[PATCH 0/3] Fix pasta-in-pasta operation (and similar)

[PATCH 0/3] Fix pasta-in-pasta operation (and similar)

[Stefano Brivio]

When pasta spawns a command (operation without pre-existing namespace),
it calls clone(2) with CLONE_NEWPID to detach the PID namespace where
this command runs, but it needs to mount /proc (in a separate mount
namespace), otherwise its contents are not consistent with the new
PID namespace.

If /proc contents are not consistent, pasta will fail to run in a
user and network namespace created by another pasta instance.

An alternative would be to drop CLONE_NEWPID altogether: pasta
itself not a container engine, and it's not meant to provide general
isolation features other than for networking aspects. This would
also make testing and debugging a bit easier, as the PIDs of
processes descending from pasta would be the same outside the
detached namespace.

However, also for testing and debugging usage itself, we would lose
two advantages: the inner environment looks more observable (from
inside) with CLONE_NEWPID, and we don't need to explicitly clean up
the environment as pasta terminates: see the ugliness of
pasta_ns_cleanup() before commit 0515adceaa8f ("passt, pasta:
Namespace-based sandboxing, defer seccomp policy application"). It
wasn't very robust either.

Now that this part works, note that writing to the uid_map procfs
entry, with 0 as domain for the map, requires (since Linux 5.12)
CAP_SETFCAP in the parent process. We need this mapping to keep the
behaviour consistent with what happens when we run directly from
the init namespace, and to set the ping_group_range sysctl. Keep
CAP_SETFCAP if we're running with UID 0 from a non-init user
namespace.

With this series, pasta finally runs in itself. I checked basic
connectivity inside a dozen of recursively nested instances.


Stefano Brivio (3):
  util, conf: Add and use ns_is_init() helper
  pasta: Detach mount namespace, (re)mount procfs before spawning
    command
  isolation: Initially Keep CAP_SETFCAP if running as UID 0 in non-init

 conf.c      | 16 +---------------
 isolation.c | 17 ++++++++++++++---
 pasta.c     |  7 ++++++-
 util.c      | 25 +++++++++++++++++++++++++
 util.h      |  2 ++
 5 files changed, 48 insertions(+), 19 deletions(-)

-- 
2.39.2

[PATCH 1/3] util, conf: Add and use ns_is_init() helper

[Stefano Brivio]

We'll need this in isolate_initial(). While at it, don't rely on
BUFSIZ: the earlier issue we had with musl reminded me it's not a
magic "everything will fit" value. Size the read buffer to what we
actually need from uid_map.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
---
 conf.c | 16 +---------------
 util.c | 25 +++++++++++++++++++++++++
 util.h |  2 ++
 3 files changed, 28 insertions(+), 15 deletions(-)

diff --git a/conf.c b/conf.c
index 447b000..984c3ce 100644
--- a/conf.c
+++ b/conf.c
@@ -1096,10 +1096,6 @@ static int conf_runas(char *opt, unsigned int *uid, unsigned int *gid)
  */
 static void conf_ugid(char *runas, uid_t *uid, gid_t *gid)
 {
-	const char root_uid_map[] = "         0          0 4294967295";
-	char buf[BUFSIZ];
-	int fd;
-
 	/* If user has specified --runas, that takes precedence... */
 	if (runas) {
 		if (conf_runas(runas, uid, gid))
@@ -1116,18 +1112,8 @@ static void conf_ugid(char *runas, uid_t *uid, gid_t *gid)
 		return;
 
 	/* ...or at least not root in the init namespace... */
-	if ((fd = open("/proc/self/uid_map", O_RDONLY | O_CLOEXEC)) < 0) {
-		die("Can't determine if we're in init namespace: %s",
-		    strerror(errno));
-	}
-
-	if (read(fd, buf, BUFSIZ) != sizeof(root_uid_map) ||
-	    strncmp(buf, root_uid_map, sizeof(root_uid_map) - 1)) {
-		close(fd);
+	if (!ns_is_init())
 		return;
-	}
-
-	close(fd);
 
 	/* ...otherwise use nobody:nobody */
 	warn("Don't run as root. Changing to nobody...");
diff --git a/util.c b/util.c
index c3e3471..5ec8a6c 100644
--- a/util.c
+++ b/util.c
@@ -390,6 +390,31 @@ int ns_enter(const struct ctx *c)
 	return 0;
 }
 
+/**
+ * ns_is_init() - Is the caller running in the "init" user namespace?
+ *
+ * Return: true if caller is in init, false otherwise, won't return on failure
+ */
+bool ns_is_init(void)
+{
+	const char root_uid_map[] = "         0          0 4294967295";
+	char buf[sizeof(root_uid_map) + 1];
+	bool ret = true;
+	int fd;
+
+	if ((fd = open("/proc/self/uid_map", O_RDONLY | O_CLOEXEC)) < 0) {
+		die("Can't determine if we're in init namespace: %s",
+		    strerror(errno));
+	}
+
+	if (read(fd, buf, sizeof(root_uid_map)) != sizeof(root_uid_map) ||
+	    strncmp(buf, root_uid_map, sizeof(root_uid_map) - 1))
+		ret = false;
+
+	close(fd);
+	return ret;
+}
+
 /**
  * pid_file() - Write PID to file, if requested to do so, and close it
  * @fd:		Open PID file descriptor, closed on exit, -1 to skip writing it
diff --git a/util.h b/util.h
index ba3e3da..26892aa 100644
--- a/util.h
+++ b/util.h
@@ -8,6 +8,7 @@
 
 #include <stdlib.h>
 #include <stdarg.h>
+#include <stdbool.h>
 
 #include "log.h"
 
@@ -216,6 +217,7 @@ char *line_read(char *buf, size_t len, int fd);
 void procfs_scan_listen(struct ctx *c, uint8_t proto, int ip_version, int ns,
 			uint8_t *map, uint8_t *exclude);
 int ns_enter(const struct ctx *c);
+bool ns_is_init(void);
 void write_pidfile(int fd, pid_t pid);
 int __daemon(int pidfile_fd, int devnull_fd);
 int fls(unsigned long x);
-- 
@@ -8,6 +8,7 @@
 
 #include <stdlib.h>
 #include <stdarg.h>
+#include <stdbool.h>
 
 #include "log.h"
 
@@ -216,6 +217,7 @@ char *line_read(char *buf, size_t len, int fd);
 void procfs_scan_listen(struct ctx *c, uint8_t proto, int ip_version, int ns,
 			uint8_t *map, uint8_t *exclude);
 int ns_enter(const struct ctx *c);
+bool ns_is_init(void);
 void write_pidfile(int fd, pid_t pid);
 int __daemon(int pidfile_fd, int devnull_fd);
 int fls(unsigned long x);
-- 
2.39.2

Re: [PATCH 1/3] util, conf: Add and use ns_is_init() helper

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 3865 bytes --]

On Mon, May 22, 2023 at 01:41:56AM +0200, Stefano Brivio wrote:
> We'll need this in isolate_initial(). While at it, don't rely on
> BUFSIZ: the earlier issue we had with musl reminded me it's not a
> magic "everything will fit" value. Size the read buffer to what we
> actually need from uid_map.
> 
> Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

Although...

> ---
>  conf.c | 16 +---------------
>  util.c | 25 +++++++++++++++++++++++++
>  util.h |  2 ++
>  3 files changed, 28 insertions(+), 15 deletions(-)
> 
> diff --git a/conf.c b/conf.c
> index 447b000..984c3ce 100644
> --- a/conf.c
> +++ b/conf.c
> @@ -1096,10 +1096,6 @@ static int conf_runas(char *opt, unsigned int *uid, unsigned int *gid)
>   */
>  static void conf_ugid(char *runas, uid_t *uid, gid_t *gid)
>  {
> -	const char root_uid_map[] = "         0          0 4294967295";
> -	char buf[BUFSIZ];
> -	int fd;
> -
>  	/* If user has specified --runas, that takes precedence... */
>  	if (runas) {
>  		if (conf_runas(runas, uid, gid))
> @@ -1116,18 +1112,8 @@ static void conf_ugid(char *runas, uid_t *uid, gid_t *gid)
>  		return;
>  
>  	/* ...or at least not root in the init namespace... */
> -	if ((fd = open("/proc/self/uid_map", O_RDONLY | O_CLOEXEC)) < 0) {
> -		die("Can't determine if we're in init namespace: %s",
> -		    strerror(errno));
> -	}
> -
> -	if (read(fd, buf, BUFSIZ) != sizeof(root_uid_map) ||
> -	    strncmp(buf, root_uid_map, sizeof(root_uid_map) - 1)) {
> -		close(fd);
> +	if (!ns_is_init())
>  		return;
> -	}
> -
> -	close(fd);
>  
>  	/* ...otherwise use nobody:nobody */
>  	warn("Don't run as root. Changing to nobody...");
> diff --git a/util.c b/util.c
> index c3e3471..5ec8a6c 100644
> --- a/util.c
> +++ b/util.c
> @@ -390,6 +390,31 @@ int ns_enter(const struct ctx *c)
>  	return 0;
>  }
>  
> +/**
> + * ns_is_init() - Is the caller running in the "init" user namespace?
> + *
> + * Return: true if caller is in init, false otherwise, won't return on failure
> + */
> +bool ns_is_init(void)
> +{
> +	const char root_uid_map[] = "         0          0 4294967295";
> +	char buf[sizeof(root_uid_map) + 1];
> +	bool ret = true;
> +	int fd;
> +
> +	if ((fd = open("/proc/self/uid_map", O_RDONLY | O_CLOEXEC)) < 0) {
> +		die("Can't determine if we're in init namespace: %s",
> +		    strerror(errno));
> +	}
> +
> +	if (read(fd, buf, sizeof(root_uid_map)) != sizeof(root_uid_map) ||

I don't think it can go bad in practice, but I think you want to pass
a slightly larger buffer than root_uid_map[], otherwise this test will
succeed if the uid_map contains the expected thing for init, followed
by something else.

> +	    strncmp(buf, root_uid_map, sizeof(root_uid_map) - 1))
> +		ret = false;
> +
> +	close(fd);
> +	return ret;
> +}
> +
>  /**
>   * pid_file() - Write PID to file, if requested to do so, and close it
>   * @fd:		Open PID file descriptor, closed on exit, -1 to skip writing it
> diff --git a/util.h b/util.h
> index ba3e3da..26892aa 100644
> --- a/util.h
> +++ b/util.h
> @@ -8,6 +8,7 @@
>  
>  #include <stdlib.h>
>  #include <stdarg.h>
> +#include <stdbool.h>
>  
>  #include "log.h"
>  
> @@ -216,6 +217,7 @@ char *line_read(char *buf, size_t len, int fd);
>  void procfs_scan_listen(struct ctx *c, uint8_t proto, int ip_version, int ns,
>  			uint8_t *map, uint8_t *exclude);
>  int ns_enter(const struct ctx *c);
> +bool ns_is_init(void);
>  void write_pidfile(int fd, pid_t pid);
>  int __daemon(int pidfile_fd, int devnull_fd);
>  int fls(unsigned long x);

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 1/3] util, conf: Add and use ns_is_init() helper

[Stefano Brivio]

On Mon, 22 May 2023 15:41:17 +1000
David Gibson <david@gibson.dropbear.id.au> wrote:

> On Mon, May 22, 2023 at 01:41:56AM +0200, Stefano Brivio wrote:
> > We'll need this in isolate_initial(). While at it, don't rely on
> > BUFSIZ: the earlier issue we had with musl reminded me it's not a
> > magic "everything will fit" value. Size the read buffer to what we
> > actually need from uid_map.
> > 
> > Signed-off-by: Stefano Brivio <sbrivio@redhat.com>  
> 
> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
> 
> Although...
> 
> > ---
> >  conf.c | 16 +---------------
> >  util.c | 25 +++++++++++++++++++++++++
> >  util.h |  2 ++
> >  3 files changed, 28 insertions(+), 15 deletions(-)
> > 
> > diff --git a/conf.c b/conf.c
> > index 447b000..984c3ce 100644
> > --- a/conf.c
> > +++ b/conf.c
> > @@ -1096,10 +1096,6 @@ static int conf_runas(char *opt, unsigned int *uid, unsigned int *gid)
> >   */
> >  static void conf_ugid(char *runas, uid_t *uid, gid_t *gid)
> >  {
> > -	const char root_uid_map[] = "         0          0 4294967295";
> > -	char buf[BUFSIZ];
> > -	int fd;
> > -
> >  	/* If user has specified --runas, that takes precedence... */
> >  	if (runas) {
> >  		if (conf_runas(runas, uid, gid))
> > @@ -1116,18 +1112,8 @@ static void conf_ugid(char *runas, uid_t *uid, gid_t *gid)
> >  		return;
> >  
> >  	/* ...or at least not root in the init namespace... */
> > -	if ((fd = open("/proc/self/uid_map", O_RDONLY | O_CLOEXEC)) < 0) {
> > -		die("Can't determine if we're in init namespace: %s",
> > -		    strerror(errno));
> > -	}
> > -
> > -	if (read(fd, buf, BUFSIZ) != sizeof(root_uid_map) ||
> > -	    strncmp(buf, root_uid_map, sizeof(root_uid_map) - 1)) {
> > -		close(fd);
> > +	if (!ns_is_init())
> >  		return;
> > -	}
> > -
> > -	close(fd);
> >  
> >  	/* ...otherwise use nobody:nobody */
> >  	warn("Don't run as root. Changing to nobody...");
> > diff --git a/util.c b/util.c
> > index c3e3471..5ec8a6c 100644
> > --- a/util.c
> > +++ b/util.c
> > @@ -390,6 +390,31 @@ int ns_enter(const struct ctx *c)
> >  	return 0;
> >  }
> >  
> > +/**
> > + * ns_is_init() - Is the caller running in the "init" user namespace?
> > + *
> > + * Return: true if caller is in init, false otherwise, won't return on failure
> > + */
> > +bool ns_is_init(void)
> > +{
> > +	const char root_uid_map[] = "         0          0 4294967295";
> > +	char buf[sizeof(root_uid_map) + 1];
> > +	bool ret = true;
> > +	int fd;
> > +
> > +	if ((fd = open("/proc/self/uid_map", O_RDONLY | O_CLOEXEC)) < 0) {
> > +		die("Can't determine if we're in init namespace: %s",
> > +		    strerror(errno));
> > +	}
> > +
> > +	if (read(fd, buf, sizeof(root_uid_map)) != sizeof(root_uid_map) ||  
> 
> I don't think it can go bad in practice, but I think you want to pass
> a slightly larger buffer than root_uid_map[], otherwise this test will
> succeed if the uid_map contains the expected thing for init, followed
> by something else.

Ah, yes, I thought of doing that, then I also thought that with
4294967295 as range size in the first line, there can't be other lines.
Then I (clumsily) went back to the original idea, because... go figure,
let's say it gets extended with something one day...

But now it's inconsistent (and probably confusing): the buffer is long
enough, but I don't use it:

	const char root_uid_map[] = "         0          0 4294967295";
32 bytes string length, sizeof() is 33

	char buf[sizeof(root_uid_map) + 1];
34 bytes (could be 33)

	read(fd, buf, sizeof(root_uid_map))
read at most 33 bytes

	!= sizeof(root_uid_map) ||  
isn't 33 bytes (should be 32)

	strncmp(buf, root_uid_map, sizeof(root_uid_map) - 1))
or the first 32 bytes, up to "5", differ (we should compare one more)

and, by the way, uid_map has _lines_:

$ hexdump -C /proc/self/uid_map 
00000000  20 20 20 20 20 20 20 20  20 30 20 20 20 20 20 20  |         0      |
00000010  20 20 20 20 30 20 34 32  39 34 39 36 37 32 39 35  |    0 4294967295|
00000020  0a                                                |.|

so it kind of works, but not for the reason I intended. Respinning
now...

-- 
Stefano

[PATCH 2/3] pasta: Detach mount namespace, (re)mount procfs before spawning command

[Stefano Brivio]

If we want /proc contents to be consistent after pasta spawns a child
process in a new PID namespace (only for operation without a
pre-existing namespace), we need to mount /proc after the clone(2)
call with CLONE_NEWPID, and we enable the child to do that by
passing, in the same call, the CLONE_NEWNS flag, as described by
pid_namespaces(7).

This is not really a remount: in fact, passing MS_REMOUNT to mount(2)
would make the call fail. We're in another mount namespace now, so
it's a fresh mount that has the effect of hiding the existing one.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
---
 pasta.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/pasta.c b/pasta.c
index 3a4d704..b30ce70 100644
--- a/pasta.c
+++ b/pasta.c
@@ -29,6 +29,7 @@
 #include <syslog.h>
 #include <sys/epoll.h>
 #include <sys/inotify.h>
+#include <sys/mount.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
@@ -172,6 +173,10 @@ static int pasta_spawn_cmd(void *arg)
 	const struct pasta_spawn_cmd_arg *a;
 	sigset_t set;
 
+	/* We run in a detached PID and mount namespace: mount /proc over */
+	if (mount("", "/proc", "proc", 0, NULL))
+		warn("Couldn't mount /proc: %s", strerror(errno));
+
 	if (write_file("/proc/sys/net/ipv4/ping_group_range", "0 0"))
 		warn("Cannot set ping_group_range, ICMP requests might fail");
 
@@ -243,7 +248,7 @@ void pasta_start_ns(struct ctx *c, uid_t uid, gid_t gid,
 	pasta_child_pid = do_clone(pasta_spawn_cmd, ns_fn_stack,
 				   sizeof(ns_fn_stack),
 				   CLONE_NEWIPC | CLONE_NEWPID | CLONE_NEWNET |
-				   CLONE_NEWUTS | SIGCHLD,
+				   CLONE_NEWUTS | CLONE_NEWNS  | SIGCHLD,
 				   (void *)&arg);
 
 	if (pasta_child_pid == -1) {
-- 
@@ -29,6 +29,7 @@
 #include <syslog.h>
 #include <sys/epoll.h>
 #include <sys/inotify.h>
+#include <sys/mount.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
@@ -172,6 +173,10 @@ static int pasta_spawn_cmd(void *arg)
 	const struct pasta_spawn_cmd_arg *a;
 	sigset_t set;
 
+	/* We run in a detached PID and mount namespace: mount /proc over */
+	if (mount("", "/proc", "proc", 0, NULL))
+		warn("Couldn't mount /proc: %s", strerror(errno));
+
 	if (write_file("/proc/sys/net/ipv4/ping_group_range", "0 0"))
 		warn("Cannot set ping_group_range, ICMP requests might fail");
 
@@ -243,7 +248,7 @@ void pasta_start_ns(struct ctx *c, uid_t uid, gid_t gid,
 	pasta_child_pid = do_clone(pasta_spawn_cmd, ns_fn_stack,
 				   sizeof(ns_fn_stack),
 				   CLONE_NEWIPC | CLONE_NEWPID | CLONE_NEWNET |
-				   CLONE_NEWUTS | SIGCHLD,
+				   CLONE_NEWUTS | CLONE_NEWNS  | SIGCHLD,
 				   (void *)&arg);
 
 	if (pasta_child_pid == -1) {
-- 
2.39.2

Re: [PATCH 2/3] pasta: Detach mount namespace, (re)mount procfs before spawning command

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 2181 bytes --]

On Mon, May 22, 2023 at 01:41:57AM +0200, Stefano Brivio wrote:
> If we want /proc contents to be consistent after pasta spawns a child
> process in a new PID namespace (only for operation without a
> pre-existing namespace), we need to mount /proc after the clone(2)
> call with CLONE_NEWPID, and we enable the child to do that by
> passing, in the same call, the CLONE_NEWNS flag, as described by
> pid_namespaces(7).
> 
> This is not really a remount: in fact, passing MS_REMOUNT to mount(2)
> would make the call fail. We're in another mount namespace now, so
> it's a fresh mount that has the effect of hiding the existing one.
> 
> Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

> ---
>  pasta.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/pasta.c b/pasta.c
> index 3a4d704..b30ce70 100644
> --- a/pasta.c
> +++ b/pasta.c
> @@ -29,6 +29,7 @@
>  #include <syslog.h>
>  #include <sys/epoll.h>
>  #include <sys/inotify.h>
> +#include <sys/mount.h>
>  #include <sys/types.h>
>  #include <sys/stat.h>
>  #include <fcntl.h>
> @@ -172,6 +173,10 @@ static int pasta_spawn_cmd(void *arg)
>  	const struct pasta_spawn_cmd_arg *a;
>  	sigset_t set;
>  
> +	/* We run in a detached PID and mount namespace: mount /proc over */
> +	if (mount("", "/proc", "proc", 0, NULL))
> +		warn("Couldn't mount /proc: %s", strerror(errno));
> +
>  	if (write_file("/proc/sys/net/ipv4/ping_group_range", "0 0"))
>  		warn("Cannot set ping_group_range, ICMP requests might fail");
>  
> @@ -243,7 +248,7 @@ void pasta_start_ns(struct ctx *c, uid_t uid, gid_t gid,
>  	pasta_child_pid = do_clone(pasta_spawn_cmd, ns_fn_stack,
>  				   sizeof(ns_fn_stack),
>  				   CLONE_NEWIPC | CLONE_NEWPID | CLONE_NEWNET |
> -				   CLONE_NEWUTS | SIGCHLD,
> +				   CLONE_NEWUTS | CLONE_NEWNS  | SIGCHLD,
>  				   (void *)&arg);
>  
>  	if (pasta_child_pid == -1) {

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[PATCH 3/3] isolation: Initially Keep CAP_SETFCAP if running as UID 0 in non-init

[Stefano Brivio]

If pasta spawns a child process while running as UID 0, which is only
allowed from a non-init namespace, we need to keep CAP_SETFCAP before
pasta_start_ns() is called: otherwise, starting from Linux 5.12, we
won't be able to update /proc/self/uid_map with the intended mapping
(from 0 to 0). See user_namespaces(7).

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
---
 isolation.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/isolation.c b/isolation.c
index 5f89047..19932bf 100644
--- a/isolation.c
+++ b/isolation.c
@@ -177,6 +177,8 @@ static void clamp_caps(void)
  */
 void isolate_initial(void)
 {
+	uint64_t keep;
+
 	/* We want to keep CAP_NET_BIND_SERVICE in the initial
 	 * namespace if we have it, so that we can forward low ports
 	 * into the guest/namespace
@@ -193,9 +195,18 @@ void isolate_initial(void)
 	 * further capabilites in isolate_user() and
 	 * isolate_prefork().
 	 */
-	drop_caps_ep_except(BIT(CAP_NET_BIND_SERVICE) |
-			    BIT(CAP_SETUID) | BIT(CAP_SETGID) |
-			    BIT(CAP_SYS_ADMIN) | BIT(CAP_NET_ADMIN));
+	keep = BIT(CAP_NET_BIND_SERVICE) | BIT(CAP_SETUID) | BIT(CAP_SETGID) |
+	       BIT(CAP_SYS_ADMIN) | BIT(CAP_NET_ADMIN);
+
+	/* Since Linux 5.12, if we want to update /proc/self/uid_map to create
+	 * a mapping from UID 0, which only happens with pasta spawning a child
+	 * from a non-init user namespace (pasta can't run as root), we need to
+	 * retain CAP_SETFCAP too.
+	 */
+	if (!ns_is_init() && !geteuid())
+		keep |= BIT(CAP_SETFCAP);
+
+	drop_caps_ep_except(keep);
 }
 
 /**
-- 
@@ -177,6 +177,8 @@ static void clamp_caps(void)
  */
 void isolate_initial(void)
 {
+	uint64_t keep;
+
 	/* We want to keep CAP_NET_BIND_SERVICE in the initial
 	 * namespace if we have it, so that we can forward low ports
 	 * into the guest/namespace
@@ -193,9 +195,18 @@ void isolate_initial(void)
 	 * further capabilites in isolate_user() and
 	 * isolate_prefork().
 	 */
-	drop_caps_ep_except(BIT(CAP_NET_BIND_SERVICE) |
-			    BIT(CAP_SETUID) | BIT(CAP_SETGID) |
-			    BIT(CAP_SYS_ADMIN) | BIT(CAP_NET_ADMIN));
+	keep = BIT(CAP_NET_BIND_SERVICE) | BIT(CAP_SETUID) | BIT(CAP_SETGID) |
+	       BIT(CAP_SYS_ADMIN) | BIT(CAP_NET_ADMIN);
+
+	/* Since Linux 5.12, if we want to update /proc/self/uid_map to create
+	 * a mapping from UID 0, which only happens with pasta spawning a child
+	 * from a non-init user namespace (pasta can't run as root), we need to
+	 * retain CAP_SETFCAP too.
+	 */
+	if (!ns_is_init() && !geteuid())
+		keep |= BIT(CAP_SETFCAP);
+
+	drop_caps_ep_except(keep);
 }
 
 /**
-- 
2.39.2

Re: [PATCH 3/3] isolation: Initially Keep CAP_SETFCAP if running as UID 0 in non-init

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 2033 bytes --]

On Mon, May 22, 2023 at 01:41:58AM +0200, Stefano Brivio wrote:
> If pasta spawns a child process while running as UID 0, which is only
> allowed from a non-init namespace, we need to keep CAP_SETFCAP before
> pasta_start_ns() is called: otherwise, starting from Linux 5.12, we
> won't be able to update /proc/self/uid_map with the intended mapping
> (from 0 to 0). See user_namespaces(7).
> 
> Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

> ---
>  isolation.c | 17 ++++++++++++++---
>  1 file changed, 14 insertions(+), 3 deletions(-)
> 
> diff --git a/isolation.c b/isolation.c
> index 5f89047..19932bf 100644
> --- a/isolation.c
> +++ b/isolation.c
> @@ -177,6 +177,8 @@ static void clamp_caps(void)
>   */
>  void isolate_initial(void)
>  {
> +	uint64_t keep;
> +
>  	/* We want to keep CAP_NET_BIND_SERVICE in the initial
>  	 * namespace if we have it, so that we can forward low ports
>  	 * into the guest/namespace
> @@ -193,9 +195,18 @@ void isolate_initial(void)
>  	 * further capabilites in isolate_user() and
>  	 * isolate_prefork().
>  	 */
> -	drop_caps_ep_except(BIT(CAP_NET_BIND_SERVICE) |
> -			    BIT(CAP_SETUID) | BIT(CAP_SETGID) |
> -			    BIT(CAP_SYS_ADMIN) | BIT(CAP_NET_ADMIN));
> +	keep = BIT(CAP_NET_BIND_SERVICE) | BIT(CAP_SETUID) | BIT(CAP_SETGID) |
> +	       BIT(CAP_SYS_ADMIN) | BIT(CAP_NET_ADMIN);
> +
> +	/* Since Linux 5.12, if we want to update /proc/self/uid_map to create
> +	 * a mapping from UID 0, which only happens with pasta spawning a child
> +	 * from a non-init user namespace (pasta can't run as root), we need to
> +	 * retain CAP_SETFCAP too.
> +	 */
> +	if (!ns_is_init() && !geteuid())
> +		keep |= BIT(CAP_SETFCAP);
> +
> +	drop_caps_ep_except(keep);
>  }
>  
>  /**

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]