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},
198 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
200 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | PR_ALLOW_RESERVED_PORTS)
201 #define JAIL_DEFAULT_ENFORCE_STATFS 2
202 #define JAIL_DEFAULT_DEVFS_RSNUM 0
203 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
204 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
205 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
206 #if defined(INET) || defined(INET6)
207 static unsigned jail_max_af_ips = 255;
211 * Initialize the parts of prison0 that can't be static-initialized with
212 * constants. This is called from proc0_init() after creating thread0 cpuset.
218 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
219 prison0.pr_osreldate = osreldate;
220 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
229 sys_jail(struct thread *td, struct jail_args *uap)
235 error = copyin(uap->jail, &version, sizeof(uint32_t));
244 /* FreeBSD single IPv4 jails. */
245 bzero(&j, sizeof(struct jail));
246 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
249 j.version = j0.version;
251 j.hostname = j0.hostname;
252 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
258 * Version 1 was used by multi-IPv4 jail implementations
259 * that never made it into the official kernel.
263 case 2: /* JAIL_API_VERSION */
264 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
265 error = copyin(uap->jail, &j, sizeof(struct jail));
271 /* Sci-Fi jails are not supported, sorry. */
274 return (kern_jail(td, &j));
278 kern_jail(struct thread *td, struct jail *j)
280 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
289 char *u_path, *u_hostname, *u_name;
290 struct bool_flags *bf;
293 struct in_addr *u_ip4;
296 struct in6_addr *u_ip6;
299 int error, enforce_statfs;
301 bzero(&optiov, sizeof(optiov));
302 opt.uio_iov = optiov;
306 opt.uio_segflg = UIO_SYSSPACE;
307 opt.uio_rw = UIO_READ;
310 /* Set permissions for top-level jails from sysctls. */
311 if (!jailed(td->td_ucred)) {
312 for (bf = pr_flag_allow;
313 bf < pr_flag_allow + nitems(pr_flag_allow) &&
316 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
317 (jail_default_allow & bf->flag)
318 ? bf->name : bf->noname);
319 optiov[opt.uio_iovcnt].iov_len =
320 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
323 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
324 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
326 enforce_statfs = jail_default_enforce_statfs;
327 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
328 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
332 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
334 ip4s = (j->version == 0) ? 1 : j->ip4s;
335 if (ip4s > jail_max_af_ips)
337 tmplen += ip4s * sizeof(struct in_addr);
343 if (j->ip6s > jail_max_af_ips)
345 tmplen += j->ip6s * sizeof(struct in6_addr);
350 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
351 u_hostname = u_path + MAXPATHLEN;
352 u_name = u_hostname + MAXHOSTNAMELEN;
354 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
358 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
360 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
363 optiov[opt.uio_iovcnt].iov_base = "path";
364 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
366 optiov[opt.uio_iovcnt].iov_base = u_path;
367 error = copyinstr(j->path, u_path, MAXPATHLEN,
368 &optiov[opt.uio_iovcnt].iov_len);
370 free(u_path, M_TEMP);
374 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
375 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
377 optiov[opt.uio_iovcnt].iov_base = u_hostname;
378 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
379 &optiov[opt.uio_iovcnt].iov_len);
381 free(u_path, M_TEMP);
385 if (j->jailname != NULL) {
386 optiov[opt.uio_iovcnt].iov_base = "name";
387 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
389 optiov[opt.uio_iovcnt].iov_base = u_name;
390 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
391 &optiov[opt.uio_iovcnt].iov_len);
393 free(u_path, M_TEMP);
399 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
400 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
402 optiov[opt.uio_iovcnt].iov_base = u_ip4;
403 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
405 u_ip4->s_addr = j->ip4s;
407 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
409 free(u_path, M_TEMP);
416 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
417 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
419 optiov[opt.uio_iovcnt].iov_base = u_ip6;
420 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
421 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
423 free(u_path, M_TEMP);
428 KASSERT(opt.uio_iovcnt <= nitems(optiov),
429 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
430 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
431 free(u_path, M_TEMP);
437 * struct jail_set_args {
438 * struct iovec *iovp;
439 * unsigned int iovcnt;
444 sys_jail_set(struct thread *td, struct jail_set_args *uap)
449 /* Check that we have an even number of iovecs. */
453 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
456 error = kern_jail_set(td, auio, uap->flags);
462 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
469 struct in6_addr *ip6;
472 struct vfsoptlist *opts;
473 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
475 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
476 char *g_path, *osrelstr;
477 struct bool_flags *bf;
478 struct jailsys_flags *jsf;
479 #if defined(INET) || defined(INET6)
484 size_t namelen, onamelen, pnamelen;
485 int born, created, cuflags, descend, enforce;
486 int error, errmsg_len, errmsg_pos;
487 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
488 int jid, jsys, len, level;
489 int childmax, osreldt, rsnum, slevel;
490 int fullpath_disabled;
491 #if defined(INET) || defined(INET6)
500 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
504 error = priv_check(td, PRIV_JAIL_SET);
505 if (!error && (flags & JAIL_ATTACH))
506 error = priv_check(td, PRIV_JAIL_ATTACH);
509 mypr = td->td_ucred->cr_prison;
510 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
512 if (flags & ~JAIL_SET_MASK)
516 * Check all the parameters before committing to anything. Not all
517 * errors can be caught early, but we may as well try. Also, this
518 * takes care of some expensive stuff (path lookup) before getting
519 * the allprison lock.
521 * XXX Jails are not filesystems, and jail parameters are not mount
522 * options. But it makes more sense to re-use the vfsopt code
523 * than duplicate it under a different name.
525 error = vfs_buildopts(optuio, &opts);
536 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
539 vfs_opterror(opts, "no valid operation (create or update)");
543 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
549 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
558 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
566 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
571 else if (enforce < 0 || enforce > 2) {
577 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
585 pr_flags = ch_flags = 0;
586 for (bf = pr_flag_bool;
587 bf < pr_flag_bool + nitems(pr_flag_bool);
589 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
590 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
592 ch_flags |= pr_flags;
593 for (jsf = pr_flag_jailsys;
594 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
596 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
602 case JAIL_SYS_DISABLE:
607 pr_flags |= jsf->disable;
610 pr_flags |= jsf->new;
612 case JAIL_SYS_INHERIT:
618 ch_flags |= jsf->new | jsf->disable;
620 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
621 && !(pr_flags & PR_PERSIST)) {
623 vfs_opterror(opts, "new jail must persist or attach");
627 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
629 vfs_opterror(opts, "vnet cannot be changed after creation");
634 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
636 vfs_opterror(opts, "ip4 cannot be changed after creation");
641 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
643 vfs_opterror(opts, "ip6 cannot be changed after creation");
648 pr_allow = ch_allow = 0;
649 for (bf = pr_flag_allow;
650 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
652 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
653 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
655 ch_allow |= pr_allow;
657 error = vfs_getopt(opts, "name", (void **)&name, &len);
663 if (len == 0 || name[len - 1] != '\0') {
667 if (len > MAXHOSTNAMELEN) {
668 error = ENAMETOOLONG;
673 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
681 if (len == 0 || host[len - 1] != '\0') {
685 if (len > MAXHOSTNAMELEN) {
686 error = ENAMETOOLONG;
691 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
699 if (len == 0 || domain[len - 1] != '\0') {
703 if (len > MAXHOSTNAMELEN) {
704 error = ENAMETOOLONG;
709 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
717 if (len == 0 || uuid[len - 1] != '\0') {
721 if (len > HOSTUUIDLEN) {
722 error = ENAMETOOLONG;
727 #ifdef COMPAT_FREEBSD32
728 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
731 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
735 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
747 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
752 else if (ip4s & (sizeof(*ip4) - 1)) {
756 ch_flags |= PR_IP4_USER;
757 pr_flags |= PR_IP4_USER;
759 ip4s /= sizeof(*ip4);
760 if (ip4s > jail_max_af_ips) {
762 vfs_opterror(opts, "too many IPv4 addresses");
765 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
766 bcopy(op, ip4, ip4s * sizeof(*ip4));
768 * IP addresses are all sorted but ip[0] to preserve
769 * the primary IP address as given from userland.
770 * This special IP is used for unbound outgoing
771 * connections as well for "loopback" traffic in case
772 * source address selection cannot find any more fitting
773 * address to connect from.
776 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
779 * Check for duplicate addresses and do some simple
780 * zero and broadcast checks. If users give other bogus
781 * addresses it is their problem.
783 * We do not have to care about byte order for these
784 * checks so we will do them in NBO.
786 for (ii = 0; ii < ip4s; ii++) {
787 if (ip4[ii].s_addr == INADDR_ANY ||
788 ip4[ii].s_addr == INADDR_BROADCAST) {
793 (ip4[0].s_addr == ip4[ii+1].s_addr ||
794 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
804 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
809 else if (ip6s & (sizeof(*ip6) - 1)) {
813 ch_flags |= PR_IP6_USER;
814 pr_flags |= PR_IP6_USER;
816 ip6s /= sizeof(*ip6);
817 if (ip6s > jail_max_af_ips) {
819 vfs_opterror(opts, "too many IPv6 addresses");
822 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
823 bcopy(op, ip6, ip6s * sizeof(*ip6));
825 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
827 for (ii = 0; ii < ip6s; ii++) {
828 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
833 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
834 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
844 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
845 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
848 "vnet jails cannot have IP address restrictions");
853 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
859 if (flags & JAIL_UPDATE) {
862 "osrelease cannot be changed after creation");
865 if (len == 0 || len >= OSRELEASELEN) {
868 "osrelease string must be 1-%d bytes long",
874 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
880 if (flags & JAIL_UPDATE) {
883 "osreldate cannot be changed after creation");
888 vfs_opterror(opts, "osreldate cannot be 0");
893 fullpath_disabled = 0;
895 error = vfs_getopt(opts, "path", (void **)&path, &len);
901 if (flags & JAIL_UPDATE) {
904 "path cannot be changed after creation");
907 if (len == 0 || path[len - 1] != '\0') {
911 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
917 NDFREE(&nd, NDF_ONLY_PNBUF);
918 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
919 strlcpy(g_path, path, MAXPATHLEN);
920 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
923 else if (error == ENODEV) {
924 /* proceed if sysctl debug.disablefullpath == 1 */
925 fullpath_disabled = 1;
926 if (len < 2 || (len == 2 && path[0] == '/'))
929 /* exit on other errors */
932 if (root->v_type != VDIR) {
938 if (fullpath_disabled) {
939 /* Leave room for a real-root full pathname. */
940 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
941 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
942 error = ENAMETOOLONG;
950 * Find the specified jail, or at least its parent.
951 * This abuses the file error codes ENOENT and EEXIST.
955 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
956 namelc = strrchr(name, '.');
957 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
961 sx_xlock(&allprison_lock);
964 * See if a requested jid already exists. There is an
965 * information leak here if the jid exists but is not within
966 * the caller's jail hierarchy. Jail creators will get EEXIST
967 * even though they cannot see the jail, and CREATE | UPDATE
968 * will return ENOENT which is not normally a valid error.
972 vfs_opterror(opts, "negative jid");
973 goto done_unlock_list;
975 pr = prison_find(jid);
978 /* Create: jid must not exist. */
979 if (cuflags == JAIL_CREATE) {
980 mtx_unlock(&pr->pr_mtx);
982 vfs_opterror(opts, "jail %d already exists",
984 goto done_unlock_list;
986 if (!prison_ischild(mypr, pr)) {
987 mtx_unlock(&pr->pr_mtx);
989 } else if (pr->pr_uref == 0) {
990 if (!(flags & JAIL_DYING)) {
991 mtx_unlock(&pr->pr_mtx);
993 vfs_opterror(opts, "jail %d is dying",
995 goto done_unlock_list;
996 } else if ((flags & JAIL_ATTACH) ||
997 (pr_flags & PR_PERSIST)) {
999 * A dying jail might be resurrected
1000 * (via attach or persist), but first
1001 * it must determine if another jail
1002 * has claimed its name. Accomplish
1003 * this by implicitly re-setting the
1007 name = prison_name(mypr, pr);
1012 /* Update: jid must exist. */
1013 if (cuflags == JAIL_UPDATE) {
1015 vfs_opterror(opts, "jail %d not found", jid);
1016 goto done_unlock_list;
1021 * If the caller provided a name, look for a jail by that name.
1022 * This has different semantics for creates and updates keyed by jid
1023 * (where the name must not already exist in a different jail),
1024 * and updates keyed by the name itself (where the name must exist
1025 * because that is the jail being updated).
1029 namelc = strrchr(name, '.');
1034 * This is a hierarchical name. Split it into the
1035 * parent and child names, and make sure the parent
1036 * exists or matches an already found jail.
1039 if (strncmp(name, ppr->pr_name, namelc - name)
1040 || ppr->pr_name[namelc - name] != '\0') {
1041 mtx_unlock(&pr->pr_mtx);
1044 "cannot change jail's parent");
1045 goto done_unlock_list;
1049 ppr = prison_find_name(mypr, name);
1053 "jail \"%s\" not found", name);
1054 goto done_unlock_list;
1056 mtx_unlock(&ppr->pr_mtx);
1061 if (namelc[0] != '\0') {
1063 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1066 FOREACH_PRISON_CHILD(ppr, tpr) {
1067 if (tpr != pr && tpr->pr_ref > 0 &&
1068 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1070 cuflags != JAIL_CREATE) {
1071 mtx_lock(&tpr->pr_mtx);
1072 if (tpr->pr_ref > 0) {
1077 if (tpr->pr_uref > 0) {
1083 mtx_unlock(&tpr->pr_mtx);
1084 } else if (tpr->pr_uref > 0) {
1086 * Create, or update(jid):
1087 * name must not exist in an
1088 * active sibling jail.
1092 mtx_unlock(&pr->pr_mtx);
1094 "jail \"%s\" already exists",
1096 goto done_unlock_list;
1100 /* If no active jail is found, use a dying one. */
1101 if (deadpr != NULL && pr == NULL) {
1102 if (flags & JAIL_DYING) {
1103 mtx_lock(&deadpr->pr_mtx);
1104 if (deadpr->pr_ref == 0) {
1105 mtx_unlock(&deadpr->pr_mtx);
1109 } else if (cuflags == JAIL_UPDATE) {
1112 "jail \"%s\" is dying", name);
1113 goto done_unlock_list;
1116 /* Update: name must exist if no jid. */
1117 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1119 vfs_opterror(opts, "jail \"%s\" not found",
1121 goto done_unlock_list;
1125 /* Update: must provide a jid or name. */
1126 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1128 vfs_opterror(opts, "update specified no jail");
1129 goto done_unlock_list;
1132 /* If there's no prison to update, create a new one and link it in. */
1134 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1135 if (tpr->pr_childcount >= tpr->pr_childmax) {
1137 vfs_opterror(opts, "prison limit exceeded");
1138 goto done_unlock_list;
1141 mtx_lock(&ppr->pr_mtx);
1142 if (ppr->pr_ref == 0) {
1143 mtx_unlock(&ppr->pr_mtx);
1145 vfs_opterror(opts, "jail \"%s\" not found",
1146 prison_name(mypr, ppr));
1147 goto done_unlock_list;
1151 mtx_unlock(&ppr->pr_mtx);
1152 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1154 /* Find the next free jid. */
1157 if (jid == JAIL_MAX)
1159 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1160 if (tpr->pr_id < jid)
1162 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1163 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1166 if (jid == lastprid) {
1169 "no available jail IDs");
1171 prison_deref(ppr, PD_DEREF |
1172 PD_DEUREF | PD_LIST_XLOCKED);
1181 * The jail already has a jid (that did not yet exist),
1182 * so just find where to insert it.
1184 TAILQ_FOREACH(tpr, &allprison, pr_list)
1185 if (tpr->pr_id >= jid) {
1186 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1191 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1192 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1193 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1194 tpr->pr_childcount++;
1196 pr->pr_parent = ppr;
1199 /* Set some default values, and inherit some from the parent. */
1204 root = mypr->pr_root;
1207 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1208 pr->pr_flags |= PR_HOST;
1209 #if defined(INET) || defined(INET6)
1211 if (!(pr_flags & PR_VNET))
1215 if (!(ch_flags & PR_IP4_USER))
1216 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1217 else if (!(pr_flags & PR_IP4_USER)) {
1218 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1219 if (ppr->pr_ip4 != NULL) {
1220 pr->pr_ip4s = ppr->pr_ip4s;
1221 pr->pr_ip4 = malloc(pr->pr_ip4s *
1222 sizeof(struct in_addr), M_PRISON,
1224 bcopy(ppr->pr_ip4, pr->pr_ip4,
1225 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1230 if (!(ch_flags & PR_IP6_USER))
1231 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1232 else if (!(pr_flags & PR_IP6_USER)) {
1233 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1234 if (ppr->pr_ip6 != NULL) {
1235 pr->pr_ip6s = ppr->pr_ip6s;
1236 pr->pr_ip6 = malloc(pr->pr_ip6s *
1237 sizeof(struct in6_addr), M_PRISON,
1239 bcopy(ppr->pr_ip6, pr->pr_ip6,
1240 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1246 /* Source address selection is always on by default. */
1247 pr->pr_flags |= _PR_IP_SADDRSEL;
1249 pr->pr_securelevel = ppr->pr_securelevel;
1250 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1251 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1252 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1254 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1255 if (osrelstr == NULL)
1256 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1258 strcpy(pr->pr_osrelease, osrelstr);
1260 LIST_INIT(&pr->pr_children);
1261 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1262 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1265 /* Allocate a new vnet if specified. */
1266 pr->pr_vnet = (pr_flags & PR_VNET)
1267 ? vnet_alloc() : ppr->pr_vnet;
1270 * Allocate a dedicated cpuset for each jail.
1271 * Unlike other initial settings, this may return an erorr.
1273 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1275 prison_deref(pr, PD_LIST_XLOCKED);
1279 mtx_lock(&pr->pr_mtx);
1281 * New prisons do not yet have a reference, because we do not
1282 * want others to see the incomplete prison once the
1283 * allprison_lock is downgraded.
1288 * Grab a reference for existing prisons, to ensure they
1289 * continue to exist for the duration of the call.
1292 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1293 if ((pr->pr_flags & PR_VNET) &&
1294 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1297 "vnet jails cannot have IP address restrictions");
1298 goto done_deref_locked;
1302 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1305 "ip4 cannot be changed after creation");
1306 goto done_deref_locked;
1310 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1313 "ip6 cannot be changed after creation");
1314 goto done_deref_locked;
1319 /* Do final error checking before setting anything. */
1321 if (slevel < ppr->pr_securelevel) {
1323 goto done_deref_locked;
1327 if (childmax >= ppr->pr_childmax) {
1329 goto done_deref_locked;
1333 if (enforce < ppr->pr_enforce_statfs) {
1335 goto done_deref_locked;
1340 * devfs_rsnum is a uint16_t
1342 if (rsnum < 0 || rsnum > 65535) {
1344 goto done_deref_locked;
1347 * Nested jails always inherit parent's devfs ruleset
1349 if (jailed(td->td_ucred)) {
1350 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1352 goto done_deref_locked;
1354 rsnum = ppr->pr_devfs_rsnum;
1359 if (ppr->pr_flags & PR_IP4) {
1361 * Make sure the new set of IP addresses is a
1362 * subset of the parent's list. Don't worry
1363 * about the parent being unlocked, as any
1364 * setting is done with allprison_lock held.
1366 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1367 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1369 if (ij == ppr->pr_ip4s) {
1371 goto done_deref_locked;
1374 for (ii = ij = 1; ii < ip4s; ii++) {
1375 if (ip4[ii].s_addr ==
1376 ppr->pr_ip4[0].s_addr)
1378 for (; ij < ppr->pr_ip4s; ij++)
1379 if (ip4[ii].s_addr ==
1380 ppr->pr_ip4[ij].s_addr)
1382 if (ij == ppr->pr_ip4s)
1385 if (ij == ppr->pr_ip4s) {
1387 goto done_deref_locked;
1392 * Check for conflicting IP addresses. We permit them
1393 * if there is no more than one IP on each jail. If
1394 * there is a duplicate on a jail with more than one
1395 * IP stop checking and return error.
1398 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1399 if (tppr->pr_flags & PR_VNET)
1404 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1407 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1409 tpr->pr_uref == 0) {
1413 if (!(tpr->pr_flags & PR_IP4_USER))
1416 if (tpr->pr_ip4 == NULL ||
1417 (ip4s == 1 && tpr->pr_ip4s == 1))
1419 for (ii = 0; ii < ip4s; ii++) {
1420 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1424 "IPv4 addresses clash");
1425 goto done_deref_locked;
1433 if (ppr->pr_flags & PR_IP6) {
1435 * Make sure the new set of IP addresses is a
1436 * subset of the parent's list.
1438 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1439 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1442 if (ij == ppr->pr_ip6s) {
1444 goto done_deref_locked;
1447 for (ii = ij = 1; ii < ip6s; ii++) {
1448 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1451 for (; ij < ppr->pr_ip6s; ij++)
1452 if (IN6_ARE_ADDR_EQUAL(
1453 &ip6[ii], &ppr->pr_ip6[ij]))
1455 if (ij == ppr->pr_ip6s)
1458 if (ij == ppr->pr_ip6s) {
1460 goto done_deref_locked;
1464 /* Check for conflicting IP addresses. */
1466 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1467 if (tppr->pr_flags & PR_VNET)
1472 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1475 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1477 tpr->pr_uref == 0) {
1481 if (!(tpr->pr_flags & PR_IP6_USER))
1484 if (tpr->pr_ip6 == NULL ||
1485 (ip6s == 1 && tpr->pr_ip6s == 1))
1487 for (ii = 0; ii < ip6s; ii++) {
1488 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1492 "IPv6 addresses clash");
1493 goto done_deref_locked;
1499 onamelen = namelen = 0;
1500 if (namelc != NULL) {
1501 /* Give a default name of the jid. Also allow the name to be
1502 * explicitly the jid - but not any other number, and only in
1503 * normal form (no leading zero/etc).
1505 if (namelc[0] == '\0')
1506 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1507 else if ((strtoul(namelc, &p, 10) != jid ||
1508 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1511 "name cannot be numeric (unless it is the jid)");
1512 goto done_deref_locked;
1515 * Make sure the name isn't too long for the prison or its
1518 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1519 onamelen = strlen(pr->pr_name + pnamelen);
1520 namelen = strlen(namelc);
1521 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1522 error = ENAMETOOLONG;
1523 goto done_deref_locked;
1525 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1526 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1527 sizeof(pr->pr_name)) {
1528 error = ENAMETOOLONG;
1529 goto done_deref_locked;
1533 if (pr_allow & ~ppr->pr_allow) {
1535 goto done_deref_locked;
1539 * Let modules check their parameters. This requires unlocking and
1540 * then re-locking the prison, but this is still a valid state as long
1541 * as allprison_lock remains xlocked.
1543 mtx_unlock(&pr->pr_mtx);
1544 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1546 prison_deref(pr, created
1548 : PD_DEREF | PD_LIST_XLOCKED);
1551 mtx_lock(&pr->pr_mtx);
1553 /* At this point, all valid parameters should have been noted. */
1554 TAILQ_FOREACH(opt, opts, link) {
1555 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1557 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1558 goto done_deref_locked;
1562 /* Set the parameters of the prison. */
1565 if (pr_flags & PR_IP4_USER) {
1566 pr->pr_flags |= PR_IP4;
1567 free(pr->pr_ip4, M_PRISON);
1571 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1573 if (tpr->pr_flags & PR_VNET) {
1578 if (prison_restrict_ip4(tpr, NULL)) {
1587 if (pr_flags & PR_IP6_USER) {
1588 pr->pr_flags |= PR_IP6;
1589 free(pr->pr_ip6, M_PRISON);
1593 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1595 if (tpr->pr_flags & PR_VNET) {
1600 if (prison_restrict_ip6(tpr, NULL)) {
1608 pr->pr_securelevel = slevel;
1609 /* Set all child jails to be at least this level. */
1610 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1611 if (tpr->pr_securelevel < slevel)
1612 tpr->pr_securelevel = slevel;
1615 pr->pr_childmax = childmax;
1616 /* Set all child jails to under this limit. */
1617 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1618 if (tpr->pr_childmax > childmax - level)
1619 tpr->pr_childmax = childmax > level
1620 ? childmax - level : 0;
1623 pr->pr_enforce_statfs = enforce;
1624 /* Pass this restriction on to the children. */
1625 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1626 if (tpr->pr_enforce_statfs < enforce)
1627 tpr->pr_enforce_statfs = enforce;
1630 pr->pr_devfs_rsnum = rsnum;
1631 /* Pass this restriction on to the children. */
1632 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1633 tpr->pr_devfs_rsnum = rsnum;
1635 if (namelc != NULL) {
1636 if (ppr == &prison0)
1637 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1639 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1640 ppr->pr_name, namelc);
1641 /* Change this component of child names. */
1642 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1643 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1644 strlen(tpr->pr_name + onamelen) + 1);
1645 bcopy(pr->pr_name, tpr->pr_name, namelen);
1649 /* Try to keep a real-rooted full pathname. */
1650 if (fullpath_disabled && path[0] == '/' &&
1651 strcmp(mypr->pr_path, "/"))
1652 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1653 mypr->pr_path, path);
1655 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1658 if (PR_HOST & ch_flags & ~pr_flags) {
1659 if (pr->pr_flags & PR_HOST) {
1661 * Copy the parent's host info. As with pr_ip4 above,
1662 * the lack of a lock on the parent is not a problem;
1663 * it is always set with allprison_lock at least
1664 * shared, and is held exclusively here.
1666 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1667 sizeof(pr->pr_hostname));
1668 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1669 sizeof(pr->pr_domainname));
1670 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1671 sizeof(pr->pr_hostuuid));
1672 pr->pr_hostid = pr->pr_parent->pr_hostid;
1674 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1675 /* Set this prison, and any descendants without PR_HOST. */
1677 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1679 strlcpy(pr->pr_domainname, domain,
1680 sizeof(pr->pr_domainname));
1682 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1684 pr->pr_hostid = hid;
1685 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1686 if (tpr->pr_flags & PR_HOST)
1690 strlcpy(tpr->pr_hostname,
1692 sizeof(tpr->pr_hostname));
1694 strlcpy(tpr->pr_domainname,
1696 sizeof(tpr->pr_domainname));
1698 strlcpy(tpr->pr_hostuuid,
1700 sizeof(tpr->pr_hostuuid));
1702 tpr->pr_hostid = hid;
1706 if ((tallow = ch_allow & ~pr_allow)) {
1707 /* Clear allow bits in all children. */
1708 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1709 tpr->pr_allow &= ~tallow;
1711 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1713 * Persistent prisons get an extra reference, and prisons losing their
1714 * persist flag lose that reference. Only do this for existing prisons
1715 * for now, so new ones will remain unseen until after the module
1716 * handlers have completed.
1718 born = pr->pr_uref == 0;
1719 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1720 if (pr_flags & PR_PERSIST) {
1728 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1729 mtx_unlock(&pr->pr_mtx);
1732 if (racct_enable && created)
1733 prison_racct_attach(pr);
1736 /* Locks may have prevented a complete restriction of child IP
1737 * addresses. If so, allocate some more memory and try again.
1742 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1743 mtx_lock(&pr->pr_mtx);
1745 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1747 if (tpr->pr_flags & PR_VNET) {
1752 if (prison_restrict_ip4(tpr, ip4)) {
1759 mtx_unlock(&pr->pr_mtx);
1765 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1766 mtx_lock(&pr->pr_mtx);
1768 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1770 if (tpr->pr_flags & PR_VNET) {
1775 if (prison_restrict_ip6(tpr, ip6)) {
1782 mtx_unlock(&pr->pr_mtx);
1786 /* Let the modules do their work. */
1787 sx_downgrade(&allprison_lock);
1789 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1791 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1792 prison_deref(pr, created
1794 : PD_DEREF | PD_LIST_SLOCKED);
1798 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1801 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1802 prison_deref(pr, created
1804 : PD_DEREF | PD_LIST_SLOCKED);
1808 /* Attach this process to the prison if requested. */
1809 if (flags & JAIL_ATTACH) {
1810 mtx_lock(&pr->pr_mtx);
1811 error = do_jail_attach(td, pr);
1813 vfs_opterror(opts, "attach failed");
1815 prison_deref(pr, PD_DEREF);
1821 if (racct_enable && !created) {
1822 if (!(flags & JAIL_ATTACH))
1823 sx_sunlock(&allprison_lock);
1824 prison_racct_modify(pr);
1825 if (!(flags & JAIL_ATTACH))
1826 sx_slock(&allprison_lock);
1830 td->td_retval[0] = pr->pr_id;
1833 * Now that it is all there, drop the temporary reference from existing
1834 * prisons. Or add a reference to newly created persistent prisons
1835 * (which was not done earlier so that the prison would not be publicly
1839 prison_deref(pr, (flags & JAIL_ATTACH)
1841 : PD_DEREF | PD_LIST_SLOCKED);
1843 if (pr_flags & PR_PERSIST) {
1844 mtx_lock(&pr->pr_mtx);
1847 mtx_unlock(&pr->pr_mtx);
1849 if (!(flags & JAIL_ATTACH))
1850 sx_sunlock(&allprison_lock);
1856 prison_deref(pr, created
1857 ? PD_LOCKED | PD_LIST_XLOCKED
1858 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1861 sx_xunlock(&allprison_lock);
1867 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1868 &errmsg_len) == 0 && errmsg_len > 0) {
1869 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1870 if (optuio->uio_segflg == UIO_SYSSPACE)
1872 optuio->uio_iov[errmsg_pos].iov_base,
1876 optuio->uio_iov[errmsg_pos].iov_base,
1882 free(ip4, M_PRISON);
1885 free(ip6, M_PRISON);
1888 free(g_path, M_TEMP);
1895 * struct jail_get_args {
1896 * struct iovec *iovp;
1897 * unsigned int iovcnt;
1902 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1907 /* Check that we have an even number of iovecs. */
1908 if (uap->iovcnt & 1)
1911 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1914 error = kern_jail_get(td, auio, uap->flags);
1916 error = copyout(auio->uio_iov, uap->iovp,
1917 uap->iovcnt * sizeof (struct iovec));
1923 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1925 struct bool_flags *bf;
1926 struct jailsys_flags *jsf;
1927 struct prison *pr, *mypr;
1929 struct vfsoptlist *opts;
1930 char *errmsg, *name;
1931 int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1934 if (flags & ~JAIL_GET_MASK)
1937 /* Get the parameter list. */
1938 error = vfs_buildopts(optuio, &opts);
1941 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1942 mypr = td->td_ucred->cr_prison;
1945 * Find the prison specified by one of: lastjid, jid, name.
1947 sx_slock(&allprison_lock);
1948 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1950 TAILQ_FOREACH(pr, &allprison, pr_list) {
1951 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1952 mtx_lock(&pr->pr_mtx);
1953 if (pr->pr_ref > 0 &&
1954 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1956 mtx_unlock(&pr->pr_mtx);
1962 vfs_opterror(opts, "no jail after %d", jid);
1963 goto done_unlock_list;
1964 } else if (error != ENOENT)
1965 goto done_unlock_list;
1967 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1970 pr = prison_find_child(mypr, jid);
1972 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1973 mtx_unlock(&pr->pr_mtx);
1975 vfs_opterror(opts, "jail %d is dying",
1977 goto done_unlock_list;
1982 vfs_opterror(opts, "jail %d not found", jid);
1983 goto done_unlock_list;
1985 } else if (error != ENOENT)
1986 goto done_unlock_list;
1988 error = vfs_getopt(opts, "name", (void **)&name, &len);
1990 if (len == 0 || name[len - 1] != '\0') {
1992 goto done_unlock_list;
1994 pr = prison_find_name(mypr, name);
1996 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1997 mtx_unlock(&pr->pr_mtx);
1999 vfs_opterror(opts, "jail \"%s\" is dying",
2001 goto done_unlock_list;
2006 vfs_opterror(opts, "jail \"%s\" not found", name);
2007 goto done_unlock_list;
2008 } else if (error != ENOENT)
2009 goto done_unlock_list;
2011 vfs_opterror(opts, "no jail specified");
2013 goto done_unlock_list;
2016 /* Get the parameters of the prison. */
2019 td->td_retval[0] = pr->pr_id;
2020 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2021 if (error != 0 && error != ENOENT)
2023 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2024 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2025 if (error != 0 && error != ENOENT)
2027 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2028 if (error != 0 && error != ENOENT)
2030 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2031 sizeof(pr->pr_cpuset->cs_id));
2032 if (error != 0 && error != ENOENT)
2034 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2035 if (error != 0 && error != ENOENT)
2038 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2039 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2040 if (error != 0 && error != ENOENT)
2044 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2045 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2046 if (error != 0 && error != ENOENT)
2049 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2050 sizeof(pr->pr_securelevel));
2051 if (error != 0 && error != ENOENT)
2053 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2054 sizeof(pr->pr_childcount));
2055 if (error != 0 && error != ENOENT)
2057 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2058 sizeof(pr->pr_childmax));
2059 if (error != 0 && error != ENOENT)
2061 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2062 if (error != 0 && error != ENOENT)
2064 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2065 if (error != 0 && error != ENOENT)
2067 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2068 if (error != 0 && error != ENOENT)
2070 #ifdef COMPAT_FREEBSD32
2071 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2072 uint32_t hid32 = pr->pr_hostid;
2074 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2077 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2078 sizeof(pr->pr_hostid));
2079 if (error != 0 && error != ENOENT)
2081 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2082 sizeof(pr->pr_enforce_statfs));
2083 if (error != 0 && error != ENOENT)
2085 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2086 sizeof(pr->pr_devfs_rsnum));
2087 if (error != 0 && error != ENOENT)
2089 for (bf = pr_flag_bool;
2090 bf < pr_flag_bool + nitems(pr_flag_bool);
2092 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2093 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2094 if (error != 0 && error != ENOENT)
2097 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2098 if (error != 0 && error != ENOENT)
2101 for (jsf = pr_flag_jailsys;
2102 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2104 f = pr->pr_flags & (jsf->disable | jsf->new);
2105 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2106 : (f == jsf->new) ? JAIL_SYS_NEW
2108 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2109 if (error != 0 && error != ENOENT)
2112 for (bf = pr_flag_allow;
2113 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2115 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2116 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2117 if (error != 0 && error != ENOENT)
2120 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2121 if (error != 0 && error != ENOENT)
2124 i = (pr->pr_uref == 0);
2125 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2126 if (error != 0 && error != ENOENT)
2129 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2130 if (error != 0 && error != ENOENT)
2132 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2133 sizeof(pr->pr_osreldate));
2134 if (error != 0 && error != ENOENT)
2136 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2137 if (error != 0 && error != ENOENT)
2140 /* Get the module parameters. */
2141 mtx_unlock(&pr->pr_mtx);
2143 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2146 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2148 /* By now, all parameters should have been noted. */
2149 TAILQ_FOREACH(opt, opts, link) {
2150 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2152 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2157 /* Write the fetched parameters back to userspace. */
2159 TAILQ_FOREACH(opt, opts, link) {
2160 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2161 pos = 2 * opt->pos + 1;
2162 optuio->uio_iov[pos].iov_len = opt->len;
2163 if (opt->value != NULL) {
2164 if (optuio->uio_segflg == UIO_SYSSPACE) {
2166 optuio->uio_iov[pos].iov_base,
2169 error = copyout(opt->value,
2170 optuio->uio_iov[pos].iov_base,
2181 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2185 sx_sunlock(&allprison_lock);
2187 if (error && errmsg_pos >= 0) {
2188 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2189 errmsg_pos = 2 * errmsg_pos + 1;
2190 if (errmsg_len > 0) {
2191 if (optuio->uio_segflg == UIO_SYSSPACE)
2193 optuio->uio_iov[errmsg_pos].iov_base,
2197 optuio->uio_iov[errmsg_pos].iov_base,
2207 * struct jail_remove_args {
2212 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2214 struct prison *pr, *cpr, *lpr, *tpr;
2217 error = priv_check(td, PRIV_JAIL_REMOVE);
2221 sx_xlock(&allprison_lock);
2222 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2224 sx_xunlock(&allprison_lock);
2228 /* Remove all descendants of this prison, then remove this prison. */
2230 if (!LIST_EMPTY(&pr->pr_children)) {
2231 mtx_unlock(&pr->pr_mtx);
2233 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2234 mtx_lock(&cpr->pr_mtx);
2235 if (cpr->pr_ref > 0) {
2239 /* Already removed - do not do it again. */
2242 mtx_unlock(&cpr->pr_mtx);
2244 mtx_lock(&lpr->pr_mtx);
2245 prison_remove_one(lpr);
2246 sx_xlock(&allprison_lock);
2251 mtx_lock(&lpr->pr_mtx);
2252 prison_remove_one(lpr);
2253 sx_xlock(&allprison_lock);
2255 mtx_lock(&pr->pr_mtx);
2257 prison_remove_one(pr);
2262 prison_remove_one(struct prison *pr)
2267 /* If the prison was persistent, it is not anymore. */
2269 if (pr->pr_flags & PR_PERSIST) {
2272 pr->pr_flags &= ~PR_PERSIST;
2276 * jail_remove added a reference. If that's the only one, remove
2279 KASSERT(pr->pr_ref > 0,
2280 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2281 if (pr->pr_ref == 1) {
2283 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2287 mtx_unlock(&pr->pr_mtx);
2288 sx_xunlock(&allprison_lock);
2290 * Kill all processes unfortunate enough to be attached to this prison.
2292 sx_slock(&allproc_lock);
2293 FOREACH_PROC_IN_SYSTEM(p) {
2295 if (p->p_state != PRS_NEW && p->p_ucred &&
2296 p->p_ucred->cr_prison == pr)
2297 kern_psignal(p, SIGKILL);
2300 sx_sunlock(&allproc_lock);
2301 /* Remove the temporary reference added by jail_remove. */
2302 prison_deref(pr, deuref | PD_DEREF);
2307 * struct jail_attach_args {
2312 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2317 error = priv_check(td, PRIV_JAIL_ATTACH);
2322 * Start with exclusive hold on allprison_lock to ensure that a possible
2323 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2324 * But then immediately downgrade it since we don't need to stop
2327 sx_xlock(&allprison_lock);
2328 sx_downgrade(&allprison_lock);
2329 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2331 sx_sunlock(&allprison_lock);
2336 * Do not allow a process to attach to a prison that is not
2337 * considered to be "alive".
2339 if (pr->pr_uref == 0) {
2340 mtx_unlock(&pr->pr_mtx);
2341 sx_sunlock(&allprison_lock);
2345 return (do_jail_attach(td, pr));
2349 do_jail_attach(struct thread *td, struct prison *pr)
2352 struct ucred *newcred, *oldcred;
2356 * XXX: Note that there is a slight race here if two threads
2357 * in the same privileged process attempt to attach to two
2358 * different jails at the same time. It is important for
2359 * user processes not to do this, or they might end up with
2360 * a process root from one prison, but attached to the jail
2365 mtx_unlock(&pr->pr_mtx);
2367 /* Let modules do whatever they need to prepare for attaching. */
2368 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2370 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2373 sx_sunlock(&allprison_lock);
2376 * Reparent the newly attached process to this jail.
2379 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2383 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2384 if ((error = change_dir(pr->pr_root, td)) != 0)
2387 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2390 VOP_UNLOCK(pr->pr_root, 0);
2391 if ((error = pwd_chroot(td, pr->pr_root)))
2396 oldcred = crcopysafe(p, newcred);
2397 newcred->cr_prison = pr;
2398 proc_set_cred(p, newcred);
2401 racct_proc_ucred_changed(p, oldcred, newcred);
2406 rctl_proc_ucred_changed(p, newcred);
2409 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2414 VOP_UNLOCK(pr->pr_root, 0);
2416 /* Tell modules this thread is still in its old jail after all. */
2417 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2418 prison_deref(pr, PD_DEREF | PD_DEUREF);
2424 * Returns a locked prison instance, or NULL on failure.
2427 prison_find(int prid)
2431 sx_assert(&allprison_lock, SX_LOCKED);
2432 TAILQ_FOREACH(pr, &allprison, pr_list) {
2433 if (pr->pr_id == prid) {
2434 mtx_lock(&pr->pr_mtx);
2437 mtx_unlock(&pr->pr_mtx);
2444 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2447 prison_find_child(struct prison *mypr, int prid)
2452 sx_assert(&allprison_lock, SX_LOCKED);
2453 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2454 if (pr->pr_id == prid) {
2455 mtx_lock(&pr->pr_mtx);
2458 mtx_unlock(&pr->pr_mtx);
2465 * Look for the name relative to mypr. Returns a locked prison or NULL.
2468 prison_find_name(struct prison *mypr, const char *name)
2470 struct prison *pr, *deadpr;
2474 sx_assert(&allprison_lock, SX_LOCKED);
2475 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2478 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2479 if (!strcmp(pr->pr_name + mylen, name)) {
2480 mtx_lock(&pr->pr_mtx);
2481 if (pr->pr_ref > 0) {
2482 if (pr->pr_uref > 0)
2486 mtx_unlock(&pr->pr_mtx);
2489 /* There was no valid prison - perhaps there was a dying one. */
2490 if (deadpr != NULL) {
2491 mtx_lock(&deadpr->pr_mtx);
2492 if (deadpr->pr_ref == 0) {
2493 mtx_unlock(&deadpr->pr_mtx);
2501 * See if a prison has the specific flag set.
2504 prison_flag(struct ucred *cred, unsigned flag)
2507 /* This is an atomic read, so no locking is necessary. */
2508 return (cred->cr_prison->pr_flags & flag);
2512 prison_allow(struct ucred *cred, unsigned flag)
2515 /* This is an atomic read, so no locking is necessary. */
2516 return (cred->cr_prison->pr_allow & flag);
2520 * Remove a prison reference. If that was the last reference, remove the
2521 * prison itself - but not in this context in case there are locks held.
2524 prison_free_locked(struct prison *pr)
2528 mtx_assert(&pr->pr_mtx, MA_OWNED);
2530 mtx_unlock(&pr->pr_mtx);
2532 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2536 prison_free(struct prison *pr)
2539 mtx_lock(&pr->pr_mtx);
2540 prison_free_locked(pr);
2544 * Complete a call to either prison_free or prison_proc_free.
2547 prison_complete(void *context, int pending)
2549 struct prison *pr = context;
2551 sx_xlock(&allprison_lock);
2552 mtx_lock(&pr->pr_mtx);
2553 prison_deref(pr, pr->pr_uref
2554 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2555 : PD_LOCKED | PD_LIST_XLOCKED);
2559 * Remove a prison reference (usually). This internal version assumes no
2560 * mutexes are held, except perhaps the prison itself. If there are no more
2561 * references, release and delist the prison. On completion, the prison lock
2562 * and the allprison lock are both unlocked.
2565 prison_deref(struct prison *pr, int flags)
2567 struct prison *ppr, *tpr;
2570 if (!(flags & PD_LOCKED))
2571 mtx_lock(&pr->pr_mtx);
2573 if (flags & PD_DEUREF) {
2574 KASSERT(pr->pr_uref > 0,
2575 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2578 lasturef = pr->pr_uref == 0;
2581 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2584 if (flags & PD_DEREF) {
2585 KASSERT(pr->pr_ref > 0,
2586 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2591 mtx_unlock(&pr->pr_mtx);
2594 * Tell the modules if the last user reference was removed
2595 * (even it sticks around in dying state).
2598 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2599 sx_xlock(&allprison_lock);
2600 flags |= PD_LIST_XLOCKED;
2602 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2603 mtx_lock(&pr->pr_mtx);
2605 mtx_unlock(&pr->pr_mtx);
2608 /* If the prison still has references, nothing else to do. */
2610 if (flags & PD_LIST_SLOCKED)
2611 sx_sunlock(&allprison_lock);
2612 else if (flags & PD_LIST_XLOCKED)
2613 sx_xunlock(&allprison_lock);
2617 if (flags & PD_LIST_SLOCKED) {
2618 if (!sx_try_upgrade(&allprison_lock)) {
2619 sx_sunlock(&allprison_lock);
2620 sx_xlock(&allprison_lock);
2622 } else if (!(flags & PD_LIST_XLOCKED))
2623 sx_xlock(&allprison_lock);
2625 TAILQ_REMOVE(&allprison, pr, pr_list);
2626 LIST_REMOVE(pr, pr_sibling);
2627 ppr = pr->pr_parent;
2628 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2629 tpr->pr_childcount--;
2630 sx_xunlock(&allprison_lock);
2633 if (pr->pr_vnet != ppr->pr_vnet)
2634 vnet_destroy(pr->pr_vnet);
2636 if (pr->pr_root != NULL)
2638 mtx_destroy(&pr->pr_mtx);
2640 free(pr->pr_ip4, M_PRISON);
2643 free(pr->pr_ip6, M_PRISON);
2645 if (pr->pr_cpuset != NULL)
2646 cpuset_rel(pr->pr_cpuset);
2650 prison_racct_detach(pr);
2654 /* Removing a prison frees a reference on its parent. */
2656 mtx_lock(&pr->pr_mtx);
2657 flags = PD_DEREF | PD_DEUREF;
2662 prison_hold_locked(struct prison *pr)
2665 mtx_assert(&pr->pr_mtx, MA_OWNED);
2666 KASSERT(pr->pr_ref > 0,
2667 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2672 prison_hold(struct prison *pr)
2675 mtx_lock(&pr->pr_mtx);
2676 prison_hold_locked(pr);
2677 mtx_unlock(&pr->pr_mtx);
2681 prison_proc_hold(struct prison *pr)
2684 mtx_lock(&pr->pr_mtx);
2685 KASSERT(pr->pr_uref > 0,
2686 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2688 mtx_unlock(&pr->pr_mtx);
2692 prison_proc_free(struct prison *pr)
2695 mtx_lock(&pr->pr_mtx);
2696 KASSERT(pr->pr_uref > 0,
2697 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2698 if (pr->pr_uref > 1)
2702 * Don't remove the last user reference in this context, which
2703 * is expected to be a process that is not only locked, but
2707 mtx_unlock(&pr->pr_mtx);
2708 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2711 mtx_unlock(&pr->pr_mtx);
2715 * Check if a jail supports the given address family.
2717 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2721 prison_check_af(struct ucred *cred, int af)
2726 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2728 pr = cred->cr_prison;
2730 /* Prisons with their own network stack are not limited. */
2731 if (prison_owns_vnet(cred))
2740 if (pr->pr_flags & PR_IP4)
2742 mtx_lock(&pr->pr_mtx);
2743 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2744 error = EAFNOSUPPORT;
2745 mtx_unlock(&pr->pr_mtx);
2751 if (pr->pr_flags & PR_IP6)
2753 mtx_lock(&pr->pr_mtx);
2754 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2755 error = EAFNOSUPPORT;
2756 mtx_unlock(&pr->pr_mtx);
2764 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2765 error = EAFNOSUPPORT;
2771 * Check if given address belongs to the jail referenced by cred (wrapper to
2772 * prison_check_ip[46]).
2774 * Returns 0 if jail doesn't restrict the address family or if address belongs
2775 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2776 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
2779 prison_if(struct ucred *cred, struct sockaddr *sa)
2782 struct sockaddr_in *sai;
2785 struct sockaddr_in6 *sai6;
2789 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2790 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2793 if (prison_owns_vnet(cred))
2798 switch (sa->sa_family)
2802 sai = (struct sockaddr_in *)sa;
2803 error = prison_check_ip4(cred, &sai->sin_addr);
2808 sai6 = (struct sockaddr_in6 *)sa;
2809 error = prison_check_ip6(cred, &sai6->sin6_addr);
2813 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2814 error = EAFNOSUPPORT;
2820 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2823 prison_check(struct ucred *cred1, struct ucred *cred2)
2826 return ((cred1->cr_prison == cred2->cr_prison ||
2827 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2831 * Return 1 if p2 is a child of p1, otherwise 0.
2834 prison_ischild(struct prison *pr1, struct prison *pr2)
2837 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2844 * Return 1 if the passed credential is in a jail, otherwise 0.
2847 jailed(struct ucred *cred)
2850 return (cred->cr_prison != &prison0);
2854 * Return 1 if the passed credential is in a jail and that jail does not
2855 * have its own virtual network stack, otherwise 0.
2858 jailed_without_vnet(struct ucred *cred)
2864 if (prison_owns_vnet(cred))
2872 * Return the correct hostname (domainname, et al) for the passed credential.
2875 getcredhostname(struct ucred *cred, char *buf, size_t size)
2880 * A NULL credential can be used to shortcut to the physical
2881 * system's hostname.
2883 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2884 mtx_lock(&pr->pr_mtx);
2885 strlcpy(buf, pr->pr_hostname, size);
2886 mtx_unlock(&pr->pr_mtx);
2890 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2893 mtx_lock(&cred->cr_prison->pr_mtx);
2894 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2895 mtx_unlock(&cred->cr_prison->pr_mtx);
2899 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2902 mtx_lock(&cred->cr_prison->pr_mtx);
2903 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2904 mtx_unlock(&cred->cr_prison->pr_mtx);
2908 getcredhostid(struct ucred *cred, unsigned long *hostid)
2911 mtx_lock(&cred->cr_prison->pr_mtx);
2912 *hostid = cred->cr_prison->pr_hostid;
2913 mtx_unlock(&cred->cr_prison->pr_mtx);
2918 * Determine whether the prison represented by cred owns
2919 * its vnet rather than having it inherited.
2921 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2924 prison_owns_vnet(struct ucred *cred)
2928 * vnets cannot be added/removed after jail creation,
2929 * so no need to lock here.
2931 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2936 * Determine whether the subject represented by cred can "see"
2937 * status of a mount point.
2938 * Returns: 0 for permitted, ENOENT otherwise.
2939 * XXX: This function should be called cr_canseemount() and should be
2940 * placed in kern_prot.c.
2943 prison_canseemount(struct ucred *cred, struct mount *mp)
2949 pr = cred->cr_prison;
2950 if (pr->pr_enforce_statfs == 0)
2952 if (pr->pr_root->v_mount == mp)
2954 if (pr->pr_enforce_statfs == 2)
2957 * If jail's chroot directory is set to "/" we should be able to see
2958 * all mount-points from inside a jail.
2959 * This is ugly check, but this is the only situation when jail's
2960 * directory ends with '/'.
2962 if (strcmp(pr->pr_path, "/") == 0)
2964 len = strlen(pr->pr_path);
2966 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2969 * Be sure that we don't have situation where jail's root directory
2970 * is "/some/path" and mount point is "/some/pathpath".
2972 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2978 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2980 char jpath[MAXPATHLEN];
2984 pr = cred->cr_prison;
2985 if (pr->pr_enforce_statfs == 0)
2987 if (prison_canseemount(cred, mp) != 0) {
2988 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2989 strlcpy(sp->f_mntonname, "[restricted]",
2990 sizeof(sp->f_mntonname));
2993 if (pr->pr_root->v_mount == mp) {
2995 * Clear current buffer data, so we are sure nothing from
2996 * the valid path left there.
2998 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2999 *sp->f_mntonname = '/';
3003 * If jail's chroot directory is set to "/" we should be able to see
3004 * all mount-points from inside a jail.
3006 if (strcmp(pr->pr_path, "/") == 0)
3008 len = strlen(pr->pr_path);
3009 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
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 if (*jpath == '\0') {
3016 /* Should never happen. */
3017 *sp->f_mntonname = '/';
3019 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3024 * Check with permission for a specific privilege is granted within jail. We
3025 * have a specific list of accepted privileges; the rest are denied.
3028 prison_priv_check(struct ucred *cred, int priv)
3036 * Privileges specific to prisons with a virtual network stack.
3037 * There might be a duplicate entry here in case the privilege
3038 * is only granted conditionally in the legacy jail case.
3043 * NFS-specific privileges.
3045 case PRIV_NFS_DAEMON:
3046 case PRIV_NFS_LOCKD:
3049 * Network stack privileges.
3051 case PRIV_NET_BRIDGE:
3054 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3055 case PRIV_NET_ROUTE:
3057 case PRIV_NET_SETIFMTU:
3058 case PRIV_NET_SETIFFLAGS:
3059 case PRIV_NET_SETIFCAP:
3060 case PRIV_NET_SETIFDESCR:
3061 case PRIV_NET_SETIFNAME :
3062 case PRIV_NET_SETIFMETRIC:
3063 case PRIV_NET_SETIFPHYS:
3064 case PRIV_NET_SETIFMAC:
3065 case PRIV_NET_SETLANPCP:
3066 case PRIV_NET_ADDMULTI:
3067 case PRIV_NET_DELMULTI:
3068 case PRIV_NET_HWIOCTL:
3069 case PRIV_NET_SETLLADDR:
3070 case PRIV_NET_ADDIFGROUP:
3071 case PRIV_NET_DELIFGROUP:
3072 case PRIV_NET_IFCREATE:
3073 case PRIV_NET_IFDESTROY:
3074 case PRIV_NET_ADDIFADDR:
3075 case PRIV_NET_DELIFADDR:
3078 case PRIV_NET_SETIFVNET:
3079 case PRIV_NET_SETIFFIB:
3082 * 802.11-related privileges.
3084 case PRIV_NET80211_GETKEY:
3086 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3093 case PRIV_NETATM_CFG:
3094 case PRIV_NETATM_ADD:
3095 case PRIV_NETATM_DEL:
3096 case PRIV_NETATM_SET:
3099 * Bluetooth privileges.
3101 case PRIV_NETBLUETOOTH_RAW:
3105 * Netgraph and netgraph module privileges.
3107 case PRIV_NETGRAPH_CONTROL:
3109 case PRIV_NETGRAPH_TTY:
3113 * IPv4 and IPv6 privileges.
3115 case PRIV_NETINET_IPFW:
3116 case PRIV_NETINET_DIVERT:
3117 case PRIV_NETINET_PF:
3118 case PRIV_NETINET_DUMMYNET:
3119 case PRIV_NETINET_CARP:
3120 case PRIV_NETINET_MROUTE:
3121 case PRIV_NETINET_RAW:
3122 case PRIV_NETINET_ADDRCTRL6:
3123 case PRIV_NETINET_ND6:
3124 case PRIV_NETINET_SCOPE6:
3125 case PRIV_NETINET_ALIFETIME6:
3126 case PRIV_NETINET_IPSEC:
3127 case PRIV_NETINET_BINDANY:
3142 * No default: or deny here.
3143 * In case of no permit fall through to next switch().
3145 if (cred->cr_prison->pr_flags & PR_VNET)
3153 * Allow ktrace privileges for root in jail.
3159 * Allow jailed processes to configure audit identity and
3160 * submit audit records (login, etc). In the future we may
3161 * want to further refine the relationship between audit and
3164 case PRIV_AUDIT_GETAUDIT:
3165 case PRIV_AUDIT_SETAUDIT:
3166 case PRIV_AUDIT_SUBMIT:
3170 * Allow jailed processes to manipulate process UNIX
3171 * credentials in any way they see fit.
3173 case PRIV_CRED_SETUID:
3174 case PRIV_CRED_SETEUID:
3175 case PRIV_CRED_SETGID:
3176 case PRIV_CRED_SETEGID:
3177 case PRIV_CRED_SETGROUPS:
3178 case PRIV_CRED_SETREUID:
3179 case PRIV_CRED_SETREGID:
3180 case PRIV_CRED_SETRESUID:
3181 case PRIV_CRED_SETRESGID:
3184 * Jail implements visibility constraints already, so allow
3185 * jailed root to override uid/gid-based constraints.
3187 case PRIV_SEEOTHERGIDS:
3188 case PRIV_SEEOTHERUIDS:
3191 * Jail implements inter-process debugging limits already, so
3192 * allow jailed root various debugging privileges.
3194 case PRIV_DEBUG_DIFFCRED:
3195 case PRIV_DEBUG_SUGID:
3196 case PRIV_DEBUG_UNPRIV:
3199 * Allow jail to set various resource limits and login
3200 * properties, and for now, exceed process resource limits.
3202 case PRIV_PROC_LIMIT:
3203 case PRIV_PROC_SETLOGIN:
3204 case PRIV_PROC_SETRLIMIT:
3207 * System V and POSIX IPC privileges are granted in jail.
3210 case PRIV_IPC_WRITE:
3211 case PRIV_IPC_ADMIN:
3212 case PRIV_IPC_MSGSIZE:
3216 * Jail operations within a jail work on child jails.
3218 case PRIV_JAIL_ATTACH:
3220 case PRIV_JAIL_REMOVE:
3223 * Jail implements its own inter-process limits, so allow
3224 * root processes in jail to change scheduling on other
3225 * processes in the same jail. Likewise for signalling.
3227 case PRIV_SCHED_DIFFCRED:
3228 case PRIV_SCHED_CPUSET:
3229 case PRIV_SIGNAL_DIFFCRED:
3230 case PRIV_SIGNAL_SUGID:
3233 * Allow jailed processes to write to sysctls marked as jail
3236 case PRIV_SYSCTL_WRITEJAIL:
3239 * Allow root in jail to manage a variety of quota
3240 * properties. These should likely be conditional on a
3241 * configuration option.
3243 case PRIV_VFS_GETQUOTA:
3244 case PRIV_VFS_SETQUOTA:
3247 * Since Jail relies on chroot() to implement file system
3248 * protections, grant many VFS privileges to root in jail.
3249 * Be careful to exclude mount-related and NFS-related
3253 case PRIV_VFS_WRITE:
3254 case PRIV_VFS_ADMIN:
3256 case PRIV_VFS_LOOKUP:
3257 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3258 case PRIV_VFS_CHFLAGS_DEV:
3259 case PRIV_VFS_CHOWN:
3260 case PRIV_VFS_CHROOT:
3261 case PRIV_VFS_RETAINSUGID:
3262 case PRIV_VFS_FCHROOT:
3264 case PRIV_VFS_SETGID:
3266 case PRIV_VFS_STICKYFILE:
3269 * As in the non-jail case, non-root users are expected to be
3270 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3271 * exists in the jail and they have permission to access it).
3273 case PRIV_KMEM_READ:
3277 * Depending on the global setting, allow privilege of
3278 * setting system flags.
3280 case PRIV_VFS_SYSFLAGS:
3281 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3287 * Depending on the global setting, allow privilege of
3288 * mounting/unmounting file systems.
3290 case PRIV_VFS_MOUNT:
3291 case PRIV_VFS_UNMOUNT:
3292 case PRIV_VFS_MOUNT_NONUSER:
3293 case PRIV_VFS_MOUNT_OWNER:
3294 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3295 cred->cr_prison->pr_enforce_statfs < 2)
3301 * Conditionnaly allow locking (unlocking) physical pages
3305 case PRIV_VM_MUNLOCK:
3306 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3312 * Conditionally allow jailed root to bind reserved ports.
3314 case PRIV_NETINET_RESERVEDPORT:
3315 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3321 * Allow jailed root to reuse in-use ports.
3323 case PRIV_NETINET_REUSEPORT:
3327 * Allow jailed root to set certain IPv4/6 (option) headers.
3329 case PRIV_NETINET_SETHDROPTS:
3333 * Conditionally allow creating raw sockets in jail.
3335 case PRIV_NETINET_RAW:
3336 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3342 * Since jail implements its own visibility limits on netstat
3343 * sysctls, allow getcred. This allows identd to work in
3346 case PRIV_NETINET_GETCRED:
3350 * Allow jailed root to set loginclass.
3352 case PRIV_PROC_SETLOGINCLASS:
3356 * Do not allow a process inside a jail to read the kernel
3357 * message buffer unless explicitly permitted.
3360 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3366 * In all remaining cases, deny the privilege request. This
3367 * includes almost all network privileges, many system
3368 * configuration privileges.
3375 * Return the part of pr2's name that is relative to pr1, or the whole name
3376 * if it does not directly follow.
3380 prison_name(struct prison *pr1, struct prison *pr2)
3384 /* Jails see themselves as "0" (if they see themselves at all). */
3387 name = pr2->pr_name;
3388 if (prison_ischild(pr1, pr2)) {
3390 * pr1 isn't locked (and allprison_lock may not be either)
3391 * so its length can't be counted on. But the number of dots
3392 * can be counted on - and counted.
3394 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3395 name = strchr(name, '.') + 1;
3401 * Return the part of pr2's path that is relative to pr1, or the whole path
3402 * if it does not directly follow.
3405 prison_path(struct prison *pr1, struct prison *pr2)
3407 char *path1, *path2;
3410 path1 = pr1->pr_path;
3411 path2 = pr2->pr_path;
3412 if (!strcmp(path1, "/"))
3414 len1 = strlen(path1);
3415 if (strncmp(path1, path2, len1))
3417 if (path2[len1] == '\0')
3419 if (path2[len1] == '/')
3420 return (path2 + len1);
3426 * Jail-related sysctls.
3428 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3432 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3435 struct prison *pr, *cpr;
3437 struct in_addr *ip4 = NULL;
3441 struct in6_addr *ip6 = NULL;
3446 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3447 pr = req->td->td_ucred->cr_prison;
3449 sx_slock(&allprison_lock);
3450 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3451 #if defined(INET) || defined(INET6)
3454 mtx_lock(&cpr->pr_mtx);
3456 if (cpr->pr_ip4s > 0) {
3457 if (ip4s < cpr->pr_ip4s) {
3458 ip4s = cpr->pr_ip4s;
3459 mtx_unlock(&cpr->pr_mtx);
3460 ip4 = realloc(ip4, ip4s *
3461 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3464 bcopy(cpr->pr_ip4, ip4,
3465 cpr->pr_ip4s * sizeof(struct in_addr));
3469 if (cpr->pr_ip6s > 0) {
3470 if (ip6s < cpr->pr_ip6s) {
3471 ip6s = cpr->pr_ip6s;
3472 mtx_unlock(&cpr->pr_mtx);
3473 ip6 = realloc(ip6, ip6s *
3474 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3477 bcopy(cpr->pr_ip6, ip6,
3478 cpr->pr_ip6s * sizeof(struct in6_addr));
3481 if (cpr->pr_ref == 0) {
3482 mtx_unlock(&cpr->pr_mtx);
3485 bzero(xp, sizeof(*xp));
3486 xp->pr_version = XPRISON_VERSION;
3487 xp->pr_id = cpr->pr_id;
3488 xp->pr_state = cpr->pr_uref > 0
3489 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3490 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3491 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3492 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3494 xp->pr_ip4s = cpr->pr_ip4s;
3497 xp->pr_ip6s = cpr->pr_ip6s;
3499 mtx_unlock(&cpr->pr_mtx);
3500 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3504 if (xp->pr_ip4s > 0) {
3505 error = SYSCTL_OUT(req, ip4,
3506 xp->pr_ip4s * sizeof(struct in_addr));
3512 if (xp->pr_ip6s > 0) {
3513 error = SYSCTL_OUT(req, ip6,
3514 xp->pr_ip6s * sizeof(struct in6_addr));
3520 sx_sunlock(&allprison_lock);
3531 SYSCTL_OID(_security_jail, OID_AUTO, list,
3532 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3533 sysctl_jail_list, "S", "List of active jails");
3536 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3540 injail = jailed(req->td->td_ucred);
3541 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3546 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3547 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3548 sysctl_jail_jailed, "I", "Process in jail?");
3551 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3553 int error, havevnet;
3555 struct ucred *cred = req->td->td_ucred;
3557 havevnet = jailed(cred) && prison_owns_vnet(cred);
3561 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3566 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3567 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3568 sysctl_jail_vnet, "I", "Jail owns vnet?");
3570 #if defined(INET) || defined(INET6)
3571 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3572 &jail_max_af_ips, 0,
3573 "Number of IP addresses a jail may have at most per address family (deprecated)");
3577 * Default parameters for jail(2) compatibility. For historical reasons,
3578 * the sysctl names have varying similarity to the parameter names. Prisons
3579 * just see their own parameters, and can't change them.
3582 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3585 int allow, error, i;
3587 pr = req->td->td_ucred->cr_prison;
3588 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3590 /* Get the current flag value, and convert it to a boolean. */
3591 i = (allow & arg2) ? 1 : 0;
3594 error = sysctl_handle_int(oidp, &i, 0, req);
3595 if (error || !req->newptr)
3601 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3604 mtx_lock(&prison0.pr_mtx);
3605 jail_default_allow = (jail_default_allow & ~arg2) | i;
3606 mtx_unlock(&prison0.pr_mtx);
3610 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3611 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3612 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3613 "Processes in jail can set their hostnames (deprecated)");
3614 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3615 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3616 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3617 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3618 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3619 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3620 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3621 "Processes in jail can use System V IPC primitives (deprecated)");
3622 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3623 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3624 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3625 "Prison root can create raw sockets (deprecated)");
3626 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3627 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3628 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3629 "Processes in jail can alter system file flags (deprecated)");
3630 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3631 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3632 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3633 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3636 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3641 pr = req->td->td_ucred->cr_prison;
3642 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3643 error = sysctl_handle_int(oidp, &level, 0, req);
3644 if (error || !req->newptr)
3646 *(int *)arg1 = level;
3650 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3651 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3652 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3653 sysctl_jail_default_level, "I",
3654 "Processes in jail cannot see all mounted file systems (deprecated)");
3656 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3657 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3658 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3659 sysctl_jail_default_level, "I",
3660 "Ruleset for the devfs filesystem in jail (deprecated)");
3663 * Nodes to describe jail parameters. Maximum length of string parameters
3664 * is returned in the string itself, and the other parameters exist merely
3665 * to make themselves and their types known.
3667 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3671 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3678 switch (oidp->oid_kind & CTLTYPE)
3684 if (!(req->flags & SCTL_MASK32))
3686 return (SYSCTL_OUT(req, &l, sizeof(l)));
3690 return (SYSCTL_OUT(req, &i, sizeof(i)));
3691 case CTLTYPE_STRING:
3692 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3694 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3695 case CTLTYPE_STRUCT:
3697 return (SYSCTL_OUT(req, &s, sizeof(s)));
3703 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3704 * jail creation time but cannot be changed in an existing jail.
3706 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3707 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3708 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3709 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3710 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3711 "I", "Jail secure level");
3712 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3713 "Jail value for kern.osreldate and uname -K");
3714 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3715 "Jail value for kern.osrelease and uname -r");
3716 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3717 "I", "Jail cannot see all mounted file systems");
3718 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3719 "I", "Ruleset for in-jail devfs mounts");
3720 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3721 "B", "Jail persistence");
3723 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3724 "E,jailsys", "Virtual network stack");
3726 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3727 "B", "Jail is in the process of shutting down");
3729 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3730 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3731 "I", "Current number of child jails");
3732 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3733 "I", "Maximum number of child jails");
3735 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3736 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3738 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3739 "Jail NIS domainname");
3740 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3742 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3743 "LU", "Jail host ID");
3745 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3746 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3749 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3750 "Jail IPv4 address virtualization");
3751 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3752 "S,in_addr,a", "Jail IPv4 addresses");
3753 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3754 "B", "Do (not) use IPv4 source address selection rather than the "
3755 "primary jail IPv4 address.");
3758 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3759 "Jail IPv6 address virtualization");
3760 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3761 "S,in6_addr,a", "Jail IPv6 addresses");
3762 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3763 "B", "Do (not) use IPv6 source address selection rather than the "
3764 "primary jail IPv6 address.");
3767 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3768 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3769 "B", "Jail may set hostname");
3770 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3771 "B", "Jail may use SYSV IPC");
3772 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3773 "B", "Jail may create raw sockets");
3774 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3775 "B", "Jail may alter system file flags");
3776 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3777 "B", "Jail may set file quotas");
3778 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3779 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3780 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3781 "B", "Jail may lock (unlock) physical pages in memory");
3782 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3783 "B", "Jail may bind sockets to reserved ports");
3784 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3785 "B", "Jail may read the kernel message buffer");
3787 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3788 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3789 "B", "Jail may mount/unmount jail-friendly file systems in general");
3792 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
3793 * its associated bit in the pr_allow bitmask, or zero if the parameter was
3797 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3800 struct bool_flags *bf;
3801 struct sysctl_oid *parent;
3802 char *allow_name, *allow_noname, *allowed;
3803 #ifndef NO_SYSCTL_DESCR
3804 char *descr_deprecated;
3806 unsigned allow_flag;
3809 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3811 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3813 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3814 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3815 free(allow_name, M_PRISON);
3820 * See if this parameter has already beed added, i.e. a module was
3821 * previously loaded/unloaded.
3823 mtx_lock(&prison0.pr_mtx);
3824 for (bf = pr_flag_allow;
3825 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3827 if (strcmp(bf->name, allow_name) == 0) {
3828 allow_flag = bf->flag;
3834 * Find a free bit in prison0's pr_allow, failing if there are none
3835 * (which shouldn't happen as long as we keep track of how many
3836 * potential dynamic flags exist).
3838 for (allow_flag = 1;; allow_flag <<= 1) {
3839 if (allow_flag == 0)
3841 if ((prison0.pr_allow & allow_flag) == 0)
3846 * Note the parameter in the next open slot in pr_flag_allow.
3847 * Set the flag last so code that checks pr_flag_allow can do so
3850 for (bf = pr_flag_allow; bf->flag != 0; bf++)
3851 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3852 /* This should never happen, but is not fatal. */
3856 prison0.pr_allow |= allow_flag;
3857 bf->name = allow_name;
3858 bf->noname = allow_noname;
3859 bf->flag = allow_flag;
3860 mtx_unlock(&prison0.pr_mtx);
3863 * Create sysctls for the paramter, and the back-compat global
3867 ? SYSCTL_ADD_NODE(NULL,
3868 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3869 OID_AUTO, prefix, 0, 0, prefix_descr)
3870 : &sysctl___security_jail_param_allow;
3871 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3872 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3873 NULL, 0, sysctl_jail_param, "B", descr);
3875 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3876 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3877 #ifndef NO_SYSCTL_DESCR
3878 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3881 (void)SYSCTL_ADD_PROC(NULL,
3882 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3883 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3884 sysctl_jail_default_allow, "I", descr_deprecated);
3885 #ifndef NO_SYSCTL_DESCR
3886 free(descr_deprecated, M_TEMP);
3888 free(allowed, M_TEMP);
3893 mtx_unlock(&prison0.pr_mtx);
3894 free(allow_name, M_PRISON);
3895 free(allow_noname, M_PRISON);
3900 * The VFS system will register jail-aware filesystems here. They each get
3901 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3902 * attempts to mount.
3905 prison_add_vfs(struct vfsconf *vfsp)
3907 #ifdef NO_SYSCTL_DESCR
3909 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3914 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3916 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3918 free(descr, M_TEMP);
3924 prison_racct_foreach(void (*callback)(struct racct *racct,
3925 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3926 void *arg2, void *arg3)
3928 struct prison_racct *prr;
3930 ASSERT_RACCT_ENABLED();
3932 sx_slock(&allprison_lock);
3935 LIST_FOREACH(prr, &allprison_racct, prr_next)
3936 (callback)(prr->prr_racct, arg2, arg3);
3939 sx_sunlock(&allprison_lock);
3942 static struct prison_racct *
3943 prison_racct_find_locked(const char *name)
3945 struct prison_racct *prr;
3947 ASSERT_RACCT_ENABLED();
3948 sx_assert(&allprison_lock, SA_XLOCKED);
3950 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3953 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3954 if (strcmp(name, prr->prr_name) != 0)
3957 /* Found prison_racct with a matching name? */
3958 prison_racct_hold(prr);
3962 /* Add new prison_racct. */
3963 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3964 racct_create(&prr->prr_racct);
3966 strcpy(prr->prr_name, name);
3967 refcount_init(&prr->prr_refcount, 1);
3968 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3973 struct prison_racct *
3974 prison_racct_find(const char *name)
3976 struct prison_racct *prr;
3978 ASSERT_RACCT_ENABLED();
3980 sx_xlock(&allprison_lock);
3981 prr = prison_racct_find_locked(name);
3982 sx_xunlock(&allprison_lock);
3987 prison_racct_hold(struct prison_racct *prr)
3990 ASSERT_RACCT_ENABLED();
3992 refcount_acquire(&prr->prr_refcount);
3996 prison_racct_free_locked(struct prison_racct *prr)
3999 ASSERT_RACCT_ENABLED();
4000 sx_assert(&allprison_lock, SA_XLOCKED);
4002 if (refcount_release(&prr->prr_refcount)) {
4003 racct_destroy(&prr->prr_racct);
4004 LIST_REMOVE(prr, prr_next);
4005 free(prr, M_PRISON_RACCT);
4010 prison_racct_free(struct prison_racct *prr)
4014 ASSERT_RACCT_ENABLED();
4015 sx_assert(&allprison_lock, SA_UNLOCKED);
4017 old = prr->prr_refcount;
4018 if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4021 sx_xlock(&allprison_lock);
4022 prison_racct_free_locked(prr);
4023 sx_xunlock(&allprison_lock);
4027 prison_racct_attach(struct prison *pr)
4029 struct prison_racct *prr;
4031 ASSERT_RACCT_ENABLED();
4032 sx_assert(&allprison_lock, SA_XLOCKED);
4034 prr = prison_racct_find_locked(pr->pr_name);
4035 KASSERT(prr != NULL, ("cannot find prison_racct"));
4037 pr->pr_prison_racct = prr;
4041 * Handle jail renaming. From the racct point of view, renaming means
4042 * moving from one prison_racct to another.
4045 prison_racct_modify(struct prison *pr)
4051 struct prison_racct *oldprr;
4053 ASSERT_RACCT_ENABLED();
4055 sx_slock(&allproc_lock);
4056 sx_xlock(&allprison_lock);
4058 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4059 sx_xunlock(&allprison_lock);
4060 sx_sunlock(&allproc_lock);
4064 oldprr = pr->pr_prison_racct;
4065 pr->pr_prison_racct = NULL;
4067 prison_racct_attach(pr);
4070 * Move resource utilisation records.
4072 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4076 * Force rctl to reattach rules to processes.
4078 FOREACH_PROC_IN_SYSTEM(p) {
4080 cred = crhold(p->p_ucred);
4082 rctl_proc_ucred_changed(p, cred);
4087 sx_sunlock(&allproc_lock);
4088 prison_racct_free_locked(oldprr);
4089 sx_xunlock(&allprison_lock);
4093 prison_racct_detach(struct prison *pr)
4096 ASSERT_RACCT_ENABLED();
4097 sx_assert(&allprison_lock, SA_UNLOCKED);
4099 if (pr->pr_prison_racct == NULL)
4101 prison_racct_free(pr->pr_prison_racct);
4102 pr->pr_prison_racct = NULL;
4109 db_show_prison(struct prison *pr)
4111 struct bool_flags *bf;
4112 struct jailsys_flags *jsf;
4113 #if defined(INET) || defined(INET6)
4118 char ip4buf[INET_ADDRSTRLEN];
4121 char ip6buf[INET6_ADDRSTRLEN];
4124 db_printf("prison %p:\n", pr);
4125 db_printf(" jid = %d\n", pr->pr_id);
4126 db_printf(" name = %s\n", pr->pr_name);
4127 db_printf(" parent = %p\n", pr->pr_parent);
4128 db_printf(" ref = %d\n", pr->pr_ref);
4129 db_printf(" uref = %d\n", pr->pr_uref);
4130 db_printf(" path = %s\n", pr->pr_path);
4131 db_printf(" cpuset = %d\n", pr->pr_cpuset
4132 ? pr->pr_cpuset->cs_id : -1);
4134 db_printf(" vnet = %p\n", pr->pr_vnet);
4136 db_printf(" root = %p\n", pr->pr_root);
4137 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4138 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4139 db_printf(" children.max = %d\n", pr->pr_childmax);
4140 db_printf(" children.cur = %d\n", pr->pr_childcount);
4141 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4142 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4143 db_printf(" flags = 0x%x", pr->pr_flags);
4144 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4145 if (pr->pr_flags & bf->flag)
4146 db_printf(" %s", bf->name);
4147 for (jsf = pr_flag_jailsys;
4148 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4150 f = pr->pr_flags & (jsf->disable | jsf->new);
4151 db_printf(" %-16s= %s\n", jsf->name,
4152 (f != 0 && f == jsf->disable) ? "disable"
4153 : (f == jsf->new) ? "new"
4156 db_printf(" allow = 0x%x", pr->pr_allow);
4157 for (bf = pr_flag_allow;
4158 bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4160 if (pr->pr_allow & bf->flag)
4161 db_printf(" %s", bf->name);
4163 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4164 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4165 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4166 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4167 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4169 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4170 for (ii = 0; ii < pr->pr_ip4s; ii++)
4171 db_printf(" %s %s\n",
4172 ii == 0 ? "ip4.addr =" : " ",
4173 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4176 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4177 for (ii = 0; ii < pr->pr_ip6s; ii++)
4178 db_printf(" %s %s\n",
4179 ii == 0 ? "ip6.addr =" : " ",
4180 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4184 DB_SHOW_COMMAND(prison, db_show_prison_command)
4190 * Show all prisons in the list, and prison0 which is not
4193 db_show_prison(&prison0);
4194 if (!db_pager_quit) {
4195 TAILQ_FOREACH(pr, &allprison, pr_list) {
4207 /* Look for a prison with the ID and with references. */
4208 TAILQ_FOREACH(pr, &allprison, pr_list)
4209 if (pr->pr_id == addr && pr->pr_ref > 0)
4212 /* Look again, without requiring a reference. */
4213 TAILQ_FOREACH(pr, &allprison, pr_list)
4214 if (pr->pr_id == addr)
4217 /* Assume address points to a valid prison. */
4218 pr = (struct prison *)addr;