[CRIU] [PATCH 4/5] eventfd: Add checkpoint/restore support
Cyrill Gorcunov
gorcunov at openvz.org
Fri Apr 27 18:13:27 EDT 2012
Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
---
Makefile | 1 +
cr-dump.c | 22 +++++++
cr-restore.c | 4 +
cr-show.c | 23 +++++++
crtools.c | 7 ++
eventfs.c | 165 +++++++++++++++++++++++++++++++++++++++++++++++++++++
include/crtools.h | 3 +
include/eventfs.h | 20 +++++++
include/image.h | 9 +++
9 files changed, 254 insertions(+), 0 deletions(-)
create mode 100644 eventfs.c
create mode 100644 include/eventfs.h
diff --git a/Makefile b/Makefile
index 8ad48c6..a13b80b 100644
--- a/Makefile
+++ b/Makefile
@@ -48,6 +48,7 @@ OBJS += file-ids.o
OBJS += namespaces.o
OBJS += uts_ns.o
OBJS += ipc_ns.o
+OBJS += eventfs.o
DEPS := $(patsubst %.o,%.d,$(OBJS))
diff --git a/cr-dump.c b/cr-dump.c
index e6384e9..b466a92 100644
--- a/cr-dump.c
+++ b/cr-dump.c
@@ -34,6 +34,7 @@
#include "parasite.h"
#include "parasite-syscall.h"
#include "files.h"
+#include "eventfs.h"
#ifndef CONFIG_X86_64
# error No x86-32 support yet
@@ -427,6 +428,9 @@ static int do_dump_gen_file(const struct fd_parms *p, int lfd,
case FDINFO_PIPE:
ret = dump_one_pipe(lfd, e.id, p);
break;
+ case FDINFO_EVENTFD:
+ ret = dump_one_eventfd(lfd, e.id, p);
+ break;
default:
ret = dump_one_reg_file(lfd, e.id, p);
break;
@@ -559,10 +563,18 @@ static int dump_chrdev(struct fd_parms *p, int lfd, const struct cr_fdset *set)
return dump_unsupp_fd(p);
}
+static int dump_eventfd(struct fd_parms *p, int lfd, const struct cr_fdset *set)
+{
+ p->id = MAKE_FD_GENID(p->stat.st_dev, p->stat.st_ino, p->pos);
+ p->type = FDINFO_EVENTFD;
+ return do_dump_gen_file(p, lfd, set);
+}
+
static int dump_one_file(pid_t pid, int fd, int lfd, char fd_flags,
const struct cr_fdset *cr_fdset)
{
struct fd_parms p;
+ struct statfs statfs;
if (fill_fd_params(pid, fd, lfd, fd_flags, &p) < 0) {
pr_perror("Can't get stat on %d", fd);
@@ -575,6 +587,16 @@ static int dump_one_file(pid_t pid, int fd, int lfd, char fd_flags,
if (S_ISCHR(p.stat.st_mode))
return dump_chrdev(&p, lfd, cr_fdset);
+ if (fstatfs(lfd, &statfs)) {
+ pr_perror("Can't obtain statfs on fd %d\n", fd);
+ return -1;
+ }
+
+ if (is_anon_inode(&statfs)) {
+ if (is_eventfd_link(lfd))
+ return dump_eventfd(&p, lfd, cr_fdset);
+ }
+
if (S_ISREG(p.stat.st_mode) ||
S_ISDIR(p.stat.st_mode) ||
S_ISFIFO(p.stat.st_mode))
diff --git a/cr-restore.c b/cr-restore.c
index 5c337d3..2ecf5a4 100644
--- a/cr-restore.c
+++ b/cr-restore.c
@@ -33,6 +33,7 @@
#include "sockets.h"
#include "lock.h"
#include "files.h"
+#include "eventfs.h"
#include "proc_parse.h"
#include "restorer-blob.h"
#include "crtools.h"
@@ -291,6 +292,9 @@ static int prepare_shared(void)
if (collect_unix_sockets())
return -1;
+ if (collect_eventfd())
+ return -1;
+
list_for_each_entry(pi, &tasks, list) {
ret = prepare_shmem_pid(pi->pid);
if (ret < 0)
diff --git a/cr-show.c b/cr-show.c
index c0193f4..f1263a3 100644
--- a/cr-show.c
+++ b/cr-show.c
@@ -62,6 +62,7 @@ static char *fdtype2s(u8 type)
[FDINFO_INETSK] = "isk",
[FDINFO_PIPE] = "pipe",
[FDINFO_UNIXSK] = "usk",
+ [FDINFO_EVENTFD] = "efd",
};
if (type > FDINFO_UND && type < FD_INFO_MAX)
@@ -99,6 +100,28 @@ void show_fown_cont(fown_t *fown)
fown->uid, fown->euid, fown->signum, fown->pid_type, fown->pid);
}
+void show_eventfds(int fd, struct cr_options *o)
+{
+ struct eventfd_file_entry efe;
+
+ pr_img_head(CR_FD_EVENTFD);
+
+ while (1) {
+ int ret;
+
+ ret = read_img_eof(fd, &efe);
+ if (ret <= 0)
+ goto out;
+ pr_msg("id: %#08x flags %#04x counter: %#016lx ",
+ efe.id, efe.flags, efe.counter);
+ show_fown_cont(&efe.fown);
+ pr_msg("\n");
+ }
+
+out:
+ pr_img_tail(CR_FD_EVENTFD);
+}
+
void show_reg_files(int fd_reg_files, struct cr_options *o)
{
struct reg_file_entry rfe;
diff --git a/crtools.c b/crtools.c
index a5a8eaa..e939d86 100644
--- a/crtools.c
+++ b/crtools.c
@@ -61,6 +61,13 @@ struct cr_fd_desc_tmpl fdset_template[CR_FD_MAX] = {
.show = show_reg_files,
},
+ /* eventfd */
+ [CR_FD_EVENTFD] = {
+ .fmt = FMT_FNAME_EVENTFD,
+ .magic = EVENTFD_MAGIC,
+ .show = show_eventfds,
+ },
+
/* core data, such as regs and vmas and such */
[CR_FD_CORE] = {
.fmt = FMT_FNAME_CORE,
diff --git a/eventfs.c b/eventfs.c
new file mode 100644
index 0000000..dfea4c9
--- /dev/null
+++ b/eventfs.c
@@ -0,0 +1,165 @@
+#include <unistd.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+#include <sys/stat.h>
+#include <sys/statfs.h>
+#include <sys/types.h>
+#include <sys/ioctl.h>
+#include <sys/eventfd.h>
+
+#include "compiler.h"
+#include "types.h"
+#include "eventfs.h"
+
+#include "crtools.h"
+#include "image.h"
+#include "util.h"
+#include "log.h"
+
+struct eventfd_file_info {
+ struct eventfd_file_entry efe;
+ struct file_desc d;
+};
+
+/* Checks if file desciptor @lfd is eventfd */
+int is_eventfd_link(int lfd)
+{
+ char link[PATH_MAX], path[32];
+ ssize_t ret;
+
+ snprintf(path, sizeof(path), "/proc/self/fd/%d", lfd);
+ ret = readlink(path, link, sizeof(link));
+ if (ret < 0) {
+ pr_perror("Can't read link of fd %d\n", lfd);
+ return 0;
+ }
+ link[ret] = 0;
+ if (!strcmp(link, "anon_inode:[eventfd]"))
+ return 1;
+
+ return 0;
+}
+
+static void pr_info_eventfd(char *action, struct eventfd_file_entry *efe)
+{
+ pr_info("%seventfd: id %#08x flags %#04x counter %#016lx\n",
+ action, efe->id, efe->flags, efe->counter);
+}
+
+int dump_one_eventfd(int lfd, u32 id, const struct fd_parms *p)
+{
+ int image_fd = fdset_fd(glob_fdset, CR_FD_EVENTFD);
+ struct eventfd_file_entry efe;
+ char buf[64];
+ char *pos;
+ int ret, fdinfo;
+
+ efe.id = id;
+ efe.flags = p->flags;
+ efe.fown = p->fown;
+
+ snprintf(buf, sizeof(buf), "/proc/self/fdinfo/%d", lfd);
+ fdinfo = open(buf, O_RDONLY);
+ if (fdinfo < 0) {
+ pr_perror("Can't open %d (%d)", p->fd, lfd);
+ return -1;
+ }
+
+ ret = read(fdinfo, buf, sizeof(buf));
+ close(fdinfo);
+
+ if (ret <= 0) {
+ pr_perror("Reading eventfd from %d (%d) failed", p->fd, lfd);
+ return -1;
+ }
+
+ pos = strstr(buf, "count-raw:");
+ if (!pos || !sscanf(pos, "count-raw: %lx", &efe.counter)) {
+ pr_err("Counter value is not found for %d (%d)\n", p->fd, lfd);
+ return -1;
+ }
+
+ pr_info_eventfd("Dumping: ", &efe);
+ if (write_img(image_fd, &efe))
+ return -1;
+
+ return 0;
+}
+
+static int eventfd_open(struct file_desc *d)
+{
+ struct eventfd_file_info *info;
+ size_t size;
+ int tmp;
+
+ info = container_of(d, struct eventfd_file_info, d);
+
+ tmp = eventfd(0, 0);
+ if (tmp < 0) {
+ pr_perror("Can't create eventfd %#08x",
+ info->efe.id);
+ return -1;
+ }
+
+ if (fcntl(tmp, F_SETFD, info->efe.flags)) {
+ pr_perror("Can't set flags on eventfd %#08x",
+ info->efe.id);
+ goto err_close;
+ }
+
+ if (restore_fown(tmp, &info->efe.fown))
+ goto err_close;
+
+ size = sizeof(info->efe.counter);
+ if (write(tmp, &info->efe.counter, size) != size) {
+ pr_perror("Can't write counter to eventfd %#08x",
+ info->efe.id);
+ goto err_close;
+ }
+
+ return tmp;
+
+err_close:
+ close(tmp);
+ return -1;
+}
+
+static struct file_desc_ops eventfd_desc_ops = {
+ .open = eventfd_open,
+};
+
+int collect_eventfd(void)
+{
+ struct eventfd_file_info *info = NULL;
+ int ret, image_fd;
+
+ image_fd = open_image_ro(CR_FD_EVENTFD);
+ if (image_fd < 0)
+ return -1;
+
+ while (1) {
+ ret = -1;
+
+ info = xmalloc(sizeof(*info));
+ if (!info)
+ break;
+
+ ret = read_img_eof(image_fd, &info->efe);
+ if (ret < 0)
+ goto err;
+ else if (!ret)
+ break;
+ pr_info_eventfd("Collected: ", &info->efe);
+ file_desc_add(&info->d, FDINFO_EVENTFD, info->efe.id, &eventfd_desc_ops);
+ }
+
+err:
+ xfree(info);
+ close(image_fd);
+ return ret;
+}
diff --git a/include/crtools.h b/include/crtools.h
index 6625448..78cf968 100644
--- a/include/crtools.h
+++ b/include/crtools.h
@@ -54,6 +54,7 @@ enum {
CR_FD_PIPES,
CR_FD_PIPES_DATA,
CR_FD_REMAP_FPATH,
+ CR_FD_EVENTFD,
_CR_FD_GLOB_TO,
CR_FD_MAX
@@ -100,6 +101,7 @@ void show_fs(int fd, struct cr_options *o);
void show_remap_files(int fd, struct cr_options *o);
void show_ghost_file(int fd, struct cr_options *o);
void show_fown_cont(fown_t *fown);
+void show_eventfds(int fd, struct cr_options *o);
extern void print_data(unsigned long addr, unsigned char *data, size_t size);
extern struct cr_fd_desc_tmpl fdset_template[CR_FD_MAX];
@@ -117,6 +119,7 @@ extern struct cr_fd_desc_tmpl fdset_template[CR_FD_MAX];
#define FMT_FNAME_SIGACTS "sigacts-%d.img"
#define FMT_FNAME_UNIXSK "unixsk.img"
#define FMT_FNAME_INETSK "inetsk.img"
+#define FMT_FNAME_EVENTFD "eventfd.img"
#define FMT_FNAME_ITIMERS "itimers-%d.img"
#define FMT_FNAME_CREDS "creds-%d.img"
#define FMT_FNAME_UTSNS "utsns-%d.img"
diff --git a/include/eventfs.h b/include/eventfs.h
new file mode 100644
index 0000000..ce3a7b7
--- /dev/null
+++ b/include/eventfs.h
@@ -0,0 +1,20 @@
+#ifndef EVENTFS_H__
+#define EVENTFS_H__
+
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "compiler.h"
+#include "types.h"
+#include "files.h"
+#include "crtools.h"
+
+#ifndef EFD_RAW_COUNTER
+#define EFD_RAW_COUNTER 0xc0044510
+#endif
+
+extern int is_eventfd_link(int lfd);
+extern int dump_one_eventfd(int lfd, u32 id, const struct fd_parms *p);
+extern int collect_eventfd(void);
+
+#endif /* EVENTFS_H__ */
diff --git a/include/image.h b/include/image.h
index 8d1c434..2349e37 100644
--- a/include/image.h
+++ b/include/image.h
@@ -32,6 +32,7 @@
#define MM_MAGIC 0x57492820 /* Pskov */
#define REMAP_FPATH_MAGIC 0x59133954 /* Vologda */
#define GHOST_FILE_MAGIC 0x52583605 /* Oryol */
+#define EVENTFD_MAGIC 0x44523722 /* Anapa */
enum fd_types {
FDINFO_UND,
@@ -39,6 +40,7 @@ enum fd_types {
FDINFO_PIPE,
FDINFO_INETSK,
FDINFO_UNIXSK,
+ FDINFO_EVENTFD,
FD_INFO_MAX
};
@@ -81,6 +83,13 @@ struct ghost_file_entry {
u32 mode;
} __packed;
+struct eventfd_file_entry {
+ u32 id;
+ u16 flags;
+ fown_t fown;
+ u64 counter;
+} __packed;
+
struct fdinfo_entry {
u32 fd;
u8 type;
--
1.7.7.6
More information about the CRIU
mailing list