[v2,07/36] ns: Set nested namespaces hookups

Submitted by Kirill Tkhai on Feb. 7, 2017, 8:22 a.m.

Details

Message ID f6ac6935-d23d-bd77-336f-4414112f1c31@virtuozzo.com
State New
Series "Nested user namespaces support"
Headers show

Commit Message

Kirill Tkhai Feb. 7, 2017, 8:22 a.m.
On 07.02.2017 04:45, Andrei Vagin wrote:
> On Fri, Feb 03, 2017 at 07:12:40PM +0300, Kirill Tkhai wrote:
>> Introduce ns_id::parent and assign a pointer to parent
>> for every ns except NS_CRIU and NS_ROOT.
>> Also populate user_ns for pid_ns.
>>
>> Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
>> ---
>>  criu/include/namespaces.h |   12 +++++++
>>  criu/namespaces.c         |   81 +++++++++++++++++++++++++++++++++++++++++++++
>>  2 files changed, 93 insertions(+)
>>
>> diff --git a/criu/include/namespaces.h b/criu/include/namespaces.h
>> index c9f3e8a74..e1f0fa79f 100644
>> --- a/criu/include/namespaces.h
>> +++ b/criu/include/namespaces.h
>> @@ -1,6 +1,8 @@
>>  #ifndef __CR_NS_H__
>>  #define __CR_NS_H__
>>  
>> +#include <sys/ioctl.h>
>> +
>>  #include "common/compiler.h"
>>  #include "files.h"
>>  #include "common/list.h"
>> @@ -39,6 +41,12 @@
>>  #define CLONE_SUBNS	(CLONE_NEWNS)
>>  #define EXTRA_SIZE	20
>>  
>> +#ifndef NSIO
>> +#define NSIO    0xb7
>> +#define NS_GET_USERNS   _IO(NSIO, 0x1)
>> +#define NS_GET_PARENT   _IO(NSIO, 0x2)
>> +#endif
>> +
>>  struct ns_desc {
>>  	unsigned int	cflag;
>>  	char		*str;
>> @@ -75,6 +83,10 @@ struct ns_id {
>>  	unsigned int id;
>>  	pid_t ns_pid;
>>  	struct ns_desc *nd;
>> +	struct ns_id *parent;
>> +	struct list_head children;
>> +	struct list_head siblings;
>> +	struct ns_id *user_ns;
>>  	struct ns_id *next;
>>  	enum ns_type type;
>>  
>> diff --git a/criu/namespaces.c b/criu/namespaces.c
>> index f655c164e..12a5408e2 100644
>> --- a/criu/namespaces.c
>> +++ b/criu/namespaces.c
>> @@ -13,6 +13,7 @@
>>  #include <sys/stat.h>
>>  #include <limits.h>
>>  #include <errno.h>
>> +#include <sys/ioctl.h>
>>  
>>  #include "page.h"
>>  #include "rst-malloc.h"
>> @@ -302,6 +303,8 @@ struct ns_id *rst_new_ns_id(unsigned int id, pid_t pid,
>>  		nsid->type = type;
>>  		nsid_add(nsid, nd, id, pid);
>>  		nsid->ns_populated = false;
>> +		INIT_LIST_HEAD(&nsid->children);
>> +		INIT_LIST_HEAD(&nsid->siblings);
>>  	}
>>  
>>  	return nsid;
>> @@ -421,6 +424,8 @@ static unsigned int generate_ns_id(int pid, unsigned int kid, struct ns_desc *nd
>>  	nsid->type = type;
>>  	nsid->kid = kid;
>>  	nsid->ns_populated = true;
>> +	INIT_LIST_HEAD(&nsid->children);
>> +	INIT_LIST_HEAD(&nsid->siblings);
>>  	nsid_add(nsid, nd, ns_next_id++, pid);
>>  
>>  found:
>> @@ -661,6 +666,82 @@ int dump_task_ns_ids(struct pstree_item *item)
>>  	return 0;
>>  }
>>  
>> +static int set_ns_opt(int ns_fd, unsigned ioc, struct ns_id **ns, struct ns_desc *nd)
>> +{
>> +	int opt_fd, ret = -1;
>> +	struct stat st;
>> +
>> +	opt_fd = ioctl(ns_fd, ioc);
>> +	if (opt_fd < 0) {
>> +		pr_info("Can't do ns ioctl %x\n", ioc);
>> +		return -errno;
>> +	}
>> +	if (fstat(opt_fd, &st) < 0) {
>> +		pr_perror("Unable to stat on ns fd");
>> +		ret = -errno;
>> +		goto out;
>> +	}
>> +	*ns = lookup_ns_by_kid(st.st_ino, nd);
>> +	if (!*ns) {
>> +		pr_err("Namespaces hierarhies with holes are not supported\n");
>> +		ret = -EINVAL;
>> +		goto out;
>> +	}
>> +	ret = 0;
>> +out:
>> +	close(opt_fd);
>> +	return ret;
>> +}
>> +
>> +static int set_ns_hookups(struct ns_id *ns)
>> +{
>> +	struct ns_desc *nd = ns->nd;
>> +	struct ns_id *u_ns;
>> +	int fd, ret = -1;
>> +
>> +	if (ns->type == NS_CRIU)
>> +		return 0;
>> +
>> +	fd = open_proc(ns->ns_pid, "ns/%s", nd->str);
>> +	if (fd < 0) {
>> +		pr_perror("Can't open %s, pid %d", nd->str, ns->ns_pid);
>> +		return -1;
>> +	}
>> +
>> +	if (ns->type != NS_ROOT && (nd == &user_ns_desc || nd == &pid_ns_desc)) {
>> +		if (set_ns_opt(fd, NS_GET_PARENT, &ns->parent, nd) < 0)
>> +			goto out;
>> +		list_add(&ns->siblings, &ns->parent->children);
>> +	}
>> +
>> +	if (nd != &user_ns_desc) {
>> +		ret = set_ns_opt(fd, NS_GET_USERNS, &ns->user_ns, &user_ns_desc);
> 
> I think we should not dump more than one userns, if NS_GET_USERNS isn't
> supported by this kernel.

Yeah, exactly yesterday worked about that :)

commit 1db8cb2cfc3184f3a22830cdfe85f6efa749300c
Author: Kirill Tkhai <ktkhai@virtuozzo.com>
Date:   Mon Feb 6 13:33:16 2017 +0300

    ns: Set nested namespaces hookups
    
    Introduce ns_id::parent and assign a pointer to parent
    for every ns except NS_CRIU and NS_ROOT.
    Also populate user_ns for pid_ns.
    
    Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>

Patch hide | download patch | download mbox

diff --git a/criu/include/namespaces.h b/criu/include/namespaces.h
index c9f3e8a74..e1f0fa79f 100644
--- a/criu/include/namespaces.h
+++ b/criu/include/namespaces.h
@@ -1,6 +1,8 @@ 
 #ifndef __CR_NS_H__
 #define __CR_NS_H__
 
+#include <sys/ioctl.h>
+
 #include "common/compiler.h"
 #include "files.h"
 #include "common/list.h"
@@ -39,6 +41,12 @@ 
 #define CLONE_SUBNS	(CLONE_NEWNS)
 #define EXTRA_SIZE	20
 
+#ifndef NSIO
+#define NSIO    0xb7
+#define NS_GET_USERNS   _IO(NSIO, 0x1)
+#define NS_GET_PARENT   _IO(NSIO, 0x2)
+#endif
+
 struct ns_desc {
 	unsigned int	cflag;
 	char		*str;
@@ -75,6 +83,10 @@  struct ns_id {
 	unsigned int id;
 	pid_t ns_pid;
 	struct ns_desc *nd;
+	struct ns_id *parent;
+	struct list_head children;
+	struct list_head siblings;
+	struct ns_id *user_ns;
 	struct ns_id *next;
 	enum ns_type type;
 
diff --git a/criu/namespaces.c b/criu/namespaces.c
index b1cca989f..7e9917c74 100644
--- a/criu/namespaces.c
+++ b/criu/namespaces.c
@@ -13,6 +13,7 @@ 
 #include <sys/stat.h>
 #include <limits.h>
 #include <errno.h>
+#include <sys/ioctl.h>
 
 #include "page.h"
 #include "rst-malloc.h"
@@ -302,6 +303,8 @@  struct ns_id *rst_new_ns_id(unsigned int id, pid_t pid,
 		nsid->type = type;
 		nsid_add(nsid, nd, id, pid);
 		nsid->ns_populated = false;
+		INIT_LIST_HEAD(&nsid->children);
+		INIT_LIST_HEAD(&nsid->siblings);
 	}
 
 	return nsid;
@@ -421,6 +424,8 @@  static unsigned int generate_ns_id(int pid, unsigned int kid, struct ns_desc *nd
 	nsid->type = type;
 	nsid->kid = kid;
 	nsid->ns_populated = true;
+	INIT_LIST_HEAD(&nsid->children);
+	INIT_LIST_HEAD(&nsid->siblings);
 	nsid_add(nsid, nd, ns_next_id++, pid);
 
 found:
@@ -661,6 +666,83 @@  int dump_task_ns_ids(struct pstree_item *item)
 	return 0;
 }
 
+static int set_ns_opt(int ns_fd, unsigned ioc, struct ns_id **ns, struct ns_desc *nd)
+{
+	int opt_fd, ret = -1;
+	struct stat st;
+
+	opt_fd = ioctl(ns_fd, ioc);
+	if (opt_fd < 0) {
+		pr_info("Can't do ns ioctl %x\n", ioc);
+		return -errno;
+	}
+	if (fstat(opt_fd, &st) < 0) {
+		pr_perror("Unable to stat on ns fd");
+		ret = -errno;
+		goto out;
+	}
+	*ns = lookup_ns_by_kid(st.st_ino, nd);
+	if (!*ns) {
+		pr_err("Namespaces hierarhies with holes are not supported\n");
+		ret = -EINVAL;
+		goto out;
+	}
+	ret = 0;
+out:
+	close(opt_fd);
+	return ret;
+}
+
+static int set_ns_hookups(struct ns_id *ns)
+{
+	struct ns_desc *nd = ns->nd;
+	struct ns_id *u_ns;
+	int fd, ret = -1;
+
+	fd = open_proc(ns->ns_pid, "ns/%s", nd->str);
+	if (fd < 0) {
+		pr_perror("Can't open %s, pid %d", nd->str, ns->ns_pid);
+		return -1;
+	}
+
+	if (ns->type != NS_ROOT && (nd == &pid_ns_desc || nd == &user_ns_desc)) {
+		if (set_ns_opt(fd, NS_GET_PARENT, &ns->parent, nd) < 0)
+			goto out;
+		if (ns->parent && ns->parent->type == NS_CRIU) {
+			pr_err("Wrong determined NS_ROOT, or root_item has NS_OTHER user_ns\n");
+			goto out;
+		}
+		list_add(&ns->siblings, &ns->parent->children);
+	}
+
+	if (nd != &user_ns_desc) {
+		ret = set_ns_opt(fd, NS_GET_USERNS, &ns->user_ns, &user_ns_desc);
+		if (ret == -ENOTTY) {
+			pr_info("Can't get user_ns of %s %u, assuming NS_ROOT\n",
+					nd->str, ns->id);
+			for (u_ns = ns_ids; u_ns; u_ns = u_ns->next) {
+				/*
+				 * Find any user_ns. If there is a NS_OTHER one,
+				 * we'll fail when we search its parent.
+				 */
+				if (u_ns->nd == &user_ns_desc &&
+				    (!(root_ns_mask & CLONE_NEWUSER) || u_ns->type == NS_ROOT))
+					break;
+			}
+			if (!u_ns) {
+				pr_err("Can't find root user_ns\n");
+				goto out;
+			}
+			ns->user_ns = u_ns;
+		} else if (ret < 0)
+			goto out;
+	}
+	ret = 0;
+out:
+	close(fd);
+	return ret;
+}
+
 static UsernsEntry userns_entry = USERNS_ENTRY__INIT;
 #define INVALID_ID (~0U)