[PATCH v2 0/8] exec: Use sane stack rlimit under secureexec

As discussed with Linus and Andy, we need to reset the stack rlimit
before we do memory layouts when execing a privilege-gaining (e.g.
setuid) program. This moves security_bprm_secureexec() earlier (with
required changes), and then lowers the stack limit when appropriate.

As a side-effect, dumpability and pdeath_signal clearing is expanded
to cover LSM definitions of secureexec (and Smack can drop its special
handler for pdeath_signal clearing).

I'd appreciate some extra eyes on this to make sure this isn't
broken in some special way. I couldn't find anything that _depended_
on security_bprm_secureexec() being called late.

Thanks!

-Kees

v2:
- fix missed current_security() uses in LSMs.
- research/consolidate dumpability setting logic
- research/consolidate pdeath_signal clearing logic
- split up logical steps a little more for easier review (and bisection)
- fix some old broken comments

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

In commit 221af7f87b97 ("Split 'flush_old_exec' into two functions"),
the comment about the point of no return should have stayed in
flush_old_exec() since it refers to "bprm->mm = NULL;" line, but prior
changes in commits c89681ed7d0e ("remove steal_locks()"), and
fd8328be874f ("sanitize handling of shared descriptor tables in failing
execve()") made it look like it meant the current->sas_ss_sp line instead.

The comment is referring to the fact that once bprm->mm is NULL, all
failures from a binfmt load_binary hook (e.g. load_elf_binary), will
get SEGV raised against current. Move this comment and expand the
explanation a bit, putting it above the assignment this time.

This also removes an erroneous commet about when credentials are being
installed. That has its own dedicated function, install_exec_creds(),
which carries a similar (and correct) comment, so remove the bogus comment
where installation is not actually happening.

Signed-off-by: Kees Cook <redacted>
---
 fs/exec.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index 904199086490..7842ae661e34 100644
--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1285,7 +1285,14 @@ int flush_old_exec(struct linux_binprm * bprm)
 	if (retval)
 		goto out;
 
-	bprm->mm = NULL;		/* We're using it now */
+	/*
+	 * After clearing bprm->mm (to mark that current is using the
+	 * prepared mm now), we are at the point of no return. If
+	 * anything from here on returns an error, the check in
+	 * search_binary_handler() will kill current (since the mm has
+	 * been replaced).
+	 */
+	bprm->mm = NULL;
 
 	set_fs(USER_DS);
 	current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |

@@ -1332,7 +1339,6 @@ void setup_new_exec(struct linux_binprm * bprm)
 {
 	arch_pick_mmap_layout(current->mm);
 
-	/* This is the point of no return */
 	current->sas_ss_sp = current->sas_ss_size = 0;
 
 	if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))

@@ -1350,7 +1356,6 @@ void setup_new_exec(struct linux_binprm * bprm)
 	 */
 	current->mm->task_size = TASK_SIZE;
 
-	/* install the new credentials */
 	if (!uid_eq(bprm->cred->uid, current_euid()) ||
 	    !gid_eq(bprm->cred->gid, current_egid())) {
 		current->pdeath_signal = 0;

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

For a secureexec, before memory layout selection has happened, reset the
stack rlimit to something sane to avoid the caller having control over
the resulting layouts.

$ ulimit -s
8192
$ ulimit -s unlimited
$ /bin/sh -c 'ulimit -s'
unlimited
$ sudo /bin/sh -c 'ulimit -s'
8192

Signed-off-by: Kees Cook <redacted>
---
 fs/exec.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/fs/exec.c b/fs/exec.c
index e0186db02f90..1e3463854a16 100644
--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1343,6 +1343,16 @@ void setup_new_exec(struct linux_binprm * bprm)
 
 		/* Make sure parent cannot signal privileged process. */
 		current->pdeath_signal = 0;
+
+		/*
+		 * For secureexec, reset the stack limit to sane default to
+		 * avoid bad behavior from the prior rlimits. This has to
+		 * happen before arch_pick_mmap_layout(), which examines
+		 * RLIMIT_STACK, but after the point of no return to avoid
+		 * needing to clean up the change on failure.
+		 */
+		if (current->signal->rlim[RLIMIT_STACK].rlim_cur > _STK_LIM)
+			current->signal->rlim[RLIMIT_STACK].rlim_cur = _STK_LIM;
 	}
 
 	arch_pick_mmap_layout(current->mm);

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

The examination of "current" to decide dumpability is wrong. This was a
check of and euid/uid (or egid/gid) mismatch in the existing process,
not the newly created one. This appears to stretch back into even the
"history.git" tree. Luckily, dumpability is later set in commit_creds().
In earlier kernel versions before creds existed, similar checks also
existed late in the exec flow, covering up the mistake as far back as I
could find.

The commit_creds() check examines differences of euid, uid, egid, gid,
and capabilities between the old and new creds. It would look like
the setup_new_exec() dumpability test could be entirely removed, but
strictly speaking, the secureexec test covers a different set of tests
than what commit_creds() checks for. So, fix this test to use secureexec,
which includes the same logical check (euid != uid || egid != gid),
but checks bprm->cred, not current->cred.

One would wonder if we need a security_commit_creds() LSM hook and to
move the existing checks in commit_creds() into commoncaps.c, which
would allow expanding the logic to all LSMs. Currently this doesn't
seem needed, though.

Signed-off-by: Kees Cook <redacted>
---
 fs/exec.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/exec.c b/fs/exec.c
index b92e37fb53aa..3e519d4f0bd3 100644
--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1346,7 +1346,7 @@ void setup_new_exec(struct linux_binprm * bprm)
 
 	current->sas_ss_sp = current->sas_ss_size = 0;
 
-	if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
+	if (!bprm->secureexec)
 		set_dumpable(current->mm, SUID_DUMP_USER);
 	else
 		set_dumpable(current->mm, suid_dumpable);

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Instead of an additional secureexec check for pdeath_signal, just move it
up into the initial secureexec test. Neither perf nor arch code touches
pdeath_signal, so the relocation shouldn't change anything.

Signed-off-by: Kees Cook <redacted>
---
 fs/exec.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index eeb8323977d1..e0186db02f90 100644
--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1340,6 +1340,9 @@ void setup_new_exec(struct linux_binprm * bprm)
 	if (security_bprm_secureexec(bprm)) {
 		/* Record for AT_SECURE. */
 		bprm->secureexec = 1;
+
+		/* Make sure parent cannot signal privileged process. */
+		current->pdeath_signal = 0;
 	}
 
 	arch_pick_mmap_layout(current->mm);

@@ -1363,10 +1366,6 @@ void setup_new_exec(struct linux_binprm * bprm)
 	 */
 	current->mm->task_size = TASK_SIZE;
 
-	if (bprm->secureexec) {
-		current->pdeath_signal = 0;
-	}
-
 	/* An exec changes our domain. We are no longer part of the thread
 	   group */
 	current->self_exec_id++;

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

There are several places where exec needs to know if a privilege-gain has
happened. These should be using the results of security_bprm_secureexec()
but it is getting (needlessly) called very late.

Instead, move this earlier in the exec code, to the start of the point
of no return in setup_new_exec(). Here, the new creds have already
been calculated (and stored in bprm->cred), which is normally what
security_bprm_secureexec() wants to examine. Since it's moved earlier,
LSMs hooking bprm_secureexec() need to be adjusted to use the creds in
bprm:

$ git grep LSM_HOOK_INIT.*bprm_secureexec
apparmor/lsm.c:    LSM_HOOK_INIT(bprm_secureexec, apparmor_bprm_secureexec),
commoncap.c:       LSM_HOOK_INIT(bprm_secureexec, cap_bprm_secureexec),
selinux/hooks.c:   LSM_HOOK_INIT(bprm_secureexec, selinux_bprm_secureexec),
smack/smack_lsm.c: LSM_HOOK_INIT(bprm_secureexec, smack_bprm_secureexec),

AppArmor does not access creds in apparmor_bprm_secureexec.

Capabilities needed to be adjusted to use bprm creds.

SELinux needed to be adjusted to use bprm creds for the security structure.

Smack needed to be adjusted to use bprm creds for the security structure.

The result of the bprm_secureexec() hook is saved in a new bprm field
"secureexec" so it can be queried later (just AT_SECURE currently).

Signed-off-by: Kees Cook <redacted>
---
 fs/binfmt_elf.c            | 2 +-
 fs/binfmt_elf_fdpic.c      | 2 +-
 fs/exec.c                  | 5 +++++
 include/linux/binfmts.h    | 3 ++-
 include/linux/lsm_hooks.h  | 3 ++-
 security/commoncap.c       | 4 +---
 security/selinux/hooks.c   | 2 +-
 security/smack/smack_lsm.c | 2 +-
 8 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index 5075fd5c62c8..7f6ec4dac13d 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c

@@ -254,7 +254,7 @@ create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
- 	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
+	NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
 #ifdef ELF_HWCAP2
 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);

diff --git a/fs/binfmt_elf_fdpic.c b/fs/binfmt_elf_fdpic.c
index cf93a4fad012..5aa9199dfb13 100644
--- a/fs/binfmt_elf_fdpic.c
+++ b/fs/binfmt_elf_fdpic.c

@@ -650,7 +650,7 @@ static int create_elf_fdpic_tables(struct linux_binprm *bprm,
 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
-	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
+	NEW_AUX_ENT(AT_SECURE,	bprm->secureexec);
 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
 
 #ifdef ARCH_DLINFO

diff --git a/fs/exec.c b/fs/exec.c
index 7842ae661e34..b92e37fb53aa 100644
--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1337,6 +1337,11 @@ EXPORT_SYMBOL(would_dump);
 
 void setup_new_exec(struct linux_binprm * bprm)
 {
+	if (security_bprm_secureexec(bprm)) {
+		/* Record for AT_SECURE. */
+		bprm->secureexec = 1;
+	}
+
 	arch_pick_mmap_layout(current->mm);
 
 	current->sas_ss_sp = current->sas_ss_size = 0;

diff --git a/include/linux/binfmts.h b/include/linux/binfmts.h
index 05488da3aee9..1afaa303cad0 100644
--- a/include/linux/binfmts.h
+++ b/include/linux/binfmts.h

@@ -27,9 +27,10 @@ struct linux_binprm {
 	unsigned int
 		cred_prepared:1,/* true if creds already prepared (multiple
 				 * preps happen for interpreters) */
-		cap_effective:1;/* true if has elevated effective capabilities,
+		cap_effective:1,/* true if has elevated effective capabilities,
 				 * false if not; except for init which inherits
 				 * its parent's caps anyway */
+		secureexec:1;	/* true when gaining privileges */
 #ifdef __alpha__
 	unsigned int taso:1;
 #endif

diff --git a/include/linux/lsm_hooks.h b/include/linux/lsm_hooks.h
index 080f34e66017..d1bd24fb4a33 100644
--- a/include/linux/lsm_hooks.h
+++ b/include/linux/lsm_hooks.h

@@ -72,7 +72,8 @@
  *	Return a boolean value (0 or 1) indicating whether a "secure exec"
  *	is required.  The flag is passed in the auxiliary table
  *	on the initial stack to the ELF interpreter to indicate whether libc
- *	should enable secure mode.
+ *	should enable secure mode. Called before bprm_committing_creds(),
+ *	so pending credentials are in @bprm->cred.
  *	@bprm contains the linux_binprm structure.
  *
  * Security hooks for filesystem operations.

diff --git a/security/commoncap.c b/security/commoncap.c
index 7abebd782d5e..482d3aac2fc6 100644
--- a/security/commoncap.c
+++ b/security/commoncap.c

@@ -624,12 +624,10 @@ int cap_bprm_set_creds(struct linux_binprm *bprm)
  * Determine whether a secure execution is required, return 1 if it is, and 0
  * if it is not.
  *
- * The credentials have been committed by this point, and so are no longer
- * available through @bprm->cred.
  */
 int cap_bprm_secureexec(struct linux_binprm *bprm)
 {
-	const struct cred *cred = current_cred();
+	const struct cred *cred = bprm->cred;
 	kuid_t root_uid = make_kuid(cred->user_ns, 0);
 
 	if (!uid_eq(cred->uid, root_uid)) {

diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
index 819fd6858b49..9381c8474cf4 100644
--- a/security/selinux/hooks.c
+++ b/security/selinux/hooks.c

@@ -2420,7 +2420,7 @@ static int selinux_bprm_set_creds(struct linux_binprm *bprm)
 
 static int selinux_bprm_secureexec(struct linux_binprm *bprm)
 {
-	const struct task_security_struct *tsec = current_security();
+	const struct task_security_struct *tsec = bprm->cred->security;
 	u32 sid, osid;
 	int atsecure = 0;
 

diff --git a/security/smack/smack_lsm.c b/security/smack/smack_lsm.c
index 658f5d8c7e76..13cf9e66d5fe 100644
--- a/security/smack/smack_lsm.c
+++ b/security/smack/smack_lsm.c

@@ -975,7 +975,7 @@ static void smack_bprm_committing_creds(struct linux_binprm *bprm)
  */
 static int smack_bprm_secureexec(struct linux_binprm *bprm)
 {
-	struct task_smack *tsp = current_security();
+	struct task_smack *tsp = bprm->cred->security;
 
 	if (tsp->smk_task != tsp->smk_forked)
 		return 1;

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Like dumpability, clearing pdeath_signal happens both in setup_new_exec()
and later in commit_creds(). The test in setup_new_exec() is different
from all other privilege comparisons, though: it is checking the new cred
(bprm) uid vs the old cred (current) euid. This appears to be a bug,
introduced by commit a6f76f23d297 ("CRED: Make execve() take advantage of
copy-on-write credentials"):

-       if (bprm->e_uid != current_euid() ||
-           bprm->e_gid != current_egid()) {
-               set_dumpable(current->mm, suid_dumpable);
+       /* install the new credentials */
+       if (bprm->cred->uid != current_euid() ||
+           bprm->cred->gid != current_egid()) {

It was bprm euid vs current euid (and egids), but the effective got
dropped. Nothing in the exec flow changes bprm->cred->uid (nor gid).
The call traces are:

	prepare_bprm_creds()
	    prepare_exec_creds()
	        prepare_creds()
	            memcpy(new_creds, old_creds, ...)
	            security_prepare_creds() (unimplemented by commoncap)
	...
	prepare_binprm()
	    bprm_fill_uid()
	        resets euid/egid to current euid/egid
	        sets euid/egid on bprm based on set*id file bits
	    security_bprm_set_creds()
		cap_bprm_set_creds()
		        handle all caps-based manipulations

so this test is effectively a test of current_uid() vs current_euid(),
which is wrong, just like the prior dumpability tests were wrong.

The commit log says "Clear pdeath_signal and set dumpable on
certain circumstances that may not be covered by commit_creds()." This
may be meaning the earlier old euid vs new euid (and egid) test that
got changed.

Luckily, as with dumpability, this is all masked by commit_creds()
which performs old/new euid and egid tests and clears pdeath_signal.

And again, like dumpability, we should include LSM secureexec logic for
pdeath_signal clearing. For example, Smack goes out of its way to clear
pdeath_signal when it finds a secureexec condition.

Signed-off-by: Kees Cook <redacted>
---
 fs/exec.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index 3e519d4f0bd3..d7bda5b60e7b 100644
--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1361,8 +1361,7 @@ void setup_new_exec(struct linux_binprm * bprm)
 	 */
 	current->mm->task_size = TASK_SIZE;
 
-	if (!uid_eq(bprm->cred->uid, current_euid()) ||
-	    !gid_eq(bprm->cred->gid, current_egid())) {
+	if (bprm->secureexec) {
 		current->pdeath_signal = 0;
 	} else {
 		if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

This removes the redundant pdeath_signal clearing in Smack: the check in
smack_bprm_committing_creds() matches the check in smack_bprm_secureexec(),
and since secureexec is now being checked for clearing pdeath_signal, this
is redundant to the common exec code.

Signed-off-by: Kees Cook <redacted>
---
 security/smack/smack_lsm.c | 15 ---------------
 1 file changed, 15 deletions(-)

diff --git a/security/smack/smack_lsm.c b/security/smack/smack_lsm.c
index 13cf9e66d5fe..4b10b782aecc 100644
--- a/security/smack/smack_lsm.c
+++ b/security/smack/smack_lsm.c

@@ -954,20 +954,6 @@ static int smack_bprm_set_creds(struct linux_binprm *bprm)
 }
 
 /**
- * smack_bprm_committing_creds - Prepare to install the new credentials
- * from bprm.
- *
- * @bprm: binprm for exec
- */
-static void smack_bprm_committing_creds(struct linux_binprm *bprm)
-{
-	struct task_smack *bsp = bprm->cred->security;
-
-	if (bsp->smk_task != bsp->smk_forked)
-		current->pdeath_signal = 0;
-}
-
-/**
  * smack_bprm_secureexec - Return the decision to use secureexec.
  * @bprm: binprm for exec
  *

@@ -4645,7 +4631,6 @@ static struct security_hook_list smack_hooks[] __lsm_ro_after_init = {
 	LSM_HOOK_INIT(sb_parse_opts_str, smack_parse_opts_str),
 
 	LSM_HOOK_INIT(bprm_set_creds, smack_bprm_set_creds),
-	LSM_HOOK_INIT(bprm_committing_creds, smack_bprm_committing_creds),
 	LSM_HOOK_INIT(bprm_secureexec, smack_bprm_secureexec),
 
 	LSM_HOOK_INIT(inode_alloc_security, smack_inode_alloc_security),

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Since it's already valid to set dumpability in the early part of
setup_new_exec(), we can consolidate the logic into a single place.
The BINPRM_FLAGS_ENFORCE_NONDUMP is set during would_dump() calls
before setup_new_exec(), so its test is safe to move as well.

Signed-off-by: Kees Cook <redacted>
---
 fs/exec.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index d7bda5b60e7b..eeb8323977d1 100644
--- a/fs/exec.c
+++ b/fs/exec.c

@@ -1346,10 +1346,12 @@ void setup_new_exec(struct linux_binprm * bprm)
 
 	current->sas_ss_sp = current->sas_ss_size = 0;
 
-	if (!bprm->secureexec)
-		set_dumpable(current->mm, SUID_DUMP_USER);
-	else
+	/* Figure out dumpability. */
+	if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
+	    bprm->secureexec)
 		set_dumpable(current->mm, suid_dumpable);
+	else
+		set_dumpable(current->mm, SUID_DUMP_USER);
 
 	arch_setup_new_exec();
 	perf_event_exec();

@@ -1363,9 +1365,6 @@ void setup_new_exec(struct linux_binprm * bprm)
 
 	if (bprm->secureexec) {
 		current->pdeath_signal = 0;
-	} else {
-		if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
-			set_dumpable(current->mm, suid_dumpable);
 	}
 
 	/* An exec changes our domain. We are no longer part of the thread

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html