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

1025 lines
21 KiB
C
Raw Normal View History

tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <string.h>
#include <limits.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/mman.h>
#include <sys/ioctl.h>
#include <termios.h>
#include <linux/major.h>
#include "compiler.h"
#include "types.h"
#include "syscall.h"
#include "files.h"
#include "crtools.h"
#include "image.h"
#include "util.h"
#include "log.h"
#include "list.h"
#include "util-net.h"
#include "proc_parse.h"
#include "file-ids.h"
#include "protobuf.h"
#include "protobuf/tty.pb-c.h"
#include "protobuf/creds.pb-c.h"
#include "parasite-syscall.h"
#include "parasite.h"
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
#include "pstree.h"
#include "tty.h"
/*
* Here are some notes about overall TTY c/r design. At moment
* we support unix98 ptys only.
*
* Usually the PTYs represent a pair of links -- master peer and slave
* peer. Master peer must be opened before slave. Internally, when kernel
* creates master peer it also generates a slave interface in a form of
* /dev/pts/N, where N is that named pty "index". Master/slave connection
* unambiguously identified by this index.
*
* Still, one master can carry multiple slaves -- for example a user opens
* one master via /dev/ptmx and appropriate /dev/pts/N in sequence.
* The result will be the following
*
* master
* `- slave 1
* `- slave 2
*
* both slave will have same master index but different file descriptors.
* Still inside the kernel pty parameters are same for both slaves. Thus
* only one slave parameters should be restored, there is no need to carry
* all parameters for every slave peer we've found.
*
*/
#undef LOG_PREFIX
#define LOG_PREFIX "tty: "
struct tty_info_entry {
struct list_head list;
TtyInfoEntry *tie;
};
struct tty_info {
struct list_head list;
struct file_desc d;
TtyFileEntry *tfe;
TtyInfoEntry *tie;
struct list_head sibling;
int major;
bool create;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
};
static LIST_HEAD(all_tty_info_entries);
static LIST_HEAD(all_ttys);
/*
* Usually an application has not that many ttys opened.
* If this won't be enough in future we simply need to
* change tracking mechanism to some more extendable.
*
* This particular bitmap requires 256 bytes of memory.
* Pretty acceptable trade off in a sake of simplicity.
*/
#define MAX_TTYS 1024
static DECLARE_BITMAP(tty_bitmap, (MAX_TTYS << 1));
static DECLARE_BITMAP(tty_active_pairs, (MAX_TTYS << 1));
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
/*
* /dev/ptmx is a shared resource between all tasks
* so we need to serialize access to it.
*/
static mutex_t *tty_mutex;
int prepare_shared_tty(void)
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
{
tty_mutex = shmalloc(sizeof(*tty_mutex));
if (!tty_mutex) {
pr_err("Can't create ptmx index mutex\n");
return -1;
}
mutex_init(tty_mutex);
return 0;
}
#define winsize_copy(d, s) \
do { \
ASSIGN_MEMBER((d), (s), ws_row); \
ASSIGN_MEMBER((d), (s), ws_col); \
ASSIGN_MEMBER((d), (s), ws_xpixel); \
ASSIGN_MEMBER((d), (s), ws_ypixel); \
} while (0)
#define termios_copy(d, s) \
do { \
struct termios __t; \
\
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
memcpy((d)->c_cc, (s)->c_cc, \
sizeof(__t.c_cc)); \
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
\
ASSIGN_MEMBER((d),(s), c_iflag); \
ASSIGN_MEMBER((d),(s), c_oflag); \
ASSIGN_MEMBER((d),(s), c_cflag); \
ASSIGN_MEMBER((d),(s), c_lflag); \
ASSIGN_MEMBER((d),(s), c_line); \
} while (0)
static int tty_gen_id(int major, int index)
{
return (index << 1) + (major == TTYAUX_MAJOR);
}
static int tty_get_index(u32 id)
{
return id >> 1;
}
/* Make sure the active pairs do exist */
int tty_verify_active_pairs(void)
{
int i;
for (i = 0; i < (MAX_TTYS << 1); i += 2) {
if (test_bit(i, tty_active_pairs) &&
!test_bit(i + 1, tty_active_pairs)) {
pr_err("Found slave peer index %d without "
"correspond master peer\n",
tty_get_index(i));
return -1;
}
}
return 0;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
static int parse_index(u32 id, int lfd, int major)
{
int index = -1;
switch (major) {
case TTYAUX_MAJOR:
if (ioctl(lfd, TIOCGPTN, &index)) {
pr_perror("Can't obtain ptmx index\n");
return -1;
}
break;
case UNIX98_PTY_SLAVE_MAJOR: {
char path[PATH_MAX];
char link[32];
int len;
snprintf(link, sizeof(link), "/proc/self/fd/%d", lfd);
len = readlink(link, path, sizeof(path) - 1);
if (len < 0) {
pr_perror("Can't readlink %s", link);
return -1;
}
path[len] = '\0';
if (sscanf(path, PTS_FMT, &index) != 1) {
pr_err("Unexpected format on path %s\n", path);
return -1;
}
break;
}
}
if (index > MAX_TTYS) {
pr_err("Index %d on tty %x is too big\n", index, id);
return -1;
}
return index;
}
static int tty_test_and_set(int bit, unsigned long *bitmap)
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
{
int ret;
BUG_ON(bit > (MAX_TTYS << 1));
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
ret = test_bit(bit, bitmap);
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
if (!ret)
set_bit(bit, bitmap);
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
return ret;
}
static int pty_open_ptmx_index(int flags, int index)
{
int fds[32], i, ret = -1, cur_idx;
memset(fds, 0xff, sizeof(fds));
mutex_lock(tty_mutex);
for (i = 0; i < ARRAY_SIZE(fds); i++) {
fds[i] = open(PTMX_PATH, flags);
if (fds[i] < 0) {
pr_perror("Can't open %s", PTMX_PATH);
break;
}
if (ioctl(fds[i], TIOCGPTN, &cur_idx)) {
pr_perror("Can't obtain current index on %s", PTMX_PATH);
break;
}
pr_debug("\t\tptmx opened with index %d\n", cur_idx);
if (cur_idx == index) {
pr_info("ptmx opened with index %d\n", cur_idx);
ret = fds[i];
fds[i] = -1;
break;
}
/*
* Maybe indices are already borrowed by
* someone else, so no need to continue.
*/
if (cur_idx < index && (index - cur_idx) < ARRAY_SIZE(fds))
continue;
pr_err("Unable to open %s with specified index %d\n", PTMX_PATH, index);
break;
}
for (i = 0; i < ARRAY_SIZE(fds); i++) {
if (fds[i] >= 0)
close(fds[i]);
}
mutex_unlock(tty_mutex);
return ret;
}
static int try_open_pts(int index, int flags, bool report)
{
char path[64];
int fd;
snprintf(path, sizeof(path), PTS_FMT, index);
fd = open(path, flags);
if (fd < 0 && report) {
pr_err("Can't open terminal %s\n", path);
return -1;
}
return fd;
}
static int unlock_pty(int fd)
{
const int lock = 0;
/*
* Usually when ptmx opened it gets locked
* by kernel and we need to unlock it to be
* able to connect slave peer.
*/
if (ioctl(fd, TIOCSPTLCK, &lock)) {
pr_err("Unable to unlock pty device via y%d\n", fd);
return -1;
}
return 0;
}
static int lock_pty(int fd)
{
const int lock = 1;
if (ioctl(fd, TIOCSPTLCK, &lock)) {
pr_err("Unable to lock pty device via %d\n", fd);
return -1;
}
return 0;
}
static int tty_set_sid(int fd)
{
if (ioctl(fd, TIOCSCTTY, 1)) {
pr_perror("Can't set sid on terminal fd %d\n", fd);
return -1;
}
return 0;
}
static int tty_set_prgp(int fd, int group)
{
if (ioctl(fd, TIOCSPGRP, &group)) {
pr_perror("Failed to set group %d on %d\n", group, fd);
return -1;
}
return 0;
}
static int tty_restore_ctl_terminal(struct file_desc *d, int fd)
{
struct tty_info *info = container_of(d, struct tty_info, d);
int slave, ret = -1;
char pts_name[64];
if (!is_service_fd(fd, CTL_TTY_OFF))
return 0;
snprintf(pts_name, sizeof(pts_name), PTS_FMT, info->tie->pty->index);
slave = open(pts_name, O_RDONLY);
if (slave < 0) {
pr_perror("Can't open %s", pts_name);
return -1;
}
pr_info("Restore session %d by %d tty (index %d)\n",
info->tie->sid, (int)getpid(),
info->tie->pty->index);
ret = tty_set_sid(slave);
if (!ret)
ret = tty_set_prgp(slave, info->tie->pgrp);
close(slave);
close(fd);
return ret;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
static char *tty_type(struct tty_info *info)
{
static char *tty_types[] = {
[UNIX98_PTY_SLAVE_MAJOR] = "pts",
[TTYAUX_MAJOR] = "ptmx",
};
static char tty_unknown[] = "unknown";
switch (info->major) {
case UNIX98_PTY_SLAVE_MAJOR:
case TTYAUX_MAJOR:
return tty_types[info->major];
}
return tty_unknown;
}
static bool pty_is_master(struct tty_info *info)
{
return info->major == TTYAUX_MAJOR;
}
static void tty_show_pty_info(char *prefix, struct tty_info *info)
{
pr_info("%s type %s id %#x index %d (master %d sid %d pgrp %d)\n",
prefix, tty_type(info), info->tfe->id, info->tie->pty->index,
pty_is_master(info), info->tie->sid, info->tie->pgrp);
}
static int restore_tty_params(int fd, struct tty_info *info)
{
struct winsize w;
struct termios t;
/*
* It's important to zeroify termios
* because it contain @c_cc array which
* is bigger than TERMIOS_NCC. Same applies
* to winsize usage, we can't guarantee the
* structure taked from the system headers will
* never be extended.
*/
if (info->tie->termios_locked) {
memzero(&t, sizeof(t));
termios_copy(&t, info->tie->termios_locked);
if (ioctl(fd, TIOCSLCKTRMIOS, &t) < 0)
goto err;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
if (info->tie->termios) {
memzero(&t, sizeof(t));
termios_copy(&t, info->tie->termios);
if (ioctl(fd, TCSETS, &t) < 0)
goto err;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
if (info->tie->winsize) {
memzero(&w, sizeof(w));
winsize_copy(&w, info->tie->winsize);
if (ioctl(fd, TIOCSWINSZ, &w) < 0)
goto err;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
return 0;
err:
pr_perror("Can't set tty params on %d", info->tfe->id);
return -1;
}
static int pty_open_slaves(struct tty_info *info)
{
int sock = -1, fd = -1, ret = -1;
struct fdinfo_list_entry *fle;
struct tty_info *slave;
char pts_name[64];
snprintf(pts_name, sizeof(pts_name), PTS_FMT, info->tie->pty->index);
sock = socket(PF_UNIX, SOCK_DGRAM, 0);
if (sock < 0) {
pr_perror("Can't create socket");
goto err;
}
list_for_each_entry(slave, &info->sibling, sibling) {
BUG_ON(pty_is_master(slave));
fd = open(pts_name, slave->tfe->flags);
if (fd < 0) {
pr_perror("Can't open slave %s", pts_name);
goto err;
}
if (restore_tty_params(fd, slave))
goto err;
fle = file_master(&slave->d);
pr_debug("send slave %#x fd %d connected on %s (pid %d)\n",
slave->tfe->id, fd, pts_name, fle->pid);
if (send_fd_to_peer(fd, fle, sock)) {
pr_perror("Can't send file descriptor");
goto err;
}
close(fd);
fd = -1;
}
ret = 0;
err:
close_safe(&fd);
close_safe(&sock);
return ret;
}
static int receive_tty(struct tty_info *info)
{
struct fdinfo_list_entry *fle;
int fd;
fle = file_master(&info->d);
pr_info("\tWaiting tty fd %d (pid %d)\n", fle->fe->fd, fle->pid);
fd = recv_fd(fle->fe->fd);
close(fle->fe->fd);
if (fd < 0) {
pr_err("Can't get fd %d\n", fd);
return -1;
}
if (rst_file_params(fd, info->tfe->fown, info->tfe->flags))
close_safe(&fd);
return fd;
}
static int pty_open_fake_ptmx(struct tty_info *slave)
{
int master = -1, ret = -1, fd = -1;
char pts_name[64];
snprintf(pts_name, sizeof(pts_name), PTS_FMT, slave->tie->pty->index);
master = pty_open_ptmx_index(O_RDONLY, slave->tie->pty->index);
if (master < 0) {
pr_perror("Can't open fale %x (index %d)",
slave->tfe->id, slave->tie->pty->index);
return -1;
}
unlock_pty(master);
fd = open(pts_name, slave->tfe->flags);
if (fd < 0) {
pr_perror("Can't open slave %s", pts_name);
goto err;
}
if (restore_tty_params(fd, slave))
goto err;
if (pty_open_slaves(slave))
goto err;
ret = fd;
fd = -1;
err:
close_safe(&master);
close_safe(&fd);
return ret;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
static int pty_open_ptmx(struct tty_info *info)
{
int master = -1;
master = pty_open_ptmx_index(info->tfe->flags, info->tie->pty->index);
if (master < 0) {
pr_perror("Can't open %x (index %d)",
info->tfe->id, info->tie->pty->index);
return -1;
}
unlock_pty(master);
if (rst_file_params(master, info->tfe->fown, info->tfe->flags))
goto err;
if (restore_tty_params(master, info))
goto err;
if (info->tie->packet_mode) {
int packet_mode = 1;
if (ioctl(master, TIOCPKT, &packet_mode) < 0) {
pr_perror("Can't set packed mode on %x",
info->tfe->id);
goto err;
}
}
if (pty_open_slaves(info))
goto err;
if (info->tie->locked)
lock_pty(master);
return master;
err:
close_safe(&master);
return -1;
}
static int tty_open(struct file_desc *d)
{
struct tty_info *info = container_of(d, struct tty_info, d);
tty_show_pty_info("open", info);
if (!info->create)
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
return receive_tty(info);
if (!pty_is_master(info))
return pty_open_fake_ptmx(info);
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
return pty_open_ptmx(info);
}
static int tty_transport(FdinfoEntry *fe, struct file_desc *d)
{
struct tty_info *info = container_of(d, struct tty_info, d);
return !info->create;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
}
static struct list_head *tty_select_pslist(struct file_desc *d, struct rst_info *ri)
{
/*
* Unix98 pty slave peers requires the master peers being
* opened before them
*/
if (pty_is_master(container_of(d, struct tty_info, d)))
return &ri->fds;
else
return &ri->tty_slaves;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
static struct file_desc_ops tty_desc_ops = {
.type = FD_TYPES__TTY,
.open = tty_open,
.post_open = tty_restore_ctl_terminal,
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
.want_transport = tty_transport,
.select_ps_list = tty_select_pslist,
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
};
static int tty_find_restoring_task(struct tty_info *info)
{
struct pstree_item *item;
if (info->tie->sid == 0)
return 0;
pr_info("Set a control terminal to %d\n", info->tie->sid);
for_each_pstree_item(item)
if (item->sid == info->tie->sid)
return prepare_ctl_tty(item->pid.virt, item->rst, info->tfe->id);
pr_err("No task found with sid %d\n", info->tie->sid);
return -1;
}
static int tty_setup_orphan_slavery(void)
{
struct tty_info *info, *peer, *m;
list_for_each_entry(info, &all_ttys, list) {
struct fdinfo_list_entry *a, *b;
bool has_leader = false;
if (pty_is_master(info))
continue;
a = file_master(&info->d);
m = info;
list_for_each_entry(peer, &info->sibling, sibling) {
if (pty_is_master(peer)) {
has_leader = true;
break;
}
/*
* Same check as in pipes and files -- need to
* order slave ends so that they do not dead lock
* waiting for each other.
*/
b = file_master(&peer->d);
if (fdinfo_rst_prio(b, a)) {
a = b;
m = peer;
}
}
if (!has_leader) {
m->create = true;
pr_debug("Found orphan slave fake leader (%#x)\n",
m->tfe->id);
}
}
return 0;
}
int tty_setup_slavery(void)
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
{
struct tty_info *info, *peer, *m;
list_for_each_entry(info, &all_ttys, list) {
if (tty_find_restoring_task(info))
return -1;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
peer = info;
list_for_each_entry_safe_continue(peer, m, &all_ttys, list) {
if (peer->tie->pty->index != info->tie->pty->index)
continue;
if (tty_find_restoring_task(peer))
return -1;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
list_add(&peer->sibling, &info->sibling);
list_del(&peer->list);
}
}
/*
* Print out information about peers.
*/
list_for_each_entry(info, &all_ttys, list) {
tty_show_pty_info("head", info);
list_for_each_entry(peer, &info->sibling, sibling)
tty_show_pty_info(" `- sibling", peer);
}
return tty_setup_orphan_slavery();
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
}
static int verify_termios(u32 id, TermiosEntry *e)
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
{
if (e && e->n_c_cc < TERMIOS_NCC) {
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
pr_err("pty ID %#x n_c_cc (%d) has wrong value\n",
id, (int)e->n_c_cc);
return -1;
}
return 0;
}
#define term_opts_missing_cmp(p, op) \
(!(p)->tie->termios op \
!(p)->tie->termios_locked op \
!(p)->tie->winsize)
#define term_opts_missing_any(p) \
term_opts_missing_cmp(p, ||)
#define term_opts_missing_all(p) \
term_opts_missing_cmp(p, &&)
static int verify_info(struct tty_info *info)
{
/*
* Master peer must have all parameters present,
* while slave peer must have either all parameters present
* or don't have them at all.
*/
if (term_opts_missing_any(info)) {
if (pty_is_master(info)) {
pr_err("Corrupted master peer %x\n", info->tfe->id);
return -1;
} else if (!term_opts_missing_all(info)) {
pr_err("Corrupted slave peer %x\n", info->tfe->id);
return -1;
}
}
if (verify_termios(info->tfe->id, info->tie->termios_locked) ||
verify_termios(info->tfe->id, info->tie->termios))
return -1;
return 0;
}
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
static TtyInfoEntry *lookup_tty_info_entry(u32 id)
{
struct tty_info_entry *e;
list_for_each_entry(e, &all_tty_info_entries, list) {
if (e->tie->id == id)
return e->tie;
}
return NULL;
}
static int collect_one_tty_info_entry(void *obj, ProtobufCMessage *msg)
{
struct tty_info_entry *info = obj;
info->tie = pb_msg(msg, TtyInfoEntry);
if (info->tie->type != TTY_TYPE__PTY) {
pr_err("Unexpected TTY type %d (id %x)\n",
info->tie->type, info->tie->id);
return -1;
}
if (!info->tie->pty) {
pr_err("No PTY data found (id %x), corrupted image?\n",
info->tie->id);
return -1;
}
INIT_LIST_HEAD(&info->list);
list_add(&info->list, &all_tty_info_entries);
return 0;
}
static int collect_one_tty(void *obj, ProtobufCMessage *msg)
{
struct tty_info *info = obj;
info->tfe = pb_msg(msg, TtyFileEntry);
info->tie = lookup_tty_info_entry(info->tfe->tty_info_id);
if (!info->tie) {
pr_err("No tty-info-id %x found on id %x\n",
info->tfe->tty_info_id, info->tfe->id);
return -1;
}
INIT_LIST_HEAD(&info->sibling);
info->major = major(info->tie->rdev);
info->create = (info->major == TTYAUX_MAJOR);
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
if (verify_info(info))
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
return -1;
pr_info("Collected tty ID %#x\n", info->tfe->id);
list_add(&info->list, &all_ttys);
file_desc_add(&info->d, info->tfe->id, &tty_desc_ops);
return 0;
}
int collect_tty(void)
{
int ret;
ret = collect_image(CR_FD_TTY_INFO, PB_TTY_INFO,
sizeof(struct tty_info_entry),
collect_one_tty_info_entry);
if (!ret)
ret = collect_image(CR_FD_TTY, PB_TTY,
sizeof(struct tty_info),
collect_one_tty);
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
return ret;
}
static int pty_get_flags(int lfd, int major, int index, TtyInfoEntry *e)
{
int slave;
e->locked = false;
e->exclusive = false;
/*
* FIXME
*
* PTYs are safe to use packet mode. While there
* is no way to fetch packet mode settings from
* the kernel, without it we see echos missing
* in `screen' application restore. So, just set
* it here for a while.
*/
e->packet_mode = true;
/*
* FIXME
*
* At moment we fetch only locked flag which
* make sense on master peer only.
*
* For exclusive and packet mode the kernel
* patching is needed.
*/
if (major != TTYAUX_MAJOR)
return 0;
slave = try_open_pts(index, O_RDONLY, false);
if (slave < 0) {
if (errno == EIO) {
e->locked = true;
return 0;
} else {
pr_err("Can't fetch flags on slave peer (index %d)\n", index);
return -1;
}
}
close(slave);
return 0;
}
static int dump_pty_info(int lfd, u32 id, const struct fd_parms *p, int major, int index)
{
TtyInfoEntry info = TTY_INFO_ENTRY__INIT;
TermiosEntry termios = TERMIOS_ENTRY__INIT;
TermiosEntry termios_locked = TERMIOS_ENTRY__INIT;
WinsizeEntry winsize = WINSIZE_ENTRY__INIT;
TtyPtyEntry pty = TTY_PTY_ENTRY__INIT;
struct parasite_tty_args *pti;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
struct termios t;
struct winsize w;
int ret = -1;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
/*
* Make sure the structures the system provides us
* correlates well with protobuf templates.
*/
BUILD_BUG_ON(ARRAY_SIZE(t.c_cc) < TERMIOS_NCC);
BUILD_BUG_ON(sizeof(termios.c_cc) != sizeof(void *));
BUILD_BUG_ON((sizeof(termios.c_cc) * TERMIOS_NCC) < sizeof(t.c_cc));
pti = parasite_dump_tty(p->ctl, p->fd);
if (!pti)
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
return -1;
info.id = id;
info.type = TTY_TYPE__PTY;
info.sid = pti->sid;
info.pgrp = pti->pgrp;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
info.rdev = p->stat.st_rdev;
info.pty = &pty;
pty.index = index;
/*
* Nothing we can do on hangin up terminal,
* just write out minimum information we can
* gather.
*/
if (pti->hangup)
return pb_write_one(fdset_fd(glob_fdset, CR_FD_TTY_INFO), &info, PB_TTY_INFO);
/*
* Now trace the paired/unpaired ttys. For example
* the task might have slave peer assigned but no
* master peer. Such "detached" master peers are
* not yet supported by our tool and better to
* inform a user about such situatio,
*/
tty_test_and_set(id, tty_active_pairs);
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
if (pty_get_flags(lfd, major, index, &info))
goto out;
info.termios = &termios;
info.termios_locked = &termios_locked;
info.winsize = &winsize;
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
termios.n_c_cc = TERMIOS_NCC;
termios.c_cc = xmalloc(pb_repeated_size(&termios, c_cc));
termios_locked.n_c_cc = TERMIOS_NCC;
termios_locked.c_cc = xmalloc(pb_repeated_size(&termios_locked, c_cc));
if (!termios.c_cc || !termios_locked.c_cc)
goto out;
memzero(&t, sizeof(t));
if (ioctl(lfd, TCGETS, &t) < 0) {
pr_perror("Can't get tty params on %x", id);
goto out;
}
termios_copy(&termios, &t);
memzero(&t, sizeof(t));
if (ioctl(lfd, TIOCGLCKTRMIOS, &t) < 0) {
pr_perror("Can't get tty locked params on %x", id);
goto out;
}
termios_copy(&termios_locked, &t);
memzero(&w, sizeof(w));
if (ioctl(lfd, TIOCGWINSZ, &w) < 0) {
pr_perror("Can't get tty window params on %x", id);
goto out;
}
winsize_copy(&winsize, &w);
ret = pb_write_one(fdset_fd(glob_fdset, CR_FD_TTY_INFO), &info, PB_TTY_INFO);
out:
xfree(termios.c_cc);
xfree(termios_locked.c_cc);
return ret;
}
static int dump_one_pty(int lfd, u32 id, const struct fd_parms *p)
{
TtyFileEntry e = TTY_FILE_ENTRY__INIT;
int ret = 0, major, index;
pr_info("Dumping tty %d with id %#x\n", lfd, id);
major = major(p->stat.st_rdev);
index = parse_index(id, lfd, major);
if (index < 0)
return -1;
e.id = id;
e.tty_info_id = tty_gen_id(major, index);
e.flags = p->flags;
e.fown = (FownEntry *)&p->fown;
/*
* FIXME
*
* Figure out how to fetch data buffered in terminal.
* For a while simply flush before dumping. Note
* we don't check for errors here since it makes
* no sense anyway, the buffered data is not handled
* properly yet.
*/
ioctl(lfd, TCFLSH, TCIOFLUSH);
if (!tty_test_and_set(e.tty_info_id, tty_bitmap))
tty: Add checkpoint/restore for unix terminals v6 Usually the PTYs represent a pair of links -- master peer and slave peer. Master peer must be opened before slave. Internally, when kernel creates master peer it also generates a slave interface in a form of /dev/pts/N, where N is that named pty "index". Master/slave connection unambiguously identified by this index. Still, one master can carry multiple slaves -- for example a user opens one master via /dev/ptmx and appropriate /dev/pts/N in sequence. The result will be the following master `- slave 1 `- slave 2 both slave will have same master index but different file descriptors. Still inside the kernel pty parameters are same for both slaves. Thus only one slave parameters should be restored, there is no need to carry all parameters for every slave peer we've found. Not yet addressed problems: - At moment of restore the master peer might be already closed for any reason so to resolve such problem we need to open a fake master peer with proper index and hook a slave on it, then we close master peer. - Need to figure out how to deal with ttys which have some data in buffers not yet flushed, at moment this data will be simply lost during c/r - Need to restore control terminals - Need to fetch tty flags such as exclusive/packet-mode, this can't be done without kernel patching [ avagin@: - ideas on contol terminals restore - overall code redesign and simplification ] v4: - drop redundant pid from dump_chrdev - make sure optional fown is passed on regular ptys - add a comments about zeroifying termios - get rid of redundant empty line in files.c v5 (by avagin@): - complete rework of tty image format, now we have two files -- tty.img and tty-info.img. The idea behind to reduce data being stored. v6 (by xemul@): - packet mode should be set to true in image, until properly fetched from the kernel - verify image data on retrieval Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> CC: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
2012-09-12 20:00:54 +04:00
ret = dump_pty_info(lfd, e.tty_info_id, p, major, index);
if (!ret)
ret = pb_write_one(fdset_fd(glob_fdset, CR_FD_TTY), &e, PB_TTY);
return ret;
}
static const struct fdtype_ops tty_ops = {
.type = FD_TYPES__TTY,
.dump = dump_one_pty,
};
int dump_tty(struct fd_parms *p, int lfd, const struct cr_fdset *set)
{
return do_dump_gen_file(p, lfd, &tty_ops, set);
}