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 fullpath_disabled = 0;
937 error = vfs_getopt(opts, "path", (void **)&path, &len);
943 if (flags & JAIL_UPDATE) {
946 "path cannot be changed after creation");
949 if (len == 0 || path[len - 1] != '\0') {
953 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
959 NDFREE(&nd, NDF_ONLY_PNBUF);
960 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
961 strlcpy(g_path, path, MAXPATHLEN);
962 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
965 else if (error == ENODEV) {
966 /* proceed if sysctl debug.disablefullpath == 1 */
967 fullpath_disabled = 1;
968 if (len < 2 || (len == 2 && path[0] == '/'))
971 /* exit on other errors */
974 if (root->v_type != VDIR) {
980 if (fullpath_disabled) {
981 /* Leave room for a real-root full pathname. */
982 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
983 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
984 error = ENAMETOOLONG;
990 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
996 if (flags & JAIL_UPDATE) {
999 "osrelease cannot be changed after creation");
1002 if (len == 0 || len >= OSRELEASELEN) {
1005 "osrelease string must be 1-%d bytes long",
1011 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1012 if (error == ENOENT)
1014 else if (error != 0)
1017 if (flags & JAIL_UPDATE) {
1020 "osreldate cannot be changed after creation");
1025 vfs_opterror(opts, "osreldate cannot be 0");
1031 * Find the specified jail, or at least its parent.
1032 * This abuses the file error codes ENOENT and EEXIST.
1036 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1037 namelc = strrchr(name, '.');
1038 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1042 sx_xlock(&allprison_lock);
1045 * See if a requested jid already exists. There is an
1046 * information leak here if the jid exists but is not within
1047 * the caller's jail hierarchy. Jail creators will get EEXIST
1048 * even though they cannot see the jail, and CREATE | UPDATE
1049 * will return ENOENT which is not normally a valid error.
1053 vfs_opterror(opts, "negative jid");
1054 goto done_unlock_list;
1056 pr = prison_find(jid);
1058 ppr = pr->pr_parent;
1059 /* Create: jid must not exist. */
1060 if (cuflags == JAIL_CREATE) {
1061 mtx_unlock(&pr->pr_mtx);
1063 vfs_opterror(opts, "jail %d already exists",
1065 goto done_unlock_list;
1067 if (!prison_ischild(mypr, pr)) {
1068 mtx_unlock(&pr->pr_mtx);
1070 } else if (pr->pr_uref == 0) {
1071 if (!(flags & JAIL_DYING)) {
1072 mtx_unlock(&pr->pr_mtx);
1074 vfs_opterror(opts, "jail %d is dying",
1076 goto done_unlock_list;
1077 } else if ((flags & JAIL_ATTACH) ||
1078 (pr_flags & PR_PERSIST)) {
1080 * A dying jail might be resurrected
1081 * (via attach or persist), but first
1082 * it must determine if another jail
1083 * has claimed its name. Accomplish
1084 * this by implicitly re-setting the
1088 name = prison_name(mypr, pr);
1093 /* Update: jid must exist. */
1094 if (cuflags == JAIL_UPDATE) {
1096 vfs_opterror(opts, "jail %d not found", jid);
1097 goto done_unlock_list;
1102 * If the caller provided a name, look for a jail by that name.
1103 * This has different semantics for creates and updates keyed by jid
1104 * (where the name must not already exist in a different jail),
1105 * and updates keyed by the name itself (where the name must exist
1106 * because that is the jail being updated).
1110 namelc = strrchr(name, '.');
1115 * This is a hierarchical name. Split it into the
1116 * parent and child names, and make sure the parent
1117 * exists or matches an already found jail.
1120 if (strncmp(name, ppr->pr_name, namelc - name)
1121 || ppr->pr_name[namelc - name] != '\0') {
1122 mtx_unlock(&pr->pr_mtx);
1125 "cannot change jail's parent");
1126 goto done_unlock_list;
1130 ppr = prison_find_name(mypr, name);
1134 "jail \"%s\" not found", name);
1135 goto done_unlock_list;
1137 mtx_unlock(&ppr->pr_mtx);
1142 if (namelc[0] != '\0') {
1144 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1147 FOREACH_PRISON_CHILD(ppr, tpr) {
1148 if (tpr != pr && tpr->pr_ref > 0 &&
1149 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1151 cuflags != JAIL_CREATE) {
1152 mtx_lock(&tpr->pr_mtx);
1153 if (tpr->pr_ref > 0) {
1158 if (tpr->pr_uref > 0) {
1164 mtx_unlock(&tpr->pr_mtx);
1165 } else if (tpr->pr_uref > 0) {
1167 * Create, or update(jid):
1168 * name must not exist in an
1169 * active sibling jail.
1173 mtx_unlock(&pr->pr_mtx);
1175 "jail \"%s\" already exists",
1177 goto done_unlock_list;
1181 /* If no active jail is found, use a dying one. */
1182 if (deadpr != NULL && pr == NULL) {
1183 if (flags & JAIL_DYING) {
1184 mtx_lock(&deadpr->pr_mtx);
1185 if (deadpr->pr_ref == 0) {
1186 mtx_unlock(&deadpr->pr_mtx);
1190 } else if (cuflags == JAIL_UPDATE) {
1193 "jail \"%s\" is dying", name);
1194 goto done_unlock_list;
1197 /* Update: name must exist if no jid. */
1198 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1200 vfs_opterror(opts, "jail \"%s\" not found",
1202 goto done_unlock_list;
1206 /* Update: must provide a jid or name. */
1207 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1209 vfs_opterror(opts, "update specified no jail");
1210 goto done_unlock_list;
1213 /* If there's no prison to update, create a new one and link it in. */
1215 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1216 if (tpr->pr_childcount >= tpr->pr_childmax) {
1218 vfs_opterror(opts, "prison limit exceeded");
1219 goto done_unlock_list;
1222 mtx_lock(&ppr->pr_mtx);
1223 if (ppr->pr_ref == 0) {
1224 mtx_unlock(&ppr->pr_mtx);
1226 vfs_opterror(opts, "jail \"%s\" not found",
1227 prison_name(mypr, ppr));
1228 goto done_unlock_list;
1232 mtx_unlock(&ppr->pr_mtx);
1233 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1235 /* Find the next free jid. */
1238 if (jid == JAIL_MAX)
1240 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1241 if (tpr->pr_id < jid)
1243 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1244 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1247 if (jid == lastprid) {
1250 "no available jail IDs");
1252 prison_deref(ppr, PD_DEREF |
1253 PD_DEUREF | PD_LIST_XLOCKED);
1262 * The jail already has a jid (that did not yet exist),
1263 * so just find where to insert it.
1265 TAILQ_FOREACH(tpr, &allprison, pr_list)
1266 if (tpr->pr_id >= jid) {
1267 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1272 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1273 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1274 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1275 tpr->pr_childcount++;
1277 pr->pr_parent = ppr;
1280 /* Set some default values, and inherit some from the parent. */
1285 root = mypr->pr_root;
1288 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1289 pr->pr_flags |= PR_HOST;
1290 #if defined(INET) || defined(INET6)
1292 if (!(pr_flags & PR_VNET))
1296 if (!(ch_flags & PR_IP4_USER))
1298 PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1299 else if (!(pr_flags & PR_IP4_USER)) {
1300 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1301 if (ppr->pr_ip4 != NULL) {
1302 pr->pr_ip4s = ppr->pr_ip4s;
1303 pr->pr_ip4 = malloc(pr->pr_ip4s *
1304 sizeof(struct in_addr), M_PRISON,
1306 bcopy(ppr->pr_ip4, pr->pr_ip4,
1307 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1312 if (!(ch_flags & PR_IP6_USER))
1314 PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1315 else if (!(pr_flags & PR_IP6_USER)) {
1316 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1317 if (ppr->pr_ip6 != NULL) {
1318 pr->pr_ip6s = ppr->pr_ip6s;
1319 pr->pr_ip6 = malloc(pr->pr_ip6s *
1320 sizeof(struct in6_addr), M_PRISON,
1322 bcopy(ppr->pr_ip6, pr->pr_ip6,
1323 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1329 /* Source address selection is always on by default. */
1330 pr->pr_flags |= _PR_IP_SADDRSEL;
1332 pr->pr_securelevel = ppr->pr_securelevel;
1333 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1334 pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1335 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1337 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1338 if (osrelstr == NULL)
1339 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1341 strcpy(pr->pr_osrelease, osrelstr);
1343 LIST_INIT(&pr->pr_children);
1344 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1345 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1348 /* Allocate a new vnet if specified. */
1349 pr->pr_vnet = (pr_flags & PR_VNET)
1350 ? vnet_alloc() : ppr->pr_vnet;
1353 * Allocate a dedicated cpuset for each jail.
1354 * Unlike other initial settings, this may return an erorr.
1356 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1358 prison_deref(pr, PD_LIST_XLOCKED);
1362 mtx_lock(&pr->pr_mtx);
1364 * New prisons do not yet have a reference, because we do not
1365 * want others to see the incomplete prison once the
1366 * allprison_lock is downgraded.
1371 * Grab a reference for existing prisons, to ensure they
1372 * continue to exist for the duration of the call.
1375 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1376 if ((pr->pr_flags & PR_VNET) &&
1377 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1380 "vnet jails cannot have IP address restrictions");
1381 goto done_deref_locked;
1385 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1388 "ip4 cannot be changed after creation");
1389 goto done_deref_locked;
1393 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1396 "ip6 cannot be changed after creation");
1397 goto done_deref_locked;
1402 /* Do final error checking before setting anything. */
1404 if (slevel < ppr->pr_securelevel) {
1406 goto done_deref_locked;
1410 if (childmax >= ppr->pr_childmax) {
1412 goto done_deref_locked;
1416 if (enforce < ppr->pr_enforce_statfs) {
1418 goto done_deref_locked;
1423 * devfs_rsnum is a uint16_t
1425 if (rsnum < 0 || rsnum > 65535) {
1427 goto done_deref_locked;
1430 * Nested jails always inherit parent's devfs ruleset
1432 if (jailed(td->td_ucred)) {
1433 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1435 goto done_deref_locked;
1437 rsnum = ppr->pr_devfs_rsnum;
1442 if (ppr->pr_flags & PR_IP4) {
1444 * Make sure the new set of IP addresses is a
1445 * subset of the parent's list. Don't worry
1446 * about the parent being unlocked, as any
1447 * setting is done with allprison_lock held.
1449 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1450 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1452 if (ij == ppr->pr_ip4s) {
1454 goto done_deref_locked;
1457 for (ii = ij = 1; ii < ip4s; ii++) {
1458 if (ip4[ii].s_addr ==
1459 ppr->pr_ip4[0].s_addr)
1461 for (; ij < ppr->pr_ip4s; ij++)
1462 if (ip4[ii].s_addr ==
1463 ppr->pr_ip4[ij].s_addr)
1465 if (ij == ppr->pr_ip4s)
1468 if (ij == ppr->pr_ip4s) {
1470 goto done_deref_locked;
1475 * Check for conflicting IP addresses. We permit them
1476 * if there is no more than one IP on each jail. If
1477 * there is a duplicate on a jail with more than one
1478 * IP stop checking and return error.
1482 for (; tppr != &prison0; tppr = tppr->pr_parent)
1483 if (tppr->pr_flags & PR_VNET)
1486 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1489 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1491 tpr->pr_uref == 0) {
1495 if (!(tpr->pr_flags & PR_IP4_USER))
1498 if (tpr->pr_ip4 == NULL ||
1499 (ip4s == 1 && tpr->pr_ip4s == 1))
1501 for (ii = 0; ii < ip4s; ii++) {
1502 if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1505 "IPv4 addresses clash");
1506 goto done_deref_locked;
1514 if (ppr->pr_flags & PR_IP6) {
1516 * Make sure the new set of IP addresses is a
1517 * subset of the parent's list.
1519 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1520 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1523 if (ij == ppr->pr_ip6s) {
1525 goto done_deref_locked;
1528 for (ii = ij = 1; ii < ip6s; ii++) {
1529 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1532 for (; ij < ppr->pr_ip6s; ij++)
1533 if (IN6_ARE_ADDR_EQUAL(
1534 &ip6[ii], &ppr->pr_ip6[ij]))
1536 if (ij == ppr->pr_ip6s)
1539 if (ij == ppr->pr_ip6s) {
1541 goto done_deref_locked;
1545 /* Check for conflicting IP addresses. */
1548 for (; tppr != &prison0; tppr = tppr->pr_parent)
1549 if (tppr->pr_flags & PR_VNET)
1552 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1555 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1557 tpr->pr_uref == 0) {
1561 if (!(tpr->pr_flags & PR_IP6_USER))
1564 if (tpr->pr_ip6 == NULL ||
1565 (ip6s == 1 && tpr->pr_ip6s == 1))
1567 for (ii = 0; ii < ip6s; ii++) {
1568 if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1571 "IPv6 addresses clash");
1572 goto done_deref_locked;
1578 onamelen = namelen = 0;
1579 if (namelc != NULL) {
1580 /* Give a default name of the jid. Also allow the name to be
1581 * explicitly the jid - but not any other number, and only in
1582 * normal form (no leading zero/etc).
1584 if (namelc[0] == '\0')
1585 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1586 else if ((strtoul(namelc, &p, 10) != jid ||
1587 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1590 "name cannot be numeric (unless it is the jid)");
1591 goto done_deref_locked;
1594 * Make sure the name isn't too long for the prison or its
1597 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1598 onamelen = strlen(pr->pr_name + pnamelen);
1599 namelen = strlen(namelc);
1600 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1601 error = ENAMETOOLONG;
1602 goto done_deref_locked;
1604 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1605 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1606 sizeof(pr->pr_name)) {
1607 error = ENAMETOOLONG;
1608 goto done_deref_locked;
1612 if (pr_allow & ~ppr->pr_allow) {
1614 goto done_deref_locked;
1618 * Let modules check their parameters. This requires unlocking and
1619 * then re-locking the prison, but this is still a valid state as long
1620 * as allprison_lock remains xlocked.
1622 mtx_unlock(&pr->pr_mtx);
1623 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1625 prison_deref(pr, created
1627 : PD_DEREF | PD_LIST_XLOCKED);
1630 mtx_lock(&pr->pr_mtx);
1632 /* At this point, all valid parameters should have been noted. */
1633 TAILQ_FOREACH(opt, opts, link) {
1634 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1636 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1637 goto done_deref_locked;
1641 /* Set the parameters of the prison. */
1644 if (pr_flags & PR_IP4_USER) {
1645 pr->pr_flags |= PR_IP4;
1646 free(pr->pr_ip4, M_PRISON);
1650 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1652 if (tpr->pr_flags & PR_VNET) {
1657 if (prison_restrict_ip4(tpr, NULL)) {
1666 if (pr_flags & PR_IP6_USER) {
1667 pr->pr_flags |= PR_IP6;
1668 free(pr->pr_ip6, M_PRISON);
1672 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1674 if (tpr->pr_flags & PR_VNET) {
1679 if (prison_restrict_ip6(tpr, NULL)) {
1687 pr->pr_securelevel = slevel;
1688 /* Set all child jails to be at least this level. */
1689 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1690 if (tpr->pr_securelevel < slevel)
1691 tpr->pr_securelevel = slevel;
1694 pr->pr_childmax = childmax;
1695 /* Set all child jails to under this limit. */
1696 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1697 if (tpr->pr_childmax > childmax - level)
1698 tpr->pr_childmax = childmax > level
1699 ? childmax - level : 0;
1702 pr->pr_enforce_statfs = enforce;
1703 /* Pass this restriction on to the children. */
1704 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1705 if (tpr->pr_enforce_statfs < enforce)
1706 tpr->pr_enforce_statfs = enforce;
1709 pr->pr_devfs_rsnum = rsnum;
1710 /* Pass this restriction on to the children. */
1711 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1712 tpr->pr_devfs_rsnum = rsnum;
1714 if (namelc != NULL) {
1715 if (ppr == &prison0)
1716 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1718 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1719 ppr->pr_name, namelc);
1720 /* Change this component of child names. */
1721 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1722 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1723 strlen(tpr->pr_name + onamelen) + 1);
1724 bcopy(pr->pr_name, tpr->pr_name, namelen);
1728 /* Try to keep a real-rooted full pathname. */
1729 if (fullpath_disabled && path[0] == '/' &&
1730 strcmp(mypr->pr_path, "/"))
1731 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1732 mypr->pr_path, path);
1734 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1737 if (PR_HOST & ch_flags & ~pr_flags) {
1738 if (pr->pr_flags & PR_HOST) {
1740 * Copy the parent's host info. As with pr_ip4 above,
1741 * the lack of a lock on the parent is not a problem;
1742 * it is always set with allprison_lock at least
1743 * shared, and is held exclusively here.
1745 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1746 sizeof(pr->pr_hostname));
1747 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1748 sizeof(pr->pr_domainname));
1749 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1750 sizeof(pr->pr_hostuuid));
1751 pr->pr_hostid = pr->pr_parent->pr_hostid;
1753 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1754 /* Set this prison, and any descendants without PR_HOST. */
1756 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1758 strlcpy(pr->pr_domainname, domain,
1759 sizeof(pr->pr_domainname));
1761 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1763 pr->pr_hostid = hid;
1764 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1765 if (tpr->pr_flags & PR_HOST)
1769 strlcpy(tpr->pr_hostname,
1771 sizeof(tpr->pr_hostname));
1773 strlcpy(tpr->pr_domainname,
1775 sizeof(tpr->pr_domainname));
1777 strlcpy(tpr->pr_hostuuid,
1779 sizeof(tpr->pr_hostuuid));
1781 tpr->pr_hostid = hid;
1785 if ((tallow = ch_allow & ~pr_allow)) {
1786 /* Clear allow bits in all children. */
1787 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1788 tpr->pr_allow &= ~tallow;
1790 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1792 * Persistent prisons get an extra reference, and prisons losing their
1793 * persist flag lose that reference. Only do this for existing prisons
1794 * for now, so new ones will remain unseen until after the module
1795 * handlers have completed.
1797 born = pr->pr_uref == 0;
1798 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1799 if (pr_flags & PR_PERSIST) {
1807 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1808 mtx_unlock(&pr->pr_mtx);
1811 if (racct_enable && created)
1812 prison_racct_attach(pr);
1815 /* Locks may have prevented a complete restriction of child IP
1816 * addresses. If so, allocate some more memory and try again.
1821 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1822 mtx_lock(&pr->pr_mtx);
1824 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1826 if (tpr->pr_flags & PR_VNET) {
1831 if (prison_restrict_ip4(tpr, ip4)) {
1838 mtx_unlock(&pr->pr_mtx);
1844 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1845 mtx_lock(&pr->pr_mtx);
1847 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1849 if (tpr->pr_flags & PR_VNET) {
1854 if (prison_restrict_ip6(tpr, ip6)) {
1861 mtx_unlock(&pr->pr_mtx);
1865 /* Let the modules do their work. */
1866 sx_downgrade(&allprison_lock);
1868 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1870 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1871 prison_deref(pr, created
1873 : PD_DEREF | PD_LIST_SLOCKED);
1877 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1880 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1881 prison_deref(pr, created
1883 : PD_DEREF | PD_LIST_SLOCKED);
1887 /* Attach this process to the prison if requested. */
1888 if (flags & JAIL_ATTACH) {
1889 mtx_lock(&pr->pr_mtx);
1890 error = do_jail_attach(td, pr);
1892 vfs_opterror(opts, "attach failed");
1894 prison_deref(pr, PD_DEREF);
1900 if (racct_enable && !created) {
1901 if (!(flags & JAIL_ATTACH))
1902 sx_sunlock(&allprison_lock);
1903 prison_racct_modify(pr);
1904 if (!(flags & JAIL_ATTACH))
1905 sx_slock(&allprison_lock);
1909 td->td_retval[0] = pr->pr_id;
1912 * Now that it is all there, drop the temporary reference from existing
1913 * prisons. Or add a reference to newly created persistent prisons
1914 * (which was not done earlier so that the prison would not be publicly
1918 prison_deref(pr, (flags & JAIL_ATTACH)
1920 : PD_DEREF | PD_LIST_SLOCKED);
1922 if (pr_flags & PR_PERSIST) {
1923 mtx_lock(&pr->pr_mtx);
1926 mtx_unlock(&pr->pr_mtx);
1928 if (!(flags & JAIL_ATTACH))
1929 sx_sunlock(&allprison_lock);
1935 prison_deref(pr, created
1936 ? PD_LOCKED | PD_LIST_XLOCKED
1937 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1940 sx_xunlock(&allprison_lock);
1946 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1947 if (errmsg_len > 0) {
1948 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1949 if (errmsg_pos > 0) {
1950 if (optuio->uio_segflg == UIO_SYSSPACE)
1952 optuio->uio_iov[errmsg_pos].iov_base,
1956 optuio->uio_iov[errmsg_pos].iov_base,
1963 free(ip4, M_PRISON);
1966 free(ip6, M_PRISON);
1969 free(g_path, M_TEMP);
1976 * struct jail_get_args {
1977 * struct iovec *iovp;
1978 * unsigned int iovcnt;
1983 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1988 /* Check that we have an even number of iovecs. */
1989 if (uap->iovcnt & 1)
1992 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1995 error = kern_jail_get(td, auio, uap->flags);
1997 error = copyout(auio->uio_iov, uap->iovp,
1998 uap->iovcnt * sizeof (struct iovec));
2004 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2006 struct prison *pr, *mypr;
2008 struct vfsoptlist *opts;
2009 char *errmsg, *name;
2010 int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
2012 if (flags & ~JAIL_GET_MASK)
2015 /* Get the parameter list. */
2016 error = vfs_buildopts(optuio, &opts);
2019 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2020 mypr = td->td_ucred->cr_prison;
2023 * Find the prison specified by one of: lastjid, jid, name.
2025 sx_slock(&allprison_lock);
2026 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2028 TAILQ_FOREACH(pr, &allprison, pr_list) {
2029 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2030 mtx_lock(&pr->pr_mtx);
2031 if (pr->pr_ref > 0 &&
2032 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
2034 mtx_unlock(&pr->pr_mtx);
2040 vfs_opterror(opts, "no jail after %d", jid);
2041 goto done_unlock_list;
2042 } else if (error != ENOENT)
2043 goto done_unlock_list;
2045 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2048 pr = prison_find_child(mypr, jid);
2050 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2051 mtx_unlock(&pr->pr_mtx);
2053 vfs_opterror(opts, "jail %d is dying",
2055 goto done_unlock_list;
2060 vfs_opterror(opts, "jail %d not found", jid);
2061 goto done_unlock_list;
2063 } else if (error != ENOENT)
2064 goto done_unlock_list;
2066 error = vfs_getopt(opts, "name", (void **)&name, &len);
2068 if (len == 0 || name[len - 1] != '\0') {
2070 goto done_unlock_list;
2072 pr = prison_find_name(mypr, name);
2074 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2075 mtx_unlock(&pr->pr_mtx);
2077 vfs_opterror(opts, "jail \"%s\" is dying",
2079 goto done_unlock_list;
2084 vfs_opterror(opts, "jail \"%s\" not found", name);
2085 goto done_unlock_list;
2086 } else if (error != ENOENT)
2087 goto done_unlock_list;
2089 vfs_opterror(opts, "no jail specified");
2091 goto done_unlock_list;
2094 /* Get the parameters of the prison. */
2097 td->td_retval[0] = pr->pr_id;
2098 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2099 if (error != 0 && error != ENOENT)
2101 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2102 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2103 if (error != 0 && error != ENOENT)
2105 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2106 if (error != 0 && error != ENOENT)
2108 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2109 sizeof(pr->pr_cpuset->cs_id));
2110 if (error != 0 && error != ENOENT)
2112 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2113 if (error != 0 && error != ENOENT)
2116 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2117 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2118 if (error != 0 && error != ENOENT)
2122 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2123 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2124 if (error != 0 && error != ENOENT)
2127 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2128 sizeof(pr->pr_securelevel));
2129 if (error != 0 && error != ENOENT)
2131 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2132 sizeof(pr->pr_childcount));
2133 if (error != 0 && error != ENOENT)
2135 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2136 sizeof(pr->pr_childmax));
2137 if (error != 0 && error != ENOENT)
2139 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2140 if (error != 0 && error != ENOENT)
2142 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2143 if (error != 0 && error != ENOENT)
2145 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2146 if (error != 0 && error != ENOENT)
2148 #ifdef COMPAT_FREEBSD32
2149 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2150 uint32_t hid32 = pr->pr_hostid;
2152 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2155 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2156 sizeof(pr->pr_hostid));
2157 if (error != 0 && error != ENOENT)
2159 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2160 sizeof(pr->pr_enforce_statfs));
2161 if (error != 0 && error != ENOENT)
2163 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2164 sizeof(pr->pr_devfs_rsnum));
2165 if (error != 0 && error != ENOENT)
2167 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2169 if (pr_flag_names[fi] == NULL)
2171 i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2172 error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2173 if (error != 0 && error != ENOENT)
2176 error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2177 if (error != 0 && error != ENOENT)
2180 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2183 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2184 i = pr_flag_jailsys[fi].disable &&
2185 (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2186 : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2189 vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2190 if (error != 0 && error != ENOENT)
2193 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2195 if (pr_allow_names[fi] == NULL)
2197 i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2198 error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2199 if (error != 0 && error != ENOENT)
2202 error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2203 if (error != 0 && error != ENOENT)
2206 i = (pr->pr_uref == 0);
2207 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2208 if (error != 0 && error != ENOENT)
2211 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2212 if (error != 0 && error != ENOENT)
2214 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2215 sizeof(pr->pr_osreldate));
2216 if (error != 0 && error != ENOENT)
2218 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2219 if (error != 0 && error != ENOENT)
2222 /* Get the module parameters. */
2223 mtx_unlock(&pr->pr_mtx);
2225 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2228 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2230 /* By now, all parameters should have been noted. */
2231 TAILQ_FOREACH(opt, opts, link) {
2232 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2234 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2239 /* Write the fetched parameters back to userspace. */
2241 TAILQ_FOREACH(opt, opts, link) {
2242 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2243 pos = 2 * opt->pos + 1;
2244 optuio->uio_iov[pos].iov_len = opt->len;
2245 if (opt->value != NULL) {
2246 if (optuio->uio_segflg == UIO_SYSSPACE) {
2248 optuio->uio_iov[pos].iov_base,
2251 error = copyout(opt->value,
2252 optuio->uio_iov[pos].iov_base,
2263 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2267 sx_sunlock(&allprison_lock);
2269 if (error && errmsg_pos >= 0) {
2270 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2271 errmsg_pos = 2 * errmsg_pos + 1;
2272 if (errmsg_len > 0) {
2273 if (optuio->uio_segflg == UIO_SYSSPACE)
2275 optuio->uio_iov[errmsg_pos].iov_base,
2279 optuio->uio_iov[errmsg_pos].iov_base,
2289 * struct jail_remove_args {
2294 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2296 struct prison *pr, *cpr, *lpr, *tpr;
2299 error = priv_check(td, PRIV_JAIL_REMOVE);
2303 sx_xlock(&allprison_lock);
2304 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2306 sx_xunlock(&allprison_lock);
2310 /* Remove all descendants of this prison, then remove this prison. */
2312 if (!LIST_EMPTY(&pr->pr_children)) {
2313 mtx_unlock(&pr->pr_mtx);
2315 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2316 mtx_lock(&cpr->pr_mtx);
2317 if (cpr->pr_ref > 0) {
2321 /* Already removed - do not do it again. */
2324 mtx_unlock(&cpr->pr_mtx);
2326 mtx_lock(&lpr->pr_mtx);
2327 prison_remove_one(lpr);
2328 sx_xlock(&allprison_lock);
2333 mtx_lock(&lpr->pr_mtx);
2334 prison_remove_one(lpr);
2335 sx_xlock(&allprison_lock);
2337 mtx_lock(&pr->pr_mtx);
2339 prison_remove_one(pr);
2344 prison_remove_one(struct prison *pr)
2349 /* If the prison was persistent, it is not anymore. */
2351 if (pr->pr_flags & PR_PERSIST) {
2354 pr->pr_flags &= ~PR_PERSIST;
2358 * jail_remove added a reference. If that's the only one, remove
2361 KASSERT(pr->pr_ref > 0,
2362 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2363 if (pr->pr_ref == 1) {
2365 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2369 mtx_unlock(&pr->pr_mtx);
2370 sx_xunlock(&allprison_lock);
2372 * Kill all processes unfortunate enough to be attached to this prison.
2374 sx_slock(&allproc_lock);
2375 LIST_FOREACH(p, &allproc, p_list) {
2377 if (p->p_state != PRS_NEW && p->p_ucred &&
2378 p->p_ucred->cr_prison == pr)
2379 kern_psignal(p, SIGKILL);
2382 sx_sunlock(&allproc_lock);
2383 /* Remove the temporary reference added by jail_remove. */
2384 prison_deref(pr, deuref | PD_DEREF);
2389 * struct jail_attach_args {
2394 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2399 error = priv_check(td, PRIV_JAIL_ATTACH);
2404 * Start with exclusive hold on allprison_lock to ensure that a possible
2405 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2406 * But then immediately downgrade it since we don't need to stop
2409 sx_xlock(&allprison_lock);
2410 sx_downgrade(&allprison_lock);
2411 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2413 sx_sunlock(&allprison_lock);
2418 * Do not allow a process to attach to a prison that is not
2419 * considered to be "alive".
2421 if (pr->pr_uref == 0) {
2422 mtx_unlock(&pr->pr_mtx);
2423 sx_sunlock(&allprison_lock);
2427 return (do_jail_attach(td, pr));
2431 do_jail_attach(struct thread *td, struct prison *pr)
2434 struct ucred *newcred, *oldcred;
2438 * XXX: Note that there is a slight race here if two threads
2439 * in the same privileged process attempt to attach to two
2440 * different jails at the same time. It is important for
2441 * user processes not to do this, or they might end up with
2442 * a process root from one prison, but attached to the jail
2447 mtx_unlock(&pr->pr_mtx);
2449 /* Let modules do whatever they need to prepare for attaching. */
2450 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2452 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2455 sx_sunlock(&allprison_lock);
2458 * Reparent the newly attached process to this jail.
2461 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2465 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2466 if ((error = change_dir(pr->pr_root, td)) != 0)
2469 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2472 VOP_UNLOCK(pr->pr_root, 0);
2473 if ((error = change_root(pr->pr_root, td)))
2478 oldcred = crcopysafe(p, newcred);
2479 newcred->cr_prison = pr;
2480 p->p_ucred = newcred;
2484 racct_proc_ucred_changed(p, oldcred, newcred);
2486 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2491 VOP_UNLOCK(pr->pr_root, 0);
2493 /* Tell modules this thread is still in its old jail after all. */
2494 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2495 prison_deref(pr, PD_DEREF | PD_DEUREF);
2501 * Returns a locked prison instance, or NULL on failure.
2504 prison_find(int prid)
2508 sx_assert(&allprison_lock, SX_LOCKED);
2509 TAILQ_FOREACH(pr, &allprison, pr_list) {
2510 if (pr->pr_id == prid) {
2511 mtx_lock(&pr->pr_mtx);
2514 mtx_unlock(&pr->pr_mtx);
2521 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2524 prison_find_child(struct prison *mypr, int prid)
2529 sx_assert(&allprison_lock, SX_LOCKED);
2530 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2531 if (pr->pr_id == prid) {
2532 mtx_lock(&pr->pr_mtx);
2535 mtx_unlock(&pr->pr_mtx);
2542 * Look for the name relative to mypr. Returns a locked prison or NULL.
2545 prison_find_name(struct prison *mypr, const char *name)
2547 struct prison *pr, *deadpr;
2551 sx_assert(&allprison_lock, SX_LOCKED);
2552 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2555 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2556 if (!strcmp(pr->pr_name + mylen, name)) {
2557 mtx_lock(&pr->pr_mtx);
2558 if (pr->pr_ref > 0) {
2559 if (pr->pr_uref > 0)
2563 mtx_unlock(&pr->pr_mtx);
2566 /* There was no valid prison - perhaps there was a dying one. */
2567 if (deadpr != NULL) {
2568 mtx_lock(&deadpr->pr_mtx);
2569 if (deadpr->pr_ref == 0) {
2570 mtx_unlock(&deadpr->pr_mtx);
2578 * See if a prison has the specific flag set.
2581 prison_flag(struct ucred *cred, unsigned flag)
2584 /* This is an atomic read, so no locking is necessary. */
2585 return (cred->cr_prison->pr_flags & flag);
2589 prison_allow(struct ucred *cred, unsigned flag)
2592 /* This is an atomic read, so no locking is necessary. */
2593 return (cred->cr_prison->pr_allow & flag);
2597 * Remove a prison reference. If that was the last reference, remove the
2598 * prison itself - but not in this context in case there are locks held.
2601 prison_free_locked(struct prison *pr)
2605 mtx_assert(&pr->pr_mtx, MA_OWNED);
2607 mtx_unlock(&pr->pr_mtx);
2609 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2613 prison_free(struct prison *pr)
2616 mtx_lock(&pr->pr_mtx);
2617 prison_free_locked(pr);
2621 * Complete a call to either prison_free or prison_proc_free.
2624 prison_complete(void *context, int pending)
2626 struct prison *pr = context;
2628 sx_xlock(&allprison_lock);
2629 mtx_lock(&pr->pr_mtx);
2630 prison_deref(pr, pr->pr_uref
2631 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2632 : PD_LOCKED | PD_LIST_XLOCKED);
2636 * Remove a prison reference (usually). This internal version assumes no
2637 * mutexes are held, except perhaps the prison itself. If there are no more
2638 * references, release and delist the prison. On completion, the prison lock
2639 * and the allprison lock are both unlocked.
2642 prison_deref(struct prison *pr, int flags)
2644 struct prison *ppr, *tpr;
2647 if (!(flags & PD_LOCKED))
2648 mtx_lock(&pr->pr_mtx);
2650 if (flags & PD_DEUREF) {
2651 KASSERT(pr->pr_uref > 0,
2652 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2655 lasturef = pr->pr_uref == 0;
2658 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2661 if (flags & PD_DEREF) {
2662 KASSERT(pr->pr_ref > 0,
2663 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2668 mtx_unlock(&pr->pr_mtx);
2671 * Tell the modules if the last user reference was removed
2672 * (even it sticks around in dying state).
2675 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2676 sx_xlock(&allprison_lock);
2677 flags |= PD_LIST_XLOCKED;
2679 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2680 mtx_lock(&pr->pr_mtx);
2682 mtx_unlock(&pr->pr_mtx);
2685 /* If the prison still has references, nothing else to do. */
2687 if (flags & PD_LIST_SLOCKED)
2688 sx_sunlock(&allprison_lock);
2689 else if (flags & PD_LIST_XLOCKED)
2690 sx_xunlock(&allprison_lock);
2694 if (flags & PD_LIST_SLOCKED) {
2695 if (!sx_try_upgrade(&allprison_lock)) {
2696 sx_sunlock(&allprison_lock);
2697 sx_xlock(&allprison_lock);
2699 } else if (!(flags & PD_LIST_XLOCKED))
2700 sx_xlock(&allprison_lock);
2702 TAILQ_REMOVE(&allprison, pr, pr_list);
2703 LIST_REMOVE(pr, pr_sibling);
2704 ppr = pr->pr_parent;
2705 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2706 tpr->pr_childcount--;
2707 sx_xunlock(&allprison_lock);
2710 if (pr->pr_vnet != ppr->pr_vnet)
2711 vnet_destroy(pr->pr_vnet);
2713 if (pr->pr_root != NULL)
2715 mtx_destroy(&pr->pr_mtx);
2717 free(pr->pr_ip4, M_PRISON);
2720 free(pr->pr_ip6, M_PRISON);
2722 if (pr->pr_cpuset != NULL)
2723 cpuset_rel(pr->pr_cpuset);
2727 prison_racct_detach(pr);
2731 /* Removing a prison frees a reference on its parent. */
2733 mtx_lock(&pr->pr_mtx);
2734 flags = PD_DEREF | PD_DEUREF;
2739 prison_hold_locked(struct prison *pr)
2742 mtx_assert(&pr->pr_mtx, MA_OWNED);
2743 KASSERT(pr->pr_ref > 0,
2744 ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2749 prison_hold(struct prison *pr)
2752 mtx_lock(&pr->pr_mtx);
2753 prison_hold_locked(pr);
2754 mtx_unlock(&pr->pr_mtx);
2758 prison_proc_hold(struct prison *pr)
2761 mtx_lock(&pr->pr_mtx);
2762 KASSERT(pr->pr_uref > 0,
2763 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2765 mtx_unlock(&pr->pr_mtx);
2769 prison_proc_free(struct prison *pr)
2772 mtx_lock(&pr->pr_mtx);
2773 KASSERT(pr->pr_uref > 0,
2774 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2775 if (pr->pr_uref > 1)
2779 * Don't remove the last user reference in this context, which
2780 * is expected to be a process that is not only locked, but
2784 mtx_unlock(&pr->pr_mtx);
2785 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2788 mtx_unlock(&pr->pr_mtx);
2794 * Restrict a prison's IP address list with its parent's, possibly replacing
2795 * it. Return true if the replacement buffer was used (or would have been).
2798 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2803 ppr = pr->pr_parent;
2804 if (!(pr->pr_flags & PR_IP4_USER)) {
2805 /* This has no user settings, so just copy the parent's list. */
2806 if (pr->pr_ip4s < ppr->pr_ip4s) {
2808 * There's no room for the parent's list. Use the
2809 * new list buffer, which is assumed to be big enough
2810 * (if it was passed). If there's no buffer, try to
2814 if (newip4 == NULL) {
2815 newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2816 M_PRISON, M_NOWAIT);
2820 if (newip4 != NULL) {
2821 bcopy(ppr->pr_ip4, newip4,
2822 ppr->pr_ip4s * sizeof(*newip4));
2823 free(pr->pr_ip4, M_PRISON);
2824 pr->pr_ip4 = newip4;
2825 pr->pr_ip4s = ppr->pr_ip4s;
2829 pr->pr_ip4s = ppr->pr_ip4s;
2830 if (pr->pr_ip4s > 0)
2831 bcopy(ppr->pr_ip4, pr->pr_ip4,
2832 pr->pr_ip4s * sizeof(*newip4));
2833 else if (pr->pr_ip4 != NULL) {
2834 free(pr->pr_ip4, M_PRISON);
2837 } else if (pr->pr_ip4s > 0) {
2838 /* Remove addresses that aren't in the parent. */
2839 for (ij = 0; ij < ppr->pr_ip4s; ij++)
2840 if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2842 if (ij < ppr->pr_ip4s)
2845 bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2846 --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2849 for (ij = 1; ii < pr->pr_ip4s; ) {
2850 if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2854 switch (ij >= ppr->pr_ip4s ? -1 :
2855 qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2857 bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2858 (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2869 if (pr->pr_ip4s == 0) {
2870 pr->pr_flags |= PR_IP4_DISABLE;
2871 free(pr->pr_ip4, M_PRISON);
2879 * Pass back primary IPv4 address of this jail.
2881 * If not restricted return success but do not alter the address. Caller has
2882 * to make sure to initialize it correctly (e.g. INADDR_ANY).
2884 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2885 * Address returned in NBO.
2888 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2892 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2893 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2895 pr = cred->cr_prison;
2896 if (!(pr->pr_flags & PR_IP4))
2898 mtx_lock(&pr->pr_mtx);
2899 if (!(pr->pr_flags & PR_IP4)) {
2900 mtx_unlock(&pr->pr_mtx);
2903 if (pr->pr_ip4 == NULL) {
2904 mtx_unlock(&pr->pr_mtx);
2905 return (EAFNOSUPPORT);
2908 ia->s_addr = pr->pr_ip4[0].s_addr;
2909 mtx_unlock(&pr->pr_mtx);
2914 * Return 1 if we should do proper source address selection or are not jailed.
2915 * We will return 0 if we should bypass source address selection in favour
2916 * of the primary jail IPv4 address. Only in this case *ia will be updated and
2918 * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2921 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2927 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2928 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2933 pr = cred->cr_prison;
2934 if (pr->pr_flags & PR_IP4_SADDRSEL)
2937 lia.s_addr = INADDR_ANY;
2938 error = prison_get_ip4(cred, &lia);
2941 if (lia.s_addr == INADDR_ANY)
2944 ia->s_addr = lia.s_addr;
2949 * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2952 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2959 * No need to lock since the PR_IP4_USER flag can't be altered for
2962 while (pr1 != &prison0 &&
2964 !(pr1->pr_flags & PR_VNET) &&
2966 !(pr1->pr_flags & PR_IP4_USER))
2967 pr1 = pr1->pr_parent;
2968 while (pr2 != &prison0 &&
2970 !(pr2->pr_flags & PR_VNET) &&
2972 !(pr2->pr_flags & PR_IP4_USER))
2973 pr2 = pr2->pr_parent;
2974 return (pr1 == pr2);
2978 * Make sure our (source) address is set to something meaningful to this
2981 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2982 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2983 * doesn't allow IPv4. Address passed in in NBO and returned in NBO.
2986 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2992 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2993 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2995 pr = cred->cr_prison;
2996 if (!(pr->pr_flags & PR_IP4))
2998 mtx_lock(&pr->pr_mtx);
2999 if (!(pr->pr_flags & PR_IP4)) {
3000 mtx_unlock(&pr->pr_mtx);
3003 if (pr->pr_ip4 == NULL) {
3004 mtx_unlock(&pr->pr_mtx);
3005 return (EAFNOSUPPORT);
3008 ia0.s_addr = ntohl(ia->s_addr);
3009 if (ia0.s_addr == INADDR_LOOPBACK) {
3010 ia->s_addr = pr->pr_ip4[0].s_addr;
3011 mtx_unlock(&pr->pr_mtx);
3015 if (ia0.s_addr == INADDR_ANY) {
3017 * In case there is only 1 IPv4 address, bind directly.
3019 if (pr->pr_ip4s == 1)
3020 ia->s_addr = pr->pr_ip4[0].s_addr;
3021 mtx_unlock(&pr->pr_mtx);
3025 error = _prison_check_ip4(pr, ia);
3026 mtx_unlock(&pr->pr_mtx);
3031 * Rewrite destination address in case we will connect to loopback address.
3033 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
3034 * Address passed in in NBO and returned in NBO.
3037 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
3041 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3042 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3044 pr = cred->cr_prison;
3045 if (!(pr->pr_flags & PR_IP4))
3047 mtx_lock(&pr->pr_mtx);
3048 if (!(pr->pr_flags & PR_IP4)) {
3049 mtx_unlock(&pr->pr_mtx);
3052 if (pr->pr_ip4 == NULL) {
3053 mtx_unlock(&pr->pr_mtx);
3054 return (EAFNOSUPPORT);
3057 if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
3058 ia->s_addr = pr->pr_ip4[0].s_addr;
3059 mtx_unlock(&pr->pr_mtx);
3064 * Return success because nothing had to be changed.
3066 mtx_unlock(&pr->pr_mtx);
3071 * Check if given address belongs to the jail referenced by cred/prison.
3073 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
3074 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3075 * doesn't allow IPv4. Address passed in in NBO.
3078 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
3083 * Check the primary IP.
3085 if (pr->pr_ip4[0].s_addr == ia->s_addr)
3089 * All the other IPs are sorted so we can do a binary search.
3092 z = pr->pr_ip4s - 2;
3095 d = qcmp_v4(&pr->pr_ip4[i+1], ia);
3104 return (EADDRNOTAVAIL);
3108 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
3113 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3114 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3116 pr = cred->cr_prison;
3117 if (!(pr->pr_flags & PR_IP4))
3119 mtx_lock(&pr->pr_mtx);
3120 if (!(pr->pr_flags & PR_IP4)) {
3121 mtx_unlock(&pr->pr_mtx);
3124 if (pr->pr_ip4 == NULL) {
3125 mtx_unlock(&pr->pr_mtx);
3126 return (EAFNOSUPPORT);
3129 error = _prison_check_ip4(pr, ia);
3130 mtx_unlock(&pr->pr_mtx);
3137 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
3142 ppr = pr->pr_parent;
3143 if (!(pr->pr_flags & PR_IP6_USER)) {
3144 /* This has no user settings, so just copy the parent's list. */
3145 if (pr->pr_ip6s < ppr->pr_ip6s) {
3147 * There's no room for the parent's list. Use the
3148 * new list buffer, which is assumed to be big enough
3149 * (if it was passed). If there's no buffer, try to
3153 if (newip6 == NULL) {
3154 newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3155 M_PRISON, M_NOWAIT);
3159 if (newip6 != NULL) {
3160 bcopy(ppr->pr_ip6, newip6,
3161 ppr->pr_ip6s * sizeof(*newip6));
3162 free(pr->pr_ip6, M_PRISON);
3163 pr->pr_ip6 = newip6;
3164 pr->pr_ip6s = ppr->pr_ip6s;
3168 pr->pr_ip6s = ppr->pr_ip6s;
3169 if (pr->pr_ip6s > 0)
3170 bcopy(ppr->pr_ip6, pr->pr_ip6,
3171 pr->pr_ip6s * sizeof(*newip6));
3172 else if (pr->pr_ip6 != NULL) {
3173 free(pr->pr_ip6, M_PRISON);
3176 } else if (pr->pr_ip6s > 0) {
3177 /* Remove addresses that aren't in the parent. */
3178 for (ij = 0; ij < ppr->pr_ip6s; ij++)
3179 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3182 if (ij < ppr->pr_ip6s)
3185 bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3186 --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3189 for (ij = 1; ii < pr->pr_ip6s; ) {
3190 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3195 switch (ij >= ppr->pr_ip6s ? -1 :
3196 qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3198 bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3199 (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3210 if (pr->pr_ip6s == 0) {
3211 pr->pr_flags |= PR_IP6_DISABLE;
3212 free(pr->pr_ip6, M_PRISON);
3220 * Pass back primary IPv6 address for this jail.
3222 * If not restricted return success but do not alter the address. Caller has
3223 * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3225 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3228 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3232 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3233 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3235 pr = cred->cr_prison;
3236 if (!(pr->pr_flags & PR_IP6))
3238 mtx_lock(&pr->pr_mtx);
3239 if (!(pr->pr_flags & PR_IP6)) {
3240 mtx_unlock(&pr->pr_mtx);
3243 if (pr->pr_ip6 == NULL) {
3244 mtx_unlock(&pr->pr_mtx);
3245 return (EAFNOSUPPORT);
3248 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3249 mtx_unlock(&pr->pr_mtx);
3254 * Return 1 if we should do proper source address selection or are not jailed.
3255 * We will return 0 if we should bypass source address selection in favour
3256 * of the primary jail IPv6 address. Only in this case *ia will be updated and
3258 * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3261 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3264 struct in6_addr lia6;
3267 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3268 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3273 pr = cred->cr_prison;
3274 if (pr->pr_flags & PR_IP6_SADDRSEL)
3278 error = prison_get_ip6(cred, &lia6);
3281 if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3284 bcopy(&lia6, ia6, sizeof(struct in6_addr));
3289 * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3292 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3298 while (pr1 != &prison0 &&
3300 !(pr1->pr_flags & PR_VNET) &&
3302 !(pr1->pr_flags & PR_IP6_USER))
3303 pr1 = pr1->pr_parent;
3304 while (pr2 != &prison0 &&
3306 !(pr2->pr_flags & PR_VNET) &&
3308 !(pr2->pr_flags & PR_IP6_USER))
3309 pr2 = pr2->pr_parent;
3310 return (pr1 == pr2);
3314 * Make sure our (source) address is set to something meaningful to this jail.
3316 * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3317 * when needed while binding.
3319 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3320 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3321 * doesn't allow IPv6.
3324 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3329 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3330 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3332 pr = cred->cr_prison;
3333 if (!(pr->pr_flags & PR_IP6))
3335 mtx_lock(&pr->pr_mtx);
3336 if (!(pr->pr_flags & PR_IP6)) {
3337 mtx_unlock(&pr->pr_mtx);
3340 if (pr->pr_ip6 == NULL) {
3341 mtx_unlock(&pr->pr_mtx);
3342 return (EAFNOSUPPORT);
3345 if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3346 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3347 mtx_unlock(&pr->pr_mtx);
3351 if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3353 * In case there is only 1 IPv6 address, and v6only is true,
3354 * then bind directly.
3356 if (v6only != 0 && pr->pr_ip6s == 1)
3357 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3358 mtx_unlock(&pr->pr_mtx);
3362 error = _prison_check_ip6(pr, ia6);
3363 mtx_unlock(&pr->pr_mtx);
3368 * Rewrite destination address in case we will connect to loopback address.
3370 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3373 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3377 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3378 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3380 pr = cred->cr_prison;
3381 if (!(pr->pr_flags & PR_IP6))
3383 mtx_lock(&pr->pr_mtx);
3384 if (!(pr->pr_flags & PR_IP6)) {
3385 mtx_unlock(&pr->pr_mtx);
3388 if (pr->pr_ip6 == NULL) {
3389 mtx_unlock(&pr->pr_mtx);
3390 return (EAFNOSUPPORT);
3393 if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3394 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3395 mtx_unlock(&pr->pr_mtx);
3400 * Return success because nothing had to be changed.
3402 mtx_unlock(&pr->pr_mtx);
3407 * Check if given address belongs to the jail referenced by cred/prison.
3409 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3410 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3411 * doesn't allow IPv6.
3414 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3419 * Check the primary IP.
3421 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3425 * All the other IPs are sorted so we can do a binary search.
3428 z = pr->pr_ip6s - 2;
3431 d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3440 return (EADDRNOTAVAIL);
3444 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3449 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3450 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3452 pr = cred->cr_prison;
3453 if (!(pr->pr_flags & PR_IP6))
3455 mtx_lock(&pr->pr_mtx);
3456 if (!(pr->pr_flags & PR_IP6)) {
3457 mtx_unlock(&pr->pr_mtx);
3460 if (pr->pr_ip6 == NULL) {
3461 mtx_unlock(&pr->pr_mtx);
3462 return (EAFNOSUPPORT);
3465 error = _prison_check_ip6(pr, ia6);
3466 mtx_unlock(&pr->pr_mtx);
3472 * Check if a jail supports the given address family.
3474 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3478 prison_check_af(struct ucred *cred, int af)
3483 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3485 pr = cred->cr_prison;
3487 /* Prisons with their own network stack are not limited. */
3488 if (prison_owns_vnet(cred))
3497 if (pr->pr_flags & PR_IP4)
3499 mtx_lock(&pr->pr_mtx);
3500 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3501 error = EAFNOSUPPORT;
3502 mtx_unlock(&pr->pr_mtx);
3508 if (pr->pr_flags & PR_IP6)
3510 mtx_lock(&pr->pr_mtx);
3511 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3512 error = EAFNOSUPPORT;
3513 mtx_unlock(&pr->pr_mtx);
3521 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3522 error = EAFNOSUPPORT;
3528 * Check if given address belongs to the jail referenced by cred (wrapper to
3529 * prison_check_ip[46]).
3531 * Returns 0 if jail doesn't restrict the address family or if address belongs
3532 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3533 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3536 prison_if(struct ucred *cred, struct sockaddr *sa)
3539 struct sockaddr_in *sai;
3542 struct sockaddr_in6 *sai6;
3546 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3547 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3550 if (prison_owns_vnet(cred))
3555 switch (sa->sa_family)
3559 sai = (struct sockaddr_in *)sa;
3560 error = prison_check_ip4(cred, &sai->sin_addr);
3565 sai6 = (struct sockaddr_in6 *)sa;
3566 error = prison_check_ip6(cred, &sai6->sin6_addr);
3570 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3571 error = EAFNOSUPPORT;
3577 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3580 prison_check(struct ucred *cred1, struct ucred *cred2)
3583 return ((cred1->cr_prison == cred2->cr_prison ||
3584 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3588 * Return 1 if p2 is a child of p1, otherwise 0.
3591 prison_ischild(struct prison *pr1, struct prison *pr2)
3594 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3601 * Return 1 if the passed credential is in a jail, otherwise 0.
3604 jailed(struct ucred *cred)
3607 return (cred->cr_prison != &prison0);
3611 * Return 1 if the passed credential is in a jail and that jail does not
3612 * have its own virtual network stack, otherwise 0.
3615 jailed_without_vnet(struct ucred *cred)
3621 if (prison_owns_vnet(cred))
3629 * Return the correct hostname (domainname, et al) for the passed credential.
3632 getcredhostname(struct ucred *cred, char *buf, size_t size)
3637 * A NULL credential can be used to shortcut to the physical
3638 * system's hostname.
3640 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3641 mtx_lock(&pr->pr_mtx);
3642 strlcpy(buf, pr->pr_hostname, size);
3643 mtx_unlock(&pr->pr_mtx);
3647 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3650 mtx_lock(&cred->cr_prison->pr_mtx);
3651 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3652 mtx_unlock(&cred->cr_prison->pr_mtx);
3656 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3659 mtx_lock(&cred->cr_prison->pr_mtx);
3660 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3661 mtx_unlock(&cred->cr_prison->pr_mtx);
3665 getcredhostid(struct ucred *cred, unsigned long *hostid)
3668 mtx_lock(&cred->cr_prison->pr_mtx);
3669 *hostid = cred->cr_prison->pr_hostid;
3670 mtx_unlock(&cred->cr_prison->pr_mtx);
3675 * Determine whether the prison represented by cred owns
3676 * its vnet rather than having it inherited.
3678 * Returns 1 in case the prison owns the vnet, 0 otherwise.
3681 prison_owns_vnet(struct ucred *cred)
3685 * vnets cannot be added/removed after jail creation,
3686 * so no need to lock here.
3688 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3693 * Determine whether the subject represented by cred can "see"
3694 * status of a mount point.
3695 * Returns: 0 for permitted, ENOENT otherwise.
3696 * XXX: This function should be called cr_canseemount() and should be
3697 * placed in kern_prot.c.
3700 prison_canseemount(struct ucred *cred, struct mount *mp)
3706 pr = cred->cr_prison;
3707 if (pr->pr_enforce_statfs == 0)
3709 if (pr->pr_root->v_mount == mp)
3711 if (pr->pr_enforce_statfs == 2)
3714 * If jail's chroot directory is set to "/" we should be able to see
3715 * all mount-points from inside a jail.
3716 * This is ugly check, but this is the only situation when jail's
3717 * directory ends with '/'.
3719 if (strcmp(pr->pr_path, "/") == 0)
3721 len = strlen(pr->pr_path);
3723 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3726 * Be sure that we don't have situation where jail's root directory
3727 * is "/some/path" and mount point is "/some/pathpath".
3729 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3735 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3737 char jpath[MAXPATHLEN];
3741 pr = cred->cr_prison;
3742 if (pr->pr_enforce_statfs == 0)
3744 if (prison_canseemount(cred, mp) != 0) {
3745 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3746 strlcpy(sp->f_mntonname, "[restricted]",
3747 sizeof(sp->f_mntonname));
3750 if (pr->pr_root->v_mount == mp) {
3752 * Clear current buffer data, so we are sure nothing from
3753 * the valid path left there.
3755 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3756 *sp->f_mntonname = '/';
3760 * If jail's chroot directory is set to "/" we should be able to see
3761 * all mount-points from inside a jail.
3763 if (strcmp(pr->pr_path, "/") == 0)
3765 len = strlen(pr->pr_path);
3766 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3768 * Clear current buffer data, so we are sure nothing from
3769 * the valid path left there.
3771 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3772 if (*jpath == '\0') {
3773 /* Should never happen. */
3774 *sp->f_mntonname = '/';
3776 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3781 * Check with permission for a specific privilege is granted within jail. We
3782 * have a specific list of accepted privileges; the rest are denied.
3785 prison_priv_check(struct ucred *cred, int priv)
3793 * Privileges specific to prisons with a virtual network stack.
3794 * There might be a duplicate entry here in case the privilege
3795 * is only granted conditionally in the legacy jail case.
3800 * NFS-specific privileges.
3802 case PRIV_NFS_DAEMON:
3803 case PRIV_NFS_LOCKD:
3806 * Network stack privileges.
3808 case PRIV_NET_BRIDGE:
3811 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3812 case PRIV_NET_ROUTE:
3814 case PRIV_NET_SETIFMTU:
3815 case PRIV_NET_SETIFFLAGS:
3816 case PRIV_NET_SETIFCAP:
3817 case PRIV_NET_SETIFDESCR:
3818 case PRIV_NET_SETIFNAME :
3819 case PRIV_NET_SETIFMETRIC:
3820 case PRIV_NET_SETIFPHYS:
3821 case PRIV_NET_SETIFMAC:
3822 case PRIV_NET_ADDMULTI:
3823 case PRIV_NET_DELMULTI:
3824 case PRIV_NET_HWIOCTL:
3825 case PRIV_NET_SETLLADDR:
3826 case PRIV_NET_ADDIFGROUP:
3827 case PRIV_NET_DELIFGROUP:
3828 case PRIV_NET_IFCREATE:
3829 case PRIV_NET_IFDESTROY:
3830 case PRIV_NET_ADDIFADDR:
3831 case PRIV_NET_DELIFADDR:
3834 case PRIV_NET_SETIFVNET:
3835 case PRIV_NET_SETIFFIB:
3838 * 802.11-related privileges.
3840 case PRIV_NET80211_GETKEY:
3842 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3847 * AppleTalk privileges.
3849 case PRIV_NETATALK_RESERVEDPORT:
3854 case PRIV_NETATM_CFG:
3855 case PRIV_NETATM_ADD:
3856 case PRIV_NETATM_DEL:
3857 case PRIV_NETATM_SET:
3860 * Bluetooth privileges.
3862 case PRIV_NETBLUETOOTH_RAW:
3866 * Netgraph and netgraph module privileges.
3868 case PRIV_NETGRAPH_CONTROL:
3870 case PRIV_NETGRAPH_TTY:
3874 * IPv4 and IPv6 privileges.
3876 case PRIV_NETINET_IPFW:
3877 case PRIV_NETINET_DIVERT:
3878 case PRIV_NETINET_PF:
3879 case PRIV_NETINET_DUMMYNET:
3880 case PRIV_NETINET_CARP:
3881 case PRIV_NETINET_MROUTE:
3882 case PRIV_NETINET_RAW:
3883 case PRIV_NETINET_ADDRCTRL6:
3884 case PRIV_NETINET_ND6:
3885 case PRIV_NETINET_SCOPE6:
3886 case PRIV_NETINET_ALIFETIME6:
3887 case PRIV_NETINET_IPSEC:
3888 case PRIV_NETINET_BINDANY:
3892 * IPX/SPX privileges.
3894 case PRIV_NETIPX_RESERVEDPORT:
3895 case PRIV_NETIPX_RAW:
3909 * No default: or deny here.
3910 * In case of no permit fall through to next switch().
3912 if (cred->cr_prison->pr_flags & PR_VNET)
3920 * Allow ktrace privileges for root in jail.
3926 * Allow jailed processes to configure audit identity and
3927 * submit audit records (login, etc). In the future we may
3928 * want to further refine the relationship between audit and
3931 case PRIV_AUDIT_GETAUDIT:
3932 case PRIV_AUDIT_SETAUDIT:
3933 case PRIV_AUDIT_SUBMIT:
3937 * Allow jailed processes to manipulate process UNIX
3938 * credentials in any way they see fit.
3940 case PRIV_CRED_SETUID:
3941 case PRIV_CRED_SETEUID:
3942 case PRIV_CRED_SETGID:
3943 case PRIV_CRED_SETEGID:
3944 case PRIV_CRED_SETGROUPS:
3945 case PRIV_CRED_SETREUID:
3946 case PRIV_CRED_SETREGID:
3947 case PRIV_CRED_SETRESUID:
3948 case PRIV_CRED_SETRESGID:
3951 * Jail implements visibility constraints already, so allow
3952 * jailed root to override uid/gid-based constraints.
3954 case PRIV_SEEOTHERGIDS:
3955 case PRIV_SEEOTHERUIDS:
3958 * Jail implements inter-process debugging limits already, so
3959 * allow jailed root various debugging privileges.
3961 case PRIV_DEBUG_DIFFCRED:
3962 case PRIV_DEBUG_SUGID:
3963 case PRIV_DEBUG_UNPRIV:
3966 * Allow jail to set various resource limits and login
3967 * properties, and for now, exceed process resource limits.
3969 case PRIV_PROC_LIMIT:
3970 case PRIV_PROC_SETLOGIN:
3971 case PRIV_PROC_SETRLIMIT:
3974 * System V and POSIX IPC privileges are granted in jail.
3977 case PRIV_IPC_WRITE:
3978 case PRIV_IPC_ADMIN:
3979 case PRIV_IPC_MSGSIZE:
3983 * Jail operations within a jail work on child jails.
3985 case PRIV_JAIL_ATTACH:
3987 case PRIV_JAIL_REMOVE:
3990 * Jail implements its own inter-process limits, so allow
3991 * root processes in jail to change scheduling on other
3992 * processes in the same jail. Likewise for signalling.
3994 case PRIV_SCHED_DIFFCRED:
3995 case PRIV_SCHED_CPUSET:
3996 case PRIV_SIGNAL_DIFFCRED:
3997 case PRIV_SIGNAL_SUGID:
4000 * Allow jailed processes to write to sysctls marked as jail
4003 case PRIV_SYSCTL_WRITEJAIL:
4006 * Allow root in jail to manage a variety of quota
4007 * properties. These should likely be conditional on a
4008 * configuration option.
4010 case PRIV_VFS_GETQUOTA:
4011 case PRIV_VFS_SETQUOTA:
4014 * Since Jail relies on chroot() to implement file system
4015 * protections, grant many VFS privileges to root in jail.
4016 * Be careful to exclude mount-related and NFS-related
4020 case PRIV_VFS_WRITE:
4021 case PRIV_VFS_ADMIN:
4023 case PRIV_VFS_LOOKUP:
4024 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4025 case PRIV_VFS_CHFLAGS_DEV:
4026 case PRIV_VFS_CHOWN:
4027 case PRIV_VFS_CHROOT:
4028 case PRIV_VFS_RETAINSUGID:
4029 case PRIV_VFS_FCHROOT:
4031 case PRIV_VFS_SETGID:
4033 case PRIV_VFS_STICKYFILE:
4036 * As in the non-jail case, non-root users are expected to be
4037 * able to read kernel/phyiscal memory (provided /dev/[k]mem
4038 * exists in the jail and they have permission to access it).
4040 case PRIV_KMEM_READ:
4044 * Depending on the global setting, allow privilege of
4045 * setting system flags.
4047 case PRIV_VFS_SYSFLAGS:
4048 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4054 * Depending on the global setting, allow privilege of
4055 * mounting/unmounting file systems.
4057 case PRIV_VFS_MOUNT:
4058 case PRIV_VFS_UNMOUNT:
4059 case PRIV_VFS_MOUNT_NONUSER:
4060 case PRIV_VFS_MOUNT_OWNER:
4061 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
4062 cred->cr_prison->pr_enforce_statfs < 2)
4068 * Allow jailed root to bind reserved ports and reuse in-use
4071 case PRIV_NETINET_RESERVEDPORT:
4072 case PRIV_NETINET_REUSEPORT:
4076 * Allow jailed root to set certian IPv4/6 (option) headers.
4078 case PRIV_NETINET_SETHDROPTS:
4082 * Conditionally allow creating raw sockets in jail.
4084 case PRIV_NETINET_RAW:
4085 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4091 * Since jail implements its own visibility limits on netstat
4092 * sysctls, allow getcred. This allows identd to work in
4095 case PRIV_NETINET_GETCRED:
4099 * Allow jailed root to set loginclass.
4101 case PRIV_PROC_SETLOGINCLASS:
4106 * In all remaining cases, deny the privilege request. This
4107 * includes almost all network privileges, many system
4108 * configuration privileges.
4115 * Return the part of pr2's name that is relative to pr1, or the whole name
4116 * if it does not directly follow.
4120 prison_name(struct prison *pr1, struct prison *pr2)
4124 /* Jails see themselves as "0" (if they see themselves at all). */
4127 name = pr2->pr_name;
4128 if (prison_ischild(pr1, pr2)) {
4130 * pr1 isn't locked (and allprison_lock may not be either)
4131 * so its length can't be counted on. But the number of dots
4132 * can be counted on - and counted.
4134 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4135 name = strchr(name, '.') + 1;
4141 * Return the part of pr2's path that is relative to pr1, or the whole path
4142 * if it does not directly follow.
4145 prison_path(struct prison *pr1, struct prison *pr2)
4147 char *path1, *path2;
4150 path1 = pr1->pr_path;
4151 path2 = pr2->pr_path;
4152 if (!strcmp(path1, "/"))
4154 len1 = strlen(path1);
4155 if (strncmp(path1, path2, len1))
4157 if (path2[len1] == '\0')
4159 if (path2[len1] == '/')
4160 return (path2 + len1);
4166 * Jail-related sysctls.
4168 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4172 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4175 struct prison *pr, *cpr;
4177 struct in_addr *ip4 = NULL;
4181 struct in6_addr *ip6 = NULL;
4186 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4187 pr = req->td->td_ucred->cr_prison;
4189 sx_slock(&allprison_lock);
4190 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4191 #if defined(INET) || defined(INET6)
4194 mtx_lock(&cpr->pr_mtx);
4196 if (cpr->pr_ip4s > 0) {
4197 if (ip4s < cpr->pr_ip4s) {
4198 ip4s = cpr->pr_ip4s;
4199 mtx_unlock(&cpr->pr_mtx);
4200 ip4 = realloc(ip4, ip4s *
4201 sizeof(struct in_addr), M_TEMP, M_WAITOK);
4204 bcopy(cpr->pr_ip4, ip4,
4205 cpr->pr_ip4s * sizeof(struct in_addr));
4209 if (cpr->pr_ip6s > 0) {
4210 if (ip6s < cpr->pr_ip6s) {
4211 ip6s = cpr->pr_ip6s;
4212 mtx_unlock(&cpr->pr_mtx);
4213 ip6 = realloc(ip6, ip6s *
4214 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4217 bcopy(cpr->pr_ip6, ip6,
4218 cpr->pr_ip6s * sizeof(struct in6_addr));
4221 if (cpr->pr_ref == 0) {
4222 mtx_unlock(&cpr->pr_mtx);
4225 bzero(xp, sizeof(*xp));
4226 xp->pr_version = XPRISON_VERSION;
4227 xp->pr_id = cpr->pr_id;
4228 xp->pr_state = cpr->pr_uref > 0
4229 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4230 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4231 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4232 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4234 xp->pr_ip4s = cpr->pr_ip4s;
4237 xp->pr_ip6s = cpr->pr_ip6s;
4239 mtx_unlock(&cpr->pr_mtx);
4240 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4244 if (xp->pr_ip4s > 0) {
4245 error = SYSCTL_OUT(req, ip4,
4246 xp->pr_ip4s * sizeof(struct in_addr));
4252 if (xp->pr_ip6s > 0) {
4253 error = SYSCTL_OUT(req, ip6,
4254 xp->pr_ip6s * sizeof(struct in6_addr));
4260 sx_sunlock(&allprison_lock);
4271 SYSCTL_OID(_security_jail, OID_AUTO, list,
4272 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4273 sysctl_jail_list, "S", "List of active jails");
4276 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4280 injail = jailed(req->td->td_ucred);
4281 error = SYSCTL_OUT(req, &injail, sizeof(injail));
4286 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4287 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4288 sysctl_jail_jailed, "I", "Process in jail?");
4291 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4293 int error, havevnet;
4295 struct ucred *cred = req->td->td_ucred;
4297 havevnet = jailed(cred) && prison_owns_vnet(cred);
4301 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4306 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4307 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4308 sysctl_jail_vnet, "I", "Jail owns VNET?");
4310 #if defined(INET) || defined(INET6)
4311 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4312 &jail_max_af_ips, 0,
4313 "Number of IP addresses a jail may have at most per address family (deprecated)");
4317 * Default parameters for jail(2) compatability. For historical reasons,
4318 * the sysctl names have varying similarity to the parameter names. Prisons
4319 * just see their own parameters, and can't change them.
4322 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4325 int allow, error, i;
4327 pr = req->td->td_ucred->cr_prison;
4328 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4330 /* Get the current flag value, and convert it to a boolean. */
4331 i = (allow & arg2) ? 1 : 0;
4334 error = sysctl_handle_int(oidp, &i, 0, req);
4335 if (error || !req->newptr)
4341 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4344 mtx_lock(&prison0.pr_mtx);
4345 jail_default_allow = (jail_default_allow & ~arg2) | i;
4346 mtx_unlock(&prison0.pr_mtx);
4350 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4351 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4352 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4353 "Processes in jail can set their hostnames (deprecated)");
4354 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4355 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4356 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4357 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4358 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4359 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4360 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4361 "Processes in jail can use System V IPC primitives (deprecated)");
4362 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4363 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4364 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4365 "Prison root can create raw sockets (deprecated)");
4366 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4367 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4368 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4369 "Processes in jail can alter system file flags (deprecated)");
4370 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4371 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4372 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4373 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4374 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4375 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4376 NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4377 "Processes in jail can mount the devfs file system (deprecated)");
4378 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4379 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4380 NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4381 "Processes in jail can mount the fdescfs file system (deprecated)");
4382 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4383 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4384 NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4385 "Processes in jail can mount the nullfs file system (deprecated)");
4386 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4387 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4388 NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4389 "Processes in jail can mount the procfs file system (deprecated)");
4390 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
4391 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4392 NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
4393 "Processes in jail can mount the linprocfs file system (deprecated)");
4394 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
4395 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4396 NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
4397 "Processes in jail can mount the linsysfs file system (deprecated)");
4398 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4399 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4400 NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4401 "Processes in jail can mount the tmpfs file system (deprecated)");
4402 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4403 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4404 NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4405 "Processes in jail can mount the zfs file system (deprecated)");
4408 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4413 pr = req->td->td_ucred->cr_prison;
4414 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4415 error = sysctl_handle_int(oidp, &level, 0, req);
4416 if (error || !req->newptr)
4418 *(int *)arg1 = level;
4422 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4423 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4424 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4425 sysctl_jail_default_level, "I",
4426 "Processes in jail cannot see all mounted file systems (deprecated)");
4428 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4429 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4430 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4431 sysctl_jail_default_level, "I",
4432 "Ruleset for the devfs filesystem in jail (deprecated)");
4435 * Nodes to describe jail parameters. Maximum length of string parameters
4436 * is returned in the string itself, and the other parameters exist merely
4437 * to make themselves and their types known.
4439 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4443 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4450 switch (oidp->oid_kind & CTLTYPE)
4456 if (!(req->flags & SCTL_MASK32))
4458 return (SYSCTL_OUT(req, &l, sizeof(l)));
4462 return (SYSCTL_OUT(req, &i, sizeof(i)));
4463 case CTLTYPE_STRING:
4464 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4466 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4467 case CTLTYPE_STRUCT:
4469 return (SYSCTL_OUT(req, &s, sizeof(s)));
4475 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4476 * jail creation time but cannot be changed in an existing jail.
4478 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4479 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4480 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4481 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4482 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4483 "I", "Jail secure level");
4484 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4485 "Jail value for kern.osreldate and uname -K");
4486 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4487 "Jail value for kern.osrelease and uname -r");
4488 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4489 "I", "Jail cannot see all mounted file systems");
4490 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4491 "I", "Ruleset for in-jail devfs mounts");
4492 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4493 "B", "Jail persistence");
4495 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4496 "E,jailsys", "Virtual network stack");
4498 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4499 "B", "Jail is in the process of shutting down");
4501 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4502 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4503 "I", "Current number of child jails");
4504 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4505 "I", "Maximum number of child jails");
4507 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4508 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4510 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4511 "Jail NIS domainname");
4512 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4514 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4515 "LU", "Jail host ID");
4517 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4518 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4521 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4522 "Jail IPv4 address virtualization");
4523 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4524 "S,in_addr,a", "Jail IPv4 addresses");
4525 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4526 "B", "Do (not) use IPv4 source address selection rather than the "
4527 "primary jail IPv4 address.");
4530 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4531 "Jail IPv6 address virtualization");
4532 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4533 "S,in6_addr,a", "Jail IPv6 addresses");
4534 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4535 "B", "Do (not) use IPv6 source address selection rather than the "
4536 "primary jail IPv6 address.");
4539 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4540 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4541 "B", "Jail may set hostname");
4542 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4543 "B", "Jail may use SYSV IPC");
4544 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4545 "B", "Jail may create raw sockets");
4546 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4547 "B", "Jail may alter system file flags");
4548 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4549 "B", "Jail may set file quotas");
4550 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4551 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4553 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4554 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4555 "B", "Jail may mount/unmount jail-friendly file systems in general");
4556 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4557 "B", "Jail may mount the devfs file system");
4558 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4559 "B", "Jail may mount the fdescfs file system");
4560 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4561 "B", "Jail may mount the nullfs file system");
4562 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4563 "B", "Jail may mount the procfs file system");
4564 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
4565 "B", "Jail may mount the linprocfs file system");
4566 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
4567 "B", "Jail may mount the linsysfs file system");
4568 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4569 "B", "Jail may mount the tmpfs file system");
4570 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4571 "B", "Jail may mount the zfs file system");
4575 prison_racct_foreach(void (*callback)(struct racct *racct,
4576 void *arg2, void *arg3), void *arg2, void *arg3)
4578 struct prison_racct *prr;
4580 ASSERT_RACCT_ENABLED();
4582 sx_slock(&allprison_lock);
4583 LIST_FOREACH(prr, &allprison_racct, prr_next)
4584 (callback)(prr->prr_racct, arg2, arg3);
4585 sx_sunlock(&allprison_lock);
4588 static struct prison_racct *
4589 prison_racct_find_locked(const char *name)
4591 struct prison_racct *prr;
4593 ASSERT_RACCT_ENABLED();
4594 sx_assert(&allprison_lock, SA_XLOCKED);
4596 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4599 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4600 if (strcmp(name, prr->prr_name) != 0)
4603 /* Found prison_racct with a matching name? */
4604 prison_racct_hold(prr);
4608 /* Add new prison_racct. */
4609 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4610 racct_create(&prr->prr_racct);
4612 strcpy(prr->prr_name, name);
4613 refcount_init(&prr->prr_refcount, 1);
4614 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4619 struct prison_racct *
4620 prison_racct_find(const char *name)
4622 struct prison_racct *prr;
4624 ASSERT_RACCT_ENABLED();
4626 sx_xlock(&allprison_lock);
4627 prr = prison_racct_find_locked(name);
4628 sx_xunlock(&allprison_lock);
4633 prison_racct_hold(struct prison_racct *prr)
4636 ASSERT_RACCT_ENABLED();
4638 refcount_acquire(&prr->prr_refcount);
4642 prison_racct_free_locked(struct prison_racct *prr)
4645 ASSERT_RACCT_ENABLED();
4646 sx_assert(&allprison_lock, SA_XLOCKED);
4648 if (refcount_release(&prr->prr_refcount)) {
4649 racct_destroy(&prr->prr_racct);
4650 LIST_REMOVE(prr, prr_next);
4651 free(prr, M_PRISON_RACCT);
4656 prison_racct_free(struct prison_racct *prr)
4660 ASSERT_RACCT_ENABLED();
4661 sx_assert(&allprison_lock, SA_UNLOCKED);
4663 old = prr->prr_refcount;
4664 if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4667 sx_xlock(&allprison_lock);
4668 prison_racct_free_locked(prr);
4669 sx_xunlock(&allprison_lock);
4673 prison_racct_attach(struct prison *pr)
4675 struct prison_racct *prr;
4677 ASSERT_RACCT_ENABLED();
4678 sx_assert(&allprison_lock, SA_XLOCKED);
4680 prr = prison_racct_find_locked(pr->pr_name);
4681 KASSERT(prr != NULL, ("cannot find prison_racct"));
4683 pr->pr_prison_racct = prr;
4687 * Handle jail renaming. From the racct point of view, renaming means
4688 * moving from one prison_racct to another.
4691 prison_racct_modify(struct prison *pr)
4695 struct prison_racct *oldprr;
4697 ASSERT_RACCT_ENABLED();
4699 sx_slock(&allproc_lock);
4700 sx_xlock(&allprison_lock);
4702 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4703 sx_xunlock(&allprison_lock);
4704 sx_sunlock(&allproc_lock);
4708 oldprr = pr->pr_prison_racct;
4709 pr->pr_prison_racct = NULL;
4711 prison_racct_attach(pr);
4714 * Move resource utilisation records.
4716 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4719 * Force rctl to reattach rules to processes.
4721 FOREACH_PROC_IN_SYSTEM(p) {
4723 cred = crhold(p->p_ucred);
4725 racct_proc_ucred_changed(p, cred, cred);
4729 sx_sunlock(&allproc_lock);
4730 prison_racct_free_locked(oldprr);
4731 sx_xunlock(&allprison_lock);
4735 prison_racct_detach(struct prison *pr)
4738 ASSERT_RACCT_ENABLED();
4739 sx_assert(&allprison_lock, SA_UNLOCKED);
4741 if (pr->pr_prison_racct == NULL)
4743 prison_racct_free(pr->pr_prison_racct);
4744 pr->pr_prison_racct = NULL;
4751 db_show_prison(struct prison *pr)
4754 #if defined(INET) || defined(INET6)
4759 char ip6buf[INET6_ADDRSTRLEN];
4762 db_printf("prison %p:\n", pr);
4763 db_printf(" jid = %d\n", pr->pr_id);
4764 db_printf(" name = %s\n", pr->pr_name);
4765 db_printf(" parent = %p\n", pr->pr_parent);
4766 db_printf(" ref = %d\n", pr->pr_ref);
4767 db_printf(" uref = %d\n", pr->pr_uref);
4768 db_printf(" path = %s\n", pr->pr_path);
4769 db_printf(" cpuset = %d\n", pr->pr_cpuset
4770 ? pr->pr_cpuset->cs_id : -1);
4772 db_printf(" vnet = %p\n", pr->pr_vnet);
4774 db_printf(" root = %p\n", pr->pr_root);
4775 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4776 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4777 db_printf(" children.max = %d\n", pr->pr_childmax);
4778 db_printf(" children.cur = %d\n", pr->pr_childcount);
4779 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4780 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4781 db_printf(" flags = 0x%x", pr->pr_flags);
4782 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4784 if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4785 db_printf(" %s", pr_flag_names[fi]);
4786 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4788 jsf = pr->pr_flags &
4789 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4790 db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4791 pr_flag_jailsys[fi].disable &&
4792 (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4793 : (jsf == pr_flag_jailsys[fi].new) ? "new"
4796 db_printf(" allow = 0x%x", pr->pr_allow);
4797 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4799 if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4800 db_printf(" %s", pr_allow_names[fi]);
4802 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4803 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4804 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4805 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4806 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4808 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4809 for (ii = 0; ii < pr->pr_ip4s; ii++)
4810 db_printf(" %s %s\n",
4811 ii == 0 ? "ip4.addr =" : " ",
4812 inet_ntoa(pr->pr_ip4[ii]));
4815 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4816 for (ii = 0; ii < pr->pr_ip6s; ii++)
4817 db_printf(" %s %s\n",
4818 ii == 0 ? "ip6.addr =" : " ",
4819 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4823 DB_SHOW_COMMAND(prison, db_show_prison_command)
4829 * Show all prisons in the list, and prison0 which is not
4832 db_show_prison(&prison0);
4833 if (!db_pager_quit) {
4834 TAILQ_FOREACH(pr, &allprison, pr_list) {
4846 /* Look for a prison with the ID and with references. */
4847 TAILQ_FOREACH(pr, &allprison, pr_list)
4848 if (pr->pr_id == addr && pr->pr_ref > 0)
4851 /* Look again, without requiring a reference. */
4852 TAILQ_FOREACH(pr, &allprison, pr_list)
4853 if (pr->pr_id == addr)
4856 /* Assume address points to a valid prison. */
4857 pr = (struct prison *)addr;