[CRIU] [PATCH 08/11 v2] net: allow to dump and restore more than one network namespace

Dmitry Safonov 0x7f454c46 at gmail.com
Wed Feb 8 04:06:56 PST 2017


2017-02-08 12:55 GMT+03:00 Pavel Emelyanov <xemul at virtuozzo.com>:
> On 02/07/2017 10:06 PM, Andrei Vagin wrote:
>> From: Andrei Vagin <avagin at virtuozzo.com>
>>
>> Restore all network namespaces from the root task and then set
>> a proper namespace for each task after restoring sockets, because
>> we need to switch network namespaces to restore sockets.
>>
>> Each socket has to be created in a proper network namespace.
>>
>> v2: fix a typo bug
>>
>> Signed-off-by: Andrei Vagin <avagin at virtuozzo.com>
>> ---
>>  criu/cr-restore.c         |  21 ++++++----
>>  criu/include/namespaces.h |   4 +-
>>  criu/include/net.h        |   6 ++-
>>  criu/namespaces.c         |   6 +--
>>  criu/net.c                | 104 +++++++++++++++++++++++++++++++++++++++++++++-
>>  criu/pstree.c             |   6 +++
>>  6 files changed, 133 insertions(+), 14 deletions(-)
>>
>> diff --git a/criu/cr-restore.c b/criu/cr-restore.c
>> index 681655d..288add3 100644
>> --- a/criu/cr-restore.c
>> +++ b/criu/cr-restore.c
>> @@ -713,6 +713,13 @@ static int restore_one_alive_task(int pid, CoreEntry *core)
>>       if (prepare_vmas(current, ta))
>>               return -1;
>>
>> +     /*
>> +      * Sockets have to be restored in their network namespaces,
>> +      * so a task namespace has to be restored after sockets.
>> +      */
>> +     if (restore_task_net_ns(current))
>> +             return -1;
>> +
>>       if (setup_uffd(pid, ta))
>>               return -1;
>>
>> @@ -1388,14 +1395,6 @@ static int restore_task_with_children(void *_arg)
>>       if (ret < 0)
>>               goto err;
>>
>> -     if (ca->clone_flags & CLONE_NEWNET) {
>> -             ret = unshare(CLONE_NEWNET);
>> -             if (ret) {
>> -                     pr_perror("Can't unshare net-namespace");
>> -                     goto err;
>> -             }
>> -     }
>> -
>>       if (!(ca->clone_flags & CLONE_FILES)) {
>>               ret = close_old_fds();
>>               if (ret)
>> @@ -2965,6 +2964,12 @@ static int sigreturn_restore(pid_t pid, struct task_restore_args *task_args, uns
>>       if (rst_prep_creds(pid, core, &creds_pos))
>>               goto err_nv;
>>
>> +     if (current->parent == NULL) {
>> +             /* Wait when all tasks restored all files */
>> +             restore_wait_other_tasks();
>> +             fini_net_namespaces();
>> +     }
>> +
>>       /*
>>        * We're about to search for free VM area and inject the restorer blob
>>        * into it. No irrelevant mmaps/mremaps beyond this point, otherwise
>> diff --git a/criu/include/namespaces.h b/criu/include/namespaces.h
>> index 18eafb2..522c098 100644
>> --- a/criu/include/namespaces.h
>> +++ b/criu/include/namespaces.h
>> @@ -36,7 +36,8 @@
>>  #define CLONE_ALLNS  (CLONE_NEWPID | CLONE_NEWNET | CLONE_NEWIPC | CLONE_NEWUTS | CLONE_NEWNS | CLONE_NEWUSER | CLONE_NEWCGROUP)
>>
>>  /* Nested namespaces are supported only for these types */
>> -#define CLONE_SUBNS  (CLONE_NEWNS)
>> +#define CLONE_SUBNS  (CLONE_NEWNS | CLONE_NEWNET)
>> +
>>  #define EXTRA_SIZE   20
>>
>>  struct ns_desc {
>> @@ -95,6 +96,7 @@ struct ns_id {
>>               } mnt;
>>
>>               struct {
>> +                     int ns_fd;      /* a file handle for the namespace */
>>                       int nlsk;       /* for sockets collection */
>>                       int seqsk;      /* to talk to parasite daemons */
>>               } net;
>> diff --git a/criu/include/net.h b/criu/include/net.h
>> index deac65f..49eca36 100644
>> --- a/criu/include/net.h
>> +++ b/criu/include/net.h
>> @@ -12,9 +12,13 @@
>>
>>  struct cr_imgset;
>>  extern int dump_net_ns(int ns_id);
>> -extern int prepare_net_ns(int pid);
>> +extern int prepare_net_namespaces(void);
>> +extern void fini_net_namespaces(void);
>>  extern int netns_keep_nsfd(void);
>>
>> +struct pstree_item;
>> +extern int restore_task_net_ns(struct pstree_item *current);
>> +
>>  struct veth_pair {
>>       struct list_head node;
>>       char *inside;
>> diff --git a/criu/namespaces.c b/criu/namespaces.c
>> index b1cca98..dce8ab5 100644
>> --- a/criu/namespaces.c
>> +++ b/criu/namespaces.c
>> @@ -1660,9 +1660,6 @@ int prepare_namespace(struct pstree_item *item, unsigned long clone_flags)
>>        * tree (i.e. -- mnt_ns restoring)
>>        */
>>
>> -     id = ns_per_id ? item->ids->net_ns_id : pid;
>> -     if ((clone_flags & CLONE_NEWNET) && prepare_net_ns(id))
>> -             return -1;
>>       id = ns_per_id ? item->ids->uts_ns_id : pid;
>>       if ((clone_flags & CLONE_NEWUTS) && prepare_utsns(id))
>>               return -1;
>> @@ -1670,6 +1667,9 @@ int prepare_namespace(struct pstree_item *item, unsigned long clone_flags)
>>       if ((clone_flags & CLONE_NEWIPC) && prepare_ipc_ns(id))
>>               return -1;
>>
>> +     if (prepare_net_namespaces())
>> +             return -1;
>> +
>>       /*
>>        * This one is special -- there can be several mount
>>        * namespaces and prepare_mnt_ns handles them itself.
>> diff --git a/criu/net.c b/criu/net.c
>> index 7dadc8d..a7e0a6e 100644
>> --- a/criu/net.c
>> +++ b/criu/net.c
>> @@ -1670,7 +1670,7 @@ int dump_net_ns(int ns_id)
>>       return ret;
>>  }
>>
>> -int prepare_net_ns(int pid)
>> +static int prepare_net_ns(int pid)
>>  {
>>       int ret = 0;
>>       NetnsEntry *netns = NULL;
>> @@ -1702,6 +1702,108 @@ int prepare_net_ns(int pid)
>>       return ret;
>>  }
>>
>> +static int open_net_ns(struct ns_id *nsid, struct rst_info *rst)
>> +{
>> +     int fd, tfd;
>> +
>> +     /* Pin one with a file descriptor */
>> +     fd = open_proc(PROC_SELF, "ns/net");
>> +     if (fd < 0)
>> +             return -1;
>> +     tfd = reopen_as_unused_fd(fd, rst);
>> +     if (tfd < 0) {
>> +             close(fd);
>> +             return -1;
>> +     }
>> +     nsid->net.ns_fd = tfd;
>> +
>> +     return 0;
>> +}
>> +
>> +int prepare_net_namespaces()
>> +{
>> +     struct ns_id *nsid;
>> +
>> +     if (!(root_ns_mask & CLONE_NEWNET))
>> +             return 0;
>> +
>> +     for (nsid = ns_ids; nsid != NULL; nsid = nsid->next) {
>> +             if (nsid->nd != &net_ns_desc)
>> +                     continue;
>> +
>> +             if (unshare(CLONE_NEWNET)) {
>> +                     pr_perror("Unable to create a new netns");
>> +                     goto err;
>> +             }
>> +
>> +             if (prepare_net_ns(nsid->id))
>> +                     goto err;
>> +
>> +             if (open_net_ns(nsid, rsti(root_item)))
>> +                     goto err;
>> +     }
>> +
>> +     return 0;
>> +err:
>> +     return -1;
>> +}
>> +
>> +void fini_net_namespaces()
>> +{
>> +     struct ns_id *nsid;
>> +
>> +     if (!(root_ns_mask & CLONE_NEWNET))
>> +             return;
>> +
>> +     for (nsid = ns_ids; nsid != NULL; nsid = nsid->next) {
>> +             if (nsid->nd != &net_ns_desc)
>> +                     continue;
>> +             close_safe(&nsid->net.ns_fd);
>> +     }
>> +}
>> +
>> +static int do_restore_task_net_ns(struct ns_id *nsid, struct pstree_item *current)
>
> From here to ...
>
>> +{
>> +     int fd;
>> +
>> +     if (!(root_ns_mask & CLONE_NEWNET))
>> +             return 0;
>> +
>> +     fd = open_proc(root_item->pid->ns[0].virt, "fd/%d", nsid->net.ns_fd);
>> +     if (fd < 0)
>> +             return -1;
>> +
>> +     if (setns(fd, CLONE_NEWNET)) {
>> +             pr_perror("Can't restore netns");
>> +             close(fd);
>> +             return -1;
>> +     }
>> +     close(fd);
>> +
>> +     return 0;
>> +}
>> +
>> +int restore_task_net_ns(struct pstree_item *current)
>> +{
>> +     if (current->ids && current->ids->has_net_ns_id) {
>> +             unsigned int id = current->ids->net_ns_id;
>> +             struct ns_id *nsid;
>> +
>> +             nsid = lookup_ns_by_id(id, &net_ns_desc);
>> +             if (nsid == NULL) {
>> +                     pr_err("Can't find mount namespace %d\n", id);
>> +                     return -1;
>> +             }
>> +
>> +             BUG_ON(nsid->type == NS_CRIU);
>> +
>> +             if (do_restore_task_net_ns(nsid, current))
>> +                     return -1;
>
> ... here. We already have very similar code in restore_task_mnt_ns(). Can
> we merge them?

Yeah, I noticed this while writing sub-UTS C/R, here is my attempt to
generalize (on top of this):
https://github.com/0x7f454c46/criu/commit/1d88b490a78c7b141b259eec030f4a258b272996
I used it in C/R of nested uts, but didn't touch mnt.
So, I suggest to do it on top with my patch, if it's suitable - as I've
already wrote the code and refactored this.

>
>> +     }
>> +
>> +     return 0;
>> +}
>> +
>>  int netns_keep_nsfd(void)
>>  {
>>       int ns_fd, ret;
>> diff --git a/criu/pstree.c b/criu/pstree.c
>> index 833b3d0..6404350 100644
>> --- a/criu/pstree.c
>> +++ b/criu/pstree.c
>> @@ -14,6 +14,8 @@
>>  #include "mount.h"
>>  #include "dump.h"
>>  #include "util.h"
>> +#include "net.h"
>> +
>>  #include "protobuf.h"
>>  #include "images/pstree.pb-c.h"
>>  #include "crtools.h"
>> @@ -473,6 +475,10 @@ static int read_pstree_ids(struct pstree_item *pi)
>>               if (rst_add_ns_id(pi->ids->mnt_ns_id, pi, &mnt_ns_desc))
>>                       return -1;
>>       }
>> +     if (pi->ids->has_net_ns_id) {
>> +             if (rst_add_ns_id(pi->ids->net_ns_id, pi, &net_ns_desc))
>> +                     return -1;
>> +     }
>>
>>       return 0;
>>  }
>>
>
> _______________________________________________
> CRIU mailing list
> CRIU at openvz.org
> https://lists.openvz.org/mailman/listinfo/criu



-- 
             Dmitry


More information about the CRIU mailing list