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>
32 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/kernel.h>
42 #include <sys/systm.h>
43 #include <sys/errno.h>
44 #include <sys/sysproto.h>
45 #include <sys/malloc.h>
49 #include <sys/epoch.h>
50 #include <sys/taskqueue.h>
51 #include <sys/fcntl.h>
53 #include <sys/linker.h>
56 #include <sys/mutex.h>
57 #include <sys/racct.h>
59 #include <sys/refcount.h>
61 #include <sys/sysent.h>
62 #include <sys/namei.h>
63 #include <sys/mount.h>
64 #include <sys/queue.h>
65 #include <sys/socket.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
69 #include <sys/vnode.h>
74 #include <netinet/in.h>
80 #include <security/mac/mac_framework.h>
82 #define PRISON0_HOSTUUID_MODULE "hostuuid"
84 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
85 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
87 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
90 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
92 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
96 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
98 #define _PR_IP_SADDRSEL 0
102 /* prison0 describes what is "real" about the system. */
103 struct prison prison0 = {
109 .pr_securelevel = -1,
111 .pr_state = PRISON_STATE_ALIVE,
112 .pr_childmax = JAIL_MAX,
113 .pr_hostuuid = DEFAULT_HOSTUUID,
114 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
116 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
118 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
120 .pr_allow = PR_ALLOW_ALL_STATIC,
122 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
129 struct jailsys_flags {
135 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
136 struct sx allprison_lock;
137 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
138 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
139 LIST_HEAD(, prison_racct) allprison_racct;
142 static int get_next_prid(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},
226 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
227 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
229 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
230 PR_ALLOW_RESERVED_PORTS | \
231 PR_ALLOW_UNPRIV_DEBUG | \
233 #define JAIL_DEFAULT_ENFORCE_STATFS 2
234 #define JAIL_DEFAULT_DEVFS_RSNUM 0
235 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
236 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
237 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
238 #if defined(INET) || defined(INET6)
239 static unsigned jail_max_af_ips = 255;
243 * Initialize the parts of prison0 that can't be static-initialized with
244 * constants. This is called from proc0_init() after creating thread0 cpuset.
249 uint8_t *file, *data;
251 char buf[sizeof(prison0.pr_hostuuid)];
254 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
255 prison0.pr_osreldate = osreldate;
256 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
258 /* If we have a preloaded hostuuid, use it. */
259 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
261 data = preload_fetch_addr(file);
262 size = preload_fetch_size(file);
265 * The preloaded data may include trailing whitespace, almost
266 * certainly a newline; skip over any whitespace or
267 * non-printable characters to be safe.
269 while (size > 0 && data[size - 1] <= 0x20) {
276 * Not NUL-terminated when passed from loader, but
277 * validate_uuid requires that due to using sscanf (as
278 * does the subsequent strlcpy, since it still reads
279 * past the given size to return the true length);
280 * bounce to a temporary buffer to fix.
282 if (size >= sizeof(buf))
285 memcpy(buf, data, size);
288 if (validate_uuid(buf, size, NULL, 0) != 0)
292 (void)strlcpy(prison0.pr_hostuuid, buf,
293 sizeof(prison0.pr_hostuuid));
296 if (bootverbose && !valid) {
297 printf("hostuuid: preload data malformed: '%.*s'\n",
303 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
312 sys_jail(struct thread *td, struct jail_args *uap)
318 error = copyin(uap->jail, &version, sizeof(uint32_t));
327 /* FreeBSD single IPv4 jails. */
328 bzero(&j, sizeof(struct jail));
329 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
332 j.version = j0.version;
334 j.hostname = j0.hostname;
335 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
341 * Version 1 was used by multi-IPv4 jail implementations
342 * that never made it into the official kernel.
346 case 2: /* JAIL_API_VERSION */
347 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
348 error = copyin(uap->jail, &j, sizeof(struct jail));
354 /* Sci-Fi jails are not supported, sorry. */
357 return (kern_jail(td, &j));
361 kern_jail(struct thread *td, struct jail *j)
363 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
372 char *u_path, *u_hostname, *u_name;
373 struct bool_flags *bf;
376 struct in_addr *u_ip4;
379 struct in6_addr *u_ip6;
382 int error, enforce_statfs;
384 bzero(&optiov, sizeof(optiov));
385 opt.uio_iov = optiov;
389 opt.uio_segflg = UIO_SYSSPACE;
390 opt.uio_rw = UIO_READ;
393 /* Set permissions for top-level jails from sysctls. */
394 if (!jailed(td->td_ucred)) {
395 for (bf = pr_flag_allow;
396 bf < pr_flag_allow + nitems(pr_flag_allow) &&
397 atomic_load_int(&bf->flag) != 0;
399 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
400 (jail_default_allow & bf->flag)
401 ? bf->name : bf->noname);
402 optiov[opt.uio_iovcnt].iov_len =
403 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
406 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
407 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
409 enforce_statfs = jail_default_enforce_statfs;
410 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
411 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
415 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
417 ip4s = (j->version == 0) ? 1 : j->ip4s;
418 if (ip4s > jail_max_af_ips)
420 tmplen += ip4s * sizeof(struct in_addr);
426 if (j->ip6s > jail_max_af_ips)
428 tmplen += j->ip6s * sizeof(struct in6_addr);
433 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
434 u_hostname = u_path + MAXPATHLEN;
435 u_name = u_hostname + MAXHOSTNAMELEN;
437 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
441 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
443 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
446 optiov[opt.uio_iovcnt].iov_base = "path";
447 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
449 optiov[opt.uio_iovcnt].iov_base = u_path;
450 error = copyinstr(j->path, u_path, MAXPATHLEN,
451 &optiov[opt.uio_iovcnt].iov_len);
453 free(u_path, M_TEMP);
457 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
458 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
460 optiov[opt.uio_iovcnt].iov_base = u_hostname;
461 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
462 &optiov[opt.uio_iovcnt].iov_len);
464 free(u_path, M_TEMP);
468 if (j->jailname != NULL) {
469 optiov[opt.uio_iovcnt].iov_base = "name";
470 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
472 optiov[opt.uio_iovcnt].iov_base = u_name;
473 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
474 &optiov[opt.uio_iovcnt].iov_len);
476 free(u_path, M_TEMP);
482 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
483 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
485 optiov[opt.uio_iovcnt].iov_base = u_ip4;
486 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
488 u_ip4->s_addr = j->ip4s;
490 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
492 free(u_path, M_TEMP);
499 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
500 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
502 optiov[opt.uio_iovcnt].iov_base = u_ip6;
503 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
504 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
506 free(u_path, M_TEMP);
511 KASSERT(opt.uio_iovcnt <= nitems(optiov),
512 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
513 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
514 free(u_path, M_TEMP);
519 * struct jail_set_args {
520 * struct iovec *iovp;
521 * unsigned int iovcnt;
526 sys_jail_set(struct thread *td, struct jail_set_args *uap)
531 /* Check that we have an even number of iovecs. */
535 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
538 error = kern_jail_set(td, auio, uap->flags);
543 #if defined(INET) || defined(INET6)
544 typedef int prison_addr_cmp_t(const void *, const void *);
545 typedef bool prison_addr_valid_t(const void *);
546 static const struct pr_family {
548 prison_addr_cmp_t *cmp;
549 prison_addr_valid_t *valid;
551 } pr_families[PR_FAMILY_MAX] = {
554 .size = sizeof(struct in_addr),
555 .cmp = prison_qcmp_v4,
556 .valid = prison_valid_v4,
557 .ip_flag = PR_IP4_USER,
562 .size = sizeof(struct in6_addr),
563 .cmp = prison_qcmp_v6,
564 .valid = prison_valid_v6,
565 .ip_flag = PR_IP6_USER,
571 * Network address lists (pr_addrs) allocation for jails. The addresses
572 * are accessed locklessly by the network stack, thus need to be protected by
576 struct epoch_context ctx;
580 * XXX Variable-length automatic arrays in union may be
581 * supported in future C.
585 struct in_addr pr_ip4[];
586 struct in6_addr pr_ip6[];
588 #else /* No future C :( */
594 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
597 MPASS(af < PR_FAMILY_MAX);
598 MPASS(idx >= 0 && idx < pip->ips);
600 return (pip->pr_ip + pr_families[af].size * idx);
603 static struct prison_ip *
604 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
606 struct prison_ip *pip;
608 pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
616 * Allocate and copyin user supplied address list, sorting and validating.
617 * kern_jail_set() helper.
619 static struct prison_ip *
620 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
622 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
623 const size_t size = pr_families[af].size;
624 struct prison_ip *pip;
626 pip = prison_ip_alloc(af, cnt, M_WAITOK);
627 bcopy(op, pip->pr_ip, cnt * size);
629 * IP addresses are all sorted but ip[0] to preserve
630 * the primary IP address as given from userland.
631 * This special IP is used for unbound outgoing
632 * connections as well for "loopback" traffic in case
633 * source address selection cannot find any more fitting
634 * address to connect from.
637 qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
639 * Check for duplicate addresses and do some simple
640 * zero and broadcast checks. If users give other bogus
641 * addresses it is their problem.
643 for (int i = 0; i < cnt; i++) {
644 if (!pr_families[af].valid(PR_IP(pip, af, i))) {
649 (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
650 cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
660 * Allocate and dup parent prison address list.
661 * kern_jail_set() helper.
664 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
666 const struct prison_ip *ppip = ppr->pr_addrs[af];
667 struct prison_ip *pip;
670 pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
671 bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
672 pr->pr_addrs[af] = pip;
677 * Make sure the new set of IP addresses is a subset of the parent's list.
678 * Don't worry about the parent being unlocked, as any setting is done with
679 * allprison_lock held.
680 * kern_jail_set() helper.
683 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
684 const pr_family_t af)
686 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
692 for (i = 0; i < ppip->ips; i++)
693 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
697 /* Main address not present in parent. */
701 for (i = j = 1; i < pip->ips; i++) {
702 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
703 /* Equals to parent primary address. */
705 for (; j < ppip->ips; j++)
706 if (cmp(PR_IP(pip, af, i),
707 PR_IP(ppip, af, j)) == 0)
713 /* Address not present in parent. */
720 * Check for conflicting IP addresses. We permit them if there is no more
721 * than one IP on each jail. If there is a duplicate on a jail with more
722 * than one IP stop checking and return error.
723 * kern_jail_set() helper.
726 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
727 struct prison_ip *pip, pr_family_t af)
729 const struct prison *tppr, *tpr;
733 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
734 if (tppr->pr_flags & PR_VNET)
739 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
742 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
744 !prison_isalive(tpr)) {
748 if (!(tpr->pr_flags & pr_families[af].ip_flag))
751 if (tpr->pr_addrs[af] == NULL ||
752 (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
754 for (int i = 0; i < pip->ips; i++)
755 if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
762 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
763 "prison must start with epoch context");
765 prison_ip_free_deferred(epoch_context_t ctx)
772 prison_ip_free(struct prison_ip *pip)
776 NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
780 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
782 struct prison_ip **mem, *old;
784 mtx_assert(&pr->pr_mtx, MA_OWNED);
786 mem = &pr->pr_addrs[af];
789 atomic_store_ptr(mem, new);
794 * Restrict a prison's IP address list with its parent's, possibly replacing
795 * it. Return true if succeed, otherwise should redo.
796 * kern_jail_set() helper.
799 prison_ip_restrict(struct prison *pr, const pr_family_t af,
800 struct prison_ip **newp)
802 struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
803 struct prison_ip *pip = pr->pr_addrs[af];
804 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
805 const size_t size = pr_families[af].size;
806 struct prison_ip *new = newp != NULL ? *newp : NULL;
809 mtx_assert(&pr->pr_mtx, MA_OWNED);
812 * Due to epoch-synchronized access to the IP address lists we always
813 * allocate a new list even if the old one has enough space. We could
814 * atomically update an IPv4 address inside a list, but that would
815 * screw up sorting, and in case of IPv6 we can't even atomically write
820 prison_ip_set(pr, af, NULL);
824 if (!(pr->pr_flags & pr_families[af].ip_flag)) {
826 new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
828 return (false); /* Redo */
830 /* This has no user settings, so just copy the parent's list. */
831 MPASS(new->ips == ppip->ips);
832 bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
833 prison_ip_set(pr, af, new);
835 *newp = NULL; /* Used */
836 } else if (pip != NULL) {
837 /* Remove addresses that aren't in the parent. */
840 i = 0; /* index in pip */
841 ips = 0; /* index in new */
844 new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
846 return (false); /* Redo */
849 for (int pi = 0; pi < ppip->ips; pi++)
850 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
851 /* Found our primary address in parent. */
852 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
858 for (int pi = 1; i < pip->ips; ) {
859 /* Check against primary, which is unsorted. */
860 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
861 /* Matches parent's primary address. */
862 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
868 /* The rest are sorted. */
869 switch (pi >= ppip->ips ? -1 :
870 cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
875 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
887 if (newp == NULL || *newp == NULL)
891 /* Shrink to real size */
892 KASSERT((new->ips >= ips),
893 ("Out-of-bounds write to prison_ip %p", new));
896 prison_ip_set(pr, af, new);
898 *newp = NULL; /* Used */
904 * Fast-path check if an address belongs to a prison.
907 prison_ip_check(const struct prison *pr, const pr_family_t af,
910 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
911 struct prison_ip *pip;
914 MPASS(mtx_owned(&pr->pr_mtx) ||
915 in_epoch(net_epoch_preempt) ||
916 sx_xlocked(&allprison_lock));
918 pip = atomic_load_ptr(&pr->pr_addrs[af]);
919 if (__predict_false(pip == NULL))
920 return (EAFNOSUPPORT);
922 /* Check the primary IP. */
923 if (cmp(PR_IP(pip, af, 0), addr) == 0)
927 * All the other IPs are sorted so we can do a binary search.
933 d = cmp(PR_IP(pip, af, i + 1), addr);
942 return (EADDRNOTAVAIL);
946 * Grab primary IP. Historically required mutex, but nothing prevents
947 * us to support epoch-protected access. Is it used in fast path?
948 * in{6}_jail.c helper
951 prison_ip_get0(const struct prison *pr, const pr_family_t af)
953 const struct prison_ip *pip = pr->pr_addrs[af];
955 mtx_assert(&pr->pr_mtx, MA_OWNED);
962 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
965 return (pr->pr_addrs[af]->ips);
967 #endif /* defined(INET) || defined(INET6) */
970 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
974 struct prison_ip *ip4;
977 struct prison_ip *ip6;
980 struct vfsoptlist *opts;
981 struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
983 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
984 char *g_path, *osrelstr;
985 struct bool_flags *bf;
986 struct jailsys_flags *jsf;
987 #if defined(INET) || defined(INET6)
991 size_t namelen, onamelen, pnamelen;
992 int born, created, cuflags, descend, drflags, enforce;
993 int error, errmsg_len, errmsg_pos;
994 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
995 int jid, jsys, len, level;
996 int childmax, osreldt, rsnum, slevel;
1005 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1006 uint64_t pr_allow_diff;
1010 error = priv_check(td, PRIV_JAIL_SET);
1011 if (!error && (flags & JAIL_ATTACH))
1012 error = priv_check(td, PRIV_JAIL_ATTACH);
1015 mypr = td->td_ucred->cr_prison;
1016 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
1018 if (flags & ~JAIL_SET_MASK)
1022 * Check all the parameters before committing to anything. Not all
1023 * errors can be caught early, but we may as well try. Also, this
1024 * takes care of some expensive stuff (path lookup) before getting
1025 * the allprison lock.
1027 * XXX Jails are not filesystems, and jail parameters are not mount
1028 * options. But it makes more sense to re-use the vfsopt code
1029 * than duplicate it under a different name.
1031 error = vfs_buildopts(optuio, &opts);
1042 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1045 vfs_opterror(opts, "no valid operation (create or update)");
1049 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1050 if (error == ENOENT)
1052 else if (error != 0)
1055 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1056 if (error == ENOENT)
1058 else if (error != 0)
1064 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1065 if (error == ENOENT)
1067 else if (error != 0)
1072 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1073 if (error == ENOENT)
1075 else if (error != 0)
1077 else if (enforce < 0 || enforce > 2) {
1083 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1084 if (error == ENOENT)
1086 else if (error != 0)
1091 pr_flags = ch_flags = 0;
1092 for (bf = pr_flag_bool;
1093 bf < pr_flag_bool + nitems(pr_flag_bool);
1095 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1096 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1098 ch_flags |= pr_flags;
1099 for (jsf = pr_flag_jailsys;
1100 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1102 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1103 if (error == ENOENT)
1108 case JAIL_SYS_DISABLE:
1109 if (!jsf->disable) {
1113 pr_flags |= jsf->disable;
1116 pr_flags |= jsf->new;
1118 case JAIL_SYS_INHERIT:
1124 ch_flags |= jsf->new | jsf->disable;
1126 if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1127 && !(pr_flags & PR_PERSIST)) {
1129 vfs_opterror(opts, "new jail must persist or attach");
1133 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1135 vfs_opterror(opts, "vnet cannot be changed after creation");
1140 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1142 vfs_opterror(opts, "ip4 cannot be changed after creation");
1147 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1149 vfs_opterror(opts, "ip6 cannot be changed after creation");
1154 pr_allow = ch_allow = 0;
1155 for (bf = pr_flag_allow;
1156 bf < pr_flag_allow + nitems(pr_flag_allow) &&
1157 atomic_load_int(&bf->flag) != 0;
1159 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1160 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1162 ch_allow |= pr_allow;
1164 error = vfs_getopt(opts, "name", (void **)&name, &len);
1165 if (error == ENOENT)
1167 else if (error != 0)
1170 if (len == 0 || name[len - 1] != '\0') {
1174 if (len > MAXHOSTNAMELEN) {
1175 error = ENAMETOOLONG;
1180 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1181 if (error == ENOENT)
1183 else if (error != 0)
1186 ch_flags |= PR_HOST;
1187 pr_flags |= PR_HOST;
1188 if (len == 0 || host[len - 1] != '\0') {
1192 if (len > MAXHOSTNAMELEN) {
1193 error = ENAMETOOLONG;
1198 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1199 if (error == ENOENT)
1201 else if (error != 0)
1204 ch_flags |= PR_HOST;
1205 pr_flags |= PR_HOST;
1206 if (len == 0 || domain[len - 1] != '\0') {
1210 if (len > MAXHOSTNAMELEN) {
1211 error = ENAMETOOLONG;
1216 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1217 if (error == ENOENT)
1219 else if (error != 0)
1222 ch_flags |= PR_HOST;
1223 pr_flags |= PR_HOST;
1224 if (len == 0 || uuid[len - 1] != '\0') {
1228 if (len > HOSTUUIDLEN) {
1229 error = ENAMETOOLONG;
1234 #ifdef COMPAT_FREEBSD32
1235 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1238 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1242 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1243 if (error == ENOENT)
1245 else if (error != 0)
1249 ch_flags |= PR_HOST;
1250 pr_flags |= PR_HOST;
1254 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1255 if (error == ENOENT)
1257 else if (error != 0)
1259 else if (ip4s & (sizeof(struct in_addr) - 1)) {
1263 ch_flags |= PR_IP4_USER;
1264 pr_flags |= PR_IP4_USER;
1266 ip4s /= sizeof(struct in_addr);
1267 if (ip4s > jail_max_af_ips) {
1269 vfs_opterror(opts, "too many IPv4 addresses");
1272 ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1282 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1283 if (error == ENOENT)
1285 else if (error != 0)
1287 else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1291 ch_flags |= PR_IP6_USER;
1292 pr_flags |= PR_IP6_USER;
1294 ip6s /= sizeof(struct in6_addr);
1295 if (ip6s > jail_max_af_ips) {
1297 vfs_opterror(opts, "too many IPv6 addresses");
1300 ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1309 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1310 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1313 "vnet jails cannot have IP address restrictions");
1318 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1319 if (error == ENOENT)
1321 else if (error != 0)
1324 if (flags & JAIL_UPDATE) {
1327 "osrelease cannot be changed after creation");
1330 if (len == 0 || osrelstr[len - 1] != '\0') {
1334 if (len >= OSRELEASELEN) {
1335 error = ENAMETOOLONG;
1337 "osrelease string must be 1-%d bytes long",
1343 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1344 if (error == ENOENT)
1346 else if (error != 0)
1349 if (flags & JAIL_UPDATE) {
1352 "osreldate cannot be changed after creation");
1357 vfs_opterror(opts, "osreldate cannot be 0");
1363 error = vfs_getopt(opts, "path", (void **)&path, &len);
1364 if (error == ENOENT)
1366 else if (error != 0)
1369 if (flags & JAIL_UPDATE) {
1372 "path cannot be changed after creation");
1375 if (len == 0 || path[len - 1] != '\0') {
1379 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1385 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1386 strlcpy(g_path, path, MAXPATHLEN);
1387 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1391 /* exit on other errors */
1394 if (root->v_type != VDIR) {
1403 * Find the specified jail, or at least its parent.
1404 * This abuses the file error codes ENOENT and EEXIST.
1408 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1409 namelc = strrchr(name, '.');
1410 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1414 sx_xlock(&allprison_lock);
1415 drflags = PD_LIST_XLOCKED;
1417 if (!prison_isalive(ppr)) {
1418 /* This jail is dying. This process will surely follow. */
1425 vfs_opterror(opts, "negative jid");
1429 * See if a requested jid already exists. Keep track of
1430 * where it can be inserted later.
1432 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1433 if (inspr->pr_id < jid)
1435 if (inspr->pr_id > jid)
1438 mtx_lock(&pr->pr_mtx);
1439 drflags |= PD_LOCKED;
1444 /* Create: jid must not exist. */
1445 if (cuflags == JAIL_CREATE) {
1447 * Even creators that cannot see the jail will
1451 vfs_opterror(opts, "jail %d already exists",
1455 if (!prison_ischild(mypr, pr)) {
1457 * Updaters get ENOENT if they cannot see the
1458 * jail. This is true even for CREATE | UPDATE,
1459 * which normally cannot give this error.
1462 vfs_opterror(opts, "jail %d not found", jid);
1465 ppr = pr->pr_parent;
1466 if (!prison_isalive(ppr)) {
1468 vfs_opterror(opts, "jail %d is dying",
1472 if (!prison_isalive(pr)) {
1473 if (!(flags & JAIL_DYING)) {
1475 vfs_opterror(opts, "jail %d is dying",
1479 if ((flags & JAIL_ATTACH) ||
1480 (pr_flags & PR_PERSIST)) {
1482 * A dying jail might be resurrected
1483 * (via attach or persist), but first
1484 * it must determine if another jail
1485 * has claimed its name. Accomplish
1486 * this by implicitly re-setting the
1490 name = prison_name(mypr, pr);
1494 /* Update: jid must exist. */
1495 if (cuflags == JAIL_UPDATE) {
1497 vfs_opterror(opts, "jail %d not found", jid);
1503 * If the caller provided a name, look for a jail by that name.
1504 * This has different semantics for creates and updates keyed by jid
1505 * (where the name must not already exist in a different jail),
1506 * and updates keyed by the name itself (where the name must exist
1507 * because that is the jail being updated).
1511 namelc = strrchr(name, '.');
1516 * This is a hierarchical name. Split it into the
1517 * parent and child names, and make sure the parent
1518 * exists or matches an already found jail.
1521 if (strncmp(name, ppr->pr_name, namelc - name)
1522 || ppr->pr_name[namelc - name] != '\0') {
1525 "cannot change jail's parent");
1530 ppr = prison_find_name(mypr, name);
1534 "jail \"%s\" not found", name);
1537 mtx_unlock(&ppr->pr_mtx);
1538 if (!prison_isalive(ppr)) {
1541 "jail \"%s\" is dying", name);
1548 if (namelc[0] != '\0') {
1550 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1552 FOREACH_PRISON_CHILD(ppr, tpr) {
1554 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1555 if (prison_isalive(tpr)) {
1557 cuflags != JAIL_CREATE) {
1563 mtx_lock(&pr->pr_mtx);
1564 drflags |= PD_LOCKED;
1568 * Create, or update(jid):
1569 * name must not exist in an
1570 * active sibling jail.
1574 "jail \"%s\" already exists",
1579 cuflags != JAIL_CREATE) {
1584 /* If no active jail is found, use a dying one. */
1585 if (deadpr != NULL && pr == NULL) {
1586 if (flags & JAIL_DYING) {
1588 mtx_lock(&pr->pr_mtx);
1589 drflags |= PD_LOCKED;
1590 } else if (cuflags == JAIL_UPDATE) {
1593 "jail \"%s\" is dying", name);
1597 /* Update: name must exist if no jid. */
1598 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1600 vfs_opterror(opts, "jail \"%s\" not found",
1606 /* Update: must provide a jid or name. */
1607 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1609 vfs_opterror(opts, "update specified no jail");
1613 /* If there's no prison to update, create a new one and link it in. */
1614 created = pr == NULL;
1616 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1617 if (tpr->pr_childcount >= tpr->pr_childmax) {
1619 vfs_opterror(opts, "prison limit exceeded");
1622 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1624 vfs_opterror(opts, "no available jail IDs");
1628 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1629 pr->pr_state = PRISON_STATE_INVALID;
1630 refcount_init(&pr->pr_ref, 1);
1631 refcount_init(&pr->pr_uref, 0);
1632 drflags |= PD_DEREF;
1633 LIST_INIT(&pr->pr_children);
1634 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1635 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1639 TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1641 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1643 pr->pr_parent = ppr;
1645 prison_proc_hold(ppr);
1646 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1647 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1648 tpr->pr_childcount++;
1650 /* Set some default values, and inherit some from the parent. */
1655 root = mypr->pr_root;
1658 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1659 pr->pr_flags |= PR_HOST;
1660 #if defined(INET) || defined(INET6)
1662 if (!(pr_flags & PR_VNET))
1666 if (!(ch_flags & PR_IP4_USER))
1667 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1668 else if (!(pr_flags & PR_IP4_USER)) {
1669 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1670 prison_ip_dup(ppr, pr, PR_INET);
1674 if (!(ch_flags & PR_IP6_USER))
1675 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1676 else if (!(pr_flags & PR_IP6_USER)) {
1677 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1678 prison_ip_dup(ppr, pr, PR_INET6);
1683 /* Source address selection is always on by default. */
1684 pr->pr_flags |= _PR_IP_SADDRSEL;
1686 pr->pr_securelevel = ppr->pr_securelevel;
1687 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1688 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1689 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1691 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1692 if (osrelstr == NULL)
1693 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1694 sizeof(pr->pr_osrelease));
1696 strlcpy(pr->pr_osrelease, osrelstr,
1697 sizeof(pr->pr_osrelease));
1700 /* Allocate a new vnet if specified. */
1701 pr->pr_vnet = (pr_flags & PR_VNET)
1702 ? vnet_alloc() : ppr->pr_vnet;
1705 * Allocate a dedicated cpuset for each jail.
1706 * Unlike other initial settings, this may return an error.
1708 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1712 mtx_lock(&pr->pr_mtx);
1713 drflags |= PD_LOCKED;
1716 * Grab a reference for existing prisons, to ensure they
1717 * continue to exist for the duration of the call.
1720 drflags |= PD_DEREF;
1721 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1722 if ((pr->pr_flags & PR_VNET) &&
1723 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1726 "vnet jails cannot have IP address restrictions");
1731 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1734 "ip4 cannot be changed after creation");
1739 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1742 "ip6 cannot be changed after creation");
1748 /* Do final error checking before setting anything. */
1750 if (slevel < ppr->pr_securelevel) {
1756 if (childmax >= ppr->pr_childmax) {
1762 if (enforce < ppr->pr_enforce_statfs) {
1769 * devfs_rsnum is a uint16_t
1771 if (rsnum < 0 || rsnum > 65535) {
1776 * Nested jails always inherit parent's devfs ruleset
1778 if (jailed(td->td_ucred)) {
1779 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1783 rsnum = ppr->pr_devfs_rsnum;
1788 if ((ppr->pr_flags & PR_IP4) &&
1789 !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1794 if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1796 vfs_opterror(opts, "IPv4 addresses clash");
1803 if ((ppr->pr_flags & PR_IP6) &&
1804 !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1809 if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1811 vfs_opterror(opts, "IPv6 addresses clash");
1816 onamelen = namelen = 0;
1817 if (namelc != NULL) {
1818 /* Give a default name of the jid. Also allow the name to be
1819 * explicitly the jid - but not any other number, and only in
1820 * normal form (no leading zero/etc).
1822 if (namelc[0] == '\0')
1823 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1824 else if ((strtoul(namelc, &p, 10) != jid ||
1825 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1828 "name cannot be numeric (unless it is the jid)");
1832 * Make sure the name isn't too long for the prison or its
1835 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1836 onamelen = strlen(pr->pr_name + pnamelen);
1837 namelen = strlen(namelc);
1838 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1839 error = ENAMETOOLONG;
1842 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1843 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1844 sizeof(pr->pr_name)) {
1845 error = ENAMETOOLONG;
1850 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1851 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1857 * Let modules check their parameters. This requires unlocking and
1858 * then re-locking the prison, but this is still a valid state as long
1859 * as allprison_lock remains xlocked.
1861 mtx_unlock(&pr->pr_mtx);
1862 drflags &= ~PD_LOCKED;
1863 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1866 mtx_lock(&pr->pr_mtx);
1867 drflags |= PD_LOCKED;
1869 /* At this point, all valid parameters should have been noted. */
1870 TAILQ_FOREACH(opt, opts, link) {
1871 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1873 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1878 /* Set the parameters of the prison. */
1881 if (pr_flags & PR_IP4_USER) {
1882 pr->pr_flags |= PR_IP4;
1883 prison_ip_set(pr, PR_INET, ip4);
1885 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1887 if (tpr->pr_flags & PR_VNET) {
1892 if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
1901 if (pr_flags & PR_IP6_USER) {
1902 pr->pr_flags |= PR_IP6;
1903 prison_ip_set(pr, PR_INET6, ip6);
1905 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1907 if (tpr->pr_flags & PR_VNET) {
1912 if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
1920 pr->pr_securelevel = slevel;
1921 /* Set all child jails to be at least this level. */
1922 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1923 if (tpr->pr_securelevel < slevel)
1924 tpr->pr_securelevel = slevel;
1927 pr->pr_childmax = childmax;
1928 /* Set all child jails to under this limit. */
1929 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1930 if (tpr->pr_childmax > childmax - level)
1931 tpr->pr_childmax = childmax > level
1932 ? childmax - level : 0;
1935 pr->pr_enforce_statfs = enforce;
1936 /* Pass this restriction on to the children. */
1937 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1938 if (tpr->pr_enforce_statfs < enforce)
1939 tpr->pr_enforce_statfs = enforce;
1942 pr->pr_devfs_rsnum = rsnum;
1943 /* Pass this restriction on to the children. */
1944 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1945 tpr->pr_devfs_rsnum = rsnum;
1947 if (namelc != NULL) {
1948 if (ppr == &prison0)
1949 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1951 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1952 ppr->pr_name, namelc);
1953 /* Change this component of child names. */
1954 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1955 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1956 strlen(tpr->pr_name + onamelen) + 1);
1957 bcopy(pr->pr_name, tpr->pr_name, namelen);
1961 /* Try to keep a real-rooted full pathname. */
1962 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1966 if (PR_HOST & ch_flags & ~pr_flags) {
1967 if (pr->pr_flags & PR_HOST) {
1969 * Copy the parent's host info. As with pr_ip4 above,
1970 * the lack of a lock on the parent is not a problem;
1971 * it is always set with allprison_lock at least
1972 * shared, and is held exclusively here.
1974 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1975 sizeof(pr->pr_hostname));
1976 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1977 sizeof(pr->pr_domainname));
1978 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1979 sizeof(pr->pr_hostuuid));
1980 pr->pr_hostid = pr->pr_parent->pr_hostid;
1982 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1983 /* Set this prison, and any descendants without PR_HOST. */
1985 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1987 strlcpy(pr->pr_domainname, domain,
1988 sizeof(pr->pr_domainname));
1990 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1992 pr->pr_hostid = hid;
1993 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1994 if (tpr->pr_flags & PR_HOST)
1998 strlcpy(tpr->pr_hostname,
2000 sizeof(tpr->pr_hostname));
2002 strlcpy(tpr->pr_domainname,
2004 sizeof(tpr->pr_domainname));
2006 strlcpy(tpr->pr_hostuuid,
2008 sizeof(tpr->pr_hostuuid));
2010 tpr->pr_hostid = hid;
2014 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2015 if ((tallow = ch_allow & ~pr_allow))
2016 prison_set_allow_locked(pr, tallow, 0);
2018 * Persistent prisons get an extra reference, and prisons losing their
2019 * persist flag lose that reference.
2021 born = !prison_isalive(pr);
2022 if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2023 if (pr_flags & PR_PERSIST) {
2026 * This may make a dead prison alive again, but wait
2027 * to label it as such until after OSD calls have had
2028 * a chance to run (and perhaps to fail).
2030 refcount_acquire(&pr->pr_uref);
2032 drflags |= PD_DEUREF;
2033 prison_free_not_last(pr);
2036 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2037 mtx_unlock(&pr->pr_mtx);
2038 drflags &= ~PD_LOCKED;
2040 * Any errors past this point will need to de-persist newly created
2041 * prisons, as well as call remove methods.
2047 if (racct_enable && created)
2048 prison_racct_attach(pr);
2051 /* Locks may have prevented a complete restriction of child IP
2052 * addresses. If so, allocate some more memory and try again.
2056 ip4s = pr->pr_addrs[PR_INET]->ips;
2058 ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2059 mtx_lock(&pr->pr_mtx);
2061 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2063 if (tpr->pr_flags & PR_VNET) {
2068 if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2071 mtx_unlock(&pr->pr_mtx);
2076 ip6s = pr->pr_addrs[PR_INET6]->ips;
2078 ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2079 mtx_lock(&pr->pr_mtx);
2081 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2083 if (tpr->pr_flags & PR_VNET) {
2088 if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2091 mtx_unlock(&pr->pr_mtx);
2095 /* Let the modules do their work. */
2097 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2101 error = osd_jail_call(pr, PR_METHOD_SET, opts);
2106 * A new prison is now ready to be seen; either it has gained a user
2107 * reference via persistence, or is about to gain one via attachment.
2110 drflags = prison_lock_xlock(pr, drflags);
2111 pr->pr_state = PRISON_STATE_ALIVE;
2114 /* Attach this process to the prison if requested. */
2115 if (flags & JAIL_ATTACH) {
2116 error = do_jail_attach(td, pr,
2117 prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2118 drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2120 vfs_opterror(opts, "attach failed");
2126 if (racct_enable && !created) {
2127 if (drflags & PD_LOCKED) {
2128 mtx_unlock(&pr->pr_mtx);
2129 drflags &= ~PD_LOCKED;
2131 if (drflags & PD_LIST_XLOCKED) {
2132 sx_xunlock(&allprison_lock);
2133 drflags &= ~PD_LIST_XLOCKED;
2135 prison_racct_modify(pr);
2139 if (born && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2140 (pr->pr_root->v_vflag & VV_ROOT) == 0)
2141 printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2142 " file system\n", pr->pr_id);
2144 drflags &= ~PD_KILL;
2145 td->td_retval[0] = pr->pr_id;
2148 /* Release any temporary prison holds and/or locks. */
2150 prison_deref(pr, drflags);
2151 else if (drflags & PD_LIST_SLOCKED)
2152 sx_sunlock(&allprison_lock);
2153 else if (drflags & PD_LIST_XLOCKED)
2154 sx_xunlock(&allprison_lock);
2159 /* Write the error message back to userspace. */
2160 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2161 &errmsg_len) == 0 && errmsg_len > 0) {
2162 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2163 if (optuio->uio_segflg == UIO_SYSSPACE)
2165 optuio->uio_iov[errmsg_pos].iov_base,
2169 optuio->uio_iov[errmsg_pos].iov_base,
2175 prison_ip_free(ip4);
2178 prison_ip_free(ip6);
2181 free(g_path, M_TEMP);
2187 * Find the next available prison ID. Return the ID on success, or zero
2188 * on failure. Also set a pointer to the allprison list entry the prison
2189 * should be inserted before.
2192 get_next_prid(struct prison **insprp)
2194 struct prison *inspr;
2197 jid = lastprid % JAIL_MAX + 1;
2198 if (TAILQ_EMPTY(&allprison) ||
2199 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2201 * A common case is for all jails to be implicitly numbered,
2202 * which means they'll go on the end of the list, at least
2203 * for the first JAIL_MAX times.
2208 * Take two passes through the allprison list: first starting
2209 * with the proposed jid, then ending with it.
2211 for (maxid = JAIL_MAX; maxid != 0; ) {
2212 TAILQ_FOREACH(inspr, &allprison, pr_list) {
2213 if (inspr->pr_id < jid)
2215 if (inspr->pr_id > jid) {
2216 /* Found an opening. */
2220 if (++jid > maxid) {
2221 if (lastprid == maxid || lastprid == 0)
2224 * The entire legal range
2225 * has been traversed
2229 /* Try again from the start. */
2235 if (inspr == NULL) {
2236 /* Found room at the end of the list. */
2247 * struct jail_get_args {
2248 * struct iovec *iovp;
2249 * unsigned int iovcnt;
2254 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2259 /* Check that we have an even number of iovecs. */
2260 if (uap->iovcnt & 1)
2263 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2266 error = kern_jail_get(td, auio, uap->flags);
2268 error = copyout(auio->uio_iov, uap->iovp,
2269 uap->iovcnt * sizeof (struct iovec));
2275 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2277 struct bool_flags *bf;
2278 struct jailsys_flags *jsf;
2279 struct prison *pr, *mypr;
2281 struct vfsoptlist *opts;
2282 char *errmsg, *name;
2283 int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2286 if (flags & ~JAIL_GET_MASK)
2289 /* Get the parameter list. */
2290 error = vfs_buildopts(optuio, &opts);
2293 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2294 mypr = td->td_ucred->cr_prison;
2298 * Find the prison specified by one of: lastjid, jid, name.
2300 sx_slock(&allprison_lock);
2301 drflags = PD_LIST_SLOCKED;
2302 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2304 TAILQ_FOREACH(pr, &allprison, pr_list) {
2305 if (pr->pr_id > jid &&
2306 ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2307 prison_ischild(mypr, pr)) {
2308 mtx_lock(&pr->pr_mtx);
2309 drflags |= PD_LOCKED;
2314 vfs_opterror(opts, "no jail after %d", jid);
2316 } else if (error != ENOENT)
2319 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2322 pr = prison_find_child(mypr, jid);
2324 drflags |= PD_LOCKED;
2325 if (!(prison_isalive(pr) ||
2326 (flags & JAIL_DYING))) {
2328 vfs_opterror(opts, "jail %d is dying",
2335 vfs_opterror(opts, "jail %d not found", jid);
2338 } else if (error != ENOENT)
2341 error = vfs_getopt(opts, "name", (void **)&name, &len);
2343 if (len == 0 || name[len - 1] != '\0') {
2347 pr = prison_find_name(mypr, name);
2349 drflags |= PD_LOCKED;
2350 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2352 vfs_opterror(opts, "jail \"%s\" is dying",
2359 vfs_opterror(opts, "jail \"%s\" not found", name);
2361 } else if (error != ENOENT)
2364 vfs_opterror(opts, "no jail specified");
2369 /* Get the parameters of the prison. */
2371 drflags |= PD_DEREF;
2372 td->td_retval[0] = pr->pr_id;
2373 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2374 if (error != 0 && error != ENOENT)
2376 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2377 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2378 if (error != 0 && error != ENOENT)
2380 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2381 if (error != 0 && error != ENOENT)
2383 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2384 sizeof(pr->pr_cpuset->cs_id));
2385 if (error != 0 && error != ENOENT)
2387 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2388 if (error != 0 && error != ENOENT)
2391 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2392 pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2393 pr_families[PR_INET].size : 0 );
2394 if (error != 0 && error != ENOENT)
2398 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2399 pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2400 pr_families[PR_INET6].size : 0 );
2401 if (error != 0 && error != ENOENT)
2404 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2405 sizeof(pr->pr_securelevel));
2406 if (error != 0 && error != ENOENT)
2408 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2409 sizeof(pr->pr_childcount));
2410 if (error != 0 && error != ENOENT)
2412 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2413 sizeof(pr->pr_childmax));
2414 if (error != 0 && error != ENOENT)
2416 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2417 if (error != 0 && error != ENOENT)
2419 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2420 if (error != 0 && error != ENOENT)
2422 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2423 if (error != 0 && error != ENOENT)
2425 #ifdef COMPAT_FREEBSD32
2426 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2427 uint32_t hid32 = pr->pr_hostid;
2429 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2432 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2433 sizeof(pr->pr_hostid));
2434 if (error != 0 && error != ENOENT)
2436 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2437 sizeof(pr->pr_enforce_statfs));
2438 if (error != 0 && error != ENOENT)
2440 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2441 sizeof(pr->pr_devfs_rsnum));
2442 if (error != 0 && error != ENOENT)
2444 for (bf = pr_flag_bool;
2445 bf < pr_flag_bool + nitems(pr_flag_bool);
2447 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2448 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2449 if (error != 0 && error != ENOENT)
2452 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2453 if (error != 0 && error != ENOENT)
2456 for (jsf = pr_flag_jailsys;
2457 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2459 f = pr->pr_flags & (jsf->disable | jsf->new);
2460 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2461 : (f == jsf->new) ? JAIL_SYS_NEW
2463 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2464 if (error != 0 && error != ENOENT)
2467 for (bf = pr_flag_allow;
2468 bf < pr_flag_allow + nitems(pr_flag_allow) &&
2469 atomic_load_int(&bf->flag) != 0;
2471 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2472 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2473 if (error != 0 && error != ENOENT)
2476 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2477 if (error != 0 && error != ENOENT)
2480 i = !prison_isalive(pr);
2481 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2482 if (error != 0 && error != ENOENT)
2485 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2486 if (error != 0 && error != ENOENT)
2488 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2489 sizeof(pr->pr_osreldate));
2490 if (error != 0 && error != ENOENT)
2492 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2493 if (error != 0 && error != ENOENT)
2496 /* Get the module parameters. */
2497 mtx_unlock(&pr->pr_mtx);
2498 drflags &= ~PD_LOCKED;
2499 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2502 prison_deref(pr, drflags);
2506 /* By now, all parameters should have been noted. */
2507 TAILQ_FOREACH(opt, opts, link) {
2508 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2510 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2515 /* Write the fetched parameters back to userspace. */
2517 TAILQ_FOREACH(opt, opts, link) {
2518 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2519 pos = 2 * opt->pos + 1;
2520 optuio->uio_iov[pos].iov_len = opt->len;
2521 if (opt->value != NULL) {
2522 if (optuio->uio_segflg == UIO_SYSSPACE) {
2524 optuio->uio_iov[pos].iov_base,
2527 error = copyout(opt->value,
2528 optuio->uio_iov[pos].iov_base,
2538 /* Release any temporary prison holds and/or locks. */
2540 prison_deref(pr, drflags);
2541 else if (drflags & PD_LIST_SLOCKED)
2542 sx_sunlock(&allprison_lock);
2543 if (error && errmsg_pos >= 0) {
2544 /* Write the error message back to userspace. */
2545 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2546 errmsg_pos = 2 * errmsg_pos + 1;
2547 if (errmsg_len > 0) {
2548 if (optuio->uio_segflg == UIO_SYSSPACE)
2550 optuio->uio_iov[errmsg_pos].iov_base,
2554 optuio->uio_iov[errmsg_pos].iov_base,
2563 * struct jail_remove_args {
2568 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2573 error = priv_check(td, PRIV_JAIL_REMOVE);
2577 sx_xlock(&allprison_lock);
2578 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2580 sx_xunlock(&allprison_lock);
2583 if (!prison_isalive(pr)) {
2584 /* Silently ignore already-dying prisons. */
2585 mtx_unlock(&pr->pr_mtx);
2586 sx_xunlock(&allprison_lock);
2589 prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2594 * struct jail_attach_args {
2599 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2604 error = priv_check(td, PRIV_JAIL_ATTACH);
2608 sx_slock(&allprison_lock);
2609 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2611 sx_sunlock(&allprison_lock);
2615 /* Do not allow a process to attach to a prison that is not alive. */
2616 if (!prison_isalive(pr)) {
2617 mtx_unlock(&pr->pr_mtx);
2618 sx_sunlock(&allprison_lock);
2622 return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2626 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2629 struct ucred *newcred, *oldcred;
2632 mtx_assert(&pr->pr_mtx, MA_OWNED);
2633 sx_assert(&allprison_lock, SX_LOCKED);
2634 drflags &= PD_LOCK_FLAGS;
2636 * XXX: Note that there is a slight race here if two threads
2637 * in the same privileged process attempt to attach to two
2638 * different jails at the same time. It is important for
2639 * user processes not to do this, or they might end up with
2640 * a process root from one prison, but attached to the jail
2644 refcount_acquire(&pr->pr_uref);
2645 drflags |= PD_DEREF | PD_DEUREF;
2646 mtx_unlock(&pr->pr_mtx);
2647 drflags &= ~PD_LOCKED;
2649 /* Let modules do whatever they need to prepare for attaching. */
2650 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2652 prison_deref(pr, drflags);
2655 sx_unlock(&allprison_lock);
2656 drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2659 * Reparent the newly attached process to this jail.
2662 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2666 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2667 if ((error = change_dir(pr->pr_root, td)) != 0)
2670 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2673 VOP_UNLOCK(pr->pr_root);
2674 if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2679 oldcred = crcopysafe(p, newcred);
2680 newcred->cr_prison = pr;
2681 proc_set_cred(p, newcred);
2684 racct_proc_ucred_changed(p, oldcred, newcred);
2689 rctl_proc_ucred_changed(p, newcred);
2692 prison_proc_relink(oldcred->cr_prison, pr, p);
2693 prison_deref(oldcred->cr_prison, drflags);
2697 * If the prison was killed while changing credentials, die along
2700 if (!prison_isalive(pr)) {
2702 kern_psignal(p, SIGKILL);
2709 VOP_UNLOCK(pr->pr_root);
2711 /* Tell modules this thread is still in its old jail after all. */
2712 sx_slock(&allprison_lock);
2713 drflags |= PD_LIST_SLOCKED;
2714 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2715 prison_deref(pr, drflags);
2720 * Returns a locked prison instance, or NULL on failure.
2723 prison_find(int prid)
2727 sx_assert(&allprison_lock, SX_LOCKED);
2728 TAILQ_FOREACH(pr, &allprison, pr_list) {
2729 if (pr->pr_id < prid)
2731 if (pr->pr_id > prid)
2733 KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2734 mtx_lock(&pr->pr_mtx);
2741 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2744 prison_find_child(struct prison *mypr, int prid)
2749 sx_assert(&allprison_lock, SX_LOCKED);
2750 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2751 if (pr->pr_id == prid) {
2752 KASSERT(prison_isvalid(pr),
2753 ("Found invalid prison %p", pr));
2754 mtx_lock(&pr->pr_mtx);
2762 * Look for the name relative to mypr. Returns a locked prison or NULL.
2765 prison_find_name(struct prison *mypr, const char *name)
2767 struct prison *pr, *deadpr;
2771 sx_assert(&allprison_lock, SX_LOCKED);
2772 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2774 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2775 if (!strcmp(pr->pr_name + mylen, name)) {
2776 KASSERT(prison_isvalid(pr),
2777 ("Found invalid prison %p", pr));
2778 if (prison_isalive(pr)) {
2779 mtx_lock(&pr->pr_mtx);
2785 /* There was no valid prison - perhaps there was a dying one. */
2787 mtx_lock(&deadpr->pr_mtx);
2792 * See if a prison has the specific flag set. The prison should be locked,
2793 * unless checking for flags that are only set at jail creation (such as
2794 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2795 * to any other prison data.
2798 prison_flag(struct ucred *cred, unsigned flag)
2801 return ((cred->cr_prison->pr_flags & flag) != 0);
2805 * See if a prison has the specific allow flag set.
2806 * The prison *should* be locked, or only a single bit is examined, without
2807 * regard to any other prison data.
2810 prison_allow(struct ucred *cred, unsigned flag)
2813 return ((cred->cr_prison->pr_allow & flag) != 0);
2817 * Hold a prison reference, by incrementing pr_ref. It is generally
2818 * an error to hold a prison that does not already have a reference.
2819 * A prison record will remain valid as long as it has at least one
2820 * reference, and will not be removed as long as either the prison
2821 * mutex or the allprison lock is held (allprison_lock may be shared).
2824 prison_hold_locked(struct prison *pr)
2827 /* Locking is no longer required. */
2832 prison_hold(struct prison *pr)
2835 int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2838 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2840 refcount_acquire(&pr->pr_ref);
2845 * Remove a prison reference. If that was the last reference, the
2846 * prison will be removed (at a later time).
2849 prison_free_locked(struct prison *pr)
2852 mtx_assert(&pr->pr_mtx, MA_OWNED);
2854 * Locking is no longer required, but unlock because the caller
2857 mtx_unlock(&pr->pr_mtx);
2862 prison_free(struct prison *pr)
2865 KASSERT(refcount_load(&pr->pr_ref) > 0,
2866 ("Trying to free dead prison %p (jid=%d).",
2868 if (!refcount_release_if_not_last(&pr->pr_ref)) {
2870 * Don't remove the last reference in this context,
2871 * in case there are locks held.
2873 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2878 prison_free_not_last(struct prison *pr)
2883 KASSERT(refcount_load(&pr->pr_ref) > 0,
2884 ("Trying to free dead prison %p (jid=%d).",
2886 lastref = refcount_release(&pr->pr_ref);
2888 ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2891 refcount_release(&pr->pr_ref);
2896 * Hold a prison for user visibility, by incrementing pr_uref.
2897 * It is generally an error to hold a prison that isn't already
2898 * user-visible, except through the jail system calls. It is also
2899 * an error to hold an invalid prison. A prison record will remain
2900 * alive as long as it has at least one user reference, and will not
2901 * be set to the dying state until the prison mutex and allprison_lock
2905 prison_proc_hold(struct prison *pr)
2908 int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2911 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2913 refcount_acquire(&pr->pr_uref);
2918 * Remove a prison user reference. If it was the last reference, the
2919 * prison will be considered "dying", and may be removed once all of
2920 * its references are dropped.
2923 prison_proc_free(struct prison *pr)
2927 * Locking is only required when releasing the last reference.
2928 * This allows assurance that a locked prison will remain alive
2929 * until it is unlocked.
2931 KASSERT(refcount_load(&pr->pr_uref) > 0,
2932 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2933 if (!refcount_release_if_not_last(&pr->pr_uref)) {
2935 * Don't remove the last user reference in this context,
2936 * which is expected to be a process that is not only locked,
2937 * but also half dead. Add a reference so any calls to
2938 * prison_free() won't re-submit the task.
2941 mtx_lock(&pr->pr_mtx);
2942 KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2943 ("Redundant last reference in prison_proc_free (jid=%d)",
2945 pr->pr_flags |= PR_COMPLETE_PROC;
2946 mtx_unlock(&pr->pr_mtx);
2947 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2952 prison_proc_free_not_last(struct prison *pr)
2957 KASSERT(refcount_load(&pr->pr_uref) > 0,
2958 ("Trying to free dead prison %p (jid=%d).",
2960 lastref = refcount_release(&pr->pr_uref);
2962 ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2965 refcount_release(&pr->pr_uref);
2970 prison_proc_link(struct prison *pr, struct proc *p)
2973 sx_assert(&allproc_lock, SA_XLOCKED);
2974 LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
2978 prison_proc_unlink(struct prison *pr, struct proc *p)
2981 sx_assert(&allproc_lock, SA_XLOCKED);
2982 LIST_REMOVE(p, p_jaillist);
2986 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
2989 sx_xlock(&allproc_lock);
2990 prison_proc_unlink(opr, p);
2991 prison_proc_link(npr, p);
2992 sx_xunlock(&allproc_lock);
2996 * Complete a call to either prison_free or prison_proc_free.
2999 prison_complete(void *context, int pending)
3001 struct prison *pr = context;
3005 * This could be called to release the last reference, or the last
3006 * user reference (plus the reference held in prison_proc_free).
3008 drflags = prison_lock_xlock(pr, PD_DEREF);
3009 if (pr->pr_flags & PR_COMPLETE_PROC) {
3010 pr->pr_flags &= ~PR_COMPLETE_PROC;
3011 drflags |= PD_DEUREF;
3013 prison_deref(pr, drflags);
3017 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3020 kern_psignal(p, SIGKILL);
3024 * Note the iteration does not guarantee acting on all processes.
3025 * Most notably there may be fork or jail_attach in progress.
3028 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3034 if (atomic_load_int(&pr->pr_childcount) == 0) {
3035 sx_slock(&allproc_lock);
3036 LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3037 if (p->p_state == PRS_NEW)
3043 sx_sunlock(&allproc_lock);
3044 if (atomic_load_int(&pr->pr_childcount) == 0)
3047 * Some jails popped up during the iteration, fall through to a
3048 * system-wide search.
3052 sx_slock(&allproc_lock);
3053 FOREACH_PROC_IN_SYSTEM(p) {
3055 if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3056 for (ppr = p->p_ucred->cr_prison;
3058 ppr = ppr->pr_parent) {
3067 sx_sunlock(&allproc_lock);
3071 * Remove a prison reference and/or user reference (usually).
3072 * This assumes context that allows sleeping (for allprison_lock),
3073 * with no non-sleeping locks held, except perhaps the prison itself.
3074 * If there are no more references, release and delist the prison.
3075 * On completion, the prison lock and the allprison lock are both
3079 prison_deref(struct prison *pr, int flags)
3081 struct prisonlist freeprison;
3082 struct prison *killpr, *rpr, *ppr, *tpr;
3085 TAILQ_INIT(&freeprison);
3087 * Release this prison as requested, which may cause its parent
3088 * to be released, and then maybe its grandparent, etc.
3091 if (flags & PD_KILL) {
3092 /* Kill the prison and its descendents. */
3093 KASSERT(pr != &prison0,
3094 ("prison_deref trying to kill prison0"));
3095 if (!(flags & PD_DEREF)) {
3099 flags = prison_lock_xlock(pr, flags);
3100 prison_deref_kill(pr, &freeprison);
3102 if (flags & PD_DEUREF) {
3103 /* Drop a user reference. */
3104 KASSERT(refcount_load(&pr->pr_uref) > 0,
3105 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3107 if (!refcount_release_if_not_last(&pr->pr_uref)) {
3108 if (!(flags & PD_DEREF)) {
3112 flags = prison_lock_xlock(pr, flags);
3113 if (refcount_release(&pr->pr_uref) &&
3114 pr->pr_state == PRISON_STATE_ALIVE) {
3116 * When the last user references goes,
3117 * this becomes a dying prison.
3120 refcount_load(&prison0.pr_uref) > 0,
3121 ("prison0 pr_uref=0"));
3122 pr->pr_state = PRISON_STATE_DYING;
3123 mtx_unlock(&pr->pr_mtx);
3124 flags &= ~PD_LOCKED;
3129 if (flags & PD_KILL) {
3131 * Any remaining user references are probably processes
3132 * that need to be killed, either in this prison or its
3135 if (refcount_load(&pr->pr_uref) > 0)
3137 /* Make sure the parent prison doesn't get killed. */
3140 if (flags & PD_DEREF) {
3141 /* Drop a reference. */
3142 KASSERT(refcount_load(&pr->pr_ref) > 0,
3143 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3145 if (!refcount_release_if_not_last(&pr->pr_ref)) {
3146 flags = prison_lock_xlock(pr, flags);
3147 if (refcount_release(&pr->pr_ref)) {
3149 * When the last reference goes,
3150 * unlink the prison and set it aside.
3153 refcount_load(&pr->pr_uref) == 0,
3154 ("prison_deref: last ref, "
3155 "but still has %d urefs (jid=%d)",
3156 pr->pr_uref, pr->pr_id));
3158 refcount_load(&prison0.pr_ref) != 0,
3159 ("prison0 pr_ref=0"));
3160 pr->pr_state = PRISON_STATE_INVALID;
3161 TAILQ_REMOVE(&allprison, pr, pr_list);
3162 LIST_REMOVE(pr, pr_sibling);
3163 TAILQ_INSERT_TAIL(&freeprison, pr,
3165 for (ppr = pr->pr_parent;
3167 ppr = ppr->pr_parent)
3168 ppr->pr_childcount--;
3170 * Removing a prison frees references
3173 mtx_unlock(&pr->pr_mtx);
3174 flags &= ~PD_LOCKED;
3176 flags |= PD_DEREF | PD_DEUREF;
3184 /* Release all the prison locks. */
3185 if (flags & PD_LOCKED)
3186 mtx_unlock(&pr->pr_mtx);
3187 if (flags & PD_LIST_SLOCKED)
3188 sx_sunlock(&allprison_lock);
3189 else if (flags & PD_LIST_XLOCKED)
3190 sx_xunlock(&allprison_lock);
3192 /* Kill any processes attached to a killed prison. */
3194 prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3197 * Finish removing any unreferenced prisons, which couldn't happen
3198 * while allprison_lock was held (to avoid a LOR on vrele).
3200 TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3202 if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3203 vnet_destroy(rpr->pr_vnet);
3205 if (rpr->pr_root != NULL)
3206 vrele(rpr->pr_root);
3207 mtx_destroy(&rpr->pr_mtx);
3209 prison_ip_free(rpr->pr_addrs[PR_INET]);
3212 prison_ip_free(rpr->pr_addrs[PR_INET6]);
3214 if (rpr->pr_cpuset != NULL)
3215 cpuset_rel(rpr->pr_cpuset);
3219 prison_racct_detach(rpr);
3221 TAILQ_REMOVE(&freeprison, rpr, pr_list);
3222 free(rpr, M_PRISON);
3227 * Kill the prison and its descendants. Mark them as dying, clear the
3228 * persist flag, and call module remove methods.
3231 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3233 struct prison *cpr, *ppr, *rpr;
3237 * Unlike the descendants, the target prison can be killed
3238 * even if it is currently dying. This is useful for failed
3239 * creation in jail_set(2).
3241 KASSERT(refcount_load(&pr->pr_ref) > 0,
3242 ("Trying to kill dead prison %p (jid=%d).",
3244 refcount_acquire(&pr->pr_uref);
3245 pr->pr_state = PRISON_STATE_DYING;
3246 mtx_unlock(&pr->pr_mtx);
3249 FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3251 if (!prison_isalive(cpr)) {
3256 prison_proc_hold(cpr);
3257 mtx_lock(&cpr->pr_mtx);
3258 cpr->pr_state = PRISON_STATE_DYING;
3259 cpr->pr_flags |= PR_REMOVE;
3260 mtx_unlock(&cpr->pr_mtx);
3263 if (!(cpr->pr_flags & PR_REMOVE))
3265 prison_cleanup(cpr);
3266 mtx_lock(&cpr->pr_mtx);
3267 cpr->pr_flags &= ~PR_REMOVE;
3268 if (cpr->pr_flags & PR_PERSIST) {
3269 cpr->pr_flags &= ~PR_PERSIST;
3270 prison_proc_free_not_last(cpr);
3271 prison_free_not_last(cpr);
3273 (void)refcount_release(&cpr->pr_uref);
3274 if (refcount_release(&cpr->pr_ref)) {
3276 * When the last reference goes, unlink the prison
3277 * and set it aside for prison_deref() to handle.
3278 * Delay unlinking the sibling list to keep the loop
3282 LIST_REMOVE(rpr, pr_sibling);
3284 rpr->pr_state = PRISON_STATE_INVALID;
3285 TAILQ_REMOVE(&allprison, rpr, pr_list);
3286 TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3288 * Removing a prison frees references from its parent.
3290 ppr = rpr->pr_parent;
3291 prison_proc_free_not_last(ppr);
3292 prison_free_not_last(ppr);
3293 for (; ppr != NULL; ppr = ppr->pr_parent)
3294 ppr->pr_childcount--;
3296 mtx_unlock(&cpr->pr_mtx);
3299 LIST_REMOVE(rpr, pr_sibling);
3302 mtx_lock(&pr->pr_mtx);
3303 if (pr->pr_flags & PR_PERSIST) {
3304 pr->pr_flags &= ~PR_PERSIST;
3305 prison_proc_free_not_last(pr);
3306 prison_free_not_last(pr);
3308 (void)refcount_release(&pr->pr_uref);
3312 * Given the current locking state in the flags, make sure allprison_lock
3313 * is held exclusive, and the prison is locked. Return flags indicating
3317 prison_lock_xlock(struct prison *pr, int flags)
3320 if (!(flags & PD_LIST_XLOCKED)) {
3322 * Get allprison_lock, which may be an upgrade,
3323 * and may require unlocking the prison.
3325 if (flags & PD_LOCKED) {
3326 mtx_unlock(&pr->pr_mtx);
3327 flags &= ~PD_LOCKED;
3329 if (flags & PD_LIST_SLOCKED) {
3330 if (!sx_try_upgrade(&allprison_lock)) {
3331 sx_sunlock(&allprison_lock);
3332 sx_xlock(&allprison_lock);
3334 flags &= ~PD_LIST_SLOCKED;
3336 sx_xlock(&allprison_lock);
3337 flags |= PD_LIST_XLOCKED;
3339 if (!(flags & PD_LOCKED)) {
3340 /* Lock the prison mutex. */
3341 mtx_lock(&pr->pr_mtx);
3348 * Release a prison's resources when it starts dying (when the last user
3349 * reference is dropped, or when it is killed).
3352 prison_cleanup(struct prison *pr)
3354 sx_assert(&allprison_lock, SA_XLOCKED);
3355 mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3356 vfs_exjail_delete(pr);
3357 shm_remove_prison(pr);
3358 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3362 * Set or clear a permission bit in the pr_allow field, passing restrictions
3363 * (cleared permission) down to child jails.
3366 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3370 pr = cred->cr_prison;
3371 sx_slock(&allprison_lock);
3372 mtx_lock(&pr->pr_mtx);
3373 prison_set_allow_locked(pr, flag, enable);
3374 mtx_unlock(&pr->pr_mtx);
3375 sx_sunlock(&allprison_lock);
3379 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3385 pr->pr_allow |= flag;
3387 pr->pr_allow &= ~flag;
3388 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3389 cpr->pr_allow &= ~flag;
3394 * Check if a jail supports the given address family.
3396 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3400 prison_check_af(struct ucred *cred, int af)
3405 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3407 pr = cred->cr_prison;
3409 /* Prisons with their own network stack are not limited. */
3410 if (prison_owns_vnet(cred))
3419 if (pr->pr_flags & PR_IP4)
3421 mtx_lock(&pr->pr_mtx);
3422 if ((pr->pr_flags & PR_IP4) &&
3423 pr->pr_addrs[PR_INET] == NULL)
3424 error = EAFNOSUPPORT;
3425 mtx_unlock(&pr->pr_mtx);
3431 if (pr->pr_flags & PR_IP6)
3433 mtx_lock(&pr->pr_mtx);
3434 if ((pr->pr_flags & PR_IP6) &&
3435 pr->pr_addrs[PR_INET6] == NULL)
3436 error = EAFNOSUPPORT;
3437 mtx_unlock(&pr->pr_mtx);
3446 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3447 error = EAFNOSUPPORT;
3453 * Check if given address belongs to the jail referenced by cred (wrapper to
3454 * prison_check_ip[46]).
3456 * Returns 0 if jail doesn't restrict the address family or if address belongs
3457 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3458 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3461 prison_if(struct ucred *cred, const struct sockaddr *sa)
3464 const struct sockaddr_in *sai;
3467 const struct sockaddr_in6 *sai6;
3471 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3472 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3475 if (prison_owns_vnet(cred))
3480 switch (sa->sa_family)
3484 sai = (const struct sockaddr_in *)sa;
3485 error = prison_check_ip4(cred, &sai->sin_addr);
3490 sai6 = (const struct sockaddr_in6 *)sa;
3491 error = prison_check_ip6(cred, &sai6->sin6_addr);
3495 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3496 error = EAFNOSUPPORT;
3502 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3505 prison_check(struct ucred *cred1, struct ucred *cred2)
3508 return ((cred1->cr_prison == cred2->cr_prison ||
3509 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3513 * For mountd/nfsd to run within a prison, it must be:
3515 * - PR_ALLOW_NFSD must be set on it.
3516 * - The root directory (pr_root) of the prison must be
3517 * a file system mount point, so the mountd can hang
3518 * export information on it.
3519 * - The prison's enforce_statfs cannot be 0, so that
3520 * mountd(8) can do exports.
3523 prison_check_nfsd(struct ucred *cred)
3526 if (jailed_without_vnet(cred))
3528 if (!prison_allow(cred, PR_ALLOW_NFSD))
3530 if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3532 if (cred->cr_prison->pr_enforce_statfs == 0)
3538 * Return true if p2 is a child of p1, otherwise false.
3541 prison_ischild(struct prison *pr1, struct prison *pr2)
3544 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3551 * Return true if the prison is currently alive. A prison is alive if it
3552 * holds user references and it isn't being removed.
3555 prison_isalive(const struct prison *pr)
3558 if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3564 * Return true if the prison is currently valid. A prison is valid if it has
3565 * been fully created, and is not being destroyed. Note that dying prisons
3566 * are still considered valid. Invalid prisons won't be found under normal
3567 * circumstances, as they're only put in that state by functions that have
3568 * an exclusive hold on allprison_lock.
3571 prison_isvalid(struct prison *pr)
3574 if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3576 if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3582 * Return true if the passed credential is in a jail and that jail does not
3583 * have its own virtual network stack, otherwise false.
3586 jailed_without_vnet(struct ucred *cred)
3592 if (prison_owns_vnet(cred))
3600 * Return the correct hostname (domainname, et al) for the passed credential.
3603 getcredhostname(struct ucred *cred, char *buf, size_t size)
3608 * A NULL credential can be used to shortcut to the physical
3609 * system's hostname.
3611 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3612 mtx_lock(&pr->pr_mtx);
3613 strlcpy(buf, pr->pr_hostname, size);
3614 mtx_unlock(&pr->pr_mtx);
3618 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3621 mtx_lock(&cred->cr_prison->pr_mtx);
3622 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3623 mtx_unlock(&cred->cr_prison->pr_mtx);
3627 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3630 mtx_lock(&cred->cr_prison->pr_mtx);
3631 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3632 mtx_unlock(&cred->cr_prison->pr_mtx);
3636 getcredhostid(struct ucred *cred, unsigned long *hostid)
3639 mtx_lock(&cred->cr_prison->pr_mtx);
3640 *hostid = cred->cr_prison->pr_hostid;
3641 mtx_unlock(&cred->cr_prison->pr_mtx);
3645 getjailname(struct ucred *cred, char *name, size_t len)
3648 mtx_lock(&cred->cr_prison->pr_mtx);
3649 strlcpy(name, cred->cr_prison->pr_name, len);
3650 mtx_unlock(&cred->cr_prison->pr_mtx);
3655 * Determine whether the prison represented by cred owns
3656 * its vnet rather than having it inherited.
3658 * Returns true in case the prison owns the vnet, false otherwise.
3661 prison_owns_vnet(struct ucred *cred)
3665 * vnets cannot be added/removed after jail creation,
3666 * so no need to lock here.
3668 return ((cred->cr_prison->pr_flags & PR_VNET) != 0);
3673 * Determine whether the subject represented by cred can "see"
3674 * status of a mount point.
3675 * Returns: 0 for permitted, ENOENT otherwise.
3676 * XXX: This function should be called cr_canseemount() and should be
3677 * placed in kern_prot.c.
3680 prison_canseemount(struct ucred *cred, struct mount *mp)
3686 pr = cred->cr_prison;
3687 if (pr->pr_enforce_statfs == 0)
3689 if (pr->pr_root->v_mount == mp)
3691 if (pr->pr_enforce_statfs == 2)
3694 * If jail's chroot directory is set to "/" we should be able to see
3695 * all mount-points from inside a jail.
3696 * This is ugly check, but this is the only situation when jail's
3697 * directory ends with '/'.
3699 if (strcmp(pr->pr_path, "/") == 0)
3701 len = strlen(pr->pr_path);
3703 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3706 * Be sure that we don't have situation where jail's root directory
3707 * is "/some/path" and mount point is "/some/pathpath".
3709 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3715 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3717 char jpath[MAXPATHLEN];
3721 pr = cred->cr_prison;
3722 if (pr->pr_enforce_statfs == 0)
3724 if (prison_canseemount(cred, mp) != 0) {
3725 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3726 strlcpy(sp->f_mntonname, "[restricted]",
3727 sizeof(sp->f_mntonname));
3730 if (pr->pr_root->v_mount == mp) {
3732 * Clear current buffer data, so we are sure nothing from
3733 * the valid path left there.
3735 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3736 *sp->f_mntonname = '/';
3740 * If jail's chroot directory is set to "/" we should be able to see
3741 * all mount-points from inside a jail.
3743 if (strcmp(pr->pr_path, "/") == 0)
3745 len = strlen(pr->pr_path);
3746 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3748 * Clear current buffer data, so we are sure nothing from
3749 * the valid path left there.
3751 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3752 if (*jpath == '\0') {
3753 /* Should never happen. */
3754 *sp->f_mntonname = '/';
3756 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3761 * Check with permission for a specific privilege is granted within jail. We
3762 * have a specific list of accepted privileges; the rest are denied.
3765 prison_priv_check(struct ucred *cred, int priv)
3771 * Some policies have custom handlers. This routine should not be
3772 * called for them. See priv_check_cred().
3775 case PRIV_VFS_LOOKUP:
3776 case PRIV_VFS_GENERATION:
3777 KASSERT(0, ("prison_priv_check instead of a custom handler "
3778 "called for %d\n", priv));
3786 * Privileges specific to prisons with a virtual network stack.
3787 * There might be a duplicate entry here in case the privilege
3788 * is only granted conditionally in the legacy jail case.
3792 * NFS-specific privileges.
3794 case PRIV_NFS_DAEMON:
3795 case PRIV_VFS_GETFH:
3796 case PRIV_VFS_MOUNT_EXPORTED:
3797 if (!prison_check_nfsd(cred))
3800 case PRIV_NFS_LOCKD:
3803 * Network stack privileges.
3805 case PRIV_NET_BRIDGE:
3808 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3809 case PRIV_NET_ROUTE:
3811 case PRIV_NET_SETIFMTU:
3812 case PRIV_NET_SETIFFLAGS:
3813 case PRIV_NET_SETIFCAP:
3814 case PRIV_NET_SETIFDESCR:
3815 case PRIV_NET_SETIFNAME :
3816 case PRIV_NET_SETIFMETRIC:
3817 case PRIV_NET_SETIFPHYS:
3818 case PRIV_NET_SETIFMAC:
3819 case PRIV_NET_SETLANPCP:
3820 case PRIV_NET_ADDMULTI:
3821 case PRIV_NET_DELMULTI:
3822 case PRIV_NET_HWIOCTL:
3823 case PRIV_NET_SETLLADDR:
3824 case PRIV_NET_ADDIFGROUP:
3825 case PRIV_NET_DELIFGROUP:
3826 case PRIV_NET_IFCREATE:
3827 case PRIV_NET_IFDESTROY:
3828 case PRIV_NET_ADDIFADDR:
3829 case PRIV_NET_DELIFADDR:
3832 case PRIV_NET_SETIFVNET:
3833 case PRIV_NET_SETIFFIB:
3839 * 802.11-related privileges.
3841 case PRIV_NET80211_VAP_GETKEY:
3842 case PRIV_NET80211_VAP_MANAGE:
3848 case PRIV_NETATM_CFG:
3849 case PRIV_NETATM_ADD:
3850 case PRIV_NETATM_DEL:
3851 case PRIV_NETATM_SET:
3854 * Bluetooth privileges.
3856 case PRIV_NETBLUETOOTH_RAW:
3860 * Netgraph and netgraph module privileges.
3862 case PRIV_NETGRAPH_CONTROL:
3864 case PRIV_NETGRAPH_TTY:
3868 * IPv4 and IPv6 privileges.
3870 case PRIV_NETINET_IPFW:
3871 case PRIV_NETINET_DIVERT:
3872 case PRIV_NETINET_PF:
3873 case PRIV_NETINET_DUMMYNET:
3874 case PRIV_NETINET_CARP:
3875 case PRIV_NETINET_MROUTE:
3876 case PRIV_NETINET_RAW:
3877 case PRIV_NETINET_ADDRCTRL6:
3878 case PRIV_NETINET_ND6:
3879 case PRIV_NETINET_SCOPE6:
3880 case PRIV_NETINET_ALIFETIME6:
3881 case PRIV_NETINET_IPSEC:
3882 case PRIV_NETINET_BINDANY:
3897 * No default: or deny here.
3898 * In case of no permit fall through to next switch().
3900 if (cred->cr_prison->pr_flags & PR_VNET)
3907 * Allow ktrace privileges for root in jail.
3913 * Allow jailed processes to configure audit identity and
3914 * submit audit records (login, etc). In the future we may
3915 * want to further refine the relationship between audit and
3918 case PRIV_AUDIT_GETAUDIT:
3919 case PRIV_AUDIT_SETAUDIT:
3920 case PRIV_AUDIT_SUBMIT:
3924 * Allow jailed processes to manipulate process UNIX
3925 * credentials in any way they see fit.
3927 case PRIV_CRED_SETUID:
3928 case PRIV_CRED_SETEUID:
3929 case PRIV_CRED_SETGID:
3930 case PRIV_CRED_SETEGID:
3931 case PRIV_CRED_SETGROUPS:
3932 case PRIV_CRED_SETREUID:
3933 case PRIV_CRED_SETREGID:
3934 case PRIV_CRED_SETRESUID:
3935 case PRIV_CRED_SETRESGID:
3938 * Jail implements visibility constraints already, so allow
3939 * jailed root to override uid/gid-based constraints.
3941 case PRIV_SEEOTHERGIDS:
3942 case PRIV_SEEOTHERUIDS:
3945 * Jail implements inter-process debugging limits already, so
3946 * allow jailed root various debugging privileges.
3948 case PRIV_DEBUG_DIFFCRED:
3949 case PRIV_DEBUG_SUGID:
3950 case PRIV_DEBUG_UNPRIV:
3953 * Allow jail to set various resource limits and login
3954 * properties, and for now, exceed process resource limits.
3956 case PRIV_PROC_LIMIT:
3957 case PRIV_PROC_SETLOGIN:
3958 case PRIV_PROC_SETRLIMIT:
3961 * System V and POSIX IPC privileges are granted in jail.
3964 case PRIV_IPC_WRITE:
3965 case PRIV_IPC_ADMIN:
3966 case PRIV_IPC_MSGSIZE:
3970 * Jail operations within a jail work on child jails.
3972 case PRIV_JAIL_ATTACH:
3974 case PRIV_JAIL_REMOVE:
3977 * Jail implements its own inter-process limits, so allow
3978 * root processes in jail to change scheduling on other
3979 * processes in the same jail. Likewise for signalling.
3981 case PRIV_SCHED_DIFFCRED:
3982 case PRIV_SCHED_CPUSET:
3983 case PRIV_SIGNAL_DIFFCRED:
3984 case PRIV_SIGNAL_SUGID:
3987 * Allow jailed processes to write to sysctls marked as jail
3990 case PRIV_SYSCTL_WRITEJAIL:
3993 * Allow root in jail to manage a variety of quota
3994 * properties. These should likely be conditional on a
3995 * configuration option.
3997 case PRIV_VFS_GETQUOTA:
3998 case PRIV_VFS_SETQUOTA:
4001 * Since Jail relies on chroot() to implement file system
4002 * protections, grant many VFS privileges to root in jail.
4003 * Be careful to exclude mount-related and NFS-related
4007 case PRIV_VFS_WRITE:
4008 case PRIV_VFS_ADMIN:
4010 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4011 case PRIV_VFS_CHFLAGS_DEV:
4012 case PRIV_VFS_CHOWN:
4013 case PRIV_VFS_CHROOT:
4014 case PRIV_VFS_RETAINSUGID:
4015 case PRIV_VFS_FCHROOT:
4017 case PRIV_VFS_SETGID:
4019 case PRIV_VFS_STICKYFILE:
4022 * As in the non-jail case, non-root users are expected to be
4023 * able to read kernel/physical memory (provided /dev/[k]mem
4024 * exists in the jail and they have permission to access it).
4026 case PRIV_KMEM_READ:
4030 * Depending on the global setting, allow privilege of
4031 * setting system flags.
4033 case PRIV_VFS_SYSFLAGS:
4034 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4040 * Depending on the global setting, allow privilege of
4041 * mounting/unmounting file systems.
4043 case PRIV_VFS_MOUNT:
4044 case PRIV_VFS_UNMOUNT:
4045 case PRIV_VFS_MOUNT_NONUSER:
4046 case PRIV_VFS_MOUNT_OWNER:
4047 pr = cred->cr_prison;
4049 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4057 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4058 * policy. priv_check_cred will not specifically allow it, and
4059 * we may want a MAC policy to allow it.
4061 case PRIV_VFS_READ_DIR:
4065 * Conditionnaly allow locking (unlocking) physical pages
4069 case PRIV_VM_MUNLOCK:
4070 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4076 * Conditionally allow jailed root to bind reserved ports.
4078 case PRIV_NETINET_RESERVEDPORT:
4079 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4085 * Allow jailed root to reuse in-use ports.
4087 case PRIV_NETINET_REUSEPORT:
4091 * Allow jailed root to set certain IPv4/6 (option) headers.
4093 case PRIV_NETINET_SETHDROPTS:
4097 * Conditionally allow creating raw sockets in jail.
4099 case PRIV_NETINET_RAW:
4100 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4106 * Since jail implements its own visibility limits on netstat
4107 * sysctls, allow getcred. This allows identd to work in
4110 case PRIV_NETINET_GETCRED:
4114 * Allow jailed root to set loginclass.
4116 case PRIV_PROC_SETLOGINCLASS:
4120 * Do not allow a process inside a jail to read the kernel
4121 * message buffer unless explicitly permitted.
4124 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4130 * In all remaining cases, deny the privilege request. This
4131 * includes almost all network privileges, many system
4132 * configuration privileges.
4139 * Return the part of pr2's name that is relative to pr1, or the whole name
4140 * if it does not directly follow.
4144 prison_name(struct prison *pr1, struct prison *pr2)
4148 /* Jails see themselves as "0" (if they see themselves at all). */
4151 name = pr2->pr_name;
4152 if (prison_ischild(pr1, pr2)) {
4154 * pr1 isn't locked (and allprison_lock may not be either)
4155 * so its length can't be counted on. But the number of dots
4156 * can be counted on - and counted.
4158 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4159 name = strchr(name, '.') + 1;
4165 * Return the part of pr2's path that is relative to pr1, or the whole path
4166 * if it does not directly follow.
4169 prison_path(struct prison *pr1, struct prison *pr2)
4171 char *path1, *path2;
4174 path1 = pr1->pr_path;
4175 path2 = pr2->pr_path;
4176 if (!strcmp(path1, "/"))
4178 len1 = strlen(path1);
4179 if (strncmp(path1, path2, len1))
4181 if (path2[len1] == '\0')
4183 if (path2[len1] == '/')
4184 return (path2 + len1);
4189 * Jail-related sysctls.
4191 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4194 #if defined(INET) || defined(INET6)
4196 * Copy address array to memory that would be then SYSCTL_OUT-ed.
4197 * sysctl_jail_list() helper.
4200 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4202 const struct prison_ip *pip;
4203 const size_t size = pr_families[af].size;
4206 mtx_assert(&pr->pr_mtx, MA_OWNED);
4207 if ((pip = pr->pr_addrs[af]) != NULL) {
4208 if (*len < pip->ips) {
4210 mtx_unlock(&pr->pr_mtx);
4211 *out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4212 mtx_lock(&pr->pr_mtx);
4215 bcopy(pip->pr_ip, *out, pip->ips * size);
4221 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4224 struct prison *pr, *cpr;
4226 struct in_addr *ip4 = NULL;
4230 struct in6_addr *ip6 = NULL;
4235 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4236 pr = req->td->td_ucred->cr_prison;
4238 sx_slock(&allprison_lock);
4239 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4240 mtx_lock(&cpr->pr_mtx);
4242 prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4245 prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4247 bzero(xp, sizeof(*xp));
4248 xp->pr_version = XPRISON_VERSION;
4249 xp->pr_id = cpr->pr_id;
4250 xp->pr_state = cpr->pr_state;
4251 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4252 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4253 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4260 mtx_unlock(&cpr->pr_mtx);
4261 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4265 if (xp->pr_ip4s > 0) {
4266 error = SYSCTL_OUT(req, ip4,
4267 xp->pr_ip4s * sizeof(struct in_addr));
4273 if (xp->pr_ip6s > 0) {
4274 error = SYSCTL_OUT(req, ip6,
4275 xp->pr_ip6s * sizeof(struct in6_addr));
4281 sx_sunlock(&allprison_lock);
4292 SYSCTL_OID(_security_jail, OID_AUTO, list,
4293 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4294 sysctl_jail_list, "S", "List of active jails");
4297 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4301 injail = jailed(req->td->td_ucred);
4302 error = SYSCTL_OUT(req, &injail, sizeof(injail));
4307 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4308 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4309 sysctl_jail_jailed, "I", "Process in jail?");
4312 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4314 int error, havevnet;
4316 struct ucred *cred = req->td->td_ucred;
4318 havevnet = jailed(cred) && prison_owns_vnet(cred);
4322 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4327 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4328 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4329 sysctl_jail_vnet, "I", "Jail owns vnet?");
4331 #if defined(INET) || defined(INET6)
4332 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4333 &jail_max_af_ips, 0,
4334 "Number of IP addresses a jail may have at most per address family (deprecated)");
4338 * Default parameters for jail(2) compatibility. For historical reasons,
4339 * the sysctl names have varying similarity to the parameter names. Prisons
4340 * just see their own parameters, and can't change them.
4343 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4347 /* Get the current flag value, and convert it to a boolean. */
4348 if (req->td->td_ucred->cr_prison == &prison0) {
4349 mtx_lock(&prison0.pr_mtx);
4350 i = (jail_default_allow & arg2) != 0;
4351 mtx_unlock(&prison0.pr_mtx);
4353 i = prison_allow(req->td->td_ucred, arg2);
4357 error = sysctl_handle_int(oidp, &i, 0, req);
4358 if (error || !req->newptr)
4364 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4367 mtx_lock(&prison0.pr_mtx);
4368 jail_default_allow = (jail_default_allow & ~arg2) | i;
4369 mtx_unlock(&prison0.pr_mtx);
4373 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4374 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4375 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4376 "Processes in jail can set their hostnames (deprecated)");
4377 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4378 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4379 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4380 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4381 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4382 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4383 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4384 "Processes in jail can use System V IPC primitives (deprecated)");
4385 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4386 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4387 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4388 "Prison root can create raw sockets (deprecated)");
4389 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4390 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4391 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4392 "Processes in jail can alter system file flags (deprecated)");
4393 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4394 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4395 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4396 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4399 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4404 pr = req->td->td_ucred->cr_prison;
4405 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4406 error = sysctl_handle_int(oidp, &level, 0, req);
4407 if (error || !req->newptr)
4409 *(int *)arg1 = level;
4413 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4414 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4415 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4416 sysctl_jail_default_level, "I",
4417 "Processes in jail cannot see all mounted file systems (deprecated)");
4419 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4420 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4421 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4422 sysctl_jail_default_level, "I",
4423 "Ruleset for the devfs filesystem in jail (deprecated)");
4426 * Nodes to describe jail parameters. Maximum length of string parameters
4427 * is returned in the string itself, and the other parameters exist merely
4428 * to make themselves and their types known.
4430 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4434 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4441 switch (oidp->oid_kind & CTLTYPE)
4447 if (!(req->flags & SCTL_MASK32))
4449 return (SYSCTL_OUT(req, &l, sizeof(l)));
4453 return (SYSCTL_OUT(req, &i, sizeof(i)));
4454 case CTLTYPE_STRING:
4455 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4457 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4458 case CTLTYPE_STRUCT:
4460 return (SYSCTL_OUT(req, &s, sizeof(s)));
4466 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4467 * jail creation time but cannot be changed in an existing jail.
4469 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4470 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4471 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4472 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4473 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4474 "I", "Jail secure level");
4475 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4476 "Jail value for kern.osreldate and uname -K");
4477 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4478 "Jail value for kern.osrelease and uname -r");
4479 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4480 "I", "Jail cannot see all mounted file systems");
4481 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4482 "I", "Ruleset for in-jail devfs mounts");
4483 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4484 "B", "Jail persistence");
4486 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4487 "E,jailsys", "Virtual network stack");
4489 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4490 "B", "Jail is in the process of shutting down");
4492 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4493 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4494 "I", "Current number of child jails");
4495 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4496 "I", "Maximum number of child jails");
4498 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4499 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4501 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4502 "Jail NIS domainname");
4503 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4505 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4506 "LU", "Jail host ID");
4508 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4509 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4512 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4513 "Jail IPv4 address virtualization");
4514 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4515 "S,in_addr,a", "Jail IPv4 addresses");
4516 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4517 "B", "Do (not) use IPv4 source address selection rather than the "
4518 "primary jail IPv4 address.");
4521 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4522 "Jail IPv6 address virtualization");
4523 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4524 "S,in6_addr,a", "Jail IPv6 addresses");
4525 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4526 "B", "Do (not) use IPv6 source address selection rather than the "
4527 "primary jail IPv6 address.");
4530 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4531 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4532 "B", "Jail may set hostname");
4533 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4534 "B", "Jail may use SYSV IPC");
4535 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4536 "B", "Jail may create raw sockets");
4537 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4538 "B", "Jail may alter system file flags");
4539 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4540 "B", "Jail may set file quotas");
4541 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4542 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4543 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4544 "B", "Jail may lock (unlock) physical pages in memory");
4545 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4546 "B", "Jail may bind sockets to reserved ports");
4547 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4548 "B", "Jail may read the kernel message buffer");
4549 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4550 "B", "Unprivileged processes may use process debugging facilities");
4551 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4552 "B", "Processes in jail with uid 0 have privilege");
4554 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
4555 "B", "Mountd/nfsd may run in the jail");
4558 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4559 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4560 "B", "Jail may mount/unmount jail-friendly file systems in general");
4563 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
4564 * its associated bit in the pr_allow bitmask, or zero if the parameter was
4568 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4571 struct bool_flags *bf;
4572 struct sysctl_oid *parent;
4573 char *allow_name, *allow_noname, *allowed;
4574 #ifndef NO_SYSCTL_DESCR
4575 char *descr_deprecated;
4580 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4582 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4584 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4585 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4586 free(allow_name, M_PRISON);
4591 * See if this parameter has already beed added, i.e. a module was
4592 * previously loaded/unloaded.
4594 mtx_lock(&prison0.pr_mtx);
4595 for (bf = pr_flag_allow;
4596 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4597 atomic_load_int(&bf->flag) != 0;
4599 if (strcmp(bf->name, allow_name) == 0) {
4600 allow_flag = bf->flag;
4606 * Find a free bit in pr_allow_all, failing if there are none
4607 * (which shouldn't happen as long as we keep track of how many
4608 * potential dynamic flags exist).
4610 for (allow_flag = 1;; allow_flag <<= 1) {
4611 if (allow_flag == 0)
4613 if ((pr_allow_all & allow_flag) == 0)
4617 /* Note the parameter in the next open slot in pr_flag_allow. */
4618 for (bf = pr_flag_allow; ; bf++) {
4619 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4620 /* This should never happen, but is not fatal. */
4624 if (atomic_load_int(&bf->flag) == 0)
4627 bf->name = allow_name;
4628 bf->noname = allow_noname;
4629 pr_allow_all |= allow_flag;
4631 * prison0 always has permission for the new parameter.
4632 * Other jails must have it granted to them.
4634 prison0.pr_allow |= allow_flag;
4635 /* The flag indicates a valid entry, so make sure it is set last. */
4636 atomic_store_rel_int(&bf->flag, allow_flag);
4637 mtx_unlock(&prison0.pr_mtx);
4640 * Create sysctls for the parameter, and the back-compat global
4644 ? SYSCTL_ADD_NODE(NULL,
4645 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4646 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4647 : &sysctl___security_jail_param_allow;
4648 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4649 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4650 NULL, 0, sysctl_jail_param, "B", descr);
4652 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4653 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4654 #ifndef NO_SYSCTL_DESCR
4655 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4658 (void)SYSCTL_ADD_PROC(NULL,
4659 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4660 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4661 sysctl_jail_default_allow, "I", descr_deprecated);
4662 #ifndef NO_SYSCTL_DESCR
4663 free(descr_deprecated, M_TEMP);
4665 free(allowed, M_TEMP);
4670 mtx_unlock(&prison0.pr_mtx);
4671 free(allow_name, M_PRISON);
4672 free(allow_noname, M_PRISON);
4677 * The VFS system will register jail-aware filesystems here. They each get
4678 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4679 * attempts to mount.
4682 prison_add_vfs(struct vfsconf *vfsp)
4684 #ifdef NO_SYSCTL_DESCR
4686 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4691 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4693 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4695 free(descr, M_TEMP);
4701 prison_racct_foreach(void (*callback)(struct racct *racct,
4702 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4703 void *arg2, void *arg3)
4705 struct prison_racct *prr;
4707 ASSERT_RACCT_ENABLED();
4709 sx_slock(&allprison_lock);
4712 LIST_FOREACH(prr, &allprison_racct, prr_next)
4713 (callback)(prr->prr_racct, arg2, arg3);
4716 sx_sunlock(&allprison_lock);
4719 static struct prison_racct *
4720 prison_racct_find_locked(const char *name)
4722 struct prison_racct *prr;
4724 ASSERT_RACCT_ENABLED();
4725 sx_assert(&allprison_lock, SA_XLOCKED);
4727 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4730 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4731 if (strcmp(name, prr->prr_name) != 0)
4734 /* Found prison_racct with a matching name? */
4735 prison_racct_hold(prr);
4739 /* Add new prison_racct. */
4740 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4741 racct_create(&prr->prr_racct);
4743 strcpy(prr->prr_name, name);
4744 refcount_init(&prr->prr_refcount, 1);
4745 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4750 struct prison_racct *
4751 prison_racct_find(const char *name)
4753 struct prison_racct *prr;
4755 ASSERT_RACCT_ENABLED();
4757 sx_xlock(&allprison_lock);
4758 prr = prison_racct_find_locked(name);
4759 sx_xunlock(&allprison_lock);
4764 prison_racct_hold(struct prison_racct *prr)
4767 ASSERT_RACCT_ENABLED();
4769 refcount_acquire(&prr->prr_refcount);
4773 prison_racct_free_locked(struct prison_racct *prr)
4776 ASSERT_RACCT_ENABLED();
4777 sx_assert(&allprison_lock, SA_XLOCKED);
4779 if (refcount_release(&prr->prr_refcount)) {
4780 racct_destroy(&prr->prr_racct);
4781 LIST_REMOVE(prr, prr_next);
4782 free(prr, M_PRISON_RACCT);
4787 prison_racct_free(struct prison_racct *prr)
4790 ASSERT_RACCT_ENABLED();
4791 sx_assert(&allprison_lock, SA_UNLOCKED);
4793 if (refcount_release_if_not_last(&prr->prr_refcount))
4796 sx_xlock(&allprison_lock);
4797 prison_racct_free_locked(prr);
4798 sx_xunlock(&allprison_lock);
4802 prison_racct_attach(struct prison *pr)
4804 struct prison_racct *prr;
4806 ASSERT_RACCT_ENABLED();
4807 sx_assert(&allprison_lock, SA_XLOCKED);
4809 prr = prison_racct_find_locked(pr->pr_name);
4810 KASSERT(prr != NULL, ("cannot find prison_racct"));
4812 pr->pr_prison_racct = prr;
4816 * Handle jail renaming. From the racct point of view, renaming means
4817 * moving from one prison_racct to another.
4820 prison_racct_modify(struct prison *pr)
4826 struct prison_racct *oldprr;
4828 ASSERT_RACCT_ENABLED();
4830 sx_slock(&allproc_lock);
4831 sx_xlock(&allprison_lock);
4833 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4834 sx_xunlock(&allprison_lock);
4835 sx_sunlock(&allproc_lock);
4839 oldprr = pr->pr_prison_racct;
4840 pr->pr_prison_racct = NULL;
4842 prison_racct_attach(pr);
4845 * Move resource utilisation records.
4847 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4851 * Force rctl to reattach rules to processes.
4853 FOREACH_PROC_IN_SYSTEM(p) {
4855 cred = crhold(p->p_ucred);
4857 rctl_proc_ucred_changed(p, cred);
4862 sx_sunlock(&allproc_lock);
4863 prison_racct_free_locked(oldprr);
4864 sx_xunlock(&allprison_lock);
4868 prison_racct_detach(struct prison *pr)
4871 ASSERT_RACCT_ENABLED();
4872 sx_assert(&allprison_lock, SA_UNLOCKED);
4874 if (pr->pr_prison_racct == NULL)
4876 prison_racct_free(pr->pr_prison_racct);
4877 pr->pr_prison_racct = NULL;
4884 db_show_prison(struct prison *pr)
4886 struct bool_flags *bf;
4887 struct jailsys_flags *jsf;
4888 #if defined(INET) || defined(INET6)
4890 struct prison_ip *pip;
4894 char ip4buf[INET_ADDRSTRLEN];
4897 char ip6buf[INET6_ADDRSTRLEN];
4900 db_printf("prison %p:\n", pr);
4901 db_printf(" jid = %d\n", pr->pr_id);
4902 db_printf(" name = %s\n", pr->pr_name);
4903 db_printf(" parent = %p\n", pr->pr_parent);
4904 db_printf(" ref = %d\n", pr->pr_ref);
4905 db_printf(" uref = %d\n", pr->pr_uref);
4906 db_printf(" state = %s\n",
4907 pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4908 pr->pr_state == PRISON_STATE_DYING ? "dying" :
4910 db_printf(" path = %s\n", pr->pr_path);
4911 db_printf(" cpuset = %d\n", pr->pr_cpuset
4912 ? pr->pr_cpuset->cs_id : -1);
4914 db_printf(" vnet = %p\n", pr->pr_vnet);
4916 db_printf(" root = %p\n", pr->pr_root);
4917 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4918 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4919 db_printf(" children.max = %d\n", pr->pr_childmax);
4920 db_printf(" children.cur = %d\n", pr->pr_childcount);
4921 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4922 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4923 db_printf(" flags = 0x%x", pr->pr_flags);
4924 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4925 if (pr->pr_flags & bf->flag)
4926 db_printf(" %s", bf->name);
4927 for (jsf = pr_flag_jailsys;
4928 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4930 f = pr->pr_flags & (jsf->disable | jsf->new);
4931 db_printf(" %-16s= %s\n", jsf->name,
4932 (f != 0 && f == jsf->disable) ? "disable"
4933 : (f == jsf->new) ? "new"
4936 db_printf(" allow = 0x%x", pr->pr_allow);
4937 for (bf = pr_flag_allow;
4938 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4939 atomic_load_int(&bf->flag) != 0;
4941 if (pr->pr_allow & bf->flag)
4942 db_printf(" %s", bf->name);
4944 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4945 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4946 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4947 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4948 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4950 if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
4951 db_printf(" ip4s = %d\n", pip->ips);
4952 for (ii = 0; ii < pip->ips; ii++)
4953 db_printf(" %s %s\n",
4954 ii == 0 ? "ip4.addr =" : " ",
4956 *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
4961 if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
4962 db_printf(" ip6s = %d\n", pip->ips);
4963 for (ii = 0; ii < pip->ips; ii++)
4964 db_printf(" %s %s\n",
4965 ii == 0 ? "ip6.addr =" : " ",
4967 (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
4972 DB_SHOW_COMMAND(prison, db_show_prison_command)
4978 * Show all prisons in the list, and prison0 which is not
4981 db_show_prison(&prison0);
4982 if (!db_pager_quit) {
4983 TAILQ_FOREACH(pr, &allprison, pr_list) {
4995 /* Look for a prison with the ID and with references. */
4996 TAILQ_FOREACH(pr, &allprison, pr_list)
4997 if (pr->pr_id == addr && pr->pr_ref > 0)
5000 /* Look again, without requiring a reference. */
5001 TAILQ_FOREACH(pr, &allprison, pr_list)
5002 if (pr->pr_id == addr)
5005 /* Assume address points to a valid prison. */
5006 pr = (struct prison *)addr;