git: a4df9e3efa30 - stable/15 - jail: prevent a race between jail_attach in different threads
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Sun, 05 Jul 2026 02:54:59 UTC
The branch stable/15 has been updated by jamie:
URL: https://cgit.FreeBSD.org/src/commit/?id=a4df9e3efa30d2486776a530e6a3548bc74d7b98
commit a4df9e3efa30d2486776a530e6a3548bc74d7b98
Author: Jamie Gritton <jamie@FreeBSD.org>
AuthorDate: 2026-07-02 22:48:07 +0000
Commit: Jamie Gritton <jamie@FreeBSD.org>
CommitDate: 2026-07-05 02:54:41 +0000
jail: prevent a race between jail_attach in different threads
Attaching to a jail changes its root directory and its process
credentials. These operations both require unlocking the jail, and also
need allprison_lock unlocked. That means that if two threads are trying
to attach to different jails at the same time, it's possible for the
process to end up with one jail's root directory but the other jail's
credentials.
Solve this by forcing the process into single-threaded mode during
system calls that attach to a jail (jail_attach, jail_attach_jd, and
sometimes jail_set).
Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D57858
(cherry picked from commit d4e0f4dab2d7f4de46bb79db1ca7e6e8a2e34746)
---
sys/kern/kern_jail.c | 72 +++++++++++---
tests/sys/kern/Makefile | 2 +
tests/sys/kern/jail_thread.c | 231 +++++++++++++++++++++++++++++++++++++++++++
3 files changed, 294 insertions(+), 11 deletions(-)
diff --git a/sys/kern/kern_jail.c b/sys/kern/kern_jail.c
index 8a6c3e1d3fd0..6b0ae8fbb8ce 100644
--- a/sys/kern/kern_jail.c
+++ b/sys/kern/kern_jail.c
@@ -155,6 +155,8 @@ int lastdeadid = 0;
static int get_next_prid(struct prison **insprp);
static int get_next_deadid(struct prison **insprp);
static int do_jail_attach(struct thread *td, struct prison *pr, int *drflagsp);
+static int prison_attach_thread_single(struct thread *td);
+static void prison_attach_thread_single_end(struct thread *td);
static void prison_complete(void *context, int pending);
static void prison_deref(struct prison *pr, int flags);
static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
@@ -1045,7 +1047,13 @@ kern_jail_set(struct thread *td, struct uio *optuio, int flags)
if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
(JAIL_USE_DESC | JAIL_AT_DESC))
return (EINVAL);
- prison_hold(mypr);
+
+ /* Only let a single thread in the process try to attach at a time. */
+ if (flags & JAIL_ATTACH) {
+ error = prison_attach_thread_single(td);
+ if (error != 0)
+ return (error);
+ }
#ifdef INET
ip4 = NULL;
@@ -1056,6 +1064,7 @@ kern_jail_set(struct thread *td, struct uio *optuio, int flags)
g_path = NULL;
jfp_out = NULL;
jfd_out = -1;
+ prison_hold(mypr);
/*
* Check all the parameters before committing to anything. Not all
* errors can be caught early, but we may as well try. Also, this
@@ -2342,6 +2351,8 @@ kern_jail_set(struct thread *td, struct uio *optuio, int flags)
if (opts != NULL)
vfs_freeopts(opts);
prison_free(mypr);
+ if (flags & JAIL_ATTACH)
+ prison_attach_thread_single_end(td);
return (error);
}
@@ -2940,6 +2951,10 @@ sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
error = priv_check(td, PRIV_JAIL_ATTACH);
if (error)
return (error);
+ /* Only let a single thread in the process try to attach at a time. */
+ error = prison_attach_thread_single(td);
+ if (error != 0)
+ return (error);
sx_slock(&allprison_lock);
drflags = PD_LIST_SLOCKED;
@@ -2960,6 +2975,7 @@ sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
done:
prison_deref(pr, drflags);
+ prison_attach_thread_single_end(td);
return (error);
}
@@ -2975,6 +2991,11 @@ sys_jail_attach_jd(struct thread *td, struct jail_attach_jd_args *uap)
struct ucred *jdcred;
int drflags, error;
+ /* Only let a single thread in the process try to attach at a time. */
+ error = prison_attach_thread_single(td);
+ if (error != 0)
+ return (error);
+
sx_slock(&allprison_lock);
drflags = PD_LIST_SLOCKED;
pr = NULL;
@@ -2997,6 +3018,7 @@ sys_jail_attach_jd(struct thread *td, struct jail_attach_jd_args *uap)
done:
prison_deref(pr, drflags);
+ prison_attach_thread_single_end(td);
return (error);
}
@@ -3014,15 +3036,6 @@ do_jail_attach(struct thread *td, struct prison *pr, int *drflagsp)
sx_assert(&allprison_lock, SX_LOCKED);
KASSERT(prison_isvalid(pr), ("Attaching to invalid prison %p", pr));
- /*
- * XXX: Note that there is a slight race here if two threads
- * in the same privileged process attempt to attach to two
- * different jails at the same time. It is important for
- * user processes not to do this, or they might end up with
- * a process root from one prison, but attached to the jail
- * of another.
- */
-
/*
* Note the caller's locking state, but gain and track our own
* references. The caller will see that locks have been
@@ -3105,7 +3118,7 @@ do_jail_attach(struct thread *td, struct prison *pr, int *drflagsp)
e_unlock:
VOP_UNLOCK(pr->pr_root);
e_revert_osd:
- /* Tell modules this thread is still in its old jail after all. */
+ /* Tell modules this process is still in its old jail after all. */
if (!(drflags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
sx_slock(&allprison_lock);
drflags |= PD_LIST_SLOCKED;
@@ -3115,6 +3128,43 @@ do_jail_attach(struct thread *td, struct prison *pr, int *drflagsp)
return (error);
}
+/*
+ * Only one thread in a process should try to attach to a jail, or
+ * they might end up with a process root from one prison, but attached
+ * to the jail of another. Enforce this by making the process run in
+ * single-threaded mode for the duration of the system call, which
+ * also prevents other related calls such as chroot.
+ */
+static int
+prison_attach_thread_single(struct thread *td)
+{
+ struct proc *p;
+ int error;
+
+ error = 0;
+ p = td->td_proc;
+ if ((atomic_load_int(&p->p_flag) & P_HADTHREADS) != 0) {
+ PROC_LOCK(p);
+ if (thread_single(p, SINGLE_BOUNDARY))
+ error = ERESTART;
+ PROC_UNLOCK(p);
+ }
+ return (error);
+}
+
+static void
+prison_attach_thread_single_end(struct thread *td)
+{
+ struct proc *p;
+
+ p = td->td_proc;
+ if ((atomic_load_int(&p->p_flag) & P_HADTHREADS) != 0) {
+ PROC_LOCK(p);
+ thread_single_end(p, SINGLE_BOUNDARY);
+ PROC_UNLOCK(p);
+ }
+}
+
/*
* Returns a locked prison instance, or NULL on failure.
*/
diff --git a/tests/sys/kern/Makefile b/tests/sys/kern/Makefile
index 4fe194771f56..03aca4f19a56 100644
--- a/tests/sys/kern/Makefile
+++ b/tests/sys/kern/Makefile
@@ -24,6 +24,7 @@ ATF_TESTS_C+= exterr_test
ATF_TESTS_C+= fdgrowtable_test
ATF_TESTS_C+= getdirentries_test
ATF_TESTS_C+= jail_lookup_root
+ATF_TESTS_C+= jail_thread
ATF_TESTS_C+= jaildesc
ATF_TESTS_C+= inotify_test
ATF_TESTS_C+= kill_zombie
@@ -96,6 +97,7 @@ PROGS+= sendfile_helper
LIBADD.aslr+= util
LIBADD.copy_file_range+= md
LIBADD.jail_lookup_root+= jail util
+LIBADD.jail_thread+= jail pthread
LIBADD.jaildesc+= kvm pthread
CFLAGS.sys_getrandom+= -I${SRCTOP}/sys/contrib/zstd/lib
LIBADD.sys_getrandom+= zstd
diff --git a/tests/sys/kern/jail_thread.c b/tests/sys/kern/jail_thread.c
new file mode 100644
index 000000000000..1af25f12386d
--- /dev/null
+++ b/tests/sys/kern/jail_thread.c
@@ -0,0 +1,231 @@
+/*-
+ * SPDX-License-Identifier: BSD-2-Clause
+ *
+ * Copyright (c) 2026 James Gritton <jamie@FreeBSD.org>
+ */
+
+#include <sys/param.h>
+#include <sys/jail.h>
+#include <sys/stat.h>
+
+#include <err.h>
+#include <errno.h>
+#include <jail.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include <atf-c.h>
+
+#define NJAILS 3 /* One master jail and two to race. */
+#define NROUNDS 10000 /* Number of attempts to make race happen. */
+
+struct jailinfo {
+ int jfd;
+ ino_t ino;
+ char path[MAXPATHLEN];
+};
+
+static pthread_barrier_t barrier;
+static struct jailinfo jinfo[NJAILS];
+
+/* Attach a thread to a jail with jail_attach_jd. */
+static void *
+thread_jail_attach_jd(void *arg)
+{
+ int error;
+
+ /*
+ * Synchronize to get as close as possible to the same time,
+ * then attach to a jail.
+ */
+ error = pthread_barrier_wait(&barrier);
+ ATF_REQUIRE_MSG(error == 0 || error == PTHREAD_BARRIER_SERIAL_THREAD,
+ "pthread_barrier_wait: %s", strerror(errno));
+ ATF_REQUIRE_MSG(jail_attach_jd(jinfo[(size_t)arg].jfd) == 0,
+ "jail_attach_jd: %s", strerror(errno));
+ return (NULL);
+}
+
+/* Attach a thread to a jail with jail_setv. */
+static void *
+thread_jail_setv(void *arg)
+{
+ int error;
+ char jdescstr[16];
+
+ error = pthread_barrier_wait(&barrier);
+ ATF_REQUIRE_MSG(error == 0 || error == PTHREAD_BARRIER_SERIAL_THREAD,
+ "pthread_barrier_wait: %s", strerror(errno));
+ snprintf(jdescstr, sizeof(jdescstr), "%d", jinfo[(size_t)arg].jfd);
+ ATF_REQUIRE_MSG(jail_setv(JAIL_UPDATE | JAIL_ATTACH | JAIL_USE_DESC,
+ "desc", jdescstr, NULL) > 0,
+ "jail_setv: %s", jail_errmsg[0] ? jail_errmsg : strerror(errno));
+ return (NULL);
+}
+
+/* Attach a thread to a jail with chroot. */
+static void *
+thread_chroot(void *arg)
+{
+ int error;
+
+ /* This is a race between jail_attach and chroot. */
+ if ((size_t)arg > 1)
+ return thread_jail_attach_jd(arg);
+ error = pthread_barrier_wait(&barrier);
+ ATF_REQUIRE_MSG(error == 0 || error == PTHREAD_BARRIER_SERIAL_THREAD,
+ "pthread_barrier_wait: %s", strerror(errno));
+ ATF_REQUIRE_MSG(chroot(jinfo[(size_t)arg].path) == 0 || errno == ENOENT,
+ "chroot: %s", strerror(errno));
+ return (NULL);
+}
+
+static void
+thread_attach_test(void *(*thread_handler)(void*), const char *jail_name,
+ const char *syscall_name, bool jail_reset)
+{
+ int ri, spn, mixed_jails;
+ size_t ji, ti, ji_hostname, ji_ino, jail_namelen;
+ char *cwd;
+ struct stat st;
+ char jnamestr[64], jdescstr[16];
+ pthread_t threads[NJAILS];
+
+ if (jinfo[0].jfd == 0) {
+ /* Start with a master jail, so we can return to real root. */
+ jdescstr[0] = '\0';
+ ATF_REQUIRE_MSG(jail_setv(JAIL_CREATE | JAIL_OWN_DESC,
+ "name", jail_name,
+ "path", "/",
+ "desc", jdescstr,
+ "persist", "true",
+ NULL) > 0,
+ "jail_setv jail 0: %s",
+ jail_errmsg[0] ? jail_errmsg : strerror(errno));
+ jinfo[0].jfd = strtol(jdescstr, NULL, 10);
+ /* Make enough jails to cause contention. */
+ cwd = getcwd(NULL, MAXPATHLEN);
+ ATF_REQUIRE_MSG(cwd != NULL, "getcwd: %s", strerror(errno));
+ for (ji = 1; ji < NJAILS; ++ji) {
+ snprintf(jnamestr, sizeof(jnamestr),
+ "%s%zu", jail_name, ji);
+ spn = snprintf(jinfo[ji].path, MAXPATHLEN,
+ "%s/jail%zu", cwd, ji);
+ ATF_REQUIRE_MSG((size_t)spn < MAXPATHLEN,
+ "snprintf exceeded MAXPATHLEN: %d", spn);
+ ATF_REQUIRE_MSG(
+ mkdir(jinfo[ji].path, 0755) == 0 || errno == EEXIST,
+ "mkdir %s: %s", jinfo[ji].path, strerror(errno));
+ ATF_REQUIRE_MSG(stat(jinfo[ji].path, &st) == 0,
+ "stat %s: %s", jinfo[ji].path, strerror(errno));
+ jinfo[ji].ino = st.st_ino;
+ jdescstr[0] = '\0';
+ ATF_REQUIRE_MSG(jail_setv(JAIL_CREATE | JAIL_OWN_DESC,
+ "name", jnamestr,
+ "host.hostname", jnamestr,
+ "path", jinfo[ji].path,
+ "desc", jdescstr,
+ "persist", "true",
+ NULL) > 0,
+ "jail_setv: %s",
+ jail_errmsg[0] ? jail_errmsg : strerror(errno));
+ jinfo[ji].jfd = strtol(jdescstr, NULL, 10);
+ }
+ } else
+ ATF_REQUIRE_MSG(jail_attach_jd(jinfo[0].jfd) == 0,
+ "jail_attach_jd: %s", strerror(errno));
+
+ /* Check the different system calls that can race. */
+ jail_namelen = strlen(jail_name);
+ mixed_jails = 0;
+ for (ri = 0; ri < NROUNDS; ++ri) {
+ ATF_REQUIRE_MSG(
+ pthread_barrier_init(&barrier, NULL, NJAILS - 1) == 0,
+ "pthread_barrier_init: %s", strerror(errno));
+ for (ti = 1; ti < NJAILS; ++ti)
+ ATF_REQUIRE_MSG(
+ pthread_create(&threads[ti], NULL, thread_handler,
+ (void*)ti) == 0,
+ "pthread_create: %s", strerror(errno));
+ for (ti = 1; ti < NJAILS; ++ti)
+ ATF_REQUIRE_MSG(
+ pthread_join(threads[ti], NULL) == 0,
+ "pthread_join: %s", strerror(errno));
+ ATF_REQUIRE_MSG(pthread_barrier_destroy(&barrier) == 0,
+ "pthread_barrier_destroy: %s", strerror(errno));
+ /*
+ * Find the current jail from the hostname, and also
+ * by the root inode. They should be the same.
+ */
+ ATF_REQUIRE_MSG(
+ gethostname(jnamestr, sizeof(jnamestr)) == 0,
+ "gethostname: %s", strerror(errno));
+ ATF_REQUIRE_MSG(strncmp(jnamestr, jail_name, jail_namelen) == 0,
+ "unexpected jail hostname %s", jnamestr);
+ ji_hostname = strtol(jnamestr + jail_namelen, NULL, 10);
+ ATF_REQUIRE_MSG(stat("/", &st) == 0,
+ "stat /: %s", strerror(errno));
+ for (ji_ino = 1; ji_ino < NJAILS; ++ji_ino)
+ if (jinfo[ji_ino].ino == st.st_ino)
+ break;
+ ATF_REQUIRE_MSG(ji_ino < NJAILS,
+ "unexpected jail root inode %lu",
+ (unsigned long)st.st_ino);
+ mixed_jails += ji_hostname != ji_ino;
+ /* Reset to the master jail, required for chroot. */
+ if (jail_reset)
+ ATF_REQUIRE_MSG(jail_attach_jd(jinfo[0].jfd) == 0,
+ "jail_attach_jd: %s", strerror(errno));
+ }
+ /* It's an error if any of the rounds had a mismatch. */
+ ATF_REQUIRE_MSG(mixed_jails == 0,
+ "%d of %d %s races with different root and "
+ "credentials", mixed_jails, NROUNDS, syscall_name);
+}
+
+#define JAIL_NAME "jail_thread_attach_test"
+
+
+ATF_TC(jail_thread_attach);
+ATF_TC_HEAD(jail_thread_attach, tc)
+{
+ atf_tc_set_md_var(tc, "require.user", "root");
+}
+ATF_TC_BODY(jail_thread_attach, tc)
+{
+ thread_attach_test(thread_jail_attach_jd,
+ "jail_thread_attach_test", "jail_attach_jd", false);
+}
+
+ATF_TC(jail_thread_attach_jail_set);
+ATF_TC_HEAD(jail_thread_attach_jail_set, tc)
+{
+ atf_tc_set_md_var(tc, "require.user", "root");
+}
+ATF_TC_BODY(jail_thread_attach_jail_set, tc)
+{
+ thread_attach_test(thread_jail_setv,
+ "jail_thread_attach_test_jail_set", "jail_set", false);
+}
+
+ATF_TC(jail_thread_attach_chroot);
+ATF_TC_HEAD(jail_thread_attach_chroot, tc)
+{
+ atf_tc_set_md_var(tc, "require.user", "root");
+}
+ATF_TC_BODY(jail_thread_attach_chroot, tc)
+{
+ thread_attach_test(thread_chroot,
+ "jail_thread_attach_test_chroot", "chroot", true);
+}
+
+ATF_TP_ADD_TCS(tp)
+{
+ ATF_TP_ADD_TC(tp, jail_thread_attach);
+ ATF_TP_ADD_TC(tp, jail_thread_attach_jail_set);
+ ATF_TP_ADD_TC(tp, jail_thread_attach_chroot);
+ return (atf_no_error());
+}