2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1999 Poul-Henning Kamp.
5 * Copyright (c) 2008 Bjoern A. Zeeb.
6 * Copyright (c) 2009 James Gritton.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
51 #include <sys/linker.h>
54 #include <sys/mutex.h>
55 #include <sys/racct.h>
57 #include <sys/refcount.h>
59 #include <sys/sysent.h>
60 #include <sys/namei.h>
61 #include <sys/mount.h>
62 #include <sys/queue.h>
63 #include <sys/socket.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
72 #include <netinet/in.h>
78 #include <security/mac/mac_framework.h>
80 #define PRISON0_HOSTUUID_MODULE "hostuuid"
82 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
83 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
85 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
88 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
90 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
94 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
96 #define _PR_IP_SADDRSEL 0
100 /* prison0 describes what is "real" about the system. */
101 struct prison prison0 = {
107 .pr_securelevel = -1,
109 .pr_state = PRISON_STATE_ALIVE,
110 .pr_childmax = JAIL_MAX,
111 .pr_hostuuid = DEFAULT_HOSTUUID,
112 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
114 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
116 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
118 .pr_allow = PR_ALLOW_ALL_STATIC,
120 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
127 struct jailsys_flags {
133 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
134 struct sx allprison_lock;
135 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
136 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
137 LIST_HEAD(, prison_racct) allprison_racct;
140 static int get_next_prid(struct prison **insprp);
141 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
142 static void prison_complete(void *context, int pending);
143 static void prison_deref(struct prison *pr, int flags);
144 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
145 static int prison_lock_xlock(struct prison *pr, int flags);
146 static void prison_cleanup(struct prison *pr);
147 static void prison_free_not_last(struct prison *pr);
148 static void prison_proc_free_not_last(struct prison *pr);
149 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
151 static char *prison_path(struct prison *pr1, struct prison *pr2);
153 static void prison_racct_attach(struct prison *pr);
154 static void prison_racct_modify(struct prison *pr);
155 static void prison_racct_detach(struct prison *pr);
158 /* Flags for prison_deref */
159 #define PD_DEREF 0x01 /* Decrement pr_ref */
160 #define PD_DEUREF 0x02 /* Decrement pr_uref */
161 #define PD_KILL 0x04 /* Remove jail, kill processes, etc */
162 #define PD_LOCKED 0x10 /* pr_mtx is held */
163 #define PD_LIST_SLOCKED 0x20 /* allprison_lock is held shared */
164 #define PD_LIST_XLOCKED 0x40 /* allprison_lock is held exclusive */
165 #define PD_OP_FLAGS 0x07 /* Operation flags */
166 #define PD_LOCK_FLAGS 0x70 /* Lock status flags */
169 * Parameter names corresponding to PR_* flag values. Size values are for kvm
170 * as we cannot figure out the size of a sparse array, or an array without a
173 static struct bool_flags pr_flag_bool[] = {
174 {"persist", "nopersist", PR_PERSIST},
176 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
179 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
182 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
184 static struct jailsys_flags pr_flag_jailsys[] = {
185 {"host", 0, PR_HOST},
187 {"vnet", 0, PR_VNET},
190 {"ip4", PR_IP4_USER, PR_IP4_USER},
193 {"ip6", PR_IP6_USER, PR_IP6_USER},
196 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
199 * Make this array full-size so dynamic parameters can be added.
200 * It is protected by prison0.mtx, but lockless reading is allowed
201 * with an atomic check of the flag values.
203 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
204 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
205 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
206 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
207 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
208 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
209 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
210 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
211 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
212 {"allow.reserved_ports", "allow.noreserved_ports",
213 PR_ALLOW_RESERVED_PORTS},
214 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
215 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
216 PR_ALLOW_UNPRIV_DEBUG},
217 {"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
219 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
220 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
222 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
223 PR_ALLOW_RESERVED_PORTS | \
224 PR_ALLOW_UNPRIV_DEBUG | \
226 #define JAIL_DEFAULT_ENFORCE_STATFS 2
227 #define JAIL_DEFAULT_DEVFS_RSNUM 0
228 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
229 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
230 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
231 #if defined(INET) || defined(INET6)
232 static unsigned jail_max_af_ips = 255;
236 * Initialize the parts of prison0 that can't be static-initialized with
237 * constants. This is called from proc0_init() after creating thread0 cpuset.
242 uint8_t *file, *data;
244 char buf[sizeof(prison0.pr_hostuuid)];
247 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
248 prison0.pr_osreldate = osreldate;
249 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
251 /* If we have a preloaded hostuuid, use it. */
252 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
254 data = preload_fetch_addr(file);
255 size = preload_fetch_size(file);
258 * The preloaded data may include trailing whitespace, almost
259 * certainly a newline; skip over any whitespace or
260 * non-printable characters to be safe.
262 while (size > 0 && data[size - 1] <= 0x20) {
269 * Not NUL-terminated when passed from loader, but
270 * validate_uuid requires that due to using sscanf (as
271 * does the subsequent strlcpy, since it still reads
272 * past the given size to return the true length);
273 * bounce to a temporary buffer to fix.
275 if (size >= sizeof(buf))
278 memcpy(buf, data, size);
281 if (validate_uuid(buf, size, NULL, 0) != 0)
285 (void)strlcpy(prison0.pr_hostuuid, buf,
286 sizeof(prison0.pr_hostuuid));
289 if (bootverbose && !valid) {
290 printf("hostuuid: preload data malformed: '%.*s'\n",
296 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
305 sys_jail(struct thread *td, struct jail_args *uap)
311 error = copyin(uap->jail, &version, sizeof(uint32_t));
320 /* FreeBSD single IPv4 jails. */
321 bzero(&j, sizeof(struct jail));
322 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
325 j.version = j0.version;
327 j.hostname = j0.hostname;
328 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
334 * Version 1 was used by multi-IPv4 jail implementations
335 * that never made it into the official kernel.
339 case 2: /* JAIL_API_VERSION */
340 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
341 error = copyin(uap->jail, &j, sizeof(struct jail));
347 /* Sci-Fi jails are not supported, sorry. */
350 return (kern_jail(td, &j));
354 kern_jail(struct thread *td, struct jail *j)
356 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
365 char *u_path, *u_hostname, *u_name;
366 struct bool_flags *bf;
369 struct in_addr *u_ip4;
372 struct in6_addr *u_ip6;
375 int error, enforce_statfs;
377 bzero(&optiov, sizeof(optiov));
378 opt.uio_iov = optiov;
382 opt.uio_segflg = UIO_SYSSPACE;
383 opt.uio_rw = UIO_READ;
386 /* Set permissions for top-level jails from sysctls. */
387 if (!jailed(td->td_ucred)) {
388 for (bf = pr_flag_allow;
389 bf < pr_flag_allow + nitems(pr_flag_allow) &&
390 atomic_load_int(&bf->flag) != 0;
392 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
393 (jail_default_allow & bf->flag)
394 ? bf->name : bf->noname);
395 optiov[opt.uio_iovcnt].iov_len =
396 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
399 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
400 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
402 enforce_statfs = jail_default_enforce_statfs;
403 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
404 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
408 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
410 ip4s = (j->version == 0) ? 1 : j->ip4s;
411 if (ip4s > jail_max_af_ips)
413 tmplen += ip4s * sizeof(struct in_addr);
419 if (j->ip6s > jail_max_af_ips)
421 tmplen += j->ip6s * sizeof(struct in6_addr);
426 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
427 u_hostname = u_path + MAXPATHLEN;
428 u_name = u_hostname + MAXHOSTNAMELEN;
430 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
434 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
436 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
439 optiov[opt.uio_iovcnt].iov_base = "path";
440 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
442 optiov[opt.uio_iovcnt].iov_base = u_path;
443 error = copyinstr(j->path, u_path, MAXPATHLEN,
444 &optiov[opt.uio_iovcnt].iov_len);
446 free(u_path, M_TEMP);
450 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
451 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
453 optiov[opt.uio_iovcnt].iov_base = u_hostname;
454 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
455 &optiov[opt.uio_iovcnt].iov_len);
457 free(u_path, M_TEMP);
461 if (j->jailname != NULL) {
462 optiov[opt.uio_iovcnt].iov_base = "name";
463 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
465 optiov[opt.uio_iovcnt].iov_base = u_name;
466 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
467 &optiov[opt.uio_iovcnt].iov_len);
469 free(u_path, M_TEMP);
475 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
476 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
478 optiov[opt.uio_iovcnt].iov_base = u_ip4;
479 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
481 u_ip4->s_addr = j->ip4s;
483 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
485 free(u_path, M_TEMP);
492 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
493 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
495 optiov[opt.uio_iovcnt].iov_base = u_ip6;
496 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
497 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
499 free(u_path, M_TEMP);
504 KASSERT(opt.uio_iovcnt <= nitems(optiov),
505 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
506 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
507 free(u_path, M_TEMP);
512 * struct jail_set_args {
513 * struct iovec *iovp;
514 * unsigned int iovcnt;
519 sys_jail_set(struct thread *td, struct jail_set_args *uap)
524 /* Check that we have an even number of iovecs. */
528 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
531 error = kern_jail_set(td, auio, uap->flags);
537 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
544 struct in6_addr *ip6;
547 struct vfsoptlist *opts;
548 struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
550 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
551 char *g_path, *osrelstr;
552 struct bool_flags *bf;
553 struct jailsys_flags *jsf;
554 #if defined(INET) || defined(INET6)
559 size_t namelen, onamelen, pnamelen;
560 int born, created, cuflags, descend, drflags, enforce;
561 int error, errmsg_len, errmsg_pos;
562 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
563 int jid, jsys, len, level;
564 int childmax, osreldt, rsnum, slevel;
565 #if defined(INET) || defined(INET6)
574 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
575 uint64_t pr_allow_diff;
579 error = priv_check(td, PRIV_JAIL_SET);
580 if (!error && (flags & JAIL_ATTACH))
581 error = priv_check(td, PRIV_JAIL_ATTACH);
584 mypr = td->td_ucred->cr_prison;
585 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
587 if (flags & ~JAIL_SET_MASK)
591 * Check all the parameters before committing to anything. Not all
592 * errors can be caught early, but we may as well try. Also, this
593 * takes care of some expensive stuff (path lookup) before getting
594 * the allprison lock.
596 * XXX Jails are not filesystems, and jail parameters are not mount
597 * options. But it makes more sense to re-use the vfsopt code
598 * than duplicate it under a different name.
600 error = vfs_buildopts(optuio, &opts);
611 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
614 vfs_opterror(opts, "no valid operation (create or update)");
618 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
624 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
633 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
641 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
646 else if (enforce < 0 || enforce > 2) {
652 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
660 pr_flags = ch_flags = 0;
661 for (bf = pr_flag_bool;
662 bf < pr_flag_bool + nitems(pr_flag_bool);
664 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
665 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
667 ch_flags |= pr_flags;
668 for (jsf = pr_flag_jailsys;
669 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
671 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
677 case JAIL_SYS_DISABLE:
682 pr_flags |= jsf->disable;
685 pr_flags |= jsf->new;
687 case JAIL_SYS_INHERIT:
693 ch_flags |= jsf->new | jsf->disable;
695 if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
696 && !(pr_flags & PR_PERSIST)) {
698 vfs_opterror(opts, "new jail must persist or attach");
702 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
704 vfs_opterror(opts, "vnet cannot be changed after creation");
709 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
711 vfs_opterror(opts, "ip4 cannot be changed after creation");
716 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
718 vfs_opterror(opts, "ip6 cannot be changed after creation");
723 pr_allow = ch_allow = 0;
724 for (bf = pr_flag_allow;
725 bf < pr_flag_allow + nitems(pr_flag_allow) &&
726 atomic_load_int(&bf->flag) != 0;
728 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
729 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
731 ch_allow |= pr_allow;
733 error = vfs_getopt(opts, "name", (void **)&name, &len);
739 if (len == 0 || name[len - 1] != '\0') {
743 if (len > MAXHOSTNAMELEN) {
744 error = ENAMETOOLONG;
749 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
757 if (len == 0 || host[len - 1] != '\0') {
761 if (len > MAXHOSTNAMELEN) {
762 error = ENAMETOOLONG;
767 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
775 if (len == 0 || domain[len - 1] != '\0') {
779 if (len > MAXHOSTNAMELEN) {
780 error = ENAMETOOLONG;
785 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
793 if (len == 0 || uuid[len - 1] != '\0') {
797 if (len > HOSTUUIDLEN) {
798 error = ENAMETOOLONG;
803 #ifdef COMPAT_FREEBSD32
804 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
807 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
811 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
823 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
828 else if (ip4s & (sizeof(*ip4) - 1)) {
832 ch_flags |= PR_IP4_USER;
833 pr_flags |= PR_IP4_USER;
835 ip4s /= sizeof(*ip4);
836 if (ip4s > jail_max_af_ips) {
838 vfs_opterror(opts, "too many IPv4 addresses");
841 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
842 bcopy(op, ip4, ip4s * sizeof(*ip4));
844 * IP addresses are all sorted but ip[0] to preserve
845 * the primary IP address as given from userland.
846 * This special IP is used for unbound outgoing
847 * connections as well for "loopback" traffic in case
848 * source address selection cannot find any more fitting
849 * address to connect from.
852 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
855 * Check for duplicate addresses and do some simple
856 * zero and broadcast checks. If users give other bogus
857 * addresses it is their problem.
859 * We do not have to care about byte order for these
860 * checks so we will do them in NBO.
862 for (ii = 0; ii < ip4s; ii++) {
863 if (ip4[ii].s_addr == INADDR_ANY ||
864 ip4[ii].s_addr == INADDR_BROADCAST) {
869 (ip4[0].s_addr == ip4[ii+1].s_addr ||
870 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
880 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
885 else if (ip6s & (sizeof(*ip6) - 1)) {
889 ch_flags |= PR_IP6_USER;
890 pr_flags |= PR_IP6_USER;
892 ip6s /= sizeof(*ip6);
893 if (ip6s > jail_max_af_ips) {
895 vfs_opterror(opts, "too many IPv6 addresses");
898 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
899 bcopy(op, ip6, ip6s * sizeof(*ip6));
901 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
903 for (ii = 0; ii < ip6s; ii++) {
904 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
909 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
910 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
920 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
921 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
924 "vnet jails cannot have IP address restrictions");
929 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
935 if (flags & JAIL_UPDATE) {
938 "osrelease cannot be changed after creation");
941 if (len == 0 || osrelstr[len - 1] != '\0') {
945 if (len >= OSRELEASELEN) {
946 error = ENAMETOOLONG;
948 "osrelease string must be 1-%d bytes long",
954 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
960 if (flags & JAIL_UPDATE) {
963 "osreldate cannot be changed after creation");
968 vfs_opterror(opts, "osreldate cannot be 0");
974 error = vfs_getopt(opts, "path", (void **)&path, &len);
980 if (flags & JAIL_UPDATE) {
983 "path cannot be changed after creation");
986 if (len == 0 || path[len - 1] != '\0') {
990 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
996 NDFREE(&nd, NDF_ONLY_PNBUF);
997 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
998 strlcpy(g_path, path, MAXPATHLEN);
999 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1003 /* exit on other errors */
1006 if (root->v_type != VDIR) {
1015 * Find the specified jail, or at least its parent.
1016 * This abuses the file error codes ENOENT and EEXIST.
1020 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1021 namelc = strrchr(name, '.');
1022 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1026 sx_xlock(&allprison_lock);
1027 drflags = PD_LIST_XLOCKED;
1029 if (!prison_isalive(ppr)) {
1030 /* This jail is dying. This process will surely follow. */
1037 vfs_opterror(opts, "negative jid");
1041 * See if a requested jid already exists. Keep track of
1042 * where it can be inserted later.
1044 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1045 if (inspr->pr_id < jid)
1047 if (inspr->pr_id > jid)
1050 mtx_lock(&pr->pr_mtx);
1051 drflags |= PD_LOCKED;
1056 /* Create: jid must not exist. */
1057 if (cuflags == JAIL_CREATE) {
1059 * Even creators that cannot see the jail will
1063 vfs_opterror(opts, "jail %d already exists",
1067 if (!prison_ischild(mypr, pr)) {
1069 * Updaters get ENOENT if they cannot see the
1070 * jail. This is true even for CREATE | UPDATE,
1071 * which normally cannot give this error.
1074 vfs_opterror(opts, "jail %d not found", jid);
1077 ppr = pr->pr_parent;
1078 if (!prison_isalive(ppr)) {
1080 vfs_opterror(opts, "jail %d is dying",
1084 if (!prison_isalive(pr)) {
1085 if (!(flags & JAIL_DYING)) {
1087 vfs_opterror(opts, "jail %d is dying",
1091 if ((flags & JAIL_ATTACH) ||
1092 (pr_flags & PR_PERSIST)) {
1094 * A dying jail might be resurrected
1095 * (via attach or persist), but first
1096 * it must determine if another jail
1097 * has claimed its name. Accomplish
1098 * this by implicitly re-setting the
1102 name = prison_name(mypr, pr);
1106 /* Update: jid must exist. */
1107 if (cuflags == JAIL_UPDATE) {
1109 vfs_opterror(opts, "jail %d not found", jid);
1115 * If the caller provided a name, look for a jail by that name.
1116 * This has different semantics for creates and updates keyed by jid
1117 * (where the name must not already exist in a different jail),
1118 * and updates keyed by the name itself (where the name must exist
1119 * because that is the jail being updated).
1123 namelc = strrchr(name, '.');
1128 * This is a hierarchical name. Split it into the
1129 * parent and child names, and make sure the parent
1130 * exists or matches an already found jail.
1133 if (strncmp(name, ppr->pr_name, namelc - name)
1134 || ppr->pr_name[namelc - name] != '\0') {
1137 "cannot change jail's parent");
1142 ppr = prison_find_name(mypr, name);
1146 "jail \"%s\" not found", name);
1149 mtx_unlock(&ppr->pr_mtx);
1150 if (!prison_isalive(ppr)) {
1153 "jail \"%s\" is dying", name);
1160 if (namelc[0] != '\0') {
1162 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1164 FOREACH_PRISON_CHILD(ppr, tpr) {
1166 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1167 if (prison_isalive(tpr)) {
1169 cuflags != JAIL_CREATE) {
1175 mtx_lock(&pr->pr_mtx);
1176 drflags |= PD_LOCKED;
1180 * Create, or update(jid):
1181 * name must not exist in an
1182 * active sibling jail.
1186 "jail \"%s\" already exists",
1191 cuflags != JAIL_CREATE) {
1196 /* If no active jail is found, use a dying one. */
1197 if (deadpr != NULL && pr == NULL) {
1198 if (flags & JAIL_DYING) {
1200 mtx_lock(&pr->pr_mtx);
1201 drflags |= PD_LOCKED;
1202 } else if (cuflags == JAIL_UPDATE) {
1205 "jail \"%s\" is dying", name);
1209 /* Update: name must exist if no jid. */
1210 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1212 vfs_opterror(opts, "jail \"%s\" not found",
1218 /* Update: must provide a jid or name. */
1219 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1221 vfs_opterror(opts, "update specified no jail");
1225 /* If there's no prison to update, create a new one and link it in. */
1226 created = pr == NULL;
1228 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1229 if (tpr->pr_childcount >= tpr->pr_childmax) {
1231 vfs_opterror(opts, "prison limit exceeded");
1234 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1236 vfs_opterror(opts, "no available jail IDs");
1240 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1241 pr->pr_state = PRISON_STATE_INVALID;
1242 refcount_init(&pr->pr_ref, 1);
1243 refcount_init(&pr->pr_uref, 0);
1244 drflags |= PD_DEREF;
1245 LIST_INIT(&pr->pr_children);
1246 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1247 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1251 TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1253 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1255 pr->pr_parent = ppr;
1257 prison_proc_hold(ppr);
1258 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1259 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1260 tpr->pr_childcount++;
1262 /* Set some default values, and inherit some from the parent. */
1267 root = mypr->pr_root;
1270 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1271 pr->pr_flags |= PR_HOST;
1272 #if defined(INET) || defined(INET6)
1274 if (!(pr_flags & PR_VNET))
1278 if (!(ch_flags & PR_IP4_USER))
1279 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1280 else if (!(pr_flags & PR_IP4_USER)) {
1281 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1282 if (ppr->pr_ip4 != NULL) {
1283 pr->pr_ip4s = ppr->pr_ip4s;
1284 pr->pr_ip4 = malloc(pr->pr_ip4s *
1285 sizeof(struct in_addr), M_PRISON,
1287 bcopy(ppr->pr_ip4, pr->pr_ip4,
1288 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1293 if (!(ch_flags & PR_IP6_USER))
1294 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1295 else if (!(pr_flags & PR_IP6_USER)) {
1296 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1297 if (ppr->pr_ip6 != NULL) {
1298 pr->pr_ip6s = ppr->pr_ip6s;
1299 pr->pr_ip6 = malloc(pr->pr_ip6s *
1300 sizeof(struct in6_addr), M_PRISON,
1302 bcopy(ppr->pr_ip6, pr->pr_ip6,
1303 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1309 /* Source address selection is always on by default. */
1310 pr->pr_flags |= _PR_IP_SADDRSEL;
1312 pr->pr_securelevel = ppr->pr_securelevel;
1313 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1314 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1315 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1317 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1318 if (osrelstr == NULL)
1319 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1320 sizeof(pr->pr_osrelease));
1322 strlcpy(pr->pr_osrelease, osrelstr,
1323 sizeof(pr->pr_osrelease));
1326 /* Allocate a new vnet if specified. */
1327 pr->pr_vnet = (pr_flags & PR_VNET)
1328 ? vnet_alloc() : ppr->pr_vnet;
1331 * Allocate a dedicated cpuset for each jail.
1332 * Unlike other initial settings, this may return an error.
1334 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1338 mtx_lock(&pr->pr_mtx);
1339 drflags |= PD_LOCKED;
1342 * Grab a reference for existing prisons, to ensure they
1343 * continue to exist for the duration of the call.
1346 drflags |= PD_DEREF;
1347 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1348 if ((pr->pr_flags & PR_VNET) &&
1349 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1352 "vnet jails cannot have IP address restrictions");
1357 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1360 "ip4 cannot be changed after creation");
1365 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1368 "ip6 cannot be changed after creation");
1374 /* Do final error checking before setting anything. */
1376 if (slevel < ppr->pr_securelevel) {
1382 if (childmax >= ppr->pr_childmax) {
1388 if (enforce < ppr->pr_enforce_statfs) {
1395 * devfs_rsnum is a uint16_t
1397 if (rsnum < 0 || rsnum > 65535) {
1402 * Nested jails always inherit parent's devfs ruleset
1404 if (jailed(td->td_ucred)) {
1405 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1409 rsnum = ppr->pr_devfs_rsnum;
1414 if (ppr->pr_flags & PR_IP4) {
1416 * Make sure the new set of IP addresses is a
1417 * subset of the parent's list. Don't worry
1418 * about the parent being unlocked, as any
1419 * setting is done with allprison_lock held.
1421 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1422 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1424 if (ij == ppr->pr_ip4s) {
1429 for (ii = ij = 1; ii < ip4s; ii++) {
1430 if (ip4[ii].s_addr ==
1431 ppr->pr_ip4[0].s_addr)
1433 for (; ij < ppr->pr_ip4s; ij++)
1434 if (ip4[ii].s_addr ==
1435 ppr->pr_ip4[ij].s_addr)
1437 if (ij == ppr->pr_ip4s)
1440 if (ij == ppr->pr_ip4s) {
1447 * Check for conflicting IP addresses. We permit them
1448 * if there is no more than one IP on each jail. If
1449 * there is a duplicate on a jail with more than one
1450 * IP stop checking and return error.
1453 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1454 if (tppr->pr_flags & PR_VNET)
1459 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1462 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1464 !prison_isalive(tpr)) {
1468 if (!(tpr->pr_flags & PR_IP4_USER))
1471 if (tpr->pr_ip4 == NULL ||
1472 (ip4s == 1 && tpr->pr_ip4s == 1))
1474 for (ii = 0; ii < ip4s; ii++) {
1475 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1479 "IPv4 addresses clash");
1488 if (ppr->pr_flags & PR_IP6) {
1490 * Make sure the new set of IP addresses is a
1491 * subset of the parent's list.
1493 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1494 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1497 if (ij == ppr->pr_ip6s) {
1502 for (ii = ij = 1; ii < ip6s; ii++) {
1503 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1506 for (; ij < ppr->pr_ip6s; ij++)
1507 if (IN6_ARE_ADDR_EQUAL(
1508 &ip6[ii], &ppr->pr_ip6[ij]))
1510 if (ij == ppr->pr_ip6s)
1513 if (ij == ppr->pr_ip6s) {
1519 /* Check for conflicting IP addresses. */
1521 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1522 if (tppr->pr_flags & PR_VNET)
1527 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1530 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1532 !prison_isalive(tpr)) {
1536 if (!(tpr->pr_flags & PR_IP6_USER))
1539 if (tpr->pr_ip6 == NULL ||
1540 (ip6s == 1 && tpr->pr_ip6s == 1))
1542 for (ii = 0; ii < ip6s; ii++) {
1543 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1547 "IPv6 addresses clash");
1554 onamelen = namelen = 0;
1555 if (namelc != NULL) {
1556 /* Give a default name of the jid. Also allow the name to be
1557 * explicitly the jid - but not any other number, and only in
1558 * normal form (no leading zero/etc).
1560 if (namelc[0] == '\0')
1561 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1562 else if ((strtoul(namelc, &p, 10) != jid ||
1563 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1566 "name cannot be numeric (unless it is the jid)");
1570 * Make sure the name isn't too long for the prison or its
1573 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1574 onamelen = strlen(pr->pr_name + pnamelen);
1575 namelen = strlen(namelc);
1576 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1577 error = ENAMETOOLONG;
1580 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1581 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1582 sizeof(pr->pr_name)) {
1583 error = ENAMETOOLONG;
1588 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1589 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1595 * Let modules check their parameters. This requires unlocking and
1596 * then re-locking the prison, but this is still a valid state as long
1597 * as allprison_lock remains xlocked.
1599 mtx_unlock(&pr->pr_mtx);
1600 drflags &= ~PD_LOCKED;
1601 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1604 mtx_lock(&pr->pr_mtx);
1605 drflags |= PD_LOCKED;
1607 /* At this point, all valid parameters should have been noted. */
1608 TAILQ_FOREACH(opt, opts, link) {
1609 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1611 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1616 /* Set the parameters of the prison. */
1619 if (pr_flags & PR_IP4_USER) {
1620 pr->pr_flags |= PR_IP4;
1621 free(pr->pr_ip4, M_PRISON);
1625 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1627 if (tpr->pr_flags & PR_VNET) {
1632 if (prison_restrict_ip4(tpr, NULL)) {
1641 if (pr_flags & PR_IP6_USER) {
1642 pr->pr_flags |= PR_IP6;
1643 free(pr->pr_ip6, M_PRISON);
1647 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1649 if (tpr->pr_flags & PR_VNET) {
1654 if (prison_restrict_ip6(tpr, NULL)) {
1662 pr->pr_securelevel = slevel;
1663 /* Set all child jails to be at least this level. */
1664 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1665 if (tpr->pr_securelevel < slevel)
1666 tpr->pr_securelevel = slevel;
1669 pr->pr_childmax = childmax;
1670 /* Set all child jails to under this limit. */
1671 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1672 if (tpr->pr_childmax > childmax - level)
1673 tpr->pr_childmax = childmax > level
1674 ? childmax - level : 0;
1677 pr->pr_enforce_statfs = enforce;
1678 /* Pass this restriction on to the children. */
1679 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1680 if (tpr->pr_enforce_statfs < enforce)
1681 tpr->pr_enforce_statfs = enforce;
1684 pr->pr_devfs_rsnum = rsnum;
1685 /* Pass this restriction on to the children. */
1686 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1687 tpr->pr_devfs_rsnum = rsnum;
1689 if (namelc != NULL) {
1690 if (ppr == &prison0)
1691 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1693 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1694 ppr->pr_name, namelc);
1695 /* Change this component of child names. */
1696 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1697 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1698 strlen(tpr->pr_name + onamelen) + 1);
1699 bcopy(pr->pr_name, tpr->pr_name, namelen);
1703 /* Try to keep a real-rooted full pathname. */
1704 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1708 if (PR_HOST & ch_flags & ~pr_flags) {
1709 if (pr->pr_flags & PR_HOST) {
1711 * Copy the parent's host info. As with pr_ip4 above,
1712 * the lack of a lock on the parent is not a problem;
1713 * it is always set with allprison_lock at least
1714 * shared, and is held exclusively here.
1716 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1717 sizeof(pr->pr_hostname));
1718 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1719 sizeof(pr->pr_domainname));
1720 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1721 sizeof(pr->pr_hostuuid));
1722 pr->pr_hostid = pr->pr_parent->pr_hostid;
1724 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1725 /* Set this prison, and any descendants without PR_HOST. */
1727 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1729 strlcpy(pr->pr_domainname, domain,
1730 sizeof(pr->pr_domainname));
1732 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1734 pr->pr_hostid = hid;
1735 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1736 if (tpr->pr_flags & PR_HOST)
1740 strlcpy(tpr->pr_hostname,
1742 sizeof(tpr->pr_hostname));
1744 strlcpy(tpr->pr_domainname,
1746 sizeof(tpr->pr_domainname));
1748 strlcpy(tpr->pr_hostuuid,
1750 sizeof(tpr->pr_hostuuid));
1752 tpr->pr_hostid = hid;
1756 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1757 if ((tallow = ch_allow & ~pr_allow))
1758 prison_set_allow_locked(pr, tallow, 0);
1760 * Persistent prisons get an extra reference, and prisons losing their
1761 * persist flag lose that reference.
1763 born = !prison_isalive(pr);
1764 if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
1765 if (pr_flags & PR_PERSIST) {
1768 * This may make a dead prison alive again, but wait
1769 * to label it as such until after OSD calls have had
1770 * a chance to run (and perhaps to fail).
1772 refcount_acquire(&pr->pr_uref);
1774 drflags |= PD_DEUREF;
1775 prison_free_not_last(pr);
1778 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1779 mtx_unlock(&pr->pr_mtx);
1780 drflags &= ~PD_LOCKED;
1782 * Any errors past this point will need to de-persist newly created
1783 * prisons, as well as call remove methods.
1789 if (racct_enable && created)
1790 prison_racct_attach(pr);
1793 /* Locks may have prevented a complete restriction of child IP
1794 * addresses. If so, allocate some more memory and try again.
1799 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1800 mtx_lock(&pr->pr_mtx);
1802 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1804 if (tpr->pr_flags & PR_VNET) {
1809 if (prison_restrict_ip4(tpr, ip4)) {
1816 mtx_unlock(&pr->pr_mtx);
1822 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1823 mtx_lock(&pr->pr_mtx);
1825 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1827 if (tpr->pr_flags & PR_VNET) {
1832 if (prison_restrict_ip6(tpr, ip6)) {
1839 mtx_unlock(&pr->pr_mtx);
1843 /* Let the modules do their work. */
1845 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1849 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1854 * A new prison is now ready to be seen; either it has gained a user
1855 * reference via persistence, or is about to gain one via attachment.
1858 drflags = prison_lock_xlock(pr, drflags);
1859 pr->pr_state = PRISON_STATE_ALIVE;
1862 /* Attach this process to the prison if requested. */
1863 if (flags & JAIL_ATTACH) {
1864 error = do_jail_attach(td, pr,
1865 prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
1866 drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
1868 vfs_opterror(opts, "attach failed");
1874 if (racct_enable && !created) {
1875 if (drflags & PD_LOCKED) {
1876 mtx_unlock(&pr->pr_mtx);
1877 drflags &= ~PD_LOCKED;
1879 if (drflags & PD_LIST_XLOCKED) {
1880 sx_xunlock(&allprison_lock);
1881 drflags &= ~PD_LIST_XLOCKED;
1883 prison_racct_modify(pr);
1887 drflags &= ~PD_KILL;
1888 td->td_retval[0] = pr->pr_id;
1891 /* Release any temporary prison holds and/or locks. */
1893 prison_deref(pr, drflags);
1894 else if (drflags & PD_LIST_SLOCKED)
1895 sx_sunlock(&allprison_lock);
1896 else if (drflags & PD_LIST_XLOCKED)
1897 sx_xunlock(&allprison_lock);
1902 /* Write the error message back to userspace. */
1903 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1904 &errmsg_len) == 0 && errmsg_len > 0) {
1905 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1906 if (optuio->uio_segflg == UIO_SYSSPACE)
1908 optuio->uio_iov[errmsg_pos].iov_base,
1912 optuio->uio_iov[errmsg_pos].iov_base,
1918 free(ip4, M_PRISON);
1921 free(ip6, M_PRISON);
1924 free(g_path, M_TEMP);
1930 * Find the next available prison ID. Return the ID on success, or zero
1931 * on failure. Also set a pointer to the allprison list entry the prison
1932 * should be inserted before.
1935 get_next_prid(struct prison **insprp)
1937 struct prison *inspr;
1940 jid = lastprid % JAIL_MAX + 1;
1941 if (TAILQ_EMPTY(&allprison) ||
1942 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
1944 * A common case is for all jails to be implicitly numbered,
1945 * which means they'll go on the end of the list, at least
1946 * for the first JAIL_MAX times.
1951 * Take two passes through the allprison list: first starting
1952 * with the proposed jid, then ending with it.
1954 for (maxid = JAIL_MAX; maxid != 0; ) {
1955 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1956 if (inspr->pr_id < jid)
1958 if (inspr->pr_id > jid) {
1959 /* Found an opening. */
1963 if (++jid > maxid) {
1964 if (lastprid == maxid || lastprid == 0)
1967 * The entire legal range
1968 * has been traversed
1972 /* Try again from the start. */
1978 if (inspr == NULL) {
1979 /* Found room at the end of the list. */
1990 * struct jail_get_args {
1991 * struct iovec *iovp;
1992 * unsigned int iovcnt;
1997 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2002 /* Check that we have an even number of iovecs. */
2003 if (uap->iovcnt & 1)
2006 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2009 error = kern_jail_get(td, auio, uap->flags);
2011 error = copyout(auio->uio_iov, uap->iovp,
2012 uap->iovcnt * sizeof (struct iovec));
2018 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2020 struct bool_flags *bf;
2021 struct jailsys_flags *jsf;
2022 struct prison *pr, *mypr;
2024 struct vfsoptlist *opts;
2025 char *errmsg, *name;
2026 int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2029 if (flags & ~JAIL_GET_MASK)
2032 /* Get the parameter list. */
2033 error = vfs_buildopts(optuio, &opts);
2036 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2037 mypr = td->td_ucred->cr_prison;
2041 * Find the prison specified by one of: lastjid, jid, name.
2043 sx_slock(&allprison_lock);
2044 drflags = PD_LIST_SLOCKED;
2045 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2047 TAILQ_FOREACH(pr, &allprison, pr_list) {
2048 if (pr->pr_id > jid &&
2049 ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2050 prison_ischild(mypr, pr)) {
2051 mtx_lock(&pr->pr_mtx);
2052 drflags |= PD_LOCKED;
2057 vfs_opterror(opts, "no jail after %d", jid);
2059 } else if (error != ENOENT)
2062 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2065 pr = prison_find_child(mypr, jid);
2067 drflags |= PD_LOCKED;
2068 if (!(prison_isalive(pr) ||
2069 (flags & JAIL_DYING))) {
2071 vfs_opterror(opts, "jail %d is dying",
2078 vfs_opterror(opts, "jail %d not found", jid);
2081 } else if (error != ENOENT)
2084 error = vfs_getopt(opts, "name", (void **)&name, &len);
2086 if (len == 0 || name[len - 1] != '\0') {
2090 pr = prison_find_name(mypr, name);
2092 drflags |= PD_LOCKED;
2093 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2095 vfs_opterror(opts, "jail \"%s\" is dying",
2102 vfs_opterror(opts, "jail \"%s\" not found", name);
2104 } else if (error != ENOENT)
2107 vfs_opterror(opts, "no jail specified");
2112 /* Get the parameters of the prison. */
2114 drflags |= PD_DEREF;
2115 td->td_retval[0] = pr->pr_id;
2116 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2117 if (error != 0 && error != ENOENT)
2119 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2120 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2121 if (error != 0 && error != ENOENT)
2123 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2124 if (error != 0 && error != ENOENT)
2126 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2127 sizeof(pr->pr_cpuset->cs_id));
2128 if (error != 0 && error != ENOENT)
2130 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2131 if (error != 0 && error != ENOENT)
2134 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2135 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2136 if (error != 0 && error != ENOENT)
2140 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2141 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2142 if (error != 0 && error != ENOENT)
2145 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2146 sizeof(pr->pr_securelevel));
2147 if (error != 0 && error != ENOENT)
2149 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2150 sizeof(pr->pr_childcount));
2151 if (error != 0 && error != ENOENT)
2153 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2154 sizeof(pr->pr_childmax));
2155 if (error != 0 && error != ENOENT)
2157 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2158 if (error != 0 && error != ENOENT)
2160 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2161 if (error != 0 && error != ENOENT)
2163 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2164 if (error != 0 && error != ENOENT)
2166 #ifdef COMPAT_FREEBSD32
2167 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2168 uint32_t hid32 = pr->pr_hostid;
2170 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2173 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2174 sizeof(pr->pr_hostid));
2175 if (error != 0 && error != ENOENT)
2177 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2178 sizeof(pr->pr_enforce_statfs));
2179 if (error != 0 && error != ENOENT)
2181 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2182 sizeof(pr->pr_devfs_rsnum));
2183 if (error != 0 && error != ENOENT)
2185 for (bf = pr_flag_bool;
2186 bf < pr_flag_bool + nitems(pr_flag_bool);
2188 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2189 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2190 if (error != 0 && error != ENOENT)
2193 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2194 if (error != 0 && error != ENOENT)
2197 for (jsf = pr_flag_jailsys;
2198 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2200 f = pr->pr_flags & (jsf->disable | jsf->new);
2201 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2202 : (f == jsf->new) ? JAIL_SYS_NEW
2204 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2205 if (error != 0 && error != ENOENT)
2208 for (bf = pr_flag_allow;
2209 bf < pr_flag_allow + nitems(pr_flag_allow) &&
2210 atomic_load_int(&bf->flag) != 0;
2212 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2213 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2214 if (error != 0 && error != ENOENT)
2217 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2218 if (error != 0 && error != ENOENT)
2221 i = !prison_isalive(pr);
2222 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2223 if (error != 0 && error != ENOENT)
2226 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2227 if (error != 0 && error != ENOENT)
2229 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2230 sizeof(pr->pr_osreldate));
2231 if (error != 0 && error != ENOENT)
2233 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2234 if (error != 0 && error != ENOENT)
2237 /* Get the module parameters. */
2238 mtx_unlock(&pr->pr_mtx);
2239 drflags &= ~PD_LOCKED;
2240 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2243 prison_deref(pr, drflags);
2247 /* By now, all parameters should have been noted. */
2248 TAILQ_FOREACH(opt, opts, link) {
2249 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2251 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2256 /* Write the fetched parameters back to userspace. */
2258 TAILQ_FOREACH(opt, opts, link) {
2259 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2260 pos = 2 * opt->pos + 1;
2261 optuio->uio_iov[pos].iov_len = opt->len;
2262 if (opt->value != NULL) {
2263 if (optuio->uio_segflg == UIO_SYSSPACE) {
2265 optuio->uio_iov[pos].iov_base,
2268 error = copyout(opt->value,
2269 optuio->uio_iov[pos].iov_base,
2279 /* Release any temporary prison holds and/or locks. */
2281 prison_deref(pr, drflags);
2282 else if (drflags & PD_LIST_SLOCKED)
2283 sx_sunlock(&allprison_lock);
2284 if (error && errmsg_pos >= 0) {
2285 /* Write the error message back to userspace. */
2286 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2287 errmsg_pos = 2 * errmsg_pos + 1;
2288 if (errmsg_len > 0) {
2289 if (optuio->uio_segflg == UIO_SYSSPACE)
2291 optuio->uio_iov[errmsg_pos].iov_base,
2295 optuio->uio_iov[errmsg_pos].iov_base,
2304 * struct jail_remove_args {
2309 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2314 error = priv_check(td, PRIV_JAIL_REMOVE);
2318 sx_xlock(&allprison_lock);
2319 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2321 sx_xunlock(&allprison_lock);
2324 if (!prison_isalive(pr)) {
2325 /* Silently ignore already-dying prisons. */
2326 mtx_unlock(&pr->pr_mtx);
2327 sx_xunlock(&allprison_lock);
2330 prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2335 * struct jail_attach_args {
2340 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2345 error = priv_check(td, PRIV_JAIL_ATTACH);
2349 sx_slock(&allprison_lock);
2350 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2352 sx_sunlock(&allprison_lock);
2356 /* Do not allow a process to attach to a prison that is not alive. */
2357 if (!prison_isalive(pr)) {
2358 mtx_unlock(&pr->pr_mtx);
2359 sx_sunlock(&allprison_lock);
2363 return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2367 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2370 struct ucred *newcred, *oldcred;
2373 mtx_assert(&pr->pr_mtx, MA_OWNED);
2374 sx_assert(&allprison_lock, SX_LOCKED);
2375 drflags &= PD_LOCK_FLAGS;
2377 * XXX: Note that there is a slight race here if two threads
2378 * in the same privileged process attempt to attach to two
2379 * different jails at the same time. It is important for
2380 * user processes not to do this, or they might end up with
2381 * a process root from one prison, but attached to the jail
2385 refcount_acquire(&pr->pr_uref);
2386 drflags |= PD_DEREF | PD_DEUREF;
2387 mtx_unlock(&pr->pr_mtx);
2388 drflags &= ~PD_LOCKED;
2390 /* Let modules do whatever they need to prepare for attaching. */
2391 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2393 prison_deref(pr, drflags);
2396 sx_unlock(&allprison_lock);
2397 drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2400 * Reparent the newly attached process to this jail.
2403 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2407 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2408 if ((error = change_dir(pr->pr_root, td)) != 0)
2411 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2414 VOP_UNLOCK(pr->pr_root);
2415 if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2420 oldcred = crcopysafe(p, newcred);
2421 newcred->cr_prison = pr;
2422 proc_set_cred(p, newcred);
2425 racct_proc_ucred_changed(p, oldcred, newcred);
2430 rctl_proc_ucred_changed(p, newcred);
2433 prison_deref(oldcred->cr_prison, drflags);
2437 * If the prison was killed while changing credentials, die along
2440 if (!prison_isalive(pr)) {
2442 kern_psignal(p, SIGKILL);
2449 VOP_UNLOCK(pr->pr_root);
2451 /* Tell modules this thread is still in its old jail after all. */
2452 sx_slock(&allprison_lock);
2453 drflags |= PD_LIST_SLOCKED;
2454 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2455 prison_deref(pr, drflags);
2460 * Returns a locked prison instance, or NULL on failure.
2463 prison_find(int prid)
2467 sx_assert(&allprison_lock, SX_LOCKED);
2468 TAILQ_FOREACH(pr, &allprison, pr_list) {
2469 if (pr->pr_id < prid)
2471 if (pr->pr_id > prid)
2473 KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2474 mtx_lock(&pr->pr_mtx);
2481 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2484 prison_find_child(struct prison *mypr, int prid)
2489 sx_assert(&allprison_lock, SX_LOCKED);
2490 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2491 if (pr->pr_id == prid) {
2492 KASSERT(prison_isvalid(pr),
2493 ("Found invalid prison %p", pr));
2494 mtx_lock(&pr->pr_mtx);
2502 * Look for the name relative to mypr. Returns a locked prison or NULL.
2505 prison_find_name(struct prison *mypr, const char *name)
2507 struct prison *pr, *deadpr;
2511 sx_assert(&allprison_lock, SX_LOCKED);
2512 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2514 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2515 if (!strcmp(pr->pr_name + mylen, name)) {
2516 KASSERT(prison_isvalid(pr),
2517 ("Found invalid prison %p", pr));
2518 if (prison_isalive(pr)) {
2519 mtx_lock(&pr->pr_mtx);
2525 /* There was no valid prison - perhaps there was a dying one. */
2527 mtx_lock(&deadpr->pr_mtx);
2532 * See if a prison has the specific flag set. The prison should be locked,
2533 * unless checking for flags that are only set at jail creation (such as
2534 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2535 * to any other prison data.
2538 prison_flag(struct ucred *cred, unsigned flag)
2541 return (cred->cr_prison->pr_flags & flag);
2545 prison_allow(struct ucred *cred, unsigned flag)
2548 return ((cred->cr_prison->pr_allow & flag) != 0);
2552 * Hold a prison reference, by incrementing pr_ref. It is generally
2553 * an error to hold a prison that does not already have a reference.
2554 * A prison record will remain valid as long as it has at least one
2555 * reference, and will not be removed as long as either the prison
2556 * mutex or the allprison lock is held (allprison_lock may be shared).
2559 prison_hold_locked(struct prison *pr)
2562 /* Locking is no longer required. */
2567 prison_hold(struct prison *pr)
2570 int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2573 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2575 refcount_acquire(&pr->pr_ref);
2580 * Remove a prison reference. If that was the last reference, the
2581 * prison will be removed (at a later time).
2584 prison_free_locked(struct prison *pr)
2587 mtx_assert(&pr->pr_mtx, MA_OWNED);
2589 * Locking is no longer required, but unlock because the caller
2592 mtx_unlock(&pr->pr_mtx);
2597 prison_free(struct prison *pr)
2600 KASSERT(refcount_load(&pr->pr_ref) > 0,
2601 ("Trying to free dead prison %p (jid=%d).",
2603 if (!refcount_release_if_not_last(&pr->pr_ref)) {
2605 * Don't remove the last reference in this context,
2606 * in case there are locks held.
2608 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2613 prison_free_not_last(struct prison *pr)
2618 KASSERT(refcount_load(&pr->pr_ref) > 0,
2619 ("Trying to free dead prison %p (jid=%d).",
2621 lastref = refcount_release(&pr->pr_ref);
2623 ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2626 refcount_release(&pr->pr_ref);
2631 * Hold a prison for user visibility, by incrementing pr_uref.
2632 * It is generally an error to hold a prison that isn't already
2633 * user-visible, except through the the jail system calls. It is also
2634 * an error to hold an invalid prison. A prison record will remain
2635 * alive as long as it has at least one user reference, and will not
2636 * be set to the dying state until the prison mutex and allprison_lock
2640 prison_proc_hold(struct prison *pr)
2643 int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2646 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2648 refcount_acquire(&pr->pr_uref);
2653 * Remove a prison user reference. If it was the last reference, the
2654 * prison will be considered "dying", and may be removed once all of
2655 * its references are dropped.
2658 prison_proc_free(struct prison *pr)
2662 * Locking is only required when releasing the last reference.
2663 * This allows assurance that a locked prison will remain alive
2664 * until it is unlocked.
2666 KASSERT(refcount_load(&pr->pr_uref) > 0,
2667 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2668 if (!refcount_release_if_not_last(&pr->pr_uref)) {
2670 * Don't remove the last user reference in this context,
2671 * which is expected to be a process that is not only locked,
2672 * but also half dead. Add a reference so any calls to
2673 * prison_free() won't re-submit the task.
2676 mtx_lock(&pr->pr_mtx);
2677 KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2678 ("Redundant last reference in prison_proc_free (jid=%d)",
2680 pr->pr_flags |= PR_COMPLETE_PROC;
2681 mtx_unlock(&pr->pr_mtx);
2682 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2687 prison_proc_free_not_last(struct prison *pr)
2692 KASSERT(refcount_load(&pr->pr_uref) > 0,
2693 ("Trying to free dead prison %p (jid=%d).",
2695 lastref = refcount_release(&pr->pr_uref);
2697 ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2700 refcount_release(&pr->pr_uref);
2705 * Complete a call to either prison_free or prison_proc_free.
2708 prison_complete(void *context, int pending)
2710 struct prison *pr = context;
2714 * This could be called to release the last reference, or the last
2715 * user reference (plus the reference held in prison_proc_free).
2717 drflags = prison_lock_xlock(pr, PD_DEREF);
2718 if (pr->pr_flags & PR_COMPLETE_PROC) {
2719 pr->pr_flags &= ~PR_COMPLETE_PROC;
2720 drflags |= PD_DEUREF;
2722 prison_deref(pr, drflags);
2726 * Remove a prison reference and/or user reference (usually).
2727 * This assumes context that allows sleeping (for allprison_lock),
2728 * with no non-sleeping locks held, except perhaps the prison itself.
2729 * If there are no more references, release and delist the prison.
2730 * On completion, the prison lock and the allprison lock are both
2734 prison_deref(struct prison *pr, int flags)
2736 struct prisonlist freeprison;
2737 struct prison *killpr, *rpr, *ppr, *tpr;
2741 TAILQ_INIT(&freeprison);
2743 * Release this prison as requested, which may cause its parent
2744 * to be released, and then maybe its grandparent, etc.
2747 if (flags & PD_KILL) {
2748 /* Kill the prison and its descendents. */
2749 KASSERT(pr != &prison0,
2750 ("prison_deref trying to kill prison0"));
2751 if (!(flags & PD_DEREF)) {
2755 flags = prison_lock_xlock(pr, flags);
2756 prison_deref_kill(pr, &freeprison);
2758 if (flags & PD_DEUREF) {
2759 /* Drop a user reference. */
2760 KASSERT(refcount_load(&pr->pr_uref) > 0,
2761 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2763 if (!refcount_release_if_not_last(&pr->pr_uref)) {
2764 if (!(flags & PD_DEREF)) {
2768 flags = prison_lock_xlock(pr, flags);
2769 if (refcount_release(&pr->pr_uref) &&
2770 pr->pr_state == PRISON_STATE_ALIVE) {
2772 * When the last user references goes,
2773 * this becomes a dying prison.
2776 refcount_load(&prison0.pr_uref) > 0,
2777 ("prison0 pr_uref=0"));
2778 pr->pr_state = PRISON_STATE_DYING;
2779 mtx_unlock(&pr->pr_mtx);
2780 flags &= ~PD_LOCKED;
2785 if (flags & PD_KILL) {
2787 * Any remaining user references are probably processes
2788 * that need to be killed, either in this prison or its
2791 if (refcount_load(&pr->pr_uref) > 0)
2793 /* Make sure the parent prison doesn't get killed. */
2796 if (flags & PD_DEREF) {
2797 /* Drop a reference. */
2798 KASSERT(refcount_load(&pr->pr_ref) > 0,
2799 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2801 if (!refcount_release_if_not_last(&pr->pr_ref)) {
2802 flags = prison_lock_xlock(pr, flags);
2803 if (refcount_release(&pr->pr_ref)) {
2805 * When the last reference goes,
2806 * unlink the prison and set it aside.
2809 refcount_load(&pr->pr_uref) == 0,
2810 ("prison_deref: last ref, "
2811 "but still has %d urefs (jid=%d)",
2812 pr->pr_uref, pr->pr_id));
2814 refcount_load(&prison0.pr_ref) != 0,
2815 ("prison0 pr_ref=0"));
2816 pr->pr_state = PRISON_STATE_INVALID;
2817 TAILQ_REMOVE(&allprison, pr, pr_list);
2818 LIST_REMOVE(pr, pr_sibling);
2819 TAILQ_INSERT_TAIL(&freeprison, pr,
2821 for (ppr = pr->pr_parent;
2823 ppr = ppr->pr_parent)
2824 ppr->pr_childcount--;
2826 * Removing a prison frees references
2829 mtx_unlock(&pr->pr_mtx);
2830 flags &= ~PD_LOCKED;
2832 flags |= PD_DEREF | PD_DEUREF;
2840 /* Release all the prison locks. */
2841 if (flags & PD_LOCKED)
2842 mtx_unlock(&pr->pr_mtx);
2843 if (flags & PD_LIST_SLOCKED)
2844 sx_sunlock(&allprison_lock);
2845 else if (flags & PD_LIST_XLOCKED)
2846 sx_xunlock(&allprison_lock);
2848 /* Kill any processes attached to a killed prison. */
2849 if (killpr != NULL) {
2850 sx_slock(&allproc_lock);
2851 FOREACH_PROC_IN_SYSTEM(p) {
2853 if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
2854 for (ppr = p->p_ucred->cr_prison;
2856 ppr = ppr->pr_parent)
2857 if (ppr == killpr) {
2858 kern_psignal(p, SIGKILL);
2864 sx_sunlock(&allproc_lock);
2868 * Finish removing any unreferenced prisons, which couldn't happen
2869 * while allprison_lock was held (to avoid a LOR on vrele).
2871 TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
2873 if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
2874 vnet_destroy(rpr->pr_vnet);
2876 if (rpr->pr_root != NULL)
2877 vrele(rpr->pr_root);
2878 mtx_destroy(&rpr->pr_mtx);
2880 free(rpr->pr_ip4, M_PRISON);
2883 free(rpr->pr_ip6, M_PRISON);
2885 if (rpr->pr_cpuset != NULL)
2886 cpuset_rel(rpr->pr_cpuset);
2890 prison_racct_detach(rpr);
2892 TAILQ_REMOVE(&freeprison, rpr, pr_list);
2893 free(rpr, M_PRISON);
2898 * Kill the prison and its descendants. Mark them as dying, clear the
2899 * persist flag, and call module remove methods.
2902 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
2904 struct prison *cpr, *ppr, *rpr;
2908 * Unlike the descendants, the target prison can be killed
2909 * even if it is currently dying. This is useful for failed
2910 * creation in jail_set(2).
2912 KASSERT(refcount_load(&pr->pr_ref) > 0,
2913 ("Trying to kill dead prison %p (jid=%d).",
2915 refcount_acquire(&pr->pr_uref);
2916 pr->pr_state = PRISON_STATE_DYING;
2917 mtx_unlock(&pr->pr_mtx);
2920 FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
2922 if (!prison_isalive(cpr)) {
2927 prison_proc_hold(cpr);
2928 mtx_lock(&cpr->pr_mtx);
2929 cpr->pr_state = PRISON_STATE_DYING;
2930 cpr->pr_flags |= PR_REMOVE;
2931 mtx_unlock(&cpr->pr_mtx);
2934 if (!(cpr->pr_flags & PR_REMOVE))
2936 prison_cleanup(cpr);
2937 mtx_lock(&cpr->pr_mtx);
2938 cpr->pr_flags &= ~PR_REMOVE;
2939 if (cpr->pr_flags & PR_PERSIST) {
2940 cpr->pr_flags &= ~PR_PERSIST;
2941 prison_proc_free_not_last(cpr);
2942 prison_free_not_last(cpr);
2944 (void)refcount_release(&cpr->pr_uref);
2945 if (refcount_release(&cpr->pr_ref)) {
2947 * When the last reference goes, unlink the prison
2948 * and set it aside for prison_deref() to handle.
2949 * Delay unlinking the sibling list to keep the loop
2953 LIST_REMOVE(rpr, pr_sibling);
2955 rpr->pr_state = PRISON_STATE_INVALID;
2956 TAILQ_REMOVE(&allprison, rpr, pr_list);
2957 TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
2959 * Removing a prison frees references from its parent.
2961 ppr = rpr->pr_parent;
2962 prison_proc_free_not_last(ppr);
2963 prison_free_not_last(ppr);
2964 for (; ppr != NULL; ppr = ppr->pr_parent)
2965 ppr->pr_childcount--;
2967 mtx_unlock(&cpr->pr_mtx);
2970 LIST_REMOVE(rpr, pr_sibling);
2973 mtx_lock(&pr->pr_mtx);
2974 if (pr->pr_flags & PR_PERSIST) {
2975 pr->pr_flags &= ~PR_PERSIST;
2976 prison_proc_free_not_last(pr);
2977 prison_free_not_last(pr);
2979 (void)refcount_release(&pr->pr_uref);
2983 * Given the current locking state in the flags, make sure allprison_lock
2984 * is held exclusive, and the prison is locked. Return flags indicating
2988 prison_lock_xlock(struct prison *pr, int flags)
2991 if (!(flags & PD_LIST_XLOCKED)) {
2993 * Get allprison_lock, which may be an upgrade,
2994 * and may require unlocking the prison.
2996 if (flags & PD_LOCKED) {
2997 mtx_unlock(&pr->pr_mtx);
2998 flags &= ~PD_LOCKED;
3000 if (flags & PD_LIST_SLOCKED) {
3001 if (!sx_try_upgrade(&allprison_lock)) {
3002 sx_sunlock(&allprison_lock);
3003 sx_xlock(&allprison_lock);
3005 flags &= ~PD_LIST_SLOCKED;
3007 sx_xlock(&allprison_lock);
3008 flags |= PD_LIST_XLOCKED;
3010 if (!(flags & PD_LOCKED)) {
3011 /* Lock the prison mutex. */
3012 mtx_lock(&pr->pr_mtx);
3019 * Release a prison's resources when it starts dying (when the last user
3020 * reference is dropped, or when it is killed).
3023 prison_cleanup(struct prison *pr)
3025 sx_assert(&allprison_lock, SA_XLOCKED);
3026 mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3027 shm_remove_prison(pr);
3028 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3032 * Set or clear a permission bit in the pr_allow field, passing restrictions
3033 * (cleared permission) down to child jails.
3036 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3040 pr = cred->cr_prison;
3041 sx_slock(&allprison_lock);
3042 mtx_lock(&pr->pr_mtx);
3043 prison_set_allow_locked(pr, flag, enable);
3044 mtx_unlock(&pr->pr_mtx);
3045 sx_sunlock(&allprison_lock);
3049 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3055 pr->pr_allow |= flag;
3057 pr->pr_allow &= ~flag;
3058 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3059 cpr->pr_allow &= ~flag;
3064 * Check if a jail supports the given address family.
3066 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3070 prison_check_af(struct ucred *cred, int af)
3075 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3077 pr = cred->cr_prison;
3079 /* Prisons with their own network stack are not limited. */
3080 if (prison_owns_vnet(cred))
3089 if (pr->pr_flags & PR_IP4)
3091 mtx_lock(&pr->pr_mtx);
3092 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3093 error = EAFNOSUPPORT;
3094 mtx_unlock(&pr->pr_mtx);
3100 if (pr->pr_flags & PR_IP6)
3102 mtx_lock(&pr->pr_mtx);
3103 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3104 error = EAFNOSUPPORT;
3105 mtx_unlock(&pr->pr_mtx);
3113 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3114 error = EAFNOSUPPORT;
3120 * Check if given address belongs to the jail referenced by cred (wrapper to
3121 * prison_check_ip[46]).
3123 * Returns 0 if jail doesn't restrict the address family or if address belongs
3124 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3125 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3128 prison_if(struct ucred *cred, const struct sockaddr *sa)
3131 const struct sockaddr_in *sai;
3134 const struct sockaddr_in6 *sai6;
3138 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3139 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3142 if (prison_owns_vnet(cred))
3147 switch (sa->sa_family)
3151 sai = (const struct sockaddr_in *)sa;
3152 error = prison_check_ip4(cred, &sai->sin_addr);
3157 sai6 = (const struct sockaddr_in6 *)sa;
3158 error = prison_check_ip6(cred, &sai6->sin6_addr);
3162 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3163 error = EAFNOSUPPORT;
3169 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3172 prison_check(struct ucred *cred1, struct ucred *cred2)
3175 return ((cred1->cr_prison == cred2->cr_prison ||
3176 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3180 * Return 1 if p2 is a child of p1, otherwise 0.
3183 prison_ischild(struct prison *pr1, struct prison *pr2)
3186 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3193 * Return true if the prison is currently alive. A prison is alive if it
3194 * holds user references and it isn't being removed.
3197 prison_isalive(struct prison *pr)
3200 if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3206 * Return true if the prison is currently valid. A prison is valid if it has
3207 * been fully created, and is not being destroyed. Note that dying prisons
3208 * are still considered valid. Invalid prisons won't be found under normal
3209 * circumstances, as they're only put in that state by functions that have
3210 * an exclusive hold on allprison_lock.
3213 prison_isvalid(struct prison *pr)
3216 if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3218 if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3224 * Return 1 if the passed credential is in a jail and that jail does not
3225 * have its own virtual network stack, otherwise 0.
3228 jailed_without_vnet(struct ucred *cred)
3234 if (prison_owns_vnet(cred))
3242 * Return the correct hostname (domainname, et al) for the passed credential.
3245 getcredhostname(struct ucred *cred, char *buf, size_t size)
3250 * A NULL credential can be used to shortcut to the physical
3251 * system's hostname.
3253 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3254 mtx_lock(&pr->pr_mtx);
3255 strlcpy(buf, pr->pr_hostname, size);
3256 mtx_unlock(&pr->pr_mtx);
3260 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3263 mtx_lock(&cred->cr_prison->pr_mtx);
3264 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3265 mtx_unlock(&cred->cr_prison->pr_mtx);
3269 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3272 mtx_lock(&cred->cr_prison->pr_mtx);
3273 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3274 mtx_unlock(&cred->cr_prison->pr_mtx);
3278 getcredhostid(struct ucred *cred, unsigned long *hostid)
3281 mtx_lock(&cred->cr_prison->pr_mtx);
3282 *hostid = cred->cr_prison->pr_hostid;
3283 mtx_unlock(&cred->cr_prison->pr_mtx);
3287 getjailname(struct ucred *cred, char *name, size_t len)
3290 mtx_lock(&cred->cr_prison->pr_mtx);
3291 strlcpy(name, cred->cr_prison->pr_name, len);
3292 mtx_unlock(&cred->cr_prison->pr_mtx);
3297 * Determine whether the prison represented by cred owns
3298 * its vnet rather than having it inherited.
3300 * Returns 1 in case the prison owns the vnet, 0 otherwise.
3303 prison_owns_vnet(struct ucred *cred)
3307 * vnets cannot be added/removed after jail creation,
3308 * so no need to lock here.
3310 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3315 * Determine whether the subject represented by cred can "see"
3316 * status of a mount point.
3317 * Returns: 0 for permitted, ENOENT otherwise.
3318 * XXX: This function should be called cr_canseemount() and should be
3319 * placed in kern_prot.c.
3322 prison_canseemount(struct ucred *cred, struct mount *mp)
3328 pr = cred->cr_prison;
3329 if (pr->pr_enforce_statfs == 0)
3331 if (pr->pr_root->v_mount == mp)
3333 if (pr->pr_enforce_statfs == 2)
3336 * If jail's chroot directory is set to "/" we should be able to see
3337 * all mount-points from inside a jail.
3338 * This is ugly check, but this is the only situation when jail's
3339 * directory ends with '/'.
3341 if (strcmp(pr->pr_path, "/") == 0)
3343 len = strlen(pr->pr_path);
3345 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3348 * Be sure that we don't have situation where jail's root directory
3349 * is "/some/path" and mount point is "/some/pathpath".
3351 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3357 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3359 char jpath[MAXPATHLEN];
3363 pr = cred->cr_prison;
3364 if (pr->pr_enforce_statfs == 0)
3366 if (prison_canseemount(cred, mp) != 0) {
3367 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3368 strlcpy(sp->f_mntonname, "[restricted]",
3369 sizeof(sp->f_mntonname));
3372 if (pr->pr_root->v_mount == mp) {
3374 * Clear current buffer data, so we are sure nothing from
3375 * the valid path left there.
3377 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3378 *sp->f_mntonname = '/';
3382 * If jail's chroot directory is set to "/" we should be able to see
3383 * all mount-points from inside a jail.
3385 if (strcmp(pr->pr_path, "/") == 0)
3387 len = strlen(pr->pr_path);
3388 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3390 * Clear current buffer data, so we are sure nothing from
3391 * the valid path left there.
3393 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3394 if (*jpath == '\0') {
3395 /* Should never happen. */
3396 *sp->f_mntonname = '/';
3398 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3403 * Check with permission for a specific privilege is granted within jail. We
3404 * have a specific list of accepted privileges; the rest are denied.
3407 prison_priv_check(struct ucred *cred, int priv)
3413 * Some policies have custom handlers. This routine should not be
3414 * called for them. See priv_check_cred().
3417 case PRIV_VFS_LOOKUP:
3418 case PRIV_VFS_GENERATION:
3419 KASSERT(0, ("prison_priv_check instead of a custom handler "
3420 "called for %d\n", priv));
3428 * Privileges specific to prisons with a virtual network stack.
3429 * There might be a duplicate entry here in case the privilege
3430 * is only granted conditionally in the legacy jail case.
3435 * NFS-specific privileges.
3437 case PRIV_NFS_DAEMON:
3438 case PRIV_NFS_LOCKD:
3441 * Network stack privileges.
3443 case PRIV_NET_BRIDGE:
3446 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3447 case PRIV_NET_ROUTE:
3449 case PRIV_NET_SETIFMTU:
3450 case PRIV_NET_SETIFFLAGS:
3451 case PRIV_NET_SETIFCAP:
3452 case PRIV_NET_SETIFDESCR:
3453 case PRIV_NET_SETIFNAME :
3454 case PRIV_NET_SETIFMETRIC:
3455 case PRIV_NET_SETIFPHYS:
3456 case PRIV_NET_SETIFMAC:
3457 case PRIV_NET_SETLANPCP:
3458 case PRIV_NET_ADDMULTI:
3459 case PRIV_NET_DELMULTI:
3460 case PRIV_NET_HWIOCTL:
3461 case PRIV_NET_SETLLADDR:
3462 case PRIV_NET_ADDIFGROUP:
3463 case PRIV_NET_DELIFGROUP:
3464 case PRIV_NET_IFCREATE:
3465 case PRIV_NET_IFDESTROY:
3466 case PRIV_NET_ADDIFADDR:
3467 case PRIV_NET_DELIFADDR:
3470 case PRIV_NET_SETIFVNET:
3471 case PRIV_NET_SETIFFIB:
3474 * 802.11-related privileges.
3476 case PRIV_NET80211_VAP_GETKEY:
3477 case PRIV_NET80211_VAP_MANAGE:
3483 case PRIV_NETATM_CFG:
3484 case PRIV_NETATM_ADD:
3485 case PRIV_NETATM_DEL:
3486 case PRIV_NETATM_SET:
3489 * Bluetooth privileges.
3491 case PRIV_NETBLUETOOTH_RAW:
3495 * Netgraph and netgraph module privileges.
3497 case PRIV_NETGRAPH_CONTROL:
3499 case PRIV_NETGRAPH_TTY:
3503 * IPv4 and IPv6 privileges.
3505 case PRIV_NETINET_IPFW:
3506 case PRIV_NETINET_DIVERT:
3507 case PRIV_NETINET_PF:
3508 case PRIV_NETINET_DUMMYNET:
3509 case PRIV_NETINET_CARP:
3510 case PRIV_NETINET_MROUTE:
3511 case PRIV_NETINET_RAW:
3512 case PRIV_NETINET_ADDRCTRL6:
3513 case PRIV_NETINET_ND6:
3514 case PRIV_NETINET_SCOPE6:
3515 case PRIV_NETINET_ALIFETIME6:
3516 case PRIV_NETINET_IPSEC:
3517 case PRIV_NETINET_BINDANY:
3532 * No default: or deny here.
3533 * In case of no permit fall through to next switch().
3535 if (cred->cr_prison->pr_flags & PR_VNET)
3542 * Allow ktrace privileges for root in jail.
3548 * Allow jailed processes to configure audit identity and
3549 * submit audit records (login, etc). In the future we may
3550 * want to further refine the relationship between audit and
3553 case PRIV_AUDIT_GETAUDIT:
3554 case PRIV_AUDIT_SETAUDIT:
3555 case PRIV_AUDIT_SUBMIT:
3559 * Allow jailed processes to manipulate process UNIX
3560 * credentials in any way they see fit.
3562 case PRIV_CRED_SETUID:
3563 case PRIV_CRED_SETEUID:
3564 case PRIV_CRED_SETGID:
3565 case PRIV_CRED_SETEGID:
3566 case PRIV_CRED_SETGROUPS:
3567 case PRIV_CRED_SETREUID:
3568 case PRIV_CRED_SETREGID:
3569 case PRIV_CRED_SETRESUID:
3570 case PRIV_CRED_SETRESGID:
3573 * Jail implements visibility constraints already, so allow
3574 * jailed root to override uid/gid-based constraints.
3576 case PRIV_SEEOTHERGIDS:
3577 case PRIV_SEEOTHERUIDS:
3580 * Jail implements inter-process debugging limits already, so
3581 * allow jailed root various debugging privileges.
3583 case PRIV_DEBUG_DIFFCRED:
3584 case PRIV_DEBUG_SUGID:
3585 case PRIV_DEBUG_UNPRIV:
3588 * Allow jail to set various resource limits and login
3589 * properties, and for now, exceed process resource limits.
3591 case PRIV_PROC_LIMIT:
3592 case PRIV_PROC_SETLOGIN:
3593 case PRIV_PROC_SETRLIMIT:
3596 * System V and POSIX IPC privileges are granted in jail.
3599 case PRIV_IPC_WRITE:
3600 case PRIV_IPC_ADMIN:
3601 case PRIV_IPC_MSGSIZE:
3605 * Jail operations within a jail work on child jails.
3607 case PRIV_JAIL_ATTACH:
3609 case PRIV_JAIL_REMOVE:
3612 * Jail implements its own inter-process limits, so allow
3613 * root processes in jail to change scheduling on other
3614 * processes in the same jail. Likewise for signalling.
3616 case PRIV_SCHED_DIFFCRED:
3617 case PRIV_SCHED_CPUSET:
3618 case PRIV_SIGNAL_DIFFCRED:
3619 case PRIV_SIGNAL_SUGID:
3622 * Allow jailed processes to write to sysctls marked as jail
3625 case PRIV_SYSCTL_WRITEJAIL:
3628 * Allow root in jail to manage a variety of quota
3629 * properties. These should likely be conditional on a
3630 * configuration option.
3632 case PRIV_VFS_GETQUOTA:
3633 case PRIV_VFS_SETQUOTA:
3636 * Since Jail relies on chroot() to implement file system
3637 * protections, grant many VFS privileges to root in jail.
3638 * Be careful to exclude mount-related and NFS-related
3642 case PRIV_VFS_WRITE:
3643 case PRIV_VFS_ADMIN:
3645 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3646 case PRIV_VFS_CHFLAGS_DEV:
3647 case PRIV_VFS_CHOWN:
3648 case PRIV_VFS_CHROOT:
3649 case PRIV_VFS_RETAINSUGID:
3650 case PRIV_VFS_FCHROOT:
3652 case PRIV_VFS_SETGID:
3654 case PRIV_VFS_STICKYFILE:
3657 * As in the non-jail case, non-root users are expected to be
3658 * able to read kernel/physical memory (provided /dev/[k]mem
3659 * exists in the jail and they have permission to access it).
3661 case PRIV_KMEM_READ:
3665 * Depending on the global setting, allow privilege of
3666 * setting system flags.
3668 case PRIV_VFS_SYSFLAGS:
3669 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3675 * Depending on the global setting, allow privilege of
3676 * mounting/unmounting file systems.
3678 case PRIV_VFS_MOUNT:
3679 case PRIV_VFS_UNMOUNT:
3680 case PRIV_VFS_MOUNT_NONUSER:
3681 case PRIV_VFS_MOUNT_OWNER:
3682 pr = cred->cr_prison;
3684 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
3692 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3693 * policy. priv_check_cred will not specifically allow it, and
3694 * we may want a MAC policy to allow it.
3696 case PRIV_VFS_READ_DIR:
3700 * Conditionnaly allow locking (unlocking) physical pages
3704 case PRIV_VM_MUNLOCK:
3705 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3711 * Conditionally allow jailed root to bind reserved ports.
3713 case PRIV_NETINET_RESERVEDPORT:
3714 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3720 * Allow jailed root to reuse in-use ports.
3722 case PRIV_NETINET_REUSEPORT:
3726 * Allow jailed root to set certain IPv4/6 (option) headers.
3728 case PRIV_NETINET_SETHDROPTS:
3732 * Conditionally allow creating raw sockets in jail.
3734 case PRIV_NETINET_RAW:
3735 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3741 * Since jail implements its own visibility limits on netstat
3742 * sysctls, allow getcred. This allows identd to work in
3745 case PRIV_NETINET_GETCRED:
3749 * Allow jailed root to set loginclass.
3751 case PRIV_PROC_SETLOGINCLASS:
3755 * Do not allow a process inside a jail to read the kernel
3756 * message buffer unless explicitly permitted.
3759 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3765 * In all remaining cases, deny the privilege request. This
3766 * includes almost all network privileges, many system
3767 * configuration privileges.
3774 * Return the part of pr2's name that is relative to pr1, or the whole name
3775 * if it does not directly follow.
3779 prison_name(struct prison *pr1, struct prison *pr2)
3783 /* Jails see themselves as "0" (if they see themselves at all). */
3786 name = pr2->pr_name;
3787 if (prison_ischild(pr1, pr2)) {
3789 * pr1 isn't locked (and allprison_lock may not be either)
3790 * so its length can't be counted on. But the number of dots
3791 * can be counted on - and counted.
3793 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3794 name = strchr(name, '.') + 1;
3800 * Return the part of pr2's path that is relative to pr1, or the whole path
3801 * if it does not directly follow.
3804 prison_path(struct prison *pr1, struct prison *pr2)
3806 char *path1, *path2;
3809 path1 = pr1->pr_path;
3810 path2 = pr2->pr_path;
3811 if (!strcmp(path1, "/"))
3813 len1 = strlen(path1);
3814 if (strncmp(path1, path2, len1))
3816 if (path2[len1] == '\0')
3818 if (path2[len1] == '/')
3819 return (path2 + len1);
3824 * Jail-related sysctls.
3826 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3830 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3833 struct prison *pr, *cpr;
3835 struct in_addr *ip4 = NULL;
3839 struct in6_addr *ip6 = NULL;
3844 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3845 pr = req->td->td_ucred->cr_prison;
3847 sx_slock(&allprison_lock);
3848 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3849 #if defined(INET) || defined(INET6)
3852 mtx_lock(&cpr->pr_mtx);
3854 if (cpr->pr_ip4s > 0) {
3855 if (ip4s < cpr->pr_ip4s) {
3856 ip4s = cpr->pr_ip4s;
3857 mtx_unlock(&cpr->pr_mtx);
3858 ip4 = realloc(ip4, ip4s *
3859 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3862 bcopy(cpr->pr_ip4, ip4,
3863 cpr->pr_ip4s * sizeof(struct in_addr));
3867 if (cpr->pr_ip6s > 0) {
3868 if (ip6s < cpr->pr_ip6s) {
3869 ip6s = cpr->pr_ip6s;
3870 mtx_unlock(&cpr->pr_mtx);
3871 ip6 = realloc(ip6, ip6s *
3872 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3875 bcopy(cpr->pr_ip6, ip6,
3876 cpr->pr_ip6s * sizeof(struct in6_addr));
3879 bzero(xp, sizeof(*xp));
3880 xp->pr_version = XPRISON_VERSION;
3881 xp->pr_id = cpr->pr_id;
3882 xp->pr_state = cpr->pr_state;
3883 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3884 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3885 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3887 xp->pr_ip4s = cpr->pr_ip4s;
3890 xp->pr_ip6s = cpr->pr_ip6s;
3892 mtx_unlock(&cpr->pr_mtx);
3893 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3897 if (xp->pr_ip4s > 0) {
3898 error = SYSCTL_OUT(req, ip4,
3899 xp->pr_ip4s * sizeof(struct in_addr));
3905 if (xp->pr_ip6s > 0) {
3906 error = SYSCTL_OUT(req, ip6,
3907 xp->pr_ip6s * sizeof(struct in6_addr));
3913 sx_sunlock(&allprison_lock);
3924 SYSCTL_OID(_security_jail, OID_AUTO, list,
3925 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3926 sysctl_jail_list, "S", "List of active jails");
3929 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3933 injail = jailed(req->td->td_ucred);
3934 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3939 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3940 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3941 sysctl_jail_jailed, "I", "Process in jail?");
3944 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3946 int error, havevnet;
3948 struct ucred *cred = req->td->td_ucred;
3950 havevnet = jailed(cred) && prison_owns_vnet(cred);
3954 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3959 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3960 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3961 sysctl_jail_vnet, "I", "Jail owns vnet?");
3963 #if defined(INET) || defined(INET6)
3964 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3965 &jail_max_af_ips, 0,
3966 "Number of IP addresses a jail may have at most per address family (deprecated)");
3970 * Default parameters for jail(2) compatibility. For historical reasons,
3971 * the sysctl names have varying similarity to the parameter names. Prisons
3972 * just see their own parameters, and can't change them.
3975 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3979 /* Get the current flag value, and convert it to a boolean. */
3980 if (req->td->td_ucred->cr_prison == &prison0) {
3981 mtx_lock(&prison0.pr_mtx);
3982 i = (jail_default_allow & arg2) != 0;
3983 mtx_unlock(&prison0.pr_mtx);
3985 i = prison_allow(req->td->td_ucred, arg2);
3989 error = sysctl_handle_int(oidp, &i, 0, req);
3990 if (error || !req->newptr)
3996 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3999 mtx_lock(&prison0.pr_mtx);
4000 jail_default_allow = (jail_default_allow & ~arg2) | i;
4001 mtx_unlock(&prison0.pr_mtx);
4005 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4006 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4007 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4008 "Processes in jail can set their hostnames (deprecated)");
4009 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4010 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4011 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4012 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4013 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4014 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4015 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4016 "Processes in jail can use System V IPC primitives (deprecated)");
4017 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4018 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4019 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4020 "Prison root can create raw sockets (deprecated)");
4021 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4022 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4023 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4024 "Processes in jail can alter system file flags (deprecated)");
4025 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4026 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4027 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4028 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4031 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4036 pr = req->td->td_ucred->cr_prison;
4037 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4038 error = sysctl_handle_int(oidp, &level, 0, req);
4039 if (error || !req->newptr)
4041 *(int *)arg1 = level;
4045 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4046 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4047 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4048 sysctl_jail_default_level, "I",
4049 "Processes in jail cannot see all mounted file systems (deprecated)");
4051 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4052 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4053 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4054 sysctl_jail_default_level, "I",
4055 "Ruleset for the devfs filesystem in jail (deprecated)");
4058 * Nodes to describe jail parameters. Maximum length of string parameters
4059 * is returned in the string itself, and the other parameters exist merely
4060 * to make themselves and their types known.
4062 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4066 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4073 switch (oidp->oid_kind & CTLTYPE)
4079 if (!(req->flags & SCTL_MASK32))
4081 return (SYSCTL_OUT(req, &l, sizeof(l)));
4085 return (SYSCTL_OUT(req, &i, sizeof(i)));
4086 case CTLTYPE_STRING:
4087 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4089 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4090 case CTLTYPE_STRUCT:
4092 return (SYSCTL_OUT(req, &s, sizeof(s)));
4098 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4099 * jail creation time but cannot be changed in an existing jail.
4101 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4102 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4103 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4104 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4105 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4106 "I", "Jail secure level");
4107 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4108 "Jail value for kern.osreldate and uname -K");
4109 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4110 "Jail value for kern.osrelease and uname -r");
4111 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4112 "I", "Jail cannot see all mounted file systems");
4113 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4114 "I", "Ruleset for in-jail devfs mounts");
4115 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4116 "B", "Jail persistence");
4118 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4119 "E,jailsys", "Virtual network stack");
4121 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4122 "B", "Jail is in the process of shutting down");
4124 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4125 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4126 "I", "Current number of child jails");
4127 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4128 "I", "Maximum number of child jails");
4130 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4131 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4133 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4134 "Jail NIS domainname");
4135 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4137 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4138 "LU", "Jail host ID");
4140 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4141 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4144 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4145 "Jail IPv4 address virtualization");
4146 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4147 "S,in_addr,a", "Jail IPv4 addresses");
4148 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4149 "B", "Do (not) use IPv4 source address selection rather than the "
4150 "primary jail IPv4 address.");
4153 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4154 "Jail IPv6 address virtualization");
4155 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4156 "S,in6_addr,a", "Jail IPv6 addresses");
4157 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4158 "B", "Do (not) use IPv6 source address selection rather than the "
4159 "primary jail IPv6 address.");
4162 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4163 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4164 "B", "Jail may set hostname");
4165 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4166 "B", "Jail may use SYSV IPC");
4167 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4168 "B", "Jail may create raw sockets");
4169 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4170 "B", "Jail may alter system file flags");
4171 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4172 "B", "Jail may set file quotas");
4173 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4174 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4175 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4176 "B", "Jail may lock (unlock) physical pages in memory");
4177 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4178 "B", "Jail may bind sockets to reserved ports");
4179 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4180 "B", "Jail may read the kernel message buffer");
4181 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4182 "B", "Unprivileged processes may use process debugging facilities");
4183 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4184 "B", "Processes in jail with uid 0 have privilege");
4186 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4187 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4188 "B", "Jail may mount/unmount jail-friendly file systems in general");
4191 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
4192 * its associated bit in the pr_allow bitmask, or zero if the parameter was
4196 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4199 struct bool_flags *bf;
4200 struct sysctl_oid *parent;
4201 char *allow_name, *allow_noname, *allowed;
4202 #ifndef NO_SYSCTL_DESCR
4203 char *descr_deprecated;
4208 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4210 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4212 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4213 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4214 free(allow_name, M_PRISON);
4219 * See if this parameter has already beed added, i.e. a module was
4220 * previously loaded/unloaded.
4222 mtx_lock(&prison0.pr_mtx);
4223 for (bf = pr_flag_allow;
4224 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4225 atomic_load_int(&bf->flag) != 0;
4227 if (strcmp(bf->name, allow_name) == 0) {
4228 allow_flag = bf->flag;
4234 * Find a free bit in pr_allow_all, failing if there are none
4235 * (which shouldn't happen as long as we keep track of how many
4236 * potential dynamic flags exist).
4238 for (allow_flag = 1;; allow_flag <<= 1) {
4239 if (allow_flag == 0)
4241 if ((pr_allow_all & allow_flag) == 0)
4245 /* Note the parameter in the next open slot in pr_flag_allow. */
4246 for (bf = pr_flag_allow; ; bf++) {
4247 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4248 /* This should never happen, but is not fatal. */
4252 if (atomic_load_int(&bf->flag) == 0)
4255 bf->name = allow_name;
4256 bf->noname = allow_noname;
4257 pr_allow_all |= allow_flag;
4259 * prison0 always has permission for the new parameter.
4260 * Other jails must have it granted to them.
4262 prison0.pr_allow |= allow_flag;
4263 /* The flag indicates a valid entry, so make sure it is set last. */
4264 atomic_store_rel_int(&bf->flag, allow_flag);
4265 mtx_unlock(&prison0.pr_mtx);
4268 * Create sysctls for the paramter, and the back-compat global
4272 ? SYSCTL_ADD_NODE(NULL,
4273 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4274 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4275 : &sysctl___security_jail_param_allow;
4276 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4277 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4278 NULL, 0, sysctl_jail_param, "B", descr);
4280 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4281 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4282 #ifndef NO_SYSCTL_DESCR
4283 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4286 (void)SYSCTL_ADD_PROC(NULL,
4287 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4288 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4289 sysctl_jail_default_allow, "I", descr_deprecated);
4290 #ifndef NO_SYSCTL_DESCR
4291 free(descr_deprecated, M_TEMP);
4293 free(allowed, M_TEMP);
4298 mtx_unlock(&prison0.pr_mtx);
4299 free(allow_name, M_PRISON);
4300 free(allow_noname, M_PRISON);
4305 * The VFS system will register jail-aware filesystems here. They each get
4306 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4307 * attempts to mount.
4310 prison_add_vfs(struct vfsconf *vfsp)
4312 #ifdef NO_SYSCTL_DESCR
4314 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4319 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4321 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4323 free(descr, M_TEMP);
4329 prison_racct_foreach(void (*callback)(struct racct *racct,
4330 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4331 void *arg2, void *arg3)
4333 struct prison_racct *prr;
4335 ASSERT_RACCT_ENABLED();
4337 sx_slock(&allprison_lock);
4340 LIST_FOREACH(prr, &allprison_racct, prr_next)
4341 (callback)(prr->prr_racct, arg2, arg3);
4344 sx_sunlock(&allprison_lock);
4347 static struct prison_racct *
4348 prison_racct_find_locked(const char *name)
4350 struct prison_racct *prr;
4352 ASSERT_RACCT_ENABLED();
4353 sx_assert(&allprison_lock, SA_XLOCKED);
4355 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4358 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4359 if (strcmp(name, prr->prr_name) != 0)
4362 /* Found prison_racct with a matching name? */
4363 prison_racct_hold(prr);
4367 /* Add new prison_racct. */
4368 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4369 racct_create(&prr->prr_racct);
4371 strcpy(prr->prr_name, name);
4372 refcount_init(&prr->prr_refcount, 1);
4373 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4378 struct prison_racct *
4379 prison_racct_find(const char *name)
4381 struct prison_racct *prr;
4383 ASSERT_RACCT_ENABLED();
4385 sx_xlock(&allprison_lock);
4386 prr = prison_racct_find_locked(name);
4387 sx_xunlock(&allprison_lock);
4392 prison_racct_hold(struct prison_racct *prr)
4395 ASSERT_RACCT_ENABLED();
4397 refcount_acquire(&prr->prr_refcount);
4401 prison_racct_free_locked(struct prison_racct *prr)
4404 ASSERT_RACCT_ENABLED();
4405 sx_assert(&allprison_lock, SA_XLOCKED);
4407 if (refcount_release(&prr->prr_refcount)) {
4408 racct_destroy(&prr->prr_racct);
4409 LIST_REMOVE(prr, prr_next);
4410 free(prr, M_PRISON_RACCT);
4415 prison_racct_free(struct prison_racct *prr)
4418 ASSERT_RACCT_ENABLED();
4419 sx_assert(&allprison_lock, SA_UNLOCKED);
4421 if (refcount_release_if_not_last(&prr->prr_refcount))
4424 sx_xlock(&allprison_lock);
4425 prison_racct_free_locked(prr);
4426 sx_xunlock(&allprison_lock);
4430 prison_racct_attach(struct prison *pr)
4432 struct prison_racct *prr;
4434 ASSERT_RACCT_ENABLED();
4435 sx_assert(&allprison_lock, SA_XLOCKED);
4437 prr = prison_racct_find_locked(pr->pr_name);
4438 KASSERT(prr != NULL, ("cannot find prison_racct"));
4440 pr->pr_prison_racct = prr;
4444 * Handle jail renaming. From the racct point of view, renaming means
4445 * moving from one prison_racct to another.
4448 prison_racct_modify(struct prison *pr)
4454 struct prison_racct *oldprr;
4456 ASSERT_RACCT_ENABLED();
4458 sx_slock(&allproc_lock);
4459 sx_xlock(&allprison_lock);
4461 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4462 sx_xunlock(&allprison_lock);
4463 sx_sunlock(&allproc_lock);
4467 oldprr = pr->pr_prison_racct;
4468 pr->pr_prison_racct = NULL;
4470 prison_racct_attach(pr);
4473 * Move resource utilisation records.
4475 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4479 * Force rctl to reattach rules to processes.
4481 FOREACH_PROC_IN_SYSTEM(p) {
4483 cred = crhold(p->p_ucred);
4485 rctl_proc_ucred_changed(p, cred);
4490 sx_sunlock(&allproc_lock);
4491 prison_racct_free_locked(oldprr);
4492 sx_xunlock(&allprison_lock);
4496 prison_racct_detach(struct prison *pr)
4499 ASSERT_RACCT_ENABLED();
4500 sx_assert(&allprison_lock, SA_UNLOCKED);
4502 if (pr->pr_prison_racct == NULL)
4504 prison_racct_free(pr->pr_prison_racct);
4505 pr->pr_prison_racct = NULL;
4512 db_show_prison(struct prison *pr)
4514 struct bool_flags *bf;
4515 struct jailsys_flags *jsf;
4516 #if defined(INET) || defined(INET6)
4521 char ip4buf[INET_ADDRSTRLEN];
4524 char ip6buf[INET6_ADDRSTRLEN];
4527 db_printf("prison %p:\n", pr);
4528 db_printf(" jid = %d\n", pr->pr_id);
4529 db_printf(" name = %s\n", pr->pr_name);
4530 db_printf(" parent = %p\n", pr->pr_parent);
4531 db_printf(" ref = %d\n", pr->pr_ref);
4532 db_printf(" uref = %d\n", pr->pr_uref);
4533 db_printf(" state = %s\n",
4534 pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4535 pr->pr_state == PRISON_STATE_DYING ? "dying" :
4537 db_printf(" path = %s\n", pr->pr_path);
4538 db_printf(" cpuset = %d\n", pr->pr_cpuset
4539 ? pr->pr_cpuset->cs_id : -1);
4541 db_printf(" vnet = %p\n", pr->pr_vnet);
4543 db_printf(" root = %p\n", pr->pr_root);
4544 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4545 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4546 db_printf(" children.max = %d\n", pr->pr_childmax);
4547 db_printf(" children.cur = %d\n", pr->pr_childcount);
4548 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4549 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4550 db_printf(" flags = 0x%x", pr->pr_flags);
4551 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4552 if (pr->pr_flags & bf->flag)
4553 db_printf(" %s", bf->name);
4554 for (jsf = pr_flag_jailsys;
4555 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4557 f = pr->pr_flags & (jsf->disable | jsf->new);
4558 db_printf(" %-16s= %s\n", jsf->name,
4559 (f != 0 && f == jsf->disable) ? "disable"
4560 : (f == jsf->new) ? "new"
4563 db_printf(" allow = 0x%x", pr->pr_allow);
4564 for (bf = pr_flag_allow;
4565 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4566 atomic_load_int(&bf->flag) != 0;
4568 if (pr->pr_allow & bf->flag)
4569 db_printf(" %s", bf->name);
4571 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4572 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4573 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4574 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4575 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4577 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4578 for (ii = 0; ii < pr->pr_ip4s; ii++)
4579 db_printf(" %s %s\n",
4580 ii == 0 ? "ip4.addr =" : " ",
4581 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4584 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4585 for (ii = 0; ii < pr->pr_ip6s; ii++)
4586 db_printf(" %s %s\n",
4587 ii == 0 ? "ip6.addr =" : " ",
4588 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4592 DB_SHOW_COMMAND(prison, db_show_prison_command)
4598 * Show all prisons in the list, and prison0 which is not
4601 db_show_prison(&prison0);
4602 if (!db_pager_quit) {
4603 TAILQ_FOREACH(pr, &allprison, pr_list) {
4615 /* Look for a prison with the ID and with references. */
4616 TAILQ_FOREACH(pr, &allprison, pr_list)
4617 if (pr->pr_id == addr && pr->pr_ref > 0)
4620 /* Look again, without requiring a reference. */
4621 TAILQ_FOREACH(pr, &allprison, pr_list)
4622 if (pr->pr_id == addr)
4625 /* Assume address points to a valid prison. */
4626 pr = (struct prison *)addr;