2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 1999 Poul-Henning Kamp.
5 * Copyright (c) 2008 Bjoern A. Zeeb.
6 * Copyright (c) 2009 James Gritton.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
34 #include "opt_inet6.h"
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
47 #include <sys/epoch.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
51 #include <sys/linker.h>
54 #include <sys/mutex.h>
55 #include <sys/racct.h>
57 #include <sys/refcount.h>
59 #include <sys/sysent.h>
60 #include <sys/namei.h>
61 #include <sys/mount.h>
62 #include <sys/queue.h>
63 #include <sys/socket.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
72 #include <netinet/in.h>
78 #include <security/mac/mac_framework.h>
80 #define PRISON0_HOSTUUID_MODULE "hostuuid"
82 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
83 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
85 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
88 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
90 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
94 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
96 #define _PR_IP_SADDRSEL 0
100 /* prison0 describes what is "real" about the system. */
101 struct prison prison0 = {
107 .pr_securelevel = -1,
109 .pr_state = PRISON_STATE_ALIVE,
110 .pr_childmax = JAIL_MAX,
111 .pr_hostuuid = DEFAULT_HOSTUUID,
112 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
114 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
116 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
118 .pr_allow = PR_ALLOW_ALL_STATIC,
120 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
127 struct jailsys_flags {
133 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
134 struct sx allprison_lock;
135 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
136 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
137 LIST_HEAD(, prison_racct) allprison_racct;
140 static int get_next_prid(struct prison **insprp);
141 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
142 static void prison_complete(void *context, int pending);
143 static void prison_deref(struct prison *pr, int flags);
144 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
145 static int prison_lock_xlock(struct prison *pr, int flags);
146 static void prison_cleanup(struct prison *pr);
147 static void prison_free_not_last(struct prison *pr);
148 static void prison_proc_free_not_last(struct prison *pr);
149 static void prison_proc_relink(struct prison *opr, struct prison *npr,
151 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
153 static char *prison_path(struct prison *pr1, struct prison *pr2);
155 static void prison_racct_attach(struct prison *pr);
156 static void prison_racct_modify(struct prison *pr);
157 static void prison_racct_detach(struct prison *pr);
160 /* Flags for prison_deref */
161 #define PD_DEREF 0x01 /* Decrement pr_ref */
162 #define PD_DEUREF 0x02 /* Decrement pr_uref */
163 #define PD_KILL 0x04 /* Remove jail, kill processes, etc */
164 #define PD_LOCKED 0x10 /* pr_mtx is held */
165 #define PD_LIST_SLOCKED 0x20 /* allprison_lock is held shared */
166 #define PD_LIST_XLOCKED 0x40 /* allprison_lock is held exclusive */
167 #define PD_OP_FLAGS 0x07 /* Operation flags */
168 #define PD_LOCK_FLAGS 0x70 /* Lock status flags */
171 * Parameter names corresponding to PR_* flag values. Size values are for kvm
172 * as we cannot figure out the size of a sparse array, or an array without a
175 static struct bool_flags pr_flag_bool[] = {
176 {"persist", "nopersist", PR_PERSIST},
178 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
181 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
184 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
186 static struct jailsys_flags pr_flag_jailsys[] = {
187 {"host", 0, PR_HOST},
189 {"vnet", 0, PR_VNET},
192 {"ip4", PR_IP4_USER, PR_IP4_USER},
195 {"ip6", PR_IP6_USER, PR_IP6_USER},
198 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
201 * Make this array full-size so dynamic parameters can be added.
202 * It is protected by prison0.mtx, but lockless reading is allowed
203 * with an atomic check of the flag values.
205 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
206 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
207 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
208 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
209 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
210 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
211 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
212 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
213 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
214 {"allow.reserved_ports", "allow.noreserved_ports",
215 PR_ALLOW_RESERVED_PORTS},
216 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
217 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
218 PR_ALLOW_UNPRIV_DEBUG},
219 {"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
221 {"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
223 {"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
225 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
226 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
228 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
229 PR_ALLOW_RESERVED_PORTS | \
230 PR_ALLOW_UNPRIV_DEBUG | \
232 #define JAIL_DEFAULT_ENFORCE_STATFS 2
233 #define JAIL_DEFAULT_DEVFS_RSNUM 0
234 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
235 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
236 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
237 #if defined(INET) || defined(INET6)
238 static unsigned jail_max_af_ips = 255;
242 * Initialize the parts of prison0 that can't be static-initialized with
243 * constants. This is called from proc0_init() after creating thread0 cpuset.
248 uint8_t *file, *data;
250 char buf[sizeof(prison0.pr_hostuuid)];
253 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
254 prison0.pr_osreldate = osreldate;
255 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
257 /* If we have a preloaded hostuuid, use it. */
258 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
260 data = preload_fetch_addr(file);
261 size = preload_fetch_size(file);
264 * The preloaded data may include trailing whitespace, almost
265 * certainly a newline; skip over any whitespace or
266 * non-printable characters to be safe.
268 while (size > 0 && data[size - 1] <= 0x20) {
275 * Not NUL-terminated when passed from loader, but
276 * validate_uuid requires that due to using sscanf (as
277 * does the subsequent strlcpy, since it still reads
278 * past the given size to return the true length);
279 * bounce to a temporary buffer to fix.
281 if (size >= sizeof(buf))
284 memcpy(buf, data, size);
287 if (validate_uuid(buf, size, NULL, 0) != 0)
291 (void)strlcpy(prison0.pr_hostuuid, buf,
292 sizeof(prison0.pr_hostuuid));
295 if (bootverbose && !valid) {
296 printf("hostuuid: preload data malformed: '%.*s'\n",
302 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
311 sys_jail(struct thread *td, struct jail_args *uap)
317 error = copyin(uap->jail, &version, sizeof(uint32_t));
326 /* FreeBSD single IPv4 jails. */
327 bzero(&j, sizeof(struct jail));
328 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
331 j.version = j0.version;
333 j.hostname = j0.hostname;
334 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
340 * Version 1 was used by multi-IPv4 jail implementations
341 * that never made it into the official kernel.
345 case 2: /* JAIL_API_VERSION */
346 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
347 error = copyin(uap->jail, &j, sizeof(struct jail));
353 /* Sci-Fi jails are not supported, sorry. */
356 return (kern_jail(td, &j));
360 kern_jail(struct thread *td, struct jail *j)
362 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
371 char *u_path, *u_hostname, *u_name;
372 struct bool_flags *bf;
375 struct in_addr *u_ip4;
378 struct in6_addr *u_ip6;
381 int error, enforce_statfs;
383 bzero(&optiov, sizeof(optiov));
384 opt.uio_iov = optiov;
388 opt.uio_segflg = UIO_SYSSPACE;
389 opt.uio_rw = UIO_READ;
392 /* Set permissions for top-level jails from sysctls. */
393 if (!jailed(td->td_ucred)) {
394 for (bf = pr_flag_allow;
395 bf < pr_flag_allow + nitems(pr_flag_allow) &&
396 atomic_load_int(&bf->flag) != 0;
398 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
399 (jail_default_allow & bf->flag)
400 ? bf->name : bf->noname);
401 optiov[opt.uio_iovcnt].iov_len =
402 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
405 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
406 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
408 enforce_statfs = jail_default_enforce_statfs;
409 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
410 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
414 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
416 ip4s = (j->version == 0) ? 1 : j->ip4s;
417 if (ip4s > jail_max_af_ips)
419 tmplen += ip4s * sizeof(struct in_addr);
425 if (j->ip6s > jail_max_af_ips)
427 tmplen += j->ip6s * sizeof(struct in6_addr);
432 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
433 u_hostname = u_path + MAXPATHLEN;
434 u_name = u_hostname + MAXHOSTNAMELEN;
436 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
440 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
442 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
445 optiov[opt.uio_iovcnt].iov_base = "path";
446 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
448 optiov[opt.uio_iovcnt].iov_base = u_path;
449 error = copyinstr(j->path, u_path, MAXPATHLEN,
450 &optiov[opt.uio_iovcnt].iov_len);
452 free(u_path, M_TEMP);
456 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
457 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
459 optiov[opt.uio_iovcnt].iov_base = u_hostname;
460 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
461 &optiov[opt.uio_iovcnt].iov_len);
463 free(u_path, M_TEMP);
467 if (j->jailname != NULL) {
468 optiov[opt.uio_iovcnt].iov_base = "name";
469 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
471 optiov[opt.uio_iovcnt].iov_base = u_name;
472 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
473 &optiov[opt.uio_iovcnt].iov_len);
475 free(u_path, M_TEMP);
481 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
482 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
484 optiov[opt.uio_iovcnt].iov_base = u_ip4;
485 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
487 u_ip4->s_addr = j->ip4s;
489 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
491 free(u_path, M_TEMP);
498 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
499 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
501 optiov[opt.uio_iovcnt].iov_base = u_ip6;
502 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
503 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
505 free(u_path, M_TEMP);
510 KASSERT(opt.uio_iovcnt <= nitems(optiov),
511 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
512 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
513 free(u_path, M_TEMP);
518 * struct jail_set_args {
519 * struct iovec *iovp;
520 * unsigned int iovcnt;
525 sys_jail_set(struct thread *td, struct jail_set_args *uap)
530 /* Check that we have an even number of iovecs. */
534 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
537 error = kern_jail_set(td, auio, uap->flags);
542 #if defined(INET) || defined(INET6)
543 typedef int prison_addr_cmp_t(const void *, const void *);
544 typedef bool prison_addr_valid_t(const void *);
545 static const struct pr_family {
547 prison_addr_cmp_t *cmp;
548 prison_addr_valid_t *valid;
550 } pr_families[PR_FAMILY_MAX] = {
553 .size = sizeof(struct in_addr),
554 .cmp = prison_qcmp_v4,
555 .valid = prison_valid_v4,
556 .ip_flag = PR_IP4_USER,
561 .size = sizeof(struct in6_addr),
562 .cmp = prison_qcmp_v6,
563 .valid = prison_valid_v6,
564 .ip_flag = PR_IP6_USER,
570 * Network address lists (pr_addrs) allocation for jails. The addresses
571 * are accessed locklessly by the network stack, thus need to be protected by
575 struct epoch_context ctx;
579 * XXX Variable-length automatic arrays in union may be
580 * supported in future C.
584 struct in_addr pr_ip4[];
585 struct in6_addr pr_ip6[];
587 #else /* No future C :( */
593 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
596 MPASS(af < PR_FAMILY_MAX);
597 MPASS(idx >= 0 && idx < pip->ips);
599 return (pip->pr_ip + pr_families[af].size * idx);
602 static struct prison_ip *
603 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
605 struct prison_ip *pip;
607 pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
615 * Allocate and copyin user supplied address list, sorting and validating.
616 * kern_jail_set() helper.
618 static struct prison_ip *
619 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
621 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
622 const size_t size = pr_families[af].size;
623 struct prison_ip *pip;
625 pip = prison_ip_alloc(af, cnt, M_WAITOK);
626 bcopy(op, pip->pr_ip, cnt * size);
628 * IP addresses are all sorted but ip[0] to preserve
629 * the primary IP address as given from userland.
630 * This special IP is used for unbound outgoing
631 * connections as well for "loopback" traffic in case
632 * source address selection cannot find any more fitting
633 * address to connect from.
636 qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
638 * Check for duplicate addresses and do some simple
639 * zero and broadcast checks. If users give other bogus
640 * addresses it is their problem.
642 for (int i = 0; i < cnt; i++) {
643 if (!pr_families[af].valid(PR_IP(pip, af, i))) {
648 (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
649 cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
659 * Allocate and dup parent prison address list.
660 * kern_jail_set() helper.
663 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
665 const struct prison_ip *ppip = ppr->pr_addrs[af];
666 struct prison_ip *pip;
669 pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
670 bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
671 pr->pr_addrs[af] = pip;
676 * Make sure the new set of IP addresses is a subset of the parent's list.
677 * Don't worry about the parent being unlocked, as any setting is done with
678 * allprison_lock held.
679 * kern_jail_set() helper.
682 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
683 const pr_family_t af)
685 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
691 for (i = 0; i < ppip->ips; i++)
692 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
696 /* Main address not present in parent. */
700 for (i = j = 1; i < pip->ips; i++) {
701 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
702 /* Equals to parent primary address. */
704 for (; j < ppip->ips; j++)
705 if (cmp(PR_IP(pip, af, i),
706 PR_IP(ppip, af, j)) == 0)
712 /* Address not present in parent. */
719 * Check for conflicting IP addresses. We permit them if there is no more
720 * than one IP on each jail. If there is a duplicate on a jail with more
721 * than one IP stop checking and return error.
722 * kern_jail_set() helper.
725 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
726 struct prison_ip *pip, pr_family_t af)
728 const struct prison *tppr, *tpr;
732 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
733 if (tppr->pr_flags & PR_VNET)
738 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
741 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
743 !prison_isalive(tpr)) {
747 if (!(tpr->pr_flags & pr_families[af].ip_flag))
750 if (tpr->pr_addrs[af] == NULL ||
751 (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
753 for (int i = 0; i < pip->ips; i++)
754 if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
761 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
762 "prison must start with epoch context");
764 prison_ip_free_deferred(epoch_context_t ctx)
771 prison_ip_free(struct prison_ip *pip)
775 NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
779 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
781 struct prison_ip **mem, *old;
783 mtx_assert(&pr->pr_mtx, MA_OWNED);
785 mem = &pr->pr_addrs[af];
788 atomic_store_ptr(mem, new);
793 * Restrict a prison's IP address list with its parent's, possibly replacing
794 * it. Return true if succeed, otherwise should redo.
795 * kern_jail_set() helper.
798 prison_ip_restrict(struct prison *pr, const pr_family_t af,
799 struct prison_ip **newp)
801 struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
802 struct prison_ip *pip = pr->pr_addrs[af];
803 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
804 const size_t size = pr_families[af].size;
805 struct prison_ip *new = newp != NULL ? *newp : NULL;
808 mtx_assert(&pr->pr_mtx, MA_OWNED);
811 * Due to epoch-synchronized access to the IP address lists we always
812 * allocate a new list even if the old one has enough space. We could
813 * atomically update an IPv4 address inside a list, but that would
814 * screw up sorting, and in case of IPv6 we can't even atomically write
819 prison_ip_set(pr, af, NULL);
823 if (!(pr->pr_flags & pr_families[af].ip_flag)) {
825 new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
827 return (false); /* Redo */
829 /* This has no user settings, so just copy the parent's list. */
830 MPASS(new->ips == ppip->ips);
831 bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
832 prison_ip_set(pr, af, new);
834 *newp = NULL; /* Used */
835 } else if (pip != NULL) {
836 /* Remove addresses that aren't in the parent. */
839 i = 0; /* index in pip */
840 ips = 0; /* index in new */
843 new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
845 return (false); /* Redo */
848 for (int pi = 0; pi < ppip->ips; pi++)
849 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
850 /* Found our primary address in parent. */
851 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
857 for (int pi = 1; i < pip->ips; ) {
858 /* Check against primary, which is unsorted. */
859 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
860 /* Matches parent's primary address. */
861 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
867 /* The rest are sorted. */
868 switch (pi >= ppip->ips ? -1 :
869 cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
874 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
886 if (newp == NULL || *newp == NULL)
890 /* Shrink to real size */
891 KASSERT((new->ips >= ips),
892 ("Out-of-bounds write to prison_ip %p", new));
895 prison_ip_set(pr, af, new);
897 *newp = NULL; /* Used */
903 * Fast-path check if an address belongs to a prison.
906 prison_ip_check(const struct prison *pr, const pr_family_t af,
909 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
910 struct prison_ip *pip;
913 MPASS(mtx_owned(&pr->pr_mtx) ||
914 in_epoch(net_epoch_preempt) ||
915 sx_xlocked(&allprison_lock));
917 pip = atomic_load_ptr(&pr->pr_addrs[af]);
918 if (__predict_false(pip == NULL))
919 return (EAFNOSUPPORT);
921 /* Check the primary IP. */
922 if (cmp(PR_IP(pip, af, 0), addr) == 0)
926 * All the other IPs are sorted so we can do a binary search.
932 d = cmp(PR_IP(pip, af, i + 1), addr);
941 return (EADDRNOTAVAIL);
945 * Grab primary IP. Historically required mutex, but nothing prevents
946 * us to support epoch-protected access. Is it used in fast path?
947 * in{6}_jail.c helper
950 prison_ip_get0(const struct prison *pr, const pr_family_t af)
952 const struct prison_ip *pip = pr->pr_addrs[af];
954 mtx_assert(&pr->pr_mtx, MA_OWNED);
961 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
964 return (pr->pr_addrs[af]->ips);
966 #endif /* defined(INET) || defined(INET6) */
969 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
973 struct prison_ip *ip4;
976 struct prison_ip *ip6;
979 struct vfsoptlist *opts;
980 struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
982 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
983 char *g_path, *osrelstr;
984 struct bool_flags *bf;
985 struct jailsys_flags *jsf;
986 #if defined(INET) || defined(INET6)
990 size_t namelen, onamelen, pnamelen;
991 int born, created, cuflags, descend, drflags, enforce;
992 int error, errmsg_len, errmsg_pos;
993 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
994 int jid, jsys, len, level;
995 int childmax, osreldt, rsnum, slevel;
1004 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1005 uint64_t pr_allow_diff;
1009 error = priv_check(td, PRIV_JAIL_SET);
1010 if (!error && (flags & JAIL_ATTACH))
1011 error = priv_check(td, PRIV_JAIL_ATTACH);
1014 mypr = td->td_ucred->cr_prison;
1015 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
1017 if (flags & ~JAIL_SET_MASK)
1021 * Check all the parameters before committing to anything. Not all
1022 * errors can be caught early, but we may as well try. Also, this
1023 * takes care of some expensive stuff (path lookup) before getting
1024 * the allprison lock.
1026 * XXX Jails are not filesystems, and jail parameters are not mount
1027 * options. But it makes more sense to re-use the vfsopt code
1028 * than duplicate it under a different name.
1030 error = vfs_buildopts(optuio, &opts);
1041 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1044 vfs_opterror(opts, "no valid operation (create or update)");
1048 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1049 if (error == ENOENT)
1051 else if (error != 0)
1054 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1055 if (error == ENOENT)
1057 else if (error != 0)
1063 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1064 if (error == ENOENT)
1066 else if (error != 0)
1071 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1072 if (error == ENOENT)
1074 else if (error != 0)
1076 else if (enforce < 0 || enforce > 2) {
1082 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1083 if (error == ENOENT)
1085 else if (error != 0)
1090 pr_flags = ch_flags = 0;
1091 for (bf = pr_flag_bool;
1092 bf < pr_flag_bool + nitems(pr_flag_bool);
1094 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1095 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1097 ch_flags |= pr_flags;
1098 for (jsf = pr_flag_jailsys;
1099 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1101 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1102 if (error == ENOENT)
1107 case JAIL_SYS_DISABLE:
1108 if (!jsf->disable) {
1112 pr_flags |= jsf->disable;
1115 pr_flags |= jsf->new;
1117 case JAIL_SYS_INHERIT:
1123 ch_flags |= jsf->new | jsf->disable;
1125 if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1126 && !(pr_flags & PR_PERSIST)) {
1128 vfs_opterror(opts, "new jail must persist or attach");
1132 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1134 vfs_opterror(opts, "vnet cannot be changed after creation");
1139 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1141 vfs_opterror(opts, "ip4 cannot be changed after creation");
1146 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1148 vfs_opterror(opts, "ip6 cannot be changed after creation");
1153 pr_allow = ch_allow = 0;
1154 for (bf = pr_flag_allow;
1155 bf < pr_flag_allow + nitems(pr_flag_allow) &&
1156 atomic_load_int(&bf->flag) != 0;
1158 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1159 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1161 ch_allow |= pr_allow;
1163 error = vfs_getopt(opts, "name", (void **)&name, &len);
1164 if (error == ENOENT)
1166 else if (error != 0)
1169 if (len == 0 || name[len - 1] != '\0') {
1173 if (len > MAXHOSTNAMELEN) {
1174 error = ENAMETOOLONG;
1179 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1180 if (error == ENOENT)
1182 else if (error != 0)
1185 ch_flags |= PR_HOST;
1186 pr_flags |= PR_HOST;
1187 if (len == 0 || host[len - 1] != '\0') {
1191 if (len > MAXHOSTNAMELEN) {
1192 error = ENAMETOOLONG;
1197 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1198 if (error == ENOENT)
1200 else if (error != 0)
1203 ch_flags |= PR_HOST;
1204 pr_flags |= PR_HOST;
1205 if (len == 0 || domain[len - 1] != '\0') {
1209 if (len > MAXHOSTNAMELEN) {
1210 error = ENAMETOOLONG;
1215 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1216 if (error == ENOENT)
1218 else if (error != 0)
1221 ch_flags |= PR_HOST;
1222 pr_flags |= PR_HOST;
1223 if (len == 0 || uuid[len - 1] != '\0') {
1227 if (len > HOSTUUIDLEN) {
1228 error = ENAMETOOLONG;
1233 #ifdef COMPAT_FREEBSD32
1234 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1237 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1241 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1242 if (error == ENOENT)
1244 else if (error != 0)
1248 ch_flags |= PR_HOST;
1249 pr_flags |= PR_HOST;
1253 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1254 if (error == ENOENT)
1256 else if (error != 0)
1258 else if (ip4s & (sizeof(struct in_addr) - 1)) {
1262 ch_flags |= PR_IP4_USER;
1263 pr_flags |= PR_IP4_USER;
1265 ip4s /= sizeof(struct in_addr);
1266 if (ip4s > jail_max_af_ips) {
1268 vfs_opterror(opts, "too many IPv4 addresses");
1271 ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1281 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1282 if (error == ENOENT)
1284 else if (error != 0)
1286 else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1290 ch_flags |= PR_IP6_USER;
1291 pr_flags |= PR_IP6_USER;
1293 ip6s /= sizeof(struct in6_addr);
1294 if (ip6s > jail_max_af_ips) {
1296 vfs_opterror(opts, "too many IPv6 addresses");
1299 ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1308 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1309 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1312 "vnet jails cannot have IP address restrictions");
1317 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1318 if (error == ENOENT)
1320 else if (error != 0)
1323 if (flags & JAIL_UPDATE) {
1326 "osrelease cannot be changed after creation");
1329 if (len == 0 || osrelstr[len - 1] != '\0') {
1333 if (len >= OSRELEASELEN) {
1334 error = ENAMETOOLONG;
1336 "osrelease string must be 1-%d bytes long",
1342 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1343 if (error == ENOENT)
1345 else if (error != 0)
1348 if (flags & JAIL_UPDATE) {
1351 "osreldate cannot be changed after creation");
1356 vfs_opterror(opts, "osreldate cannot be 0");
1362 error = vfs_getopt(opts, "path", (void **)&path, &len);
1363 if (error == ENOENT)
1365 else if (error != 0)
1368 if (flags & JAIL_UPDATE) {
1371 "path cannot be changed after creation");
1374 if (len == 0 || path[len - 1] != '\0') {
1378 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1384 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1385 strlcpy(g_path, path, MAXPATHLEN);
1386 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1390 /* exit on other errors */
1393 if (root->v_type != VDIR) {
1402 * Find the specified jail, or at least its parent.
1403 * This abuses the file error codes ENOENT and EEXIST.
1407 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1408 namelc = strrchr(name, '.');
1409 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1413 sx_xlock(&allprison_lock);
1414 drflags = PD_LIST_XLOCKED;
1416 if (!prison_isalive(ppr)) {
1417 /* This jail is dying. This process will surely follow. */
1424 vfs_opterror(opts, "negative jid");
1428 * See if a requested jid already exists. Keep track of
1429 * where it can be inserted later.
1431 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1432 if (inspr->pr_id < jid)
1434 if (inspr->pr_id > jid)
1437 mtx_lock(&pr->pr_mtx);
1438 drflags |= PD_LOCKED;
1443 /* Create: jid must not exist. */
1444 if (cuflags == JAIL_CREATE) {
1446 * Even creators that cannot see the jail will
1450 vfs_opterror(opts, "jail %d already exists",
1454 if (!prison_ischild(mypr, pr)) {
1456 * Updaters get ENOENT if they cannot see the
1457 * jail. This is true even for CREATE | UPDATE,
1458 * which normally cannot give this error.
1461 vfs_opterror(opts, "jail %d not found", jid);
1464 ppr = pr->pr_parent;
1465 if (!prison_isalive(ppr)) {
1467 vfs_opterror(opts, "jail %d is dying",
1471 if (!prison_isalive(pr)) {
1472 if (!(flags & JAIL_DYING)) {
1474 vfs_opterror(opts, "jail %d is dying",
1478 if ((flags & JAIL_ATTACH) ||
1479 (pr_flags & PR_PERSIST)) {
1481 * A dying jail might be resurrected
1482 * (via attach or persist), but first
1483 * it must determine if another jail
1484 * has claimed its name. Accomplish
1485 * this by implicitly re-setting the
1489 name = prison_name(mypr, pr);
1493 /* Update: jid must exist. */
1494 if (cuflags == JAIL_UPDATE) {
1496 vfs_opterror(opts, "jail %d not found", jid);
1502 * If the caller provided a name, look for a jail by that name.
1503 * This has different semantics for creates and updates keyed by jid
1504 * (where the name must not already exist in a different jail),
1505 * and updates keyed by the name itself (where the name must exist
1506 * because that is the jail being updated).
1510 namelc = strrchr(name, '.');
1515 * This is a hierarchical name. Split it into the
1516 * parent and child names, and make sure the parent
1517 * exists or matches an already found jail.
1520 if (strncmp(name, ppr->pr_name, namelc - name)
1521 || ppr->pr_name[namelc - name] != '\0') {
1524 "cannot change jail's parent");
1529 ppr = prison_find_name(mypr, name);
1533 "jail \"%s\" not found", name);
1536 mtx_unlock(&ppr->pr_mtx);
1537 if (!prison_isalive(ppr)) {
1540 "jail \"%s\" is dying", name);
1547 if (namelc[0] != '\0') {
1549 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1551 FOREACH_PRISON_CHILD(ppr, tpr) {
1553 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1554 if (prison_isalive(tpr)) {
1556 cuflags != JAIL_CREATE) {
1562 mtx_lock(&pr->pr_mtx);
1563 drflags |= PD_LOCKED;
1567 * Create, or update(jid):
1568 * name must not exist in an
1569 * active sibling jail.
1573 "jail \"%s\" already exists",
1578 cuflags != JAIL_CREATE) {
1583 /* If no active jail is found, use a dying one. */
1584 if (deadpr != NULL && pr == NULL) {
1585 if (flags & JAIL_DYING) {
1587 mtx_lock(&pr->pr_mtx);
1588 drflags |= PD_LOCKED;
1589 } else if (cuflags == JAIL_UPDATE) {
1592 "jail \"%s\" is dying", name);
1596 /* Update: name must exist if no jid. */
1597 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1599 vfs_opterror(opts, "jail \"%s\" not found",
1605 /* Update: must provide a jid or name. */
1606 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1608 vfs_opterror(opts, "update specified no jail");
1612 /* If there's no prison to update, create a new one and link it in. */
1613 created = pr == NULL;
1615 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1616 if (tpr->pr_childcount >= tpr->pr_childmax) {
1618 vfs_opterror(opts, "prison limit exceeded");
1621 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1623 vfs_opterror(opts, "no available jail IDs");
1627 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1628 pr->pr_state = PRISON_STATE_INVALID;
1629 refcount_init(&pr->pr_ref, 1);
1630 refcount_init(&pr->pr_uref, 0);
1631 drflags |= PD_DEREF;
1632 LIST_INIT(&pr->pr_children);
1633 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1634 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1638 TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1640 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1642 pr->pr_parent = ppr;
1644 prison_proc_hold(ppr);
1645 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1646 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1647 tpr->pr_childcount++;
1649 /* Set some default values, and inherit some from the parent. */
1654 root = mypr->pr_root;
1657 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1658 pr->pr_flags |= PR_HOST;
1659 #if defined(INET) || defined(INET6)
1661 if (!(pr_flags & PR_VNET))
1665 if (!(ch_flags & PR_IP4_USER))
1666 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1667 else if (!(pr_flags & PR_IP4_USER)) {
1668 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1669 prison_ip_dup(ppr, pr, PR_INET);
1673 if (!(ch_flags & PR_IP6_USER))
1674 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1675 else if (!(pr_flags & PR_IP6_USER)) {
1676 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1677 prison_ip_dup(ppr, pr, PR_INET6);
1682 /* Source address selection is always on by default. */
1683 pr->pr_flags |= _PR_IP_SADDRSEL;
1685 pr->pr_securelevel = ppr->pr_securelevel;
1686 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1687 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1688 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1690 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1691 if (osrelstr == NULL)
1692 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1693 sizeof(pr->pr_osrelease));
1695 strlcpy(pr->pr_osrelease, osrelstr,
1696 sizeof(pr->pr_osrelease));
1699 /* Allocate a new vnet if specified. */
1700 pr->pr_vnet = (pr_flags & PR_VNET)
1701 ? vnet_alloc() : ppr->pr_vnet;
1704 * Allocate a dedicated cpuset for each jail.
1705 * Unlike other initial settings, this may return an error.
1707 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1711 mtx_lock(&pr->pr_mtx);
1712 drflags |= PD_LOCKED;
1715 * Grab a reference for existing prisons, to ensure they
1716 * continue to exist for the duration of the call.
1719 drflags |= PD_DEREF;
1720 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1721 if ((pr->pr_flags & PR_VNET) &&
1722 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1725 "vnet jails cannot have IP address restrictions");
1730 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1733 "ip4 cannot be changed after creation");
1738 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1741 "ip6 cannot be changed after creation");
1747 /* Do final error checking before setting anything. */
1749 if (slevel < ppr->pr_securelevel) {
1755 if (childmax >= ppr->pr_childmax) {
1761 if (enforce < ppr->pr_enforce_statfs) {
1768 * devfs_rsnum is a uint16_t
1770 if (rsnum < 0 || rsnum > 65535) {
1775 * Nested jails always inherit parent's devfs ruleset
1777 if (jailed(td->td_ucred)) {
1778 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1782 rsnum = ppr->pr_devfs_rsnum;
1787 if ((ppr->pr_flags & PR_IP4) &&
1788 !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1793 if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1795 vfs_opterror(opts, "IPv4 addresses clash");
1802 if ((ppr->pr_flags & PR_IP6) &&
1803 !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1808 if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1810 vfs_opterror(opts, "IPv6 addresses clash");
1815 onamelen = namelen = 0;
1816 if (namelc != NULL) {
1817 /* Give a default name of the jid. Also allow the name to be
1818 * explicitly the jid - but not any other number, and only in
1819 * normal form (no leading zero/etc).
1821 if (namelc[0] == '\0')
1822 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1823 else if ((strtoul(namelc, &p, 10) != jid ||
1824 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1827 "name cannot be numeric (unless it is the jid)");
1831 * Make sure the name isn't too long for the prison or its
1834 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1835 onamelen = strlen(pr->pr_name + pnamelen);
1836 namelen = strlen(namelc);
1837 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1838 error = ENAMETOOLONG;
1841 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1842 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1843 sizeof(pr->pr_name)) {
1844 error = ENAMETOOLONG;
1849 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1850 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1856 * Let modules check their parameters. This requires unlocking and
1857 * then re-locking the prison, but this is still a valid state as long
1858 * as allprison_lock remains xlocked.
1860 mtx_unlock(&pr->pr_mtx);
1861 drflags &= ~PD_LOCKED;
1862 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1865 mtx_lock(&pr->pr_mtx);
1866 drflags |= PD_LOCKED;
1868 /* At this point, all valid parameters should have been noted. */
1869 TAILQ_FOREACH(opt, opts, link) {
1870 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1872 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1877 /* Set the parameters of the prison. */
1880 if (pr_flags & PR_IP4_USER) {
1881 pr->pr_flags |= PR_IP4;
1882 prison_ip_set(pr, PR_INET, ip4);
1884 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1886 if (tpr->pr_flags & PR_VNET) {
1891 if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
1900 if (pr_flags & PR_IP6_USER) {
1901 pr->pr_flags |= PR_IP6;
1902 prison_ip_set(pr, PR_INET6, ip6);
1904 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1906 if (tpr->pr_flags & PR_VNET) {
1911 if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
1919 pr->pr_securelevel = slevel;
1920 /* Set all child jails to be at least this level. */
1921 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1922 if (tpr->pr_securelevel < slevel)
1923 tpr->pr_securelevel = slevel;
1926 pr->pr_childmax = childmax;
1927 /* Set all child jails to under this limit. */
1928 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1929 if (tpr->pr_childmax > childmax - level)
1930 tpr->pr_childmax = childmax > level
1931 ? childmax - level : 0;
1934 pr->pr_enforce_statfs = enforce;
1935 /* Pass this restriction on to the children. */
1936 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1937 if (tpr->pr_enforce_statfs < enforce)
1938 tpr->pr_enforce_statfs = enforce;
1941 pr->pr_devfs_rsnum = rsnum;
1942 /* Pass this restriction on to the children. */
1943 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1944 tpr->pr_devfs_rsnum = rsnum;
1946 if (namelc != NULL) {
1947 if (ppr == &prison0)
1948 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1950 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1951 ppr->pr_name, namelc);
1952 /* Change this component of child names. */
1953 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1954 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1955 strlen(tpr->pr_name + onamelen) + 1);
1956 bcopy(pr->pr_name, tpr->pr_name, namelen);
1960 /* Try to keep a real-rooted full pathname. */
1961 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1965 if (PR_HOST & ch_flags & ~pr_flags) {
1966 if (pr->pr_flags & PR_HOST) {
1968 * Copy the parent's host info. As with pr_ip4 above,
1969 * the lack of a lock on the parent is not a problem;
1970 * it is always set with allprison_lock at least
1971 * shared, and is held exclusively here.
1973 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1974 sizeof(pr->pr_hostname));
1975 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1976 sizeof(pr->pr_domainname));
1977 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1978 sizeof(pr->pr_hostuuid));
1979 pr->pr_hostid = pr->pr_parent->pr_hostid;
1981 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1982 /* Set this prison, and any descendants without PR_HOST. */
1984 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1986 strlcpy(pr->pr_domainname, domain,
1987 sizeof(pr->pr_domainname));
1989 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1991 pr->pr_hostid = hid;
1992 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1993 if (tpr->pr_flags & PR_HOST)
1997 strlcpy(tpr->pr_hostname,
1999 sizeof(tpr->pr_hostname));
2001 strlcpy(tpr->pr_domainname,
2003 sizeof(tpr->pr_domainname));
2005 strlcpy(tpr->pr_hostuuid,
2007 sizeof(tpr->pr_hostuuid));
2009 tpr->pr_hostid = hid;
2013 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2014 if ((tallow = ch_allow & ~pr_allow))
2015 prison_set_allow_locked(pr, tallow, 0);
2017 * Persistent prisons get an extra reference, and prisons losing their
2018 * persist flag lose that reference.
2020 born = !prison_isalive(pr);
2021 if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2022 if (pr_flags & PR_PERSIST) {
2025 * This may make a dead prison alive again, but wait
2026 * to label it as such until after OSD calls have had
2027 * a chance to run (and perhaps to fail).
2029 refcount_acquire(&pr->pr_uref);
2031 drflags |= PD_DEUREF;
2032 prison_free_not_last(pr);
2035 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2036 mtx_unlock(&pr->pr_mtx);
2037 drflags &= ~PD_LOCKED;
2039 * Any errors past this point will need to de-persist newly created
2040 * prisons, as well as call remove methods.
2046 if (racct_enable && created)
2047 prison_racct_attach(pr);
2050 /* Locks may have prevented a complete restriction of child IP
2051 * addresses. If so, allocate some more memory and try again.
2055 ip4s = pr->pr_addrs[PR_INET]->ips;
2057 ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2058 mtx_lock(&pr->pr_mtx);
2060 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2062 if (tpr->pr_flags & PR_VNET) {
2067 if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2070 mtx_unlock(&pr->pr_mtx);
2075 ip6s = pr->pr_addrs[PR_INET6]->ips;
2077 ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2078 mtx_lock(&pr->pr_mtx);
2080 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2082 if (tpr->pr_flags & PR_VNET) {
2087 if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2090 mtx_unlock(&pr->pr_mtx);
2094 /* Let the modules do their work. */
2096 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2100 error = osd_jail_call(pr, PR_METHOD_SET, opts);
2105 * A new prison is now ready to be seen; either it has gained a user
2106 * reference via persistence, or is about to gain one via attachment.
2109 drflags = prison_lock_xlock(pr, drflags);
2110 pr->pr_state = PRISON_STATE_ALIVE;
2113 /* Attach this process to the prison if requested. */
2114 if (flags & JAIL_ATTACH) {
2115 error = do_jail_attach(td, pr,
2116 prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2117 drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2119 vfs_opterror(opts, "attach failed");
2125 if (racct_enable && !created) {
2126 if (drflags & PD_LOCKED) {
2127 mtx_unlock(&pr->pr_mtx);
2128 drflags &= ~PD_LOCKED;
2130 if (drflags & PD_LIST_XLOCKED) {
2131 sx_xunlock(&allprison_lock);
2132 drflags &= ~PD_LIST_XLOCKED;
2134 prison_racct_modify(pr);
2138 if (born && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2139 (pr->pr_root->v_vflag & VV_ROOT) == 0)
2140 printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2141 " file system\n", pr->pr_id);
2143 drflags &= ~PD_KILL;
2144 td->td_retval[0] = pr->pr_id;
2147 /* Release any temporary prison holds and/or locks. */
2149 prison_deref(pr, drflags);
2150 else if (drflags & PD_LIST_SLOCKED)
2151 sx_sunlock(&allprison_lock);
2152 else if (drflags & PD_LIST_XLOCKED)
2153 sx_xunlock(&allprison_lock);
2158 /* Write the error message back to userspace. */
2159 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2160 &errmsg_len) == 0 && errmsg_len > 0) {
2161 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2162 if (optuio->uio_segflg == UIO_SYSSPACE)
2164 optuio->uio_iov[errmsg_pos].iov_base,
2168 optuio->uio_iov[errmsg_pos].iov_base,
2174 prison_ip_free(ip4);
2177 prison_ip_free(ip6);
2180 free(g_path, M_TEMP);
2186 * Find the next available prison ID. Return the ID on success, or zero
2187 * on failure. Also set a pointer to the allprison list entry the prison
2188 * should be inserted before.
2191 get_next_prid(struct prison **insprp)
2193 struct prison *inspr;
2196 jid = lastprid % JAIL_MAX + 1;
2197 if (TAILQ_EMPTY(&allprison) ||
2198 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2200 * A common case is for all jails to be implicitly numbered,
2201 * which means they'll go on the end of the list, at least
2202 * for the first JAIL_MAX times.
2207 * Take two passes through the allprison list: first starting
2208 * with the proposed jid, then ending with it.
2210 for (maxid = JAIL_MAX; maxid != 0; ) {
2211 TAILQ_FOREACH(inspr, &allprison, pr_list) {
2212 if (inspr->pr_id < jid)
2214 if (inspr->pr_id > jid) {
2215 /* Found an opening. */
2219 if (++jid > maxid) {
2220 if (lastprid == maxid || lastprid == 0)
2223 * The entire legal range
2224 * has been traversed
2228 /* Try again from the start. */
2234 if (inspr == NULL) {
2235 /* Found room at the end of the list. */
2246 * struct jail_get_args {
2247 * struct iovec *iovp;
2248 * unsigned int iovcnt;
2253 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2258 /* Check that we have an even number of iovecs. */
2259 if (uap->iovcnt & 1)
2262 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2265 error = kern_jail_get(td, auio, uap->flags);
2267 error = copyout(auio->uio_iov, uap->iovp,
2268 uap->iovcnt * sizeof (struct iovec));
2274 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2276 struct bool_flags *bf;
2277 struct jailsys_flags *jsf;
2278 struct prison *pr, *mypr;
2280 struct vfsoptlist *opts;
2281 char *errmsg, *name;
2282 int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2285 if (flags & ~JAIL_GET_MASK)
2288 /* Get the parameter list. */
2289 error = vfs_buildopts(optuio, &opts);
2292 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2293 mypr = td->td_ucred->cr_prison;
2297 * Find the prison specified by one of: lastjid, jid, name.
2299 sx_slock(&allprison_lock);
2300 drflags = PD_LIST_SLOCKED;
2301 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2303 TAILQ_FOREACH(pr, &allprison, pr_list) {
2304 if (pr->pr_id > jid &&
2305 ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2306 prison_ischild(mypr, pr)) {
2307 mtx_lock(&pr->pr_mtx);
2308 drflags |= PD_LOCKED;
2313 vfs_opterror(opts, "no jail after %d", jid);
2315 } else if (error != ENOENT)
2318 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2321 pr = prison_find_child(mypr, jid);
2323 drflags |= PD_LOCKED;
2324 if (!(prison_isalive(pr) ||
2325 (flags & JAIL_DYING))) {
2327 vfs_opterror(opts, "jail %d is dying",
2334 vfs_opterror(opts, "jail %d not found", jid);
2337 } else if (error != ENOENT)
2340 error = vfs_getopt(opts, "name", (void **)&name, &len);
2342 if (len == 0 || name[len - 1] != '\0') {
2346 pr = prison_find_name(mypr, name);
2348 drflags |= PD_LOCKED;
2349 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2351 vfs_opterror(opts, "jail \"%s\" is dying",
2358 vfs_opterror(opts, "jail \"%s\" not found", name);
2360 } else if (error != ENOENT)
2363 vfs_opterror(opts, "no jail specified");
2368 /* Get the parameters of the prison. */
2370 drflags |= PD_DEREF;
2371 td->td_retval[0] = pr->pr_id;
2372 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2373 if (error != 0 && error != ENOENT)
2375 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2376 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2377 if (error != 0 && error != ENOENT)
2379 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2380 if (error != 0 && error != ENOENT)
2382 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2383 sizeof(pr->pr_cpuset->cs_id));
2384 if (error != 0 && error != ENOENT)
2386 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2387 if (error != 0 && error != ENOENT)
2390 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2391 pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2392 pr_families[PR_INET].size : 0 );
2393 if (error != 0 && error != ENOENT)
2397 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2398 pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2399 pr_families[PR_INET6].size : 0 );
2400 if (error != 0 && error != ENOENT)
2403 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2404 sizeof(pr->pr_securelevel));
2405 if (error != 0 && error != ENOENT)
2407 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2408 sizeof(pr->pr_childcount));
2409 if (error != 0 && error != ENOENT)
2411 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2412 sizeof(pr->pr_childmax));
2413 if (error != 0 && error != ENOENT)
2415 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2416 if (error != 0 && error != ENOENT)
2418 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2419 if (error != 0 && error != ENOENT)
2421 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2422 if (error != 0 && error != ENOENT)
2424 #ifdef COMPAT_FREEBSD32
2425 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2426 uint32_t hid32 = pr->pr_hostid;
2428 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2431 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2432 sizeof(pr->pr_hostid));
2433 if (error != 0 && error != ENOENT)
2435 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2436 sizeof(pr->pr_enforce_statfs));
2437 if (error != 0 && error != ENOENT)
2439 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2440 sizeof(pr->pr_devfs_rsnum));
2441 if (error != 0 && error != ENOENT)
2443 for (bf = pr_flag_bool;
2444 bf < pr_flag_bool + nitems(pr_flag_bool);
2446 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2447 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2448 if (error != 0 && error != ENOENT)
2451 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2452 if (error != 0 && error != ENOENT)
2455 for (jsf = pr_flag_jailsys;
2456 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2458 f = pr->pr_flags & (jsf->disable | jsf->new);
2459 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2460 : (f == jsf->new) ? JAIL_SYS_NEW
2462 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2463 if (error != 0 && error != ENOENT)
2466 for (bf = pr_flag_allow;
2467 bf < pr_flag_allow + nitems(pr_flag_allow) &&
2468 atomic_load_int(&bf->flag) != 0;
2470 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2471 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2472 if (error != 0 && error != ENOENT)
2475 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2476 if (error != 0 && error != ENOENT)
2479 i = !prison_isalive(pr);
2480 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2481 if (error != 0 && error != ENOENT)
2484 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2485 if (error != 0 && error != ENOENT)
2487 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2488 sizeof(pr->pr_osreldate));
2489 if (error != 0 && error != ENOENT)
2491 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2492 if (error != 0 && error != ENOENT)
2495 /* Get the module parameters. */
2496 mtx_unlock(&pr->pr_mtx);
2497 drflags &= ~PD_LOCKED;
2498 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2501 prison_deref(pr, drflags);
2505 /* By now, all parameters should have been noted. */
2506 TAILQ_FOREACH(opt, opts, link) {
2507 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2509 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2514 /* Write the fetched parameters back to userspace. */
2516 TAILQ_FOREACH(opt, opts, link) {
2517 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2518 pos = 2 * opt->pos + 1;
2519 optuio->uio_iov[pos].iov_len = opt->len;
2520 if (opt->value != NULL) {
2521 if (optuio->uio_segflg == UIO_SYSSPACE) {
2523 optuio->uio_iov[pos].iov_base,
2526 error = copyout(opt->value,
2527 optuio->uio_iov[pos].iov_base,
2537 /* Release any temporary prison holds and/or locks. */
2539 prison_deref(pr, drflags);
2540 else if (drflags & PD_LIST_SLOCKED)
2541 sx_sunlock(&allprison_lock);
2542 if (error && errmsg_pos >= 0) {
2543 /* Write the error message back to userspace. */
2544 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2545 errmsg_pos = 2 * errmsg_pos + 1;
2546 if (errmsg_len > 0) {
2547 if (optuio->uio_segflg == UIO_SYSSPACE)
2549 optuio->uio_iov[errmsg_pos].iov_base,
2553 optuio->uio_iov[errmsg_pos].iov_base,
2562 * struct jail_remove_args {
2567 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2572 error = priv_check(td, PRIV_JAIL_REMOVE);
2576 sx_xlock(&allprison_lock);
2577 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2579 sx_xunlock(&allprison_lock);
2582 if (!prison_isalive(pr)) {
2583 /* Silently ignore already-dying prisons. */
2584 mtx_unlock(&pr->pr_mtx);
2585 sx_xunlock(&allprison_lock);
2588 prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2593 * struct jail_attach_args {
2598 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2603 error = priv_check(td, PRIV_JAIL_ATTACH);
2607 sx_slock(&allprison_lock);
2608 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2610 sx_sunlock(&allprison_lock);
2614 /* Do not allow a process to attach to a prison that is not alive. */
2615 if (!prison_isalive(pr)) {
2616 mtx_unlock(&pr->pr_mtx);
2617 sx_sunlock(&allprison_lock);
2621 return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2625 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2628 struct ucred *newcred, *oldcred;
2631 mtx_assert(&pr->pr_mtx, MA_OWNED);
2632 sx_assert(&allprison_lock, SX_LOCKED);
2633 drflags &= PD_LOCK_FLAGS;
2635 * XXX: Note that there is a slight race here if two threads
2636 * in the same privileged process attempt to attach to two
2637 * different jails at the same time. It is important for
2638 * user processes not to do this, or they might end up with
2639 * a process root from one prison, but attached to the jail
2643 refcount_acquire(&pr->pr_uref);
2644 drflags |= PD_DEREF | PD_DEUREF;
2645 mtx_unlock(&pr->pr_mtx);
2646 drflags &= ~PD_LOCKED;
2648 /* Let modules do whatever they need to prepare for attaching. */
2649 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2651 prison_deref(pr, drflags);
2654 sx_unlock(&allprison_lock);
2655 drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2658 * Reparent the newly attached process to this jail.
2661 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2665 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2666 if ((error = change_dir(pr->pr_root, td)) != 0)
2669 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2672 VOP_UNLOCK(pr->pr_root);
2673 if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2678 oldcred = crcopysafe(p, newcred);
2679 newcred->cr_prison = pr;
2680 proc_set_cred(p, newcred);
2683 racct_proc_ucred_changed(p, oldcred, newcred);
2688 rctl_proc_ucred_changed(p, newcred);
2691 prison_proc_relink(oldcred->cr_prison, pr, p);
2692 prison_deref(oldcred->cr_prison, drflags);
2696 * If the prison was killed while changing credentials, die along
2699 if (!prison_isalive(pr)) {
2701 kern_psignal(p, SIGKILL);
2708 VOP_UNLOCK(pr->pr_root);
2710 /* Tell modules this thread is still in its old jail after all. */
2711 sx_slock(&allprison_lock);
2712 drflags |= PD_LIST_SLOCKED;
2713 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2714 prison_deref(pr, drflags);
2719 * Returns a locked prison instance, or NULL on failure.
2722 prison_find(int prid)
2726 sx_assert(&allprison_lock, SX_LOCKED);
2727 TAILQ_FOREACH(pr, &allprison, pr_list) {
2728 if (pr->pr_id < prid)
2730 if (pr->pr_id > prid)
2732 KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2733 mtx_lock(&pr->pr_mtx);
2740 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2743 prison_find_child(struct prison *mypr, int prid)
2748 sx_assert(&allprison_lock, SX_LOCKED);
2749 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2750 if (pr->pr_id == prid) {
2751 KASSERT(prison_isvalid(pr),
2752 ("Found invalid prison %p", pr));
2753 mtx_lock(&pr->pr_mtx);
2761 * Look for the name relative to mypr. Returns a locked prison or NULL.
2764 prison_find_name(struct prison *mypr, const char *name)
2766 struct prison *pr, *deadpr;
2770 sx_assert(&allprison_lock, SX_LOCKED);
2771 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2773 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2774 if (!strcmp(pr->pr_name + mylen, name)) {
2775 KASSERT(prison_isvalid(pr),
2776 ("Found invalid prison %p", pr));
2777 if (prison_isalive(pr)) {
2778 mtx_lock(&pr->pr_mtx);
2784 /* There was no valid prison - perhaps there was a dying one. */
2786 mtx_lock(&deadpr->pr_mtx);
2791 * See if a prison has the specific flag set. The prison should be locked,
2792 * unless checking for flags that are only set at jail creation (such as
2793 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2794 * to any other prison data.
2797 prison_flag(struct ucred *cred, unsigned flag)
2800 return ((cred->cr_prison->pr_flags & flag) != 0);
2804 * See if a prison has the specific allow flag set.
2805 * The prison *should* be locked, or only a single bit is examined, without
2806 * regard to any other prison data.
2809 prison_allow(struct ucred *cred, unsigned flag)
2812 return ((cred->cr_prison->pr_allow & flag) != 0);
2816 * Hold a prison reference, by incrementing pr_ref. It is generally
2817 * an error to hold a prison that does not already have a reference.
2818 * A prison record will remain valid as long as it has at least one
2819 * reference, and will not be removed as long as either the prison
2820 * mutex or the allprison lock is held (allprison_lock may be shared).
2823 prison_hold_locked(struct prison *pr)
2826 /* Locking is no longer required. */
2831 prison_hold(struct prison *pr)
2834 int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2837 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2839 refcount_acquire(&pr->pr_ref);
2844 * Remove a prison reference. If that was the last reference, the
2845 * prison will be removed (at a later time).
2848 prison_free_locked(struct prison *pr)
2851 mtx_assert(&pr->pr_mtx, MA_OWNED);
2853 * Locking is no longer required, but unlock because the caller
2856 mtx_unlock(&pr->pr_mtx);
2861 prison_free(struct prison *pr)
2864 KASSERT(refcount_load(&pr->pr_ref) > 0,
2865 ("Trying to free dead prison %p (jid=%d).",
2867 if (!refcount_release_if_not_last(&pr->pr_ref)) {
2869 * Don't remove the last reference in this context,
2870 * in case there are locks held.
2872 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2877 prison_free_not_last(struct prison *pr)
2882 KASSERT(refcount_load(&pr->pr_ref) > 0,
2883 ("Trying to free dead prison %p (jid=%d).",
2885 lastref = refcount_release(&pr->pr_ref);
2887 ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2890 refcount_release(&pr->pr_ref);
2895 * Hold a prison for user visibility, by incrementing pr_uref.
2896 * It is generally an error to hold a prison that isn't already
2897 * user-visible, except through the jail system calls. It is also
2898 * an error to hold an invalid prison. A prison record will remain
2899 * alive as long as it has at least one user reference, and will not
2900 * be set to the dying state until the prison mutex and allprison_lock
2904 prison_proc_hold(struct prison *pr)
2907 int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2910 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2912 refcount_acquire(&pr->pr_uref);
2917 * Remove a prison user reference. If it was the last reference, the
2918 * prison will be considered "dying", and may be removed once all of
2919 * its references are dropped.
2922 prison_proc_free(struct prison *pr)
2926 * Locking is only required when releasing the last reference.
2927 * This allows assurance that a locked prison will remain alive
2928 * until it is unlocked.
2930 KASSERT(refcount_load(&pr->pr_uref) > 0,
2931 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2932 if (!refcount_release_if_not_last(&pr->pr_uref)) {
2934 * Don't remove the last user reference in this context,
2935 * which is expected to be a process that is not only locked,
2936 * but also half dead. Add a reference so any calls to
2937 * prison_free() won't re-submit the task.
2940 mtx_lock(&pr->pr_mtx);
2941 KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2942 ("Redundant last reference in prison_proc_free (jid=%d)",
2944 pr->pr_flags |= PR_COMPLETE_PROC;
2945 mtx_unlock(&pr->pr_mtx);
2946 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2951 prison_proc_free_not_last(struct prison *pr)
2956 KASSERT(refcount_load(&pr->pr_uref) > 0,
2957 ("Trying to free dead prison %p (jid=%d).",
2959 lastref = refcount_release(&pr->pr_uref);
2961 ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2964 refcount_release(&pr->pr_uref);
2969 prison_proc_link(struct prison *pr, struct proc *p)
2972 sx_assert(&allproc_lock, SA_XLOCKED);
2973 LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
2977 prison_proc_unlink(struct prison *pr, struct proc *p)
2980 sx_assert(&allproc_lock, SA_XLOCKED);
2981 LIST_REMOVE(p, p_jaillist);
2985 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
2988 sx_xlock(&allproc_lock);
2989 prison_proc_unlink(opr, p);
2990 prison_proc_link(npr, p);
2991 sx_xunlock(&allproc_lock);
2995 * Complete a call to either prison_free or prison_proc_free.
2998 prison_complete(void *context, int pending)
3000 struct prison *pr = context;
3004 * This could be called to release the last reference, or the last
3005 * user reference (plus the reference held in prison_proc_free).
3007 drflags = prison_lock_xlock(pr, PD_DEREF);
3008 if (pr->pr_flags & PR_COMPLETE_PROC) {
3009 pr->pr_flags &= ~PR_COMPLETE_PROC;
3010 drflags |= PD_DEUREF;
3012 prison_deref(pr, drflags);
3016 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3019 kern_psignal(p, SIGKILL);
3023 * Note the iteration does not guarantee acting on all processes.
3024 * Most notably there may be fork or jail_attach in progress.
3027 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3033 if (atomic_load_int(&pr->pr_childcount) == 0) {
3034 sx_slock(&allproc_lock);
3035 LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3036 if (p->p_state == PRS_NEW)
3042 sx_sunlock(&allproc_lock);
3043 if (atomic_load_int(&pr->pr_childcount) == 0)
3046 * Some jails popped up during the iteration, fall through to a
3047 * system-wide search.
3051 sx_slock(&allproc_lock);
3052 FOREACH_PROC_IN_SYSTEM(p) {
3054 if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3055 for (ppr = p->p_ucred->cr_prison;
3057 ppr = ppr->pr_parent) {
3066 sx_sunlock(&allproc_lock);
3070 * Remove a prison reference and/or user reference (usually).
3071 * This assumes context that allows sleeping (for allprison_lock),
3072 * with no non-sleeping locks held, except perhaps the prison itself.
3073 * If there are no more references, release and delist the prison.
3074 * On completion, the prison lock and the allprison lock are both
3078 prison_deref(struct prison *pr, int flags)
3080 struct prisonlist freeprison;
3081 struct prison *killpr, *rpr, *ppr, *tpr;
3084 TAILQ_INIT(&freeprison);
3086 * Release this prison as requested, which may cause its parent
3087 * to be released, and then maybe its grandparent, etc.
3090 if (flags & PD_KILL) {
3091 /* Kill the prison and its descendents. */
3092 KASSERT(pr != &prison0,
3093 ("prison_deref trying to kill prison0"));
3094 if (!(flags & PD_DEREF)) {
3098 flags = prison_lock_xlock(pr, flags);
3099 prison_deref_kill(pr, &freeprison);
3101 if (flags & PD_DEUREF) {
3102 /* Drop a user reference. */
3103 KASSERT(refcount_load(&pr->pr_uref) > 0,
3104 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3106 if (!refcount_release_if_not_last(&pr->pr_uref)) {
3107 if (!(flags & PD_DEREF)) {
3111 flags = prison_lock_xlock(pr, flags);
3112 if (refcount_release(&pr->pr_uref) &&
3113 pr->pr_state == PRISON_STATE_ALIVE) {
3115 * When the last user references goes,
3116 * this becomes a dying prison.
3119 refcount_load(&prison0.pr_uref) > 0,
3120 ("prison0 pr_uref=0"));
3121 pr->pr_state = PRISON_STATE_DYING;
3122 mtx_unlock(&pr->pr_mtx);
3123 flags &= ~PD_LOCKED;
3128 if (flags & PD_KILL) {
3130 * Any remaining user references are probably processes
3131 * that need to be killed, either in this prison or its
3134 if (refcount_load(&pr->pr_uref) > 0)
3136 /* Make sure the parent prison doesn't get killed. */
3139 if (flags & PD_DEREF) {
3140 /* Drop a reference. */
3141 KASSERT(refcount_load(&pr->pr_ref) > 0,
3142 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3144 if (!refcount_release_if_not_last(&pr->pr_ref)) {
3145 flags = prison_lock_xlock(pr, flags);
3146 if (refcount_release(&pr->pr_ref)) {
3148 * When the last reference goes,
3149 * unlink the prison and set it aside.
3152 refcount_load(&pr->pr_uref) == 0,
3153 ("prison_deref: last ref, "
3154 "but still has %d urefs (jid=%d)",
3155 pr->pr_uref, pr->pr_id));
3157 refcount_load(&prison0.pr_ref) != 0,
3158 ("prison0 pr_ref=0"));
3159 pr->pr_state = PRISON_STATE_INVALID;
3160 TAILQ_REMOVE(&allprison, pr, pr_list);
3161 LIST_REMOVE(pr, pr_sibling);
3162 TAILQ_INSERT_TAIL(&freeprison, pr,
3164 for (ppr = pr->pr_parent;
3166 ppr = ppr->pr_parent)
3167 ppr->pr_childcount--;
3169 * Removing a prison frees references
3172 mtx_unlock(&pr->pr_mtx);
3173 flags &= ~PD_LOCKED;
3175 flags |= PD_DEREF | PD_DEUREF;
3183 /* Release all the prison locks. */
3184 if (flags & PD_LOCKED)
3185 mtx_unlock(&pr->pr_mtx);
3186 if (flags & PD_LIST_SLOCKED)
3187 sx_sunlock(&allprison_lock);
3188 else if (flags & PD_LIST_XLOCKED)
3189 sx_xunlock(&allprison_lock);
3191 /* Kill any processes attached to a killed prison. */
3193 prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3196 * Finish removing any unreferenced prisons, which couldn't happen
3197 * while allprison_lock was held (to avoid a LOR on vrele).
3199 TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3201 if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3202 vnet_destroy(rpr->pr_vnet);
3204 if (rpr->pr_root != NULL)
3205 vrele(rpr->pr_root);
3206 mtx_destroy(&rpr->pr_mtx);
3208 prison_ip_free(rpr->pr_addrs[PR_INET]);
3211 prison_ip_free(rpr->pr_addrs[PR_INET6]);
3213 if (rpr->pr_cpuset != NULL)
3214 cpuset_rel(rpr->pr_cpuset);
3218 prison_racct_detach(rpr);
3220 TAILQ_REMOVE(&freeprison, rpr, pr_list);
3221 free(rpr, M_PRISON);
3226 * Kill the prison and its descendants. Mark them as dying, clear the
3227 * persist flag, and call module remove methods.
3230 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3232 struct prison *cpr, *ppr, *rpr;
3236 * Unlike the descendants, the target prison can be killed
3237 * even if it is currently dying. This is useful for failed
3238 * creation in jail_set(2).
3240 KASSERT(refcount_load(&pr->pr_ref) > 0,
3241 ("Trying to kill dead prison %p (jid=%d).",
3243 refcount_acquire(&pr->pr_uref);
3244 pr->pr_state = PRISON_STATE_DYING;
3245 mtx_unlock(&pr->pr_mtx);
3248 FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3250 if (!prison_isalive(cpr)) {
3255 prison_proc_hold(cpr);
3256 mtx_lock(&cpr->pr_mtx);
3257 cpr->pr_state = PRISON_STATE_DYING;
3258 cpr->pr_flags |= PR_REMOVE;
3259 mtx_unlock(&cpr->pr_mtx);
3262 if (!(cpr->pr_flags & PR_REMOVE))
3264 prison_cleanup(cpr);
3265 mtx_lock(&cpr->pr_mtx);
3266 cpr->pr_flags &= ~PR_REMOVE;
3267 if (cpr->pr_flags & PR_PERSIST) {
3268 cpr->pr_flags &= ~PR_PERSIST;
3269 prison_proc_free_not_last(cpr);
3270 prison_free_not_last(cpr);
3272 (void)refcount_release(&cpr->pr_uref);
3273 if (refcount_release(&cpr->pr_ref)) {
3275 * When the last reference goes, unlink the prison
3276 * and set it aside for prison_deref() to handle.
3277 * Delay unlinking the sibling list to keep the loop
3281 LIST_REMOVE(rpr, pr_sibling);
3283 rpr->pr_state = PRISON_STATE_INVALID;
3284 TAILQ_REMOVE(&allprison, rpr, pr_list);
3285 TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3287 * Removing a prison frees references from its parent.
3289 ppr = rpr->pr_parent;
3290 prison_proc_free_not_last(ppr);
3291 prison_free_not_last(ppr);
3292 for (; ppr != NULL; ppr = ppr->pr_parent)
3293 ppr->pr_childcount--;
3295 mtx_unlock(&cpr->pr_mtx);
3298 LIST_REMOVE(rpr, pr_sibling);
3301 mtx_lock(&pr->pr_mtx);
3302 if (pr->pr_flags & PR_PERSIST) {
3303 pr->pr_flags &= ~PR_PERSIST;
3304 prison_proc_free_not_last(pr);
3305 prison_free_not_last(pr);
3307 (void)refcount_release(&pr->pr_uref);
3311 * Given the current locking state in the flags, make sure allprison_lock
3312 * is held exclusive, and the prison is locked. Return flags indicating
3316 prison_lock_xlock(struct prison *pr, int flags)
3319 if (!(flags & PD_LIST_XLOCKED)) {
3321 * Get allprison_lock, which may be an upgrade,
3322 * and may require unlocking the prison.
3324 if (flags & PD_LOCKED) {
3325 mtx_unlock(&pr->pr_mtx);
3326 flags &= ~PD_LOCKED;
3328 if (flags & PD_LIST_SLOCKED) {
3329 if (!sx_try_upgrade(&allprison_lock)) {
3330 sx_sunlock(&allprison_lock);
3331 sx_xlock(&allprison_lock);
3333 flags &= ~PD_LIST_SLOCKED;
3335 sx_xlock(&allprison_lock);
3336 flags |= PD_LIST_XLOCKED;
3338 if (!(flags & PD_LOCKED)) {
3339 /* Lock the prison mutex. */
3340 mtx_lock(&pr->pr_mtx);
3347 * Release a prison's resources when it starts dying (when the last user
3348 * reference is dropped, or when it is killed).
3351 prison_cleanup(struct prison *pr)
3353 sx_assert(&allprison_lock, SA_XLOCKED);
3354 mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3355 vfs_exjail_delete(pr);
3356 shm_remove_prison(pr);
3357 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3361 * Set or clear a permission bit in the pr_allow field, passing restrictions
3362 * (cleared permission) down to child jails.
3365 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3369 pr = cred->cr_prison;
3370 sx_slock(&allprison_lock);
3371 mtx_lock(&pr->pr_mtx);
3372 prison_set_allow_locked(pr, flag, enable);
3373 mtx_unlock(&pr->pr_mtx);
3374 sx_sunlock(&allprison_lock);
3378 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3384 pr->pr_allow |= flag;
3386 pr->pr_allow &= ~flag;
3387 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3388 cpr->pr_allow &= ~flag;
3393 * Check if a jail supports the given address family.
3395 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3399 prison_check_af(struct ucred *cred, int af)
3404 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3406 pr = cred->cr_prison;
3408 /* Prisons with their own network stack are not limited. */
3409 if (prison_owns_vnet(cred))
3418 if (pr->pr_flags & PR_IP4)
3420 mtx_lock(&pr->pr_mtx);
3421 if ((pr->pr_flags & PR_IP4) &&
3422 pr->pr_addrs[PR_INET] == NULL)
3423 error = EAFNOSUPPORT;
3424 mtx_unlock(&pr->pr_mtx);
3430 if (pr->pr_flags & PR_IP6)
3432 mtx_lock(&pr->pr_mtx);
3433 if ((pr->pr_flags & PR_IP6) &&
3434 pr->pr_addrs[PR_INET6] == NULL)
3435 error = EAFNOSUPPORT;
3436 mtx_unlock(&pr->pr_mtx);
3445 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3446 error = EAFNOSUPPORT;
3452 * Check if given address belongs to the jail referenced by cred (wrapper to
3453 * prison_check_ip[46]).
3455 * Returns 0 if jail doesn't restrict the address family or if address belongs
3456 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3457 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3460 prison_if(struct ucred *cred, const struct sockaddr *sa)
3463 const struct sockaddr_in *sai;
3466 const struct sockaddr_in6 *sai6;
3470 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3471 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3474 if (prison_owns_vnet(cred))
3479 switch (sa->sa_family)
3483 sai = (const struct sockaddr_in *)sa;
3484 error = prison_check_ip4(cred, &sai->sin_addr);
3489 sai6 = (const struct sockaddr_in6 *)sa;
3490 error = prison_check_ip6(cred, &sai6->sin6_addr);
3494 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3495 error = EAFNOSUPPORT;
3501 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3504 prison_check(struct ucred *cred1, struct ucred *cred2)
3507 return ((cred1->cr_prison == cred2->cr_prison ||
3508 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3512 * For mountd/nfsd to run within a prison, it must be:
3514 * - PR_ALLOW_NFSD must be set on it.
3515 * - The root directory (pr_root) of the prison must be
3516 * a file system mount point, so the mountd can hang
3517 * export information on it.
3518 * - The prison's enforce_statfs cannot be 0, so that
3519 * mountd(8) can do exports.
3522 prison_check_nfsd(struct ucred *cred)
3525 if (jailed_without_vnet(cred))
3527 if (!prison_allow(cred, PR_ALLOW_NFSD))
3529 if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3531 if (cred->cr_prison->pr_enforce_statfs == 0)
3537 * Return true if p2 is a child of p1, otherwise false.
3540 prison_ischild(struct prison *pr1, struct prison *pr2)
3543 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3550 * Return true if the prison is currently alive. A prison is alive if it
3551 * holds user references and it isn't being removed.
3554 prison_isalive(const struct prison *pr)
3557 if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3563 * Return true if the prison is currently valid. A prison is valid if it has
3564 * been fully created, and is not being destroyed. Note that dying prisons
3565 * are still considered valid. Invalid prisons won't be found under normal
3566 * circumstances, as they're only put in that state by functions that have
3567 * an exclusive hold on allprison_lock.
3570 prison_isvalid(struct prison *pr)
3573 if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3575 if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3581 * Return true if the passed credential is in a jail and that jail does not
3582 * have its own virtual network stack, otherwise false.
3585 jailed_without_vnet(struct ucred *cred)
3591 if (prison_owns_vnet(cred))
3599 * Return the correct hostname (domainname, et al) for the passed credential.
3602 getcredhostname(struct ucred *cred, char *buf, size_t size)
3607 * A NULL credential can be used to shortcut to the physical
3608 * system's hostname.
3610 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3611 mtx_lock(&pr->pr_mtx);
3612 strlcpy(buf, pr->pr_hostname, size);
3613 mtx_unlock(&pr->pr_mtx);
3617 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3620 mtx_lock(&cred->cr_prison->pr_mtx);
3621 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3622 mtx_unlock(&cred->cr_prison->pr_mtx);
3626 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3629 mtx_lock(&cred->cr_prison->pr_mtx);
3630 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3631 mtx_unlock(&cred->cr_prison->pr_mtx);
3635 getcredhostid(struct ucred *cred, unsigned long *hostid)
3638 mtx_lock(&cred->cr_prison->pr_mtx);
3639 *hostid = cred->cr_prison->pr_hostid;
3640 mtx_unlock(&cred->cr_prison->pr_mtx);
3644 getjailname(struct ucred *cred, char *name, size_t len)
3647 mtx_lock(&cred->cr_prison->pr_mtx);
3648 strlcpy(name, cred->cr_prison->pr_name, len);
3649 mtx_unlock(&cred->cr_prison->pr_mtx);
3654 * Determine whether the prison represented by cred owns
3655 * its vnet rather than having it inherited.
3657 * Returns true in case the prison owns the vnet, false otherwise.
3660 prison_owns_vnet(struct ucred *cred)
3664 * vnets cannot be added/removed after jail creation,
3665 * so no need to lock here.
3667 return ((cred->cr_prison->pr_flags & PR_VNET) != 0);
3672 * Determine whether the subject represented by cred can "see"
3673 * status of a mount point.
3674 * Returns: 0 for permitted, ENOENT otherwise.
3675 * XXX: This function should be called cr_canseemount() and should be
3676 * placed in kern_prot.c.
3679 prison_canseemount(struct ucred *cred, struct mount *mp)
3685 pr = cred->cr_prison;
3686 if (pr->pr_enforce_statfs == 0)
3688 if (pr->pr_root->v_mount == mp)
3690 if (pr->pr_enforce_statfs == 2)
3693 * If jail's chroot directory is set to "/" we should be able to see
3694 * all mount-points from inside a jail.
3695 * This is ugly check, but this is the only situation when jail's
3696 * directory ends with '/'.
3698 if (strcmp(pr->pr_path, "/") == 0)
3700 len = strlen(pr->pr_path);
3702 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3705 * Be sure that we don't have situation where jail's root directory
3706 * is "/some/path" and mount point is "/some/pathpath".
3708 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3714 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3716 char jpath[MAXPATHLEN];
3720 pr = cred->cr_prison;
3721 if (pr->pr_enforce_statfs == 0)
3723 if (prison_canseemount(cred, mp) != 0) {
3724 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3725 strlcpy(sp->f_mntonname, "[restricted]",
3726 sizeof(sp->f_mntonname));
3729 if (pr->pr_root->v_mount == mp) {
3731 * Clear current buffer data, so we are sure nothing from
3732 * the valid path left there.
3734 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3735 *sp->f_mntonname = '/';
3739 * If jail's chroot directory is set to "/" we should be able to see
3740 * all mount-points from inside a jail.
3742 if (strcmp(pr->pr_path, "/") == 0)
3744 len = strlen(pr->pr_path);
3745 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3747 * Clear current buffer data, so we are sure nothing from
3748 * the valid path left there.
3750 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3751 if (*jpath == '\0') {
3752 /* Should never happen. */
3753 *sp->f_mntonname = '/';
3755 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3760 * Check with permission for a specific privilege is granted within jail. We
3761 * have a specific list of accepted privileges; the rest are denied.
3764 prison_priv_check(struct ucred *cred, int priv)
3770 * Some policies have custom handlers. This routine should not be
3771 * called for them. See priv_check_cred().
3774 case PRIV_VFS_LOOKUP:
3775 case PRIV_VFS_GENERATION:
3776 KASSERT(0, ("prison_priv_check instead of a custom handler "
3777 "called for %d\n", priv));
3785 * Privileges specific to prisons with a virtual network stack.
3786 * There might be a duplicate entry here in case the privilege
3787 * is only granted conditionally in the legacy jail case.
3791 * NFS-specific privileges.
3793 case PRIV_NFS_DAEMON:
3794 case PRIV_VFS_GETFH:
3795 case PRIV_VFS_MOUNT_EXPORTED:
3796 if (!prison_check_nfsd(cred))
3799 case PRIV_NFS_LOCKD:
3802 * Network stack privileges.
3804 case PRIV_NET_BRIDGE:
3807 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3808 case PRIV_NET_ROUTE:
3810 case PRIV_NET_SETIFMTU:
3811 case PRIV_NET_SETIFFLAGS:
3812 case PRIV_NET_SETIFCAP:
3813 case PRIV_NET_SETIFDESCR:
3814 case PRIV_NET_SETIFNAME :
3815 case PRIV_NET_SETIFMETRIC:
3816 case PRIV_NET_SETIFPHYS:
3817 case PRIV_NET_SETIFMAC:
3818 case PRIV_NET_SETLANPCP:
3819 case PRIV_NET_ADDMULTI:
3820 case PRIV_NET_DELMULTI:
3821 case PRIV_NET_HWIOCTL:
3822 case PRIV_NET_SETLLADDR:
3823 case PRIV_NET_ADDIFGROUP:
3824 case PRIV_NET_DELIFGROUP:
3825 case PRIV_NET_IFCREATE:
3826 case PRIV_NET_IFDESTROY:
3827 case PRIV_NET_ADDIFADDR:
3828 case PRIV_NET_DELIFADDR:
3831 case PRIV_NET_SETIFVNET:
3832 case PRIV_NET_SETIFFIB:
3838 * 802.11-related privileges.
3840 case PRIV_NET80211_VAP_GETKEY:
3841 case PRIV_NET80211_VAP_MANAGE:
3847 case PRIV_NETATM_CFG:
3848 case PRIV_NETATM_ADD:
3849 case PRIV_NETATM_DEL:
3850 case PRIV_NETATM_SET:
3853 * Bluetooth privileges.
3855 case PRIV_NETBLUETOOTH_RAW:
3859 * Netgraph and netgraph module privileges.
3861 case PRIV_NETGRAPH_CONTROL:
3863 case PRIV_NETGRAPH_TTY:
3867 * IPv4 and IPv6 privileges.
3869 case PRIV_NETINET_IPFW:
3870 case PRIV_NETINET_DIVERT:
3871 case PRIV_NETINET_PF:
3872 case PRIV_NETINET_DUMMYNET:
3873 case PRIV_NETINET_CARP:
3874 case PRIV_NETINET_MROUTE:
3875 case PRIV_NETINET_RAW:
3876 case PRIV_NETINET_ADDRCTRL6:
3877 case PRIV_NETINET_ND6:
3878 case PRIV_NETINET_SCOPE6:
3879 case PRIV_NETINET_ALIFETIME6:
3880 case PRIV_NETINET_IPSEC:
3881 case PRIV_NETINET_BINDANY:
3896 * No default: or deny here.
3897 * In case of no permit fall through to next switch().
3899 if (cred->cr_prison->pr_flags & PR_VNET)
3906 * Allow ktrace privileges for root in jail.
3912 * Allow jailed processes to configure audit identity and
3913 * submit audit records (login, etc). In the future we may
3914 * want to further refine the relationship between audit and
3917 case PRIV_AUDIT_GETAUDIT:
3918 case PRIV_AUDIT_SETAUDIT:
3919 case PRIV_AUDIT_SUBMIT:
3923 * Allow jailed processes to manipulate process UNIX
3924 * credentials in any way they see fit.
3926 case PRIV_CRED_SETUID:
3927 case PRIV_CRED_SETEUID:
3928 case PRIV_CRED_SETGID:
3929 case PRIV_CRED_SETEGID:
3930 case PRIV_CRED_SETGROUPS:
3931 case PRIV_CRED_SETREUID:
3932 case PRIV_CRED_SETREGID:
3933 case PRIV_CRED_SETRESUID:
3934 case PRIV_CRED_SETRESGID:
3937 * Jail implements visibility constraints already, so allow
3938 * jailed root to override uid/gid-based constraints.
3940 case PRIV_SEEOTHERGIDS:
3941 case PRIV_SEEOTHERUIDS:
3944 * Jail implements inter-process debugging limits already, so
3945 * allow jailed root various debugging privileges.
3947 case PRIV_DEBUG_DIFFCRED:
3948 case PRIV_DEBUG_SUGID:
3949 case PRIV_DEBUG_UNPRIV:
3952 * Allow jail to set various resource limits and login
3953 * properties, and for now, exceed process resource limits.
3955 case PRIV_PROC_LIMIT:
3956 case PRIV_PROC_SETLOGIN:
3957 case PRIV_PROC_SETRLIMIT:
3960 * System V and POSIX IPC privileges are granted in jail.
3963 case PRIV_IPC_WRITE:
3964 case PRIV_IPC_ADMIN:
3965 case PRIV_IPC_MSGSIZE:
3969 * Jail operations within a jail work on child jails.
3971 case PRIV_JAIL_ATTACH:
3973 case PRIV_JAIL_REMOVE:
3976 * Jail implements its own inter-process limits, so allow
3977 * root processes in jail to change scheduling on other
3978 * processes in the same jail. Likewise for signalling.
3980 case PRIV_SCHED_DIFFCRED:
3981 case PRIV_SCHED_CPUSET:
3982 case PRIV_SIGNAL_DIFFCRED:
3983 case PRIV_SIGNAL_SUGID:
3986 * Allow jailed processes to write to sysctls marked as jail
3989 case PRIV_SYSCTL_WRITEJAIL:
3992 * Allow root in jail to manage a variety of quota
3993 * properties. These should likely be conditional on a
3994 * configuration option.
3996 case PRIV_VFS_GETQUOTA:
3997 case PRIV_VFS_SETQUOTA:
4000 * Since Jail relies on chroot() to implement file system
4001 * protections, grant many VFS privileges to root in jail.
4002 * Be careful to exclude mount-related and NFS-related
4006 case PRIV_VFS_WRITE:
4007 case PRIV_VFS_ADMIN:
4009 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4010 case PRIV_VFS_CHFLAGS_DEV:
4011 case PRIV_VFS_CHOWN:
4012 case PRIV_VFS_CHROOT:
4013 case PRIV_VFS_RETAINSUGID:
4014 case PRIV_VFS_FCHROOT:
4016 case PRIV_VFS_SETGID:
4018 case PRIV_VFS_STICKYFILE:
4021 * As in the non-jail case, non-root users are expected to be
4022 * able to read kernel/physical memory (provided /dev/[k]mem
4023 * exists in the jail and they have permission to access it).
4025 case PRIV_KMEM_READ:
4029 * Depending on the global setting, allow privilege of
4030 * setting system flags.
4032 case PRIV_VFS_SYSFLAGS:
4033 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4039 * Depending on the global setting, allow privilege of
4040 * mounting/unmounting file systems.
4042 case PRIV_VFS_MOUNT:
4043 case PRIV_VFS_UNMOUNT:
4044 case PRIV_VFS_MOUNT_NONUSER:
4045 case PRIV_VFS_MOUNT_OWNER:
4046 pr = cred->cr_prison;
4048 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4056 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4057 * policy. priv_check_cred will not specifically allow it, and
4058 * we may want a MAC policy to allow it.
4060 case PRIV_VFS_READ_DIR:
4064 * Conditionally allow privileged process in the jail to
4065 * manipulate filesystem extended attributes in the system
4068 case PRIV_VFS_EXTATTR_SYSTEM:
4069 if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4075 * Conditionnaly allow locking (unlocking) physical pages
4079 case PRIV_VM_MUNLOCK:
4080 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4086 * Conditionally allow jailed root to bind reserved ports.
4088 case PRIV_NETINET_RESERVEDPORT:
4089 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4095 * Allow jailed root to reuse in-use ports.
4097 case PRIV_NETINET_REUSEPORT:
4101 * Allow jailed root to set certain IPv4/6 (option) headers.
4103 case PRIV_NETINET_SETHDROPTS:
4107 * Conditionally allow creating raw sockets in jail.
4109 case PRIV_NETINET_RAW:
4110 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4116 * Since jail implements its own visibility limits on netstat
4117 * sysctls, allow getcred. This allows identd to work in
4120 case PRIV_NETINET_GETCRED:
4124 * Allow jailed root to set loginclass.
4126 case PRIV_PROC_SETLOGINCLASS:
4130 * Do not allow a process inside a jail to read the kernel
4131 * message buffer unless explicitly permitted.
4134 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4140 * In all remaining cases, deny the privilege request. This
4141 * includes almost all network privileges, many system
4142 * configuration privileges.
4149 * Return the part of pr2's name that is relative to pr1, or the whole name
4150 * if it does not directly follow.
4154 prison_name(struct prison *pr1, struct prison *pr2)
4158 /* Jails see themselves as "0" (if they see themselves at all). */
4161 name = pr2->pr_name;
4162 if (prison_ischild(pr1, pr2)) {
4164 * pr1 isn't locked (and allprison_lock may not be either)
4165 * so its length can't be counted on. But the number of dots
4166 * can be counted on - and counted.
4168 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4169 name = strchr(name, '.') + 1;
4175 * Return the part of pr2's path that is relative to pr1, or the whole path
4176 * if it does not directly follow.
4179 prison_path(struct prison *pr1, struct prison *pr2)
4181 char *path1, *path2;
4184 path1 = pr1->pr_path;
4185 path2 = pr2->pr_path;
4186 if (!strcmp(path1, "/"))
4188 len1 = strlen(path1);
4189 if (strncmp(path1, path2, len1))
4191 if (path2[len1] == '\0')
4193 if (path2[len1] == '/')
4194 return (path2 + len1);
4199 * Jail-related sysctls.
4201 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4204 #if defined(INET) || defined(INET6)
4206 * Copy address array to memory that would be then SYSCTL_OUT-ed.
4207 * sysctl_jail_list() helper.
4210 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4212 const struct prison_ip *pip;
4213 const size_t size = pr_families[af].size;
4216 mtx_assert(&pr->pr_mtx, MA_OWNED);
4217 if ((pip = pr->pr_addrs[af]) != NULL) {
4218 if (*len < pip->ips) {
4220 mtx_unlock(&pr->pr_mtx);
4221 *out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4222 mtx_lock(&pr->pr_mtx);
4225 bcopy(pip->pr_ip, *out, pip->ips * size);
4231 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4234 struct prison *pr, *cpr;
4236 struct in_addr *ip4 = NULL;
4240 struct in6_addr *ip6 = NULL;
4245 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4246 pr = req->td->td_ucred->cr_prison;
4248 sx_slock(&allprison_lock);
4249 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4250 mtx_lock(&cpr->pr_mtx);
4252 prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4255 prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4257 bzero(xp, sizeof(*xp));
4258 xp->pr_version = XPRISON_VERSION;
4259 xp->pr_id = cpr->pr_id;
4260 xp->pr_state = cpr->pr_state;
4261 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4262 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4263 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4270 mtx_unlock(&cpr->pr_mtx);
4271 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4275 if (xp->pr_ip4s > 0) {
4276 error = SYSCTL_OUT(req, ip4,
4277 xp->pr_ip4s * sizeof(struct in_addr));
4283 if (xp->pr_ip6s > 0) {
4284 error = SYSCTL_OUT(req, ip6,
4285 xp->pr_ip6s * sizeof(struct in6_addr));
4291 sx_sunlock(&allprison_lock);
4302 SYSCTL_OID(_security_jail, OID_AUTO, list,
4303 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4304 sysctl_jail_list, "S", "List of active jails");
4307 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4311 injail = jailed(req->td->td_ucred);
4312 error = SYSCTL_OUT(req, &injail, sizeof(injail));
4317 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4318 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4319 sysctl_jail_jailed, "I", "Process in jail?");
4322 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4324 int error, havevnet;
4326 struct ucred *cred = req->td->td_ucred;
4328 havevnet = jailed(cred) && prison_owns_vnet(cred);
4332 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4337 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4338 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4339 sysctl_jail_vnet, "I", "Jail owns vnet?");
4341 #if defined(INET) || defined(INET6)
4342 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4343 &jail_max_af_ips, 0,
4344 "Number of IP addresses a jail may have at most per address family (deprecated)");
4348 * Default parameters for jail(2) compatibility. For historical reasons,
4349 * the sysctl names have varying similarity to the parameter names. Prisons
4350 * just see their own parameters, and can't change them.
4353 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4357 /* Get the current flag value, and convert it to a boolean. */
4358 if (req->td->td_ucred->cr_prison == &prison0) {
4359 mtx_lock(&prison0.pr_mtx);
4360 i = (jail_default_allow & arg2) != 0;
4361 mtx_unlock(&prison0.pr_mtx);
4363 i = prison_allow(req->td->td_ucred, arg2);
4367 error = sysctl_handle_int(oidp, &i, 0, req);
4368 if (error || !req->newptr)
4374 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4377 mtx_lock(&prison0.pr_mtx);
4378 jail_default_allow = (jail_default_allow & ~arg2) | i;
4379 mtx_unlock(&prison0.pr_mtx);
4383 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4384 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4385 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4386 "Processes in jail can set their hostnames (deprecated)");
4387 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4388 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4389 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4390 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4391 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4392 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4393 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4394 "Processes in jail can use System V IPC primitives (deprecated)");
4395 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4396 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4397 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4398 "Prison root can create raw sockets (deprecated)");
4399 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4400 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4401 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4402 "Processes in jail can alter system file flags (deprecated)");
4403 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4404 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4405 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4406 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4409 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4414 pr = req->td->td_ucred->cr_prison;
4415 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4416 error = sysctl_handle_int(oidp, &level, 0, req);
4417 if (error || !req->newptr)
4419 *(int *)arg1 = level;
4423 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4424 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4425 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4426 sysctl_jail_default_level, "I",
4427 "Processes in jail cannot see all mounted file systems (deprecated)");
4429 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4430 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4431 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4432 sysctl_jail_default_level, "I",
4433 "Ruleset for the devfs filesystem in jail (deprecated)");
4436 * Nodes to describe jail parameters. Maximum length of string parameters
4437 * is returned in the string itself, and the other parameters exist merely
4438 * to make themselves and their types known.
4440 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4444 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4451 switch (oidp->oid_kind & CTLTYPE)
4457 if (!(req->flags & SCTL_MASK32))
4459 return (SYSCTL_OUT(req, &l, sizeof(l)));
4463 return (SYSCTL_OUT(req, &i, sizeof(i)));
4464 case CTLTYPE_STRING:
4465 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4467 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4468 case CTLTYPE_STRUCT:
4470 return (SYSCTL_OUT(req, &s, sizeof(s)));
4476 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4477 * jail creation time but cannot be changed in an existing jail.
4479 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4480 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4481 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4482 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4483 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4484 "I", "Jail secure level");
4485 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4486 "Jail value for kern.osreldate and uname -K");
4487 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4488 "Jail value for kern.osrelease and uname -r");
4489 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4490 "I", "Jail cannot see all mounted file systems");
4491 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4492 "I", "Ruleset for in-jail devfs mounts");
4493 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4494 "B", "Jail persistence");
4496 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4497 "E,jailsys", "Virtual network stack");
4499 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4500 "B", "Jail is in the process of shutting down");
4502 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4503 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4504 "I", "Current number of child jails");
4505 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4506 "I", "Maximum number of child jails");
4508 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4509 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4511 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4512 "Jail NIS domainname");
4513 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4515 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4516 "LU", "Jail host ID");
4518 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4519 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4522 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4523 "Jail IPv4 address virtualization");
4524 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4525 "S,in_addr,a", "Jail IPv4 addresses");
4526 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4527 "B", "Do (not) use IPv4 source address selection rather than the "
4528 "primary jail IPv4 address.");
4531 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4532 "Jail IPv6 address virtualization");
4533 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4534 "S,in6_addr,a", "Jail IPv6 addresses");
4535 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4536 "B", "Do (not) use IPv6 source address selection rather than the "
4537 "primary jail IPv6 address.");
4540 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4541 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4542 "B", "Jail may set hostname");
4543 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4544 "B", "Jail may use SYSV IPC");
4545 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4546 "B", "Jail may create raw sockets");
4547 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4548 "B", "Jail may alter system file flags");
4549 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4550 "B", "Jail may set file quotas");
4551 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4552 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4553 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4554 "B", "Jail may lock (unlock) physical pages in memory");
4555 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4556 "B", "Jail may bind sockets to reserved ports");
4557 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4558 "B", "Jail may read the kernel message buffer");
4559 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4560 "B", "Unprivileged processes may use process debugging facilities");
4561 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4562 "B", "Processes in jail with uid 0 have privilege");
4564 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
4565 "B", "Mountd/nfsd may run in the jail");
4567 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
4568 "B", "Jail may set system-level filesystem extended attributes");
4570 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4571 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4572 "B", "Jail may mount/unmount jail-friendly file systems in general");
4575 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
4576 * its associated bit in the pr_allow bitmask, or zero if the parameter was
4580 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4583 struct bool_flags *bf;
4584 struct sysctl_oid *parent;
4585 char *allow_name, *allow_noname, *allowed;
4586 #ifndef NO_SYSCTL_DESCR
4587 char *descr_deprecated;
4592 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4594 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4596 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4597 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4598 free(allow_name, M_PRISON);
4603 * See if this parameter has already beed added, i.e. a module was
4604 * previously loaded/unloaded.
4606 mtx_lock(&prison0.pr_mtx);
4607 for (bf = pr_flag_allow;
4608 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4609 atomic_load_int(&bf->flag) != 0;
4611 if (strcmp(bf->name, allow_name) == 0) {
4612 allow_flag = bf->flag;
4618 * Find a free bit in pr_allow_all, failing if there are none
4619 * (which shouldn't happen as long as we keep track of how many
4620 * potential dynamic flags exist).
4622 for (allow_flag = 1;; allow_flag <<= 1) {
4623 if (allow_flag == 0)
4625 if ((pr_allow_all & allow_flag) == 0)
4629 /* Note the parameter in the next open slot in pr_flag_allow. */
4630 for (bf = pr_flag_allow; ; bf++) {
4631 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4632 /* This should never happen, but is not fatal. */
4636 if (atomic_load_int(&bf->flag) == 0)
4639 bf->name = allow_name;
4640 bf->noname = allow_noname;
4641 pr_allow_all |= allow_flag;
4643 * prison0 always has permission for the new parameter.
4644 * Other jails must have it granted to them.
4646 prison0.pr_allow |= allow_flag;
4647 /* The flag indicates a valid entry, so make sure it is set last. */
4648 atomic_store_rel_int(&bf->flag, allow_flag);
4649 mtx_unlock(&prison0.pr_mtx);
4652 * Create sysctls for the parameter, and the back-compat global
4656 ? SYSCTL_ADD_NODE(NULL,
4657 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4658 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4659 : &sysctl___security_jail_param_allow;
4660 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4661 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4662 NULL, 0, sysctl_jail_param, "B", descr);
4664 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4665 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4666 #ifndef NO_SYSCTL_DESCR
4667 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4670 (void)SYSCTL_ADD_PROC(NULL,
4671 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4672 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4673 sysctl_jail_default_allow, "I", descr_deprecated);
4674 #ifndef NO_SYSCTL_DESCR
4675 free(descr_deprecated, M_TEMP);
4677 free(allowed, M_TEMP);
4682 mtx_unlock(&prison0.pr_mtx);
4683 free(allow_name, M_PRISON);
4684 free(allow_noname, M_PRISON);
4689 * The VFS system will register jail-aware filesystems here. They each get
4690 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4691 * attempts to mount.
4694 prison_add_vfs(struct vfsconf *vfsp)
4696 #ifdef NO_SYSCTL_DESCR
4698 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4703 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4705 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4707 free(descr, M_TEMP);
4713 prison_racct_foreach(void (*callback)(struct racct *racct,
4714 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4715 void *arg2, void *arg3)
4717 struct prison_racct *prr;
4719 ASSERT_RACCT_ENABLED();
4721 sx_slock(&allprison_lock);
4724 LIST_FOREACH(prr, &allprison_racct, prr_next)
4725 (callback)(prr->prr_racct, arg2, arg3);
4728 sx_sunlock(&allprison_lock);
4731 static struct prison_racct *
4732 prison_racct_find_locked(const char *name)
4734 struct prison_racct *prr;
4736 ASSERT_RACCT_ENABLED();
4737 sx_assert(&allprison_lock, SA_XLOCKED);
4739 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4742 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4743 if (strcmp(name, prr->prr_name) != 0)
4746 /* Found prison_racct with a matching name? */
4747 prison_racct_hold(prr);
4751 /* Add new prison_racct. */
4752 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4753 racct_create(&prr->prr_racct);
4755 strcpy(prr->prr_name, name);
4756 refcount_init(&prr->prr_refcount, 1);
4757 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4762 struct prison_racct *
4763 prison_racct_find(const char *name)
4765 struct prison_racct *prr;
4767 ASSERT_RACCT_ENABLED();
4769 sx_xlock(&allprison_lock);
4770 prr = prison_racct_find_locked(name);
4771 sx_xunlock(&allprison_lock);
4776 prison_racct_hold(struct prison_racct *prr)
4779 ASSERT_RACCT_ENABLED();
4781 refcount_acquire(&prr->prr_refcount);
4785 prison_racct_free_locked(struct prison_racct *prr)
4788 ASSERT_RACCT_ENABLED();
4789 sx_assert(&allprison_lock, SA_XLOCKED);
4791 if (refcount_release(&prr->prr_refcount)) {
4792 racct_destroy(&prr->prr_racct);
4793 LIST_REMOVE(prr, prr_next);
4794 free(prr, M_PRISON_RACCT);
4799 prison_racct_free(struct prison_racct *prr)
4802 ASSERT_RACCT_ENABLED();
4803 sx_assert(&allprison_lock, SA_UNLOCKED);
4805 if (refcount_release_if_not_last(&prr->prr_refcount))
4808 sx_xlock(&allprison_lock);
4809 prison_racct_free_locked(prr);
4810 sx_xunlock(&allprison_lock);
4814 prison_racct_attach(struct prison *pr)
4816 struct prison_racct *prr;
4818 ASSERT_RACCT_ENABLED();
4819 sx_assert(&allprison_lock, SA_XLOCKED);
4821 prr = prison_racct_find_locked(pr->pr_name);
4822 KASSERT(prr != NULL, ("cannot find prison_racct"));
4824 pr->pr_prison_racct = prr;
4828 * Handle jail renaming. From the racct point of view, renaming means
4829 * moving from one prison_racct to another.
4832 prison_racct_modify(struct prison *pr)
4838 struct prison_racct *oldprr;
4840 ASSERT_RACCT_ENABLED();
4842 sx_slock(&allproc_lock);
4843 sx_xlock(&allprison_lock);
4845 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4846 sx_xunlock(&allprison_lock);
4847 sx_sunlock(&allproc_lock);
4851 oldprr = pr->pr_prison_racct;
4852 pr->pr_prison_racct = NULL;
4854 prison_racct_attach(pr);
4857 * Move resource utilisation records.
4859 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4863 * Force rctl to reattach rules to processes.
4865 FOREACH_PROC_IN_SYSTEM(p) {
4867 cred = crhold(p->p_ucred);
4869 rctl_proc_ucred_changed(p, cred);
4874 sx_sunlock(&allproc_lock);
4875 prison_racct_free_locked(oldprr);
4876 sx_xunlock(&allprison_lock);
4880 prison_racct_detach(struct prison *pr)
4883 ASSERT_RACCT_ENABLED();
4884 sx_assert(&allprison_lock, SA_UNLOCKED);
4886 if (pr->pr_prison_racct == NULL)
4888 prison_racct_free(pr->pr_prison_racct);
4889 pr->pr_prison_racct = NULL;
4896 db_show_prison(struct prison *pr)
4898 struct bool_flags *bf;
4899 struct jailsys_flags *jsf;
4900 #if defined(INET) || defined(INET6)
4902 struct prison_ip *pip;
4906 char ip4buf[INET_ADDRSTRLEN];
4909 char ip6buf[INET6_ADDRSTRLEN];
4912 db_printf("prison %p:\n", pr);
4913 db_printf(" jid = %d\n", pr->pr_id);
4914 db_printf(" name = %s\n", pr->pr_name);
4915 db_printf(" parent = %p\n", pr->pr_parent);
4916 db_printf(" ref = %d\n", pr->pr_ref);
4917 db_printf(" uref = %d\n", pr->pr_uref);
4918 db_printf(" state = %s\n",
4919 pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4920 pr->pr_state == PRISON_STATE_DYING ? "dying" :
4922 db_printf(" path = %s\n", pr->pr_path);
4923 db_printf(" cpuset = %d\n", pr->pr_cpuset
4924 ? pr->pr_cpuset->cs_id : -1);
4926 db_printf(" vnet = %p\n", pr->pr_vnet);
4928 db_printf(" root = %p\n", pr->pr_root);
4929 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4930 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4931 db_printf(" children.max = %d\n", pr->pr_childmax);
4932 db_printf(" children.cur = %d\n", pr->pr_childcount);
4933 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4934 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4935 db_printf(" flags = 0x%x", pr->pr_flags);
4936 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4937 if (pr->pr_flags & bf->flag)
4938 db_printf(" %s", bf->name);
4939 for (jsf = pr_flag_jailsys;
4940 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4942 f = pr->pr_flags & (jsf->disable | jsf->new);
4943 db_printf(" %-16s= %s\n", jsf->name,
4944 (f != 0 && f == jsf->disable) ? "disable"
4945 : (f == jsf->new) ? "new"
4948 db_printf(" allow = 0x%x", pr->pr_allow);
4949 for (bf = pr_flag_allow;
4950 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4951 atomic_load_int(&bf->flag) != 0;
4953 if (pr->pr_allow & bf->flag)
4954 db_printf(" %s", bf->name);
4956 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4957 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4958 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4959 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4960 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4962 if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
4963 db_printf(" ip4s = %d\n", pip->ips);
4964 for (ii = 0; ii < pip->ips; ii++)
4965 db_printf(" %s %s\n",
4966 ii == 0 ? "ip4.addr =" : " ",
4968 *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
4973 if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
4974 db_printf(" ip6s = %d\n", pip->ips);
4975 for (ii = 0; ii < pip->ips; ii++)
4976 db_printf(" %s %s\n",
4977 ii == 0 ? "ip6.addr =" : " ",
4979 (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
4984 DB_SHOW_COMMAND(prison, db_show_prison_command)
4990 * Show all prisons in the list, and prison0 which is not
4993 db_show_prison(&prison0);
4994 if (!db_pager_quit) {
4995 TAILQ_FOREACH(pr, &allprison, pr_list) {
5007 /* Look for a prison with the ID and with references. */
5008 TAILQ_FOREACH(pr, &allprison, pr_list)
5009 if (pr->pr_id == addr && pr->pr_ref > 0)
5012 /* Look again, without requiring a reference. */
5013 TAILQ_FOREACH(pr, &allprison, pr_list)
5014 if (pr->pr_id == addr)
5017 /* Assume address points to a valid prison. */
5018 pr = (struct prison *)addr;