2 * Copyright (c) 1999 Poul-Henning Kamp.
3 * Copyright (c) 2008 Bjoern A. Zeeb.
4 * Copyright (c) 2009 James Gritton.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include "opt_compat.h"
35 #include "opt_inet6.h"
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
54 #include <sys/refcount.h>
56 #include <sys/sysent.h>
57 #include <sys/namei.h>
58 #include <sys/mount.h>
59 #include <sys/queue.h>
60 #include <sys/socket.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysctl.h>
63 #include <sys/vnode.h>
68 #include <netinet/in.h>
74 #include <security/mac/mac_framework.h>
76 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000"
78 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
79 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
81 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
84 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
86 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
90 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
92 #define _PR_IP_SADDRSEL 0
96 /* prison0 describes what is "real" about the system. */
97 struct prison prison0 = {
103 .pr_securelevel = -1,
105 .pr_childmax = JAIL_MAX,
106 .pr_hostuuid = DEFAULT_HOSTUUID,
107 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
109 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
111 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
113 .pr_allow = PR_ALLOW_ALL,
115 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
117 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
118 struct sx allprison_lock;
119 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
120 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
121 LIST_HEAD(, prison_racct) allprison_racct;
124 static int do_jail_attach(struct thread *td, struct prison *pr);
125 static void prison_complete(void *context, int pending);
126 static void prison_deref(struct prison *pr, int flags);
127 static char *prison_path(struct prison *pr1, struct prison *pr2);
128 static void prison_remove_one(struct prison *pr);
130 static void prison_racct_attach(struct prison *pr);
131 static void prison_racct_modify(struct prison *pr);
132 static void prison_racct_detach(struct prison *pr);
135 /* Flags for prison_deref */
136 #define PD_DEREF 0x01
137 #define PD_DEUREF 0x02
138 #define PD_LOCKED 0x04
139 #define PD_LIST_SLOCKED 0x08
140 #define PD_LIST_XLOCKED 0x10
143 * Parameter names corresponding to PR_* flag values. Size values are for kvm
144 * as we cannot figure out the size of a sparse array, or an array without a
147 static char *pr_flag_names[] = {
150 [7] = "ip4.saddrsel",
153 [8] = "ip6.saddrsel",
156 const size_t pr_flag_names_size = sizeof(pr_flag_names);
158 static char *pr_flag_nonames[] = {
161 [7] = "ip4.nosaddrsel",
164 [8] = "ip6.nosaddrsel",
167 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
169 struct jailsys_flags {
173 } 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 static char *pr_allow_names[] = {
188 "allow.set_hostname",
196 "allow.mount.nullfs",
198 "allow.mount.procfs",
200 "allow.mount.fdescfs",
201 "allow.mount.linprocfs",
202 "allow.mount.linsysfs",
205 const size_t pr_allow_names_size = sizeof(pr_allow_names);
207 static char *pr_allow_nonames[] = {
208 "allow.noset_hostname",
210 "allow.noraw_sockets",
215 "allow.mount.nodevfs",
216 "allow.mount.nonullfs",
218 "allow.mount.noprocfs",
219 "allow.mount.notmpfs",
220 "allow.mount.nofdescfs",
221 "allow.mount.nolinprocfs",
222 "allow.mount.nolinsysfs",
223 "allow.noread_msgbuf",
225 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
227 #define JAIL_DEFAULT_ALLOW PR_ALLOW_SET_HOSTNAME
228 #define JAIL_DEFAULT_ENFORCE_STATFS 2
229 #define JAIL_DEFAULT_DEVFS_RSNUM 0
230 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
231 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
232 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
233 #if defined(INET) || defined(INET6)
234 static unsigned jail_max_af_ips = 255;
238 * Initialize the parts of prison0 that can't be static-initialized with
239 * constants. This is called from proc0_init() after creating thread0 cpuset.
245 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
246 prison0.pr_osreldate = osreldate;
247 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
256 sys_jail(struct thread *td, struct jail_args *uap)
262 error = copyin(uap->jail, &version, sizeof(uint32_t));
271 /* FreeBSD single IPv4 jails. */
272 bzero(&j, sizeof(struct jail));
273 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
276 j.version = j0.version;
278 j.hostname = j0.hostname;
279 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
285 * Version 1 was used by multi-IPv4 jail implementations
286 * that never made it into the official kernel.
290 case 2: /* JAIL_API_VERSION */
291 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
292 error = copyin(uap->jail, &j, sizeof(struct jail));
298 /* Sci-Fi jails are not supported, sorry. */
301 return (kern_jail(td, &j));
305 kern_jail(struct thread *td, struct jail *j)
307 struct iovec optiov[2 * (4 + nitems(pr_allow_names)
316 char *u_path, *u_hostname, *u_name;
319 struct in_addr *u_ip4;
322 struct in6_addr *u_ip6;
325 int error, enforce_statfs, fi;
327 bzero(&optiov, sizeof(optiov));
328 opt.uio_iov = optiov;
332 opt.uio_segflg = UIO_SYSSPACE;
333 opt.uio_rw = UIO_READ;
336 /* Set permissions for top-level jails from sysctls. */
337 if (!jailed(td->td_ucred)) {
338 for (fi = 0; fi < nitems(pr_allow_names); fi++) {
339 optiov[opt.uio_iovcnt].iov_base =
340 (jail_default_allow & (1 << fi))
341 ? pr_allow_names[fi] : pr_allow_nonames[fi];
342 optiov[opt.uio_iovcnt].iov_len =
343 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
346 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
347 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
349 enforce_statfs = jail_default_enforce_statfs;
350 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
351 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
355 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
357 ip4s = (j->version == 0) ? 1 : j->ip4s;
358 if (ip4s > jail_max_af_ips)
360 tmplen += ip4s * sizeof(struct in_addr);
366 if (j->ip6s > jail_max_af_ips)
368 tmplen += j->ip6s * sizeof(struct in6_addr);
373 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
374 u_hostname = u_path + MAXPATHLEN;
375 u_name = u_hostname + MAXHOSTNAMELEN;
377 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
381 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
383 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
386 optiov[opt.uio_iovcnt].iov_base = "path";
387 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
389 optiov[opt.uio_iovcnt].iov_base = u_path;
390 error = copyinstr(j->path, u_path, MAXPATHLEN,
391 &optiov[opt.uio_iovcnt].iov_len);
393 free(u_path, M_TEMP);
397 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
398 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
400 optiov[opt.uio_iovcnt].iov_base = u_hostname;
401 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
402 &optiov[opt.uio_iovcnt].iov_len);
404 free(u_path, M_TEMP);
408 if (j->jailname != NULL) {
409 optiov[opt.uio_iovcnt].iov_base = "name";
410 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
412 optiov[opt.uio_iovcnt].iov_base = u_name;
413 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
414 &optiov[opt.uio_iovcnt].iov_len);
416 free(u_path, M_TEMP);
422 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
423 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
425 optiov[opt.uio_iovcnt].iov_base = u_ip4;
426 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
428 u_ip4->s_addr = j->ip4s;
430 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
432 free(u_path, M_TEMP);
439 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
440 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
442 optiov[opt.uio_iovcnt].iov_base = u_ip6;
443 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
444 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
446 free(u_path, M_TEMP);
451 KASSERT(opt.uio_iovcnt <= nitems(optiov),
452 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
453 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
454 free(u_path, M_TEMP);
460 * struct jail_set_args {
461 * struct iovec *iovp;
462 * unsigned int iovcnt;
467 sys_jail_set(struct thread *td, struct jail_set_args *uap)
472 /* Check that we have an even number of iovecs. */
476 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
479 error = kern_jail_set(td, auio, uap->flags);
485 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
492 struct in6_addr *ip6;
495 struct vfsoptlist *opts;
496 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
498 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
499 char *g_path, *osrelstr;
500 #if defined(INET) || defined(INET6)
505 size_t namelen, onamelen, pnamelen;
506 int born, created, cuflags, descend, enforce;
507 int error, errmsg_len, errmsg_pos;
508 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
509 int fi, jid, jsys, len, level;
510 int childmax, osreldt, rsnum, slevel;
511 int fullpath_disabled;
512 #if defined(INET) || defined(INET6)
521 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
525 error = priv_check(td, PRIV_JAIL_SET);
526 if (!error && (flags & JAIL_ATTACH))
527 error = priv_check(td, PRIV_JAIL_ATTACH);
530 mypr = td->td_ucred->cr_prison;
531 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
533 if (flags & ~JAIL_SET_MASK)
537 * Check all the parameters before committing to anything. Not all
538 * errors can be caught early, but we may as well try. Also, this
539 * takes care of some expensive stuff (path lookup) before getting
540 * the allprison lock.
542 * XXX Jails are not filesystems, and jail parameters are not mount
543 * options. But it makes more sense to re-use the vfsopt code
544 * than duplicate it under a different name.
546 error = vfs_buildopts(optuio, &opts);
557 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
560 vfs_opterror(opts, "no valid operation (create or update)");
564 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
570 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
579 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
587 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
592 else if (enforce < 0 || enforce > 2) {
598 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
606 pr_flags = ch_flags = 0;
607 for (fi = 0; fi < nitems(pr_flag_names); fi++) {
608 if (pr_flag_names[fi] == NULL)
610 vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
611 vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
613 ch_flags |= pr_flags;
614 for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
615 error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
622 case JAIL_SYS_DISABLE:
623 if (!pr_flag_jailsys[fi].disable) {
627 pr_flags |= pr_flag_jailsys[fi].disable;
630 pr_flags |= pr_flag_jailsys[fi].new;
632 case JAIL_SYS_INHERIT:
639 pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
641 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
642 && !(pr_flags & PR_PERSIST)) {
644 vfs_opterror(opts, "new jail must persist or attach");
648 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
650 vfs_opterror(opts, "vnet cannot be changed after creation");
655 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
657 vfs_opterror(opts, "ip4 cannot be changed after creation");
662 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
664 vfs_opterror(opts, "ip6 cannot be changed after creation");
669 pr_allow = ch_allow = 0;
670 for (fi = 0; fi < nitems(pr_allow_names); fi++) {
671 vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
672 vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
674 ch_allow |= pr_allow;
676 error = vfs_getopt(opts, "name", (void **)&name, &len);
682 if (len == 0 || name[len - 1] != '\0') {
686 if (len > MAXHOSTNAMELEN) {
687 error = ENAMETOOLONG;
692 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
700 if (len == 0 || host[len - 1] != '\0') {
704 if (len > MAXHOSTNAMELEN) {
705 error = ENAMETOOLONG;
710 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
718 if (len == 0 || domain[len - 1] != '\0') {
722 if (len > MAXHOSTNAMELEN) {
723 error = ENAMETOOLONG;
728 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
736 if (len == 0 || uuid[len - 1] != '\0') {
740 if (len > HOSTUUIDLEN) {
741 error = ENAMETOOLONG;
746 #ifdef COMPAT_FREEBSD32
747 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
750 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
754 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
766 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
771 else if (ip4s & (sizeof(*ip4) - 1)) {
775 ch_flags |= PR_IP4_USER;
776 pr_flags |= PR_IP4_USER;
778 ip4s /= sizeof(*ip4);
779 if (ip4s > jail_max_af_ips) {
781 vfs_opterror(opts, "too many IPv4 addresses");
784 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
785 bcopy(op, ip4, ip4s * sizeof(*ip4));
787 * IP addresses are all sorted but ip[0] to preserve
788 * the primary IP address as given from userland.
789 * This special IP is used for unbound outgoing
790 * connections as well for "loopback" traffic in case
791 * source address selection cannot find any more fitting
792 * address to connect from.
795 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
798 * Check for duplicate addresses and do some simple
799 * zero and broadcast checks. If users give other bogus
800 * addresses it is their problem.
802 * We do not have to care about byte order for these
803 * checks so we will do them in NBO.
805 for (ii = 0; ii < ip4s; ii++) {
806 if (ip4[ii].s_addr == INADDR_ANY ||
807 ip4[ii].s_addr == INADDR_BROADCAST) {
812 (ip4[0].s_addr == ip4[ii+1].s_addr ||
813 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
823 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
828 else if (ip6s & (sizeof(*ip6) - 1)) {
832 ch_flags |= PR_IP6_USER;
833 pr_flags |= PR_IP6_USER;
835 ip6s /= sizeof(*ip6);
836 if (ip6s > jail_max_af_ips) {
838 vfs_opterror(opts, "too many IPv6 addresses");
841 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
842 bcopy(op, ip6, ip6s * sizeof(*ip6));
844 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
846 for (ii = 0; ii < ip6s; ii++) {
847 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
852 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
853 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
863 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
864 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
867 "vnet jails cannot have IP address restrictions");
872 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
878 if (flags & JAIL_UPDATE) {
881 "osrelease cannot be changed after creation");
884 if (len == 0 || len >= OSRELEASELEN) {
887 "osrelease string must be 1-%d bytes long",
893 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
899 if (flags & JAIL_UPDATE) {
902 "osreldate cannot be changed after creation");
907 vfs_opterror(opts, "osreldate cannot be 0");
912 fullpath_disabled = 0;
914 error = vfs_getopt(opts, "path", (void **)&path, &len);
920 if (flags & JAIL_UPDATE) {
923 "path cannot be changed after creation");
926 if (len == 0 || path[len - 1] != '\0') {
930 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
936 NDFREE(&nd, NDF_ONLY_PNBUF);
937 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
938 strlcpy(g_path, path, MAXPATHLEN);
939 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
942 else if (error == ENODEV) {
943 /* proceed if sysctl debug.disablefullpath == 1 */
944 fullpath_disabled = 1;
945 if (len < 2 || (len == 2 && path[0] == '/'))
948 /* exit on other errors */
951 if (root->v_type != VDIR) {
957 if (fullpath_disabled) {
958 /* Leave room for a real-root full pathname. */
959 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
960 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
961 error = ENAMETOOLONG;
969 * Find the specified jail, or at least its parent.
970 * This abuses the file error codes ENOENT and EEXIST.
974 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
975 namelc = strrchr(name, '.');
976 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
980 sx_xlock(&allprison_lock);
983 * See if a requested jid already exists. There is an
984 * information leak here if the jid exists but is not within
985 * the caller's jail hierarchy. Jail creators will get EEXIST
986 * even though they cannot see the jail, and CREATE | UPDATE
987 * will return ENOENT which is not normally a valid error.
991 vfs_opterror(opts, "negative jid");
992 goto done_unlock_list;
994 pr = prison_find(jid);
997 /* Create: jid must not exist. */
998 if (cuflags == JAIL_CREATE) {
999 mtx_unlock(&pr->pr_mtx);
1001 vfs_opterror(opts, "jail %d already exists",
1003 goto done_unlock_list;
1005 if (!prison_ischild(mypr, pr)) {
1006 mtx_unlock(&pr->pr_mtx);
1008 } else if (pr->pr_uref == 0) {
1009 if (!(flags & JAIL_DYING)) {
1010 mtx_unlock(&pr->pr_mtx);
1012 vfs_opterror(opts, "jail %d is dying",
1014 goto done_unlock_list;
1015 } else if ((flags & JAIL_ATTACH) ||
1016 (pr_flags & PR_PERSIST)) {
1018 * A dying jail might be resurrected
1019 * (via attach or persist), but first
1020 * it must determine if another jail
1021 * has claimed its name. Accomplish
1022 * this by implicitly re-setting the
1026 name = prison_name(mypr, pr);
1031 /* Update: jid must exist. */
1032 if (cuflags == JAIL_UPDATE) {
1034 vfs_opterror(opts, "jail %d not found", jid);
1035 goto done_unlock_list;
1040 * If the caller provided a name, look for a jail by that name.
1041 * This has different semantics for creates and updates keyed by jid
1042 * (where the name must not already exist in a different jail),
1043 * and updates keyed by the name itself (where the name must exist
1044 * because that is the jail being updated).
1048 namelc = strrchr(name, '.');
1053 * This is a hierarchical name. Split it into the
1054 * parent and child names, and make sure the parent
1055 * exists or matches an already found jail.
1058 if (strncmp(name, ppr->pr_name, namelc - name)
1059 || ppr->pr_name[namelc - name] != '\0') {
1060 mtx_unlock(&pr->pr_mtx);
1063 "cannot change jail's parent");
1064 goto done_unlock_list;
1068 ppr = prison_find_name(mypr, name);
1072 "jail \"%s\" not found", name);
1073 goto done_unlock_list;
1075 mtx_unlock(&ppr->pr_mtx);
1080 if (namelc[0] != '\0') {
1082 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1085 FOREACH_PRISON_CHILD(ppr, tpr) {
1086 if (tpr != pr && tpr->pr_ref > 0 &&
1087 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1089 cuflags != JAIL_CREATE) {
1090 mtx_lock(&tpr->pr_mtx);
1091 if (tpr->pr_ref > 0) {
1096 if (tpr->pr_uref > 0) {
1102 mtx_unlock(&tpr->pr_mtx);
1103 } else if (tpr->pr_uref > 0) {
1105 * Create, or update(jid):
1106 * name must not exist in an
1107 * active sibling jail.
1111 mtx_unlock(&pr->pr_mtx);
1113 "jail \"%s\" already exists",
1115 goto done_unlock_list;
1119 /* If no active jail is found, use a dying one. */
1120 if (deadpr != NULL && pr == NULL) {
1121 if (flags & JAIL_DYING) {
1122 mtx_lock(&deadpr->pr_mtx);
1123 if (deadpr->pr_ref == 0) {
1124 mtx_unlock(&deadpr->pr_mtx);
1128 } else if (cuflags == JAIL_UPDATE) {
1131 "jail \"%s\" is dying", name);
1132 goto done_unlock_list;
1135 /* Update: name must exist if no jid. */
1136 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1138 vfs_opterror(opts, "jail \"%s\" not found",
1140 goto done_unlock_list;
1144 /* Update: must provide a jid or name. */
1145 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1147 vfs_opterror(opts, "update specified no jail");
1148 goto done_unlock_list;
1151 /* If there's no prison to update, create a new one and link it in. */
1153 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1154 if (tpr->pr_childcount >= tpr->pr_childmax) {
1156 vfs_opterror(opts, "prison limit exceeded");
1157 goto done_unlock_list;
1160 mtx_lock(&ppr->pr_mtx);
1161 if (ppr->pr_ref == 0) {
1162 mtx_unlock(&ppr->pr_mtx);
1164 vfs_opterror(opts, "jail \"%s\" not found",
1165 prison_name(mypr, ppr));
1166 goto done_unlock_list;
1170 mtx_unlock(&ppr->pr_mtx);
1171 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1173 /* Find the next free jid. */
1176 if (jid == JAIL_MAX)
1178 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1179 if (tpr->pr_id < jid)
1181 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1182 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1185 if (jid == lastprid) {
1188 "no available jail IDs");
1190 prison_deref(ppr, PD_DEREF |
1191 PD_DEUREF | PD_LIST_XLOCKED);
1200 * The jail already has a jid (that did not yet exist),
1201 * so just find where to insert it.
1203 TAILQ_FOREACH(tpr, &allprison, pr_list)
1204 if (tpr->pr_id >= jid) {
1205 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1210 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1211 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1212 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1213 tpr->pr_childcount++;
1215 pr->pr_parent = ppr;
1218 /* Set some default values, and inherit some from the parent. */
1223 root = mypr->pr_root;
1226 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1227 pr->pr_flags |= PR_HOST;
1228 #if defined(INET) || defined(INET6)
1230 if (!(pr_flags & PR_VNET))
1234 if (!(ch_flags & PR_IP4_USER))
1235 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1236 else if (!(pr_flags & PR_IP4_USER)) {
1237 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1238 if (ppr->pr_ip4 != NULL) {
1239 pr->pr_ip4s = ppr->pr_ip4s;
1240 pr->pr_ip4 = malloc(pr->pr_ip4s *
1241 sizeof(struct in_addr), M_PRISON,
1243 bcopy(ppr->pr_ip4, pr->pr_ip4,
1244 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1249 if (!(ch_flags & PR_IP6_USER))
1250 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1251 else if (!(pr_flags & PR_IP6_USER)) {
1252 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1253 if (ppr->pr_ip6 != NULL) {
1254 pr->pr_ip6s = ppr->pr_ip6s;
1255 pr->pr_ip6 = malloc(pr->pr_ip6s *
1256 sizeof(struct in6_addr), M_PRISON,
1258 bcopy(ppr->pr_ip6, pr->pr_ip6,
1259 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1265 /* Source address selection is always on by default. */
1266 pr->pr_flags |= _PR_IP_SADDRSEL;
1268 pr->pr_securelevel = ppr->pr_securelevel;
1269 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1270 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1271 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1273 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1274 if (osrelstr == NULL)
1275 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1277 strcpy(pr->pr_osrelease, osrelstr);
1279 LIST_INIT(&pr->pr_children);
1280 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1281 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1284 /* Allocate a new vnet if specified. */
1285 pr->pr_vnet = (pr_flags & PR_VNET)
1286 ? vnet_alloc() : ppr->pr_vnet;
1289 * Allocate a dedicated cpuset for each jail.
1290 * Unlike other initial settings, this may return an erorr.
1292 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1294 prison_deref(pr, PD_LIST_XLOCKED);
1298 mtx_lock(&pr->pr_mtx);
1300 * New prisons do not yet have a reference, because we do not
1301 * want others to see the incomplete prison once the
1302 * allprison_lock is downgraded.
1307 * Grab a reference for existing prisons, to ensure they
1308 * continue to exist for the duration of the call.
1311 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1312 if ((pr->pr_flags & PR_VNET) &&
1313 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1316 "vnet jails cannot have IP address restrictions");
1317 goto done_deref_locked;
1321 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1324 "ip4 cannot be changed after creation");
1325 goto done_deref_locked;
1329 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1332 "ip6 cannot be changed after creation");
1333 goto done_deref_locked;
1338 /* Do final error checking before setting anything. */
1340 if (slevel < ppr->pr_securelevel) {
1342 goto done_deref_locked;
1346 if (childmax >= ppr->pr_childmax) {
1348 goto done_deref_locked;
1352 if (enforce < ppr->pr_enforce_statfs) {
1354 goto done_deref_locked;
1359 * devfs_rsnum is a uint16_t
1361 if (rsnum < 0 || rsnum > 65535) {
1363 goto done_deref_locked;
1366 * Nested jails always inherit parent's devfs ruleset
1368 if (jailed(td->td_ucred)) {
1369 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1371 goto done_deref_locked;
1373 rsnum = ppr->pr_devfs_rsnum;
1378 if (ppr->pr_flags & PR_IP4) {
1380 * Make sure the new set of IP addresses is a
1381 * subset of the parent's list. Don't worry
1382 * about the parent being unlocked, as any
1383 * setting is done with allprison_lock held.
1385 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1386 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1388 if (ij == ppr->pr_ip4s) {
1390 goto done_deref_locked;
1393 for (ii = ij = 1; ii < ip4s; ii++) {
1394 if (ip4[ii].s_addr ==
1395 ppr->pr_ip4[0].s_addr)
1397 for (; ij < ppr->pr_ip4s; ij++)
1398 if (ip4[ii].s_addr ==
1399 ppr->pr_ip4[ij].s_addr)
1401 if (ij == ppr->pr_ip4s)
1404 if (ij == ppr->pr_ip4s) {
1406 goto done_deref_locked;
1411 * Check for conflicting IP addresses. We permit them
1412 * if there is no more than one IP on each jail. If
1413 * there is a duplicate on a jail with more than one
1414 * IP stop checking and return error.
1417 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1418 if (tppr->pr_flags & PR_VNET)
1423 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1426 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1428 tpr->pr_uref == 0) {
1432 if (!(tpr->pr_flags & PR_IP4_USER))
1435 if (tpr->pr_ip4 == NULL ||
1436 (ip4s == 1 && tpr->pr_ip4s == 1))
1438 for (ii = 0; ii < ip4s; ii++) {
1439 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1443 "IPv4 addresses clash");
1444 goto done_deref_locked;
1452 if (ppr->pr_flags & PR_IP6) {
1454 * Make sure the new set of IP addresses is a
1455 * subset of the parent's list.
1457 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1458 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1461 if (ij == ppr->pr_ip6s) {
1463 goto done_deref_locked;
1466 for (ii = ij = 1; ii < ip6s; ii++) {
1467 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1470 for (; ij < ppr->pr_ip6s; ij++)
1471 if (IN6_ARE_ADDR_EQUAL(
1472 &ip6[ii], &ppr->pr_ip6[ij]))
1474 if (ij == ppr->pr_ip6s)
1477 if (ij == ppr->pr_ip6s) {
1479 goto done_deref_locked;
1483 /* Check for conflicting IP addresses. */
1485 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1486 if (tppr->pr_flags & PR_VNET)
1491 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1494 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1496 tpr->pr_uref == 0) {
1500 if (!(tpr->pr_flags & PR_IP6_USER))
1503 if (tpr->pr_ip6 == NULL ||
1504 (ip6s == 1 && tpr->pr_ip6s == 1))
1506 for (ii = 0; ii < ip6s; ii++) {
1507 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1511 "IPv6 addresses clash");
1512 goto done_deref_locked;
1518 onamelen = namelen = 0;
1519 if (namelc != NULL) {
1520 /* Give a default name of the jid. Also allow the name to be
1521 * explicitly the jid - but not any other number, and only in
1522 * normal form (no leading zero/etc).
1524 if (namelc[0] == '\0')
1525 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1526 else if ((strtoul(namelc, &p, 10) != jid ||
1527 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1530 "name cannot be numeric (unless it is the jid)");
1531 goto done_deref_locked;
1534 * Make sure the name isn't too long for the prison or its
1537 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1538 onamelen = strlen(pr->pr_name + pnamelen);
1539 namelen = strlen(namelc);
1540 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1541 error = ENAMETOOLONG;
1542 goto done_deref_locked;
1544 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1545 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1546 sizeof(pr->pr_name)) {
1547 error = ENAMETOOLONG;
1548 goto done_deref_locked;
1552 if (pr_allow & ~ppr->pr_allow) {
1554 goto done_deref_locked;
1558 * Let modules check their parameters. This requires unlocking and
1559 * then re-locking the prison, but this is still a valid state as long
1560 * as allprison_lock remains xlocked.
1562 mtx_unlock(&pr->pr_mtx);
1563 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1565 prison_deref(pr, created
1567 : PD_DEREF | PD_LIST_XLOCKED);
1570 mtx_lock(&pr->pr_mtx);
1572 /* At this point, all valid parameters should have been noted. */
1573 TAILQ_FOREACH(opt, opts, link) {
1574 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1576 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1577 goto done_deref_locked;
1581 /* Set the parameters of the prison. */
1584 if (pr_flags & PR_IP4_USER) {
1585 pr->pr_flags |= PR_IP4;
1586 free(pr->pr_ip4, M_PRISON);
1590 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1592 if (tpr->pr_flags & PR_VNET) {
1597 if (prison_restrict_ip4(tpr, NULL)) {
1606 if (pr_flags & PR_IP6_USER) {
1607 pr->pr_flags |= PR_IP6;
1608 free(pr->pr_ip6, M_PRISON);
1612 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1614 if (tpr->pr_flags & PR_VNET) {
1619 if (prison_restrict_ip6(tpr, NULL)) {
1627 pr->pr_securelevel = slevel;
1628 /* Set all child jails to be at least this level. */
1629 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1630 if (tpr->pr_securelevel < slevel)
1631 tpr->pr_securelevel = slevel;
1634 pr->pr_childmax = childmax;
1635 /* Set all child jails to under this limit. */
1636 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1637 if (tpr->pr_childmax > childmax - level)
1638 tpr->pr_childmax = childmax > level
1639 ? childmax - level : 0;
1642 pr->pr_enforce_statfs = enforce;
1643 /* Pass this restriction on to the children. */
1644 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1645 if (tpr->pr_enforce_statfs < enforce)
1646 tpr->pr_enforce_statfs = enforce;
1649 pr->pr_devfs_rsnum = rsnum;
1650 /* Pass this restriction on to the children. */
1651 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1652 tpr->pr_devfs_rsnum = rsnum;
1654 if (namelc != NULL) {
1655 if (ppr == &prison0)
1656 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1658 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1659 ppr->pr_name, namelc);
1660 /* Change this component of child names. */
1661 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1662 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1663 strlen(tpr->pr_name + onamelen) + 1);
1664 bcopy(pr->pr_name, tpr->pr_name, namelen);
1668 /* Try to keep a real-rooted full pathname. */
1669 if (fullpath_disabled && path[0] == '/' &&
1670 strcmp(mypr->pr_path, "/"))
1671 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1672 mypr->pr_path, path);
1674 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1677 if (PR_HOST & ch_flags & ~pr_flags) {
1678 if (pr->pr_flags & PR_HOST) {
1680 * Copy the parent's host info. As with pr_ip4 above,
1681 * the lack of a lock on the parent is not a problem;
1682 * it is always set with allprison_lock at least
1683 * shared, and is held exclusively here.
1685 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1686 sizeof(pr->pr_hostname));
1687 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1688 sizeof(pr->pr_domainname));
1689 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1690 sizeof(pr->pr_hostuuid));
1691 pr->pr_hostid = pr->pr_parent->pr_hostid;
1693 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1694 /* Set this prison, and any descendants without PR_HOST. */
1696 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1698 strlcpy(pr->pr_domainname, domain,
1699 sizeof(pr->pr_domainname));
1701 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1703 pr->pr_hostid = hid;
1704 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1705 if (tpr->pr_flags & PR_HOST)
1709 strlcpy(tpr->pr_hostname,
1711 sizeof(tpr->pr_hostname));
1713 strlcpy(tpr->pr_domainname,
1715 sizeof(tpr->pr_domainname));
1717 strlcpy(tpr->pr_hostuuid,
1719 sizeof(tpr->pr_hostuuid));
1721 tpr->pr_hostid = hid;
1725 if ((tallow = ch_allow & ~pr_allow)) {
1726 /* Clear allow bits in all children. */
1727 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1728 tpr->pr_allow &= ~tallow;
1730 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1732 * Persistent prisons get an extra reference, and prisons losing their
1733 * persist flag lose that reference. Only do this for existing prisons
1734 * for now, so new ones will remain unseen until after the module
1735 * handlers have completed.
1737 born = pr->pr_uref == 0;
1738 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1739 if (pr_flags & PR_PERSIST) {
1747 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1748 mtx_unlock(&pr->pr_mtx);
1751 if (racct_enable && created)
1752 prison_racct_attach(pr);
1755 /* Locks may have prevented a complete restriction of child IP
1756 * addresses. If so, allocate some more memory and try again.
1761 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1762 mtx_lock(&pr->pr_mtx);
1764 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1766 if (tpr->pr_flags & PR_VNET) {
1771 if (prison_restrict_ip4(tpr, ip4)) {
1778 mtx_unlock(&pr->pr_mtx);
1784 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1785 mtx_lock(&pr->pr_mtx);
1787 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1789 if (tpr->pr_flags & PR_VNET) {
1794 if (prison_restrict_ip6(tpr, ip6)) {
1801 mtx_unlock(&pr->pr_mtx);
1805 /* Let the modules do their work. */
1806 sx_downgrade(&allprison_lock);
1808 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1810 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1811 prison_deref(pr, created
1813 : PD_DEREF | PD_LIST_SLOCKED);
1817 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1820 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1821 prison_deref(pr, created
1823 : PD_DEREF | PD_LIST_SLOCKED);
1827 /* Attach this process to the prison if requested. */
1828 if (flags & JAIL_ATTACH) {
1829 mtx_lock(&pr->pr_mtx);
1830 error = do_jail_attach(td, pr);
1832 vfs_opterror(opts, "attach failed");
1834 prison_deref(pr, PD_DEREF);
1840 if (racct_enable && !created) {
1841 if (!(flags & JAIL_ATTACH))
1842 sx_sunlock(&allprison_lock);
1843 prison_racct_modify(pr);
1844 if (!(flags & JAIL_ATTACH))
1845 sx_slock(&allprison_lock);
1849 td->td_retval[0] = pr->pr_id;
1852 * Now that it is all there, drop the temporary reference from existing
1853 * prisons. Or add a reference to newly created persistent prisons
1854 * (which was not done earlier so that the prison would not be publicly
1858 prison_deref(pr, (flags & JAIL_ATTACH)
1860 : PD_DEREF | PD_LIST_SLOCKED);
1862 if (pr_flags & PR_PERSIST) {
1863 mtx_lock(&pr->pr_mtx);
1866 mtx_unlock(&pr->pr_mtx);
1868 if (!(flags & JAIL_ATTACH))
1869 sx_sunlock(&allprison_lock);
1875 prison_deref(pr, created
1876 ? PD_LOCKED | PD_LIST_XLOCKED
1877 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1880 sx_xunlock(&allprison_lock);
1886 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1887 &errmsg_len) == 0 && errmsg_len > 0) {
1888 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1889 if (optuio->uio_segflg == UIO_SYSSPACE)
1891 optuio->uio_iov[errmsg_pos].iov_base,
1895 optuio->uio_iov[errmsg_pos].iov_base,
1901 free(ip4, M_PRISON);
1904 free(ip6, M_PRISON);
1907 free(g_path, M_TEMP);
1914 * struct jail_get_args {
1915 * struct iovec *iovp;
1916 * unsigned int iovcnt;
1921 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1926 /* Check that we have an even number of iovecs. */
1927 if (uap->iovcnt & 1)
1930 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1933 error = kern_jail_get(td, auio, uap->flags);
1935 error = copyout(auio->uio_iov, uap->iovp,
1936 uap->iovcnt * sizeof (struct iovec));
1942 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1944 struct prison *pr, *mypr;
1946 struct vfsoptlist *opts;
1947 char *errmsg, *name;
1948 int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1950 if (flags & ~JAIL_GET_MASK)
1953 /* Get the parameter list. */
1954 error = vfs_buildopts(optuio, &opts);
1957 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1958 mypr = td->td_ucred->cr_prison;
1961 * Find the prison specified by one of: lastjid, jid, name.
1963 sx_slock(&allprison_lock);
1964 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1966 TAILQ_FOREACH(pr, &allprison, pr_list) {
1967 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1968 mtx_lock(&pr->pr_mtx);
1969 if (pr->pr_ref > 0 &&
1970 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1972 mtx_unlock(&pr->pr_mtx);
1978 vfs_opterror(opts, "no jail after %d", jid);
1979 goto done_unlock_list;
1980 } else if (error != ENOENT)
1981 goto done_unlock_list;
1983 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1986 pr = prison_find_child(mypr, jid);
1988 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1989 mtx_unlock(&pr->pr_mtx);
1991 vfs_opterror(opts, "jail %d is dying",
1993 goto done_unlock_list;
1998 vfs_opterror(opts, "jail %d not found", jid);
1999 goto done_unlock_list;
2001 } else if (error != ENOENT)
2002 goto done_unlock_list;
2004 error = vfs_getopt(opts, "name", (void **)&name, &len);
2006 if (len == 0 || name[len - 1] != '\0') {
2008 goto done_unlock_list;
2010 pr = prison_find_name(mypr, name);
2012 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2013 mtx_unlock(&pr->pr_mtx);
2015 vfs_opterror(opts, "jail \"%s\" is dying",
2017 goto done_unlock_list;
2022 vfs_opterror(opts, "jail \"%s\" not found", name);
2023 goto done_unlock_list;
2024 } else if (error != ENOENT)
2025 goto done_unlock_list;
2027 vfs_opterror(opts, "no jail specified");
2029 goto done_unlock_list;
2032 /* Get the parameters of the prison. */
2035 td->td_retval[0] = pr->pr_id;
2036 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2037 if (error != 0 && error != ENOENT)
2039 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2040 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2041 if (error != 0 && error != ENOENT)
2043 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2044 if (error != 0 && error != ENOENT)
2046 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2047 sizeof(pr->pr_cpuset->cs_id));
2048 if (error != 0 && error != ENOENT)
2050 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2051 if (error != 0 && error != ENOENT)
2054 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2055 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2056 if (error != 0 && error != ENOENT)
2060 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2061 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2062 if (error != 0 && error != ENOENT)
2065 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2066 sizeof(pr->pr_securelevel));
2067 if (error != 0 && error != ENOENT)
2069 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2070 sizeof(pr->pr_childcount));
2071 if (error != 0 && error != ENOENT)
2073 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2074 sizeof(pr->pr_childmax));
2075 if (error != 0 && error != ENOENT)
2077 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2078 if (error != 0 && error != ENOENT)
2080 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2081 if (error != 0 && error != ENOENT)
2083 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2084 if (error != 0 && error != ENOENT)
2086 #ifdef COMPAT_FREEBSD32
2087 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2088 uint32_t hid32 = pr->pr_hostid;
2090 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2093 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2094 sizeof(pr->pr_hostid));
2095 if (error != 0 && error != ENOENT)
2097 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2098 sizeof(pr->pr_enforce_statfs));
2099 if (error != 0 && error != ENOENT)
2101 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2102 sizeof(pr->pr_devfs_rsnum));
2103 if (error != 0 && error != ENOENT)
2105 for (fi = 0; fi < nitems(pr_flag_names); fi++) {
2106 if (pr_flag_names[fi] == NULL)
2108 i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2109 error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2110 if (error != 0 && error != ENOENT)
2113 error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2114 if (error != 0 && error != ENOENT)
2117 for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
2119 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2120 i = pr_flag_jailsys[fi].disable &&
2121 (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2122 : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2125 vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2126 if (error != 0 && error != ENOENT)
2129 for (fi = 0; fi < nitems(pr_allow_names); fi++) {
2130 if (pr_allow_names[fi] == NULL)
2132 i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2133 error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2134 if (error != 0 && error != ENOENT)
2137 error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2138 if (error != 0 && error != ENOENT)
2141 i = (pr->pr_uref == 0);
2142 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2143 if (error != 0 && error != ENOENT)
2146 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2147 if (error != 0 && error != ENOENT)
2149 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2150 sizeof(pr->pr_osreldate));
2151 if (error != 0 && error != ENOENT)
2153 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2154 if (error != 0 && error != ENOENT)
2157 /* Get the module parameters. */
2158 mtx_unlock(&pr->pr_mtx);
2160 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2163 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2165 /* By now, all parameters should have been noted. */
2166 TAILQ_FOREACH(opt, opts, link) {
2167 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2169 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2174 /* Write the fetched parameters back to userspace. */
2176 TAILQ_FOREACH(opt, opts, link) {
2177 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2178 pos = 2 * opt->pos + 1;
2179 optuio->uio_iov[pos].iov_len = opt->len;
2180 if (opt->value != NULL) {
2181 if (optuio->uio_segflg == UIO_SYSSPACE) {
2183 optuio->uio_iov[pos].iov_base,
2186 error = copyout(opt->value,
2187 optuio->uio_iov[pos].iov_base,
2198 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2202 sx_sunlock(&allprison_lock);
2204 if (error && errmsg_pos >= 0) {
2205 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2206 errmsg_pos = 2 * errmsg_pos + 1;
2207 if (errmsg_len > 0) {
2208 if (optuio->uio_segflg == UIO_SYSSPACE)
2210 optuio->uio_iov[errmsg_pos].iov_base,
2214 optuio->uio_iov[errmsg_pos].iov_base,
2224 * struct jail_remove_args {
2229 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2231 struct prison *pr, *cpr, *lpr, *tpr;
2234 error = priv_check(td, PRIV_JAIL_REMOVE);
2238 sx_xlock(&allprison_lock);
2239 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2241 sx_xunlock(&allprison_lock);
2245 /* Remove all descendants of this prison, then remove this prison. */
2247 if (!LIST_EMPTY(&pr->pr_children)) {
2248 mtx_unlock(&pr->pr_mtx);
2250 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2251 mtx_lock(&cpr->pr_mtx);
2252 if (cpr->pr_ref > 0) {
2256 /* Already removed - do not do it again. */
2259 mtx_unlock(&cpr->pr_mtx);
2261 mtx_lock(&lpr->pr_mtx);
2262 prison_remove_one(lpr);
2263 sx_xlock(&allprison_lock);
2268 mtx_lock(&lpr->pr_mtx);
2269 prison_remove_one(lpr);
2270 sx_xlock(&allprison_lock);
2272 mtx_lock(&pr->pr_mtx);
2274 prison_remove_one(pr);
2279 prison_remove_one(struct prison *pr)
2284 /* If the prison was persistent, it is not anymore. */
2286 if (pr->pr_flags & PR_PERSIST) {
2289 pr->pr_flags &= ~PR_PERSIST;
2293 * jail_remove added a reference. If that's the only one, remove
2296 KASSERT(pr->pr_ref > 0,
2297 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2298 if (pr->pr_ref == 1) {
2300 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2304 mtx_unlock(&pr->pr_mtx);
2305 sx_xunlock(&allprison_lock);
2307 * Kill all processes unfortunate enough to be attached to this prison.
2309 sx_slock(&allproc_lock);
2310 LIST_FOREACH(p, &allproc, p_list) {
2312 if (p->p_state != PRS_NEW && p->p_ucred &&
2313 p->p_ucred->cr_prison == pr)
2314 kern_psignal(p, SIGKILL);
2317 sx_sunlock(&allproc_lock);
2318 /* Remove the temporary reference added by jail_remove. */
2319 prison_deref(pr, deuref | PD_DEREF);
2324 * struct jail_attach_args {
2329 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2334 error = priv_check(td, PRIV_JAIL_ATTACH);
2339 * Start with exclusive hold on allprison_lock to ensure that a possible
2340 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2341 * But then immediately downgrade it since we don't need to stop
2344 sx_xlock(&allprison_lock);
2345 sx_downgrade(&allprison_lock);
2346 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2348 sx_sunlock(&allprison_lock);
2353 * Do not allow a process to attach to a prison that is not
2354 * considered to be "alive".
2356 if (pr->pr_uref == 0) {
2357 mtx_unlock(&pr->pr_mtx);
2358 sx_sunlock(&allprison_lock);
2362 return (do_jail_attach(td, pr));
2366 do_jail_attach(struct thread *td, struct prison *pr)
2369 struct ucred *newcred, *oldcred;
2373 * XXX: Note that there is a slight race here if two threads
2374 * in the same privileged process attempt to attach to two
2375 * different jails at the same time. It is important for
2376 * user processes not to do this, or they might end up with
2377 * a process root from one prison, but attached to the jail
2382 mtx_unlock(&pr->pr_mtx);
2384 /* Let modules do whatever they need to prepare for attaching. */
2385 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2387 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2390 sx_sunlock(&allprison_lock);
2393 * Reparent the newly attached process to this jail.
2396 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2400 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2401 if ((error = change_dir(pr->pr_root, td)) != 0)
2404 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2407 VOP_UNLOCK(pr->pr_root, 0);
2408 if ((error = pwd_chroot(td, pr->pr_root)))
2413 oldcred = crcopysafe(p, newcred);
2414 newcred->cr_prison = pr;
2415 proc_set_cred(p, newcred);
2418 racct_proc_ucred_changed(p, oldcred, newcred);
2423 rctl_proc_ucred_changed(p, newcred);
2426 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2431 VOP_UNLOCK(pr->pr_root, 0);
2433 /* Tell modules this thread is still in its old jail after all. */
2434 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2435 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 (jid=%d).", 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, struct sockaddr *sa)
2799 struct sockaddr_in *sai;
2802 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 = (struct sockaddr_in *)sa;
2820 error = prison_check_ip4(cred, &sai->sin_addr);
2825 sai6 = (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, otherwise 0.
2864 jailed(struct ucred *cred)
2867 return (cred->cr_prison != &prison0);
2871 * Return 1 if the passed credential is in a jail and that jail does not
2872 * have its own virtual network stack, otherwise 0.
2875 jailed_without_vnet(struct ucred *cred)
2881 if (prison_owns_vnet(cred))
2889 * Return the correct hostname (domainname, et al) for the passed credential.
2892 getcredhostname(struct ucred *cred, char *buf, size_t size)
2897 * A NULL credential can be used to shortcut to the physical
2898 * system's hostname.
2900 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2901 mtx_lock(&pr->pr_mtx);
2902 strlcpy(buf, pr->pr_hostname, size);
2903 mtx_unlock(&pr->pr_mtx);
2907 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2910 mtx_lock(&cred->cr_prison->pr_mtx);
2911 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2912 mtx_unlock(&cred->cr_prison->pr_mtx);
2916 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2919 mtx_lock(&cred->cr_prison->pr_mtx);
2920 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2921 mtx_unlock(&cred->cr_prison->pr_mtx);
2925 getcredhostid(struct ucred *cred, unsigned long *hostid)
2928 mtx_lock(&cred->cr_prison->pr_mtx);
2929 *hostid = cred->cr_prison->pr_hostid;
2930 mtx_unlock(&cred->cr_prison->pr_mtx);
2935 * Determine whether the prison represented by cred owns
2936 * its vnet rather than having it inherited.
2938 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2941 prison_owns_vnet(struct ucred *cred)
2945 * vnets cannot be added/removed after jail creation,
2946 * so no need to lock here.
2948 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2953 * Determine whether the subject represented by cred can "see"
2954 * status of a mount point.
2955 * Returns: 0 for permitted, ENOENT otherwise.
2956 * XXX: This function should be called cr_canseemount() and should be
2957 * placed in kern_prot.c.
2960 prison_canseemount(struct ucred *cred, struct mount *mp)
2966 pr = cred->cr_prison;
2967 if (pr->pr_enforce_statfs == 0)
2969 if (pr->pr_root->v_mount == mp)
2971 if (pr->pr_enforce_statfs == 2)
2974 * If jail's chroot directory is set to "/" we should be able to see
2975 * all mount-points from inside a jail.
2976 * This is ugly check, but this is the only situation when jail's
2977 * directory ends with '/'.
2979 if (strcmp(pr->pr_path, "/") == 0)
2981 len = strlen(pr->pr_path);
2983 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2986 * Be sure that we don't have situation where jail's root directory
2987 * is "/some/path" and mount point is "/some/pathpath".
2989 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2995 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2997 char jpath[MAXPATHLEN];
3001 pr = cred->cr_prison;
3002 if (pr->pr_enforce_statfs == 0)
3004 if (prison_canseemount(cred, mp) != 0) {
3005 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3006 strlcpy(sp->f_mntonname, "[restricted]",
3007 sizeof(sp->f_mntonname));
3010 if (pr->pr_root->v_mount == mp) {
3012 * Clear current buffer data, so we are sure nothing from
3013 * the valid path left there.
3015 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3016 *sp->f_mntonname = '/';
3020 * If jail's chroot directory is set to "/" we should be able to see
3021 * all mount-points from inside a jail.
3023 if (strcmp(pr->pr_path, "/") == 0)
3025 len = strlen(pr->pr_path);
3026 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3028 * Clear current buffer data, so we are sure nothing from
3029 * the valid path left there.
3031 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3032 if (*jpath == '\0') {
3033 /* Should never happen. */
3034 *sp->f_mntonname = '/';
3036 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3041 * Check with permission for a specific privilege is granted within jail. We
3042 * have a specific list of accepted privileges; the rest are denied.
3045 prison_priv_check(struct ucred *cred, int priv)
3053 * Privileges specific to prisons with a virtual network stack.
3054 * There might be a duplicate entry here in case the privilege
3055 * is only granted conditionally in the legacy jail case.
3060 * NFS-specific privileges.
3062 case PRIV_NFS_DAEMON:
3063 case PRIV_NFS_LOCKD:
3066 * Network stack privileges.
3068 case PRIV_NET_BRIDGE:
3071 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3072 case PRIV_NET_ROUTE:
3074 case PRIV_NET_SETIFMTU:
3075 case PRIV_NET_SETIFFLAGS:
3076 case PRIV_NET_SETIFCAP:
3077 case PRIV_NET_SETIFDESCR:
3078 case PRIV_NET_SETIFNAME :
3079 case PRIV_NET_SETIFMETRIC:
3080 case PRIV_NET_SETIFPHYS:
3081 case PRIV_NET_SETIFMAC:
3082 case PRIV_NET_ADDMULTI:
3083 case PRIV_NET_DELMULTI:
3084 case PRIV_NET_HWIOCTL:
3085 case PRIV_NET_SETLLADDR:
3086 case PRIV_NET_ADDIFGROUP:
3087 case PRIV_NET_DELIFGROUP:
3088 case PRIV_NET_IFCREATE:
3089 case PRIV_NET_IFDESTROY:
3090 case PRIV_NET_ADDIFADDR:
3091 case PRIV_NET_DELIFADDR:
3094 case PRIV_NET_SETIFVNET:
3095 case PRIV_NET_SETIFFIB:
3098 * 802.11-related privileges.
3100 case PRIV_NET80211_GETKEY:
3102 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3109 case PRIV_NETATM_CFG:
3110 case PRIV_NETATM_ADD:
3111 case PRIV_NETATM_DEL:
3112 case PRIV_NETATM_SET:
3115 * Bluetooth privileges.
3117 case PRIV_NETBLUETOOTH_RAW:
3121 * Netgraph and netgraph module privileges.
3123 case PRIV_NETGRAPH_CONTROL:
3125 case PRIV_NETGRAPH_TTY:
3129 * IPv4 and IPv6 privileges.
3131 case PRIV_NETINET_IPFW:
3132 case PRIV_NETINET_DIVERT:
3133 case PRIV_NETINET_PF:
3134 case PRIV_NETINET_DUMMYNET:
3135 case PRIV_NETINET_CARP:
3136 case PRIV_NETINET_MROUTE:
3137 case PRIV_NETINET_RAW:
3138 case PRIV_NETINET_ADDRCTRL6:
3139 case PRIV_NETINET_ND6:
3140 case PRIV_NETINET_SCOPE6:
3141 case PRIV_NETINET_ALIFETIME6:
3142 case PRIV_NETINET_IPSEC:
3143 case PRIV_NETINET_BINDANY:
3158 * No default: or deny here.
3159 * In case of no permit fall through to next switch().
3161 if (cred->cr_prison->pr_flags & PR_VNET)
3169 * Allow ktrace privileges for root in jail.
3175 * Allow jailed processes to configure audit identity and
3176 * submit audit records (login, etc). In the future we may
3177 * want to further refine the relationship between audit and
3180 case PRIV_AUDIT_GETAUDIT:
3181 case PRIV_AUDIT_SETAUDIT:
3182 case PRIV_AUDIT_SUBMIT:
3186 * Allow jailed processes to manipulate process UNIX
3187 * credentials in any way they see fit.
3189 case PRIV_CRED_SETUID:
3190 case PRIV_CRED_SETEUID:
3191 case PRIV_CRED_SETGID:
3192 case PRIV_CRED_SETEGID:
3193 case PRIV_CRED_SETGROUPS:
3194 case PRIV_CRED_SETREUID:
3195 case PRIV_CRED_SETREGID:
3196 case PRIV_CRED_SETRESUID:
3197 case PRIV_CRED_SETRESGID:
3200 * Jail implements visibility constraints already, so allow
3201 * jailed root to override uid/gid-based constraints.
3203 case PRIV_SEEOTHERGIDS:
3204 case PRIV_SEEOTHERUIDS:
3207 * Jail implements inter-process debugging limits already, so
3208 * allow jailed root various debugging privileges.
3210 case PRIV_DEBUG_DIFFCRED:
3211 case PRIV_DEBUG_SUGID:
3212 case PRIV_DEBUG_UNPRIV:
3215 * Allow jail to set various resource limits and login
3216 * properties, and for now, exceed process resource limits.
3218 case PRIV_PROC_LIMIT:
3219 case PRIV_PROC_SETLOGIN:
3220 case PRIV_PROC_SETRLIMIT:
3223 * System V and POSIX IPC privileges are granted in jail.
3226 case PRIV_IPC_WRITE:
3227 case PRIV_IPC_ADMIN:
3228 case PRIV_IPC_MSGSIZE:
3232 * Jail operations within a jail work on child jails.
3234 case PRIV_JAIL_ATTACH:
3236 case PRIV_JAIL_REMOVE:
3239 * Jail implements its own inter-process limits, so allow
3240 * root processes in jail to change scheduling on other
3241 * processes in the same jail. Likewise for signalling.
3243 case PRIV_SCHED_DIFFCRED:
3244 case PRIV_SCHED_CPUSET:
3245 case PRIV_SIGNAL_DIFFCRED:
3246 case PRIV_SIGNAL_SUGID:
3249 * Allow jailed processes to write to sysctls marked as jail
3252 case PRIV_SYSCTL_WRITEJAIL:
3255 * Allow root in jail to manage a variety of quota
3256 * properties. These should likely be conditional on a
3257 * configuration option.
3259 case PRIV_VFS_GETQUOTA:
3260 case PRIV_VFS_SETQUOTA:
3263 * Since Jail relies on chroot() to implement file system
3264 * protections, grant many VFS privileges to root in jail.
3265 * Be careful to exclude mount-related and NFS-related
3269 case PRIV_VFS_WRITE:
3270 case PRIV_VFS_ADMIN:
3272 case PRIV_VFS_LOOKUP:
3273 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3274 case PRIV_VFS_CHFLAGS_DEV:
3275 case PRIV_VFS_CHOWN:
3276 case PRIV_VFS_CHROOT:
3277 case PRIV_VFS_RETAINSUGID:
3278 case PRIV_VFS_FCHROOT:
3280 case PRIV_VFS_SETGID:
3282 case PRIV_VFS_STICKYFILE:
3285 * As in the non-jail case, non-root users are expected to be
3286 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3287 * exists in the jail and they have permission to access it).
3289 case PRIV_KMEM_READ:
3293 * Depending on the global setting, allow privilege of
3294 * setting system flags.
3296 case PRIV_VFS_SYSFLAGS:
3297 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3303 * Depending on the global setting, allow privilege of
3304 * mounting/unmounting file systems.
3306 case PRIV_VFS_MOUNT:
3307 case PRIV_VFS_UNMOUNT:
3308 case PRIV_VFS_MOUNT_NONUSER:
3309 case PRIV_VFS_MOUNT_OWNER:
3310 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3311 cred->cr_prison->pr_enforce_statfs < 2)
3317 * Allow jailed root to bind reserved ports and reuse in-use
3320 case PRIV_NETINET_RESERVEDPORT:
3321 case PRIV_NETINET_REUSEPORT:
3325 * Allow jailed root to set certain IPv4/6 (option) headers.
3327 case PRIV_NETINET_SETHDROPTS:
3331 * Conditionally allow creating raw sockets in jail.
3333 case PRIV_NETINET_RAW:
3334 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3340 * Since jail implements its own visibility limits on netstat
3341 * sysctls, allow getcred. This allows identd to work in
3344 case PRIV_NETINET_GETCRED:
3348 * Allow jailed root to set loginclass.
3350 case PRIV_PROC_SETLOGINCLASS:
3354 * Do not allow a process inside a jail to read the kernel
3355 * message buffer unless explicitly permitted.
3358 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3364 * In all remaining cases, deny the privilege request. This
3365 * includes almost all network privileges, many system
3366 * configuration privileges.
3373 * Return the part of pr2's name that is relative to pr1, or the whole name
3374 * if it does not directly follow.
3378 prison_name(struct prison *pr1, struct prison *pr2)
3382 /* Jails see themselves as "0" (if they see themselves at all). */
3385 name = pr2->pr_name;
3386 if (prison_ischild(pr1, pr2)) {
3388 * pr1 isn't locked (and allprison_lock may not be either)
3389 * so its length can't be counted on. But the number of dots
3390 * can be counted on - and counted.
3392 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3393 name = strchr(name, '.') + 1;
3399 * Return the part of pr2's path that is relative to pr1, or the whole path
3400 * if it does not directly follow.
3403 prison_path(struct prison *pr1, struct prison *pr2)
3405 char *path1, *path2;
3408 path1 = pr1->pr_path;
3409 path2 = pr2->pr_path;
3410 if (!strcmp(path1, "/"))
3412 len1 = strlen(path1);
3413 if (strncmp(path1, path2, len1))
3415 if (path2[len1] == '\0')
3417 if (path2[len1] == '/')
3418 return (path2 + len1);
3424 * Jail-related sysctls.
3426 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3430 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3433 struct prison *pr, *cpr;
3435 struct in_addr *ip4 = NULL;
3439 struct in6_addr *ip6 = NULL;
3444 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3445 pr = req->td->td_ucred->cr_prison;
3447 sx_slock(&allprison_lock);
3448 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3449 #if defined(INET) || defined(INET6)
3452 mtx_lock(&cpr->pr_mtx);
3454 if (cpr->pr_ip4s > 0) {
3455 if (ip4s < cpr->pr_ip4s) {
3456 ip4s = cpr->pr_ip4s;
3457 mtx_unlock(&cpr->pr_mtx);
3458 ip4 = realloc(ip4, ip4s *
3459 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3462 bcopy(cpr->pr_ip4, ip4,
3463 cpr->pr_ip4s * sizeof(struct in_addr));
3467 if (cpr->pr_ip6s > 0) {
3468 if (ip6s < cpr->pr_ip6s) {
3469 ip6s = cpr->pr_ip6s;
3470 mtx_unlock(&cpr->pr_mtx);
3471 ip6 = realloc(ip6, ip6s *
3472 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3475 bcopy(cpr->pr_ip6, ip6,
3476 cpr->pr_ip6s * sizeof(struct in6_addr));
3479 if (cpr->pr_ref == 0) {
3480 mtx_unlock(&cpr->pr_mtx);
3483 bzero(xp, sizeof(*xp));
3484 xp->pr_version = XPRISON_VERSION;
3485 xp->pr_id = cpr->pr_id;
3486 xp->pr_state = cpr->pr_uref > 0
3487 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3488 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3489 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3490 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3492 xp->pr_ip4s = cpr->pr_ip4s;
3495 xp->pr_ip6s = cpr->pr_ip6s;
3497 mtx_unlock(&cpr->pr_mtx);
3498 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3502 if (xp->pr_ip4s > 0) {
3503 error = SYSCTL_OUT(req, ip4,
3504 xp->pr_ip4s * sizeof(struct in_addr));
3510 if (xp->pr_ip6s > 0) {
3511 error = SYSCTL_OUT(req, ip6,
3512 xp->pr_ip6s * sizeof(struct in6_addr));
3518 sx_sunlock(&allprison_lock);
3529 SYSCTL_OID(_security_jail, OID_AUTO, list,
3530 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3531 sysctl_jail_list, "S", "List of active jails");
3534 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3538 injail = jailed(req->td->td_ucred);
3539 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3544 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3545 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3546 sysctl_jail_jailed, "I", "Process in jail?");
3549 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3551 int error, havevnet;
3553 struct ucred *cred = req->td->td_ucred;
3555 havevnet = jailed(cred) && prison_owns_vnet(cred);
3559 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3564 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3565 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3566 sysctl_jail_vnet, "I", "Jail owns VNET?");
3568 #if defined(INET) || defined(INET6)
3569 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3570 &jail_max_af_ips, 0,
3571 "Number of IP addresses a jail may have at most per address family (deprecated)");
3575 * Default parameters for jail(2) compatibility. For historical reasons,
3576 * the sysctl names have varying similarity to the parameter names. Prisons
3577 * just see their own parameters, and can't change them.
3580 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3583 int allow, error, i;
3585 pr = req->td->td_ucred->cr_prison;
3586 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3588 /* Get the current flag value, and convert it to a boolean. */
3589 i = (allow & arg2) ? 1 : 0;
3592 error = sysctl_handle_int(oidp, &i, 0, req);
3593 if (error || !req->newptr)
3599 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3602 mtx_lock(&prison0.pr_mtx);
3603 jail_default_allow = (jail_default_allow & ~arg2) | i;
3604 mtx_unlock(&prison0.pr_mtx);
3608 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3609 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3610 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3611 "Processes in jail can set their hostnames (deprecated)");
3612 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3613 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3614 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3615 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3616 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3617 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3618 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3619 "Processes in jail can use System V IPC primitives (deprecated)");
3620 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3621 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3622 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3623 "Prison root can create raw sockets (deprecated)");
3624 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3625 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3626 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3627 "Processes in jail can alter system file flags (deprecated)");
3628 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3629 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3630 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3631 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3632 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
3633 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3634 NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
3635 "Processes in jail can mount the devfs file system (deprecated)");
3636 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
3637 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3638 NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
3639 "Processes in jail can mount the fdescfs file system (deprecated)");
3640 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
3641 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3642 NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
3643 "Processes in jail can mount the nullfs file system (deprecated)");
3644 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
3645 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3646 NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
3647 "Processes in jail can mount the procfs file system (deprecated)");
3648 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
3649 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3650 NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
3651 "Processes in jail can mount the linprocfs file system (deprecated)");
3652 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
3653 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3654 NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
3655 "Processes in jail can mount the linsysfs file system (deprecated)");
3656 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
3657 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3658 NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
3659 "Processes in jail can mount the tmpfs file system (deprecated)");
3660 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
3661 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3662 NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
3663 "Processes in jail can mount the zfs file system (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, 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, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3811 "B", "Jail may read the kernel message buffer");
3813 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3814 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3815 "B", "Jail may mount/unmount jail-friendly file systems in general");
3816 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
3817 "B", "Jail may mount the devfs file system");
3818 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
3819 "B", "Jail may mount the fdescfs file system");
3820 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
3821 "B", "Jail may mount the nullfs file system");
3822 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
3823 "B", "Jail may mount the procfs file system");
3824 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
3825 "B", "Jail may mount the linprocfs file system");
3826 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
3827 "B", "Jail may mount the linsysfs file system");
3828 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
3829 "B", "Jail may mount the tmpfs file system");
3830 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
3831 "B", "Jail may mount the zfs file system");
3835 prison_racct_foreach(void (*callback)(struct racct *racct,
3836 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3837 void *arg2, void *arg3)
3839 struct prison_racct *prr;
3841 ASSERT_RACCT_ENABLED();
3843 sx_slock(&allprison_lock);
3846 LIST_FOREACH(prr, &allprison_racct, prr_next)
3847 (callback)(prr->prr_racct, arg2, arg3);
3850 sx_sunlock(&allprison_lock);
3853 static struct prison_racct *
3854 prison_racct_find_locked(const char *name)
3856 struct prison_racct *prr;
3858 ASSERT_RACCT_ENABLED();
3859 sx_assert(&allprison_lock, SA_XLOCKED);
3861 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3864 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3865 if (strcmp(name, prr->prr_name) != 0)
3868 /* Found prison_racct with a matching name? */
3869 prison_racct_hold(prr);
3873 /* Add new prison_racct. */
3874 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3875 racct_create(&prr->prr_racct);
3877 strcpy(prr->prr_name, name);
3878 refcount_init(&prr->prr_refcount, 1);
3879 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3884 struct prison_racct *
3885 prison_racct_find(const char *name)
3887 struct prison_racct *prr;
3889 ASSERT_RACCT_ENABLED();
3891 sx_xlock(&allprison_lock);
3892 prr = prison_racct_find_locked(name);
3893 sx_xunlock(&allprison_lock);
3898 prison_racct_hold(struct prison_racct *prr)
3901 ASSERT_RACCT_ENABLED();
3903 refcount_acquire(&prr->prr_refcount);
3907 prison_racct_free_locked(struct prison_racct *prr)
3910 ASSERT_RACCT_ENABLED();
3911 sx_assert(&allprison_lock, SA_XLOCKED);
3913 if (refcount_release(&prr->prr_refcount)) {
3914 racct_destroy(&prr->prr_racct);
3915 LIST_REMOVE(prr, prr_next);
3916 free(prr, M_PRISON_RACCT);
3921 prison_racct_free(struct prison_racct *prr)
3925 ASSERT_RACCT_ENABLED();
3926 sx_assert(&allprison_lock, SA_UNLOCKED);
3928 old = prr->prr_refcount;
3929 if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
3932 sx_xlock(&allprison_lock);
3933 prison_racct_free_locked(prr);
3934 sx_xunlock(&allprison_lock);
3938 prison_racct_attach(struct prison *pr)
3940 struct prison_racct *prr;
3942 ASSERT_RACCT_ENABLED();
3943 sx_assert(&allprison_lock, SA_XLOCKED);
3945 prr = prison_racct_find_locked(pr->pr_name);
3946 KASSERT(prr != NULL, ("cannot find prison_racct"));
3948 pr->pr_prison_racct = prr;
3952 * Handle jail renaming. From the racct point of view, renaming means
3953 * moving from one prison_racct to another.
3956 prison_racct_modify(struct prison *pr)
3962 struct prison_racct *oldprr;
3964 ASSERT_RACCT_ENABLED();
3966 sx_slock(&allproc_lock);
3967 sx_xlock(&allprison_lock);
3969 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
3970 sx_xunlock(&allprison_lock);
3971 sx_sunlock(&allproc_lock);
3975 oldprr = pr->pr_prison_racct;
3976 pr->pr_prison_racct = NULL;
3978 prison_racct_attach(pr);
3981 * Move resource utilisation records.
3983 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
3987 * Force rctl to reattach rules to processes.
3989 FOREACH_PROC_IN_SYSTEM(p) {
3991 cred = crhold(p->p_ucred);
3993 rctl_proc_ucred_changed(p, cred);
3998 sx_sunlock(&allproc_lock);
3999 prison_racct_free_locked(oldprr);
4000 sx_xunlock(&allprison_lock);
4004 prison_racct_detach(struct prison *pr)
4007 ASSERT_RACCT_ENABLED();
4008 sx_assert(&allprison_lock, SA_UNLOCKED);
4010 if (pr->pr_prison_racct == NULL)
4012 prison_racct_free(pr->pr_prison_racct);
4013 pr->pr_prison_racct = NULL;
4020 db_show_prison(struct prison *pr)
4023 #if defined(INET) || defined(INET6)
4028 char ip4buf[INET_ADDRSTRLEN];
4031 char ip6buf[INET6_ADDRSTRLEN];
4034 db_printf("prison %p:\n", pr);
4035 db_printf(" jid = %d\n", pr->pr_id);
4036 db_printf(" name = %s\n", pr->pr_name);
4037 db_printf(" parent = %p\n", pr->pr_parent);
4038 db_printf(" ref = %d\n", pr->pr_ref);
4039 db_printf(" uref = %d\n", pr->pr_uref);
4040 db_printf(" path = %s\n", pr->pr_path);
4041 db_printf(" cpuset = %d\n", pr->pr_cpuset
4042 ? pr->pr_cpuset->cs_id : -1);
4044 db_printf(" vnet = %p\n", pr->pr_vnet);
4046 db_printf(" root = %p\n", pr->pr_root);
4047 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4048 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4049 db_printf(" children.max = %d\n", pr->pr_childmax);
4050 db_printf(" children.cur = %d\n", pr->pr_childcount);
4051 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4052 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4053 db_printf(" flags = 0x%x", pr->pr_flags);
4054 for (fi = 0; fi < nitems(pr_flag_names); fi++)
4055 if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4056 db_printf(" %s", pr_flag_names[fi]);
4057 for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
4058 jsf = pr->pr_flags &
4059 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4060 db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4061 pr_flag_jailsys[fi].disable &&
4062 (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4063 : (jsf == pr_flag_jailsys[fi].new) ? "new"
4066 db_printf(" allow = 0x%x", pr->pr_allow);
4067 for (fi = 0; fi < nitems(pr_allow_names); fi++)
4068 if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4069 db_printf(" %s", pr_allow_names[fi]);
4071 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4072 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4073 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4074 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4075 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4077 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4078 for (ii = 0; ii < pr->pr_ip4s; ii++)
4079 db_printf(" %s %s\n",
4080 ii == 0 ? "ip4.addr =" : " ",
4081 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4084 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4085 for (ii = 0; ii < pr->pr_ip6s; ii++)
4086 db_printf(" %s %s\n",
4087 ii == 0 ? "ip6.addr =" : " ",
4088 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4092 DB_SHOW_COMMAND(prison, db_show_prison_command)
4098 * Show all prisons in the list, and prison0 which is not
4101 db_show_prison(&prison0);
4102 if (!db_pager_quit) {
4103 TAILQ_FOREACH(pr, &allprison, pr_list) {
4115 /* Look for a prison with the ID and with references. */
4116 TAILQ_FOREACH(pr, &allprison, pr_list)
4117 if (pr->pr_id == addr && pr->pr_ref > 0)
4120 /* Look again, without requiring a reference. */
4121 TAILQ_FOREACH(pr, &allprison, pr_list)
4122 if (pr->pr_id == addr)
4125 /* Assume address points to a valid prison. */
4126 pr = (struct prison *)addr;