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>
53 #include <sys/refcount.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
67 #include <netinet/in.h>
73 #include <security/mac/mac_framework.h>
75 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000"
77 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
78 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
80 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
83 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
85 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
89 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
91 #define _PR_IP_SADDRSEL 0
95 /* prison0 describes what is "real" about the system. */
96 struct prison prison0 = {
102 .pr_securelevel = -1,
104 .pr_childmax = JAIL_MAX,
105 .pr_hostuuid = DEFAULT_HOSTUUID,
106 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
108 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
110 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
112 .pr_allow = PR_ALLOW_ALL,
114 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
116 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
117 struct sx allprison_lock;
118 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
119 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
120 LIST_HEAD(, prison_racct) allprison_racct;
123 static int do_jail_attach(struct thread *td, struct prison *pr);
124 static void prison_complete(void *context, int pending);
125 static void prison_deref(struct prison *pr, int flags);
126 static char *prison_path(struct prison *pr1, struct prison *pr2);
127 static void prison_remove_one(struct prison *pr);
129 static void prison_racct_attach(struct prison *pr);
130 static void prison_racct_modify(struct prison *pr);
131 static void prison_racct_detach(struct prison *pr);
134 /* Flags for prison_deref */
135 #define PD_DEREF 0x01
136 #define PD_DEUREF 0x02
137 #define PD_LOCKED 0x04
138 #define PD_LIST_SLOCKED 0x08
139 #define PD_LIST_XLOCKED 0x10
142 * Parameter names corresponding to PR_* flag values. Size values are for kvm
143 * as we cannot figure out the size of a sparse array, or an array without a
146 static char *pr_flag_names[] = {
149 [7] = "ip4.saddrsel",
152 [8] = "ip6.saddrsel",
155 const size_t pr_flag_names_size = sizeof(pr_flag_names);
157 static char *pr_flag_nonames[] = {
160 [7] = "ip4.nosaddrsel",
163 [8] = "ip6.nosaddrsel",
166 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
168 struct jailsys_flags {
172 } pr_flag_jailsys[] = {
173 { "host", 0, PR_HOST },
175 { "vnet", 0, PR_VNET },
178 { "ip4", PR_IP4_USER, PR_IP4_USER },
181 { "ip6", PR_IP6_USER, PR_IP6_USER },
184 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
186 static char *pr_allow_names[] = {
187 "allow.set_hostname",
195 "allow.mount.nullfs",
197 "allow.mount.procfs",
199 "allow.mount.fdescfs",
200 "allow.mount.linprocfs",
201 "allow.mount.linsysfs",
202 "allow.reserved_ports",
204 const size_t pr_allow_names_size = sizeof(pr_allow_names);
206 static char *pr_allow_nonames[] = {
207 "allow.noset_hostname",
209 "allow.noraw_sockets",
214 "allow.mount.nodevfs",
215 "allow.mount.nonullfs",
217 "allow.mount.noprocfs",
218 "allow.mount.notmpfs",
219 "allow.mount.nofdescfs",
220 "allow.mount.nolinprocfs",
221 "allow.mount.nolinsysfs",
222 "allow.noreserved_ports",
224 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
226 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | PR_ALLOW_RESERVED_PORTS)
227 #define JAIL_DEFAULT_ENFORCE_STATFS 2
228 #define JAIL_DEFAULT_DEVFS_RSNUM 0
229 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
230 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
231 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
232 #if defined(INET) || defined(INET6)
233 static unsigned jail_max_af_ips = 255;
237 * Initialize the parts of prison0 that can't be static-initialized with
238 * constants. This is called from proc0_init() after creating thread0 cpuset.
244 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
245 prison0.pr_osreldate = osreldate;
246 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
255 sys_jail(struct thread *td, struct jail_args *uap)
261 error = copyin(uap->jail, &version, sizeof(uint32_t));
270 /* FreeBSD single IPv4 jails. */
271 bzero(&j, sizeof(struct jail));
272 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
275 j.version = j0.version;
277 j.hostname = j0.hostname;
278 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
284 * Version 1 was used by multi-IPv4 jail implementations
285 * that never made it into the official kernel.
289 case 2: /* JAIL_API_VERSION */
290 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
291 error = copyin(uap->jail, &j, sizeof(struct jail));
297 /* Sci-Fi jails are not supported, sorry. */
300 return (kern_jail(td, &j));
304 kern_jail(struct thread *td, struct jail *j)
306 struct iovec optiov[2 * (4 + nitems(pr_allow_names)
315 char *u_path, *u_hostname, *u_name;
318 struct in_addr *u_ip4;
321 struct in6_addr *u_ip6;
324 int error, enforce_statfs, fi;
326 bzero(&optiov, sizeof(optiov));
327 opt.uio_iov = optiov;
331 opt.uio_segflg = UIO_SYSSPACE;
332 opt.uio_rw = UIO_READ;
335 /* Set permissions for top-level jails from sysctls. */
336 if (!jailed(td->td_ucred)) {
337 for (fi = 0; fi < nitems(pr_allow_names); fi++) {
338 optiov[opt.uio_iovcnt].iov_base =
339 (jail_default_allow & (1 << fi))
340 ? pr_allow_names[fi] : pr_allow_nonames[fi];
341 optiov[opt.uio_iovcnt].iov_len =
342 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
345 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
346 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
348 enforce_statfs = jail_default_enforce_statfs;
349 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
350 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
354 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
356 ip4s = (j->version == 0) ? 1 : j->ip4s;
357 if (ip4s > jail_max_af_ips)
359 tmplen += ip4s * sizeof(struct in_addr);
365 if (j->ip6s > jail_max_af_ips)
367 tmplen += j->ip6s * sizeof(struct in6_addr);
372 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
373 u_hostname = u_path + MAXPATHLEN;
374 u_name = u_hostname + MAXHOSTNAMELEN;
376 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
380 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
382 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
385 optiov[opt.uio_iovcnt].iov_base = "path";
386 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
388 optiov[opt.uio_iovcnt].iov_base = u_path;
389 error = copyinstr(j->path, u_path, MAXPATHLEN,
390 &optiov[opt.uio_iovcnt].iov_len);
392 free(u_path, M_TEMP);
396 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
397 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
399 optiov[opt.uio_iovcnt].iov_base = u_hostname;
400 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
401 &optiov[opt.uio_iovcnt].iov_len);
403 free(u_path, M_TEMP);
407 if (j->jailname != NULL) {
408 optiov[opt.uio_iovcnt].iov_base = "name";
409 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
411 optiov[opt.uio_iovcnt].iov_base = u_name;
412 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
413 &optiov[opt.uio_iovcnt].iov_len);
415 free(u_path, M_TEMP);
421 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
422 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
424 optiov[opt.uio_iovcnt].iov_base = u_ip4;
425 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
427 u_ip4->s_addr = j->ip4s;
429 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
431 free(u_path, M_TEMP);
438 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
439 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
441 optiov[opt.uio_iovcnt].iov_base = u_ip6;
442 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
443 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
445 free(u_path, M_TEMP);
450 KASSERT(opt.uio_iovcnt <= nitems(optiov),
451 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
452 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
453 free(u_path, M_TEMP);
459 * struct jail_set_args {
460 * struct iovec *iovp;
461 * unsigned int iovcnt;
466 sys_jail_set(struct thread *td, struct jail_set_args *uap)
471 /* Check that we have an even number of iovecs. */
475 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
478 error = kern_jail_set(td, auio, uap->flags);
484 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
491 struct in6_addr *ip6;
494 struct vfsoptlist *opts;
495 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
497 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
498 char *g_path, *osrelstr;
499 #if defined(INET) || defined(INET6)
504 size_t namelen, onamelen, pnamelen;
505 int born, created, cuflags, descend, enforce;
506 int error, errmsg_len, errmsg_pos;
507 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
508 int fi, jid, jsys, len, level;
509 int childmax, osreldt, rsnum, slevel;
510 int fullpath_disabled;
511 #if defined(INET) || defined(INET6)
520 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
524 error = priv_check(td, PRIV_JAIL_SET);
525 if (!error && (flags & JAIL_ATTACH))
526 error = priv_check(td, PRIV_JAIL_ATTACH);
529 mypr = td->td_ucred->cr_prison;
530 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
532 if (flags & ~JAIL_SET_MASK)
536 * Check all the parameters before committing to anything. Not all
537 * errors can be caught early, but we may as well try. Also, this
538 * takes care of some expensive stuff (path lookup) before getting
539 * the allprison lock.
541 * XXX Jails are not filesystems, and jail parameters are not mount
542 * options. But it makes more sense to re-use the vfsopt code
543 * than duplicate it under a different name.
545 error = vfs_buildopts(optuio, &opts);
556 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
559 vfs_opterror(opts, "no valid operation (create or update)");
563 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
569 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
578 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
586 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
591 else if (enforce < 0 || enforce > 2) {
597 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
605 pr_flags = ch_flags = 0;
606 for (fi = 0; fi < nitems(pr_flag_names); fi++) {
607 if (pr_flag_names[fi] == NULL)
609 vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
610 vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
612 ch_flags |= pr_flags;
613 for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
614 error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
621 case JAIL_SYS_DISABLE:
622 if (!pr_flag_jailsys[fi].disable) {
626 pr_flags |= pr_flag_jailsys[fi].disable;
629 pr_flags |= pr_flag_jailsys[fi].new;
631 case JAIL_SYS_INHERIT:
638 pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
640 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
641 && !(pr_flags & PR_PERSIST)) {
643 vfs_opterror(opts, "new jail must persist or attach");
647 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
649 vfs_opterror(opts, "vnet cannot be changed after creation");
654 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
656 vfs_opterror(opts, "ip4 cannot be changed after creation");
661 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
663 vfs_opterror(opts, "ip6 cannot be changed after creation");
668 pr_allow = ch_allow = 0;
669 for (fi = 0; fi < nitems(pr_allow_names); fi++) {
670 vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
671 vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
673 ch_allow |= pr_allow;
675 error = vfs_getopt(opts, "name", (void **)&name, &len);
681 if (len == 0 || name[len - 1] != '\0') {
685 if (len > MAXHOSTNAMELEN) {
686 error = ENAMETOOLONG;
691 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
699 if (len == 0 || host[len - 1] != '\0') {
703 if (len > MAXHOSTNAMELEN) {
704 error = ENAMETOOLONG;
709 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
717 if (len == 0 || domain[len - 1] != '\0') {
721 if (len > MAXHOSTNAMELEN) {
722 error = ENAMETOOLONG;
727 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
735 if (len == 0 || uuid[len - 1] != '\0') {
739 if (len > HOSTUUIDLEN) {
740 error = ENAMETOOLONG;
745 #ifdef COMPAT_FREEBSD32
746 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
749 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
753 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
765 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
770 else if (ip4s & (sizeof(*ip4) - 1)) {
774 ch_flags |= PR_IP4_USER;
775 pr_flags |= PR_IP4_USER;
777 ip4s /= sizeof(*ip4);
778 if (ip4s > jail_max_af_ips) {
780 vfs_opterror(opts, "too many IPv4 addresses");
783 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
784 bcopy(op, ip4, ip4s * sizeof(*ip4));
786 * IP addresses are all sorted but ip[0] to preserve
787 * the primary IP address as given from userland.
788 * This special IP is used for unbound outgoing
789 * connections as well for "loopback" traffic in case
790 * source address selection cannot find any more fitting
791 * address to connect from.
794 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
797 * Check for duplicate addresses and do some simple
798 * zero and broadcast checks. If users give other bogus
799 * addresses it is their problem.
801 * We do not have to care about byte order for these
802 * checks so we will do them in NBO.
804 for (ii = 0; ii < ip4s; ii++) {
805 if (ip4[ii].s_addr == INADDR_ANY ||
806 ip4[ii].s_addr == INADDR_BROADCAST) {
811 (ip4[0].s_addr == ip4[ii+1].s_addr ||
812 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
822 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
827 else if (ip6s & (sizeof(*ip6) - 1)) {
831 ch_flags |= PR_IP6_USER;
832 pr_flags |= PR_IP6_USER;
834 ip6s /= sizeof(*ip6);
835 if (ip6s > jail_max_af_ips) {
837 vfs_opterror(opts, "too many IPv6 addresses");
840 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
841 bcopy(op, ip6, ip6s * sizeof(*ip6));
843 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
845 for (ii = 0; ii < ip6s; ii++) {
846 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
851 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
852 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
862 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
863 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
866 "vnet jails cannot have IP address restrictions");
871 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
877 if (flags & JAIL_UPDATE) {
880 "osrelease cannot be changed after creation");
883 if (len == 0 || len >= OSRELEASELEN) {
886 "osrelease string must be 1-%d bytes long",
892 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
898 if (flags & JAIL_UPDATE) {
901 "osreldate cannot be changed after creation");
906 vfs_opterror(opts, "osreldate cannot be 0");
911 fullpath_disabled = 0;
913 error = vfs_getopt(opts, "path", (void **)&path, &len);
919 if (flags & JAIL_UPDATE) {
922 "path cannot be changed after creation");
925 if (len == 0 || path[len - 1] != '\0') {
929 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
935 NDFREE(&nd, NDF_ONLY_PNBUF);
936 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
937 strlcpy(g_path, path, MAXPATHLEN);
938 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
941 else if (error == ENODEV) {
942 /* proceed if sysctl debug.disablefullpath == 1 */
943 fullpath_disabled = 1;
944 if (len < 2 || (len == 2 && path[0] == '/'))
947 /* exit on other errors */
950 if (root->v_type != VDIR) {
956 if (fullpath_disabled) {
957 /* Leave room for a real-root full pathname. */
958 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
959 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
960 error = ENAMETOOLONG;
968 * Find the specified jail, or at least its parent.
969 * This abuses the file error codes ENOENT and EEXIST.
973 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
974 namelc = strrchr(name, '.');
975 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
979 sx_xlock(&allprison_lock);
982 * See if a requested jid already exists. There is an
983 * information leak here if the jid exists but is not within
984 * the caller's jail hierarchy. Jail creators will get EEXIST
985 * even though they cannot see the jail, and CREATE | UPDATE
986 * will return ENOENT which is not normally a valid error.
990 vfs_opterror(opts, "negative jid");
991 goto done_unlock_list;
993 pr = prison_find(jid);
996 /* Create: jid must not exist. */
997 if (cuflags == JAIL_CREATE) {
998 mtx_unlock(&pr->pr_mtx);
1000 vfs_opterror(opts, "jail %d already exists",
1002 goto done_unlock_list;
1004 if (!prison_ischild(mypr, pr)) {
1005 mtx_unlock(&pr->pr_mtx);
1007 } else if (pr->pr_uref == 0) {
1008 if (!(flags & JAIL_DYING)) {
1009 mtx_unlock(&pr->pr_mtx);
1011 vfs_opterror(opts, "jail %d is dying",
1013 goto done_unlock_list;
1014 } else if ((flags & JAIL_ATTACH) ||
1015 (pr_flags & PR_PERSIST)) {
1017 * A dying jail might be resurrected
1018 * (via attach or persist), but first
1019 * it must determine if another jail
1020 * has claimed its name. Accomplish
1021 * this by implicitly re-setting the
1025 name = prison_name(mypr, pr);
1030 /* Update: jid must exist. */
1031 if (cuflags == JAIL_UPDATE) {
1033 vfs_opterror(opts, "jail %d not found", jid);
1034 goto done_unlock_list;
1039 * If the caller provided a name, look for a jail by that name.
1040 * This has different semantics for creates and updates keyed by jid
1041 * (where the name must not already exist in a different jail),
1042 * and updates keyed by the name itself (where the name must exist
1043 * because that is the jail being updated).
1047 namelc = strrchr(name, '.');
1052 * This is a hierarchical name. Split it into the
1053 * parent and child names, and make sure the parent
1054 * exists or matches an already found jail.
1057 if (strncmp(name, ppr->pr_name, namelc - name)
1058 || ppr->pr_name[namelc - name] != '\0') {
1059 mtx_unlock(&pr->pr_mtx);
1062 "cannot change jail's parent");
1063 goto done_unlock_list;
1067 ppr = prison_find_name(mypr, name);
1071 "jail \"%s\" not found", name);
1072 goto done_unlock_list;
1074 mtx_unlock(&ppr->pr_mtx);
1079 if (namelc[0] != '\0') {
1081 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1084 FOREACH_PRISON_CHILD(ppr, tpr) {
1085 if (tpr != pr && tpr->pr_ref > 0 &&
1086 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1088 cuflags != JAIL_CREATE) {
1089 mtx_lock(&tpr->pr_mtx);
1090 if (tpr->pr_ref > 0) {
1095 if (tpr->pr_uref > 0) {
1101 mtx_unlock(&tpr->pr_mtx);
1102 } else if (tpr->pr_uref > 0) {
1104 * Create, or update(jid):
1105 * name must not exist in an
1106 * active sibling jail.
1110 mtx_unlock(&pr->pr_mtx);
1112 "jail \"%s\" already exists",
1114 goto done_unlock_list;
1118 /* If no active jail is found, use a dying one. */
1119 if (deadpr != NULL && pr == NULL) {
1120 if (flags & JAIL_DYING) {
1121 mtx_lock(&deadpr->pr_mtx);
1122 if (deadpr->pr_ref == 0) {
1123 mtx_unlock(&deadpr->pr_mtx);
1127 } else if (cuflags == JAIL_UPDATE) {
1130 "jail \"%s\" is dying", name);
1131 goto done_unlock_list;
1134 /* Update: name must exist if no jid. */
1135 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1137 vfs_opterror(opts, "jail \"%s\" not found",
1139 goto done_unlock_list;
1143 /* Update: must provide a jid or name. */
1144 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1146 vfs_opterror(opts, "update specified no jail");
1147 goto done_unlock_list;
1150 /* If there's no prison to update, create a new one and link it in. */
1152 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1153 if (tpr->pr_childcount >= tpr->pr_childmax) {
1155 vfs_opterror(opts, "prison limit exceeded");
1156 goto done_unlock_list;
1159 mtx_lock(&ppr->pr_mtx);
1160 if (ppr->pr_ref == 0) {
1161 mtx_unlock(&ppr->pr_mtx);
1163 vfs_opterror(opts, "jail \"%s\" not found",
1164 prison_name(mypr, ppr));
1165 goto done_unlock_list;
1169 mtx_unlock(&ppr->pr_mtx);
1170 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1172 /* Find the next free jid. */
1175 if (jid == JAIL_MAX)
1177 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1178 if (tpr->pr_id < jid)
1180 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1181 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1184 if (jid == lastprid) {
1187 "no available jail IDs");
1189 prison_deref(ppr, PD_DEREF |
1190 PD_DEUREF | PD_LIST_XLOCKED);
1199 * The jail already has a jid (that did not yet exist),
1200 * so just find where to insert it.
1202 TAILQ_FOREACH(tpr, &allprison, pr_list)
1203 if (tpr->pr_id >= jid) {
1204 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1209 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1210 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1211 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1212 tpr->pr_childcount++;
1214 pr->pr_parent = ppr;
1217 /* Set some default values, and inherit some from the parent. */
1222 root = mypr->pr_root;
1225 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1226 pr->pr_flags |= PR_HOST;
1227 #if defined(INET) || defined(INET6)
1229 if (!(pr_flags & PR_VNET))
1233 if (!(ch_flags & PR_IP4_USER))
1234 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1235 else if (!(pr_flags & PR_IP4_USER)) {
1236 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1237 if (ppr->pr_ip4 != NULL) {
1238 pr->pr_ip4s = ppr->pr_ip4s;
1239 pr->pr_ip4 = malloc(pr->pr_ip4s *
1240 sizeof(struct in_addr), M_PRISON,
1242 bcopy(ppr->pr_ip4, pr->pr_ip4,
1243 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1248 if (!(ch_flags & PR_IP6_USER))
1249 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1250 else if (!(pr_flags & PR_IP6_USER)) {
1251 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1252 if (ppr->pr_ip6 != NULL) {
1253 pr->pr_ip6s = ppr->pr_ip6s;
1254 pr->pr_ip6 = malloc(pr->pr_ip6s *
1255 sizeof(struct in6_addr), M_PRISON,
1257 bcopy(ppr->pr_ip6, pr->pr_ip6,
1258 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1264 /* Source address selection is always on by default. */
1265 pr->pr_flags |= _PR_IP_SADDRSEL;
1267 pr->pr_securelevel = ppr->pr_securelevel;
1268 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1269 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1270 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1272 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1273 if (osrelstr == NULL)
1274 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1276 strcpy(pr->pr_osrelease, osrelstr);
1278 LIST_INIT(&pr->pr_children);
1279 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1280 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1283 /* Allocate a new vnet if specified. */
1284 pr->pr_vnet = (pr_flags & PR_VNET)
1285 ? vnet_alloc() : ppr->pr_vnet;
1288 * Allocate a dedicated cpuset for each jail.
1289 * Unlike other initial settings, this may return an erorr.
1291 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1293 prison_deref(pr, PD_LIST_XLOCKED);
1297 mtx_lock(&pr->pr_mtx);
1299 * New prisons do not yet have a reference, because we do not
1300 * want others to see the incomplete prison once the
1301 * allprison_lock is downgraded.
1306 * Grab a reference for existing prisons, to ensure they
1307 * continue to exist for the duration of the call.
1310 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1311 if ((pr->pr_flags & PR_VNET) &&
1312 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1315 "vnet jails cannot have IP address restrictions");
1316 goto done_deref_locked;
1320 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1323 "ip4 cannot be changed after creation");
1324 goto done_deref_locked;
1328 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1331 "ip6 cannot be changed after creation");
1332 goto done_deref_locked;
1337 /* Do final error checking before setting anything. */
1339 if (slevel < ppr->pr_securelevel) {
1341 goto done_deref_locked;
1345 if (childmax >= ppr->pr_childmax) {
1347 goto done_deref_locked;
1351 if (enforce < ppr->pr_enforce_statfs) {
1353 goto done_deref_locked;
1358 * devfs_rsnum is a uint16_t
1360 if (rsnum < 0 || rsnum > 65535) {
1362 goto done_deref_locked;
1365 * Nested jails always inherit parent's devfs ruleset
1367 if (jailed(td->td_ucred)) {
1368 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1370 goto done_deref_locked;
1372 rsnum = ppr->pr_devfs_rsnum;
1377 if (ppr->pr_flags & PR_IP4) {
1379 * Make sure the new set of IP addresses is a
1380 * subset of the parent's list. Don't worry
1381 * about the parent being unlocked, as any
1382 * setting is done with allprison_lock held.
1384 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1385 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1387 if (ij == ppr->pr_ip4s) {
1389 goto done_deref_locked;
1392 for (ii = ij = 1; ii < ip4s; ii++) {
1393 if (ip4[ii].s_addr ==
1394 ppr->pr_ip4[0].s_addr)
1396 for (; ij < ppr->pr_ip4s; ij++)
1397 if (ip4[ii].s_addr ==
1398 ppr->pr_ip4[ij].s_addr)
1400 if (ij == ppr->pr_ip4s)
1403 if (ij == ppr->pr_ip4s) {
1405 goto done_deref_locked;
1410 * Check for conflicting IP addresses. We permit them
1411 * if there is no more than one IP on each jail. If
1412 * there is a duplicate on a jail with more than one
1413 * IP stop checking and return error.
1417 for (; tppr != &prison0; tppr = tppr->pr_parent)
1418 if (tppr->pr_flags & PR_VNET)
1421 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1424 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1426 tpr->pr_uref == 0) {
1430 if (!(tpr->pr_flags & PR_IP4_USER))
1433 if (tpr->pr_ip4 == NULL ||
1434 (ip4s == 1 && tpr->pr_ip4s == 1))
1436 for (ii = 0; ii < ip4s; ii++) {
1437 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1441 "IPv4 addresses clash");
1442 goto done_deref_locked;
1450 if (ppr->pr_flags & PR_IP6) {
1452 * Make sure the new set of IP addresses is a
1453 * subset of the parent's list.
1455 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1456 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1459 if (ij == ppr->pr_ip6s) {
1461 goto done_deref_locked;
1464 for (ii = ij = 1; ii < ip6s; ii++) {
1465 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1468 for (; ij < ppr->pr_ip6s; ij++)
1469 if (IN6_ARE_ADDR_EQUAL(
1470 &ip6[ii], &ppr->pr_ip6[ij]))
1472 if (ij == ppr->pr_ip6s)
1475 if (ij == ppr->pr_ip6s) {
1477 goto done_deref_locked;
1481 /* Check for conflicting IP addresses. */
1484 for (; tppr != &prison0; tppr = tppr->pr_parent)
1485 if (tppr->pr_flags & PR_VNET)
1488 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1491 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1493 tpr->pr_uref == 0) {
1497 if (!(tpr->pr_flags & PR_IP6_USER))
1500 if (tpr->pr_ip6 == NULL ||
1501 (ip6s == 1 && tpr->pr_ip6s == 1))
1503 for (ii = 0; ii < ip6s; ii++) {
1504 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1508 "IPv6 addresses clash");
1509 goto done_deref_locked;
1515 onamelen = namelen = 0;
1516 if (namelc != NULL) {
1517 /* Give a default name of the jid. Also allow the name to be
1518 * explicitly the jid - but not any other number, and only in
1519 * normal form (no leading zero/etc).
1521 if (namelc[0] == '\0')
1522 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1523 else if ((strtoul(namelc, &p, 10) != jid ||
1524 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1527 "name cannot be numeric (unless it is the jid)");
1528 goto done_deref_locked;
1531 * Make sure the name isn't too long for the prison or its
1534 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1535 onamelen = strlen(pr->pr_name + pnamelen);
1536 namelen = strlen(namelc);
1537 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1538 error = ENAMETOOLONG;
1539 goto done_deref_locked;
1541 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1542 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1543 sizeof(pr->pr_name)) {
1544 error = ENAMETOOLONG;
1545 goto done_deref_locked;
1549 if (pr_allow & ~ppr->pr_allow) {
1551 goto done_deref_locked;
1555 * Let modules check their parameters. This requires unlocking and
1556 * then re-locking the prison, but this is still a valid state as long
1557 * as allprison_lock remains xlocked.
1559 mtx_unlock(&pr->pr_mtx);
1560 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1562 prison_deref(pr, created
1564 : PD_DEREF | PD_LIST_XLOCKED);
1567 mtx_lock(&pr->pr_mtx);
1569 /* At this point, all valid parameters should have been noted. */
1570 TAILQ_FOREACH(opt, opts, link) {
1571 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1573 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1574 goto done_deref_locked;
1578 /* Set the parameters of the prison. */
1581 if (pr_flags & PR_IP4_USER) {
1582 pr->pr_flags |= PR_IP4;
1583 free(pr->pr_ip4, M_PRISON);
1587 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1589 if (tpr->pr_flags & PR_VNET) {
1594 if (prison_restrict_ip4(tpr, NULL)) {
1603 if (pr_flags & PR_IP6_USER) {
1604 pr->pr_flags |= PR_IP6;
1605 free(pr->pr_ip6, M_PRISON);
1609 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1611 if (tpr->pr_flags & PR_VNET) {
1616 if (prison_restrict_ip6(tpr, NULL)) {
1624 pr->pr_securelevel = slevel;
1625 /* Set all child jails to be at least this level. */
1626 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1627 if (tpr->pr_securelevel < slevel)
1628 tpr->pr_securelevel = slevel;
1631 pr->pr_childmax = childmax;
1632 /* Set all child jails to under this limit. */
1633 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1634 if (tpr->pr_childmax > childmax - level)
1635 tpr->pr_childmax = childmax > level
1636 ? childmax - level : 0;
1639 pr->pr_enforce_statfs = enforce;
1640 /* Pass this restriction on to the children. */
1641 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1642 if (tpr->pr_enforce_statfs < enforce)
1643 tpr->pr_enforce_statfs = enforce;
1646 pr->pr_devfs_rsnum = rsnum;
1647 /* Pass this restriction on to the children. */
1648 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1649 tpr->pr_devfs_rsnum = rsnum;
1651 if (namelc != NULL) {
1652 if (ppr == &prison0)
1653 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1655 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1656 ppr->pr_name, namelc);
1657 /* Change this component of child names. */
1658 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1659 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1660 strlen(tpr->pr_name + onamelen) + 1);
1661 bcopy(pr->pr_name, tpr->pr_name, namelen);
1665 /* Try to keep a real-rooted full pathname. */
1666 if (fullpath_disabled && path[0] == '/' &&
1667 strcmp(mypr->pr_path, "/"))
1668 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1669 mypr->pr_path, path);
1671 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1674 if (PR_HOST & ch_flags & ~pr_flags) {
1675 if (pr->pr_flags & PR_HOST) {
1677 * Copy the parent's host info. As with pr_ip4 above,
1678 * the lack of a lock on the parent is not a problem;
1679 * it is always set with allprison_lock at least
1680 * shared, and is held exclusively here.
1682 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1683 sizeof(pr->pr_hostname));
1684 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1685 sizeof(pr->pr_domainname));
1686 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1687 sizeof(pr->pr_hostuuid));
1688 pr->pr_hostid = pr->pr_parent->pr_hostid;
1690 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1691 /* Set this prison, and any descendants without PR_HOST. */
1693 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1695 strlcpy(pr->pr_domainname, domain,
1696 sizeof(pr->pr_domainname));
1698 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1700 pr->pr_hostid = hid;
1701 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1702 if (tpr->pr_flags & PR_HOST)
1706 strlcpy(tpr->pr_hostname,
1708 sizeof(tpr->pr_hostname));
1710 strlcpy(tpr->pr_domainname,
1712 sizeof(tpr->pr_domainname));
1714 strlcpy(tpr->pr_hostuuid,
1716 sizeof(tpr->pr_hostuuid));
1718 tpr->pr_hostid = hid;
1722 if ((tallow = ch_allow & ~pr_allow)) {
1723 /* Clear allow bits in all children. */
1724 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1725 tpr->pr_allow &= ~tallow;
1727 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1729 * Persistent prisons get an extra reference, and prisons losing their
1730 * persist flag lose that reference. Only do this for existing prisons
1731 * for now, so new ones will remain unseen until after the module
1732 * handlers have completed.
1734 born = pr->pr_uref == 0;
1735 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1736 if (pr_flags & PR_PERSIST) {
1744 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1745 mtx_unlock(&pr->pr_mtx);
1748 if (racct_enable && created)
1749 prison_racct_attach(pr);
1752 /* Locks may have prevented a complete restriction of child IP
1753 * addresses. If so, allocate some more memory and try again.
1758 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1759 mtx_lock(&pr->pr_mtx);
1761 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1763 if (tpr->pr_flags & PR_VNET) {
1768 if (prison_restrict_ip4(tpr, ip4)) {
1775 mtx_unlock(&pr->pr_mtx);
1781 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1782 mtx_lock(&pr->pr_mtx);
1784 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1786 if (tpr->pr_flags & PR_VNET) {
1791 if (prison_restrict_ip6(tpr, ip6)) {
1798 mtx_unlock(&pr->pr_mtx);
1802 /* Let the modules do their work. */
1803 sx_downgrade(&allprison_lock);
1805 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1807 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1808 prison_deref(pr, created
1810 : PD_DEREF | PD_LIST_SLOCKED);
1814 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1817 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1818 prison_deref(pr, created
1820 : PD_DEREF | PD_LIST_SLOCKED);
1824 /* Attach this process to the prison if requested. */
1825 if (flags & JAIL_ATTACH) {
1826 mtx_lock(&pr->pr_mtx);
1827 error = do_jail_attach(td, pr);
1829 vfs_opterror(opts, "attach failed");
1831 prison_deref(pr, PD_DEREF);
1837 if (racct_enable && !created) {
1838 if (!(flags & JAIL_ATTACH))
1839 sx_sunlock(&allprison_lock);
1840 prison_racct_modify(pr);
1841 if (!(flags & JAIL_ATTACH))
1842 sx_slock(&allprison_lock);
1846 td->td_retval[0] = pr->pr_id;
1849 * Now that it is all there, drop the temporary reference from existing
1850 * prisons. Or add a reference to newly created persistent prisons
1851 * (which was not done earlier so that the prison would not be publicly
1855 prison_deref(pr, (flags & JAIL_ATTACH)
1857 : PD_DEREF | PD_LIST_SLOCKED);
1859 if (pr_flags & PR_PERSIST) {
1860 mtx_lock(&pr->pr_mtx);
1863 mtx_unlock(&pr->pr_mtx);
1865 if (!(flags & JAIL_ATTACH))
1866 sx_sunlock(&allprison_lock);
1872 prison_deref(pr, created
1873 ? PD_LOCKED | PD_LIST_XLOCKED
1874 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1877 sx_xunlock(&allprison_lock);
1883 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1884 &errmsg_len) == 0 && errmsg_len > 0) {
1885 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1886 if (optuio->uio_segflg == UIO_SYSSPACE)
1888 optuio->uio_iov[errmsg_pos].iov_base,
1892 optuio->uio_iov[errmsg_pos].iov_base,
1898 free(ip4, M_PRISON);
1901 free(ip6, M_PRISON);
1904 free(g_path, M_TEMP);
1911 * struct jail_get_args {
1912 * struct iovec *iovp;
1913 * unsigned int iovcnt;
1918 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1923 /* Check that we have an even number of iovecs. */
1924 if (uap->iovcnt & 1)
1927 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1930 error = kern_jail_get(td, auio, uap->flags);
1932 error = copyout(auio->uio_iov, uap->iovp,
1933 uap->iovcnt * sizeof (struct iovec));
1939 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1941 struct prison *pr, *mypr;
1943 struct vfsoptlist *opts;
1944 char *errmsg, *name;
1945 int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1947 if (flags & ~JAIL_GET_MASK)
1950 /* Get the parameter list. */
1951 error = vfs_buildopts(optuio, &opts);
1954 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1955 mypr = td->td_ucred->cr_prison;
1958 * Find the prison specified by one of: lastjid, jid, name.
1960 sx_slock(&allprison_lock);
1961 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1963 TAILQ_FOREACH(pr, &allprison, pr_list) {
1964 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1965 mtx_lock(&pr->pr_mtx);
1966 if (pr->pr_ref > 0 &&
1967 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1969 mtx_unlock(&pr->pr_mtx);
1975 vfs_opterror(opts, "no jail after %d", jid);
1976 goto done_unlock_list;
1977 } else if (error != ENOENT)
1978 goto done_unlock_list;
1980 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1983 pr = prison_find_child(mypr, jid);
1985 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1986 mtx_unlock(&pr->pr_mtx);
1988 vfs_opterror(opts, "jail %d is dying",
1990 goto done_unlock_list;
1995 vfs_opterror(opts, "jail %d not found", jid);
1996 goto done_unlock_list;
1998 } else if (error != ENOENT)
1999 goto done_unlock_list;
2001 error = vfs_getopt(opts, "name", (void **)&name, &len);
2003 if (len == 0 || name[len - 1] != '\0') {
2005 goto done_unlock_list;
2007 pr = prison_find_name(mypr, name);
2009 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2010 mtx_unlock(&pr->pr_mtx);
2012 vfs_opterror(opts, "jail \"%s\" is dying",
2014 goto done_unlock_list;
2019 vfs_opterror(opts, "jail \"%s\" not found", name);
2020 goto done_unlock_list;
2021 } else if (error != ENOENT)
2022 goto done_unlock_list;
2024 vfs_opterror(opts, "no jail specified");
2026 goto done_unlock_list;
2029 /* Get the parameters of the prison. */
2032 td->td_retval[0] = pr->pr_id;
2033 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2034 if (error != 0 && error != ENOENT)
2036 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2037 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2038 if (error != 0 && error != ENOENT)
2040 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2041 if (error != 0 && error != ENOENT)
2043 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2044 sizeof(pr->pr_cpuset->cs_id));
2045 if (error != 0 && error != ENOENT)
2047 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2048 if (error != 0 && error != ENOENT)
2051 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2052 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2053 if (error != 0 && error != ENOENT)
2057 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2058 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2059 if (error != 0 && error != ENOENT)
2062 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2063 sizeof(pr->pr_securelevel));
2064 if (error != 0 && error != ENOENT)
2066 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2067 sizeof(pr->pr_childcount));
2068 if (error != 0 && error != ENOENT)
2070 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2071 sizeof(pr->pr_childmax));
2072 if (error != 0 && error != ENOENT)
2074 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2075 if (error != 0 && error != ENOENT)
2077 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2078 if (error != 0 && error != ENOENT)
2080 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2081 if (error != 0 && error != ENOENT)
2083 #ifdef COMPAT_FREEBSD32
2084 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2085 uint32_t hid32 = pr->pr_hostid;
2087 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2090 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2091 sizeof(pr->pr_hostid));
2092 if (error != 0 && error != ENOENT)
2094 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2095 sizeof(pr->pr_enforce_statfs));
2096 if (error != 0 && error != ENOENT)
2098 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2099 sizeof(pr->pr_devfs_rsnum));
2100 if (error != 0 && error != ENOENT)
2102 for (fi = 0; fi < nitems(pr_flag_names); fi++) {
2103 if (pr_flag_names[fi] == NULL)
2105 i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2106 error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2107 if (error != 0 && error != ENOENT)
2110 error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2111 if (error != 0 && error != ENOENT)
2114 for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
2116 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2117 i = pr_flag_jailsys[fi].disable &&
2118 (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2119 : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2122 vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2123 if (error != 0 && error != ENOENT)
2126 for (fi = 0; fi < nitems(pr_allow_names); fi++) {
2127 if (pr_allow_names[fi] == NULL)
2129 i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2130 error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2131 if (error != 0 && error != ENOENT)
2134 error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2135 if (error != 0 && error != ENOENT)
2138 i = (pr->pr_uref == 0);
2139 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2140 if (error != 0 && error != ENOENT)
2143 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2144 if (error != 0 && error != ENOENT)
2146 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2147 sizeof(pr->pr_osreldate));
2148 if (error != 0 && error != ENOENT)
2150 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2151 if (error != 0 && error != ENOENT)
2154 /* Get the module parameters. */
2155 mtx_unlock(&pr->pr_mtx);
2157 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2160 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2162 /* By now, all parameters should have been noted. */
2163 TAILQ_FOREACH(opt, opts, link) {
2164 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2166 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2171 /* Write the fetched parameters back to userspace. */
2173 TAILQ_FOREACH(opt, opts, link) {
2174 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2175 pos = 2 * opt->pos + 1;
2176 optuio->uio_iov[pos].iov_len = opt->len;
2177 if (opt->value != NULL) {
2178 if (optuio->uio_segflg == UIO_SYSSPACE) {
2180 optuio->uio_iov[pos].iov_base,
2183 error = copyout(opt->value,
2184 optuio->uio_iov[pos].iov_base,
2195 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2199 sx_sunlock(&allprison_lock);
2201 if (error && errmsg_pos >= 0) {
2202 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2203 errmsg_pos = 2 * errmsg_pos + 1;
2204 if (errmsg_len > 0) {
2205 if (optuio->uio_segflg == UIO_SYSSPACE)
2207 optuio->uio_iov[errmsg_pos].iov_base,
2211 optuio->uio_iov[errmsg_pos].iov_base,
2221 * struct jail_remove_args {
2226 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2228 struct prison *pr, *cpr, *lpr, *tpr;
2231 error = priv_check(td, PRIV_JAIL_REMOVE);
2235 sx_xlock(&allprison_lock);
2236 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2238 sx_xunlock(&allprison_lock);
2242 /* Remove all descendants of this prison, then remove this prison. */
2244 if (!LIST_EMPTY(&pr->pr_children)) {
2245 mtx_unlock(&pr->pr_mtx);
2247 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2248 mtx_lock(&cpr->pr_mtx);
2249 if (cpr->pr_ref > 0) {
2253 /* Already removed - do not do it again. */
2256 mtx_unlock(&cpr->pr_mtx);
2258 mtx_lock(&lpr->pr_mtx);
2259 prison_remove_one(lpr);
2260 sx_xlock(&allprison_lock);
2265 mtx_lock(&lpr->pr_mtx);
2266 prison_remove_one(lpr);
2267 sx_xlock(&allprison_lock);
2269 mtx_lock(&pr->pr_mtx);
2271 prison_remove_one(pr);
2276 prison_remove_one(struct prison *pr)
2281 /* If the prison was persistent, it is not anymore. */
2283 if (pr->pr_flags & PR_PERSIST) {
2286 pr->pr_flags &= ~PR_PERSIST;
2290 * jail_remove added a reference. If that's the only one, remove
2293 KASSERT(pr->pr_ref > 0,
2294 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2295 if (pr->pr_ref == 1) {
2297 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2301 mtx_unlock(&pr->pr_mtx);
2302 sx_xunlock(&allprison_lock);
2304 * Kill all processes unfortunate enough to be attached to this prison.
2306 sx_slock(&allproc_lock);
2307 LIST_FOREACH(p, &allproc, p_list) {
2309 if (p->p_state != PRS_NEW && p->p_ucred &&
2310 p->p_ucred->cr_prison == pr)
2311 kern_psignal(p, SIGKILL);
2314 sx_sunlock(&allproc_lock);
2315 /* Remove the temporary reference added by jail_remove. */
2316 prison_deref(pr, deuref | PD_DEREF);
2321 * struct jail_attach_args {
2326 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2331 error = priv_check(td, PRIV_JAIL_ATTACH);
2336 * Start with exclusive hold on allprison_lock to ensure that a possible
2337 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2338 * But then immediately downgrade it since we don't need to stop
2341 sx_xlock(&allprison_lock);
2342 sx_downgrade(&allprison_lock);
2343 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2345 sx_sunlock(&allprison_lock);
2350 * Do not allow a process to attach to a prison that is not
2351 * considered to be "alive".
2353 if (pr->pr_uref == 0) {
2354 mtx_unlock(&pr->pr_mtx);
2355 sx_sunlock(&allprison_lock);
2359 return (do_jail_attach(td, pr));
2363 do_jail_attach(struct thread *td, struct prison *pr)
2366 struct ucred *newcred, *oldcred;
2370 * XXX: Note that there is a slight race here if two threads
2371 * in the same privileged process attempt to attach to two
2372 * different jails at the same time. It is important for
2373 * user processes not to do this, or they might end up with
2374 * a process root from one prison, but attached to the jail
2379 mtx_unlock(&pr->pr_mtx);
2381 /* Let modules do whatever they need to prepare for attaching. */
2382 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2384 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2387 sx_sunlock(&allprison_lock);
2390 * Reparent the newly attached process to this jail.
2393 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2397 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2398 if ((error = change_dir(pr->pr_root, td)) != 0)
2401 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2404 VOP_UNLOCK(pr->pr_root, 0);
2405 if ((error = pwd_chroot(td, pr->pr_root)))
2410 oldcred = crcopysafe(p, newcred);
2411 newcred->cr_prison = pr;
2412 proc_set_cred(p, newcred);
2416 racct_proc_ucred_changed(p, oldcred, newcred);
2418 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2423 VOP_UNLOCK(pr->pr_root, 0);
2425 /* Tell modules this thread is still in its old jail after all. */
2426 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2427 prison_deref(pr, PD_DEREF | PD_DEUREF);
2433 * Returns a locked prison instance, or NULL on failure.
2436 prison_find(int prid)
2440 sx_assert(&allprison_lock, SX_LOCKED);
2441 TAILQ_FOREACH(pr, &allprison, pr_list) {
2442 if (pr->pr_id == prid) {
2443 mtx_lock(&pr->pr_mtx);
2446 mtx_unlock(&pr->pr_mtx);
2453 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2456 prison_find_child(struct prison *mypr, int prid)
2461 sx_assert(&allprison_lock, SX_LOCKED);
2462 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2463 if (pr->pr_id == prid) {
2464 mtx_lock(&pr->pr_mtx);
2467 mtx_unlock(&pr->pr_mtx);
2474 * Look for the name relative to mypr. Returns a locked prison or NULL.
2477 prison_find_name(struct prison *mypr, const char *name)
2479 struct prison *pr, *deadpr;
2483 sx_assert(&allprison_lock, SX_LOCKED);
2484 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2487 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2488 if (!strcmp(pr->pr_name + mylen, name)) {
2489 mtx_lock(&pr->pr_mtx);
2490 if (pr->pr_ref > 0) {
2491 if (pr->pr_uref > 0)
2495 mtx_unlock(&pr->pr_mtx);
2498 /* There was no valid prison - perhaps there was a dying one. */
2499 if (deadpr != NULL) {
2500 mtx_lock(&deadpr->pr_mtx);
2501 if (deadpr->pr_ref == 0) {
2502 mtx_unlock(&deadpr->pr_mtx);
2510 * See if a prison has the specific flag set.
2513 prison_flag(struct ucred *cred, unsigned flag)
2516 /* This is an atomic read, so no locking is necessary. */
2517 return (cred->cr_prison->pr_flags & flag);
2521 prison_allow(struct ucred *cred, unsigned flag)
2524 /* This is an atomic read, so no locking is necessary. */
2525 return (cred->cr_prison->pr_allow & flag);
2529 * Remove a prison reference. If that was the last reference, remove the
2530 * prison itself - but not in this context in case there are locks held.
2533 prison_free_locked(struct prison *pr)
2537 mtx_assert(&pr->pr_mtx, MA_OWNED);
2539 mtx_unlock(&pr->pr_mtx);
2541 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2545 prison_free(struct prison *pr)
2548 mtx_lock(&pr->pr_mtx);
2549 prison_free_locked(pr);
2553 * Complete a call to either prison_free or prison_proc_free.
2556 prison_complete(void *context, int pending)
2558 struct prison *pr = context;
2560 sx_xlock(&allprison_lock);
2561 mtx_lock(&pr->pr_mtx);
2562 prison_deref(pr, pr->pr_uref
2563 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2564 : PD_LOCKED | PD_LIST_XLOCKED);
2568 * Remove a prison reference (usually). This internal version assumes no
2569 * mutexes are held, except perhaps the prison itself. If there are no more
2570 * references, release and delist the prison. On completion, the prison lock
2571 * and the allprison lock are both unlocked.
2574 prison_deref(struct prison *pr, int flags)
2576 struct prison *ppr, *tpr;
2579 if (!(flags & PD_LOCKED))
2580 mtx_lock(&pr->pr_mtx);
2582 if (flags & PD_DEUREF) {
2583 KASSERT(pr->pr_uref > 0,
2584 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2587 lasturef = pr->pr_uref == 0;
2590 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2593 if (flags & PD_DEREF) {
2594 KASSERT(pr->pr_ref > 0,
2595 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2600 mtx_unlock(&pr->pr_mtx);
2603 * Tell the modules if the last user reference was removed
2604 * (even it sticks around in dying state).
2607 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2608 sx_xlock(&allprison_lock);
2609 flags |= PD_LIST_XLOCKED;
2611 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2612 mtx_lock(&pr->pr_mtx);
2614 mtx_unlock(&pr->pr_mtx);
2617 /* If the prison still has references, nothing else to do. */
2619 if (flags & PD_LIST_SLOCKED)
2620 sx_sunlock(&allprison_lock);
2621 else if (flags & PD_LIST_XLOCKED)
2622 sx_xunlock(&allprison_lock);
2626 if (flags & PD_LIST_SLOCKED) {
2627 if (!sx_try_upgrade(&allprison_lock)) {
2628 sx_sunlock(&allprison_lock);
2629 sx_xlock(&allprison_lock);
2631 } else if (!(flags & PD_LIST_XLOCKED))
2632 sx_xlock(&allprison_lock);
2634 TAILQ_REMOVE(&allprison, pr, pr_list);
2635 LIST_REMOVE(pr, pr_sibling);
2636 ppr = pr->pr_parent;
2637 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2638 tpr->pr_childcount--;
2639 sx_xunlock(&allprison_lock);
2642 if (pr->pr_vnet != ppr->pr_vnet)
2643 vnet_destroy(pr->pr_vnet);
2645 if (pr->pr_root != NULL)
2647 mtx_destroy(&pr->pr_mtx);
2649 free(pr->pr_ip4, M_PRISON);
2652 free(pr->pr_ip6, M_PRISON);
2654 if (pr->pr_cpuset != NULL)
2655 cpuset_rel(pr->pr_cpuset);
2659 prison_racct_detach(pr);
2663 /* Removing a prison frees a reference on its parent. */
2665 mtx_lock(&pr->pr_mtx);
2666 flags = PD_DEREF | PD_DEUREF;
2671 prison_hold_locked(struct prison *pr)
2674 mtx_assert(&pr->pr_mtx, MA_OWNED);
2675 KASSERT(pr->pr_ref > 0,
2676 ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2681 prison_hold(struct prison *pr)
2684 mtx_lock(&pr->pr_mtx);
2685 prison_hold_locked(pr);
2686 mtx_unlock(&pr->pr_mtx);
2690 prison_proc_hold(struct prison *pr)
2693 mtx_lock(&pr->pr_mtx);
2694 KASSERT(pr->pr_uref > 0,
2695 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2697 mtx_unlock(&pr->pr_mtx);
2701 prison_proc_free(struct prison *pr)
2704 mtx_lock(&pr->pr_mtx);
2705 KASSERT(pr->pr_uref > 0,
2706 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2707 if (pr->pr_uref > 1)
2711 * Don't remove the last user reference in this context, which
2712 * is expected to be a process that is not only locked, but
2716 mtx_unlock(&pr->pr_mtx);
2717 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2720 mtx_unlock(&pr->pr_mtx);
2724 * Check if a jail supports the given address family.
2726 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2730 prison_check_af(struct ucred *cred, int af)
2735 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2737 pr = cred->cr_prison;
2739 /* Prisons with their own network stack are not limited. */
2740 if (prison_owns_vnet(cred))
2749 if (pr->pr_flags & PR_IP4)
2751 mtx_lock(&pr->pr_mtx);
2752 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2753 error = EAFNOSUPPORT;
2754 mtx_unlock(&pr->pr_mtx);
2760 if (pr->pr_flags & PR_IP6)
2762 mtx_lock(&pr->pr_mtx);
2763 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2764 error = EAFNOSUPPORT;
2765 mtx_unlock(&pr->pr_mtx);
2773 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2774 error = EAFNOSUPPORT;
2780 * Check if given address belongs to the jail referenced by cred (wrapper to
2781 * prison_check_ip[46]).
2783 * Returns 0 if jail doesn't restrict the address family or if address belongs
2784 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2785 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
2788 prison_if(struct ucred *cred, struct sockaddr *sa)
2791 struct sockaddr_in *sai;
2794 struct sockaddr_in6 *sai6;
2798 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2799 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2802 if (prison_owns_vnet(cred))
2807 switch (sa->sa_family)
2811 sai = (struct sockaddr_in *)sa;
2812 error = prison_check_ip4(cred, &sai->sin_addr);
2817 sai6 = (struct sockaddr_in6 *)sa;
2818 error = prison_check_ip6(cred, &sai6->sin6_addr);
2822 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2823 error = EAFNOSUPPORT;
2829 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2832 prison_check(struct ucred *cred1, struct ucred *cred2)
2835 return ((cred1->cr_prison == cred2->cr_prison ||
2836 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2840 * Return 1 if p2 is a child of p1, otherwise 0.
2843 prison_ischild(struct prison *pr1, struct prison *pr2)
2846 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2853 * Return 1 if the passed credential is in a jail, otherwise 0.
2856 jailed(struct ucred *cred)
2859 return (cred->cr_prison != &prison0);
2863 * Return 1 if the passed credential is in a jail and that jail does not
2864 * have its own virtual network stack, otherwise 0.
2867 jailed_without_vnet(struct ucred *cred)
2873 if (prison_owns_vnet(cred))
2881 * Return the correct hostname (domainname, et al) for the passed credential.
2884 getcredhostname(struct ucred *cred, char *buf, size_t size)
2889 * A NULL credential can be used to shortcut to the physical
2890 * system's hostname.
2892 pr = (cred != NULL) ? cred->cr_prison : &prison0;
2893 mtx_lock(&pr->pr_mtx);
2894 strlcpy(buf, pr->pr_hostname, size);
2895 mtx_unlock(&pr->pr_mtx);
2899 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2902 mtx_lock(&cred->cr_prison->pr_mtx);
2903 strlcpy(buf, cred->cr_prison->pr_domainname, size);
2904 mtx_unlock(&cred->cr_prison->pr_mtx);
2908 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2911 mtx_lock(&cred->cr_prison->pr_mtx);
2912 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2913 mtx_unlock(&cred->cr_prison->pr_mtx);
2917 getcredhostid(struct ucred *cred, unsigned long *hostid)
2920 mtx_lock(&cred->cr_prison->pr_mtx);
2921 *hostid = cred->cr_prison->pr_hostid;
2922 mtx_unlock(&cred->cr_prison->pr_mtx);
2927 * Determine whether the prison represented by cred owns
2928 * its vnet rather than having it inherited.
2930 * Returns 1 in case the prison owns the vnet, 0 otherwise.
2933 prison_owns_vnet(struct ucred *cred)
2937 * vnets cannot be added/removed after jail creation,
2938 * so no need to lock here.
2940 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2945 * Determine whether the subject represented by cred can "see"
2946 * status of a mount point.
2947 * Returns: 0 for permitted, ENOENT otherwise.
2948 * XXX: This function should be called cr_canseemount() and should be
2949 * placed in kern_prot.c.
2952 prison_canseemount(struct ucred *cred, struct mount *mp)
2958 pr = cred->cr_prison;
2959 if (pr->pr_enforce_statfs == 0)
2961 if (pr->pr_root->v_mount == mp)
2963 if (pr->pr_enforce_statfs == 2)
2966 * If jail's chroot directory is set to "/" we should be able to see
2967 * all mount-points from inside a jail.
2968 * This is ugly check, but this is the only situation when jail's
2969 * directory ends with '/'.
2971 if (strcmp(pr->pr_path, "/") == 0)
2973 len = strlen(pr->pr_path);
2975 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2978 * Be sure that we don't have situation where jail's root directory
2979 * is "/some/path" and mount point is "/some/pathpath".
2981 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2987 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2989 char jpath[MAXPATHLEN];
2993 pr = cred->cr_prison;
2994 if (pr->pr_enforce_statfs == 0)
2996 if (prison_canseemount(cred, mp) != 0) {
2997 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2998 strlcpy(sp->f_mntonname, "[restricted]",
2999 sizeof(sp->f_mntonname));
3002 if (pr->pr_root->v_mount == mp) {
3004 * Clear current buffer data, so we are sure nothing from
3005 * the valid path left there.
3007 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3008 *sp->f_mntonname = '/';
3012 * If jail's chroot directory is set to "/" we should be able to see
3013 * all mount-points from inside a jail.
3015 if (strcmp(pr->pr_path, "/") == 0)
3017 len = strlen(pr->pr_path);
3018 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3020 * Clear current buffer data, so we are sure nothing from
3021 * the valid path left there.
3023 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3024 if (*jpath == '\0') {
3025 /* Should never happen. */
3026 *sp->f_mntonname = '/';
3028 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3033 * Check with permission for a specific privilege is granted within jail. We
3034 * have a specific list of accepted privileges; the rest are denied.
3037 prison_priv_check(struct ucred *cred, int priv)
3045 * Privileges specific to prisons with a virtual network stack.
3046 * There might be a duplicate entry here in case the privilege
3047 * is only granted conditionally in the legacy jail case.
3052 * NFS-specific privileges.
3054 case PRIV_NFS_DAEMON:
3055 case PRIV_NFS_LOCKD:
3058 * Network stack privileges.
3060 case PRIV_NET_BRIDGE:
3063 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3064 case PRIV_NET_ROUTE:
3066 case PRIV_NET_SETIFMTU:
3067 case PRIV_NET_SETIFFLAGS:
3068 case PRIV_NET_SETIFCAP:
3069 case PRIV_NET_SETIFDESCR:
3070 case PRIV_NET_SETIFNAME :
3071 case PRIV_NET_SETIFMETRIC:
3072 case PRIV_NET_SETIFPHYS:
3073 case PRIV_NET_SETIFMAC:
3074 case PRIV_NET_ADDMULTI:
3075 case PRIV_NET_DELMULTI:
3076 case PRIV_NET_HWIOCTL:
3077 case PRIV_NET_SETLLADDR:
3078 case PRIV_NET_ADDIFGROUP:
3079 case PRIV_NET_DELIFGROUP:
3080 case PRIV_NET_IFCREATE:
3081 case PRIV_NET_IFDESTROY:
3082 case PRIV_NET_ADDIFADDR:
3083 case PRIV_NET_DELIFADDR:
3086 case PRIV_NET_SETIFVNET:
3087 case PRIV_NET_SETIFFIB:
3090 * 802.11-related privileges.
3092 case PRIV_NET80211_GETKEY:
3094 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3101 case PRIV_NETATM_CFG:
3102 case PRIV_NETATM_ADD:
3103 case PRIV_NETATM_DEL:
3104 case PRIV_NETATM_SET:
3107 * Bluetooth privileges.
3109 case PRIV_NETBLUETOOTH_RAW:
3113 * Netgraph and netgraph module privileges.
3115 case PRIV_NETGRAPH_CONTROL:
3117 case PRIV_NETGRAPH_TTY:
3121 * IPv4 and IPv6 privileges.
3123 case PRIV_NETINET_IPFW:
3124 case PRIV_NETINET_DIVERT:
3125 case PRIV_NETINET_PF:
3126 case PRIV_NETINET_DUMMYNET:
3127 case PRIV_NETINET_CARP:
3128 case PRIV_NETINET_MROUTE:
3129 case PRIV_NETINET_RAW:
3130 case PRIV_NETINET_ADDRCTRL6:
3131 case PRIV_NETINET_ND6:
3132 case PRIV_NETINET_SCOPE6:
3133 case PRIV_NETINET_ALIFETIME6:
3134 case PRIV_NETINET_IPSEC:
3135 case PRIV_NETINET_BINDANY:
3150 * No default: or deny here.
3151 * In case of no permit fall through to next switch().
3153 if (cred->cr_prison->pr_flags & PR_VNET)
3161 * Allow ktrace privileges for root in jail.
3167 * Allow jailed processes to configure audit identity and
3168 * submit audit records (login, etc). In the future we may
3169 * want to further refine the relationship between audit and
3172 case PRIV_AUDIT_GETAUDIT:
3173 case PRIV_AUDIT_SETAUDIT:
3174 case PRIV_AUDIT_SUBMIT:
3178 * Allow jailed processes to manipulate process UNIX
3179 * credentials in any way they see fit.
3181 case PRIV_CRED_SETUID:
3182 case PRIV_CRED_SETEUID:
3183 case PRIV_CRED_SETGID:
3184 case PRIV_CRED_SETEGID:
3185 case PRIV_CRED_SETGROUPS:
3186 case PRIV_CRED_SETREUID:
3187 case PRIV_CRED_SETREGID:
3188 case PRIV_CRED_SETRESUID:
3189 case PRIV_CRED_SETRESGID:
3192 * Jail implements visibility constraints already, so allow
3193 * jailed root to override uid/gid-based constraints.
3195 case PRIV_SEEOTHERGIDS:
3196 case PRIV_SEEOTHERUIDS:
3199 * Jail implements inter-process debugging limits already, so
3200 * allow jailed root various debugging privileges.
3202 case PRIV_DEBUG_DIFFCRED:
3203 case PRIV_DEBUG_SUGID:
3204 case PRIV_DEBUG_UNPRIV:
3207 * Allow jail to set various resource limits and login
3208 * properties, and for now, exceed process resource limits.
3210 case PRIV_PROC_LIMIT:
3211 case PRIV_PROC_SETLOGIN:
3212 case PRIV_PROC_SETRLIMIT:
3215 * System V and POSIX IPC privileges are granted in jail.
3218 case PRIV_IPC_WRITE:
3219 case PRIV_IPC_ADMIN:
3220 case PRIV_IPC_MSGSIZE:
3224 * Jail operations within a jail work on child jails.
3226 case PRIV_JAIL_ATTACH:
3228 case PRIV_JAIL_REMOVE:
3231 * Jail implements its own inter-process limits, so allow
3232 * root processes in jail to change scheduling on other
3233 * processes in the same jail. Likewise for signalling.
3235 case PRIV_SCHED_DIFFCRED:
3236 case PRIV_SCHED_CPUSET:
3237 case PRIV_SIGNAL_DIFFCRED:
3238 case PRIV_SIGNAL_SUGID:
3241 * Allow jailed processes to write to sysctls marked as jail
3244 case PRIV_SYSCTL_WRITEJAIL:
3247 * Allow root in jail to manage a variety of quota
3248 * properties. These should likely be conditional on a
3249 * configuration option.
3251 case PRIV_VFS_GETQUOTA:
3252 case PRIV_VFS_SETQUOTA:
3255 * Since Jail relies on chroot() to implement file system
3256 * protections, grant many VFS privileges to root in jail.
3257 * Be careful to exclude mount-related and NFS-related
3261 case PRIV_VFS_WRITE:
3262 case PRIV_VFS_ADMIN:
3264 case PRIV_VFS_LOOKUP:
3265 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3266 case PRIV_VFS_CHFLAGS_DEV:
3267 case PRIV_VFS_CHOWN:
3268 case PRIV_VFS_CHROOT:
3269 case PRIV_VFS_RETAINSUGID:
3270 case PRIV_VFS_FCHROOT:
3272 case PRIV_VFS_SETGID:
3274 case PRIV_VFS_STICKYFILE:
3277 * As in the non-jail case, non-root users are expected to be
3278 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3279 * exists in the jail and they have permission to access it).
3281 case PRIV_KMEM_READ:
3285 * Depending on the global setting, allow privilege of
3286 * setting system flags.
3288 case PRIV_VFS_SYSFLAGS:
3289 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3295 * Depending on the global setting, allow privilege of
3296 * mounting/unmounting file systems.
3298 case PRIV_VFS_MOUNT:
3299 case PRIV_VFS_UNMOUNT:
3300 case PRIV_VFS_MOUNT_NONUSER:
3301 case PRIV_VFS_MOUNT_OWNER:
3302 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3303 cred->cr_prison->pr_enforce_statfs < 2)
3309 * Conditionally allow jailed root to bind reserved ports.
3311 case PRIV_NETINET_RESERVEDPORT:
3312 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3318 * Allow jailed root to reuse in-use ports.
3320 case PRIV_NETINET_REUSEPORT:
3324 * Allow jailed root to set certain IPv4/6 (option) headers.
3326 case PRIV_NETINET_SETHDROPTS:
3330 * Conditionally allow creating raw sockets in jail.
3332 case PRIV_NETINET_RAW:
3333 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3339 * Since jail implements its own visibility limits on netstat
3340 * sysctls, allow getcred. This allows identd to work in
3343 case PRIV_NETINET_GETCRED:
3347 * Allow jailed root to set loginclass.
3349 case PRIV_PROC_SETLOGINCLASS:
3354 * In all remaining cases, deny the privilege request. This
3355 * includes almost all network privileges, many system
3356 * configuration privileges.
3363 * Return the part of pr2's name that is relative to pr1, or the whole name
3364 * if it does not directly follow.
3368 prison_name(struct prison *pr1, struct prison *pr2)
3372 /* Jails see themselves as "0" (if they see themselves at all). */
3375 name = pr2->pr_name;
3376 if (prison_ischild(pr1, pr2)) {
3378 * pr1 isn't locked (and allprison_lock may not be either)
3379 * so its length can't be counted on. But the number of dots
3380 * can be counted on - and counted.
3382 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3383 name = strchr(name, '.') + 1;
3389 * Return the part of pr2's path that is relative to pr1, or the whole path
3390 * if it does not directly follow.
3393 prison_path(struct prison *pr1, struct prison *pr2)
3395 char *path1, *path2;
3398 path1 = pr1->pr_path;
3399 path2 = pr2->pr_path;
3400 if (!strcmp(path1, "/"))
3402 len1 = strlen(path1);
3403 if (strncmp(path1, path2, len1))
3405 if (path2[len1] == '\0')
3407 if (path2[len1] == '/')
3408 return (path2 + len1);
3414 * Jail-related sysctls.
3416 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3420 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3423 struct prison *pr, *cpr;
3425 struct in_addr *ip4 = NULL;
3429 struct in6_addr *ip6 = NULL;
3434 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3435 pr = req->td->td_ucred->cr_prison;
3437 sx_slock(&allprison_lock);
3438 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3439 #if defined(INET) || defined(INET6)
3442 mtx_lock(&cpr->pr_mtx);
3444 if (cpr->pr_ip4s > 0) {
3445 if (ip4s < cpr->pr_ip4s) {
3446 ip4s = cpr->pr_ip4s;
3447 mtx_unlock(&cpr->pr_mtx);
3448 ip4 = realloc(ip4, ip4s *
3449 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3452 bcopy(cpr->pr_ip4, ip4,
3453 cpr->pr_ip4s * sizeof(struct in_addr));
3457 if (cpr->pr_ip6s > 0) {
3458 if (ip6s < cpr->pr_ip6s) {
3459 ip6s = cpr->pr_ip6s;
3460 mtx_unlock(&cpr->pr_mtx);
3461 ip6 = realloc(ip6, ip6s *
3462 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3465 bcopy(cpr->pr_ip6, ip6,
3466 cpr->pr_ip6s * sizeof(struct in6_addr));
3469 if (cpr->pr_ref == 0) {
3470 mtx_unlock(&cpr->pr_mtx);
3473 bzero(xp, sizeof(*xp));
3474 xp->pr_version = XPRISON_VERSION;
3475 xp->pr_id = cpr->pr_id;
3476 xp->pr_state = cpr->pr_uref > 0
3477 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3478 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3479 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3480 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3482 xp->pr_ip4s = cpr->pr_ip4s;
3485 xp->pr_ip6s = cpr->pr_ip6s;
3487 mtx_unlock(&cpr->pr_mtx);
3488 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3492 if (xp->pr_ip4s > 0) {
3493 error = SYSCTL_OUT(req, ip4,
3494 xp->pr_ip4s * sizeof(struct in_addr));
3500 if (xp->pr_ip6s > 0) {
3501 error = SYSCTL_OUT(req, ip6,
3502 xp->pr_ip6s * sizeof(struct in6_addr));
3508 sx_sunlock(&allprison_lock);
3519 SYSCTL_OID(_security_jail, OID_AUTO, list,
3520 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3521 sysctl_jail_list, "S", "List of active jails");
3524 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3528 injail = jailed(req->td->td_ucred);
3529 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3534 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3535 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3536 sysctl_jail_jailed, "I", "Process in jail?");
3539 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3541 int error, havevnet;
3543 struct ucred *cred = req->td->td_ucred;
3545 havevnet = jailed(cred) && prison_owns_vnet(cred);
3549 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3554 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3555 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3556 sysctl_jail_vnet, "I", "Jail owns VNET?");
3558 #if defined(INET) || defined(INET6)
3559 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3560 &jail_max_af_ips, 0,
3561 "Number of IP addresses a jail may have at most per address family (deprecated)");
3565 * Default parameters for jail(2) compatibility. For historical reasons,
3566 * the sysctl names have varying similarity to the parameter names. Prisons
3567 * just see their own parameters, and can't change them.
3570 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3573 int allow, error, i;
3575 pr = req->td->td_ucred->cr_prison;
3576 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3578 /* Get the current flag value, and convert it to a boolean. */
3579 i = (allow & arg2) ? 1 : 0;
3582 error = sysctl_handle_int(oidp, &i, 0, req);
3583 if (error || !req->newptr)
3589 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3592 mtx_lock(&prison0.pr_mtx);
3593 jail_default_allow = (jail_default_allow & ~arg2) | i;
3594 mtx_unlock(&prison0.pr_mtx);
3598 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3599 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3600 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3601 "Processes in jail can set their hostnames (deprecated)");
3602 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3603 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3604 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3605 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3606 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3607 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3608 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3609 "Processes in jail can use System V IPC primitives (deprecated)");
3610 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3611 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3612 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3613 "Prison root can create raw sockets (deprecated)");
3614 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3615 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3616 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3617 "Processes in jail can alter system file flags (deprecated)");
3618 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3619 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3620 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3621 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3622 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
3623 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3624 NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
3625 "Processes in jail can mount the devfs file system (deprecated)");
3626 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
3627 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3628 NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
3629 "Processes in jail can mount the fdescfs file system (deprecated)");
3630 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
3631 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3632 NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
3633 "Processes in jail can mount the nullfs file system (deprecated)");
3634 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
3635 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3636 NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
3637 "Processes in jail can mount the procfs file system (deprecated)");
3638 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
3639 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3640 NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
3641 "Processes in jail can mount the linprocfs file system (deprecated)");
3642 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
3643 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3644 NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
3645 "Processes in jail can mount the linsysfs file system (deprecated)");
3646 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
3647 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3648 NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
3649 "Processes in jail can mount the tmpfs file system (deprecated)");
3650 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
3651 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3652 NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
3653 "Processes in jail can mount the zfs file system (deprecated)");
3656 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3661 pr = req->td->td_ucred->cr_prison;
3662 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3663 error = sysctl_handle_int(oidp, &level, 0, req);
3664 if (error || !req->newptr)
3666 *(int *)arg1 = level;
3670 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3671 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3672 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3673 sysctl_jail_default_level, "I",
3674 "Processes in jail cannot see all mounted file systems (deprecated)");
3676 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3677 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3678 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3679 sysctl_jail_default_level, "I",
3680 "Ruleset for the devfs filesystem in jail (deprecated)");
3683 * Nodes to describe jail parameters. Maximum length of string parameters
3684 * is returned in the string itself, and the other parameters exist merely
3685 * to make themselves and their types known.
3687 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3691 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3698 switch (oidp->oid_kind & CTLTYPE)
3704 if (!(req->flags & SCTL_MASK32))
3706 return (SYSCTL_OUT(req, &l, sizeof(l)));
3710 return (SYSCTL_OUT(req, &i, sizeof(i)));
3711 case CTLTYPE_STRING:
3712 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3714 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3715 case CTLTYPE_STRUCT:
3717 return (SYSCTL_OUT(req, &s, sizeof(s)));
3723 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3724 * jail creation time but cannot be changed in an existing jail.
3726 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3727 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3728 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3729 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3730 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3731 "I", "Jail secure level");
3732 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3733 "Jail value for kern.osreldate and uname -K");
3734 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3735 "Jail value for kern.osrelease and uname -r");
3736 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3737 "I", "Jail cannot see all mounted file systems");
3738 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3739 "I", "Ruleset for in-jail devfs mounts");
3740 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3741 "B", "Jail persistence");
3743 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3744 "E,jailsys", "Virtual network stack");
3746 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3747 "B", "Jail is in the process of shutting down");
3749 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3750 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3751 "I", "Current number of child jails");
3752 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3753 "I", "Maximum number of child jails");
3755 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3756 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3758 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3759 "Jail NIS domainname");
3760 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3762 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3763 "LU", "Jail host ID");
3765 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3766 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3769 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3770 "Jail IPv4 address virtualization");
3771 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3772 "S,in_addr,a", "Jail IPv4 addresses");
3773 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3774 "B", "Do (not) use IPv4 source address selection rather than the "
3775 "primary jail IPv4 address.");
3778 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3779 "Jail IPv6 address virtualization");
3780 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3781 "S,in6_addr,a", "Jail IPv6 addresses");
3782 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3783 "B", "Do (not) use IPv6 source address selection rather than the "
3784 "primary jail IPv6 address.");
3787 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3788 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3789 "B", "Jail may set hostname");
3790 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3791 "B", "Jail may use SYSV IPC");
3792 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3793 "B", "Jail may create raw sockets");
3794 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3795 "B", "Jail may alter system file flags");
3796 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3797 "B", "Jail may set file quotas");
3798 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3799 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3800 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3801 "B", "Jail may bind sockets to reserved ports");
3803 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3804 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3805 "B", "Jail may mount/unmount jail-friendly file systems in general");
3806 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
3807 "B", "Jail may mount the devfs file system");
3808 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
3809 "B", "Jail may mount the fdescfs file system");
3810 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
3811 "B", "Jail may mount the nullfs file system");
3812 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
3813 "B", "Jail may mount the procfs file system");
3814 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
3815 "B", "Jail may mount the linprocfs file system");
3816 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
3817 "B", "Jail may mount the linsysfs file system");
3818 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
3819 "B", "Jail may mount the tmpfs file system");
3820 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
3821 "B", "Jail may mount the zfs file system");
3825 prison_racct_foreach(void (*callback)(struct racct *racct,
3826 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3827 void *arg2, void *arg3)
3829 struct prison_racct *prr;
3831 ASSERT_RACCT_ENABLED();
3833 sx_slock(&allprison_lock);
3836 LIST_FOREACH(prr, &allprison_racct, prr_next)
3837 (callback)(prr->prr_racct, arg2, arg3);
3840 sx_sunlock(&allprison_lock);
3843 static struct prison_racct *
3844 prison_racct_find_locked(const char *name)
3846 struct prison_racct *prr;
3848 ASSERT_RACCT_ENABLED();
3849 sx_assert(&allprison_lock, SA_XLOCKED);
3851 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3854 LIST_FOREACH(prr, &allprison_racct, prr_next) {
3855 if (strcmp(name, prr->prr_name) != 0)
3858 /* Found prison_racct with a matching name? */
3859 prison_racct_hold(prr);
3863 /* Add new prison_racct. */
3864 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3865 racct_create(&prr->prr_racct);
3867 strcpy(prr->prr_name, name);
3868 refcount_init(&prr->prr_refcount, 1);
3869 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3874 struct prison_racct *
3875 prison_racct_find(const char *name)
3877 struct prison_racct *prr;
3879 ASSERT_RACCT_ENABLED();
3881 sx_xlock(&allprison_lock);
3882 prr = prison_racct_find_locked(name);
3883 sx_xunlock(&allprison_lock);
3888 prison_racct_hold(struct prison_racct *prr)
3891 ASSERT_RACCT_ENABLED();
3893 refcount_acquire(&prr->prr_refcount);
3897 prison_racct_free_locked(struct prison_racct *prr)
3900 ASSERT_RACCT_ENABLED();
3901 sx_assert(&allprison_lock, SA_XLOCKED);
3903 if (refcount_release(&prr->prr_refcount)) {
3904 racct_destroy(&prr->prr_racct);
3905 LIST_REMOVE(prr, prr_next);
3906 free(prr, M_PRISON_RACCT);
3911 prison_racct_free(struct prison_racct *prr)
3915 ASSERT_RACCT_ENABLED();
3916 sx_assert(&allprison_lock, SA_UNLOCKED);
3918 old = prr->prr_refcount;
3919 if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
3922 sx_xlock(&allprison_lock);
3923 prison_racct_free_locked(prr);
3924 sx_xunlock(&allprison_lock);
3928 prison_racct_attach(struct prison *pr)
3930 struct prison_racct *prr;
3932 ASSERT_RACCT_ENABLED();
3933 sx_assert(&allprison_lock, SA_XLOCKED);
3935 prr = prison_racct_find_locked(pr->pr_name);
3936 KASSERT(prr != NULL, ("cannot find prison_racct"));
3938 pr->pr_prison_racct = prr;
3942 * Handle jail renaming. From the racct point of view, renaming means
3943 * moving from one prison_racct to another.
3946 prison_racct_modify(struct prison *pr)
3950 struct prison_racct *oldprr;
3952 ASSERT_RACCT_ENABLED();
3954 sx_slock(&allproc_lock);
3955 sx_xlock(&allprison_lock);
3957 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
3958 sx_xunlock(&allprison_lock);
3959 sx_sunlock(&allproc_lock);
3963 oldprr = pr->pr_prison_racct;
3964 pr->pr_prison_racct = NULL;
3966 prison_racct_attach(pr);
3969 * Move resource utilisation records.
3971 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
3974 * Force rctl to reattach rules to processes.
3976 FOREACH_PROC_IN_SYSTEM(p) {
3978 cred = crhold(p->p_ucred);
3980 racct_proc_ucred_changed(p, cred, cred);
3984 sx_sunlock(&allproc_lock);
3985 prison_racct_free_locked(oldprr);
3986 sx_xunlock(&allprison_lock);
3990 prison_racct_detach(struct prison *pr)
3993 ASSERT_RACCT_ENABLED();
3994 sx_assert(&allprison_lock, SA_UNLOCKED);
3996 if (pr->pr_prison_racct == NULL)
3998 prison_racct_free(pr->pr_prison_racct);
3999 pr->pr_prison_racct = NULL;
4006 db_show_prison(struct prison *pr)
4009 #if defined(INET) || defined(INET6)
4014 char ip4buf[INET_ADDRSTRLEN];
4017 char ip6buf[INET6_ADDRSTRLEN];
4020 db_printf("prison %p:\n", pr);
4021 db_printf(" jid = %d\n", pr->pr_id);
4022 db_printf(" name = %s\n", pr->pr_name);
4023 db_printf(" parent = %p\n", pr->pr_parent);
4024 db_printf(" ref = %d\n", pr->pr_ref);
4025 db_printf(" uref = %d\n", pr->pr_uref);
4026 db_printf(" path = %s\n", pr->pr_path);
4027 db_printf(" cpuset = %d\n", pr->pr_cpuset
4028 ? pr->pr_cpuset->cs_id : -1);
4030 db_printf(" vnet = %p\n", pr->pr_vnet);
4032 db_printf(" root = %p\n", pr->pr_root);
4033 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4034 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4035 db_printf(" children.max = %d\n", pr->pr_childmax);
4036 db_printf(" children.cur = %d\n", pr->pr_childcount);
4037 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4038 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4039 db_printf(" flags = 0x%x", pr->pr_flags);
4040 for (fi = 0; fi < nitems(pr_flag_names); fi++)
4041 if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4042 db_printf(" %s", pr_flag_names[fi]);
4043 for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
4044 jsf = pr->pr_flags &
4045 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4046 db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4047 pr_flag_jailsys[fi].disable &&
4048 (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4049 : (jsf == pr_flag_jailsys[fi].new) ? "new"
4052 db_printf(" allow = 0x%x", pr->pr_allow);
4053 for (fi = 0; fi < nitems(pr_allow_names); fi++)
4054 if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4055 db_printf(" %s", pr_allow_names[fi]);
4057 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4058 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4059 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4060 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4061 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4063 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4064 for (ii = 0; ii < pr->pr_ip4s; ii++)
4065 db_printf(" %s %s\n",
4066 ii == 0 ? "ip4.addr =" : " ",
4067 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4070 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4071 for (ii = 0; ii < pr->pr_ip6s; ii++)
4072 db_printf(" %s %s\n",
4073 ii == 0 ? "ip6.addr =" : " ",
4074 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4078 DB_SHOW_COMMAND(prison, db_show_prison_command)
4084 * Show all prisons in the list, and prison0 which is not
4087 db_show_prison(&prison0);
4088 if (!db_pager_quit) {
4089 TAILQ_FOREACH(pr, &allprison, pr_list) {
4101 /* Look for a prison with the ID and with references. */
4102 TAILQ_FOREACH(pr, &allprison, pr_list)
4103 if (pr->pr_id == addr && pr->pr_ref > 0)
4106 /* Look again, without requiring a reference. */
4107 TAILQ_FOREACH(pr, &allprison, pr_list)
4108 if (pr->pr_id == addr)
4111 /* Assume address points to a valid prison. */
4112 pr = (struct prison *)addr;