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>
72 #include <netinet6/in6_var.h>
76 #include <security/mac/mac_framework.h>
78 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000"
80 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
81 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
83 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
86 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
88 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
92 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
94 #define _PR_IP_SADDRSEL 0
98 /* prison0 describes what is "real" about the system. */
99 struct prison prison0 = {
105 .pr_securelevel = -1,
107 .pr_childmax = JAIL_MAX,
108 .pr_hostuuid = DEFAULT_HOSTUUID,
109 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
111 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
113 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
115 .pr_allow = PR_ALLOW_ALL,
117 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
119 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
120 struct sx allprison_lock;
121 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
122 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
123 LIST_HEAD(, prison_racct) allprison_racct;
126 static int do_jail_attach(struct thread *td, struct prison *pr);
127 static void prison_complete(void *context, int pending);
128 static void prison_deref(struct prison *pr, int flags);
129 static char *prison_path(struct prison *pr1, struct prison *pr2);
130 static void prison_remove_one(struct prison *pr);
132 static void prison_racct_attach(struct prison *pr);
133 static void prison_racct_modify(struct prison *pr);
134 static void prison_racct_detach(struct prison *pr);
137 static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
138 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
141 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
142 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
145 /* Flags for prison_deref */
146 #define PD_DEREF 0x01
147 #define PD_DEUREF 0x02
148 #define PD_LOCKED 0x04
149 #define PD_LIST_SLOCKED 0x08
150 #define PD_LIST_XLOCKED 0x10
153 * Parameter names corresponding to PR_* flag values. Size values are for kvm
154 * as we cannot figure out the size of a sparse array, or an array without a
157 static char *pr_flag_names[] = {
160 [7] = "ip4.saddrsel",
163 [8] = "ip6.saddrsel",
166 const size_t pr_flag_names_size = sizeof(pr_flag_names);
168 static char *pr_flag_nonames[] = {
171 [7] = "ip4.nosaddrsel",
174 [8] = "ip6.nosaddrsel",
177 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
179 struct jailsys_flags {
183 } pr_flag_jailsys[] = {
184 { "host", 0, PR_HOST },
186 { "vnet", 0, PR_VNET },
189 { "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
192 { "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
195 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
197 static char *pr_allow_names[] = {
198 "allow.set_hostname",
206 "allow.mount.nullfs",
208 "allow.mount.procfs",
210 "allow.mount.fdescfs",
211 "allow.mount.linprocfs",
212 "allow.mount.linsysfs",
214 const size_t pr_allow_names_size = sizeof(pr_allow_names);
216 static char *pr_allow_nonames[] = {
217 "allow.noset_hostname",
219 "allow.noraw_sockets",
224 "allow.mount.nodevfs",
225 "allow.mount.nonullfs",
227 "allow.mount.noprocfs",
228 "allow.mount.notmpfs",
229 "allow.mount.nofdescfs",
230 "allow.mount.nolinprocfs",
231 "allow.mount.nolinsysfs",
233 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
235 #define JAIL_DEFAULT_ALLOW PR_ALLOW_SET_HOSTNAME
236 #define JAIL_DEFAULT_ENFORCE_STATFS 2
237 #define JAIL_DEFAULT_DEVFS_RSNUM 0
238 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
239 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
240 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
241 #if defined(INET) || defined(INET6)
242 static unsigned jail_max_af_ips = 255;
246 * Initialize the parts of prison0 that can't be static-initialized with
247 * constants. This is called from proc0_init() after creating thread0 cpuset.
253 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
254 prison0.pr_osreldate = osreldate;
255 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
260 qcmp_v4(const void *ip1, const void *ip2)
265 * We need to compare in HBO here to get the list sorted as expected
266 * by the result of the code. Sorting NBO addresses gives you
267 * interesting results. If you do not understand, do not try.
269 iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
270 iab = ntohl(((const struct in_addr *)ip2)->s_addr);
273 * Do not simply return the difference of the two numbers, the int is
287 qcmp_v6(const void *ip1, const void *ip2)
289 const struct in6_addr *ia6a, *ia6b;
292 ia6a = (const struct in6_addr *)ip1;
293 ia6b = (const struct in6_addr *)ip2;
296 for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
297 if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
299 else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
312 sys_jail(struct thread *td, struct jail_args *uap)
318 error = copyin(uap->jail, &version, sizeof(uint32_t));
327 /* FreeBSD single IPv4 jails. */
328 bzero(&j, sizeof(struct jail));
329 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
332 j.version = j0.version;
334 j.hostname = j0.hostname;
335 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
341 * Version 1 was used by multi-IPv4 jail implementations
342 * that never made it into the official kernel.
346 case 2: /* JAIL_API_VERSION */
347 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
348 error = copyin(uap->jail, &j, sizeof(struct jail));
354 /* Sci-Fi jails are not supported, sorry. */
357 return (kern_jail(td, &j));
361 kern_jail(struct thread *td, struct jail *j)
363 struct iovec optiov[2 * (4
364 + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
373 char *u_path, *u_hostname, *u_name;
376 struct in_addr *u_ip4;
379 struct in6_addr *u_ip6;
382 int error, enforce_statfs, fi;
384 bzero(&optiov, sizeof(optiov));
385 opt.uio_iov = optiov;
389 opt.uio_segflg = UIO_SYSSPACE;
390 opt.uio_rw = UIO_READ;
393 /* Set permissions for top-level jails from sysctls. */
394 if (!jailed(td->td_ucred)) {
395 for (fi = 0; fi < sizeof(pr_allow_names) /
396 sizeof(pr_allow_names[0]); fi++) {
397 optiov[opt.uio_iovcnt].iov_base =
398 (jail_default_allow & (1 << fi))
399 ? pr_allow_names[fi] : pr_allow_nonames[fi];
400 optiov[opt.uio_iovcnt].iov_len =
401 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
404 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
405 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
407 enforce_statfs = jail_default_enforce_statfs;
408 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
409 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
413 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
415 ip4s = (j->version == 0) ? 1 : j->ip4s;
416 if (ip4s > jail_max_af_ips)
418 tmplen += ip4s * sizeof(struct in_addr);
424 if (j->ip6s > jail_max_af_ips)
426 tmplen += j->ip6s * sizeof(struct in6_addr);
431 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
432 u_hostname = u_path + MAXPATHLEN;
433 u_name = u_hostname + MAXHOSTNAMELEN;
435 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
439 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
441 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
444 optiov[opt.uio_iovcnt].iov_base = "path";
445 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
447 optiov[opt.uio_iovcnt].iov_base = u_path;
448 error = copyinstr(j->path, u_path, MAXPATHLEN,
449 &optiov[opt.uio_iovcnt].iov_len);
451 free(u_path, M_TEMP);
455 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
456 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
458 optiov[opt.uio_iovcnt].iov_base = u_hostname;
459 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
460 &optiov[opt.uio_iovcnt].iov_len);
462 free(u_path, M_TEMP);
466 if (j->jailname != NULL) {
467 optiov[opt.uio_iovcnt].iov_base = "name";
468 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
470 optiov[opt.uio_iovcnt].iov_base = u_name;
471 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
472 &optiov[opt.uio_iovcnt].iov_len);
474 free(u_path, M_TEMP);
480 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
481 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
483 optiov[opt.uio_iovcnt].iov_base = u_ip4;
484 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
486 u_ip4->s_addr = j->ip4s;
488 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
490 free(u_path, M_TEMP);
497 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
498 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
500 optiov[opt.uio_iovcnt].iov_base = u_ip6;
501 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
502 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
504 free(u_path, M_TEMP);
509 KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
510 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
511 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
512 free(u_path, M_TEMP);
518 * struct jail_set_args {
519 * struct iovec *iovp;
520 * unsigned int iovcnt;
525 sys_jail_set(struct thread *td, struct jail_set_args *uap)
530 /* Check that we have an even number of iovecs. */
534 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
537 error = kern_jail_set(td, auio, uap->flags);
543 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
550 struct in6_addr *ip6;
553 struct vfsoptlist *opts;
554 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
556 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
557 char *g_path, *osrelstr;
558 #if defined(INET) || defined(INET6)
563 size_t namelen, onamelen, pnamelen;
564 int born, created, cuflags, descend, enforce;
565 int error, errmsg_len, errmsg_pos;
566 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
567 int fi, jid, jsys, len, level;
568 int childmax, osreldt, rsnum, slevel;
569 int fullpath_disabled;
570 #if defined(INET) || defined(INET6)
579 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
583 error = priv_check(td, PRIV_JAIL_SET);
584 if (!error && (flags & JAIL_ATTACH))
585 error = priv_check(td, PRIV_JAIL_ATTACH);
588 mypr = td->td_ucred->cr_prison;
589 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
591 if (flags & ~JAIL_SET_MASK)
595 * Check all the parameters before committing to anything. Not all
596 * errors can be caught early, but we may as well try. Also, this
597 * takes care of some expensive stuff (path lookup) before getting
598 * the allprison lock.
600 * XXX Jails are not filesystems, and jail parameters are not mount
601 * options. But it makes more sense to re-use the vfsopt code
602 * than duplicate it under a different name.
604 error = vfs_buildopts(optuio, &opts);
615 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
618 vfs_opterror(opts, "no valid operation (create or update)");
622 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
628 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
637 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
645 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
650 else if (enforce < 0 || enforce > 2) {
656 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
664 pr_flags = ch_flags = 0;
665 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
667 if (pr_flag_names[fi] == NULL)
669 vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
670 vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
672 ch_flags |= pr_flags;
673 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
675 error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
682 case JAIL_SYS_DISABLE:
683 if (!pr_flag_jailsys[fi].disable) {
687 pr_flags |= pr_flag_jailsys[fi].disable;
690 pr_flags |= pr_flag_jailsys[fi].new;
692 case JAIL_SYS_INHERIT:
699 pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
701 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
702 && !(pr_flags & PR_PERSIST)) {
704 vfs_opterror(opts, "new jail must persist or attach");
708 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
710 vfs_opterror(opts, "vnet cannot be changed after creation");
715 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
717 vfs_opterror(opts, "ip4 cannot be changed after creation");
722 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
724 vfs_opterror(opts, "ip6 cannot be changed after creation");
729 pr_allow = ch_allow = 0;
730 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
732 vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
733 vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
735 ch_allow |= pr_allow;
737 error = vfs_getopt(opts, "name", (void **)&name, &len);
743 if (len == 0 || name[len - 1] != '\0') {
747 if (len > MAXHOSTNAMELEN) {
748 error = ENAMETOOLONG;
753 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
761 if (len == 0 || host[len - 1] != '\0') {
765 if (len > MAXHOSTNAMELEN) {
766 error = ENAMETOOLONG;
771 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
779 if (len == 0 || domain[len - 1] != '\0') {
783 if (len > MAXHOSTNAMELEN) {
784 error = ENAMETOOLONG;
789 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
797 if (len == 0 || uuid[len - 1] != '\0') {
801 if (len > HOSTUUIDLEN) {
802 error = ENAMETOOLONG;
807 #ifdef COMPAT_FREEBSD32
808 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
811 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
815 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
827 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
832 else if (ip4s & (sizeof(*ip4) - 1)) {
836 ch_flags |= PR_IP4_USER | PR_IP4_DISABLE;
838 pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
840 pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
841 ip4s /= sizeof(*ip4);
842 if (ip4s > jail_max_af_ips) {
844 vfs_opterror(opts, "too many IPv4 addresses");
847 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
848 bcopy(op, ip4, ip4s * sizeof(*ip4));
850 * IP addresses are all sorted but ip[0] to preserve
851 * the primary IP address as given from userland.
852 * This special IP is used for unbound outgoing
853 * connections as well for "loopback" traffic in case
854 * source address selection cannot find any more fitting
855 * address to connect from.
858 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
860 * Check for duplicate addresses and do some simple
861 * zero and broadcast checks. If users give other bogus
862 * addresses it is their problem.
864 * We do not have to care about byte order for these
865 * checks so we will do them in NBO.
867 for (ii = 0; ii < ip4s; ii++) {
868 if (ip4[ii].s_addr == INADDR_ANY ||
869 ip4[ii].s_addr == INADDR_BROADCAST) {
874 (ip4[0].s_addr == ip4[ii+1].s_addr ||
875 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
885 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
890 else if (ip6s & (sizeof(*ip6) - 1)) {
894 ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
896 pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
898 pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
899 ip6s /= sizeof(*ip6);
900 if (ip6s > jail_max_af_ips) {
902 vfs_opterror(opts, "too many IPv6 addresses");
905 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
906 bcopy(op, ip6, ip6s * sizeof(*ip6));
908 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
909 for (ii = 0; ii < ip6s; ii++) {
910 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
915 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
916 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
926 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
927 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
930 "vnet jails cannot have IP address restrictions");
935 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
941 if (flags & JAIL_UPDATE) {
944 "osrelease cannot be changed after creation");
947 if (len == 0 || len >= OSRELEASELEN) {
950 "osrelease string must be 1-%d bytes long",
956 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
962 if (flags & JAIL_UPDATE) {
965 "osreldate cannot be changed after creation");
970 vfs_opterror(opts, "osreldate cannot be 0");
975 fullpath_disabled = 0;
977 error = vfs_getopt(opts, "path", (void **)&path, &len);
983 if (flags & JAIL_UPDATE) {
986 "path cannot be changed after creation");
989 if (len == 0 || path[len - 1] != '\0') {
993 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
999 NDFREE(&nd, NDF_ONLY_PNBUF);
1000 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1001 strlcpy(g_path, path, MAXPATHLEN);
1002 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1005 else if (error == ENODEV) {
1006 /* proceed if sysctl debug.disablefullpath == 1 */
1007 fullpath_disabled = 1;
1008 if (len < 2 || (len == 2 && path[0] == '/'))
1011 /* exit on other errors */
1014 if (root->v_type != VDIR) {
1019 VOP_UNLOCK(root, 0);
1020 if (fullpath_disabled) {
1021 /* Leave room for a real-root full pathname. */
1022 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
1023 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
1024 error = ENAMETOOLONG;
1032 * Find the specified jail, or at least its parent.
1033 * This abuses the file error codes ENOENT and EEXIST.
1037 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1038 namelc = strrchr(name, '.');
1039 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1043 sx_xlock(&allprison_lock);
1046 * See if a requested jid already exists. There is an
1047 * information leak here if the jid exists but is not within
1048 * the caller's jail hierarchy. Jail creators will get EEXIST
1049 * even though they cannot see the jail, and CREATE | UPDATE
1050 * will return ENOENT which is not normally a valid error.
1054 vfs_opterror(opts, "negative jid");
1055 goto done_unlock_list;
1057 pr = prison_find(jid);
1059 ppr = pr->pr_parent;
1060 /* Create: jid must not exist. */
1061 if (cuflags == JAIL_CREATE) {
1062 mtx_unlock(&pr->pr_mtx);
1064 vfs_opterror(opts, "jail %d already exists",
1066 goto done_unlock_list;
1068 if (!prison_ischild(mypr, pr)) {
1069 mtx_unlock(&pr->pr_mtx);
1071 } else if (pr->pr_uref == 0) {
1072 if (!(flags & JAIL_DYING)) {
1073 mtx_unlock(&pr->pr_mtx);
1075 vfs_opterror(opts, "jail %d is dying",
1077 goto done_unlock_list;
1078 } else if ((flags & JAIL_ATTACH) ||
1079 (pr_flags & PR_PERSIST)) {
1081 * A dying jail might be resurrected
1082 * (via attach or persist), but first
1083 * it must determine if another jail
1084 * has claimed its name. Accomplish
1085 * this by implicitly re-setting the
1089 name = prison_name(mypr, pr);
1094 /* Update: jid must exist. */
1095 if (cuflags == JAIL_UPDATE) {
1097 vfs_opterror(opts, "jail %d not found", jid);
1098 goto done_unlock_list;
1103 * If the caller provided a name, look for a jail by that name.
1104 * This has different semantics for creates and updates keyed by jid
1105 * (where the name must not already exist in a different jail),
1106 * and updates keyed by the name itself (where the name must exist
1107 * because that is the jail being updated).
1111 namelc = strrchr(name, '.');
1116 * This is a hierarchical name. Split it into the
1117 * parent and child names, and make sure the parent
1118 * exists or matches an already found jail.
1121 if (strncmp(name, ppr->pr_name, namelc - name)
1122 || ppr->pr_name[namelc - name] != '\0') {
1123 mtx_unlock(&pr->pr_mtx);
1126 "cannot change jail's parent");
1127 goto done_unlock_list;
1131 ppr = prison_find_name(mypr, name);
1135 "jail \"%s\" not found", name);
1136 goto done_unlock_list;
1138 mtx_unlock(&ppr->pr_mtx);
1143 if (namelc[0] != '\0') {
1145 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1148 FOREACH_PRISON_CHILD(ppr, tpr) {
1149 if (tpr != pr && tpr->pr_ref > 0 &&
1150 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1152 cuflags != JAIL_CREATE) {
1153 mtx_lock(&tpr->pr_mtx);
1154 if (tpr->pr_ref > 0) {
1159 if (tpr->pr_uref > 0) {
1165 mtx_unlock(&tpr->pr_mtx);
1166 } else if (tpr->pr_uref > 0) {
1168 * Create, or update(jid):
1169 * name must not exist in an
1170 * active sibling jail.
1174 mtx_unlock(&pr->pr_mtx);
1176 "jail \"%s\" already exists",
1178 goto done_unlock_list;
1182 /* If no active jail is found, use a dying one. */
1183 if (deadpr != NULL && pr == NULL) {
1184 if (flags & JAIL_DYING) {
1185 mtx_lock(&deadpr->pr_mtx);
1186 if (deadpr->pr_ref == 0) {
1187 mtx_unlock(&deadpr->pr_mtx);
1191 } else if (cuflags == JAIL_UPDATE) {
1194 "jail \"%s\" is dying", name);
1195 goto done_unlock_list;
1198 /* Update: name must exist if no jid. */
1199 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1201 vfs_opterror(opts, "jail \"%s\" not found",
1203 goto done_unlock_list;
1207 /* Update: must provide a jid or name. */
1208 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1210 vfs_opterror(opts, "update specified no jail");
1211 goto done_unlock_list;
1214 /* If there's no prison to update, create a new one and link it in. */
1216 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1217 if (tpr->pr_childcount >= tpr->pr_childmax) {
1219 vfs_opterror(opts, "prison limit exceeded");
1220 goto done_unlock_list;
1223 mtx_lock(&ppr->pr_mtx);
1224 if (ppr->pr_ref == 0) {
1225 mtx_unlock(&ppr->pr_mtx);
1227 vfs_opterror(opts, "jail \"%s\" not found",
1228 prison_name(mypr, ppr));
1229 goto done_unlock_list;
1233 mtx_unlock(&ppr->pr_mtx);
1234 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1236 /* Find the next free jid. */
1239 if (jid == JAIL_MAX)
1241 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1242 if (tpr->pr_id < jid)
1244 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1245 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1248 if (jid == lastprid) {
1251 "no available jail IDs");
1253 prison_deref(ppr, PD_DEREF |
1254 PD_DEUREF | PD_LIST_XLOCKED);
1263 * The jail already has a jid (that did not yet exist),
1264 * so just find where to insert it.
1266 TAILQ_FOREACH(tpr, &allprison, pr_list)
1267 if (tpr->pr_id >= jid) {
1268 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1273 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1274 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1275 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1276 tpr->pr_childcount++;
1278 pr->pr_parent = ppr;
1281 /* Set some default values, and inherit some from the parent. */
1286 root = mypr->pr_root;
1289 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1290 pr->pr_flags |= PR_HOST;
1291 #if defined(INET) || defined(INET6)
1293 if (!(pr_flags & PR_VNET))
1297 if (!(ch_flags & PR_IP4_USER))
1299 PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1300 else if (!(pr_flags & PR_IP4_USER)) {
1301 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1302 if (ppr->pr_ip4 != NULL) {
1303 pr->pr_ip4s = ppr->pr_ip4s;
1304 pr->pr_ip4 = malloc(pr->pr_ip4s *
1305 sizeof(struct in_addr), M_PRISON,
1307 bcopy(ppr->pr_ip4, pr->pr_ip4,
1308 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1313 if (!(ch_flags & PR_IP6_USER))
1315 PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1316 else if (!(pr_flags & PR_IP6_USER)) {
1317 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1318 if (ppr->pr_ip6 != NULL) {
1319 pr->pr_ip6s = ppr->pr_ip6s;
1320 pr->pr_ip6 = malloc(pr->pr_ip6s *
1321 sizeof(struct in6_addr), M_PRISON,
1323 bcopy(ppr->pr_ip6, pr->pr_ip6,
1324 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1330 /* Source address selection is always on by default. */
1331 pr->pr_flags |= _PR_IP_SADDRSEL;
1333 pr->pr_securelevel = ppr->pr_securelevel;
1334 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1335 pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1336 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1338 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1339 if (osrelstr == NULL)
1340 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1342 strcpy(pr->pr_osrelease, osrelstr);
1344 LIST_INIT(&pr->pr_children);
1345 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1346 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1349 /* Allocate a new vnet if specified. */
1350 pr->pr_vnet = (pr_flags & PR_VNET)
1351 ? vnet_alloc() : ppr->pr_vnet;
1354 * Allocate a dedicated cpuset for each jail.
1355 * Unlike other initial settings, this may return an erorr.
1357 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1359 prison_deref(pr, PD_LIST_XLOCKED);
1363 mtx_lock(&pr->pr_mtx);
1365 * New prisons do not yet have a reference, because we do not
1366 * want others to see the incomplete prison once the
1367 * allprison_lock is downgraded.
1372 * Grab a reference for existing prisons, to ensure they
1373 * continue to exist for the duration of the call.
1376 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1377 if ((pr->pr_flags & PR_VNET) &&
1378 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1381 "vnet jails cannot have IP address restrictions");
1382 goto done_deref_locked;
1386 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1389 "ip4 cannot be changed after creation");
1390 goto done_deref_locked;
1394 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1397 "ip6 cannot be changed after creation");
1398 goto done_deref_locked;
1403 /* Do final error checking before setting anything. */
1405 if (slevel < ppr->pr_securelevel) {
1407 goto done_deref_locked;
1411 if (childmax >= ppr->pr_childmax) {
1413 goto done_deref_locked;
1417 if (enforce < ppr->pr_enforce_statfs) {
1419 goto done_deref_locked;
1424 * devfs_rsnum is a uint16_t
1426 if (rsnum < 0 || rsnum > 65535) {
1428 goto done_deref_locked;
1431 * Nested jails always inherit parent's devfs ruleset
1433 if (jailed(td->td_ucred)) {
1434 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1436 goto done_deref_locked;
1438 rsnum = ppr->pr_devfs_rsnum;
1443 if (ppr->pr_flags & PR_IP4) {
1445 * Make sure the new set of IP addresses is a
1446 * subset of the parent's list. Don't worry
1447 * about the parent being unlocked, as any
1448 * setting is done with allprison_lock held.
1450 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1451 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1453 if (ij == ppr->pr_ip4s) {
1455 goto done_deref_locked;
1458 for (ii = ij = 1; ii < ip4s; ii++) {
1459 if (ip4[ii].s_addr ==
1460 ppr->pr_ip4[0].s_addr)
1462 for (; ij < ppr->pr_ip4s; ij++)
1463 if (ip4[ii].s_addr ==
1464 ppr->pr_ip4[ij].s_addr)
1466 if (ij == ppr->pr_ip4s)
1469 if (ij == ppr->pr_ip4s) {
1471 goto done_deref_locked;
1476 * Check for conflicting IP addresses. We permit them
1477 * if there is no more than one IP on each jail. If
1478 * there is a duplicate on a jail with more than one
1479 * IP stop checking and return error.
1483 for (; tppr != &prison0; tppr = tppr->pr_parent)
1484 if (tppr->pr_flags & PR_VNET)
1487 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1490 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1492 tpr->pr_uref == 0) {
1496 if (!(tpr->pr_flags & PR_IP4_USER))
1499 if (tpr->pr_ip4 == NULL ||
1500 (ip4s == 1 && tpr->pr_ip4s == 1))
1502 for (ii = 0; ii < ip4s; ii++) {
1503 if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1506 "IPv4 addresses clash");
1507 goto done_deref_locked;
1515 if (ppr->pr_flags & PR_IP6) {
1517 * Make sure the new set of IP addresses is a
1518 * subset of the parent's list.
1520 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1521 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1524 if (ij == ppr->pr_ip6s) {
1526 goto done_deref_locked;
1529 for (ii = ij = 1; ii < ip6s; ii++) {
1530 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1533 for (; ij < ppr->pr_ip6s; ij++)
1534 if (IN6_ARE_ADDR_EQUAL(
1535 &ip6[ii], &ppr->pr_ip6[ij]))
1537 if (ij == ppr->pr_ip6s)
1540 if (ij == ppr->pr_ip6s) {
1542 goto done_deref_locked;
1546 /* Check for conflicting IP addresses. */
1549 for (; tppr != &prison0; tppr = tppr->pr_parent)
1550 if (tppr->pr_flags & PR_VNET)
1553 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1556 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1558 tpr->pr_uref == 0) {
1562 if (!(tpr->pr_flags & PR_IP6_USER))
1565 if (tpr->pr_ip6 == NULL ||
1566 (ip6s == 1 && tpr->pr_ip6s == 1))
1568 for (ii = 0; ii < ip6s; ii++) {
1569 if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1572 "IPv6 addresses clash");
1573 goto done_deref_locked;
1579 onamelen = namelen = 0;
1580 if (namelc != NULL) {
1581 /* Give a default name of the jid. Also allow the name to be
1582 * explicitly the jid - but not any other number, and only in
1583 * normal form (no leading zero/etc).
1585 if (namelc[0] == '\0')
1586 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1587 else if ((strtoul(namelc, &p, 10) != jid ||
1588 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1591 "name cannot be numeric (unless it is the jid)");
1592 goto done_deref_locked;
1595 * Make sure the name isn't too long for the prison or its
1598 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1599 onamelen = strlen(pr->pr_name + pnamelen);
1600 namelen = strlen(namelc);
1601 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1602 error = ENAMETOOLONG;
1603 goto done_deref_locked;
1605 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1606 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1607 sizeof(pr->pr_name)) {
1608 error = ENAMETOOLONG;
1609 goto done_deref_locked;
1613 if (pr_allow & ~ppr->pr_allow) {
1615 goto done_deref_locked;
1619 * Let modules check their parameters. This requires unlocking and
1620 * then re-locking the prison, but this is still a valid state as long
1621 * as allprison_lock remains xlocked.
1623 mtx_unlock(&pr->pr_mtx);
1624 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1626 prison_deref(pr, created
1628 : PD_DEREF | PD_LIST_XLOCKED);
1631 mtx_lock(&pr->pr_mtx);
1633 /* At this point, all valid parameters should have been noted. */
1634 TAILQ_FOREACH(opt, opts, link) {
1635 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1637 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1638 goto done_deref_locked;
1642 /* Set the parameters of the prison. */
1645 if (pr_flags & PR_IP4_USER) {
1646 pr->pr_flags |= PR_IP4;
1647 free(pr->pr_ip4, M_PRISON);
1651 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1653 if (tpr->pr_flags & PR_VNET) {
1658 if (prison_restrict_ip4(tpr, NULL)) {
1667 if (pr_flags & PR_IP6_USER) {
1668 pr->pr_flags |= PR_IP6;
1669 free(pr->pr_ip6, M_PRISON);
1673 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1675 if (tpr->pr_flags & PR_VNET) {
1680 if (prison_restrict_ip6(tpr, NULL)) {
1688 pr->pr_securelevel = slevel;
1689 /* Set all child jails to be at least this level. */
1690 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1691 if (tpr->pr_securelevel < slevel)
1692 tpr->pr_securelevel = slevel;
1695 pr->pr_childmax = childmax;
1696 /* Set all child jails to under this limit. */
1697 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1698 if (tpr->pr_childmax > childmax - level)
1699 tpr->pr_childmax = childmax > level
1700 ? childmax - level : 0;
1703 pr->pr_enforce_statfs = enforce;
1704 /* Pass this restriction on to the children. */
1705 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1706 if (tpr->pr_enforce_statfs < enforce)
1707 tpr->pr_enforce_statfs = enforce;
1710 pr->pr_devfs_rsnum = rsnum;
1711 /* Pass this restriction on to the children. */
1712 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1713 tpr->pr_devfs_rsnum = rsnum;
1715 if (namelc != NULL) {
1716 if (ppr == &prison0)
1717 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1719 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1720 ppr->pr_name, namelc);
1721 /* Change this component of child names. */
1722 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1723 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1724 strlen(tpr->pr_name + onamelen) + 1);
1725 bcopy(pr->pr_name, tpr->pr_name, namelen);
1729 /* Try to keep a real-rooted full pathname. */
1730 if (fullpath_disabled && path[0] == '/' &&
1731 strcmp(mypr->pr_path, "/"))
1732 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1733 mypr->pr_path, path);
1735 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1738 if (PR_HOST & ch_flags & ~pr_flags) {
1739 if (pr->pr_flags & PR_HOST) {
1741 * Copy the parent's host info. As with pr_ip4 above,
1742 * the lack of a lock on the parent is not a problem;
1743 * it is always set with allprison_lock at least
1744 * shared, and is held exclusively here.
1746 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1747 sizeof(pr->pr_hostname));
1748 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1749 sizeof(pr->pr_domainname));
1750 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1751 sizeof(pr->pr_hostuuid));
1752 pr->pr_hostid = pr->pr_parent->pr_hostid;
1754 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1755 /* Set this prison, and any descendants without PR_HOST. */
1757 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1759 strlcpy(pr->pr_domainname, domain,
1760 sizeof(pr->pr_domainname));
1762 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1764 pr->pr_hostid = hid;
1765 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1766 if (tpr->pr_flags & PR_HOST)
1770 strlcpy(tpr->pr_hostname,
1772 sizeof(tpr->pr_hostname));
1774 strlcpy(tpr->pr_domainname,
1776 sizeof(tpr->pr_domainname));
1778 strlcpy(tpr->pr_hostuuid,
1780 sizeof(tpr->pr_hostuuid));
1782 tpr->pr_hostid = hid;
1786 if ((tallow = ch_allow & ~pr_allow)) {
1787 /* Clear allow bits in all children. */
1788 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1789 tpr->pr_allow &= ~tallow;
1791 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1793 * Persistent prisons get an extra reference, and prisons losing their
1794 * persist flag lose that reference. Only do this for existing prisons
1795 * for now, so new ones will remain unseen until after the module
1796 * handlers have completed.
1798 born = pr->pr_uref == 0;
1799 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1800 if (pr_flags & PR_PERSIST) {
1808 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1809 mtx_unlock(&pr->pr_mtx);
1812 if (racct_enable && created)
1813 prison_racct_attach(pr);
1816 /* Locks may have prevented a complete restriction of child IP
1817 * addresses. If so, allocate some more memory and try again.
1822 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1823 mtx_lock(&pr->pr_mtx);
1825 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1827 if (tpr->pr_flags & PR_VNET) {
1832 if (prison_restrict_ip4(tpr, ip4)) {
1839 mtx_unlock(&pr->pr_mtx);
1845 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1846 mtx_lock(&pr->pr_mtx);
1848 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1850 if (tpr->pr_flags & PR_VNET) {
1855 if (prison_restrict_ip6(tpr, ip6)) {
1862 mtx_unlock(&pr->pr_mtx);
1866 /* Let the modules do their work. */
1867 sx_downgrade(&allprison_lock);
1869 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1871 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1872 prison_deref(pr, created
1874 : PD_DEREF | PD_LIST_SLOCKED);
1878 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1881 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1882 prison_deref(pr, created
1884 : PD_DEREF | PD_LIST_SLOCKED);
1888 /* Attach this process to the prison if requested. */
1889 if (flags & JAIL_ATTACH) {
1890 mtx_lock(&pr->pr_mtx);
1891 error = do_jail_attach(td, pr);
1893 vfs_opterror(opts, "attach failed");
1895 prison_deref(pr, PD_DEREF);
1901 if (racct_enable && !created) {
1902 if (!(flags & JAIL_ATTACH))
1903 sx_sunlock(&allprison_lock);
1904 prison_racct_modify(pr);
1905 if (!(flags & JAIL_ATTACH))
1906 sx_slock(&allprison_lock);
1910 td->td_retval[0] = pr->pr_id;
1913 * Now that it is all there, drop the temporary reference from existing
1914 * prisons. Or add a reference to newly created persistent prisons
1915 * (which was not done earlier so that the prison would not be publicly
1919 prison_deref(pr, (flags & JAIL_ATTACH)
1921 : PD_DEREF | PD_LIST_SLOCKED);
1923 if (pr_flags & PR_PERSIST) {
1924 mtx_lock(&pr->pr_mtx);
1927 mtx_unlock(&pr->pr_mtx);
1929 if (!(flags & JAIL_ATTACH))
1930 sx_sunlock(&allprison_lock);
1936 prison_deref(pr, created
1937 ? PD_LOCKED | PD_LIST_XLOCKED
1938 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1941 sx_xunlock(&allprison_lock);
1947 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1948 &errmsg_len) == 0 && errmsg_len > 0) {
1949 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1950 if (optuio->uio_segflg == UIO_SYSSPACE)
1952 optuio->uio_iov[errmsg_pos].iov_base,
1956 optuio->uio_iov[errmsg_pos].iov_base,
1962 free(ip4, M_PRISON);
1965 free(ip6, M_PRISON);
1968 free(g_path, M_TEMP);
1975 * struct jail_get_args {
1976 * struct iovec *iovp;
1977 * unsigned int iovcnt;
1982 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1987 /* Check that we have an even number of iovecs. */
1988 if (uap->iovcnt & 1)
1991 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1994 error = kern_jail_get(td, auio, uap->flags);
1996 error = copyout(auio->uio_iov, uap->iovp,
1997 uap->iovcnt * sizeof (struct iovec));
2003 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2005 struct prison *pr, *mypr;
2007 struct vfsoptlist *opts;
2008 char *errmsg, *name;
2009 int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
2011 if (flags & ~JAIL_GET_MASK)
2014 /* Get the parameter list. */
2015 error = vfs_buildopts(optuio, &opts);
2018 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2019 mypr = td->td_ucred->cr_prison;
2022 * Find the prison specified by one of: lastjid, jid, name.
2024 sx_slock(&allprison_lock);
2025 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2027 TAILQ_FOREACH(pr, &allprison, pr_list) {
2028 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2029 mtx_lock(&pr->pr_mtx);
2030 if (pr->pr_ref > 0 &&
2031 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
2033 mtx_unlock(&pr->pr_mtx);
2039 vfs_opterror(opts, "no jail after %d", jid);
2040 goto done_unlock_list;
2041 } else if (error != ENOENT)
2042 goto done_unlock_list;
2044 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2047 pr = prison_find_child(mypr, jid);
2049 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2050 mtx_unlock(&pr->pr_mtx);
2052 vfs_opterror(opts, "jail %d is dying",
2054 goto done_unlock_list;
2059 vfs_opterror(opts, "jail %d not found", jid);
2060 goto done_unlock_list;
2062 } else if (error != ENOENT)
2063 goto done_unlock_list;
2065 error = vfs_getopt(opts, "name", (void **)&name, &len);
2067 if (len == 0 || name[len - 1] != '\0') {
2069 goto done_unlock_list;
2071 pr = prison_find_name(mypr, name);
2073 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2074 mtx_unlock(&pr->pr_mtx);
2076 vfs_opterror(opts, "jail \"%s\" is dying",
2078 goto done_unlock_list;
2083 vfs_opterror(opts, "jail \"%s\" not found", name);
2084 goto done_unlock_list;
2085 } else if (error != ENOENT)
2086 goto done_unlock_list;
2088 vfs_opterror(opts, "no jail specified");
2090 goto done_unlock_list;
2093 /* Get the parameters of the prison. */
2096 td->td_retval[0] = pr->pr_id;
2097 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2098 if (error != 0 && error != ENOENT)
2100 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2101 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2102 if (error != 0 && error != ENOENT)
2104 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2105 if (error != 0 && error != ENOENT)
2107 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2108 sizeof(pr->pr_cpuset->cs_id));
2109 if (error != 0 && error != ENOENT)
2111 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2112 if (error != 0 && error != ENOENT)
2115 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2116 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2117 if (error != 0 && error != ENOENT)
2121 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2122 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2123 if (error != 0 && error != ENOENT)
2126 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2127 sizeof(pr->pr_securelevel));
2128 if (error != 0 && error != ENOENT)
2130 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2131 sizeof(pr->pr_childcount));
2132 if (error != 0 && error != ENOENT)
2134 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2135 sizeof(pr->pr_childmax));
2136 if (error != 0 && error != ENOENT)
2138 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2139 if (error != 0 && error != ENOENT)
2141 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2142 if (error != 0 && error != ENOENT)
2144 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2145 if (error != 0 && error != ENOENT)
2147 #ifdef COMPAT_FREEBSD32
2148 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2149 uint32_t hid32 = pr->pr_hostid;
2151 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2154 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2155 sizeof(pr->pr_hostid));
2156 if (error != 0 && error != ENOENT)
2158 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2159 sizeof(pr->pr_enforce_statfs));
2160 if (error != 0 && error != ENOENT)
2162 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2163 sizeof(pr->pr_devfs_rsnum));
2164 if (error != 0 && error != ENOENT)
2166 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2168 if (pr_flag_names[fi] == NULL)
2170 i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2171 error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2172 if (error != 0 && error != ENOENT)
2175 error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2176 if (error != 0 && error != ENOENT)
2179 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2182 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2183 i = pr_flag_jailsys[fi].disable &&
2184 (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2185 : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2188 vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2189 if (error != 0 && error != ENOENT)
2192 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2194 if (pr_allow_names[fi] == NULL)
2196 i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2197 error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2198 if (error != 0 && error != ENOENT)
2201 error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2202 if (error != 0 && error != ENOENT)
2205 i = (pr->pr_uref == 0);
2206 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2207 if (error != 0 && error != ENOENT)
2210 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2211 if (error != 0 && error != ENOENT)
2213 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2214 sizeof(pr->pr_osreldate));
2215 if (error != 0 && error != ENOENT)
2217 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2218 if (error != 0 && error != ENOENT)
2221 /* Get the module parameters. */
2222 mtx_unlock(&pr->pr_mtx);
2224 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2227 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2229 /* By now, all parameters should have been noted. */
2230 TAILQ_FOREACH(opt, opts, link) {
2231 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2233 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2238 /* Write the fetched parameters back to userspace. */
2240 TAILQ_FOREACH(opt, opts, link) {
2241 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2242 pos = 2 * opt->pos + 1;
2243 optuio->uio_iov[pos].iov_len = opt->len;
2244 if (opt->value != NULL) {
2245 if (optuio->uio_segflg == UIO_SYSSPACE) {
2247 optuio->uio_iov[pos].iov_base,
2250 error = copyout(opt->value,
2251 optuio->uio_iov[pos].iov_base,
2262 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2266 sx_sunlock(&allprison_lock);
2268 if (error && errmsg_pos >= 0) {
2269 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2270 errmsg_pos = 2 * errmsg_pos + 1;
2271 if (errmsg_len > 0) {
2272 if (optuio->uio_segflg == UIO_SYSSPACE)
2274 optuio->uio_iov[errmsg_pos].iov_base,
2278 optuio->uio_iov[errmsg_pos].iov_base,
2288 * struct jail_remove_args {
2293 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2295 struct prison *pr, *cpr, *lpr, *tpr;
2298 error = priv_check(td, PRIV_JAIL_REMOVE);
2302 sx_xlock(&allprison_lock);
2303 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2305 sx_xunlock(&allprison_lock);
2309 /* Remove all descendants of this prison, then remove this prison. */
2311 if (!LIST_EMPTY(&pr->pr_children)) {
2312 mtx_unlock(&pr->pr_mtx);
2314 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2315 mtx_lock(&cpr->pr_mtx);
2316 if (cpr->pr_ref > 0) {
2320 /* Already removed - do not do it again. */
2323 mtx_unlock(&cpr->pr_mtx);
2325 mtx_lock(&lpr->pr_mtx);
2326 prison_remove_one(lpr);
2327 sx_xlock(&allprison_lock);
2332 mtx_lock(&lpr->pr_mtx);
2333 prison_remove_one(lpr);
2334 sx_xlock(&allprison_lock);
2336 mtx_lock(&pr->pr_mtx);
2338 prison_remove_one(pr);
2343 prison_remove_one(struct prison *pr)
2348 /* If the prison was persistent, it is not anymore. */
2350 if (pr->pr_flags & PR_PERSIST) {
2353 pr->pr_flags &= ~PR_PERSIST;
2357 * jail_remove added a reference. If that's the only one, remove
2360 KASSERT(pr->pr_ref > 0,
2361 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2362 if (pr->pr_ref == 1) {
2364 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2368 mtx_unlock(&pr->pr_mtx);
2369 sx_xunlock(&allprison_lock);
2371 * Kill all processes unfortunate enough to be attached to this prison.
2373 sx_slock(&allproc_lock);
2374 LIST_FOREACH(p, &allproc, p_list) {
2376 if (p->p_state != PRS_NEW && p->p_ucred &&
2377 p->p_ucred->cr_prison == pr)
2378 kern_psignal(p, SIGKILL);
2381 sx_sunlock(&allproc_lock);
2382 /* Remove the temporary reference added by jail_remove. */
2383 prison_deref(pr, deuref | PD_DEREF);
2388 * struct jail_attach_args {
2393 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2398 error = priv_check(td, PRIV_JAIL_ATTACH);
2403 * Start with exclusive hold on allprison_lock to ensure that a possible
2404 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2405 * But then immediately downgrade it since we don't need to stop
2408 sx_xlock(&allprison_lock);
2409 sx_downgrade(&allprison_lock);
2410 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2412 sx_sunlock(&allprison_lock);
2417 * Do not allow a process to attach to a prison that is not
2418 * considered to be "alive".
2420 if (pr->pr_uref == 0) {
2421 mtx_unlock(&pr->pr_mtx);
2422 sx_sunlock(&allprison_lock);
2426 return (do_jail_attach(td, pr));
2430 do_jail_attach(struct thread *td, struct prison *pr)
2433 struct ucred *newcred, *oldcred;
2437 * XXX: Note that there is a slight race here if two threads
2438 * in the same privileged process attempt to attach to two
2439 * different jails at the same time. It is important for
2440 * user processes not to do this, or they might end up with
2441 * a process root from one prison, but attached to the jail
2446 mtx_unlock(&pr->pr_mtx);
2448 /* Let modules do whatever they need to prepare for attaching. */
2449 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2451 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2454 sx_sunlock(&allprison_lock);
2457 * Reparent the newly attached process to this jail.
2460 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2464 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2465 if ((error = change_dir(pr->pr_root, td)) != 0)
2468 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2471 VOP_UNLOCK(pr->pr_root, 0);
2472 if ((error = change_root(pr->pr_root, td)))
2477 oldcred = crcopysafe(p, newcred);
2478 newcred->cr_prison = pr;
2479 proc_set_cred(p, newcred);
2483 racct_proc_ucred_changed(p, oldcred, newcred);
2485 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2490 VOP_UNLOCK(pr->pr_root, 0);
2492 /* Tell modules this thread is still in its old jail after all. */
2493 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2494 prison_deref(pr, PD_DEREF | PD_DEUREF);
2500 * Returns a locked prison instance, or NULL on failure.
2503 prison_find(int prid)
2507 sx_assert(&allprison_lock, SX_LOCKED);
2508 TAILQ_FOREACH(pr, &allprison, pr_list) {
2509 if (pr->pr_id == prid) {
2510 mtx_lock(&pr->pr_mtx);
2513 mtx_unlock(&pr->pr_mtx);
2520 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2523 prison_find_child(struct prison *mypr, int prid)
2528 sx_assert(&allprison_lock, SX_LOCKED);
2529 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2530 if (pr->pr_id == prid) {
2531 mtx_lock(&pr->pr_mtx);
2534 mtx_unlock(&pr->pr_mtx);
2541 * Look for the name relative to mypr. Returns a locked prison or NULL.
2544 prison_find_name(struct prison *mypr, const char *name)
2546 struct prison *pr, *deadpr;
2550 sx_assert(&allprison_lock, SX_LOCKED);
2551 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2554 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2555 if (!strcmp(pr->pr_name + mylen, name)) {
2556 mtx_lock(&pr->pr_mtx);
2557 if (pr->pr_ref > 0) {
2558 if (pr->pr_uref > 0)
2562 mtx_unlock(&pr->pr_mtx);
2565 /* There was no valid prison - perhaps there was a dying one. */
2566 if (deadpr != NULL) {
2567 mtx_lock(&deadpr->pr_mtx);
2568 if (deadpr->pr_ref == 0) {
2569 mtx_unlock(&deadpr->pr_mtx);
2577 * See if a prison has the specific flag set.
2580 prison_flag(struct ucred *cred, unsigned flag)
2583 /* This is an atomic read, so no locking is necessary. */
2584 return (cred->cr_prison->pr_flags & flag);
2588 prison_allow(struct ucred *cred, unsigned flag)
2591 /* This is an atomic read, so no locking is necessary. */
2592 return (cred->cr_prison->pr_allow & flag);
2596 * Remove a prison reference. If that was the last reference, remove the
2597 * prison itself - but not in this context in case there are locks held.
2600 prison_free_locked(struct prison *pr)
2604 mtx_assert(&pr->pr_mtx, MA_OWNED);
2606 mtx_unlock(&pr->pr_mtx);
2608 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2612 prison_free(struct prison *pr)
2615 mtx_lock(&pr->pr_mtx);
2616 prison_free_locked(pr);
2620 * Complete a call to either prison_free or prison_proc_free.
2623 prison_complete(void *context, int pending)
2625 struct prison *pr = context;
2627 sx_xlock(&allprison_lock);
2628 mtx_lock(&pr->pr_mtx);
2629 prison_deref(pr, pr->pr_uref
2630 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2631 : PD_LOCKED | PD_LIST_XLOCKED);
2635 * Remove a prison reference (usually). This internal version assumes no
2636 * mutexes are held, except perhaps the prison itself. If there are no more
2637 * references, release and delist the prison. On completion, the prison lock
2638 * and the allprison lock are both unlocked.
2641 prison_deref(struct prison *pr, int flags)
2643 struct prison *ppr, *tpr;
2646 if (!(flags & PD_LOCKED))
2647 mtx_lock(&pr->pr_mtx);
2649 if (flags & PD_DEUREF) {
2650 KASSERT(pr->pr_uref > 0,
2651 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2654 lasturef = pr->pr_uref == 0;
2657 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2660 if (flags & PD_DEREF) {
2661 KASSERT(pr->pr_ref > 0,
2662 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2667 mtx_unlock(&pr->pr_mtx);
2670 * Tell the modules if the last user reference was removed
2671 * (even it sticks around in dying state).
2674 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2675 sx_xlock(&allprison_lock);
2676 flags |= PD_LIST_XLOCKED;
2678 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2679 mtx_lock(&pr->pr_mtx);
2681 mtx_unlock(&pr->pr_mtx);
2684 /* If the prison still has references, nothing else to do. */
2686 if (flags & PD_LIST_SLOCKED)
2687 sx_sunlock(&allprison_lock);
2688 else if (flags & PD_LIST_XLOCKED)
2689 sx_xunlock(&allprison_lock);
2693 if (flags & PD_LIST_SLOCKED) {
2694 if (!sx_try_upgrade(&allprison_lock)) {
2695 sx_sunlock(&allprison_lock);
2696 sx_xlock(&allprison_lock);
2698 } else if (!(flags & PD_LIST_XLOCKED))
2699 sx_xlock(&allprison_lock);
2701 TAILQ_REMOVE(&allprison, pr, pr_list);
2702 LIST_REMOVE(pr, pr_sibling);
2703 ppr = pr->pr_parent;
2704 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2705 tpr->pr_childcount--;
2706 sx_xunlock(&allprison_lock);
2709 if (pr->pr_vnet != ppr->pr_vnet)
2710 vnet_destroy(pr->pr_vnet);
2712 if (pr->pr_root != NULL)
2714 mtx_destroy(&pr->pr_mtx);
2716 free(pr->pr_ip4, M_PRISON);
2719 free(pr->pr_ip6, M_PRISON);
2721 if (pr->pr_cpuset != NULL)
2722 cpuset_rel(pr->pr_cpuset);
2726 prison_racct_detach(pr);
2730 /* Removing a prison frees a reference on its parent. */
2732 mtx_lock(&pr->pr_mtx);
2733 flags = PD_DEREF | PD_DEUREF;
2738 prison_hold_locked(struct prison *pr)
2741 mtx_assert(&pr->pr_mtx, MA_OWNED);
2742 KASSERT(pr->pr_ref > 0,
2743 ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2748 prison_hold(struct prison *pr)
2751 mtx_lock(&pr->pr_mtx);
2752 prison_hold_locked(pr);
2753 mtx_unlock(&pr->pr_mtx);
2757 prison_proc_hold(struct prison *pr)
2760 mtx_lock(&pr->pr_mtx);
2761 KASSERT(pr->pr_uref > 0,
2762 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2764 mtx_unlock(&pr->pr_mtx);
2768 prison_proc_free(struct prison *pr)
2771 mtx_lock(&pr->pr_mtx);
2772 KASSERT(pr->pr_uref > 0,
2773 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2774 if (pr->pr_uref > 1)
2778 * Don't remove the last user reference in this context, which
2779 * is expected to be a process that is not only locked, but
2783 mtx_unlock(&pr->pr_mtx);
2784 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2787 mtx_unlock(&pr->pr_mtx);
2793 * Restrict a prison's IP address list with its parent's, possibly replacing
2794 * it. Return true if the replacement buffer was used (or would have been).
2797 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2802 ppr = pr->pr_parent;
2803 if (!(pr->pr_flags & PR_IP4_USER)) {
2804 /* This has no user settings, so just copy the parent's list. */
2805 if (pr->pr_ip4s < ppr->pr_ip4s) {
2807 * There's no room for the parent's list. Use the
2808 * new list buffer, which is assumed to be big enough
2809 * (if it was passed). If there's no buffer, try to
2813 if (newip4 == NULL) {
2814 newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2815 M_PRISON, M_NOWAIT);
2819 if (newip4 != NULL) {
2820 bcopy(ppr->pr_ip4, newip4,
2821 ppr->pr_ip4s * sizeof(*newip4));
2822 free(pr->pr_ip4, M_PRISON);
2823 pr->pr_ip4 = newip4;
2824 pr->pr_ip4s = ppr->pr_ip4s;
2828 pr->pr_ip4s = ppr->pr_ip4s;
2829 if (pr->pr_ip4s > 0)
2830 bcopy(ppr->pr_ip4, pr->pr_ip4,
2831 pr->pr_ip4s * sizeof(*newip4));
2832 else if (pr->pr_ip4 != NULL) {
2833 free(pr->pr_ip4, M_PRISON);
2836 } else if (pr->pr_ip4s > 0) {
2837 /* Remove addresses that aren't in the parent. */
2838 for (ij = 0; ij < ppr->pr_ip4s; ij++)
2839 if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2841 if (ij < ppr->pr_ip4s)
2844 bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2845 --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2848 for (ij = 1; ii < pr->pr_ip4s; ) {
2849 if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2853 switch (ij >= ppr->pr_ip4s ? -1 :
2854 qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2856 bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2857 (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2868 if (pr->pr_ip4s == 0) {
2869 pr->pr_flags |= PR_IP4_DISABLE;
2870 free(pr->pr_ip4, M_PRISON);
2878 * Pass back primary IPv4 address of this jail.
2880 * If not restricted return success but do not alter the address. Caller has
2881 * to make sure to initialize it correctly (e.g. INADDR_ANY).
2883 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2884 * Address returned in NBO.
2887 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2891 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2892 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2894 pr = cred->cr_prison;
2895 if (!(pr->pr_flags & PR_IP4))
2897 mtx_lock(&pr->pr_mtx);
2898 if (!(pr->pr_flags & PR_IP4)) {
2899 mtx_unlock(&pr->pr_mtx);
2902 if (pr->pr_ip4 == NULL) {
2903 mtx_unlock(&pr->pr_mtx);
2904 return (EAFNOSUPPORT);
2907 ia->s_addr = pr->pr_ip4[0].s_addr;
2908 mtx_unlock(&pr->pr_mtx);
2913 * Return 1 if we should do proper source address selection or are not jailed.
2914 * We will return 0 if we should bypass source address selection in favour
2915 * of the primary jail IPv4 address. Only in this case *ia will be updated and
2917 * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2920 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2926 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2927 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2932 pr = cred->cr_prison;
2933 if (pr->pr_flags & PR_IP4_SADDRSEL)
2936 lia.s_addr = INADDR_ANY;
2937 error = prison_get_ip4(cred, &lia);
2940 if (lia.s_addr == INADDR_ANY)
2943 ia->s_addr = lia.s_addr;
2948 * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2951 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2958 * No need to lock since the PR_IP4_USER flag can't be altered for
2961 while (pr1 != &prison0 &&
2963 !(pr1->pr_flags & PR_VNET) &&
2965 !(pr1->pr_flags & PR_IP4_USER))
2966 pr1 = pr1->pr_parent;
2967 while (pr2 != &prison0 &&
2969 !(pr2->pr_flags & PR_VNET) &&
2971 !(pr2->pr_flags & PR_IP4_USER))
2972 pr2 = pr2->pr_parent;
2973 return (pr1 == pr2);
2977 * Make sure our (source) address is set to something meaningful to this
2980 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2981 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2982 * doesn't allow IPv4. Address passed in in NBO and returned in NBO.
2985 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2991 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2992 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2994 pr = cred->cr_prison;
2995 if (!(pr->pr_flags & PR_IP4))
2997 mtx_lock(&pr->pr_mtx);
2998 if (!(pr->pr_flags & PR_IP4)) {
2999 mtx_unlock(&pr->pr_mtx);
3002 if (pr->pr_ip4 == NULL) {
3003 mtx_unlock(&pr->pr_mtx);
3004 return (EAFNOSUPPORT);
3007 ia0.s_addr = ntohl(ia->s_addr);
3008 if (ia0.s_addr == INADDR_LOOPBACK) {
3009 ia->s_addr = pr->pr_ip4[0].s_addr;
3010 mtx_unlock(&pr->pr_mtx);
3014 if (ia0.s_addr == INADDR_ANY) {
3016 * In case there is only 1 IPv4 address, bind directly.
3018 if (pr->pr_ip4s == 1)
3019 ia->s_addr = pr->pr_ip4[0].s_addr;
3020 mtx_unlock(&pr->pr_mtx);
3024 error = _prison_check_ip4(pr, ia);
3025 mtx_unlock(&pr->pr_mtx);
3030 * Rewrite destination address in case we will connect to loopback address.
3032 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
3033 * Address passed in in NBO and returned in NBO.
3036 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
3040 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3041 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3043 pr = cred->cr_prison;
3044 if (!(pr->pr_flags & PR_IP4))
3046 mtx_lock(&pr->pr_mtx);
3047 if (!(pr->pr_flags & PR_IP4)) {
3048 mtx_unlock(&pr->pr_mtx);
3051 if (pr->pr_ip4 == NULL) {
3052 mtx_unlock(&pr->pr_mtx);
3053 return (EAFNOSUPPORT);
3056 if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
3057 ia->s_addr = pr->pr_ip4[0].s_addr;
3058 mtx_unlock(&pr->pr_mtx);
3063 * Return success because nothing had to be changed.
3065 mtx_unlock(&pr->pr_mtx);
3070 * Check if given address belongs to the jail referenced by cred/prison.
3072 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
3073 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3074 * doesn't allow IPv4. Address passed in in NBO.
3077 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
3082 * Check the primary IP.
3084 if (pr->pr_ip4[0].s_addr == ia->s_addr)
3088 * All the other IPs are sorted so we can do a binary search.
3091 z = pr->pr_ip4s - 2;
3094 d = qcmp_v4(&pr->pr_ip4[i+1], ia);
3103 return (EADDRNOTAVAIL);
3107 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
3112 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3113 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3115 pr = cred->cr_prison;
3116 if (!(pr->pr_flags & PR_IP4))
3118 mtx_lock(&pr->pr_mtx);
3119 if (!(pr->pr_flags & PR_IP4)) {
3120 mtx_unlock(&pr->pr_mtx);
3123 if (pr->pr_ip4 == NULL) {
3124 mtx_unlock(&pr->pr_mtx);
3125 return (EAFNOSUPPORT);
3128 error = _prison_check_ip4(pr, ia);
3129 mtx_unlock(&pr->pr_mtx);
3136 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
3141 ppr = pr->pr_parent;
3142 if (!(pr->pr_flags & PR_IP6_USER)) {
3143 /* This has no user settings, so just copy the parent's list. */
3144 if (pr->pr_ip6s < ppr->pr_ip6s) {
3146 * There's no room for the parent's list. Use the
3147 * new list buffer, which is assumed to be big enough
3148 * (if it was passed). If there's no buffer, try to
3152 if (newip6 == NULL) {
3153 newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3154 M_PRISON, M_NOWAIT);
3158 if (newip6 != NULL) {
3159 bcopy(ppr->pr_ip6, newip6,
3160 ppr->pr_ip6s * sizeof(*newip6));
3161 free(pr->pr_ip6, M_PRISON);
3162 pr->pr_ip6 = newip6;
3163 pr->pr_ip6s = ppr->pr_ip6s;
3167 pr->pr_ip6s = ppr->pr_ip6s;
3168 if (pr->pr_ip6s > 0)
3169 bcopy(ppr->pr_ip6, pr->pr_ip6,
3170 pr->pr_ip6s * sizeof(*newip6));
3171 else if (pr->pr_ip6 != NULL) {
3172 free(pr->pr_ip6, M_PRISON);
3175 } else if (pr->pr_ip6s > 0) {
3176 /* Remove addresses that aren't in the parent. */
3177 for (ij = 0; ij < ppr->pr_ip6s; ij++)
3178 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3181 if (ij < ppr->pr_ip6s)
3184 bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3185 --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3188 for (ij = 1; ii < pr->pr_ip6s; ) {
3189 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3194 switch (ij >= ppr->pr_ip6s ? -1 :
3195 qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3197 bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3198 (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3209 if (pr->pr_ip6s == 0) {
3210 pr->pr_flags |= PR_IP6_DISABLE;
3211 free(pr->pr_ip6, M_PRISON);
3219 * Pass back primary IPv6 address for this jail.
3221 * If not restricted return success but do not alter the address. Caller has
3222 * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3224 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3227 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3231 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3232 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3234 pr = cred->cr_prison;
3235 if (!(pr->pr_flags & PR_IP6))
3237 mtx_lock(&pr->pr_mtx);
3238 if (!(pr->pr_flags & PR_IP6)) {
3239 mtx_unlock(&pr->pr_mtx);
3242 if (pr->pr_ip6 == NULL) {
3243 mtx_unlock(&pr->pr_mtx);
3244 return (EAFNOSUPPORT);
3247 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3248 mtx_unlock(&pr->pr_mtx);
3253 * Return 1 if we should do proper source address selection or are not jailed.
3254 * We will return 0 if we should bypass source address selection in favour
3255 * of the primary jail IPv6 address. Only in this case *ia will be updated and
3257 * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3260 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3263 struct in6_addr lia6;
3266 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3267 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3272 pr = cred->cr_prison;
3273 if (pr->pr_flags & PR_IP6_SADDRSEL)
3277 error = prison_get_ip6(cred, &lia6);
3280 if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3283 bcopy(&lia6, ia6, sizeof(struct in6_addr));
3288 * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3291 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3297 while (pr1 != &prison0 &&
3299 !(pr1->pr_flags & PR_VNET) &&
3301 !(pr1->pr_flags & PR_IP6_USER))
3302 pr1 = pr1->pr_parent;
3303 while (pr2 != &prison0 &&
3305 !(pr2->pr_flags & PR_VNET) &&
3307 !(pr2->pr_flags & PR_IP6_USER))
3308 pr2 = pr2->pr_parent;
3309 return (pr1 == pr2);
3313 * Make sure our (source) address is set to something meaningful to this jail.
3315 * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3316 * when needed while binding.
3318 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3319 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3320 * doesn't allow IPv6.
3323 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3328 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3329 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3331 pr = cred->cr_prison;
3332 if (!(pr->pr_flags & PR_IP6))
3334 mtx_lock(&pr->pr_mtx);
3335 if (!(pr->pr_flags & PR_IP6)) {
3336 mtx_unlock(&pr->pr_mtx);
3339 if (pr->pr_ip6 == NULL) {
3340 mtx_unlock(&pr->pr_mtx);
3341 return (EAFNOSUPPORT);
3344 if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3345 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3346 mtx_unlock(&pr->pr_mtx);
3350 if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3352 * In case there is only 1 IPv6 address, and v6only is true,
3353 * then bind directly.
3355 if (v6only != 0 && pr->pr_ip6s == 1)
3356 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3357 mtx_unlock(&pr->pr_mtx);
3361 error = _prison_check_ip6(pr, ia6);
3362 mtx_unlock(&pr->pr_mtx);
3367 * Rewrite destination address in case we will connect to loopback address.
3369 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3372 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3376 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3377 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3379 pr = cred->cr_prison;
3380 if (!(pr->pr_flags & PR_IP6))
3382 mtx_lock(&pr->pr_mtx);
3383 if (!(pr->pr_flags & PR_IP6)) {
3384 mtx_unlock(&pr->pr_mtx);
3387 if (pr->pr_ip6 == NULL) {
3388 mtx_unlock(&pr->pr_mtx);
3389 return (EAFNOSUPPORT);
3392 if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3393 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3394 mtx_unlock(&pr->pr_mtx);
3399 * Return success because nothing had to be changed.
3401 mtx_unlock(&pr->pr_mtx);
3406 * Check if given address belongs to the jail referenced by cred/prison.
3408 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3409 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3410 * doesn't allow IPv6.
3413 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3418 * Check the primary IP.
3420 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3424 * All the other IPs are sorted so we can do a binary search.
3427 z = pr->pr_ip6s - 2;
3430 d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3439 return (EADDRNOTAVAIL);
3443 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3448 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3449 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3451 pr = cred->cr_prison;
3452 if (!(pr->pr_flags & PR_IP6))
3454 mtx_lock(&pr->pr_mtx);
3455 if (!(pr->pr_flags & PR_IP6)) {
3456 mtx_unlock(&pr->pr_mtx);
3459 if (pr->pr_ip6 == NULL) {
3460 mtx_unlock(&pr->pr_mtx);
3461 return (EAFNOSUPPORT);
3464 error = _prison_check_ip6(pr, ia6);
3465 mtx_unlock(&pr->pr_mtx);
3471 * Check if a jail supports the given address family.
3473 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3477 prison_check_af(struct ucred *cred, int af)
3482 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3484 pr = cred->cr_prison;
3486 /* Prisons with their own network stack are not limited. */
3487 if (prison_owns_vnet(cred))
3496 if (pr->pr_flags & PR_IP4)
3498 mtx_lock(&pr->pr_mtx);
3499 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3500 error = EAFNOSUPPORT;
3501 mtx_unlock(&pr->pr_mtx);
3507 if (pr->pr_flags & PR_IP6)
3509 mtx_lock(&pr->pr_mtx);
3510 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3511 error = EAFNOSUPPORT;
3512 mtx_unlock(&pr->pr_mtx);
3520 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3521 error = EAFNOSUPPORT;
3527 * Check if given address belongs to the jail referenced by cred (wrapper to
3528 * prison_check_ip[46]).
3530 * Returns 0 if jail doesn't restrict the address family or if address belongs
3531 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3532 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3535 prison_if(struct ucred *cred, struct sockaddr *sa)
3538 struct sockaddr_in *sai;
3541 struct sockaddr_in6 *sai6;
3545 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3546 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3549 if (prison_owns_vnet(cred))
3554 switch (sa->sa_family)
3558 sai = (struct sockaddr_in *)sa;
3559 error = prison_check_ip4(cred, &sai->sin_addr);
3564 sai6 = (struct sockaddr_in6 *)sa;
3565 error = prison_check_ip6(cred, &sai6->sin6_addr);
3569 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3570 error = EAFNOSUPPORT;
3576 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3579 prison_check(struct ucred *cred1, struct ucred *cred2)
3582 return ((cred1->cr_prison == cred2->cr_prison ||
3583 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3587 * Return 1 if p2 is a child of p1, otherwise 0.
3590 prison_ischild(struct prison *pr1, struct prison *pr2)
3593 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3600 * Return 1 if the passed credential is in a jail, otherwise 0.
3603 jailed(struct ucred *cred)
3606 return (cred->cr_prison != &prison0);
3610 * Return 1 if the passed credential is in a jail and that jail does not
3611 * have its own virtual network stack, otherwise 0.
3614 jailed_without_vnet(struct ucred *cred)
3620 if (prison_owns_vnet(cred))
3628 * Return the correct hostname (domainname, et al) for the passed credential.
3631 getcredhostname(struct ucred *cred, char *buf, size_t size)
3636 * A NULL credential can be used to shortcut to the physical
3637 * system's hostname.
3639 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3640 mtx_lock(&pr->pr_mtx);
3641 strlcpy(buf, pr->pr_hostname, size);
3642 mtx_unlock(&pr->pr_mtx);
3646 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3649 mtx_lock(&cred->cr_prison->pr_mtx);
3650 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3651 mtx_unlock(&cred->cr_prison->pr_mtx);
3655 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3658 mtx_lock(&cred->cr_prison->pr_mtx);
3659 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3660 mtx_unlock(&cred->cr_prison->pr_mtx);
3664 getcredhostid(struct ucred *cred, unsigned long *hostid)
3667 mtx_lock(&cred->cr_prison->pr_mtx);
3668 *hostid = cred->cr_prison->pr_hostid;
3669 mtx_unlock(&cred->cr_prison->pr_mtx);
3674 * Determine whether the prison represented by cred owns
3675 * its vnet rather than having it inherited.
3677 * Returns 1 in case the prison owns the vnet, 0 otherwise.
3680 prison_owns_vnet(struct ucred *cred)
3684 * vnets cannot be added/removed after jail creation,
3685 * so no need to lock here.
3687 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3692 * Determine whether the subject represented by cred can "see"
3693 * status of a mount point.
3694 * Returns: 0 for permitted, ENOENT otherwise.
3695 * XXX: This function should be called cr_canseemount() and should be
3696 * placed in kern_prot.c.
3699 prison_canseemount(struct ucred *cred, struct mount *mp)
3705 pr = cred->cr_prison;
3706 if (pr->pr_enforce_statfs == 0)
3708 if (pr->pr_root->v_mount == mp)
3710 if (pr->pr_enforce_statfs == 2)
3713 * If jail's chroot directory is set to "/" we should be able to see
3714 * all mount-points from inside a jail.
3715 * This is ugly check, but this is the only situation when jail's
3716 * directory ends with '/'.
3718 if (strcmp(pr->pr_path, "/") == 0)
3720 len = strlen(pr->pr_path);
3722 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3725 * Be sure that we don't have situation where jail's root directory
3726 * is "/some/path" and mount point is "/some/pathpath".
3728 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3734 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3736 char jpath[MAXPATHLEN];
3740 pr = cred->cr_prison;
3741 if (pr->pr_enforce_statfs == 0)
3743 if (prison_canseemount(cred, mp) != 0) {
3744 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3745 strlcpy(sp->f_mntonname, "[restricted]",
3746 sizeof(sp->f_mntonname));
3749 if (pr->pr_root->v_mount == mp) {
3751 * Clear current buffer data, so we are sure nothing from
3752 * the valid path left there.
3754 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3755 *sp->f_mntonname = '/';
3759 * If jail's chroot directory is set to "/" we should be able to see
3760 * all mount-points from inside a jail.
3762 if (strcmp(pr->pr_path, "/") == 0)
3764 len = strlen(pr->pr_path);
3765 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3767 * Clear current buffer data, so we are sure nothing from
3768 * the valid path left there.
3770 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3771 if (*jpath == '\0') {
3772 /* Should never happen. */
3773 *sp->f_mntonname = '/';
3775 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3780 * Check with permission for a specific privilege is granted within jail. We
3781 * have a specific list of accepted privileges; the rest are denied.
3784 prison_priv_check(struct ucred *cred, int priv)
3792 * Privileges specific to prisons with a virtual network stack.
3793 * There might be a duplicate entry here in case the privilege
3794 * is only granted conditionally in the legacy jail case.
3799 * NFS-specific privileges.
3801 case PRIV_NFS_DAEMON:
3802 case PRIV_NFS_LOCKD:
3805 * Network stack privileges.
3807 case PRIV_NET_BRIDGE:
3810 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3811 case PRIV_NET_ROUTE:
3813 case PRIV_NET_SETIFMTU:
3814 case PRIV_NET_SETIFFLAGS:
3815 case PRIV_NET_SETIFCAP:
3816 case PRIV_NET_SETIFDESCR:
3817 case PRIV_NET_SETIFNAME :
3818 case PRIV_NET_SETIFMETRIC:
3819 case PRIV_NET_SETIFPHYS:
3820 case PRIV_NET_SETIFMAC:
3821 case PRIV_NET_ADDMULTI:
3822 case PRIV_NET_DELMULTI:
3823 case PRIV_NET_HWIOCTL:
3824 case PRIV_NET_SETLLADDR:
3825 case PRIV_NET_ADDIFGROUP:
3826 case PRIV_NET_DELIFGROUP:
3827 case PRIV_NET_IFCREATE:
3828 case PRIV_NET_IFDESTROY:
3829 case PRIV_NET_ADDIFADDR:
3830 case PRIV_NET_DELIFADDR:
3833 case PRIV_NET_SETIFVNET:
3834 case PRIV_NET_SETIFFIB:
3837 * 802.11-related privileges.
3839 case PRIV_NET80211_GETKEY:
3841 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3846 * AppleTalk privileges.
3848 case PRIV_NETATALK_RESERVEDPORT:
3853 case PRIV_NETATM_CFG:
3854 case PRIV_NETATM_ADD:
3855 case PRIV_NETATM_DEL:
3856 case PRIV_NETATM_SET:
3859 * Bluetooth privileges.
3861 case PRIV_NETBLUETOOTH_RAW:
3865 * Netgraph and netgraph module privileges.
3867 case PRIV_NETGRAPH_CONTROL:
3869 case PRIV_NETGRAPH_TTY:
3873 * IPv4 and IPv6 privileges.
3875 case PRIV_NETINET_IPFW:
3876 case PRIV_NETINET_DIVERT:
3877 case PRIV_NETINET_PF:
3878 case PRIV_NETINET_DUMMYNET:
3879 case PRIV_NETINET_CARP:
3880 case PRIV_NETINET_MROUTE:
3881 case PRIV_NETINET_RAW:
3882 case PRIV_NETINET_ADDRCTRL6:
3883 case PRIV_NETINET_ND6:
3884 case PRIV_NETINET_SCOPE6:
3885 case PRIV_NETINET_ALIFETIME6:
3886 case PRIV_NETINET_IPSEC:
3887 case PRIV_NETINET_BINDANY:
3891 * IPX/SPX privileges.
3893 case PRIV_NETIPX_RESERVEDPORT:
3894 case PRIV_NETIPX_RAW:
3908 * No default: or deny here.
3909 * In case of no permit fall through to next switch().
3911 if (cred->cr_prison->pr_flags & PR_VNET)
3919 * Allow ktrace privileges for root in jail.
3925 * Allow jailed processes to configure audit identity and
3926 * submit audit records (login, etc). In the future we may
3927 * want to further refine the relationship between audit and
3930 case PRIV_AUDIT_GETAUDIT:
3931 case PRIV_AUDIT_SETAUDIT:
3932 case PRIV_AUDIT_SUBMIT:
3936 * Allow jailed processes to manipulate process UNIX
3937 * credentials in any way they see fit.
3939 case PRIV_CRED_SETUID:
3940 case PRIV_CRED_SETEUID:
3941 case PRIV_CRED_SETGID:
3942 case PRIV_CRED_SETEGID:
3943 case PRIV_CRED_SETGROUPS:
3944 case PRIV_CRED_SETREUID:
3945 case PRIV_CRED_SETREGID:
3946 case PRIV_CRED_SETRESUID:
3947 case PRIV_CRED_SETRESGID:
3950 * Jail implements visibility constraints already, so allow
3951 * jailed root to override uid/gid-based constraints.
3953 case PRIV_SEEOTHERGIDS:
3954 case PRIV_SEEOTHERUIDS:
3957 * Jail implements inter-process debugging limits already, so
3958 * allow jailed root various debugging privileges.
3960 case PRIV_DEBUG_DIFFCRED:
3961 case PRIV_DEBUG_SUGID:
3962 case PRIV_DEBUG_UNPRIV:
3965 * Allow jail to set various resource limits and login
3966 * properties, and for now, exceed process resource limits.
3968 case PRIV_PROC_LIMIT:
3969 case PRIV_PROC_SETLOGIN:
3970 case PRIV_PROC_SETRLIMIT:
3973 * System V and POSIX IPC privileges are granted in jail.
3976 case PRIV_IPC_WRITE:
3977 case PRIV_IPC_ADMIN:
3978 case PRIV_IPC_MSGSIZE:
3982 * Jail operations within a jail work on child jails.
3984 case PRIV_JAIL_ATTACH:
3986 case PRIV_JAIL_REMOVE:
3989 * Jail implements its own inter-process limits, so allow
3990 * root processes in jail to change scheduling on other
3991 * processes in the same jail. Likewise for signalling.
3993 case PRIV_SCHED_DIFFCRED:
3994 case PRIV_SCHED_CPUSET:
3995 case PRIV_SIGNAL_DIFFCRED:
3996 case PRIV_SIGNAL_SUGID:
3999 * Allow jailed processes to write to sysctls marked as jail
4002 case PRIV_SYSCTL_WRITEJAIL:
4005 * Allow root in jail to manage a variety of quota
4006 * properties. These should likely be conditional on a
4007 * configuration option.
4009 case PRIV_VFS_GETQUOTA:
4010 case PRIV_VFS_SETQUOTA:
4013 * Since Jail relies on chroot() to implement file system
4014 * protections, grant many VFS privileges to root in jail.
4015 * Be careful to exclude mount-related and NFS-related
4019 case PRIV_VFS_WRITE:
4020 case PRIV_VFS_ADMIN:
4022 case PRIV_VFS_LOOKUP:
4023 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4024 case PRIV_VFS_CHFLAGS_DEV:
4025 case PRIV_VFS_CHOWN:
4026 case PRIV_VFS_CHROOT:
4027 case PRIV_VFS_RETAINSUGID:
4028 case PRIV_VFS_FCHROOT:
4030 case PRIV_VFS_SETGID:
4032 case PRIV_VFS_STICKYFILE:
4035 * As in the non-jail case, non-root users are expected to be
4036 * able to read kernel/phyiscal memory (provided /dev/[k]mem
4037 * exists in the jail and they have permission to access it).
4039 case PRIV_KMEM_READ:
4043 * Depending on the global setting, allow privilege of
4044 * setting system flags.
4046 case PRIV_VFS_SYSFLAGS:
4047 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4053 * Depending on the global setting, allow privilege of
4054 * mounting/unmounting file systems.
4056 case PRIV_VFS_MOUNT:
4057 case PRIV_VFS_UNMOUNT:
4058 case PRIV_VFS_MOUNT_NONUSER:
4059 case PRIV_VFS_MOUNT_OWNER:
4060 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
4061 cred->cr_prison->pr_enforce_statfs < 2)
4067 * Allow jailed root to bind reserved ports and reuse in-use
4070 case PRIV_NETINET_RESERVEDPORT:
4071 case PRIV_NETINET_REUSEPORT:
4075 * Allow jailed root to set certain IPv4/6 (option) headers.
4077 case PRIV_NETINET_SETHDROPTS:
4081 * Conditionally allow creating raw sockets in jail.
4083 case PRIV_NETINET_RAW:
4084 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4090 * Since jail implements its own visibility limits on netstat
4091 * sysctls, allow getcred. This allows identd to work in
4094 case PRIV_NETINET_GETCRED:
4098 * Allow jailed root to set loginclass.
4100 case PRIV_PROC_SETLOGINCLASS:
4105 * In all remaining cases, deny the privilege request. This
4106 * includes almost all network privileges, many system
4107 * configuration privileges.
4114 * Return the part of pr2's name that is relative to pr1, or the whole name
4115 * if it does not directly follow.
4119 prison_name(struct prison *pr1, struct prison *pr2)
4123 /* Jails see themselves as "0" (if they see themselves at all). */
4126 name = pr2->pr_name;
4127 if (prison_ischild(pr1, pr2)) {
4129 * pr1 isn't locked (and allprison_lock may not be either)
4130 * so its length can't be counted on. But the number of dots
4131 * can be counted on - and counted.
4133 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4134 name = strchr(name, '.') + 1;
4140 * Return the part of pr2's path that is relative to pr1, or the whole path
4141 * if it does not directly follow.
4144 prison_path(struct prison *pr1, struct prison *pr2)
4146 char *path1, *path2;
4149 path1 = pr1->pr_path;
4150 path2 = pr2->pr_path;
4151 if (!strcmp(path1, "/"))
4153 len1 = strlen(path1);
4154 if (strncmp(path1, path2, len1))
4156 if (path2[len1] == '\0')
4158 if (path2[len1] == '/')
4159 return (path2 + len1);
4165 * Jail-related sysctls.
4167 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4171 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4174 struct prison *pr, *cpr;
4176 struct in_addr *ip4 = NULL;
4180 struct in6_addr *ip6 = NULL;
4185 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4186 pr = req->td->td_ucred->cr_prison;
4188 sx_slock(&allprison_lock);
4189 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4190 #if defined(INET) || defined(INET6)
4193 mtx_lock(&cpr->pr_mtx);
4195 if (cpr->pr_ip4s > 0) {
4196 if (ip4s < cpr->pr_ip4s) {
4197 ip4s = cpr->pr_ip4s;
4198 mtx_unlock(&cpr->pr_mtx);
4199 ip4 = realloc(ip4, ip4s *
4200 sizeof(struct in_addr), M_TEMP, M_WAITOK);
4203 bcopy(cpr->pr_ip4, ip4,
4204 cpr->pr_ip4s * sizeof(struct in_addr));
4208 if (cpr->pr_ip6s > 0) {
4209 if (ip6s < cpr->pr_ip6s) {
4210 ip6s = cpr->pr_ip6s;
4211 mtx_unlock(&cpr->pr_mtx);
4212 ip6 = realloc(ip6, ip6s *
4213 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4216 bcopy(cpr->pr_ip6, ip6,
4217 cpr->pr_ip6s * sizeof(struct in6_addr));
4220 if (cpr->pr_ref == 0) {
4221 mtx_unlock(&cpr->pr_mtx);
4224 bzero(xp, sizeof(*xp));
4225 xp->pr_version = XPRISON_VERSION;
4226 xp->pr_id = cpr->pr_id;
4227 xp->pr_state = cpr->pr_uref > 0
4228 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4229 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4230 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4231 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4233 xp->pr_ip4s = cpr->pr_ip4s;
4236 xp->pr_ip6s = cpr->pr_ip6s;
4238 mtx_unlock(&cpr->pr_mtx);
4239 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4243 if (xp->pr_ip4s > 0) {
4244 error = SYSCTL_OUT(req, ip4,
4245 xp->pr_ip4s * sizeof(struct in_addr));
4251 if (xp->pr_ip6s > 0) {
4252 error = SYSCTL_OUT(req, ip6,
4253 xp->pr_ip6s * sizeof(struct in6_addr));
4259 sx_sunlock(&allprison_lock);
4270 SYSCTL_OID(_security_jail, OID_AUTO, list,
4271 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4272 sysctl_jail_list, "S", "List of active jails");
4275 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4279 injail = jailed(req->td->td_ucred);
4280 error = SYSCTL_OUT(req, &injail, sizeof(injail));
4285 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4286 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4287 sysctl_jail_jailed, "I", "Process in jail?");
4290 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4292 int error, havevnet;
4294 struct ucred *cred = req->td->td_ucred;
4296 havevnet = jailed(cred) && prison_owns_vnet(cred);
4300 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4305 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4306 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4307 sysctl_jail_vnet, "I", "Jail owns VNET?");
4309 #if defined(INET) || defined(INET6)
4310 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4311 &jail_max_af_ips, 0,
4312 "Number of IP addresses a jail may have at most per address family (deprecated)");
4316 * Default parameters for jail(2) compatibility. For historical reasons,
4317 * the sysctl names have varying similarity to the parameter names. Prisons
4318 * just see their own parameters, and can't change them.
4321 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4324 int allow, error, i;
4326 pr = req->td->td_ucred->cr_prison;
4327 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4329 /* Get the current flag value, and convert it to a boolean. */
4330 i = (allow & arg2) ? 1 : 0;
4333 error = sysctl_handle_int(oidp, &i, 0, req);
4334 if (error || !req->newptr)
4340 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4343 mtx_lock(&prison0.pr_mtx);
4344 jail_default_allow = (jail_default_allow & ~arg2) | i;
4345 mtx_unlock(&prison0.pr_mtx);
4349 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4350 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4351 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4352 "Processes in jail can set their hostnames (deprecated)");
4353 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4354 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4355 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4356 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4357 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4358 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4359 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4360 "Processes in jail can use System V IPC primitives (deprecated)");
4361 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4362 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4363 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4364 "Prison root can create raw sockets (deprecated)");
4365 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4366 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4367 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4368 "Processes in jail can alter system file flags (deprecated)");
4369 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4370 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4371 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4372 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4373 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4374 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4375 NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4376 "Processes in jail can mount the devfs file system (deprecated)");
4377 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4378 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4379 NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4380 "Processes in jail can mount the fdescfs file system (deprecated)");
4381 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4382 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4383 NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4384 "Processes in jail can mount the nullfs file system (deprecated)");
4385 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4386 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4387 NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4388 "Processes in jail can mount the procfs file system (deprecated)");
4389 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
4390 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4391 NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
4392 "Processes in jail can mount the linprocfs file system (deprecated)");
4393 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
4394 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4395 NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
4396 "Processes in jail can mount the linsysfs file system (deprecated)");
4397 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4398 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4399 NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4400 "Processes in jail can mount the tmpfs file system (deprecated)");
4401 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4402 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4403 NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4404 "Processes in jail can mount the zfs file system (deprecated)");
4407 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4412 pr = req->td->td_ucred->cr_prison;
4413 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4414 error = sysctl_handle_int(oidp, &level, 0, req);
4415 if (error || !req->newptr)
4417 *(int *)arg1 = level;
4421 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4422 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4423 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4424 sysctl_jail_default_level, "I",
4425 "Processes in jail cannot see all mounted file systems (deprecated)");
4427 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4428 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4429 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4430 sysctl_jail_default_level, "I",
4431 "Ruleset for the devfs filesystem in jail (deprecated)");
4434 * Nodes to describe jail parameters. Maximum length of string parameters
4435 * is returned in the string itself, and the other parameters exist merely
4436 * to make themselves and their types known.
4438 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4442 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4449 switch (oidp->oid_kind & CTLTYPE)
4455 if (!(req->flags & SCTL_MASK32))
4457 return (SYSCTL_OUT(req, &l, sizeof(l)));
4461 return (SYSCTL_OUT(req, &i, sizeof(i)));
4462 case CTLTYPE_STRING:
4463 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4465 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4466 case CTLTYPE_STRUCT:
4468 return (SYSCTL_OUT(req, &s, sizeof(s)));
4474 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4475 * jail creation time but cannot be changed in an existing jail.
4477 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4478 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4479 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4480 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4481 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4482 "I", "Jail secure level");
4483 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4484 "Jail value for kern.osreldate and uname -K");
4485 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4486 "Jail value for kern.osrelease and uname -r");
4487 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4488 "I", "Jail cannot see all mounted file systems");
4489 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4490 "I", "Ruleset for in-jail devfs mounts");
4491 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4492 "B", "Jail persistence");
4494 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4495 "E,jailsys", "Virtual network stack");
4497 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4498 "B", "Jail is in the process of shutting down");
4500 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4501 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4502 "I", "Current number of child jails");
4503 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4504 "I", "Maximum number of child jails");
4506 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4507 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4509 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4510 "Jail NIS domainname");
4511 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4513 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4514 "LU", "Jail host ID");
4516 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4517 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4520 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4521 "Jail IPv4 address virtualization");
4522 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4523 "S,in_addr,a", "Jail IPv4 addresses");
4524 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4525 "B", "Do (not) use IPv4 source address selection rather than the "
4526 "primary jail IPv4 address.");
4529 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4530 "Jail IPv6 address virtualization");
4531 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4532 "S,in6_addr,a", "Jail IPv6 addresses");
4533 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4534 "B", "Do (not) use IPv6 source address selection rather than the "
4535 "primary jail IPv6 address.");
4538 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4539 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4540 "B", "Jail may set hostname");
4541 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4542 "B", "Jail may use SYSV IPC");
4543 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4544 "B", "Jail may create raw sockets");
4545 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4546 "B", "Jail may alter system file flags");
4547 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4548 "B", "Jail may set file quotas");
4549 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4550 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4552 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4553 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4554 "B", "Jail may mount/unmount jail-friendly file systems in general");
4555 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4556 "B", "Jail may mount the devfs file system");
4557 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4558 "B", "Jail may mount the fdescfs file system");
4559 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4560 "B", "Jail may mount the nullfs file system");
4561 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4562 "B", "Jail may mount the procfs file system");
4563 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
4564 "B", "Jail may mount the linprocfs file system");
4565 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
4566 "B", "Jail may mount the linsysfs file system");
4567 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4568 "B", "Jail may mount the tmpfs file system");
4569 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4570 "B", "Jail may mount the zfs file system");
4574 prison_racct_foreach(void (*callback)(struct racct *racct,
4575 void *arg2, void *arg3), void *arg2, void *arg3)
4577 struct prison_racct *prr;
4579 ASSERT_RACCT_ENABLED();
4581 sx_slock(&allprison_lock);
4582 LIST_FOREACH(prr, &allprison_racct, prr_next)
4583 (callback)(prr->prr_racct, arg2, arg3);
4584 sx_sunlock(&allprison_lock);
4587 static struct prison_racct *
4588 prison_racct_find_locked(const char *name)
4590 struct prison_racct *prr;
4592 ASSERT_RACCT_ENABLED();
4593 sx_assert(&allprison_lock, SA_XLOCKED);
4595 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4598 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4599 if (strcmp(name, prr->prr_name) != 0)
4602 /* Found prison_racct with a matching name? */
4603 prison_racct_hold(prr);
4607 /* Add new prison_racct. */
4608 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4609 racct_create(&prr->prr_racct);
4611 strcpy(prr->prr_name, name);
4612 refcount_init(&prr->prr_refcount, 1);
4613 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4618 struct prison_racct *
4619 prison_racct_find(const char *name)
4621 struct prison_racct *prr;
4623 ASSERT_RACCT_ENABLED();
4625 sx_xlock(&allprison_lock);
4626 prr = prison_racct_find_locked(name);
4627 sx_xunlock(&allprison_lock);
4632 prison_racct_hold(struct prison_racct *prr)
4635 ASSERT_RACCT_ENABLED();
4637 refcount_acquire(&prr->prr_refcount);
4641 prison_racct_free_locked(struct prison_racct *prr)
4644 ASSERT_RACCT_ENABLED();
4645 sx_assert(&allprison_lock, SA_XLOCKED);
4647 if (refcount_release(&prr->prr_refcount)) {
4648 racct_destroy(&prr->prr_racct);
4649 LIST_REMOVE(prr, prr_next);
4650 free(prr, M_PRISON_RACCT);
4655 prison_racct_free(struct prison_racct *prr)
4659 ASSERT_RACCT_ENABLED();
4660 sx_assert(&allprison_lock, SA_UNLOCKED);
4662 old = prr->prr_refcount;
4663 if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4666 sx_xlock(&allprison_lock);
4667 prison_racct_free_locked(prr);
4668 sx_xunlock(&allprison_lock);
4672 prison_racct_attach(struct prison *pr)
4674 struct prison_racct *prr;
4676 ASSERT_RACCT_ENABLED();
4677 sx_assert(&allprison_lock, SA_XLOCKED);
4679 prr = prison_racct_find_locked(pr->pr_name);
4680 KASSERT(prr != NULL, ("cannot find prison_racct"));
4682 pr->pr_prison_racct = prr;
4686 * Handle jail renaming. From the racct point of view, renaming means
4687 * moving from one prison_racct to another.
4690 prison_racct_modify(struct prison *pr)
4694 struct prison_racct *oldprr;
4696 ASSERT_RACCT_ENABLED();
4698 sx_slock(&allproc_lock);
4699 sx_xlock(&allprison_lock);
4701 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4702 sx_xunlock(&allprison_lock);
4703 sx_sunlock(&allproc_lock);
4707 oldprr = pr->pr_prison_racct;
4708 pr->pr_prison_racct = NULL;
4710 prison_racct_attach(pr);
4713 * Move resource utilisation records.
4715 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4718 * Force rctl to reattach rules to processes.
4720 FOREACH_PROC_IN_SYSTEM(p) {
4722 cred = crhold(p->p_ucred);
4724 racct_proc_ucred_changed(p, cred, cred);
4728 sx_sunlock(&allproc_lock);
4729 prison_racct_free_locked(oldprr);
4730 sx_xunlock(&allprison_lock);
4734 prison_racct_detach(struct prison *pr)
4737 ASSERT_RACCT_ENABLED();
4738 sx_assert(&allprison_lock, SA_UNLOCKED);
4740 if (pr->pr_prison_racct == NULL)
4742 prison_racct_free(pr->pr_prison_racct);
4743 pr->pr_prison_racct = NULL;
4750 db_show_prison(struct prison *pr)
4753 #if defined(INET) || defined(INET6)
4758 char ip6buf[INET6_ADDRSTRLEN];
4761 db_printf("prison %p:\n", pr);
4762 db_printf(" jid = %d\n", pr->pr_id);
4763 db_printf(" name = %s\n", pr->pr_name);
4764 db_printf(" parent = %p\n", pr->pr_parent);
4765 db_printf(" ref = %d\n", pr->pr_ref);
4766 db_printf(" uref = %d\n", pr->pr_uref);
4767 db_printf(" path = %s\n", pr->pr_path);
4768 db_printf(" cpuset = %d\n", pr->pr_cpuset
4769 ? pr->pr_cpuset->cs_id : -1);
4771 db_printf(" vnet = %p\n", pr->pr_vnet);
4773 db_printf(" root = %p\n", pr->pr_root);
4774 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4775 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4776 db_printf(" children.max = %d\n", pr->pr_childmax);
4777 db_printf(" children.cur = %d\n", pr->pr_childcount);
4778 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4779 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4780 db_printf(" flags = 0x%x", pr->pr_flags);
4781 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4783 if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4784 db_printf(" %s", pr_flag_names[fi]);
4785 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4787 jsf = pr->pr_flags &
4788 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4789 db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4790 pr_flag_jailsys[fi].disable &&
4791 (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4792 : (jsf == pr_flag_jailsys[fi].new) ? "new"
4795 db_printf(" allow = 0x%x", pr->pr_allow);
4796 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4798 if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4799 db_printf(" %s", pr_allow_names[fi]);
4801 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4802 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4803 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4804 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4805 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4807 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4808 for (ii = 0; ii < pr->pr_ip4s; ii++)
4809 db_printf(" %s %s\n",
4810 ii == 0 ? "ip4.addr =" : " ",
4811 inet_ntoa(pr->pr_ip4[ii]));
4814 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4815 for (ii = 0; ii < pr->pr_ip6s; ii++)
4816 db_printf(" %s %s\n",
4817 ii == 0 ? "ip6.addr =" : " ",
4818 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4822 DB_SHOW_COMMAND(prison, db_show_prison_command)
4828 * Show all prisons in the list, and prison0 which is not
4831 db_show_prison(&prison0);
4832 if (!db_pager_quit) {
4833 TAILQ_FOREACH(pr, &allprison, pr_list) {
4845 /* Look for a prison with the ID and with references. */
4846 TAILQ_FOREACH(pr, &allprison, pr_list)
4847 if (pr->pr_id == addr && pr->pr_ref > 0)
4850 /* Look again, without requiring a reference. */
4851 TAILQ_FOREACH(pr, &allprison, pr_list)
4852 if (pr->pr_id == addr)
4855 /* Assume address points to a valid prison. */
4856 pr = (struct prison *)addr;