->pr_hostuuid));
1723 tpr->pr_hostid = hid;
1727 if ((tallow = ch_allow & ~pr_allow)) {
1728 /* Clear allow bits in all children. */
1729 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1730 tpr->pr_allow &= ~tallow;
1732 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1734 * Persistent prisons get an extra reference, and prisons losing their
1735 * persist flag lose that reference. Only do this for existing prisons
1736 * for now, so new ones will remain unseen until after the module
1737 * handlers have completed.
1739 born = pr->pr_uref == 0;
1740 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1741 if (pr_flags & PR_PERSIST) {
1749 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1750 mtx_unlock(&pr->pr_mtx);
1753 if (racct_enable && created)
1754 prison_racct_attach(pr);
1757 /* Locks may have prevented a complete restriction of child IP
1758 * addresses. If so, allocate some more memory and try again.
1763 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1764 mtx_lock(&pr->pr_mtx);
1766 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1768 if (tpr->pr_flags & PR_VNET) {
1773 if (prison_restrict_ip4(tpr, ip4)) {
1780 mtx_unlock(&pr->pr_mtx);
1786 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1787 mtx_lock(&pr->pr_mtx);
1789 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1791 if (tpr->pr_flags & PR_VNET) {
1796 if (prison_restrict_ip6(tpr, ip6)) {
1803 mtx_unlock(&pr->pr_mtx);
1807 /* Let the modules do their work. */
1808 sx_downgrade(&allprison_lock);
1810 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1812 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1813 prison_deref(pr, created
1815 : PD_DEREF | PD_LIST_SLOCKED);
1819 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1822 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1823 prison_deref(pr, created
1825 : PD_DEREF | PD_LIST_SLOCKED);
1829 /* Attach this process to the prison if requested. */
1830 if (flags & JAIL_ATTACH) {
1831 mtx_lock(&pr->pr_mtx);
1832 error = do_jail_attach(td, pr);
1834 vfs_opterror(opts, "attach failed");
1836 prison_deref(pr, PD_DEREF);
1842 if (racct_enable && !created) {
1843 if (!(flags & JAIL_ATTACH))
1844 sx_sunlock(&allprison_lock);
1845 prison_racct_modify(pr);
1846 if (!(flags & JAIL_ATTACH))
1847 sx_slock(&allprison_lock);
1851 td->td_retval[0] = pr->pr_id;
1854 * Now that it is all there, drop the temporary reference from existing
1855 * prisons. Or add a reference to newly created persistent prisons
1856 * (which was not done earlier so that the prison would not be publicly
1860 prison_deref(pr, (flags & JAIL_ATTACH)
1862 : PD_DEREF | PD_LIST_SLOCKED);
1864 if (pr_flags & PR_PERSIST) {
1865 mtx_lock(&pr->pr_mtx);
1868 mtx_unlock(&pr->pr_mtx);
1870 if (!(flags & JAIL_ATTACH))
1871 sx_sunlock(&allprison_lock);
1877 prison_deref(pr, created
1878 ? PD_LOCKED | PD_LIST_XLOCKED
1879 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1882 sx_xunlock(&allprison_lock);
1888 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1889 &errmsg_len) == 0 && errmsg_len > 0) {
1890 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1891 if (optuio->uio_segflg == UIO_SYSSPACE)
1893 optuio->uio_iov[errmsg_pos].iov_base,
1897 optuio->uio_iov[errmsg_pos].iov_base,
1903 free(ip4, M_PRISON);
1906 free(ip6, M_PRISON);
1909 free(g_path, M_TEMP);
1915 * struct jail_get_args {
1916 * struct iovec *iovp;
1917 * unsigned int iovcnt;
1922 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1927 /* Check that we have an even number of iovecs. */
1928 if (uap->iovcnt & 1)
1931 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1934 error = kern_jail_get(td, auio, uap->flags);
1936 error = copyout(auio->uio_iov, uap->iovp,
1937 uap->iovcnt * sizeof (struct iovec));
1943 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1945 struct bool_flags *bf;
1946 struct jailsys_flags *jsf;
1947 struct prison *pr, *mypr;
1949 struct vfsoptlist *opts;
1950 char *errmsg, *name;
1951 int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1954 if (flags & ~JAIL_GET_MASK)
1957 /* Get the parameter list. */
1958 error = vfs_buildopts(optuio, &opts);
1961 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1962 mypr = td->td_ucred->cr_prison;
1965 * Find the prison specified by one of: lastjid, jid, name.
1967 sx_slock(&allprison_lock);
1968 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1970 TAILQ_FOREACH(pr, &allprison, pr_list) {
1971 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1972 mtx_lock(&pr->pr_mtx);
1973 if (pr->pr_ref > 0 &&
1974 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1976 mtx_unlock(&pr->pr_mtx);
1982 vfs_opterror(opts, "no jail after %d", jid);
1983 goto done_unlock_list;
1984 } else if (error != ENOENT)
1985 goto done_unlock_list;
1987 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1990 pr = prison_find_child(mypr, jid);
1992 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1993 mtx_unlock(&pr->pr_mtx);
1995 vfs_opterror(opts, "jail %d is dying",
1997 goto done_unlock_list;
2002 vfs_opterror(opts, "jail %d not found", jid);
2003 goto done_unlock_list;
2005 } else if (error != ENOENT)
2006 goto done_unlock_list;
2008 error = vfs_getopt(opts, "name", (void **)&name, &len);
2010 if (len == 0 || name[len - 1] != '\0') {
2012 goto done_unlock_list;
2014 pr = prison_find_name(mypr, name);
2016 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2017 mtx_unlock(&pr->pr_mtx);
2019 vfs_opterror(opts, "jail \"%s\" is dying",
2021 goto done_unlock_list;
2026 vfs_opterror(opts, "jail \"%s\" not found", name);
2027 goto done_unlock_list;
2028 } else if (error != ENOENT)
2029 goto done_unlock_list;
2031 vfs_opterror(opts, "no jail specified");
2033 goto done_unlock_list;
2036 /* Get the parameters of the prison. */
2039 td->td_retval[0] = pr->pr_id;
2040 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2041 if (error != 0 && error != ENOENT)
2043 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2044 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2045 if (error != 0 && error != ENOENT)
2047 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2048 if (error != 0 && error != ENOENT)
2050 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2051 sizeof(pr->pr_cpuset->cs_id));
2052 if (error != 0 && error != ENOENT)
2054 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2055 if (error != 0 && error != ENOENT)
2058 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2059 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2060 if (error != 0 && error != ENOENT)
2064 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2065 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2066 if (error != 0 && error != ENOENT)
2069 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2070 sizeof(pr->pr_securelevel));
2071 if (error != 0 && error != ENOENT)
2073 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2074 sizeof(pr->pr_childcount));
2075 if (error != 0 && error != ENOENT)
2077 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2078 sizeof(pr->pr_childmax));
2079 if (error != 0 && error != ENOENT)
2081 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2082 if (error != 0 && error != ENOENT)
2084 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2085 if (error != 0 && error != ENOENT)
2087 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2088 if (error != 0 && error != ENOENT)
2090 #ifdef COMPAT_FREEBSD32
2091 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2092 uint32_t hid32 = pr->pr_hostid;
2094 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2097 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2098 sizeof(pr->pr_hostid));
2099 if (error != 0 && error != ENOENT)
2101 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2102 sizeof(pr->pr_enforce_statfs));
2103 if (error != 0 && error != ENOENT)
2105 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2106 sizeof(pr->pr_devfs_rsnum));
2107 if (error != 0 && error != ENOENT)
2109 for (bf = pr_flag_bool;
2110 bf < pr_flag_bool + nitems(pr_flag_bool);
2112 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2113 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2114 if (error != 0 && error != ENOENT)
2117 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2118 if (error != 0 && error != ENOENT)
2121 for (jsf = pr_flag_jailsys;
2122 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2124 f = pr->pr_flags & (jsf->disable | jsf->new);
2125 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2126 : (f == jsf->new) ? JAIL_SYS_NEW
2128 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2129 if (error != 0 && error != ENOENT)
2132 for (bf = pr_flag_allow;
2133 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2135 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2136 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2137 if (error != 0 && error != ENOENT)
2140 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2141 if (error != 0 && error != ENOENT)
2144 i = (pr->pr_uref == 0);
2145 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2146 if (error != 0 && error != ENOENT)
2149 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2150 if (error != 0 && error != ENOENT)
2152 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2153 sizeof(pr->pr_osreldate));
2154 if (error != 0 && error != ENOENT)
2156 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2157 if (error != 0 && error != ENOENT)
2160 /* Get the module parameters. */
2161 mtx_unlock(&pr->pr_mtx);
2163 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2166 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2168 /* By now, all parameters should have been noted. */
2169 TAILQ_FOREACH(opt, opts, link) {
2170 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2172 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2177 /* Write the fetched parameters back to userspace. */
2179 TAILQ_FOREACH(opt, opts, link) {
2180 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2181 pos = 2 * opt->pos + 1;
2182 optuio->uio_iov[pos].iov_len = opt->len;
2183 if (opt->value != NULL) {
2184 if (optuio->uio_segflg == UIO_SYSSPACE) {
2186 optuio->uio_iov[pos].iov_base,
2189 error = copyout(opt->value,
2190 optuio->uio_iov[pos].iov_base,
2201 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2205 sx_sunlock(&allprison_lock);
2207 if (error && errmsg_pos >= 0) {
2208 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2209 errmsg_pos = 2 * errmsg_pos + 1;
2210 if (errmsg_len > 0) {
2211 if (optuio->uio_segflg == UIO_SYSSPACE)
2213 optuio->uio_iov[errmsg_pos].iov_base,
2217 optuio->uio_iov[errmsg_pos].iov_base,
2226 * struct jail_remove_args {
2231 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2233 struct prison *pr, *cpr, *lpr, *tpr;
2236 error = priv_check(td, PRIV_JAIL_REMOVE);
2240 sx_xlock(&allprison_lock);
2241 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2243 sx_xunlock(&allprison_lock);
2247 /* Remove all descendants of this prison, then remove this prison. */
2249 if (!LIST_EMPTY(&pr->pr_children)) {
2250 mtx_unlock(&pr->pr_mtx);
2252 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2253 mtx_lock(&cpr->pr_mtx);
2254 if (cpr->pr_ref > 0) {
2258 /* Already removed - do not do it again. */
2261 mtx_unlock(&cpr->pr_mtx);
2263 mtx_lock(&lpr->pr_mtx);
2264 prison_remove_one(lpr);
2265 sx_xlock(&allprison_lock);
2270 mtx_lock(&lpr->pr_mtx);
2271 prison_remove_one(lpr);
2272 sx_xlock(&allprison_lock);
2274 mtx_lock(&pr->pr_mtx);
2276 prison_remove_one(pr);
2281 prison_remove_one(struct prison *pr)
2286 /* If the prison was persistent, it is not anymore. */
2288 if (pr->pr_flags & PR_PERSIST) {
2291 pr->pr_flags &= ~PR_PERSIST;
2295 * jail_remove added a reference. If that's the only one, remove
2298 KASSERT(pr->pr_ref > 0,
2299 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2300 if (pr->pr_ref == 1) {
2302 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2306 mtx_unlock(&pr->pr_mtx);
2307 sx_xunlock(&allprison_lock);
2309 * Kill all processes unfortunate enough to be attached to this prison.
2311 sx_slock(&allproc_lock);
2312 FOREACH_PROC_IN_SYSTEM(p) {
2314 if (p->p_state != PRS_NEW && p->p_ucred &&
2315 p->p_ucred->cr_prison == pr)
2316 kern_psignal(p, SIGKILL);
2319 sx_sunlock(&allproc_lock);
2320 /* Remove the temporary reference added by jail_remove. */
2321 prison_deref(pr, deuref | PD_DEREF);
2325 * struct jail_attach_args {
2330 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2335 error = priv_check(td, PRIV_JAIL_ATTACH);
2340 * Start with exclusive hold on allprison_lock to ensure that a possible
2341 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2342 * But then immediately downgrade it since we don't need to stop
2345 sx_xlock(&allprison_lock);
2346 sx_downgrade(&allprison_lock);
2347 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2349 sx_sunlock(&allprison_lock);
2354 * Do not allow a process to attach to a prison that is not
2355 * considered to be "alive".
2357 if (pr->pr_uref == 0) {
2358 mtx_unlock(&pr->pr_mtx);
2359 sx_sunlock(&allprison_lock);
2363 return (do_jail_attach(td, pr));
2367 do_jail_attach(struct thread *td, struct prison *pr)
2370 struct ucred *newcred, *oldcred;
2374 * XXX: Note that there is a slight race here if two threads
2375 * in the same privileged process attempt to attach to two
2376 * different jails at the same time. It is important for
2377 * user processes not to do this, or they might end up with
2378 * a process root from one prison, but attached to the jail
2383 mtx_unlock(&pr->pr_mtx);
2385 /* Let modules do whatever they need to prepare for attaching. */
2386 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2388 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2391 sx_sunlock(&allprison_lock);
2394 * Reparent the newly attached process to this jail.
2397 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2401 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2402 if ((error = change_dir(pr->pr_root, td)) != 0)
2405 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2408 VOP_UNLOCK(pr->pr_root);
2409 if ((error = pwd_chroot(td, pr->pr_root)))
2414 oldcred = crcopysafe(p, newcred);
2415 newcred->cr_prison = pr;
2416 proc_set_cred(p, newcred);
2419 racct_proc_ucred_changed(p, oldcred, newcred);
2424 rctl_proc_ucred_changed(p, newcred);
2427 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2432 VOP_UNLOCK(pr->pr_root);
2434 /* Tell modules this thread is still in its old jail after all. */
2435 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2436 prison_deref(pr, PD_DEREF | PD_DEUREF);
2441 * Returns a locked prison instance, or NULL on failure.
2444 prison_find(int prid)
2448 sx_assert(&allprison_lock, SX_LOCKED);
2449 TAILQ_FOREACH(pr, &allprison, pr_list) {
2450 if (pr->pr_id == prid) {
2451 mtx_lock(&pr->pr_mtx);
2454 mtx_unlock(&pr->pr_mtx);
2461 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2464 prison_find_child(struct prison *mypr, int prid)
2469 sx_assert(&allprison_lock, SX_LOCKED);
2470 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2471 if (pr->pr_id == prid) {
2472 mtx_lock(&pr->pr_mtx);
2475 mtx_unlock(&pr->pr_mtx);
2482 * Look for the name relative to mypr. Returns a locked prison or NULL.
2485 prison_find_name(struct prison *mypr, const char *name)
2487 struct prison *pr, *deadpr;
2491 sx_assert(&allprison_lock, SX_LOCKED);
2492 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2495 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2496 if (!strcmp(pr->pr_name + mylen, name)) {
2497 mtx_lock(&pr->pr_mtx);
2498 if (pr->pr_ref > 0) {
2499 if (pr->pr_uref > 0)
2503 mtx_unlock(&pr->pr_mtx);
2506 /* There was no valid prison - perhaps there was a dying one. */
2507 if (deadpr != NULL) {
2508 mtx_lock(&deadpr->pr_mtx);
2509 if (deadpr->pr_ref == 0) {
2510 mtx_unlock(&deadpr->pr_mtx);
2518 * See if a prison has the specific flag set.
2521 prison_flag(struct ucred *cred, unsigned flag)
2524 /* This is an atomic read, so no locking is necessary. */
2525 return (cred->cr_prison->pr_flags & flag);
2529 prison_allow(struct ucred *cred, unsigned flag)
2532 /* This is an atomic read, so no locking is necessary. */
2533 return (cred->cr_prison->pr_allow & flag);
2537 * Remove a prison reference. If that was the last reference, remove the
2538 * prison itself - but not in this context in case there are locks held.
2541 prison_free_locked(struct prison *pr)
2545 mtx_assert(&pr->pr_mtx, MA_OWNED);
2547 mtx_unlock(&pr->pr_mtx);
2549 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2553 prison_free(struct prison *pr)
2556 mtx_lock(&pr->pr_mtx);
2557 prison_free_locked(pr);
2561 * Complete a call to either prison_free or prison_proc_free.
2564 prison_complete(void *context, int pending)
2566 struct prison *pr = context;
2568 sx_xlock(&allprison_lock);
2569 mtx_lock(&pr->pr_mtx);
2570 prison_deref(pr, pr->pr_uref
2571 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2572 : PD_LOCKED | PD_LIST_XLOCKED);
2576 * Remove a prison reference (usually). This internal version assumes no
2577 * mutexes are held, except perhaps the prison itself. If there are no more
2578 * references, release and delist the prison. On completion, the prison lock
2579 * and the allprison lock are both unlocked.
2582 prison_deref(struct prison *pr, int flags)
2584 struct prison *ppr, *tpr;
2587 if (!(flags & PD_LOCKED))
2588 mtx_lock(&pr->pr_mtx);
2590 if (flags & PD_DEUREF) {
2591 KASSERT(pr->pr_uref > 0,
2592 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2595 lasturef = pr->pr_uref == 0;
2598 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2601 if (flags & PD_DEREF) {
2602 KASSERT(pr->pr_ref > 0,
2603 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2608 mtx_unlock(&pr->pr_mtx);
2611 * Tell the modules if the last user reference was removed
2612 * (even it sticks around in dying state).
2615 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2616 sx_xlock(&allprison_lock);
2617 flags |= PD_LIST_XLOCKED;
2619 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2620 mtx_lock(&pr->pr_mtx);
2622 mtx_unlock(&pr->pr_mtx);
2625 /* If the prison still has references, nothing else to do. */
2627 if (flags & PD_LIST_SLOCKED)
2628 sx_sunlock(&allprison_lock);
2629 else if (flags & PD_LIST_XLOCKED)
2630 sx_xunlock(&allprison_lock);
2634 if (flags & PD_LIST_SLOCKED) {
2635 if (!sx_try_upgrade(&allprison_lock)) {
2636 sx_sunlock(&allprison_lock);
2637 sx_xlock(&allprison_lock);
2639 } else if (!(flags & PD_LIST_XLOCKED))
2640 sx_xlock(&allprison_lock);
2642 TAILQ_REMOVE(&allprison, pr, pr_list);
2643 LIST_REMOVE(pr, pr_sibling);
2644 ppr = pr->pr_parent;
2645 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2646 tpr->pr_childcount--;
2647 sx_xunlock(&allprison_lock);
2650 if (pr->pr_vnet != ppr->pr_vnet)
2651 vnet_destroy(pr->pr_vnet);
2653 if (pr->pr_root != NULL)
2655 mtx_destroy(&pr->pr_mtx);
2657 free(pr->pr_ip4, M_PRISON);
2660 free(pr->pr_ip6, M_PRISON);
2662 if (pr->pr_cpuset != NULL)
2663 cpuset_rel(pr->pr_cpuset);
2667 prison_racct_detach(pr);
2671 /* Removing a prison frees a reference on its parent. */
2673 mtx_lock(&pr->pr_mtx);
2674 flags = PD_DEREF | PD_DEUREF;
2679 prison_hold_locked(struct prison *pr)
2682 mtx_assert(&pr->pr_mtx, MA_OWNED);
2683 KASSERT(pr->pr_ref > 0,
2684 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2689 prison_hold(struct prison *pr)
2692 mtx_lock(&pr->pr_mtx);
2693 prison_hold_locked(pr);
2694 mtx_unlock(&pr->pr_mtx);
2698 prison_proc_hold(struct prison *pr)
2701 mtx_lock(&pr->pr_mtx);
2702 KASSERT(pr->pr_uref > 0,
2703 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2705 mtx_unlock(&pr->pr_mtx);
2709 prison_proc_free(struct prison *pr)
2712 mtx_lock(&pr->pr_mtx);
2713 KASSERT(pr->pr_uref > 0,
2714 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2715 if (pr->pr_uref > 1)
2719 * Don't remove the last user reference in this context, which
2720 * is expected to be a process that is not only locked, but
2724 mtx_unlock(&pr->pr_mtx);
2725 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2728 mtx_unlock(&pr->pr_mtx);
2732 * Check if a jail supports the given address family.
2734 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2738 prison_check_af(struct ucred *cred, int af)
2743 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2745 pr = cred->cr_prison;
2747 /* Prisons with their own network stack are not limited. */
2748 if (prison_owns_vnet(cred))
2757 if (pr->pr_flags & PR_IP4)
2759 mtx_lock(&pr->pr_mtx);
2760 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2761 error = EAFNOSUPPORT;
2762 mtx_unlock(&pr->pr_mtx);
2768 if (pr->pr_flags & PR_IP6)
2770 mtx_lock(&pr->pr_mtx);
2771 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2772 error = EAFNOSUPPORT;
2773 mtx_unlock(&pr->pr_mtx);
2781 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2782 error = EAFNOSUPPORT;
2788 * Check if given address belongs to the jail referenced by cred (wrapper to
2789 * prison_check_ip[46]).
2791 * Returns 0 if jail doesn't restrict the address family or if address belongs
2792 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2793 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
2796 prison_if(struct ucred *cred, const struct sockaddr *sa)
2799 const struct sockaddr_in *sai;
2802 const struct sockaddr_in6 *sai6;
2806 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2807 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2810 if (prison_owns_vnet(cred))
2815 switch (sa->sa_family)
2819 sai = (const struct sockaddr_in *)sa;
2820 error = prison_check_ip4(cred, &sai->sin_addr);
2825 sai6 = (const struct sockaddr_in6 *)sa;
2826 error = prison_check_ip6(cred, &sai6->sin6_addr);
2830 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2831 error = EAFNOSUPPORT;
2837 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2840 prison_check(struct ucred *cred1, struct ucred *cred2)
2843 return ((cred1->cr_prison == cred2->cr_prison ||
2844 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2848 * Return 1 if p2 is a child of p1, otherwise 0.
2851 prison_ischild(struct prison *pr1, struct prison *pr2)
2854 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2861 * Return 1 if the passed credential is in a jail and that jail does not
2862 * have its own virtual network stack, otherwise 0.
2865 jailed_without_vnet(struct ucred *cred)
2871 if (prison_owns_vnet(cred))
2879 * Return the correct hostname (domainname, et al) for the passed credential.
2882 getcredhostname(struct ucred *cred, char *buf, size_t size)
2887 * A NULL credential can be used to shortcut to the physical
2888 * system's hostname.
2890 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2891 mtx_lock(&pr->pr_mtx);
2892 strlcpy(buf, pr->pr_hostname, size);
2893 mtx_unlock(&pr->pr_mtx);
2897 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2900 mtx_lock(&cred->cr_prison->pr_mtx);
2901 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2902 mtx_unlock(&cred->cr_prison->pr_mtx);
2906 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2909 mtx_lock(&cred->cr_prison->pr_mtx);
2910 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2911 mtx_unlock(&cred->cr_prison->pr_mtx);
2915 getcredhostid(struct ucred *cred, unsigned long *hostid)
2918 mtx_lock(&cred->cr_prison->pr_mtx);
2919 *hostid = cred->cr_prison->pr_hostid;
2920 mtx_unlock(&cred->cr_prison->pr_mtx);
2924 getjailname(struct ucred *cred, char *name, size_t len)
2927 mtx_lock(&cred->cr_prison->pr_mtx);
2928 strlcpy(name, cred->cr_prison->pr_name, len);
2929 mtx_unlock(&cred->cr_prison->pr_mtx);
2934 * Determine whether the prison represented by cred owns
2935 * its vnet rather than having it inherited.
2937 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2940 prison_owns_vnet(struct ucred *cred)
2944 * vnets cannot be added/removed after jail creation,
2945 * so no need to lock here.
2947 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2952 * Determine whether the subject represented by cred can "see"
2953 * status of a mount point.
2954 * Returns: 0 for permitted, ENOENT otherwise.
2955 * XXX: This function should be called cr_canseemount() and should be
2956 * placed in kern_prot.c.
2959 prison_canseemount(struct ucred *cred, struct mount *mp)
2965 pr = cred->cr_prison;
2966 if (pr->pr_enforce_statfs == 0)
2968 if (pr->pr_root->v_mount == mp)
2970 if (pr->pr_enforce_statfs == 2)
2973 * If jail's chroot directory is set to "/" we should be able to see
2974 * all mount-points from inside a jail.
2975 * This is ugly check, but this is the only situation when jail's
2976 * directory ends with '/'.
2978 if (strcmp(pr->pr_path, "/") == 0)
2980 len = strlen(pr->pr_path);
2982 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2985 * Be sure that we don't have situation where jail's root directory
2986 * is "/some/path" and mount point is "/some/pathpath".
2988 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2994 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2996 char jpath[MAXPATHLEN];
3000 pr = cred->cr_prison;
3001 if (pr->pr_enforce_statfs == 0)
3003 if (prison_canseemount(cred, mp) != 0) {
3004 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3005 strlcpy(sp->f_mntonname, "[restricted]",
3006 sizeof(sp->f_mntonname));
3009 if (pr->pr_root->v_mount == mp) {
3011 * Clear current buffer data, so we are sure nothing from
3012 * the valid path left there.
3014 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3015 *sp->f_mntonname = '/';
3019 * If jail's chroot directory is set to "/" we should be able to see
3020 * all mount-points from inside a jail.
3022 if (strcmp(pr->pr_path, "/") == 0)
3024 len = strlen(pr->pr_path);
3025 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3027 * Clear current buffer data, so we are sure nothing from
3028 * the valid path left there.
3030 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3031 if (*jpath == '\0') {
3032 /* Should never happen. */
3033 *sp->f_mntonname = '/';
3035 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3040 * Check with permission for a specific privilege is granted within jail. We
3041 * have a specific list of accepted privileges; the rest are denied.
3044 prison_priv_check(struct ucred *cred, int priv)
3048 * Some policies have custom handlers. This routine should not be
3049 * called for them. See priv_check_cred().
3052 case PRIV_VFS_LOOKUP:
3053 case PRIV_VFS_GENERATION:
3054 KASSERT(0, ("prison_priv_check instead of a custom handler "
3055 "called for %d\n", priv));
3063 * Privileges specific to prisons with a virtual network stack.
3064 * There might be a duplicate entry here in case the privilege
3065 * is only granted conditionally in the legacy jail case.
3070 * NFS-specific privileges.
3072 case PRIV_NFS_DAEMON:
3073 case PRIV_NFS_LOCKD:
3076 * Network stack privileges.
3078 case PRIV_NET_BRIDGE:
3081 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3082 case PRIV_NET_ROUTE:
3084 case PRIV_NET_SETIFMTU:
3085 case PRIV_NET_SETIFFLAGS:
3086 case PRIV_NET_SETIFCAP:
3087 case PRIV_NET_SETIFDESCR:
3088 case PRIV_NET_SETIFNAME :
3089 case PRIV_NET_SETIFMETRIC:
3090 case PRIV_NET_SETIFPHYS:
3091 case PRIV_NET_SETIFMAC:
3092 case PRIV_NET_SETLANPCP:
3093 case PRIV_NET_ADDMULTI:
3094 case PRIV_NET_DELMULTI:
3095 case PRIV_NET_HWIOCTL:
3096 case PRIV_NET_SETLLADDR:
3097 case PRIV_NET_ADDIFGROUP:
3098 case PRIV_NET_DELIFGROUP:
3099 case PRIV_NET_IFCREATE:
3100 case PRIV_NET_IFDESTROY:
3101 case PRIV_NET_ADDIFADDR:
3102 case PRIV_NET_DELIFADDR:
3105 case PRIV_NET_SETIFVNET:
3106 case PRIV_NET_SETIFFIB:
3109 * 802.11-related privileges.
3111 case PRIV_NET80211_VAP_GETKEY:
3112 case PRIV_NET80211_VAP_MANAGE:
3118 case PRIV_NETATM_CFG:
3119 case PRIV_NETATM_ADD:
3120 case PRIV_NETATM_DEL:
3121 case PRIV_NETATM_SET:
3124 * Bluetooth privileges.
3126 case PRIV_NETBLUETOOTH_RAW:
3130 * Netgraph and netgraph module privileges.
3132 case PRIV_NETGRAPH_CONTROL:
3134 case PRIV_NETGRAPH_TTY:
3138 * IPv4 and IPv6 privileges.
3140 case PRIV_NETINET_IPFW:
3141 case PRIV_NETINET_DIVERT:
3142 case PRIV_NETINET_PF:
3143 case PRIV_NETINET_DUMMYNET:
3144 case PRIV_NETINET_CARP:
3145 case PRIV_NETINET_MROUTE:
3146 case PRIV_NETINET_RAW:
3147 case PRIV_NETINET_ADDRCTRL6:
3148 case PRIV_NETINET_ND6:
3149 case PRIV_NETINET_SCOPE6:
3150 case PRIV_NETINET_ALIFETIME6:
3151 case PRIV_NETINET_IPSEC:
3152 case PRIV_NETINET_BINDANY:
3167 * No default: or deny here.
3168 * In case of no permit fall through to next switch().
3170 if (cred->cr_prison->pr_flags & PR_VNET)
3177 * Allow ktrace privileges for root in jail.
3183 * Allow jailed processes to configure audit identity and
3184 * submit audit records (login, etc). In the future we may
3185 * want to further refine the relationship between audit and
3188 case PRIV_AUDIT_GETAUDIT:
3189 case PRIV_AUDIT_SETAUDIT:
3190 case PRIV_AUDIT_SUBMIT:
3194 * Allow jailed processes to manipulate process UNIX
3195 * credentials in any way they see fit.
3197 case PRIV_CRED_SETUID:
3198 case PRIV_CRED_SETEUID:
3199 case PRIV_CRED_SETGID:
3200 case PRIV_CRED_SETEGID:
3201 case PRIV_CRED_SETGROUPS:
3202 case PRIV_CRED_SETREUID:
3203 case PRIV_CRED_SETREGID:
3204 case PRIV_CRED_SETRESUID:
3205 case PRIV_CRED_SETRESGID:
3208 * Jail implements visibility constraints already, so allow
3209 * jailed root to override uid/gid-based constraints.
3211 case PRIV_SEEOTHERGIDS:
3212 case PRIV_SEEOTHERUIDS:
3215 * Jail implements inter-process debugging limits already, so
3216 * allow jailed root various debugging privileges.
3218 case PRIV_DEBUG_DIFFCRED:
3219 case PRIV_DEBUG_SUGID:
3220 case PRIV_DEBUG_UNPRIV:
3223 * Allow jail to set various resource limits and login
3224 * properties, and for now, exceed process resource limits.
3226 case PRIV_PROC_LIMIT:
3227 case PRIV_PROC_SETLOGIN:
3228 case PRIV_PROC_SETRLIMIT:
3231 * System V and POSIX IPC privileges are granted in jail.
3234 case PRIV_IPC_WRITE:
3235 case PRIV_IPC_ADMIN:
3236 case PRIV_IPC_MSGSIZE:
3240 * Jail operations within a jail work on child jails.
3242 case PRIV_JAIL_ATTACH:
3244 case PRIV_JAIL_REMOVE:
3247 * Jail implements its own inter-process limits, so allow
3248 * root processes in jail to change scheduling on other
3249 * processes in the same jail. Likewise for signalling.
3251 case PRIV_SCHED_DIFFCRED:
3252 case PRIV_SCHED_CPUSET:
3253 case PRIV_SIGNAL_DIFFCRED:
3254 case PRIV_SIGNAL_SUGID:
3257 * Allow jailed processes to write to sysctls marked as jail
3260 case PRIV_SYSCTL_WRITEJAIL:
3263 * Allow root in jail to manage a variety of quota
3264 * properties. These should likely be conditional on a
3265 * configuration option.
3267 case PRIV_VFS_GETQUOTA:
3268 case PRIV_VFS_SETQUOTA:
3271 * Since Jail relies on chroot() to implement file system
3272 * protections, grant many VFS privileges to root in jail.
3273 * Be careful to exclude mount-related and NFS-related
3277 case PRIV_VFS_WRITE:
3278 case PRIV_VFS_ADMIN:
3280 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3281 case PRIV_VFS_CHFLAGS_DEV:
3282 case PRIV_VFS_CHOWN:
3283 case PRIV_VFS_CHROOT:
3284 case PRIV_VFS_RETAINSUGID:
3285 case PRIV_VFS_FCHROOT:
3287 case PRIV_VFS_SETGID:
3289 case PRIV_VFS_STICKYFILE:
3292 * As in the non-jail case, non-root users are expected to be
3293 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3294 * exists in the jail and they have permission to access it).
3296 case PRIV_KMEM_READ:
3300 * Depending on the global setting, allow privilege of
3301 * setting system flags.
3303 case PRIV_VFS_SYSFLAGS:
3304 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3310 * Depending on the global setting, allow privilege of
3311 * mounting/unmounting file systems.
3313 case PRIV_VFS_MOUNT:
3314 case PRIV_VFS_UNMOUNT:
3315 case PRIV_VFS_MOUNT_NONUSER:
3316 case PRIV_VFS_MOUNT_OWNER:
3317 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3318 cred->cr_prison->pr_enforce_statfs < 2)
3324 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3325 * policy. priv_check_cred will not specifically allow it, and
3326 * we may want a MAC policy to allow it.
3328 case PRIV_VFS_READ_DIR:
3332 * Conditionnaly allow locking (unlocking) physical pages
3336 case PRIV_VM_MUNLOCK:
3337 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3343 * Conditionally allow jailed root to bind reserved ports.
3345 case PRIV_NETINET_RESERVEDPORT:
3346 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3352 * Allow jailed root to reuse in-use ports.
3354 case PRIV_NETINET_REUSEPORT:
3358 * Allow jailed root to set certain IPv4/6 (option) headers.
3360 case PRIV_NETINET_SETHDROPTS:
3364 * Conditionally allow creating raw sockets in jail.
3366 case PRIV_NETINET_RAW:
3367 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3373 * Since jail implements its own visibility limits on netstat
3374 * sysctls, allow getcred. This allows identd to work in
3377 case PRIV_NETINET_GETCRED:
3381 * Allow jailed root to set loginclass.
3383 case PRIV_PROC_SETLOGINCLASS:
3387 * Do not allow a process inside a jail to read the kernel
3388 * message buffer unless explicitly permitted.
3391 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3397 * In all remaining cases, deny the privilege request. This
3398 * includes almost all network privileges, many system
3399 * configuration privileges.
3406 * Return the part of pr2's name that is relative to pr1, or the whole name
3407 * if it does not directly follow.
3411 prison_name(struct prison *pr1, struct prison *pr2)
3415 /* Jails see themselves as "0" (if they see themselves at all). */
3418 name = pr2->pr_name;
3419 if (prison_ischild(pr1, pr2)) {
3421 * pr1 isn't locked (and allprison_lock may not be either)
3422 * so its length can't be counted on. But the number of dots
3423 * can be counted on - and counted.
3425 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3426 name = strchr(name, '.') + 1;
3432 * Return the part of pr2's path that is relative to pr1, or the whole path
3433 * if it does not directly follow.
3436 prison_path(struct prison *pr1, struct prison *pr2)
3438 char *path1, *path2;
3441 path1 = pr1->pr_path;
3442 path2 = pr2->pr_path;
3443 if (!strcmp(path1, "/"))
3445 len1 = strlen(path1);
3446 if (strncmp(path1, path2, len1))
3448 if (path2[len1] == '\0')
3450 if (path2[len1] == '/')
3451 return (path2 + len1);
3456 * Jail-related sysctls.
3458 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3462 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3465 struct prison *pr, *cpr;
3467 struct in_addr *ip4 = NULL;
3471 struct in6_addr *ip6 = NULL;
3476 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3477 pr = req->td->td_ucred->cr_prison;
3479 sx_slock(&allprison_lock);
3480 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3481 #if defined(INET) || defined(INET6)
3484 mtx_lock(&cpr->pr_mtx);
3486 if (cpr->pr_ip4s > 0) {
3487 if (ip4s < cpr->pr_ip4s) {
3488 ip4s = cpr->pr_ip4s;
3489 mtx_unlock(&cpr->pr_mtx);
3490 ip4 = realloc(ip4, ip4s *
3491 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3494 bcopy(cpr->pr_ip4, ip4,
3495 cpr->pr_ip4s * sizeof(struct in_addr));
3499 if (cpr->pr_ip6s > 0) {
3500 if (ip6s < cpr->pr_ip6s) {
3501 ip6s = cpr->pr_ip6s;
3502 mtx_unlock(&cpr->pr_mtx);
3503 ip6 = realloc(ip6, ip6s *
3504 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3507 bcopy(cpr->pr_ip6, ip6,
3508 cpr->pr_ip6s * sizeof(struct in6_addr));
3511 if (cpr->pr_ref == 0) {
3512 mtx_unlock(&cpr->pr_mtx);
3515 bzero(xp, sizeof(*xp));
3516 xp->pr_version = XPRISON_VERSION;
3517 xp->pr_id = cpr->pr_id;
3518 xp->pr_state = cpr->pr_uref > 0
3519 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3520 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3521 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3522 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3524 xp->pr_ip4s = cpr->pr_ip4s;
3527 xp->pr_ip6s = cpr->pr_ip6s;
3529 mtx_unlock(&cpr->pr_mtx);
3530 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3534 if (xp->pr_ip4s > 0) {
3535 error = SYSCTL_OUT(req, ip4,
3536 xp->pr_ip4s * sizeof(struct in_addr));
3542 if (xp->pr_ip6s > 0) {
3543 error = SYSCTL_OUT(req, ip6,
3544 xp->pr_ip6s * sizeof(struct in6_addr));
3550 sx_sunlock(&allprison_lock);
3561 SYSCTL_OID(_security_jail, OID_AUTO, list,
3562 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3563 sysctl_jail_list, "S", "List of active jails");
3566 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3570 injail = jailed(req->td->td_ucred);
3571 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3576 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3577 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3578 sysctl_jail_jailed, "I", "Process in jail?");
3581 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3583 int error, havevnet;
3585 struct ucred *cred = req->td->td_ucred;
3587 havevnet = jailed(cred) && prison_owns_vnet(cred);
3591 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3596 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3597 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3598 sysctl_jail_vnet, "I", "Jail owns vnet?");
3600 #if defined(INET) || defined(INET6)
3601 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3602 &jail_max_af_ips, 0,
3603 "Number of IP addresses a jail may have at most per address family (deprecated)");
3607 * Default parameters for jail(2) compatibility. For historical reasons,
3608 * the sysctl names have varying similarity to the parameter names. Prisons
3609 * just see their own parameters, and can't change them.
3612 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3615 int allow, error, i;
3617 pr = req->td->td_ucred->cr_prison;
3618 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3620 /* Get the current flag value, and convert it to a boolean. */
3621 i = (allow & arg2) ? 1 : 0;
3624 error = sysctl_handle_int(oidp, &i, 0, req);
3625 if (error || !req->newptr)
3631 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3634 mtx_lock(&prison0.pr_mtx);
3635 jail_default_allow = (jail_default_allow & ~arg2) | i;
3636 mtx_unlock(&prison0.pr_mtx);
3640 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3641 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3642 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3643 "Processes in jail can set their hostnames (deprecated)");
3644 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3645 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3646 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3647 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3648 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3649 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3650 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3651 "Processes in jail can use System V IPC primitives (deprecated)");
3652 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3653 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3654 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3655 "Prison root can create raw sockets (deprecated)");
3656 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3657 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3658 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3659 "Processes in jail can alter system file flags (deprecated)");
3660 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3661 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3662 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3663 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3666 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3671 pr = req->td->td_ucred->cr_prison;
3672 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3673 error = sysctl_handle_int(oidp, &level, 0, req);
3674 if (error || !req->newptr)
3676 *(int *)arg1 = level;
3680 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3681 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3682 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3683 sysctl_jail_default_level, "I",
3684 "Processes in jail cannot see all mounted file systems (deprecated)");
3686 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3687 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3688 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3689 sysctl_jail_default_level, "I",
3690 "Ruleset for the devfs filesystem in jail (deprecated)");
3693 * Nodes to describe jail parameters. Maximum length of string parameters
3694 * is returned in the string itself, and the other parameters exist merely
3695 * to make themselves and their types known.
3697 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3701 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3708 switch (oidp->oid_kind & CTLTYPE)
3714 if (!(req->flags & SCTL_MASK32))
3716 return (SYSCTL_OUT(req, &l, sizeof(l)));
3720 return (SYSCTL_OUT(req, &i, sizeof(i)));
3721 case CTLTYPE_STRING:
3722 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3724 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3725 case CTLTYPE_STRUCT:
3727 return (SYSCTL_OUT(req, &s, sizeof(s)));
3733 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3734 * jail creation time but cannot be changed in an existing jail.
3736 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3737 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3738 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3739 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3740 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3741 "I", "Jail secure level");
3742 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3743 "Jail value for kern.osreldate and uname -K");
3744 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3745 "Jail value for kern.osrelease and uname -r");
3746 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3747 "I", "Jail cannot see all mounted file systems");
3748 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3749 "I", "Ruleset for in-jail devfs mounts");
3750 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3751 "B", "Jail persistence");
3753 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3754 "E,jailsys", "Virtual network stack");
3756 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3757 "B", "Jail is in the process of shutting down");
3759 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3760 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3761 "I", "Current number of child jails");
3762 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3763 "I", "Maximum number of child jails");
3765 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3766 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3768 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3769 "Jail NIS domainname");
3770 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3772 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3773 "LU", "Jail host ID");
3775 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3776 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3779 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3780 "Jail IPv4 address virtualization");
3781 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3782 "S,in_addr,a", "Jail IPv4 addresses");
3783 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3784 "B", "Do (not) use IPv4 source address selection rather than the "
3785 "primary jail IPv4 address.");
3788 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3789 "Jail IPv6 address virtualization");
3790 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3791 "S,in6_addr,a", "Jail IPv6 addresses");
3792 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3793 "B", "Do (not) use IPv6 source address selection rather than the "
3794 "primary jail IPv6 address.");
3797 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3798 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3799 "B", "Jail may set hostname");
3800 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3801 "B", "Jail may use SYSV IPC");
3802 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3803 "B", "Jail may create raw sockets");
3804 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3805 "B", "Jail may alter system file flags");
3806 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3807 "B", "Jail may set file quotas");
3808 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3809 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3810 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3811 "B", "Jail may lock (unlock) physical pages in memory");
3812 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3813 "B", "Jail may bind sockets to reserved ports");
3814 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3815 "B", "Jail may read the kernel message buffer");
3816 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
3817 "B", "Unprivileged processes may use process debugging facilities");
3819 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3820 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3821 "B", "Jail may mount/unmount jail-friendly file systems in general");
3824 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
3825 * its associated bit in the pr_allow bitmask, or zero if the parameter was
3829 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3832 struct bool_flags *bf;
3833 struct sysctl_oid *parent;
3834 char *allow_name, *allow_noname, *allowed;
3835 #ifndef NO_SYSCTL_DESCR
3836 char *descr_deprecated;
3838 unsigned allow_flag;
3841 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3843 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3845 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3846 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3847 free(allow_name, M_PRISON);
3852 * See if this parameter has already beed added, i.e. a module was
3853 * previously loaded/unloaded.
3855 mtx_lock(&prison0.pr_mtx);
3856 for (bf = pr_flag_allow;
3857 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3859 if (strcmp(bf->name, allow_name) == 0) {
3860 allow_flag = bf->flag;
3866 * Find a free bit in prison0's pr_allow, failing if there are none
3867 * (which shouldn't happen as long as we keep track of how many
3868 * potential dynamic flags exist).
3870 * Due to per-jail unprivileged process debugging support
3871 * using pr_allow, also verify against PR_ALLOW_ALL_STATIC.
3872 * prison0 may have unprivileged process debugging unset.
3874 for (allow_flag = 1;; allow_flag <<= 1) {
3875 if (allow_flag == 0)
3877 if (allow_flag & PR_ALLOW_ALL_STATIC)
3879 if ((prison0.pr_allow & allow_flag) == 0)
3884 * Note the parameter in the next open slot in pr_flag_allow.
3885 * Set the flag last so code that checks pr_flag_allow can do so
3888 for (bf = pr_flag_allow; bf->flag != 0; bf++)
3889 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3890 /* This should never happen, but is not fatal. */
3894 prison0.pr_allow |= allow_flag;
3895 bf->name = allow_name;
3896 bf->noname = allow_noname;
3897 bf->flag = allow_flag;
3898 mtx_unlock(&prison0.pr_mtx);
3901 * Create sysctls for the paramter, and the back-compat global
3905 ? SYSCTL_ADD_NODE(NULL,
3906 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3907 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
3908 : &sysctl___security_jail_param_allow;
3909 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3910 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3911 NULL, 0, sysctl_jail_param, "B", descr);
3913 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3914 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3915 #ifndef NO_SYSCTL_DESCR
3916 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3919 (void)SYSCTL_ADD_PROC(NULL,
3920 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3921 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3922 sysctl_jail_default_allow, "I", descr_deprecated);
3923 #ifndef NO_SYSCTL_DESCR
3924 free(descr_deprecated, M_TEMP);
3926 free(allowed, M_TEMP);
3931 mtx_unlock(&prison0.pr_mtx);
3932 free(allow_name, M_PRISON);
3933 free(allow_noname, M_PRISON);
3938 * The VFS system will register jail-aware filesystems here. They each get
3939 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3940 * attempts to mount.
3943 prison_add_vfs(struct vfsconf *vfsp)
3945 #ifdef NO_SYSCTL_DESCR
3947 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3952 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3954 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3956 free(descr, M_TEMP);
3962 prison_racct_foreach(void (*callback)(struct racct *racct,
3963 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3964 void *arg2, void *arg3)
3966 struct prison_racct *prr;
3968 ASSERT_RACCT_ENABLED();
3970 sx_slock(&allprison_lock);
3973 LIST_FOREACH(prr, &allprison_racct, prr_next)
3974 (callback)(prr->prr_racct, arg2, arg3);
3977 sx_sunlock(&allprison_lock);
3980 static struct prison_racct *
3981 prison_racct_find_locked(const char *name)
3983 struct prison_racct *prr;
3985 ASSERT_RACCT_ENABLED();
3986 sx_assert(&allprison_lock, SA_XLOCKED);
3988 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3991 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3992 if (strcmp(name, prr->prr_name) != 0)
3995 /* Found prison_racct with a matching name? */
3996 prison_racct_hold(prr);
4000 /* Add new prison_racct. */
4001 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4002 racct_create(&prr->prr_racct);
4004 strcpy(prr->prr_name, name);
4005 refcount_init(&prr->prr_refcount, 1);
4006 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4011 struct prison_racct *
4012 prison_racct_find(const char *name)
4014 struct prison_racct *prr;
4016 ASSERT_RACCT_ENABLED();
4018 sx_xlock(&allprison_lock);
4019 prr = prison_racct_find_locked(name);
4020 sx_xunlock(&allprison_lock);
4025 prison_racct_hold(struct prison_racct *prr)
4028 ASSERT_RACCT_ENABLED();
4030 refcount_acquire(&prr->prr_refcount);
4034 prison_racct_free_locked(struct prison_racct *prr)
4037 ASSERT_RACCT_ENABLED();
4038 sx_assert(&allprison_lock, SA_XLOCKED);
4040 if (refcount_release(&prr->prr_refcount)) {
4041 racct_destroy(&prr->prr_racct);
4042 LIST_REMOVE(prr, prr_next);
4043 free(prr, M_PRISON_RACCT);
4048 prison_racct_free(struct prison_racct *prr)
4051 ASSERT_RACCT_ENABLED();
4052 sx_assert(&allprison_lock, SA_UNLOCKED);
4054 if (refcount_release_if_not_last(&prr->prr_refcount))
4057 sx_xlock(&allprison_lock);
4058 prison_racct_free_locked(prr);
4059 sx_xunlock(&allprison_lock);
4063 prison_racct_attach(struct prison *pr)
4065 struct prison_racct *prr;
4067 ASSERT_RACCT_ENABLED();
4068 sx_assert(&allprison_lock, SA_XLOCKED);
4070 prr = prison_racct_find_locked(pr->pr_name);
4071 KASSERT(prr != NULL, ("cannot find prison_racct"));
4073 pr->pr_prison_racct = prr;
4077 * Handle jail renaming. From the racct point of view, renaming means
4078 * moving from one prison_racct to another.
4081 prison_racct_modify(struct prison *pr)
4087 struct prison_racct *oldprr;
4089 ASSERT_RACCT_ENABLED();
4091 sx_slock(&allproc_lock);
4092 sx_xlock(&allprison_lock);
4094 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4095 sx_xunlock(&allprison_lock);
4096 sx_sunlock(&allproc_lock);
4100 oldprr = pr->pr_prison_racct;
4101 pr->pr_prison_racct = NULL;
4103 prison_racct_attach(pr);
4106 * Move resource utilisation records.
4108 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4112 * Force rctl to reattach rules to processes.
4114 FOREACH_PROC_IN_SYSTEM(p) {
4116 cred = crhold(p->p_ucred);
4118 rctl_proc_ucred_changed(p, cred);
4123 sx_sunlock(&allproc_lock);
4124 prison_racct_free_locked(oldprr);
4125 sx_xunlock(&allprison_lock);
4129 prison_racct_detach(struct prison *pr)
4132 ASSERT_RACCT_ENABLED();
4133 sx_assert(&allprison_lock, SA_UNLOCKED);
4135 if (pr->pr_prison_racct == NULL)
4137 prison_racct_free(pr->pr_prison_racct);
4138 pr->pr_prison_racct = NULL;
4145 db_show_prison(struct prison *pr)
4147 struct bool_flags *bf;
4148 struct jailsys_flags *jsf;
4149 #if defined(INET) || defined(INET6)
4154 char ip4buf[INET_ADDRSTRLEN];
4157 char ip6buf[INET6_ADDRSTRLEN];
4160 db_printf("prison %p:\n", pr);
4161 db_printf(" jid = %d\n", pr->pr_id);
4162 db_printf(" name = %s\n", pr->pr_name);
4163 db_printf(" parent = %p\n", pr->pr_parent);
4164 db_printf(" ref = %d\n", pr->pr_ref);
4165 db_printf(" uref = %d\n", pr->pr_uref);
4166 db_printf(" path = %s\n", pr->pr_path);
4167 db_printf(" cpuset = %d\n", pr->pr_cpuset
4168 ? pr->pr_cpuset->cs_id : -1);
4170 db_printf(" vnet = %p\n", pr->pr_vnet);
4172 db_printf(" root = %p\n", pr->pr_root);
4173 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4174 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4175 db_printf(" children.max = %d\n", pr->pr_childmax);
4176 db_printf(" children.cur = %d\n", pr->pr_childcount);
4177 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4178 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4179 db_printf(" flags = 0x%x", pr->pr_flags);
4180 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4181 if (pr->pr_flags & bf->flag)
4182 db_printf(" %s", bf->name);
4183 for (jsf = pr_flag_jailsys;
4184 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4186 f = pr->pr_flags & (jsf->disable | jsf->new);
4187 db_printf(" %-16s= %s\n", jsf->name,
4188 (f != 0 && f == jsf->disable) ? "disable"
4189 : (f == jsf->new) ? "new"
4192 db_printf(" allow = 0x%x", pr->pr_allow);
4193 for (bf = pr_flag_allow;
4194 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4196 if (pr->pr_allow & bf->flag)
4197 db_printf(" %s", bf->name);
4199 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4200 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4201 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4202 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4203 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4205 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4206 for (ii = 0; ii < pr->pr_ip4s; ii++)
4207 db_printf(" %s %s\n",
4208 ii == 0 ? "ip4.addr =" : " ",
4209 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4212 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4213 for (ii = 0; ii < pr->pr_ip6s; ii++)
4214 db_printf(" %s %s\n",
4215 ii == 0 ? "ip6.addr =" : " ",
4216 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4220 DB_SHOW_COMMAND(prison, db_show_prison_command)
4226 * Show all prisons in the list, and prison0 which is not
4229 db_show_prison(&prison0);
4230 if (!db_pager_quit) {
4231 TAILQ_FOREACH(pr, &allprison, pr_list) {
4243 /* Look for a prison with the ID and with references. */
4244 TAILQ_FOREACH(pr, &allprison, pr_list)
4245 if (pr->pr_id == addr && pr->pr_ref > 0)
4248 /* Look again, without requiring a reference. */
4249 TAILQ_FOREACH(pr, &allprison, pr_list)
4250 if (pr->pr_id == addr)
4253 /* Assume address points to a valid prison. */
4254 pr = (struct prison *)addr;