2 * SPDX-License-Identifier: BSD-2-Clause
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>
34 #include "opt_inet6.h"
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
47 #include <sys/epoch.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;
141 static int get_next_prid(struct prison **insprp);
142 static int get_next_deadid(struct prison **insprp);
143 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
144 static void prison_complete(void *context, int pending);
145 static void prison_deref(struct prison *pr, int flags);
146 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
147 static int prison_lock_xlock(struct prison *pr, int flags);
148 static void prison_cleanup(struct prison *pr);
149 static void prison_free_not_last(struct prison *pr);
150 static void prison_proc_free_not_last(struct prison *pr);
151 static void prison_proc_relink(struct prison *opr, struct prison *npr,
153 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
155 static char *prison_path(struct prison *pr1, struct prison *pr2);
157 static void prison_racct_attach(struct prison *pr);
158 static void prison_racct_modify(struct prison *pr);
159 static void prison_racct_detach(struct prison *pr);
162 /* Flags for prison_deref */
163 #define PD_DEREF 0x01 /* Decrement pr_ref */
164 #define PD_DEUREF 0x02 /* Decrement pr_uref */
165 #define PD_KILL 0x04 /* Remove jail, kill processes, etc */
166 #define PD_LOCKED 0x10 /* pr_mtx is held */
167 #define PD_LIST_SLOCKED 0x20 /* allprison_lock is held shared */
168 #define PD_LIST_XLOCKED 0x40 /* allprison_lock is held exclusive */
169 #define PD_OP_FLAGS 0x07 /* Operation flags */
170 #define PD_LOCK_FLAGS 0x70 /* Lock status flags */
173 * Parameter names corresponding to PR_* flag values. Size values are for kvm
174 * as we cannot figure out the size of a sparse array, or an array without a
177 static struct bool_flags pr_flag_bool[] = {
178 {"persist", "nopersist", PR_PERSIST},
180 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
183 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
186 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
188 static struct jailsys_flags pr_flag_jailsys[] = {
189 {"host", 0, PR_HOST},
191 {"vnet", 0, PR_VNET},
194 {"ip4", PR_IP4_USER, PR_IP4_USER},
197 {"ip6", PR_IP6_USER, PR_IP6_USER},
200 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
203 * Make this array full-size so dynamic parameters can be added.
204 * It is protected by prison0.mtx, but lockless reading is allowed
205 * with an atomic check of the flag values.
207 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
208 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
209 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
210 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
211 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
212 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
213 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
214 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
215 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
216 {"allow.reserved_ports", "allow.noreserved_ports",
217 PR_ALLOW_RESERVED_PORTS},
218 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
219 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
220 PR_ALLOW_UNPRIV_DEBUG},
221 {"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
223 {"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
225 {"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
227 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
228 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
230 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
231 PR_ALLOW_RESERVED_PORTS | \
232 PR_ALLOW_UNPRIV_DEBUG | \
234 #define JAIL_DEFAULT_ENFORCE_STATFS 2
235 #define JAIL_DEFAULT_DEVFS_RSNUM 0
236 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
237 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
238 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
239 #if defined(INET) || defined(INET6)
240 static unsigned jail_max_af_ips = 255;
244 * Initialize the parts of prison0 that can't be static-initialized with
245 * constants. This is called from proc0_init() after creating thread0 cpuset.
250 uint8_t *file, *data;
252 char buf[sizeof(prison0.pr_hostuuid)];
255 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
256 prison0.pr_osreldate = osreldate;
257 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
259 /* If we have a preloaded hostuuid, use it. */
260 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
262 data = preload_fetch_addr(file);
263 size = preload_fetch_size(file);
266 * The preloaded data may include trailing whitespace, almost
267 * certainly a newline; skip over any whitespace or
268 * non-printable characters to be safe.
270 while (size > 0 && data[size - 1] <= 0x20) {
277 * Not NUL-terminated when passed from loader, but
278 * validate_uuid requires that due to using sscanf (as
279 * does the subsequent strlcpy, since it still reads
280 * past the given size to return the true length);
281 * bounce to a temporary buffer to fix.
283 if (size >= sizeof(buf))
286 memcpy(buf, data, size);
289 if (validate_uuid(buf, size, NULL, 0) != 0)
293 (void)strlcpy(prison0.pr_hostuuid, buf,
294 sizeof(prison0.pr_hostuuid));
297 if (bootverbose && !valid) {
298 printf("hostuuid: preload data malformed: '%.*s'\n",
304 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
313 sys_jail(struct thread *td, struct jail_args *uap)
319 error = copyin(uap->jail, &version, sizeof(uint32_t));
328 /* FreeBSD single IPv4 jails. */
329 bzero(&j, sizeof(struct jail));
330 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
333 j.version = j0.version;
335 j.hostname = j0.hostname;
336 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
342 * Version 1 was used by multi-IPv4 jail implementations
343 * that never made it into the official kernel.
347 case 2: /* JAIL_API_VERSION */
348 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
349 error = copyin(uap->jail, &j, sizeof(struct jail));
355 /* Sci-Fi jails are not supported, sorry. */
358 return (kern_jail(td, &j));
362 kern_jail(struct thread *td, struct jail *j)
364 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
373 char *u_path, *u_hostname, *u_name;
374 struct bool_flags *bf;
377 struct in_addr *u_ip4;
380 struct in6_addr *u_ip6;
383 int error, enforce_statfs;
385 bzero(&optiov, sizeof(optiov));
386 opt.uio_iov = optiov;
390 opt.uio_segflg = UIO_SYSSPACE;
391 opt.uio_rw = UIO_READ;
394 /* Set permissions for top-level jails from sysctls. */
395 if (!jailed(td->td_ucred)) {
396 for (bf = pr_flag_allow;
397 bf < pr_flag_allow + nitems(pr_flag_allow) &&
398 atomic_load_int(&bf->flag) != 0;
400 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
401 (jail_default_allow & bf->flag)
402 ? bf->name : bf->noname);
403 optiov[opt.uio_iovcnt].iov_len =
404 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
407 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
408 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
410 enforce_statfs = jail_default_enforce_statfs;
411 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
412 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
416 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
418 ip4s = (j->version == 0) ? 1 : j->ip4s;
419 if (ip4s > jail_max_af_ips)
421 tmplen += ip4s * sizeof(struct in_addr);
427 if (j->ip6s > jail_max_af_ips)
429 tmplen += j->ip6s * sizeof(struct in6_addr);
434 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
435 u_hostname = u_path + MAXPATHLEN;
436 u_name = u_hostname + MAXHOSTNAMELEN;
438 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
442 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
444 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
447 optiov[opt.uio_iovcnt].iov_base = "path";
448 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
450 optiov[opt.uio_iovcnt].iov_base = u_path;
451 error = copyinstr(j->path, u_path, MAXPATHLEN,
452 &optiov[opt.uio_iovcnt].iov_len);
454 free(u_path, M_TEMP);
458 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
459 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
461 optiov[opt.uio_iovcnt].iov_base = u_hostname;
462 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
463 &optiov[opt.uio_iovcnt].iov_len);
465 free(u_path, M_TEMP);
469 if (j->jailname != NULL) {
470 optiov[opt.uio_iovcnt].iov_base = "name";
471 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
473 optiov[opt.uio_iovcnt].iov_base = u_name;
474 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
475 &optiov[opt.uio_iovcnt].iov_len);
477 free(u_path, M_TEMP);
483 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
484 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
486 optiov[opt.uio_iovcnt].iov_base = u_ip4;
487 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
489 u_ip4->s_addr = j->ip4s;
491 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
493 free(u_path, M_TEMP);
500 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
501 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
503 optiov[opt.uio_iovcnt].iov_base = u_ip6;
504 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
505 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
507 free(u_path, M_TEMP);
512 KASSERT(opt.uio_iovcnt <= nitems(optiov),
513 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
514 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
515 free(u_path, M_TEMP);
520 * struct jail_set_args {
521 * struct iovec *iovp;
522 * unsigned int iovcnt;
527 sys_jail_set(struct thread *td, struct jail_set_args *uap)
532 /* Check that we have an even number of iovecs. */
536 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
539 error = kern_jail_set(td, auio, uap->flags);
544 #if defined(INET) || defined(INET6)
545 typedef int prison_addr_cmp_t(const void *, const void *);
546 typedef bool prison_addr_valid_t(const void *);
547 static const struct pr_family {
549 prison_addr_cmp_t *cmp;
550 prison_addr_valid_t *valid;
552 } pr_families[PR_FAMILY_MAX] = {
555 .size = sizeof(struct in_addr),
556 .cmp = prison_qcmp_v4,
557 .valid = prison_valid_v4,
558 .ip_flag = PR_IP4_USER,
563 .size = sizeof(struct in6_addr),
564 .cmp = prison_qcmp_v6,
565 .valid = prison_valid_v6,
566 .ip_flag = PR_IP6_USER,
572 * Network address lists (pr_addrs) allocation for jails. The addresses
573 * are accessed locklessly by the network stack, thus need to be protected by
577 struct epoch_context ctx;
581 * XXX Variable-length automatic arrays in union may be
582 * supported in future C.
586 struct in_addr pr_ip4[];
587 struct in6_addr pr_ip6[];
589 #else /* No future C :( */
595 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
598 MPASS(af < PR_FAMILY_MAX);
599 MPASS(idx >= 0 && idx < pip->ips);
601 return (pip->pr_ip + pr_families[af].size * idx);
604 static struct prison_ip *
605 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
607 struct prison_ip *pip;
609 pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
617 * Allocate and copyin user supplied address list, sorting and validating.
618 * kern_jail_set() helper.
620 static struct prison_ip *
621 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
623 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
624 const size_t size = pr_families[af].size;
625 struct prison_ip *pip;
627 pip = prison_ip_alloc(af, cnt, M_WAITOK);
628 bcopy(op, pip->pr_ip, cnt * size);
630 * IP addresses are all sorted but ip[0] to preserve
631 * the primary IP address as given from userland.
632 * This special IP is used for unbound outgoing
633 * connections as well for "loopback" traffic in case
634 * source address selection cannot find any more fitting
635 * address to connect from.
638 qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
640 * Check for duplicate addresses and do some simple
641 * zero and broadcast checks. If users give other bogus
642 * addresses it is their problem.
644 for (int i = 0; i < cnt; i++) {
645 if (!pr_families[af].valid(PR_IP(pip, af, i))) {
650 (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
651 cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
661 * Allocate and dup parent prison address list.
662 * kern_jail_set() helper.
665 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
667 const struct prison_ip *ppip = ppr->pr_addrs[af];
668 struct prison_ip *pip;
671 pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
672 bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
673 pr->pr_addrs[af] = pip;
678 * Make sure the new set of IP addresses is a subset of the parent's list.
679 * Don't worry about the parent being unlocked, as any setting is done with
680 * allprison_lock held.
681 * kern_jail_set() helper.
684 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
685 const pr_family_t af)
687 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
693 for (i = 0; i < ppip->ips; i++)
694 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
698 /* Main address not present in parent. */
702 for (i = j = 1; i < pip->ips; i++) {
703 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
704 /* Equals to parent primary address. */
706 for (; j < ppip->ips; j++)
707 if (cmp(PR_IP(pip, af, i),
708 PR_IP(ppip, af, j)) == 0)
714 /* Address not present in parent. */
721 * Check for conflicting IP addresses. We permit them if there is no more
722 * than one IP on each jail. If there is a duplicate on a jail with more
723 * than one IP stop checking and return error.
724 * kern_jail_set() helper.
727 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
728 struct prison_ip *pip, pr_family_t af)
730 const struct prison *tppr, *tpr;
734 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
735 if (tppr->pr_flags & PR_VNET)
740 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
743 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
745 !prison_isalive(tpr)) {
749 if (!(tpr->pr_flags & pr_families[af].ip_flag))
752 if (tpr->pr_addrs[af] == NULL ||
753 (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
755 for (int i = 0; i < pip->ips; i++)
756 if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
763 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
764 "prison must start with epoch context");
766 prison_ip_free_deferred(epoch_context_t ctx)
773 prison_ip_free(struct prison_ip *pip)
777 NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
781 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
783 struct prison_ip **mem, *old;
785 mtx_assert(&pr->pr_mtx, MA_OWNED);
787 mem = &pr->pr_addrs[af];
790 atomic_store_ptr(mem, new);
795 * Restrict a prison's IP address list with its parent's, possibly replacing
796 * it. Return true if succeed, otherwise should redo.
797 * kern_jail_set() helper.
800 prison_ip_restrict(struct prison *pr, const pr_family_t af,
801 struct prison_ip **newp)
803 struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
804 struct prison_ip *pip = pr->pr_addrs[af];
805 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
806 const size_t size = pr_families[af].size;
807 struct prison_ip *new = newp != NULL ? *newp : NULL;
810 mtx_assert(&pr->pr_mtx, MA_OWNED);
813 * Due to epoch-synchronized access to the IP address lists we always
814 * allocate a new list even if the old one has enough space. We could
815 * atomically update an IPv4 address inside a list, but that would
816 * screw up sorting, and in case of IPv6 we can't even atomically write
821 prison_ip_set(pr, af, NULL);
825 if (!(pr->pr_flags & pr_families[af].ip_flag)) {
827 new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
829 return (false); /* Redo */
831 /* This has no user settings, so just copy the parent's list. */
832 MPASS(new->ips == ppip->ips);
833 bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
834 prison_ip_set(pr, af, new);
836 *newp = NULL; /* Used */
837 } else if (pip != NULL) {
838 /* Remove addresses that aren't in the parent. */
841 i = 0; /* index in pip */
842 ips = 0; /* index in new */
845 new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
847 return (false); /* Redo */
850 for (int pi = 0; pi < ppip->ips; pi++)
851 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
852 /* Found our primary address in parent. */
853 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
859 for (int pi = 1; i < pip->ips; ) {
860 /* Check against primary, which is unsorted. */
861 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
862 /* Matches parent's primary address. */
863 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
869 /* The rest are sorted. */
870 switch (pi >= ppip->ips ? -1 :
871 cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
876 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
888 if (newp == NULL || *newp == NULL)
892 /* Shrink to real size */
893 KASSERT((new->ips >= ips),
894 ("Out-of-bounds write to prison_ip %p", new));
897 prison_ip_set(pr, af, new);
899 *newp = NULL; /* Used */
905 * Fast-path check if an address belongs to a prison.
908 prison_ip_check(const struct prison *pr, const pr_family_t af,
911 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
912 struct prison_ip *pip;
915 MPASS(mtx_owned(&pr->pr_mtx) ||
916 in_epoch(net_epoch_preempt) ||
917 sx_xlocked(&allprison_lock));
919 pip = atomic_load_ptr(&pr->pr_addrs[af]);
920 if (__predict_false(pip == NULL))
921 return (EAFNOSUPPORT);
923 /* Check the primary IP. */
924 if (cmp(PR_IP(pip, af, 0), addr) == 0)
928 * All the other IPs are sorted so we can do a binary search.
934 d = cmp(PR_IP(pip, af, i + 1), addr);
943 return (EADDRNOTAVAIL);
947 * Grab primary IP. Historically required mutex, but nothing prevents
948 * us to support epoch-protected access. Is it used in fast path?
949 * in{6}_jail.c helper
952 prison_ip_get0(const struct prison *pr, const pr_family_t af)
954 const struct prison_ip *pip = pr->pr_addrs[af];
956 mtx_assert(&pr->pr_mtx, MA_OWNED);
963 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
966 return (pr->pr_addrs[af]->ips);
968 #endif /* defined(INET) || defined(INET6) */
971 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
975 struct prison_ip *ip4;
978 struct prison_ip *ip6;
981 struct vfsoptlist *opts;
982 struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
984 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
985 char *g_path, *osrelstr;
986 struct bool_flags *bf;
987 struct jailsys_flags *jsf;
988 #if defined(INET) || defined(INET6)
992 size_t namelen, onamelen, pnamelen;
993 int created, cuflags, descend, drflags, enforce;
994 int error, errmsg_len, errmsg_pos;
995 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
996 int deadid, jid, jsys, len, level;
997 int childmax, osreldt, rsnum, slevel;
1006 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1007 uint64_t pr_allow_diff;
1011 error = priv_check(td, PRIV_JAIL_SET);
1012 if (!error && (flags & JAIL_ATTACH))
1013 error = priv_check(td, PRIV_JAIL_ATTACH);
1016 mypr = td->td_ucred->cr_prison;
1017 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
1019 if (flags & ~JAIL_SET_MASK)
1023 * Check all the parameters before committing to anything. Not all
1024 * errors can be caught early, but we may as well try. Also, this
1025 * takes care of some expensive stuff (path lookup) before getting
1026 * the allprison lock.
1028 * XXX Jails are not filesystems, and jail parameters are not mount
1029 * options. But it makes more sense to re-use the vfsopt code
1030 * than duplicate it under a different name.
1032 error = vfs_buildopts(optuio, &opts);
1043 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1046 vfs_opterror(opts, "no valid operation (create or update)");
1050 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1051 if (error == ENOENT)
1053 else if (error != 0)
1056 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1057 if (error == ENOENT)
1059 else if (error != 0)
1065 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1066 if (error == ENOENT)
1068 else if (error != 0)
1073 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1074 if (error == ENOENT)
1076 else if (error != 0)
1078 else if (enforce < 0 || enforce > 2) {
1084 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1085 if (error == ENOENT)
1087 else if (error != 0)
1092 pr_flags = ch_flags = 0;
1093 for (bf = pr_flag_bool;
1094 bf < pr_flag_bool + nitems(pr_flag_bool);
1096 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1097 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1099 ch_flags |= pr_flags;
1100 for (jsf = pr_flag_jailsys;
1101 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1103 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1104 if (error == ENOENT)
1109 case JAIL_SYS_DISABLE:
1110 if (!jsf->disable) {
1114 pr_flags |= jsf->disable;
1117 pr_flags |= jsf->new;
1119 case JAIL_SYS_INHERIT:
1125 ch_flags |= jsf->new | jsf->disable;
1127 if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1128 && !(pr_flags & PR_PERSIST)) {
1130 vfs_opterror(opts, "new jail must persist or attach");
1134 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1136 vfs_opterror(opts, "vnet cannot be changed after creation");
1141 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1143 vfs_opterror(opts, "ip4 cannot be changed after creation");
1148 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1150 vfs_opterror(opts, "ip6 cannot be changed after creation");
1155 pr_allow = ch_allow = 0;
1156 for (bf = pr_flag_allow;
1157 bf < pr_flag_allow + nitems(pr_flag_allow) &&
1158 atomic_load_int(&bf->flag) != 0;
1160 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1161 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1163 ch_allow |= pr_allow;
1165 error = vfs_getopt(opts, "name", (void **)&name, &len);
1166 if (error == ENOENT)
1168 else if (error != 0)
1171 if (len == 0 || name[len - 1] != '\0') {
1175 if (len > MAXHOSTNAMELEN) {
1176 error = ENAMETOOLONG;
1181 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1182 if (error == ENOENT)
1184 else if (error != 0)
1187 ch_flags |= PR_HOST;
1188 pr_flags |= PR_HOST;
1189 if (len == 0 || host[len - 1] != '\0') {
1193 if (len > MAXHOSTNAMELEN) {
1194 error = ENAMETOOLONG;
1199 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1200 if (error == ENOENT)
1202 else if (error != 0)
1205 ch_flags |= PR_HOST;
1206 pr_flags |= PR_HOST;
1207 if (len == 0 || domain[len - 1] != '\0') {
1211 if (len > MAXHOSTNAMELEN) {
1212 error = ENAMETOOLONG;
1217 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1218 if (error == ENOENT)
1220 else if (error != 0)
1223 ch_flags |= PR_HOST;
1224 pr_flags |= PR_HOST;
1225 if (len == 0 || uuid[len - 1] != '\0') {
1229 if (len > HOSTUUIDLEN) {
1230 error = ENAMETOOLONG;
1235 #ifdef COMPAT_FREEBSD32
1236 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1239 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1243 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1244 if (error == ENOENT)
1246 else if (error != 0)
1250 ch_flags |= PR_HOST;
1251 pr_flags |= PR_HOST;
1255 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1256 if (error == ENOENT)
1258 else if (error != 0)
1260 else if (ip4s & (sizeof(struct in_addr) - 1)) {
1264 ch_flags |= PR_IP4_USER;
1265 pr_flags |= PR_IP4_USER;
1267 ip4s /= sizeof(struct in_addr);
1268 if (ip4s > jail_max_af_ips) {
1270 vfs_opterror(opts, "too many IPv4 addresses");
1273 ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1283 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1284 if (error == ENOENT)
1286 else if (error != 0)
1288 else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1292 ch_flags |= PR_IP6_USER;
1293 pr_flags |= PR_IP6_USER;
1295 ip6s /= sizeof(struct in6_addr);
1296 if (ip6s > jail_max_af_ips) {
1298 vfs_opterror(opts, "too many IPv6 addresses");
1301 ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1310 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1311 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1314 "vnet jails cannot have IP address restrictions");
1319 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1320 if (error == ENOENT)
1322 else if (error != 0)
1325 if (flags & JAIL_UPDATE) {
1328 "osrelease cannot be changed after creation");
1331 if (len == 0 || osrelstr[len - 1] != '\0') {
1335 if (len >= OSRELEASELEN) {
1336 error = ENAMETOOLONG;
1338 "osrelease string must be 1-%d bytes long",
1344 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1345 if (error == ENOENT)
1347 else if (error != 0)
1350 if (flags & JAIL_UPDATE) {
1353 "osreldate cannot be changed after creation");
1358 vfs_opterror(opts, "osreldate cannot be 0");
1364 error = vfs_getopt(opts, "path", (void **)&path, &len);
1365 if (error == ENOENT)
1367 else if (error != 0)
1370 if (flags & JAIL_UPDATE) {
1373 "path cannot be changed after creation");
1376 if (len == 0 || path[len - 1] != '\0') {
1380 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1386 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1387 strlcpy(g_path, path, MAXPATHLEN);
1388 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1392 /* exit on other errors */
1395 if (root->v_type != VDIR) {
1404 * Find the specified jail, or at least its parent.
1405 * This abuses the file error codes ENOENT and EEXIST.
1410 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1411 namelc = strrchr(name, '.');
1412 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1416 sx_xlock(&allprison_lock);
1417 drflags = PD_LIST_XLOCKED;
1419 if (!prison_isalive(ppr)) {
1420 /* This jail is dying. This process will surely follow. */
1427 vfs_opterror(opts, "negative jid");
1431 * See if a requested jid already exists. Keep track of
1432 * where it can be inserted later.
1434 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1435 if (inspr->pr_id < jid)
1437 if (inspr->pr_id > jid)
1439 if (prison_isalive(inspr)) {
1441 mtx_lock(&pr->pr_mtx);
1442 drflags |= PD_LOCKED;
1444 /* Note a dying jail to handle later. */
1450 if (cuflags == JAIL_CREATE && pr != NULL) {
1452 * Even creators that cannot see the jail will
1456 vfs_opterror(opts, "jail %d already exists", jid);
1460 ? cuflags == JAIL_UPDATE
1461 : !prison_ischild(mypr, pr)) {
1463 * Updaters get ENOENT for nonexistent jails,
1464 * or for jails they cannot see. The latter
1465 * case is true even for CREATE | UPDATE,
1466 * which normally cannot give this error.
1469 vfs_opterror(opts, "jail %d not found", jid);
1474 * If the caller provided a name, look for a jail by that name.
1475 * This has different semantics for creates and updates keyed by jid
1476 * (where the name must not already exist in a different jail),
1477 * and updates keyed by the name itself (where the name must exist
1478 * because that is the jail being updated).
1482 namelc = strrchr(name, '.');
1487 * This is a hierarchical name. Split it into the
1488 * parent and child names, and make sure the parent
1489 * exists or matches an already found jail.
1492 if (strncmp(name, ppr->pr_name, namelc - name)
1493 || ppr->pr_name[namelc - name] != '\0') {
1496 "cannot change jail's parent");
1501 ppr = prison_find_name(mypr, name);
1505 "jail \"%s\" not found", name);
1508 mtx_unlock(&ppr->pr_mtx);
1509 if (!prison_isalive(ppr)) {
1512 "jail \"%s\" is dying", name);
1519 if (namelc[0] != '\0') {
1521 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1522 FOREACH_PRISON_CHILD(ppr, tpr) {
1523 if (tpr == pr || !prison_isalive(tpr) ||
1524 strcmp(tpr->pr_name + pnamelen, namelc))
1526 if (cuflags == JAIL_CREATE || pr != NULL) {
1528 * Create, or update(jid): name must
1529 * not exist in an active sibling jail.
1533 "jail \"%s\" already exists", name);
1536 /* Use this jail for updates. */
1538 mtx_lock(&pr->pr_mtx);
1539 drflags |= PD_LOCKED;
1543 * Update: name must exist if no jid is specified.
1544 * As with the jid case, the jail must be currently
1545 * visible, or else even CREATE | UPDATE will get
1549 ? cuflags == JAIL_UPDATE
1550 : !prison_isalive(pr)) {
1552 vfs_opterror(opts, "jail \"%s\" not found",
1558 /* Update: must provide a jid or name. */
1559 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1561 vfs_opterror(opts, "update specified no jail");
1565 /* If there's no prison to update, create a new one and link it in. */
1566 created = pr == NULL;
1568 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1569 if (tpr->pr_childcount >= tpr->pr_childmax) {
1571 vfs_opterror(opts, "prison limit exceeded");
1575 if (deadpr != NULL) {
1577 * The prison being created has the same ID as a dying
1578 * one. Handle this by giving the dying jail a new ID.
1579 * This may cause some confusion to user space, but
1580 * only to those listing dying jails.
1582 deadid = get_next_deadid(&dinspr);
1585 vfs_opterror(opts, "no available jail IDs");
1588 mtx_lock(&deadpr->pr_mtx);
1589 deadpr->pr_id = deadid;
1590 mtx_unlock(&deadpr->pr_mtx);
1591 if (dinspr == deadpr)
1594 inspr = TAILQ_NEXT(deadpr, pr_list);
1595 TAILQ_REMOVE(&allprison, deadpr, pr_list);
1597 TAILQ_INSERT_AFTER(&allprison, dinspr,
1600 TAILQ_INSERT_HEAD(&allprison, deadpr,
1604 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1606 vfs_opterror(opts, "no available jail IDs");
1610 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1611 pr->pr_state = PRISON_STATE_INVALID;
1612 refcount_init(&pr->pr_ref, 1);
1613 refcount_init(&pr->pr_uref, 0);
1614 drflags |= PD_DEREF;
1615 LIST_INIT(&pr->pr_children);
1616 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1617 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1621 TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1623 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1625 pr->pr_parent = ppr;
1627 prison_proc_hold(ppr);
1628 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1629 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1630 tpr->pr_childcount++;
1632 /* Set some default values, and inherit some from the parent. */
1637 root = mypr->pr_root;
1640 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1641 pr->pr_flags |= PR_HOST;
1642 #if defined(INET) || defined(INET6)
1644 if (!(pr_flags & PR_VNET))
1648 if (!(ch_flags & PR_IP4_USER))
1649 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1650 else if (!(pr_flags & PR_IP4_USER)) {
1651 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1652 prison_ip_dup(ppr, pr, PR_INET);
1656 if (!(ch_flags & PR_IP6_USER))
1657 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1658 else if (!(pr_flags & PR_IP6_USER)) {
1659 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1660 prison_ip_dup(ppr, pr, PR_INET6);
1665 /* Source address selection is always on by default. */
1666 pr->pr_flags |= _PR_IP_SADDRSEL;
1668 pr->pr_securelevel = ppr->pr_securelevel;
1669 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1670 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1671 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1673 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1674 if (osrelstr == NULL)
1675 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1676 sizeof(pr->pr_osrelease));
1678 strlcpy(pr->pr_osrelease, osrelstr,
1679 sizeof(pr->pr_osrelease));
1682 /* Allocate a new vnet if specified. */
1683 pr->pr_vnet = (pr_flags & PR_VNET)
1684 ? vnet_alloc() : ppr->pr_vnet;
1687 * Allocate a dedicated cpuset for each jail.
1688 * Unlike other initial settings, this may return an error.
1690 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1694 mtx_lock(&pr->pr_mtx);
1695 drflags |= PD_LOCKED;
1698 * Grab a reference for existing prisons, to ensure they
1699 * continue to exist for the duration of the call.
1702 drflags |= PD_DEREF;
1703 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1704 if ((pr->pr_flags & PR_VNET) &&
1705 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1708 "vnet jails cannot have IP address restrictions");
1713 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1716 "ip4 cannot be changed after creation");
1721 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1724 "ip6 cannot be changed after creation");
1730 /* Do final error checking before setting anything. */
1732 if (slevel < ppr->pr_securelevel) {
1738 if (childmax >= ppr->pr_childmax) {
1744 if (enforce < ppr->pr_enforce_statfs) {
1751 * devfs_rsnum is a uint16_t
1753 if (rsnum < 0 || rsnum > 65535) {
1758 * Nested jails always inherit parent's devfs ruleset
1760 if (jailed(td->td_ucred)) {
1761 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1765 rsnum = ppr->pr_devfs_rsnum;
1770 if ((ppr->pr_flags & PR_IP4) &&
1771 !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1776 if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1778 vfs_opterror(opts, "IPv4 addresses clash");
1785 if ((ppr->pr_flags & PR_IP6) &&
1786 !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1791 if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1793 vfs_opterror(opts, "IPv6 addresses clash");
1798 onamelen = namelen = 0;
1799 if (namelc != NULL) {
1800 /* Give a default name of the jid. Also allow the name to be
1801 * explicitly the jid - but not any other number, and only in
1802 * normal form (no leading zero/etc).
1804 if (namelc[0] == '\0')
1805 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1806 else if ((strtoul(namelc, &p, 10) != jid ||
1807 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1810 "name cannot be numeric (unless it is the jid)");
1814 * Make sure the name isn't too long for the prison or its
1817 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1818 onamelen = strlen(pr->pr_name + pnamelen);
1819 namelen = strlen(namelc);
1820 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1821 error = ENAMETOOLONG;
1824 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1825 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1826 sizeof(pr->pr_name)) {
1827 error = ENAMETOOLONG;
1832 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1833 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1839 * Let modules check their parameters. This requires unlocking and
1840 * then re-locking the prison, but this is still a valid state as long
1841 * as allprison_lock remains xlocked.
1843 mtx_unlock(&pr->pr_mtx);
1844 drflags &= ~PD_LOCKED;
1845 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1848 mtx_lock(&pr->pr_mtx);
1849 drflags |= PD_LOCKED;
1851 /* At this point, all valid parameters should have been noted. */
1852 TAILQ_FOREACH(opt, opts, link) {
1853 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1855 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1860 /* Set the parameters of the prison. */
1863 if (pr_flags & PR_IP4_USER) {
1864 pr->pr_flags |= PR_IP4;
1865 prison_ip_set(pr, PR_INET, ip4);
1867 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1869 if (tpr->pr_flags & PR_VNET) {
1874 if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
1883 if (pr_flags & PR_IP6_USER) {
1884 pr->pr_flags |= PR_IP6;
1885 prison_ip_set(pr, PR_INET6, ip6);
1887 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1889 if (tpr->pr_flags & PR_VNET) {
1894 if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
1902 pr->pr_securelevel = slevel;
1903 /* Set all child jails to be at least this level. */
1904 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1905 if (tpr->pr_securelevel < slevel)
1906 tpr->pr_securelevel = slevel;
1909 pr->pr_childmax = childmax;
1910 /* Set all child jails to under this limit. */
1911 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1912 if (tpr->pr_childmax > childmax - level)
1913 tpr->pr_childmax = childmax > level
1914 ? childmax - level : 0;
1917 pr->pr_enforce_statfs = enforce;
1918 /* Pass this restriction on to the children. */
1919 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1920 if (tpr->pr_enforce_statfs < enforce)
1921 tpr->pr_enforce_statfs = enforce;
1924 pr->pr_devfs_rsnum = rsnum;
1925 /* Pass this restriction on to the children. */
1926 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1927 tpr->pr_devfs_rsnum = rsnum;
1929 if (namelc != NULL) {
1930 if (ppr == &prison0)
1931 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1933 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1934 ppr->pr_name, namelc);
1935 /* Change this component of child names. */
1936 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1937 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1938 strlen(tpr->pr_name + onamelen) + 1);
1939 bcopy(pr->pr_name, tpr->pr_name, namelen);
1943 /* Try to keep a real-rooted full pathname. */
1944 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1948 if (PR_HOST & ch_flags & ~pr_flags) {
1949 if (pr->pr_flags & PR_HOST) {
1951 * Copy the parent's host info. As with pr_ip4 above,
1952 * the lack of a lock on the parent is not a problem;
1953 * it is always set with allprison_lock at least
1954 * shared, and is held exclusively here.
1956 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1957 sizeof(pr->pr_hostname));
1958 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1959 sizeof(pr->pr_domainname));
1960 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1961 sizeof(pr->pr_hostuuid));
1962 pr->pr_hostid = pr->pr_parent->pr_hostid;
1964 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1965 /* Set this prison, and any descendants without PR_HOST. */
1967 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1969 strlcpy(pr->pr_domainname, domain,
1970 sizeof(pr->pr_domainname));
1972 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1974 pr->pr_hostid = hid;
1975 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1976 if (tpr->pr_flags & PR_HOST)
1980 strlcpy(tpr->pr_hostname,
1982 sizeof(tpr->pr_hostname));
1984 strlcpy(tpr->pr_domainname,
1986 sizeof(tpr->pr_domainname));
1988 strlcpy(tpr->pr_hostuuid,
1990 sizeof(tpr->pr_hostuuid));
1992 tpr->pr_hostid = hid;
1996 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1997 if ((tallow = ch_allow & ~pr_allow))
1998 prison_set_allow_locked(pr, tallow, 0);
2000 * Persistent prisons get an extra reference, and prisons losing their
2001 * persist flag lose that reference.
2003 if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2004 if (pr_flags & PR_PERSIST) {
2007 * This may be a new prison's first user reference,
2008 * but wait to call it alive until after OSD calls
2009 * have had a chance to run (and perhaps to fail).
2011 refcount_acquire(&pr->pr_uref);
2013 drflags |= PD_DEUREF;
2014 prison_free_not_last(pr);
2017 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2018 mtx_unlock(&pr->pr_mtx);
2019 drflags &= ~PD_LOCKED;
2021 * Any errors past this point will need to de-persist newly created
2022 * prisons, as well as call remove methods.
2028 if (racct_enable && created)
2029 prison_racct_attach(pr);
2032 /* Locks may have prevented a complete restriction of child IP
2033 * addresses. If so, allocate some more memory and try again.
2037 ip4s = pr->pr_addrs[PR_INET]->ips;
2039 ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2040 mtx_lock(&pr->pr_mtx);
2042 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2044 if (tpr->pr_flags & PR_VNET) {
2049 if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2052 mtx_unlock(&pr->pr_mtx);
2057 ip6s = pr->pr_addrs[PR_INET6]->ips;
2059 ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2060 mtx_lock(&pr->pr_mtx);
2062 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2064 if (tpr->pr_flags & PR_VNET) {
2069 if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2072 mtx_unlock(&pr->pr_mtx);
2076 /* Let the modules do their work. */
2078 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2082 error = osd_jail_call(pr, PR_METHOD_SET, opts);
2087 * A new prison is now ready to be seen; either it has gained a user
2088 * reference via persistence, or is about to gain one via attachment.
2091 drflags = prison_lock_xlock(pr, drflags);
2092 pr->pr_state = PRISON_STATE_ALIVE;
2095 /* Attach this process to the prison if requested. */
2096 if (flags & JAIL_ATTACH) {
2097 error = do_jail_attach(td, pr,
2098 prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2099 drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2101 vfs_opterror(opts, "attach failed");
2107 if (racct_enable && !created) {
2108 if (drflags & PD_LOCKED) {
2109 mtx_unlock(&pr->pr_mtx);
2110 drflags &= ~PD_LOCKED;
2112 if (drflags & PD_LIST_XLOCKED) {
2113 sx_xunlock(&allprison_lock);
2114 drflags &= ~PD_LIST_XLOCKED;
2116 prison_racct_modify(pr);
2120 if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2121 (pr->pr_root->v_vflag & VV_ROOT) == 0)
2122 printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2123 " file system\n", pr->pr_id);
2125 drflags &= ~PD_KILL;
2126 td->td_retval[0] = pr->pr_id;
2129 /* Release any temporary prison holds and/or locks. */
2131 prison_deref(pr, drflags);
2132 else if (drflags & PD_LIST_SLOCKED)
2133 sx_sunlock(&allprison_lock);
2134 else if (drflags & PD_LIST_XLOCKED)
2135 sx_xunlock(&allprison_lock);
2140 /* Write the error message back to userspace. */
2141 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2142 &errmsg_len) == 0 && errmsg_len > 0) {
2143 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2144 if (optuio->uio_segflg == UIO_SYSSPACE)
2146 optuio->uio_iov[errmsg_pos].iov_base,
2150 optuio->uio_iov[errmsg_pos].iov_base,
2156 prison_ip_free(ip4);
2159 prison_ip_free(ip6);
2162 free(g_path, M_TEMP);
2168 * Find the next available prison ID. Return the ID on success, or zero
2169 * on failure. Also set a pointer to the allprison list entry the prison
2170 * should be inserted before.
2173 get_next_prid(struct prison **insprp)
2175 struct prison *inspr;
2178 jid = lastprid % JAIL_MAX + 1;
2179 if (TAILQ_EMPTY(&allprison) ||
2180 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2182 * A common case is for all jails to be implicitly numbered,
2183 * which means they'll go on the end of the list, at least
2184 * for the first JAIL_MAX times.
2189 * Take two passes through the allprison list: first starting
2190 * with the proposed jid, then ending with it.
2192 for (maxid = JAIL_MAX; maxid != 0; ) {
2193 TAILQ_FOREACH(inspr, &allprison, pr_list) {
2194 if (inspr->pr_id < jid)
2196 if (inspr->pr_id > jid) {
2197 /* Found an opening. */
2201 if (++jid > maxid) {
2202 if (lastprid == maxid || lastprid == 0)
2205 * The entire legal range
2206 * has been traversed
2210 /* Try again from the start. */
2216 if (inspr == NULL) {
2217 /* Found room at the end of the list. */
2228 * Find the next available ID for a renumbered dead prison. This is the same
2229 * as get_next_prid, but counting backward from the end of the range.
2232 get_next_deadid(struct prison **dinsprp)
2234 struct prison *dinspr;
2237 deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2239 * Take two reverse passes through the allprison list: first
2240 * starting with the proposed deadid, then ending with it.
2242 for (minid = 1; minid != 0; ) {
2243 TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2244 if (dinspr->pr_id > deadid)
2246 if (dinspr->pr_id < deadid) {
2247 /* Found an opening. */
2251 if (--deadid < minid) {
2252 if (lastdeadid == minid || lastdeadid == 0)
2255 * The entire legal range
2256 * has been traversed
2260 /* Try again from the end. */
2266 if (dinspr == NULL) {
2267 /* Found room at the beginning of the list. */
2272 lastdeadid = deadid;
2277 * struct jail_get_args {
2278 * struct iovec *iovp;
2279 * unsigned int iovcnt;
2284 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2289 /* Check that we have an even number of iovecs. */
2290 if (uap->iovcnt & 1)
2293 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2296 error = kern_jail_get(td, auio, uap->flags);
2298 error = copyout(auio->uio_iov, uap->iovp,
2299 uap->iovcnt * sizeof (struct iovec));
2305 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2307 struct bool_flags *bf;
2308 struct jailsys_flags *jsf;
2309 struct prison *pr, *mypr;
2311 struct vfsoptlist *opts;
2312 char *errmsg, *name;
2313 int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2316 if (flags & ~JAIL_GET_MASK)
2319 /* Get the parameter list. */
2320 error = vfs_buildopts(optuio, &opts);
2323 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2324 mypr = td->td_ucred->cr_prison;
2328 * Find the prison specified by one of: lastjid, jid, name.
2330 sx_slock(&allprison_lock);
2331 drflags = PD_LIST_SLOCKED;
2332 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2334 TAILQ_FOREACH(pr, &allprison, pr_list) {
2335 if (pr->pr_id > jid &&
2336 ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2337 prison_ischild(mypr, pr)) {
2338 mtx_lock(&pr->pr_mtx);
2339 drflags |= PD_LOCKED;
2344 vfs_opterror(opts, "no jail after %d", jid);
2346 } else if (error != ENOENT)
2349 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2352 pr = prison_find_child(mypr, jid);
2354 drflags |= PD_LOCKED;
2355 if (!(prison_isalive(pr) ||
2356 (flags & JAIL_DYING))) {
2358 vfs_opterror(opts, "jail %d is dying",
2365 vfs_opterror(opts, "jail %d not found", jid);
2368 } else if (error != ENOENT)
2371 error = vfs_getopt(opts, "name", (void **)&name, &len);
2373 if (len == 0 || name[len - 1] != '\0') {
2377 pr = prison_find_name(mypr, name);
2379 drflags |= PD_LOCKED;
2380 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2382 vfs_opterror(opts, "jail \"%s\" is dying",
2389 vfs_opterror(opts, "jail \"%s\" not found", name);
2391 } else if (error != ENOENT)
2394 vfs_opterror(opts, "no jail specified");
2399 /* Get the parameters of the prison. */
2401 drflags |= PD_DEREF;
2402 td->td_retval[0] = pr->pr_id;
2403 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2404 if (error != 0 && error != ENOENT)
2406 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2407 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2408 if (error != 0 && error != ENOENT)
2410 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2411 if (error != 0 && error != ENOENT)
2413 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2414 sizeof(pr->pr_cpuset->cs_id));
2415 if (error != 0 && error != ENOENT)
2417 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2418 if (error != 0 && error != ENOENT)
2421 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2422 pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2423 pr_families[PR_INET].size : 0 );
2424 if (error != 0 && error != ENOENT)
2428 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2429 pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2430 pr_families[PR_INET6].size : 0 );
2431 if (error != 0 && error != ENOENT)
2434 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2435 sizeof(pr->pr_securelevel));
2436 if (error != 0 && error != ENOENT)
2438 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2439 sizeof(pr->pr_childcount));
2440 if (error != 0 && error != ENOENT)
2442 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2443 sizeof(pr->pr_childmax));
2444 if (error != 0 && error != ENOENT)
2446 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2447 if (error != 0 && error != ENOENT)
2449 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2450 if (error != 0 && error != ENOENT)
2452 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2453 if (error != 0 && error != ENOENT)
2455 #ifdef COMPAT_FREEBSD32
2456 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2457 uint32_t hid32 = pr->pr_hostid;
2459 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2462 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2463 sizeof(pr->pr_hostid));
2464 if (error != 0 && error != ENOENT)
2466 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2467 sizeof(pr->pr_enforce_statfs));
2468 if (error != 0 && error != ENOENT)
2470 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2471 sizeof(pr->pr_devfs_rsnum));
2472 if (error != 0 && error != ENOENT)
2474 for (bf = pr_flag_bool;
2475 bf < pr_flag_bool + nitems(pr_flag_bool);
2477 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2478 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2479 if (error != 0 && error != ENOENT)
2482 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2483 if (error != 0 && error != ENOENT)
2486 for (jsf = pr_flag_jailsys;
2487 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2489 f = pr->pr_flags & (jsf->disable | jsf->new);
2490 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2491 : (f == jsf->new) ? JAIL_SYS_NEW
2493 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2494 if (error != 0 && error != ENOENT)
2497 for (bf = pr_flag_allow;
2498 bf < pr_flag_allow + nitems(pr_flag_allow) &&
2499 atomic_load_int(&bf->flag) != 0;
2501 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2502 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2503 if (error != 0 && error != ENOENT)
2506 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2507 if (error != 0 && error != ENOENT)
2510 i = !prison_isalive(pr);
2511 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2512 if (error != 0 && error != ENOENT)
2515 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2516 if (error != 0 && error != ENOENT)
2518 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2519 sizeof(pr->pr_osreldate));
2520 if (error != 0 && error != ENOENT)
2522 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2523 if (error != 0 && error != ENOENT)
2526 /* Get the module parameters. */
2527 mtx_unlock(&pr->pr_mtx);
2528 drflags &= ~PD_LOCKED;
2529 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2532 prison_deref(pr, drflags);
2536 /* By now, all parameters should have been noted. */
2537 TAILQ_FOREACH(opt, opts, link) {
2538 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2540 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2545 /* Write the fetched parameters back to userspace. */
2547 TAILQ_FOREACH(opt, opts, link) {
2548 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2549 pos = 2 * opt->pos + 1;
2550 optuio->uio_iov[pos].iov_len = opt->len;
2551 if (opt->value != NULL) {
2552 if (optuio->uio_segflg == UIO_SYSSPACE) {
2554 optuio->uio_iov[pos].iov_base,
2557 error = copyout(opt->value,
2558 optuio->uio_iov[pos].iov_base,
2568 /* Release any temporary prison holds and/or locks. */
2570 prison_deref(pr, drflags);
2571 else if (drflags & PD_LIST_SLOCKED)
2572 sx_sunlock(&allprison_lock);
2573 if (error && errmsg_pos >= 0) {
2574 /* Write the error message back to userspace. */
2575 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2576 errmsg_pos = 2 * errmsg_pos + 1;
2577 if (errmsg_len > 0) {
2578 if (optuio->uio_segflg == UIO_SYSSPACE)
2580 optuio->uio_iov[errmsg_pos].iov_base,
2584 optuio->uio_iov[errmsg_pos].iov_base,
2593 * struct jail_remove_args {
2598 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2603 error = priv_check(td, PRIV_JAIL_REMOVE);
2607 sx_xlock(&allprison_lock);
2608 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2610 sx_xunlock(&allprison_lock);
2613 if (!prison_isalive(pr)) {
2614 /* Silently ignore already-dying prisons. */
2615 mtx_unlock(&pr->pr_mtx);
2616 sx_xunlock(&allprison_lock);
2619 prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2624 * struct jail_attach_args {
2629 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2634 error = priv_check(td, PRIV_JAIL_ATTACH);
2638 sx_slock(&allprison_lock);
2639 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2641 sx_sunlock(&allprison_lock);
2645 /* Do not allow a process to attach to a prison that is not alive. */
2646 if (!prison_isalive(pr)) {
2647 mtx_unlock(&pr->pr_mtx);
2648 sx_sunlock(&allprison_lock);
2652 return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2656 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2659 struct ucred *newcred, *oldcred;
2662 mtx_assert(&pr->pr_mtx, MA_OWNED);
2663 sx_assert(&allprison_lock, SX_LOCKED);
2664 drflags &= PD_LOCK_FLAGS;
2666 * XXX: Note that there is a slight race here if two threads
2667 * in the same privileged process attempt to attach to two
2668 * different jails at the same time. It is important for
2669 * user processes not to do this, or they might end up with
2670 * a process root from one prison, but attached to the jail
2674 refcount_acquire(&pr->pr_uref);
2675 drflags |= PD_DEREF | PD_DEUREF;
2676 mtx_unlock(&pr->pr_mtx);
2677 drflags &= ~PD_LOCKED;
2679 /* Let modules do whatever they need to prepare for attaching. */
2680 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2682 prison_deref(pr, drflags);
2685 sx_unlock(&allprison_lock);
2686 drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2689 * Reparent the newly attached process to this jail.
2692 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2696 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2697 if ((error = change_dir(pr->pr_root, td)) != 0)
2700 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2703 VOP_UNLOCK(pr->pr_root);
2704 if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2709 oldcred = crcopysafe(p, newcred);
2710 newcred->cr_prison = pr;
2711 proc_set_cred(p, newcred);
2714 racct_proc_ucred_changed(p, oldcred, newcred);
2719 rctl_proc_ucred_changed(p, newcred);
2722 prison_proc_relink(oldcred->cr_prison, pr, p);
2723 prison_deref(oldcred->cr_prison, drflags);
2727 * If the prison was killed while changing credentials, die along
2730 if (!prison_isalive(pr)) {
2732 kern_psignal(p, SIGKILL);
2739 VOP_UNLOCK(pr->pr_root);
2741 /* Tell modules this thread is still in its old jail after all. */
2742 sx_slock(&allprison_lock);
2743 drflags |= PD_LIST_SLOCKED;
2744 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2745 prison_deref(pr, drflags);
2750 * Returns a locked prison instance, or NULL on failure.
2753 prison_find(int prid)
2757 sx_assert(&allprison_lock, SX_LOCKED);
2758 TAILQ_FOREACH(pr, &allprison, pr_list) {
2759 if (pr->pr_id < prid)
2761 if (pr->pr_id > prid)
2763 KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2764 mtx_lock(&pr->pr_mtx);
2771 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2774 prison_find_child(struct prison *mypr, int prid)
2779 sx_assert(&allprison_lock, SX_LOCKED);
2780 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2781 if (pr->pr_id == prid) {
2782 KASSERT(prison_isvalid(pr),
2783 ("Found invalid prison %p", pr));
2784 mtx_lock(&pr->pr_mtx);
2792 * Look for the name relative to mypr. Returns a locked prison or NULL.
2795 prison_find_name(struct prison *mypr, const char *name)
2797 struct prison *pr, *deadpr;
2801 sx_assert(&allprison_lock, SX_LOCKED);
2802 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2804 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2805 if (!strcmp(pr->pr_name + mylen, name)) {
2806 KASSERT(prison_isvalid(pr),
2807 ("Found invalid prison %p", pr));
2808 if (prison_isalive(pr)) {
2809 mtx_lock(&pr->pr_mtx);
2815 /* There was no valid prison - perhaps there was a dying one. */
2817 mtx_lock(&deadpr->pr_mtx);
2822 * See if a prison has the specific flag set. The prison should be locked,
2823 * unless checking for flags that are only set at jail creation (such as
2824 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2825 * to any other prison data.
2828 prison_flag(struct ucred *cred, unsigned flag)
2831 return ((cred->cr_prison->pr_flags & flag) != 0);
2835 * See if a prison has the specific allow flag set.
2836 * The prison *should* be locked, or only a single bit is examined, without
2837 * regard to any other prison data.
2840 prison_allow(struct ucred *cred, unsigned flag)
2843 return ((cred->cr_prison->pr_allow & flag) != 0);
2847 * Hold a prison reference, by incrementing pr_ref. It is generally
2848 * an error to hold a prison that does not already have a reference.
2849 * A prison record will remain valid as long as it has at least one
2850 * reference, and will not be removed as long as either the prison
2851 * mutex or the allprison lock is held (allprison_lock may be shared).
2854 prison_hold_locked(struct prison *pr)
2857 /* Locking is no longer required. */
2862 prison_hold(struct prison *pr)
2865 int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2868 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2870 refcount_acquire(&pr->pr_ref);
2875 * Remove a prison reference. If that was the last reference, the
2876 * prison will be removed (at a later time).
2879 prison_free_locked(struct prison *pr)
2882 mtx_assert(&pr->pr_mtx, MA_OWNED);
2884 * Locking is no longer required, but unlock because the caller
2887 mtx_unlock(&pr->pr_mtx);
2892 prison_free(struct prison *pr)
2895 KASSERT(refcount_load(&pr->pr_ref) > 0,
2896 ("Trying to free dead prison %p (jid=%d).",
2898 if (!refcount_release_if_not_last(&pr->pr_ref)) {
2900 * Don't remove the last reference in this context,
2901 * in case there are locks held.
2903 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2908 prison_free_not_last(struct prison *pr)
2913 KASSERT(refcount_load(&pr->pr_ref) > 0,
2914 ("Trying to free dead prison %p (jid=%d).",
2916 lastref = refcount_release(&pr->pr_ref);
2918 ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2921 refcount_release(&pr->pr_ref);
2926 * Hold a prison for user visibility, by incrementing pr_uref.
2927 * It is generally an error to hold a prison that isn't already
2928 * user-visible, except through the jail system calls. It is also
2929 * an error to hold an invalid prison. A prison record will remain
2930 * alive as long as it has at least one user reference, and will not
2931 * be set to the dying state until the prison mutex and allprison_lock
2935 prison_proc_hold(struct prison *pr)
2938 int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2941 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2943 refcount_acquire(&pr->pr_uref);
2948 * Remove a prison user reference. If it was the last reference, the
2949 * prison will be considered "dying", and may be removed once all of
2950 * its references are dropped.
2953 prison_proc_free(struct prison *pr)
2957 * Locking is only required when releasing the last reference.
2958 * This allows assurance that a locked prison will remain alive
2959 * until it is unlocked.
2961 KASSERT(refcount_load(&pr->pr_uref) > 0,
2962 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2963 if (!refcount_release_if_not_last(&pr->pr_uref)) {
2965 * Don't remove the last user reference in this context,
2966 * which is expected to be a process that is not only locked,
2967 * but also half dead. Add a reference so any calls to
2968 * prison_free() won't re-submit the task.
2971 mtx_lock(&pr->pr_mtx);
2972 KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2973 ("Redundant last reference in prison_proc_free (jid=%d)",
2975 pr->pr_flags |= PR_COMPLETE_PROC;
2976 mtx_unlock(&pr->pr_mtx);
2977 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2982 prison_proc_free_not_last(struct prison *pr)
2987 KASSERT(refcount_load(&pr->pr_uref) > 0,
2988 ("Trying to free dead prison %p (jid=%d).",
2990 lastref = refcount_release(&pr->pr_uref);
2992 ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2995 refcount_release(&pr->pr_uref);
3000 prison_proc_link(struct prison *pr, struct proc *p)
3003 sx_assert(&allproc_lock, SA_XLOCKED);
3004 LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3008 prison_proc_unlink(struct prison *pr, struct proc *p)
3011 sx_assert(&allproc_lock, SA_XLOCKED);
3012 LIST_REMOVE(p, p_jaillist);
3016 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3019 sx_xlock(&allproc_lock);
3020 prison_proc_unlink(opr, p);
3021 prison_proc_link(npr, p);
3022 sx_xunlock(&allproc_lock);
3026 * Complete a call to either prison_free or prison_proc_free.
3029 prison_complete(void *context, int pending)
3031 struct prison *pr = context;
3035 * This could be called to release the last reference, or the last
3036 * user reference (plus the reference held in prison_proc_free).
3038 drflags = prison_lock_xlock(pr, PD_DEREF);
3039 if (pr->pr_flags & PR_COMPLETE_PROC) {
3040 pr->pr_flags &= ~PR_COMPLETE_PROC;
3041 drflags |= PD_DEUREF;
3043 prison_deref(pr, drflags);
3047 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3050 kern_psignal(p, SIGKILL);
3054 * Note the iteration does not guarantee acting on all processes.
3055 * Most notably there may be fork or jail_attach in progress.
3058 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3064 if (atomic_load_int(&pr->pr_childcount) == 0) {
3065 sx_slock(&allproc_lock);
3066 LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3067 if (p->p_state == PRS_NEW)
3073 sx_sunlock(&allproc_lock);
3074 if (atomic_load_int(&pr->pr_childcount) == 0)
3077 * Some jails popped up during the iteration, fall through to a
3078 * system-wide search.
3082 sx_slock(&allproc_lock);
3083 FOREACH_PROC_IN_SYSTEM(p) {
3085 if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3086 for (ppr = p->p_ucred->cr_prison;
3088 ppr = ppr->pr_parent) {
3097 sx_sunlock(&allproc_lock);
3101 * Remove a prison reference and/or user reference (usually).
3102 * This assumes context that allows sleeping (for allprison_lock),
3103 * with no non-sleeping locks held, except perhaps the prison itself.
3104 * If there are no more references, release and delist the prison.
3105 * On completion, the prison lock and the allprison lock are both
3109 prison_deref(struct prison *pr, int flags)
3111 struct prisonlist freeprison;
3112 struct prison *killpr, *rpr, *ppr, *tpr;
3115 TAILQ_INIT(&freeprison);
3117 * Release this prison as requested, which may cause its parent
3118 * to be released, and then maybe its grandparent, etc.
3121 if (flags & PD_KILL) {
3122 /* Kill the prison and its descendents. */
3123 KASSERT(pr != &prison0,
3124 ("prison_deref trying to kill prison0"));
3125 if (!(flags & PD_DEREF)) {
3129 flags = prison_lock_xlock(pr, flags);
3130 prison_deref_kill(pr, &freeprison);
3132 if (flags & PD_DEUREF) {
3133 /* Drop a user reference. */
3134 KASSERT(refcount_load(&pr->pr_uref) > 0,
3135 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3137 if (!refcount_release_if_not_last(&pr->pr_uref)) {
3138 if (!(flags & PD_DEREF)) {
3142 flags = prison_lock_xlock(pr, flags);
3143 if (refcount_release(&pr->pr_uref) &&
3144 pr->pr_state == PRISON_STATE_ALIVE) {
3146 * When the last user references goes,
3147 * this becomes a dying prison.
3150 refcount_load(&prison0.pr_uref) > 0,
3151 ("prison0 pr_uref=0"));
3152 pr->pr_state = PRISON_STATE_DYING;
3153 mtx_unlock(&pr->pr_mtx);
3154 flags &= ~PD_LOCKED;
3159 if (flags & PD_KILL) {
3161 * Any remaining user references are probably processes
3162 * that need to be killed, either in this prison or its
3165 if (refcount_load(&pr->pr_uref) > 0)
3167 /* Make sure the parent prison doesn't get killed. */
3170 if (flags & PD_DEREF) {
3171 /* Drop a reference. */
3172 KASSERT(refcount_load(&pr->pr_ref) > 0,
3173 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3175 if (!refcount_release_if_not_last(&pr->pr_ref)) {
3176 flags = prison_lock_xlock(pr, flags);
3177 if (refcount_release(&pr->pr_ref)) {
3179 * When the last reference goes,
3180 * unlink the prison and set it aside.
3183 refcount_load(&pr->pr_uref) == 0,
3184 ("prison_deref: last ref, "
3185 "but still has %d urefs (jid=%d)",
3186 pr->pr_uref, pr->pr_id));
3188 refcount_load(&prison0.pr_ref) != 0,
3189 ("prison0 pr_ref=0"));
3190 pr->pr_state = PRISON_STATE_INVALID;
3191 TAILQ_REMOVE(&allprison, pr, pr_list);
3192 LIST_REMOVE(pr, pr_sibling);
3193 TAILQ_INSERT_TAIL(&freeprison, pr,
3195 for (ppr = pr->pr_parent;
3197 ppr = ppr->pr_parent)
3198 ppr->pr_childcount--;
3200 * Removing a prison frees references
3203 mtx_unlock(&pr->pr_mtx);
3204 flags &= ~PD_LOCKED;
3206 flags |= PD_DEREF | PD_DEUREF;
3214 /* Release all the prison locks. */
3215 if (flags & PD_LOCKED)
3216 mtx_unlock(&pr->pr_mtx);
3217 if (flags & PD_LIST_SLOCKED)
3218 sx_sunlock(&allprison_lock);
3219 else if (flags & PD_LIST_XLOCKED)
3220 sx_xunlock(&allprison_lock);
3222 /* Kill any processes attached to a killed prison. */
3224 prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3227 * Finish removing any unreferenced prisons, which couldn't happen
3228 * while allprison_lock was held (to avoid a LOR on vrele).
3230 TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3232 if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3233 vnet_destroy(rpr->pr_vnet);
3235 if (rpr->pr_root != NULL)
3236 vrele(rpr->pr_root);
3237 mtx_destroy(&rpr->pr_mtx);
3239 prison_ip_free(rpr->pr_addrs[PR_INET]);
3242 prison_ip_free(rpr->pr_addrs[PR_INET6]);
3244 if (rpr->pr_cpuset != NULL)
3245 cpuset_rel(rpr->pr_cpuset);
3249 prison_racct_detach(rpr);
3251 TAILQ_REMOVE(&freeprison, rpr, pr_list);
3252 free(rpr, M_PRISON);
3257 * Kill the prison and its descendants. Mark them as dying, clear the
3258 * persist flag, and call module remove methods.
3261 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3263 struct prison *cpr, *ppr, *rpr;
3267 * Unlike the descendants, the target prison can be killed
3268 * even if it is currently dying. This is useful for failed
3269 * creation in jail_set(2).
3271 KASSERT(refcount_load(&pr->pr_ref) > 0,
3272 ("Trying to kill dead prison %p (jid=%d).",
3274 refcount_acquire(&pr->pr_uref);
3275 pr->pr_state = PRISON_STATE_DYING;
3276 mtx_unlock(&pr->pr_mtx);
3279 FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3281 if (!prison_isalive(cpr)) {
3286 prison_proc_hold(cpr);
3287 mtx_lock(&cpr->pr_mtx);
3288 cpr->pr_state = PRISON_STATE_DYING;
3289 cpr->pr_flags |= PR_REMOVE;
3290 mtx_unlock(&cpr->pr_mtx);
3293 if (!(cpr->pr_flags & PR_REMOVE))
3295 prison_cleanup(cpr);
3296 mtx_lock(&cpr->pr_mtx);
3297 cpr->pr_flags &= ~PR_REMOVE;
3298 if (cpr->pr_flags & PR_PERSIST) {
3299 cpr->pr_flags &= ~PR_PERSIST;
3300 prison_proc_free_not_last(cpr);
3301 prison_free_not_last(cpr);
3303 (void)refcount_release(&cpr->pr_uref);
3304 if (refcount_release(&cpr->pr_ref)) {
3306 * When the last reference goes, unlink the prison
3307 * and set it aside for prison_deref() to handle.
3308 * Delay unlinking the sibling list to keep the loop
3312 LIST_REMOVE(rpr, pr_sibling);
3314 rpr->pr_state = PRISON_STATE_INVALID;
3315 TAILQ_REMOVE(&allprison, rpr, pr_list);
3316 TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3318 * Removing a prison frees references from its parent.
3320 ppr = rpr->pr_parent;
3321 prison_proc_free_not_last(ppr);
3322 prison_free_not_last(ppr);
3323 for (; ppr != NULL; ppr = ppr->pr_parent)
3324 ppr->pr_childcount--;
3326 mtx_unlock(&cpr->pr_mtx);
3329 LIST_REMOVE(rpr, pr_sibling);
3332 mtx_lock(&pr->pr_mtx);
3333 if (pr->pr_flags & PR_PERSIST) {
3334 pr->pr_flags &= ~PR_PERSIST;
3335 prison_proc_free_not_last(pr);
3336 prison_free_not_last(pr);
3338 (void)refcount_release(&pr->pr_uref);
3342 * Given the current locking state in the flags, make sure allprison_lock
3343 * is held exclusive, and the prison is locked. Return flags indicating
3347 prison_lock_xlock(struct prison *pr, int flags)
3350 if (!(flags & PD_LIST_XLOCKED)) {
3352 * Get allprison_lock, which may be an upgrade,
3353 * and may require unlocking the prison.
3355 if (flags & PD_LOCKED) {
3356 mtx_unlock(&pr->pr_mtx);
3357 flags &= ~PD_LOCKED;
3359 if (flags & PD_LIST_SLOCKED) {
3360 if (!sx_try_upgrade(&allprison_lock)) {
3361 sx_sunlock(&allprison_lock);
3362 sx_xlock(&allprison_lock);
3364 flags &= ~PD_LIST_SLOCKED;
3366 sx_xlock(&allprison_lock);
3367 flags |= PD_LIST_XLOCKED;
3369 if (!(flags & PD_LOCKED)) {
3370 /* Lock the prison mutex. */
3371 mtx_lock(&pr->pr_mtx);
3378 * Release a prison's resources when it starts dying (when the last user
3379 * reference is dropped, or when it is killed).
3382 prison_cleanup(struct prison *pr)
3384 sx_assert(&allprison_lock, SA_XLOCKED);
3385 mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3386 vfs_exjail_delete(pr);
3387 shm_remove_prison(pr);
3388 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3392 * Set or clear a permission bit in the pr_allow field, passing restrictions
3393 * (cleared permission) down to child jails.
3396 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3400 pr = cred->cr_prison;
3401 sx_slock(&allprison_lock);
3402 mtx_lock(&pr->pr_mtx);
3403 prison_set_allow_locked(pr, flag, enable);
3404 mtx_unlock(&pr->pr_mtx);
3405 sx_sunlock(&allprison_lock);
3409 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3415 pr->pr_allow |= flag;
3417 pr->pr_allow &= ~flag;
3418 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3419 cpr->pr_allow &= ~flag;
3424 * Check if a jail supports the given address family.
3426 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3430 prison_check_af(struct ucred *cred, int af)
3435 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3437 pr = cred->cr_prison;
3439 /* Prisons with their own network stack are not limited. */
3440 if (prison_owns_vnet(cred))
3449 if (pr->pr_flags & PR_IP4)
3451 mtx_lock(&pr->pr_mtx);
3452 if ((pr->pr_flags & PR_IP4) &&
3453 pr->pr_addrs[PR_INET] == NULL)
3454 error = EAFNOSUPPORT;
3455 mtx_unlock(&pr->pr_mtx);
3461 if (pr->pr_flags & PR_IP6)
3463 mtx_lock(&pr->pr_mtx);
3464 if ((pr->pr_flags & PR_IP6) &&
3465 pr->pr_addrs[PR_INET6] == NULL)
3466 error = EAFNOSUPPORT;
3467 mtx_unlock(&pr->pr_mtx);
3476 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3477 error = EAFNOSUPPORT;
3483 * Check if given address belongs to the jail referenced by cred (wrapper to
3484 * prison_check_ip[46]).
3486 * Returns 0 if jail doesn't restrict the address family or if address belongs
3487 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3488 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3491 prison_if(struct ucred *cred, const struct sockaddr *sa)
3494 const struct sockaddr_in *sai;
3497 const struct sockaddr_in6 *sai6;
3501 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3502 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3505 if (prison_owns_vnet(cred))
3510 switch (sa->sa_family)
3514 sai = (const struct sockaddr_in *)sa;
3515 error = prison_check_ip4(cred, &sai->sin_addr);
3520 sai6 = (const struct sockaddr_in6 *)sa;
3521 error = prison_check_ip6(cred, &sai6->sin6_addr);
3525 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3526 error = EAFNOSUPPORT;
3532 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3535 prison_check(struct ucred *cred1, struct ucred *cred2)
3538 return ((cred1->cr_prison == cred2->cr_prison ||
3539 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3543 * For mountd/nfsd to run within a prison, it must be:
3545 * - PR_ALLOW_NFSD must be set on it.
3546 * - The root directory (pr_root) of the prison must be
3547 * a file system mount point, so the mountd can hang
3548 * export information on it.
3549 * - The prison's enforce_statfs cannot be 0, so that
3550 * mountd(8) can do exports.
3553 prison_check_nfsd(struct ucred *cred)
3556 if (jailed_without_vnet(cred))
3558 if (!prison_allow(cred, PR_ALLOW_NFSD))
3560 if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3562 if (cred->cr_prison->pr_enforce_statfs == 0)
3568 * Return true if p2 is a child of p1, otherwise false.
3571 prison_ischild(struct prison *pr1, struct prison *pr2)
3574 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3581 * Return true if the prison is currently alive. A prison is alive if it
3582 * holds user references and it isn't being removed.
3585 prison_isalive(const struct prison *pr)
3588 if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3594 * Return true if the prison is currently valid. A prison is valid if it has
3595 * been fully created, and is not being destroyed. Note that dying prisons
3596 * are still considered valid. Invalid prisons won't be found under normal
3597 * circumstances, as they're only put in that state by functions that have
3598 * an exclusive hold on allprison_lock.
3601 prison_isvalid(struct prison *pr)
3604 if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3606 if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3612 * Return true if the passed credential is in a jail and that jail does not
3613 * have its own virtual network stack, otherwise false.
3616 jailed_without_vnet(struct ucred *cred)
3622 if (prison_owns_vnet(cred))
3630 * Return the correct hostname (domainname, et al) for the passed credential.
3633 getcredhostname(struct ucred *cred, char *buf, size_t size)
3638 * A NULL credential can be used to shortcut to the physical
3639 * system's hostname.
3641 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3642 mtx_lock(&pr->pr_mtx);
3643 strlcpy(buf, pr->pr_hostname, size);
3644 mtx_unlock(&pr->pr_mtx);
3648 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3651 mtx_lock(&cred->cr_prison->pr_mtx);
3652 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3653 mtx_unlock(&cred->cr_prison->pr_mtx);
3657 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3660 mtx_lock(&cred->cr_prison->pr_mtx);
3661 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3662 mtx_unlock(&cred->cr_prison->pr_mtx);
3666 getcredhostid(struct ucred *cred, unsigned long *hostid)
3669 mtx_lock(&cred->cr_prison->pr_mtx);
3670 *hostid = cred->cr_prison->pr_hostid;
3671 mtx_unlock(&cred->cr_prison->pr_mtx);
3675 getjailname(struct ucred *cred, char *name, size_t len)
3678 mtx_lock(&cred->cr_prison->pr_mtx);
3679 strlcpy(name, cred->cr_prison->pr_name, len);
3680 mtx_unlock(&cred->cr_prison->pr_mtx);
3685 * Determine whether the prison represented by cred owns
3686 * its vnet rather than having it inherited.
3688 * Returns true in case the prison owns the vnet, false otherwise.
3691 prison_owns_vnet(struct ucred *cred)
3695 * vnets cannot be added/removed after jail creation,
3696 * so no need to lock here.
3698 return ((cred->cr_prison->pr_flags & PR_VNET) != 0);
3703 * Determine whether the subject represented by cred can "see"
3704 * status of a mount point.
3705 * Returns: 0 for permitted, ENOENT otherwise.
3706 * XXX: This function should be called cr_canseemount() and should be
3707 * placed in kern_prot.c.
3710 prison_canseemount(struct ucred *cred, struct mount *mp)
3716 pr = cred->cr_prison;
3717 if (pr->pr_enforce_statfs == 0)
3719 if (pr->pr_root->v_mount == mp)
3721 if (pr->pr_enforce_statfs == 2)
3724 * If jail's chroot directory is set to "/" we should be able to see
3725 * all mount-points from inside a jail.
3726 * This is ugly check, but this is the only situation when jail's
3727 * directory ends with '/'.
3729 if (strcmp(pr->pr_path, "/") == 0)
3731 len = strlen(pr->pr_path);
3733 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3736 * Be sure that we don't have situation where jail's root directory
3737 * is "/some/path" and mount point is "/some/pathpath".
3739 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3745 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3747 char jpath[MAXPATHLEN];
3751 pr = cred->cr_prison;
3752 if (pr->pr_enforce_statfs == 0)
3754 if (prison_canseemount(cred, mp) != 0) {
3755 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3756 strlcpy(sp->f_mntonname, "[restricted]",
3757 sizeof(sp->f_mntonname));
3760 if (pr->pr_root->v_mount == mp) {
3762 * Clear current buffer data, so we are sure nothing from
3763 * the valid path left there.
3765 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3766 *sp->f_mntonname = '/';
3770 * If jail's chroot directory is set to "/" we should be able to see
3771 * all mount-points from inside a jail.
3773 if (strcmp(pr->pr_path, "/") == 0)
3775 len = strlen(pr->pr_path);
3776 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3778 * Clear current buffer data, so we are sure nothing from
3779 * the valid path left there.
3781 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3782 if (*jpath == '\0') {
3783 /* Should never happen. */
3784 *sp->f_mntonname = '/';
3786 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3791 * Check with permission for a specific privilege is granted within jail. We
3792 * have a specific list of accepted privileges; the rest are denied.
3795 prison_priv_check(struct ucred *cred, int priv)
3801 * Some policies have custom handlers. This routine should not be
3802 * called for them. See priv_check_cred().
3805 case PRIV_VFS_LOOKUP:
3806 case PRIV_VFS_GENERATION:
3807 KASSERT(0, ("prison_priv_check instead of a custom handler "
3808 "called for %d\n", priv));
3816 * Privileges specific to prisons with a virtual network stack.
3817 * There might be a duplicate entry here in case the privilege
3818 * is only granted conditionally in the legacy jail case.
3822 * NFS-specific privileges.
3824 case PRIV_NFS_DAEMON:
3825 case PRIV_VFS_GETFH:
3826 case PRIV_VFS_MOUNT_EXPORTED:
3827 if (!prison_check_nfsd(cred))
3830 case PRIV_NFS_LOCKD:
3833 * Network stack privileges.
3835 case PRIV_NET_BRIDGE:
3838 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3839 case PRIV_NET_ROUTE:
3841 case PRIV_NET_SETIFMTU:
3842 case PRIV_NET_SETIFFLAGS:
3843 case PRIV_NET_SETIFCAP:
3844 case PRIV_NET_SETIFDESCR:
3845 case PRIV_NET_SETIFNAME :
3846 case PRIV_NET_SETIFMETRIC:
3847 case PRIV_NET_SETIFPHYS:
3848 case PRIV_NET_SETIFMAC:
3849 case PRIV_NET_SETLANPCP:
3850 case PRIV_NET_ADDMULTI:
3851 case PRIV_NET_DELMULTI:
3852 case PRIV_NET_HWIOCTL:
3853 case PRIV_NET_SETLLADDR:
3854 case PRIV_NET_ADDIFGROUP:
3855 case PRIV_NET_DELIFGROUP:
3856 case PRIV_NET_IFCREATE:
3857 case PRIV_NET_IFDESTROY:
3858 case PRIV_NET_ADDIFADDR:
3859 case PRIV_NET_DELIFADDR:
3862 case PRIV_NET_SETIFVNET:
3863 case PRIV_NET_SETIFFIB:
3869 * 802.11-related privileges.
3871 case PRIV_NET80211_VAP_GETKEY:
3872 case PRIV_NET80211_VAP_MANAGE:
3878 case PRIV_NETATM_CFG:
3879 case PRIV_NETATM_ADD:
3880 case PRIV_NETATM_DEL:
3881 case PRIV_NETATM_SET:
3884 * Bluetooth privileges.
3886 case PRIV_NETBLUETOOTH_RAW:
3890 * Netgraph and netgraph module privileges.
3892 case PRIV_NETGRAPH_CONTROL:
3894 case PRIV_NETGRAPH_TTY:
3898 * IPv4 and IPv6 privileges.
3900 case PRIV_NETINET_IPFW:
3901 case PRIV_NETINET_DIVERT:
3902 case PRIV_NETINET_PF:
3903 case PRIV_NETINET_DUMMYNET:
3904 case PRIV_NETINET_CARP:
3905 case PRIV_NETINET_MROUTE:
3906 case PRIV_NETINET_RAW:
3907 case PRIV_NETINET_ADDRCTRL6:
3908 case PRIV_NETINET_ND6:
3909 case PRIV_NETINET_SCOPE6:
3910 case PRIV_NETINET_ALIFETIME6:
3911 case PRIV_NETINET_IPSEC:
3912 case PRIV_NETINET_BINDANY:
3927 * No default: or deny here.
3928 * In case of no permit fall through to next switch().
3930 if (cred->cr_prison->pr_flags & PR_VNET)
3937 * Allow ktrace privileges for root in jail.
3943 * Allow jailed processes to configure audit identity and
3944 * submit audit records (login, etc). In the future we may
3945 * want to further refine the relationship between audit and
3948 case PRIV_AUDIT_GETAUDIT:
3949 case PRIV_AUDIT_SETAUDIT:
3950 case PRIV_AUDIT_SUBMIT:
3954 * Allow jailed processes to manipulate process UNIX
3955 * credentials in any way they see fit.
3957 case PRIV_CRED_SETUID:
3958 case PRIV_CRED_SETEUID:
3959 case PRIV_CRED_SETGID:
3960 case PRIV_CRED_SETEGID:
3961 case PRIV_CRED_SETGROUPS:
3962 case PRIV_CRED_SETREUID:
3963 case PRIV_CRED_SETREGID:
3964 case PRIV_CRED_SETRESUID:
3965 case PRIV_CRED_SETRESGID:
3968 * Jail implements visibility constraints already, so allow
3969 * jailed root to override uid/gid-based constraints.
3971 case PRIV_SEEOTHERGIDS:
3972 case PRIV_SEEOTHERUIDS:
3973 case PRIV_SEEJAILPROC:
3976 * Jail implements inter-process debugging limits already, so
3977 * allow jailed root various debugging privileges.
3979 case PRIV_DEBUG_DIFFCRED:
3980 case PRIV_DEBUG_SUGID:
3981 case PRIV_DEBUG_UNPRIV:
3984 * Allow jail to set various resource limits and login
3985 * properties, and for now, exceed process resource limits.
3987 case PRIV_PROC_LIMIT:
3988 case PRIV_PROC_SETLOGIN:
3989 case PRIV_PROC_SETRLIMIT:
3992 * System V and POSIX IPC privileges are granted in jail.
3995 case PRIV_IPC_WRITE:
3996 case PRIV_IPC_ADMIN:
3997 case PRIV_IPC_MSGSIZE:
4001 * Jail operations within a jail work on child jails.
4003 case PRIV_JAIL_ATTACH:
4005 case PRIV_JAIL_REMOVE:
4008 * Jail implements its own inter-process limits, so allow
4009 * root processes in jail to change scheduling on other
4010 * processes in the same jail. Likewise for signalling.
4012 case PRIV_SCHED_DIFFCRED:
4013 case PRIV_SCHED_CPUSET:
4014 case PRIV_SIGNAL_DIFFCRED:
4015 case PRIV_SIGNAL_SUGID:
4018 * Allow jailed processes to write to sysctls marked as jail
4021 case PRIV_SYSCTL_WRITEJAIL:
4024 * Allow root in jail to manage a variety of quota
4025 * properties. These should likely be conditional on a
4026 * configuration option.
4028 case PRIV_VFS_GETQUOTA:
4029 case PRIV_VFS_SETQUOTA:
4032 * Since Jail relies on chroot() to implement file system
4033 * protections, grant many VFS privileges to root in jail.
4034 * Be careful to exclude mount-related and NFS-related
4038 case PRIV_VFS_WRITE:
4039 case PRIV_VFS_ADMIN:
4041 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4042 case PRIV_VFS_CHFLAGS_DEV:
4043 case PRIV_VFS_CHOWN:
4044 case PRIV_VFS_CHROOT:
4045 case PRIV_VFS_RETAINSUGID:
4046 case PRIV_VFS_FCHROOT:
4048 case PRIV_VFS_SETGID:
4050 case PRIV_VFS_STICKYFILE:
4053 * As in the non-jail case, non-root users are expected to be
4054 * able to read kernel/physical memory (provided /dev/[k]mem
4055 * exists in the jail and they have permission to access it).
4057 case PRIV_KMEM_READ:
4061 * Depending on the global setting, allow privilege of
4062 * setting system flags.
4064 case PRIV_VFS_SYSFLAGS:
4065 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4071 * Depending on the global setting, allow privilege of
4072 * mounting/unmounting file systems.
4074 case PRIV_VFS_MOUNT:
4075 case PRIV_VFS_UNMOUNT:
4076 case PRIV_VFS_MOUNT_NONUSER:
4077 case PRIV_VFS_MOUNT_OWNER:
4078 pr = cred->cr_prison;
4080 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4088 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4089 * policy. priv_check_cred will not specifically allow it, and
4090 * we may want a MAC policy to allow it.
4092 case PRIV_VFS_READ_DIR:
4096 * Conditionally allow privileged process in the jail to
4097 * manipulate filesystem extended attributes in the system
4100 case PRIV_VFS_EXTATTR_SYSTEM:
4101 if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4107 * Conditionnaly allow locking (unlocking) physical pages
4111 case PRIV_VM_MUNLOCK:
4112 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4118 * Conditionally allow jailed root to bind reserved ports.
4120 case PRIV_NETINET_RESERVEDPORT:
4121 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4127 * Allow jailed root to reuse in-use ports.
4129 case PRIV_NETINET_REUSEPORT:
4133 * Allow jailed root to set certain IPv4/6 (option) headers.
4135 case PRIV_NETINET_SETHDROPTS:
4139 * Conditionally allow creating raw sockets in jail.
4141 case PRIV_NETINET_RAW:
4142 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4148 * Since jail implements its own visibility limits on netstat
4149 * sysctls, allow getcred. This allows identd to work in
4152 case PRIV_NETINET_GETCRED:
4156 * Allow jailed root to set loginclass.
4158 case PRIV_PROC_SETLOGINCLASS:
4162 * Do not allow a process inside a jail to read the kernel
4163 * message buffer unless explicitly permitted.
4166 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4172 * In all remaining cases, deny the privilege request. This
4173 * includes almost all network privileges, many system
4174 * configuration privileges.
4181 * Return the part of pr2's name that is relative to pr1, or the whole name
4182 * if it does not directly follow.
4186 prison_name(struct prison *pr1, struct prison *pr2)
4190 /* Jails see themselves as "0" (if they see themselves at all). */
4193 name = pr2->pr_name;
4194 if (prison_ischild(pr1, pr2)) {
4196 * pr1 isn't locked (and allprison_lock may not be either)
4197 * so its length can't be counted on. But the number of dots
4198 * can be counted on - and counted.
4200 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4201 name = strchr(name, '.') + 1;
4207 * Return the part of pr2's path that is relative to pr1, or the whole path
4208 * if it does not directly follow.
4211 prison_path(struct prison *pr1, struct prison *pr2)
4213 char *path1, *path2;
4216 path1 = pr1->pr_path;
4217 path2 = pr2->pr_path;
4218 if (!strcmp(path1, "/"))
4220 len1 = strlen(path1);
4221 if (strncmp(path1, path2, len1))
4223 if (path2[len1] == '\0')
4225 if (path2[len1] == '/')
4226 return (path2 + len1);
4231 * Jail-related sysctls.
4233 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4236 #if defined(INET) || defined(INET6)
4238 * Copy address array to memory that would be then SYSCTL_OUT-ed.
4239 * sysctl_jail_list() helper.
4242 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4244 const struct prison_ip *pip;
4245 const size_t size = pr_families[af].size;
4248 mtx_assert(&pr->pr_mtx, MA_OWNED);
4249 if ((pip = pr->pr_addrs[af]) != NULL) {
4250 if (*len < pip->ips) {
4252 mtx_unlock(&pr->pr_mtx);
4253 *out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4254 mtx_lock(&pr->pr_mtx);
4257 bcopy(pip->pr_ip, *out, pip->ips * size);
4263 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4266 struct prison *pr, *cpr;
4268 struct in_addr *ip4 = NULL;
4272 struct in6_addr *ip6 = NULL;
4277 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4278 pr = req->td->td_ucred->cr_prison;
4280 sx_slock(&allprison_lock);
4281 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4282 mtx_lock(&cpr->pr_mtx);
4284 prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4287 prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4289 bzero(xp, sizeof(*xp));
4290 xp->pr_version = XPRISON_VERSION;
4291 xp->pr_id = cpr->pr_id;
4292 xp->pr_state = cpr->pr_state;
4293 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4294 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4295 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4302 mtx_unlock(&cpr->pr_mtx);
4303 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4307 if (xp->pr_ip4s > 0) {
4308 error = SYSCTL_OUT(req, ip4,
4309 xp->pr_ip4s * sizeof(struct in_addr));
4315 if (xp->pr_ip6s > 0) {
4316 error = SYSCTL_OUT(req, ip6,
4317 xp->pr_ip6s * sizeof(struct in6_addr));
4323 sx_sunlock(&allprison_lock);
4334 SYSCTL_OID(_security_jail, OID_AUTO, list,
4335 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4336 sysctl_jail_list, "S", "List of active jails");
4339 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4343 injail = jailed(req->td->td_ucred);
4344 error = SYSCTL_OUT(req, &injail, sizeof(injail));
4349 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4350 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4351 sysctl_jail_jailed, "I", "Process in jail?");
4354 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4356 int error, havevnet;
4358 struct ucred *cred = req->td->td_ucred;
4360 havevnet = jailed(cred) && prison_owns_vnet(cred);
4364 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4369 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4370 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4371 sysctl_jail_vnet, "I", "Jail owns vnet?");
4373 #if defined(INET) || defined(INET6)
4374 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4375 &jail_max_af_ips, 0,
4376 "Number of IP addresses a jail may have at most per address family (deprecated)");
4380 * Default parameters for jail(2) compatibility. For historical reasons,
4381 * the sysctl names have varying similarity to the parameter names. Prisons
4382 * just see their own parameters, and can't change them.
4385 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4389 /* Get the current flag value, and convert it to a boolean. */
4390 if (req->td->td_ucred->cr_prison == &prison0) {
4391 mtx_lock(&prison0.pr_mtx);
4392 i = (jail_default_allow & arg2) != 0;
4393 mtx_unlock(&prison0.pr_mtx);
4395 i = prison_allow(req->td->td_ucred, arg2);
4399 error = sysctl_handle_int(oidp, &i, 0, req);
4400 if (error || !req->newptr)
4406 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4409 mtx_lock(&prison0.pr_mtx);
4410 jail_default_allow = (jail_default_allow & ~arg2) | i;
4411 mtx_unlock(&prison0.pr_mtx);
4415 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4416 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4417 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4418 "Processes in jail can set their hostnames (deprecated)");
4419 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4420 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4421 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4422 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4423 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4424 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4425 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4426 "Processes in jail can use System V IPC primitives (deprecated)");
4427 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4428 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4429 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4430 "Prison root can create raw sockets (deprecated)");
4431 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4432 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4433 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4434 "Processes in jail can alter system file flags (deprecated)");
4435 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4436 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4437 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4438 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4441 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4446 pr = req->td->td_ucred->cr_prison;
4447 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4448 error = sysctl_handle_int(oidp, &level, 0, req);
4449 if (error || !req->newptr)
4451 *(int *)arg1 = level;
4455 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4456 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4457 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4458 sysctl_jail_default_level, "I",
4459 "Processes in jail cannot see all mounted file systems (deprecated)");
4461 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4462 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4463 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4464 sysctl_jail_default_level, "I",
4465 "Ruleset for the devfs filesystem in jail (deprecated)");
4468 * Nodes to describe jail parameters. Maximum length of string parameters
4469 * is returned in the string itself, and the other parameters exist merely
4470 * to make themselves and their types known.
4472 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4476 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4483 switch (oidp->oid_kind & CTLTYPE)
4489 if (!(req->flags & SCTL_MASK32))
4491 return (SYSCTL_OUT(req, &l, sizeof(l)));
4495 return (SYSCTL_OUT(req, &i, sizeof(i)));
4496 case CTLTYPE_STRING:
4497 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4499 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4500 case CTLTYPE_STRUCT:
4502 return (SYSCTL_OUT(req, &s, sizeof(s)));
4508 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4509 * jail creation time but cannot be changed in an existing jail.
4511 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4512 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4513 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4514 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4515 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4516 "I", "Jail secure level");
4517 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4518 "Jail value for kern.osreldate and uname -K");
4519 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4520 "Jail value for kern.osrelease and uname -r");
4521 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4522 "I", "Jail cannot see all mounted file systems");
4523 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4524 "I", "Ruleset for in-jail devfs mounts");
4525 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4526 "B", "Jail persistence");
4528 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4529 "E,jailsys", "Virtual network stack");
4531 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4532 "B", "Jail is in the process of shutting down");
4534 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4535 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4536 "I", "Current number of child jails");
4537 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4538 "I", "Maximum number of child jails");
4540 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4541 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4543 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4544 "Jail NIS domainname");
4545 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4547 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4548 "LU", "Jail host ID");
4550 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4551 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4554 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4555 "Jail IPv4 address virtualization");
4556 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4557 "S,in_addr,a", "Jail IPv4 addresses");
4558 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4559 "B", "Do (not) use IPv4 source address selection rather than the "
4560 "primary jail IPv4 address.");
4563 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4564 "Jail IPv6 address virtualization");
4565 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4566 "S,in6_addr,a", "Jail IPv6 addresses");
4567 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4568 "B", "Do (not) use IPv6 source address selection rather than the "
4569 "primary jail IPv6 address.");
4572 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4573 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4574 "B", "Jail may set hostname");
4575 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4576 "B", "Jail may use SYSV IPC");
4577 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4578 "B", "Jail may create raw sockets");
4579 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4580 "B", "Jail may alter system file flags");
4581 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4582 "B", "Jail may set file quotas");
4583 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4584 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4585 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4586 "B", "Jail may lock (unlock) physical pages in memory");
4587 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4588 "B", "Jail may bind sockets to reserved ports");
4589 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4590 "B", "Jail may read the kernel message buffer");
4591 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4592 "B", "Unprivileged processes may use process debugging facilities");
4593 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4594 "B", "Processes in jail with uid 0 have privilege");
4596 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
4597 "B", "Mountd/nfsd may run in the jail");
4599 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
4600 "B", "Jail may set system-level filesystem extended attributes");
4602 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4603 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4604 "B", "Jail may mount/unmount jail-friendly file systems in general");
4607 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
4608 * its associated bit in the pr_allow bitmask, or zero if the parameter was
4612 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4615 struct bool_flags *bf;
4616 struct sysctl_oid *parent;
4617 char *allow_name, *allow_noname, *allowed;
4618 #ifndef NO_SYSCTL_DESCR
4619 char *descr_deprecated;
4624 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4626 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4628 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4629 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4630 free(allow_name, M_PRISON);
4635 * See if this parameter has already beed added, i.e. a module was
4636 * previously loaded/unloaded.
4638 mtx_lock(&prison0.pr_mtx);
4639 for (bf = pr_flag_allow;
4640 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4641 atomic_load_int(&bf->flag) != 0;
4643 if (strcmp(bf->name, allow_name) == 0) {
4644 allow_flag = bf->flag;
4650 * Find a free bit in pr_allow_all, failing if there are none
4651 * (which shouldn't happen as long as we keep track of how many
4652 * potential dynamic flags exist).
4654 for (allow_flag = 1;; allow_flag <<= 1) {
4655 if (allow_flag == 0)
4657 if ((pr_allow_all & allow_flag) == 0)
4661 /* Note the parameter in the next open slot in pr_flag_allow. */
4662 for (bf = pr_flag_allow; ; bf++) {
4663 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4664 /* This should never happen, but is not fatal. */
4668 if (atomic_load_int(&bf->flag) == 0)
4671 bf->name = allow_name;
4672 bf->noname = allow_noname;
4673 pr_allow_all |= allow_flag;
4675 * prison0 always has permission for the new parameter.
4676 * Other jails must have it granted to them.
4678 prison0.pr_allow |= allow_flag;
4679 /* The flag indicates a valid entry, so make sure it is set last. */
4680 atomic_store_rel_int(&bf->flag, allow_flag);
4681 mtx_unlock(&prison0.pr_mtx);
4684 * Create sysctls for the parameter, and the back-compat global
4688 ? SYSCTL_ADD_NODE(NULL,
4689 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4690 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4691 : &sysctl___security_jail_param_allow;
4692 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4693 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4694 NULL, 0, sysctl_jail_param, "B", descr);
4696 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4697 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4698 #ifndef NO_SYSCTL_DESCR
4699 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4702 (void)SYSCTL_ADD_PROC(NULL,
4703 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4704 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4705 sysctl_jail_default_allow, "I", descr_deprecated);
4706 #ifndef NO_SYSCTL_DESCR
4707 free(descr_deprecated, M_TEMP);
4709 free(allowed, M_TEMP);
4714 mtx_unlock(&prison0.pr_mtx);
4715 free(allow_name, M_PRISON);
4716 free(allow_noname, M_PRISON);
4721 * The VFS system will register jail-aware filesystems here. They each get
4722 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4723 * attempts to mount.
4726 prison_add_vfs(struct vfsconf *vfsp)
4728 #ifdef NO_SYSCTL_DESCR
4730 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4735 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4737 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4739 free(descr, M_TEMP);
4745 prison_racct_foreach(void (*callback)(struct racct *racct,
4746 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4747 void *arg2, void *arg3)
4749 struct prison_racct *prr;
4751 ASSERT_RACCT_ENABLED();
4753 sx_slock(&allprison_lock);
4756 LIST_FOREACH(prr, &allprison_racct, prr_next)
4757 (callback)(prr->prr_racct, arg2, arg3);
4760 sx_sunlock(&allprison_lock);
4763 static struct prison_racct *
4764 prison_racct_find_locked(const char *name)
4766 struct prison_racct *prr;
4768 ASSERT_RACCT_ENABLED();
4769 sx_assert(&allprison_lock, SA_XLOCKED);
4771 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4774 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4775 if (strcmp(name, prr->prr_name) != 0)
4778 /* Found prison_racct with a matching name? */
4779 prison_racct_hold(prr);
4783 /* Add new prison_racct. */
4784 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4785 racct_create(&prr->prr_racct);
4787 strcpy(prr->prr_name, name);
4788 refcount_init(&prr->prr_refcount, 1);
4789 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4794 struct prison_racct *
4795 prison_racct_find(const char *name)
4797 struct prison_racct *prr;
4799 ASSERT_RACCT_ENABLED();
4801 sx_xlock(&allprison_lock);
4802 prr = prison_racct_find_locked(name);
4803 sx_xunlock(&allprison_lock);
4808 prison_racct_hold(struct prison_racct *prr)
4811 ASSERT_RACCT_ENABLED();
4813 refcount_acquire(&prr->prr_refcount);
4817 prison_racct_free_locked(struct prison_racct *prr)
4820 ASSERT_RACCT_ENABLED();
4821 sx_assert(&allprison_lock, SA_XLOCKED);
4823 if (refcount_release(&prr->prr_refcount)) {
4824 racct_destroy(&prr->prr_racct);
4825 LIST_REMOVE(prr, prr_next);
4826 free(prr, M_PRISON_RACCT);
4831 prison_racct_free(struct prison_racct *prr)
4834 ASSERT_RACCT_ENABLED();
4835 sx_assert(&allprison_lock, SA_UNLOCKED);
4837 if (refcount_release_if_not_last(&prr->prr_refcount))
4840 sx_xlock(&allprison_lock);
4841 prison_racct_free_locked(prr);
4842 sx_xunlock(&allprison_lock);
4846 prison_racct_attach(struct prison *pr)
4848 struct prison_racct *prr;
4850 ASSERT_RACCT_ENABLED();
4851 sx_assert(&allprison_lock, SA_XLOCKED);
4853 prr = prison_racct_find_locked(pr->pr_name);
4854 KASSERT(prr != NULL, ("cannot find prison_racct"));
4856 pr->pr_prison_racct = prr;
4860 * Handle jail renaming. From the racct point of view, renaming means
4861 * moving from one prison_racct to another.
4864 prison_racct_modify(struct prison *pr)
4870 struct prison_racct *oldprr;
4872 ASSERT_RACCT_ENABLED();
4874 sx_slock(&allproc_lock);
4875 sx_xlock(&allprison_lock);
4877 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4878 sx_xunlock(&allprison_lock);
4879 sx_sunlock(&allproc_lock);
4883 oldprr = pr->pr_prison_racct;
4884 pr->pr_prison_racct = NULL;
4886 prison_racct_attach(pr);
4889 * Move resource utilisation records.
4891 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4895 * Force rctl to reattach rules to processes.
4897 FOREACH_PROC_IN_SYSTEM(p) {
4899 cred = crhold(p->p_ucred);
4901 rctl_proc_ucred_changed(p, cred);
4906 sx_sunlock(&allproc_lock);
4907 prison_racct_free_locked(oldprr);
4908 sx_xunlock(&allprison_lock);
4912 prison_racct_detach(struct prison *pr)
4915 ASSERT_RACCT_ENABLED();
4916 sx_assert(&allprison_lock, SA_UNLOCKED);
4918 if (pr->pr_prison_racct == NULL)
4920 prison_racct_free(pr->pr_prison_racct);
4921 pr->pr_prison_racct = NULL;
4928 db_show_prison(struct prison *pr)
4930 struct bool_flags *bf;
4931 struct jailsys_flags *jsf;
4932 #if defined(INET) || defined(INET6)
4934 struct prison_ip *pip;
4938 char ip4buf[INET_ADDRSTRLEN];
4941 char ip6buf[INET6_ADDRSTRLEN];
4944 db_printf("prison %p:\n", pr);
4945 db_printf(" jid = %d\n", pr->pr_id);
4946 db_printf(" name = %s\n", pr->pr_name);
4947 db_printf(" parent = %p\n", pr->pr_parent);
4948 db_printf(" ref = %d\n", pr->pr_ref);
4949 db_printf(" uref = %d\n", pr->pr_uref);
4950 db_printf(" state = %s\n",
4951 pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4952 pr->pr_state == PRISON_STATE_DYING ? "dying" :
4954 db_printf(" path = %s\n", pr->pr_path);
4955 db_printf(" cpuset = %d\n", pr->pr_cpuset
4956 ? pr->pr_cpuset->cs_id : -1);
4958 db_printf(" vnet = %p\n", pr->pr_vnet);
4960 db_printf(" root = %p\n", pr->pr_root);
4961 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4962 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4963 db_printf(" children.max = %d\n", pr->pr_childmax);
4964 db_printf(" children.cur = %d\n", pr->pr_childcount);
4965 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4966 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4967 db_printf(" flags = 0x%x", pr->pr_flags);
4968 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4969 if (pr->pr_flags & bf->flag)
4970 db_printf(" %s", bf->name);
4971 for (jsf = pr_flag_jailsys;
4972 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4974 f = pr->pr_flags & (jsf->disable | jsf->new);
4975 db_printf(" %-16s= %s\n", jsf->name,
4976 (f != 0 && f == jsf->disable) ? "disable"
4977 : (f == jsf->new) ? "new"
4980 db_printf(" allow = 0x%x", pr->pr_allow);
4981 for (bf = pr_flag_allow;
4982 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4983 atomic_load_int(&bf->flag) != 0;
4985 if (pr->pr_allow & bf->flag)
4986 db_printf(" %s", bf->name);
4988 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4989 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4990 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4991 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4992 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4994 if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
4995 db_printf(" ip4s = %d\n", pip->ips);
4996 for (ii = 0; ii < pip->ips; ii++)
4997 db_printf(" %s %s\n",
4998 ii == 0 ? "ip4.addr =" : " ",
5000 *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5005 if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5006 db_printf(" ip6s = %d\n", pip->ips);
5007 for (ii = 0; ii < pip->ips; ii++)
5008 db_printf(" %s %s\n",
5009 ii == 0 ? "ip6.addr =" : " ",
5011 (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5016 DB_SHOW_COMMAND(prison, db_show_prison_command)
5022 * Show all prisons in the list, and prison0 which is not
5025 db_show_prison(&prison0);
5026 if (!db_pager_quit) {
5027 TAILQ_FOREACH(pr, &allprison, pr_list) {
5039 /* Look for a prison with the ID and with references. */
5040 TAILQ_FOREACH(pr, &allprison, pr_list)
5041 if (pr->pr_id == addr && pr->pr_ref > 0)
5044 /* Look again, without requiring a reference. */
5045 TAILQ_FOREACH(pr, &allprison, pr_list)
5046 if (pr->pr_id == addr)
5049 /* Assume address points to a valid prison. */
5050 pr = (struct prison *)addr;