2
0
mirror of https://github.com/checkpoint-restore/criu synced 2025-08-22 09:58:09 +00:00
criu/sockets.c

669 lines
13 KiB
C
Raw Normal View History

Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
#include <sys/socket.h>
#include <linux/netlink.h>
#include <linux/types.h>
#include <linux/net.h>
#include <linux/un.h>
#include <sys/types.h>
#include <sys/vfs.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <netinet/tcp.h>
#include <errno.h>
#include <unistd.h>
#include <limits.h>
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
#include "types.h"
#include "libnetlink.h"
#include "sockets.h"
#include "unix_diag.h"
#include "image.h"
#include "crtools.h"
#include "util.h"
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
static char buf[4096];
#ifndef NETLINK_SOCK_DIAG
#define NETLINK_SOCK_DIAG NETLINK_INET_DIAG
#endif
#ifndef SOCK_DIAG_BY_FAMILY
#define SOCK_DIAG_BY_FAMILY 20
#endif
#ifndef SOCKFS_MAGIC
#define SOCKFS_MAGIC 0x534F434B
#endif
struct socket_desc {
unsigned int family;
unsigned int ino;
struct socket_desc *next;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
};
struct unix_sk_desc {
struct socket_desc sd;
unsigned int type;
unsigned int state;
unsigned int peer_ino;
unsigned int rqlen;
unsigned int namelen;
char *name;
unsigned int *icons;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
};
#define SK_HASH_SIZE 32
static struct socket_desc *sockets[SK_HASH_SIZE];
static struct socket_desc *lookup_socket(int ino)
{
struct socket_desc *d;
for (d = sockets[ino % SK_HASH_SIZE]; d; d = d->next) {
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (d->ino == ino)
break;
}
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return d;
}
static int sk_collect_one(int ino, int family, struct socket_desc *d)
{
d->ino = ino;
d->family = family;
d->next = sockets[ino % SK_HASH_SIZE];
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
sockets[ino % SK_HASH_SIZE] = d;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return 0;
}
static void show_one_unix(char *act, struct unix_sk_desc *sk)
{
dprintk("\t%s: ino %d type %d state %d name %s\n",
act, sk->sd.ino, sk->type, sk->state, sk->name);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
}
static void show_one_unix_img(char *act, struct unix_sk_entry *e)
{
dprintk("\t%s: fd %d type %d state %d name %d bytes\n",
act, e->fd, e->type, e->state, e->namelen);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
}
static int can_dump_unix_sk(struct unix_sk_desc *sk)
{
if (sk->type != SOCK_STREAM) {
/*
* Dgram sockets connect and accept jobs at
* restore time should be fixed
*/
pr_err("Only stream sockets for now\n");
return 0;
}
switch (sk->state) {
case TCP_LISTEN:
if (sk->rqlen != 0) {
/*
* Currently the ICONS nla reports the conn
* requests for listen sockets. Need to pick
* those up and fix the connect job respectively
*/
pr_err("In-flight connection (l)\n");
return 0;
}
break;
case TCP_ESTABLISHED:
if (!sk->peer_ino) {
/*
* Read above
*/
pr_err("In-flight connection\n");
return 0;
}
if (sk->rqlen) {
/*
* The hard case :( Currentl there's no way to
* clone the sk queue. Even the MSG_PEEK doesn't
* help, since it picks up the head of the queue
* always. Some more patches should go
*/
pr_err("Non empty queue\n");
return 0;
}
break;
default:
pr_err("Unknown state %d\n", sk->state);
return 0;
}
return 1;
}
#define USK_DEF_BACKLOG 16
static int dump_one_unix(struct socket_desc *_sk, char *fd, struct cr_fdset *cr_fdset)
{
struct unix_sk_desc *sk = (struct unix_sk_desc *)_sk;
struct unix_sk_entry ue;
if (!can_dump_unix_sk(sk))
goto err;
ue.fd = atoi(fd);
ue.id = sk->sd.ino;
ue.type = sk->type;
ue.state = sk->state;
ue.namelen = sk->namelen;
ue.backlog = USK_DEF_BACKLOG; /* FIXME */
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
ue.pad = 0;
ue.peer = sk->peer_ino;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
write_ptr_safe(cr_fdset->desc[CR_FD_UNIXSK].fd, &ue, err);
write_safe(cr_fdset->desc[CR_FD_UNIXSK].fd, sk->name, ue.namelen, err);
pr_info("Dumping unix socket at %s\n", fd);
show_one_unix("Dumping", sk);
show_one_unix_img("Dumped", &ue);
return 0;
err:
return -1;
}
int try_dump_socket(char *dir, char *fd, struct cr_fdset *cr_fdset)
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
{
struct socket_desc *sk;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
struct statfs fst;
struct stat st;
snprintf(buf, sizeof(buf), "%s/%s", dir, fd);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (statfs(buf, &fst)) {
pr_err("Can't statfs %s\n", buf);
return -1;
}
if (stat(buf, &st)) {
pr_err("Can't stat %s\n", buf);
return -1;
}
if (fst.f_type != SOCKFS_MAGIC)
return 1; /* not a socket, proceed with caller error */
sk = lookup_socket(st.st_ino);
if (!sk) {
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
pr_err("Uncollected socket %d\n", st.st_ino);
return -1;
}
switch (sk->family) {
case AF_UNIX:
return dump_one_unix(sk, fd, cr_fdset);
default:
pr_err("BUG! Unknown socket collected\n");
return -1;
}
}
static int unix_collect_one(struct unix_diag_msg *m, struct rtattr **tb)
{
struct unix_sk_desc *d, **h;
d = xzalloc(sizeof(*d));
if (!d)
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return -1;
d->type = m->udiag_type;
d->state= m->udiag_state;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (tb[UNIX_DIAG_PEER])
d->peer_ino = *(int *)RTA_DATA(tb[UNIX_DIAG_PEER]);
if (tb[UNIX_DIAG_NAME]) {
int len = RTA_PAYLOAD(tb[UNIX_DIAG_NAME]);
char *name = xmalloc(len + 1);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (!name)
goto err;
memcpy(name, RTA_DATA(tb[UNIX_DIAG_NAME]), len);
name[len] = '\0';
if (name[0] != '\0' && d->state == TCP_LISTEN) {
struct unix_diag_vfs *uv;
struct stat st;
if (name[0] != '/') {
pr_err("Relative bind path unsupported\n");
goto err;
}
if (!tb[UNIX_DIAG_VFS]) {
pr_err("Bound socket w/o inode %d\n",
m->udiag_ino);
goto err;
}
uv = RTA_DATA(tb[UNIX_DIAG_VFS]);
if (stat(name, &st)) {
pr_perror("Can't stat socket %d(%s)\n",
m->udiag_ino, name);
goto err;
}
if ((st.st_ino != uv->udiag_vfs_ino) ||
(st.st_dev == uv->udiag_vfs_dev)) {
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
/*
* When a listen socket is bound to
* unlinked file, we just drop his name,
* since noone will access it via one.
*/
xfree(name);
len = 0;
name = NULL;
}
}
d->namelen = len;
d->name = name;
}
if (tb[UNIX_DIAG_ICONS]) {
int len = RTA_PAYLOAD(tb[UNIX_DIAG_ICONS]);
d->icons = xmalloc(len + sizeof(u32));
if (!d->icons)
goto err;
memcpy(d->icons, RTA_DATA(tb[UNIX_DIAG_ICONS]), len);
d->icons[len / sizeof(u32)] = 0;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
}
if (tb[UNIX_DIAG_RQLEN])
d->rqlen = *(int *)RTA_DATA(tb[UNIX_DIAG_RQLEN]);
show_one_unix("Collected", d);
return sk_collect_one(m->udiag_ino, AF_UNIX, &d->sd);
err:
xfree(d->icons);
xfree(d->name);
xfree(d);
return -1;
}
static int unix_receive_one(struct nlmsghdr *h)
{
struct unix_diag_msg *m = NLMSG_DATA(h);
struct rtattr *tb[UNIX_DIAG_MAX+1];
parse_rtattr(tb, UNIX_DIAG_MAX, (struct rtattr *)(m + 1),
h->nlmsg_len - NLMSG_LENGTH(sizeof(*m)));
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return unix_collect_one(m, tb);
}
static int collect_unix_sockets(int nl)
{
struct msghdr msg;
struct sockaddr_nl nladdr;
struct iovec iov;
struct {
struct nlmsghdr hdr;
struct unix_diag_req r;
} req;
memset(&msg, 0, sizeof(msg));
msg.msg_name = &nladdr;
msg.msg_namelen = sizeof(nladdr);
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
memset(&nladdr, 0, sizeof(nladdr));
nladdr.nl_family= AF_NETLINK;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
iov.iov_base = &req;
iov.iov_len = sizeof(req);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
memset(&req, 0, sizeof(req));
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = SOCK_DIAG_BY_FAMILY;
req.hdr.nlmsg_flags = NLM_F_ROOT | NLM_F_MATCH | NLM_F_REQUEST;
req.hdr.nlmsg_seq = CR_NLMSG_SEQ;
req.r.sdiag_family = AF_UNIX;
req.r.udiag_states = -1; /* All */
req.r.udiag_show = UDIAG_SHOW_NAME | UDIAG_SHOW_VFS | UDIAG_SHOW_PEER |
UDIAG_SHOW_ICONS | UDIAG_SHOW_RQLEN;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (sendmsg(nl, &msg, 0) < 0) {
pr_perror("Can't send request message\n");
goto err;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
}
iov.iov_base = buf;
iov.iov_len = sizeof(buf);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
while (1) {
int err;
memset(&msg, 0, sizeof(msg));
msg.msg_name = &nladdr;
msg.msg_namelen = sizeof(nladdr);
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
err = recvmsg(nl, &msg, 0);
if (err < 0) {
if (errno == EINTR)
continue;
else
goto err;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
}
if (err == 0)
break;
err = nlmsg_receive(buf, err, unix_receive_one);
if (err < 0)
goto err;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (err == 0)
break;
}
return 0;
err:
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return -1;
}
int collect_sockets(void)
{
int err;
int nl;
nl = socket(PF_NETLINK, SOCK_RAW, NETLINK_SOCK_DIAG);
if (nl < 0) {
pr_err("Can't create sock diag socket\n");
return -1;
}
err = collect_unix_sockets(nl);
close(nl);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return err;
}
struct unix_conn_job {
struct sockaddr_un addr;
int addrlen;
int fd;
struct unix_conn_job *next;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
};
static void unix_show_job(char *type, int fd, int id)
{
dprintk("%s job fd %d id %d\n", type, fd, id);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
}
static struct unix_conn_job *conn_jobs;
static int run_connect_jobs(void)
{
struct unix_conn_job *cj, *next;
cj = conn_jobs;
while (cj) {
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
int attempts = 8;
unix_show_job("Run conn", cj->fd, -1);
try_again:
if (connect(cj->fd, (struct sockaddr *)&cj->addr, cj->addrlen) < 0) {
if (attempts) {
usleep(1000);
attempts--;
goto try_again; /* FIXME - use avagin@'s waiters */
}
pr_perror("Can't restore connection (c)\n");
return -1;
}
unix_show_job("Fin conn", cj->fd, -1);
next = cj->next;
xfree(cj);
cj = next;
}
return 0;
}
struct unix_accept_job {
int fd;
struct unix_accept_job *next;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
};
static struct unix_accept_job *accept_jobs;
static int run_accept_jobs(void)
{
struct unix_accept_job *aj, *next;
aj = accept_jobs;
while (aj) {
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
int fd;
unix_show_job("Run acc", aj->fd, -1);
fd = accept(aj->fd, NULL, NULL);
if (fd < 0) {
pr_perror("Can't restore connection (s)\n");
return -1;
}
if (reopen_fd_as(aj->fd, fd))
return -1;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
unix_show_job("Fin acc", aj->fd, -1);
next = aj->next;
xfree(aj);
aj = next;
}
return 0;
}
static void prep_conn_addr(int id, struct sockaddr_un *addr, int *addrlen)
{
addr->sun_family = AF_UNIX;
addr->sun_path[0] = '\0';
snprintf(addr->sun_path + 1, UNIX_PATH_MAX - 1, "crtools-sk-%10d", id);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
*addrlen = sizeof(addr->sun_family) + sizeof("crtools-sk-") - 1 + 10;
}
static int open_unix_sk(struct unix_sk_entry *ue, int *img_fd)
{
int sk;
show_one_unix_img("Restore", ue);
sk = socket(PF_UNIX, ue->type, 0);
if (sk < 0) {
pr_perror("Can't create unix socket\n");
return -1;
}
if (ue->state == TCP_LISTEN) {
struct sockaddr_un addr;
int ret;
if (!ue->namelen || ue->namelen >= UNIX_PATH_MAX) {
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
pr_err("Bad unix name len %d\n", ue->namelen);
goto err;
}
memset(&addr, 0, sizeof(addr));
addr.sun_family = AF_UNIX;
ret = read(*img_fd, &addr.sun_path, ue->namelen);
if (ret != ue->namelen) {
pr_err("Error reading socket name from image (%d)", ret);
goto err;
}
if (addr.sun_path[0] != '\0')
unlink(addr.sun_path);
if (bind(sk, (struct sockaddr *)&addr,
sizeof(addr.sun_family) + ue->namelen) < 0) {
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
pr_perror("Can't bind socket\n");
goto err;
}
if (listen(sk, ue->backlog) < 0) {
pr_perror("Can't listen socket\n");
goto err;
}
} else if (ue->state == TCP_ESTABLISHED) {
if (ue->peer < ue->id) {
struct sockaddr_un addr;
int len;
struct unix_accept_job *aj;
/*
* Will become a server
*/
prep_conn_addr(ue->id, &addr, &len);
if (bind(sk, (struct sockaddr *)&addr, len) < 0) {
pr_perror("Can't bind socket\n");
goto err;
}
if (listen(sk, 1) < 0) {
pr_perror("Can't listen socket\n");
goto err;
}
aj = xmalloc(sizeof(*aj));
if (aj == NULL)
goto err;
aj->fd = ue->fd;
aj->next = accept_jobs;
accept_jobs = aj;
unix_show_job("Sched acc", ue->fd, ue->id);
} else {
struct unix_conn_job *cj;
/*
* Will do the connect
*/
cj = xmalloc(sizeof(*cj));
if (!cj)
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
goto err;
prep_conn_addr(ue->peer, &cj->addr, &cj->addrlen);
cj->fd = ue->fd;
cj->next = conn_jobs;
conn_jobs = cj;
unix_show_job("Sched conn", ue->fd, ue->peer);
}
} else {
pr_err("Unknown state %d\n", ue->state);
goto err;
}
if (move_img_fd(img_fd, ue->fd))
return -1;
return reopen_fd_as(ue->fd, sk);
err:
close(sk);
return -1;
}
static int prepare_unix_sockets(int pid)
{
int usk_fd, ret = -1;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
usk_fd = open_image_ro(CR_FD_UNIXSK, pid);
if (usk_fd < 0)
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return -1;
while (1) {
struct unix_sk_entry ue;
ret = read(usk_fd, &ue, sizeof(ue));
if (ret == 0)
break;
if (ret != sizeof(ue)) {
pr_perror("%d: Bad unix sk entry (ret %d)\n", pid, ret);
ret = -1;
break;
}
ret = open_unix_sk(&ue, &usk_fd);
if (ret)
break;
}
close(usk_fd);
if (!ret)
ret = run_connect_jobs();
if (!ret)
ret = run_accept_jobs();
err:
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
return ret;
}
int prepare_sockets(int pid)
{
return prepare_unix_sockets(pid);
}
void show_unixsk(char *name, int fd, bool show_header)
{
struct unix_sk_entry ue;
int ret = 0;
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (show_header) {
pr_info("\n");
pr_info("CR_FD_UNIXSK: %s\n", name);
pr_info("----------------------------------------\n");
}
while (1) {
ret = read_ptr_safe_eof(fd, &ue, out);
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (!ret)
goto out;
pr_info("fd %4d type %2d state %2d namelen %4d id %6d peer %6d",
ue.fd, ue.type, ue.state, ue.namelen, ue.id, ue.peer);
if (ue.namelen) {
ret = read_safe_eof(fd, buf, ue.namelen, out);
if (!ret)
goto out;
if (!buf[0])
buf[0] = '@';
pr_info("\t---> [%s]\n", buf);
} else
pr_info("\n");
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
}
out:
if (ret)
pr_info("\n");
Unix sockets initial support Currently it can only work with stream sockets, which have no skbs in queues (listening or established -- both work OK). The cpt part uses the sock_diag engine that was merged to Dave recently to collect sockets. Then it dumps sockets by checking the filesystem ID of a failed-to-open through /proc/pid/fd descriptors (sockets do not allow for such tricks with opens through proc) against SOCKFS_TYPE. The rst part is more tricky. Listen sockets are just restored, this is simple. Connected sockets are restored like this: 1. One end establishes a listening anon socket at the desired descriptor; 2. The other end just creates a socket at the desired descriptor; 3. All sockets, that are to be connect()-ed call connect. Unix sockets do not block connect() till the accept() time and thus we continue with... 4. ... all listening sockets call accept() and ... dup2 the new fd into the accepting end. There's a problem with this approach -- socket names are not preserved, but looking into our OpenVZ implementation I think this is OK for existing apps. What should be done next is: 1. Need to merge the file IDs patches in our tree and make Andrey to support files sharing. This will solve the sk = socket(); fork(); case. Currently it simply doesn't work :( 2. Need to add support for DGRAM sockets -- I wrote comment how to do it in the can_dump_unix_sk() 3. Need to add support for in-flight connections 4. Implement support for UDP sockets (quite simple) 5. Implement support for listening TCP sockets (also not very complex) 6. Implement support for connected TCP scokets (hard one, Tejun's patches are not very good for this from my POV) Cyrill, plz, apply this patch and put the above descriptions onto wiki docs (do we have the plans page yet?). Andrey, plz, take care of unix sockets tests in zdtm. Most likely it won't work till you do the shared files support for sockets. Signed-off-by: Pavel Emelyanov <xemul@parallels.com> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
2011-12-26 22:12:03 +04:00
if (show_header)
pr_info("----------------------------------------\n");
}