[CRIU] Re: [PATCH 1/7] sockets: Move packet queue handling into
sk-queue.h
Cyrill Gorcunov
gorcunov at openvz.org
Mon Apr 23 06:12:39 EDT 2012
On Mon, Apr 23, 2012 at 02:05:40PM +0400, Pavel Emelyanov wrote:
> On 04/21/2012 02:56 AM, Cyrill Gorcunov wrote:
> > Easier to read.
> >
> > Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
>
> Tabs screwed up. Resend in attach please.
Here
-------------- next part --------------
>From b0fd7519366fde74d770d86661fc41781141e9fb Mon Sep 17 00:00:00 2001
From: Cyrill Gorcunov <gorcunov at openvz.org>
Date: Fri, 20 Apr 2012 13:47:03 +0400
Subject: [PATCH] sockets: Move packet queue handling into sk-queue.h
Easier to read.
Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
---
Makefile | 1 +
include/sk-queue.h | 20 +++++
sk-queue.c | 219 ++++++++++++++++++++++++++++++++++++++++++++++++++++
sockets.c | 211 ++------------------------------------------------
4 files changed, 246 insertions(+), 205 deletions(-)
create mode 100644 include/sk-queue.h
create mode 100644 sk-queue.c
diff --git a/Makefile b/Makefile
index b16ee4f..0ef29a4 100644
--- a/Makefile
+++ b/Makefile
@@ -39,6 +39,7 @@ OBJS += rbtree.o
OBJS += log.o
OBJS += libnetlink.o
OBJS += sockets.o
+OBJS += sk-queue.o
OBJS += files.o
OBJS += pipes.o
OBJS += file-ids.o
diff --git a/include/sk-queue.h b/include/sk-queue.h
new file mode 100644
index 0000000..2f38759
--- /dev/null
+++ b/include/sk-queue.h
@@ -0,0 +1,20 @@
+#ifndef SK_QUEUE_H__
+#define SK_QUEUE_H__
+
+#include "types.h"
+#include "list.h"
+#include "crtools.h"
+#include "image.h"
+
+struct sk_packet {
+ struct list_head list;
+ struct sk_packet_entry entry;
+ off_t img_off;
+};
+
+extern int read_sk_queues(void);
+extern int dump_sk_queue(int sock_fd, int sock_id);
+extern void show_sk_queues(int fd, struct cr_options *o);
+extern int restore_sk_queue(int fd, unsigned int peer_id);
+
+#endif /* SK_QUEUE_H__ */
diff --git a/sk-queue.c b/sk-queue.c
new file mode 100644
index 0000000..a63ea02
--- /dev/null
+++ b/sk-queue.c
@@ -0,0 +1,219 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <limits.h>
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <sys/sendfile.h>
+
+#include "types.h"
+#include "list.h"
+#include "image.h"
+#include "crtools.h"
+#include "util.h"
+#include "util-net.h"
+
+#include "sk-queue.h"
+
+static LIST_HEAD(packets_list);
+
+int read_sk_queues(void)
+{
+ struct sk_packet *pkt;
+ int ret, fd;
+
+ pr_info("Trying to read socket queues image\n");
+
+ fd = open_image_ro(CR_FD_SK_QUEUES);
+ if (fd < 0)
+ return -1;
+
+ while (1) {
+ struct sk_packet_entry tmp;
+
+ pkt = xmalloc(sizeof(*pkt));
+ if (!pkt) {
+ pr_err("Failed to allocate packet header\n");
+ return -ENOMEM;
+ }
+ ret = read_img_eof(fd, &pkt->entry);
+ if (ret <= 0)
+ break;
+
+ pkt->img_off = lseek(fd, 0, SEEK_CUR);
+ /*
+ * NOTE: packet must be added to the tail. Otherwise sequence
+ * will be broken.
+ */
+ list_add_tail(&pkt->list, &packets_list);
+ lseek(fd, pkt->entry.length, SEEK_CUR);
+ }
+ close(fd);
+ xfree(pkt);
+
+ return ret;
+}
+
+int dump_sk_queue(int sock_fd, int sock_id)
+{
+ struct sk_packet_entry *pe;
+ unsigned long size;
+ socklen_t tmp;
+ int ret, orig_peek_off;
+
+ /*
+ * Save original peek offset.
+ */
+ tmp = sizeof(orig_peek_off);
+ ret = getsockopt(sock_fd, SOL_SOCKET, SO_PEEK_OFF, &orig_peek_off, &tmp);
+ if (ret < 0) {
+ pr_perror("getsockopt failed\n");
+ return ret;
+ }
+ /*
+ * Discover max DGRAM size
+ */
+ tmp = sizeof(size);
+ ret = getsockopt(sock_fd, SOL_SOCKET, SO_SNDBUF, &size, &tmp);
+ if (ret < 0) {
+ pr_perror("getsockopt failed\n");
+ return ret;
+ }
+
+ /* Note: 32 bytes will be used by kernel for protocol header. */
+ size -= 32;
+ /*
+ * Try to alloc buffer for max supported DGRAM + our header.
+ * Note: STREAM queue will be written by chunks of this size.
+ */
+ pe = xmalloc(size + sizeof(struct sk_packet_entry));
+ if (!pe)
+ return -ENOMEM;
+
+ /*
+ * Enable peek offset incrementation.
+ */
+ ret = setsockopt(sock_fd, SOL_SOCKET, SO_PEEK_OFF, &ret, sizeof(int));
+ if (ret < 0) {
+ pr_perror("setsockopt fail\n");
+ goto err_brk;
+ }
+
+ pe->id_for = sock_id;
+
+ while (1) {
+ struct iovec iov = {
+ .iov_base = pe->data,
+ .iov_len = size,
+ };
+ struct msghdr msg = {
+ .msg_iov = &iov,
+ .msg_iovlen = 1,
+ };
+
+ ret = pe->length = recvmsg(sock_fd, &msg, MSG_DONTWAIT | MSG_PEEK);
+ if (ret < 0) {
+ if (ret == -EAGAIN)
+ break; /* we're done */
+ pr_perror("sys_recvmsg fail: error\n");
+ goto err_set_sock;
+ }
+ if (msg.msg_flags & MSG_TRUNC) {
+ /*
+ * DGRAM thuncated. This should not happen. But we have
+ * to check...
+ */
+ pr_err("sys_recvmsg failed: truncated\n");
+ ret = -E2BIG;
+ goto err_set_sock;
+ }
+ ret = write_img_buf(fdset_fd(glob_fdset, CR_FD_SK_QUEUES),
+ pe, sizeof(pe) + pe->length);
+ if (ret < 0) {
+ ret = -EIO;
+ goto err_set_sock;
+ }
+ }
+ ret = 0;
+
+err_set_sock:
+ /*
+ * Restore original peek offset.
+ */
+ ret = setsockopt(sock_fd, SOL_SOCKET, SO_PEEK_OFF, &orig_peek_off, sizeof(int));
+ if (ret < 0)
+ pr_perror("setsockopt failed on restore\n");
+err_brk:
+ xfree(pe);
+ return ret;
+}
+
+void show_sk_queues(int fd, struct cr_options *o)
+{
+ struct sk_packet_entry pe;
+ char *buf = NULL, *p;
+ int ret;
+
+ pr_img_head(CR_FD_SK_QUEUES);
+ while (1) {
+ ret = read_img_eof(fd, &pe);
+ if (ret <= 0)
+ break;
+ p = xrealloc(buf, pe.length);
+ if (!p)
+ break;
+ buf = p;
+ pr_info("pkt for %u length %u bytes\n",
+ pe.id_for, pe.length);
+
+ ret = read_img_buf(fd, (unsigned char *)buf, pe.length);
+ if (ret < 0)
+ break;
+
+ print_data(0, (unsigned char *)buf, pe.length);
+ }
+ xfree(buf);
+ pr_img_tail(CR_FD_SK_QUEUES);
+}
+
+int restore_sk_queue(int fd, unsigned int peer_id)
+{
+ struct sk_packet *pkt, *tmp;
+ int ret, img_fd;
+
+ pr_info("Trying to restore recv queue for %u\n", peer_id);
+
+ img_fd = open_image_ro(CR_FD_SK_QUEUES);
+ if (img_fd < 0)
+ return -1;
+
+ list_for_each_entry_safe(pkt, tmp, &packets_list, list) {
+ struct sk_packet_entry *entry = &pkt->entry;
+
+ if (entry->id_for != peer_id)
+ continue;
+
+ pr_info("\tRestoring %d-bytes skb for %u\n",
+ entry->length, peer_id);
+
+ ret = sendfile(fd, img_fd, &pkt->img_off, entry->length);
+ if (ret < 0) {
+ pr_perror("Failed to sendfile packet");
+ return -1;
+ }
+ if (ret != entry->length) {
+ pr_err("Restored skb trimmed to %d/%d\n",
+ ret, entry->length);
+ return -1;
+ }
+ list_del(&pkt->list);
+ xfree(pkt);
+ }
+
+ close(img_fd);
+ return 0;
+}
diff --git a/sockets.c b/sockets.c
index 06ab9ae..e387f1a 100644
--- a/sockets.c
+++ b/sockets.c
@@ -16,6 +16,7 @@
#include "types.h"
#include "libnetlink.h"
#include "sockets.h"
+#include "sk-queue.h"
#include "unix_diag.h"
#include "image.h"
#include "crtools.h"
@@ -80,100 +81,6 @@ struct inet_sk_desc {
unsigned int dst_addr[4];
};
-static int dump_socket_queue(int sock_fd, int sock_id)
-{
- struct sk_packet_entry *pe;
- unsigned long size;
- socklen_t tmp;
- int ret, orig_peek_off;
-
- /*
- * Save original peek offset.
- */
- tmp = sizeof(orig_peek_off);
- ret = getsockopt(sock_fd, SOL_SOCKET, SO_PEEK_OFF, &orig_peek_off, &tmp);
- if (ret < 0) {
- pr_perror("getsockopt failed\n");
- return ret;
- }
- /*
- * Discover max DGRAM size
- */
- tmp = sizeof(size);
- ret = getsockopt(sock_fd, SOL_SOCKET, SO_SNDBUF, &size, &tmp);
- if (ret < 0) {
- pr_perror("getsockopt failed\n");
- return ret;
- }
-
- /* Note: 32 bytes will be used by kernel for protocol header. */
- size -= 32;
- /*
- * Try to alloc buffer for max supported DGRAM + our header.
- * Note: STREAM queue will be written by chunks of this size.
- */
- pe = xmalloc(size + sizeof(struct sk_packet_entry));
- if (!pe)
- return -ENOMEM;
-
- /*
- * Enable peek offset incrementation.
- */
- ret = setsockopt(sock_fd, SOL_SOCKET, SO_PEEK_OFF, &ret, sizeof(int));
- if (ret < 0) {
- pr_perror("setsockopt fail\n");
- goto err_brk;
- }
-
- pe->id_for = sock_id;
-
- while (1) {
- struct iovec iov = {
- .iov_base = pe->data,
- .iov_len = size,
- };
- struct msghdr msg = {
- .msg_iov = &iov,
- .msg_iovlen = 1,
- };
-
- ret = pe->length = recvmsg(sock_fd, &msg, MSG_DONTWAIT | MSG_PEEK);
- if (ret < 0) {
- if (ret == -EAGAIN)
- break; /* we're done */
- pr_perror("sys_recvmsg fail: error\n");
- goto err_set_sock;
- }
- if (msg.msg_flags & MSG_TRUNC) {
- /*
- * DGRAM thuncated. This should not happen. But we have
- * to check...
- */
- pr_err("sys_recvmsg failed: truncated\n");
- ret = -E2BIG;
- goto err_set_sock;
- }
- ret = write_img_buf(fdset_fd(glob_fdset, CR_FD_SK_QUEUES),
- pe, sizeof(pe) + pe->length);
- if (ret < 0) {
- ret = -EIO;
- goto err_set_sock;
- }
- }
- ret = 0;
-
-err_set_sock:
- /*
- * Restore original peek offset.
- */
- ret = setsockopt(sock_fd, SOL_SOCKET, SO_PEEK_OFF, &orig_peek_off, sizeof(int));
- if (ret < 0)
- pr_perror("setsockopt failed on restore\n");
-err_brk:
- xfree(pe);
- return ret;
-}
-
#define SK_HASH_SIZE 32
#define SK_HASH_LINK(head, key, elem) \
do { \
@@ -465,7 +372,7 @@ static int dump_one_unix(const struct socket_desc *_sk, struct fd_parms *p,
if (sk->rqlen != 0 && !(sk->type == SOCK_STREAM &&
sk->state == TCP_LISTEN))
- if (dump_socket_queue(lfd, ue.id))
+ if (dump_sk_queue(lfd, ue.id))
goto err;
pr_info("Dumping unix socket at %d\n", p->fd);
@@ -890,89 +797,6 @@ static struct unix_sk_info *find_unix_sk(int id)
return NULL;
}
-struct sk_packet {
- struct list_head list;
- struct sk_packet_entry entry;
- off_t img_off;
-};
-
-static LIST_HEAD(packets_list);
-
-static int read_sockets_queues(void)
-{
- struct sk_packet *pkt;
- int ret, fd;
-
- pr_info("Trying to read socket queues image\n");
-
- fd = open_image_ro(CR_FD_SK_QUEUES);
- if (fd < 0)
- return -1;
-
- while (1) {
- struct sk_packet_entry tmp;
-
- pkt = xmalloc(sizeof(*pkt));
- if (!pkt) {
- pr_err("Failed to allocate packet header\n");
- return -ENOMEM;
- }
- ret = read_img_eof(fd, &pkt->entry);
- if (ret <= 0)
- break;
-
- pkt->img_off = lseek(fd, 0, SEEK_CUR);
- /*
- * NOTE: packet must be added to the tail. Otherwise sequence
- * will be broken.
- */
- list_add_tail(&pkt->list, &packets_list);
- lseek(fd, pkt->entry.length, SEEK_CUR);
- }
- close(fd);
- xfree(pkt);
-
- return ret;
-}
-
-static int restore_socket_queue(int fd, unsigned int peer_id)
-{
- struct sk_packet *pkt, *tmp;
- int ret, img_fd;
-
- pr_info("Trying to restore recv queue for %u\n", peer_id);
-
- img_fd = open_image_ro(CR_FD_SK_QUEUES);
- if (img_fd < 0)
- return -1;
-
- list_for_each_entry_safe(pkt, tmp, &packets_list, list) {
- struct sk_packet_entry *entry = &pkt->entry;
-
- if (entry->id_for != peer_id)
- continue;
-
- pr_info("\tRestoring %d-bytes skb for %u\n",
- entry->length, peer_id);
-
- ret = sendfile(fd, img_fd, &pkt->img_off, entry->length);
- if (ret < 0) {
- pr_perror("Failed to sendfile packet");
- return -1;
- }
- if (ret != entry->length) {
- pr_err("Restored skb trimmed to %d/%d\n",
- ret, entry->length);
- return -1;
- }
- list_del(&pkt->list);
- xfree(pkt);
- }
-
- close(img_fd);
- return 0;
-}
-
struct inet_sk_info {
struct inet_sk_entry ie;
struct file_desc d;
@@ -1218,29 +1042,6 @@ out:
pr_img_tail(CR_FD_UNIXSK);
}
-void show_sk_queues(int fd, struct cr_options *o)
-{
- struct sk_packet_entry pe;
- int ret;
-
- pr_img_head(CR_FD_SK_QUEUES);
- while (1) {
- ret = read_img_eof(fd, &pe);
- if (ret <= 0)
- break;
-
- pr_info("pkt for %u length %u bytes\n",
- pe.id_for, pe.length);
-
- ret = read_img_buf(fd, (unsigned char *)buf, pe.length);
- if (ret < 0)
- break;
-
- print_data(0, (unsigned char *)buf, pe.length);
- }
- pr_img_tail(CR_FD_SK_QUEUES);
-}
-
struct unix_conn_job {
struct unix_sk_info *sk;
struct unix_conn_job *next;
@@ -1298,7 +1099,7 @@ try_again:
return -1;
}
- if (restore_socket_queue(fle->fe.fd, peer->ue.id))
+ if (restore_sk_queue(fle->fe.fd, peer->ue.id))
return -1;
if (set_fd_flags(fle->fe.fd, ui->ue.flags))
@@ -1360,9 +1161,9 @@ static int open_unixsk_pair_master(struct unix_sk_info *ui)
return -1;
}
- if (restore_socket_queue(sk[0], peer->ue.id))
+ if (restore_sk_queue(sk[0], peer->ue.id))
return -1;
- if (restore_socket_queue(sk[1], ui->ue.id))
+ if (restore_sk_queue(sk[1], ui->ue.id))
return -1;
if (set_fd_flags(sk[0], ui->ue.flags))
@@ -1535,7 +1336,7 @@ int collect_unix_sockets(void)
close(fd);
- return read_sockets_queues();
+ return read_sk_queues();
}
int resolve_unix_peers(void)
--
1.7.7.6
More information about the CRIU
mailing list