[PATCH 0/2] ptrace: add ability to attach a file descriptor to another process

[PATCH 0/2] ptrace: add ability to attach a file descriptor to another process

[Andrew Vagin]

We need this functionality for checkpointing processes.  Now some parts
are dumped with help PTRACE_SEIZE.  Parasite code is injected to process
and it collects information.  This code should save data to somewhere.
I want to suggest a scheme, when a dumper creates file descriptor and
attaches it to a target process, then execute parasite code, which closes
this descriptor at the end.

We can't create unix sockets or open files, because a process may
be in another namespaces.

Signed-off-by: Andrew Vagin <avagin@openvz.org>

Andrew Vagin (2):
  fs: add ability attaching and allocating fd for non current tasks
  ptrace: add ability to attach a file descriptor to another process

 fs/file.c              |    9 +++++++--
 fs/open.c              |    9 +++++++--
 include/linux/file.h   |    4 ++++
 include/linux/ptrace.h |    1 +
 kernel/ptrace.c        |   27 +++++++++++++++++++++++++++
 5 files changed, 46 insertions(+), 4 deletions(-)

[PATCH 1/2] fs: add ability to attach and to allocate fd for non current tasks

[Andrew Vagin]

Signed-off-by: Andrew Vagin <avagin@openvz.org>
---
 fs/file.c            |    9 +++++++--
 fs/open.c            |    9 +++++++--
 include/linux/file.h |    4 ++++
 3 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/fs/file.c b/fs/file.c
index 4c6992d..fbf2879 100644
--- a/fs/file.c
+++ b/fs/file.c
@@ -428,9 +428,9 @@ struct files_struct init_files = {
 /*
  * allocate a file descriptor, mark it busy.
  */
-int alloc_fd(unsigned start, unsigned flags)
+int alloc_task_fd(struct task_struct *tsk, unsigned start, unsigned flags)
 {
-	struct files_struct *files = current->files;
+	struct files_struct *files = tsk->files;
 	unsigned int fd;
 	int error;
 	struct fdtable *fdt;
@@ -479,6 +479,11 @@ out:
 	return error;
 }
 
+int alloc_fd(unsigned start, unsigned flags)
+{
+	return alloc_task_fd(current, start, flags);
+}
+
 int get_unused_fd(void)
 {
 	return alloc_fd(0, 0);
diff --git a/fs/open.c b/fs/open.c
index 22c41b5..9d340cc 100644
--- a/fs/open.c
+++ b/fs/open.c
@@ -864,9 +864,9 @@ EXPORT_SYMBOL(put_unused_fd);
  * will follow.
  */
 
-void fd_install(unsigned int fd, struct file *file)
+void fd_task_install(struct task_struct *tsk, unsigned int fd, struct file *file)
 {
-	struct files_struct *files = current->files;
+	struct files_struct *files = tsk->files;
 	struct fdtable *fdt;
 	spin_lock(&files->file_lock);
 	fdt = files_fdtable(files);
@@ -875,6 +875,11 @@ void fd_install(unsigned int fd, struct file *file)
 	spin_unlock(&files->file_lock);
 }
 
+void fd_install(unsigned int fd, struct file *file)
+{
+	return fd_task_install(current, fd, file);
+}
+
 EXPORT_SYMBOL(fd_install);
 
 static inline int build_open_flags(int flags, int mode, struct open_flags *op)
diff --git a/include/linux/file.h b/include/linux/file.h
index 21a7995..0250cf4 100644
--- a/include/linux/file.h
+++ b/include/linux/file.h
@@ -34,10 +34,14 @@ extern struct file *fget_raw_light(unsigned int fd, int *fput_needed);
 extern void set_close_on_exec(unsigned int fd, int flag);
 extern void put_filp(struct file *);
 extern int alloc_fd(unsigned start, unsigned flags);
+struct task_struct;
+extern int alloc_task_fd(struct task_struct *tsk, unsigned start, unsigned flags);
 extern int get_unused_fd(void);
 #define get_unused_fd_flags(flags) alloc_fd(0, (flags))
 extern void put_unused_fd(unsigned int fd);
 
+extern void fd_task_install(struct task_struct *tsk,
+				unsigned int fd, struct file *file);
 extern void fd_install(unsigned int fd, struct file *file);
 
 #endif /* __LINUX_FILE_H */
-- 
1.7.1

[PATCH 2/2] ptrace: add ability to attach a file descriptor to another task

[Andrew Vagin]

Signed-off-by: Andrew Vagin <avagin@openvz.org>
---
 include/linux/ptrace.h |    1 +
 kernel/ptrace.c        |   27 +++++++++++++++++++++++++++
 2 files changed, 28 insertions(+), 0 deletions(-)

diff --git a/include/linux/ptrace.h b/include/linux/ptrace.h
index 800f113..8439ce9 100644
--- a/include/linux/ptrace.h
+++ b/include/linux/ptrace.h
@@ -50,6 +50,7 @@
 #define PTRACE_SEIZE		0x4206
 #define PTRACE_INTERRUPT	0x4207
 #define PTRACE_LISTEN		0x4208
+#define PTRACE_DUPFD		0x4209
 
 /* flags in @data for PTRACE_SEIZE */
 #define PTRACE_SEIZE_DEVEL	0x80000000 /* temp flag for development */
diff --git a/kernel/ptrace.c b/kernel/ptrace.c
index 24d0447..a646d01 100644
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -24,6 +24,7 @@
 #include <linux/regset.h>
 #include <linux/hw_breakpoint.h>
 #include <linux/cn_proc.h>
+#include <linux/file.h>
 
 
 static int ptrace_trapping_sleep_fn(void *flags)
@@ -826,6 +827,32 @@ int ptrace_request(struct task_struct *child, long request,
 		break;
 	}
 #endif
+	case PTRACE_DUPFD:
+	{
+		struct file *file = fget_raw(data);
+		ret = -EBADF;
+
+		if (!file)
+			break;
+
+		task_lock(child);
+		if (!child->files)
+			goto out_getfd;
+
+		ret = alloc_task_fd(child, 0, 0);
+		if (ret < 0) {
+			fput(file);
+			goto out_getfd;
+		}
+
+		fd_task_install(child, ret, file);
+
+out_getfd:
+		task_unlock(child);
+
+		break;
+	}
+
 	default:
 		break;
 	}
-- 
1.7.1

Re: [PATCH 0/2] ptrace: add ability to attach a file descriptor to another process

[Tejun Heo]

Hello,

On Fri, Dec 16, 2011 at 01:46:21PM +0300, Andrew Vagin wrote:
> We need this functionality for checkpointing processes.  Now some parts
> are dumped with help PTRACE_SEIZE.  Parasite code is injected to process
> and it collects information.  This code should save data to somewhere.
> I want to suggest a scheme, when a dumper creates file descriptor and
> attaches it to a target process, then execute parasite code, which closes
> this descriptor at the end.
> 
> We can't create unix sockets or open files, because a process may
> be in another namespaces.

I really dislike this.  This doesn't belong in ptrace at all.  With
parasite, you have full control of the process, there gotta be some
other way to talk back.  Can't you create a management process w/ open
UNIX socket when spawning a namespace?  If that doesn't work, wouldn't
letting it dump to filesystem and retrieving it afterwards work?

Thanks.

-- 
tejun

Re: [PATCH 2/2] ptrace: add ability to attach a file descriptor to another task

[Oleg Nesterov]

On 12/16, Andrew Vagin wrote:
>
> Signed-off-by: Andrew Vagin <avagin@openvz.org>

Nice changelog ;)

I agree with Tejun, this doesn't look like the good idea to me...



As for the patch itself,

> @@ -826,6 +827,32 @@ int ptrace_request(struct task_struct *child, long request,
>  		break;
>  	}
>  #endif
> +	case PTRACE_DUPFD:
> +	{
> +		struct file *file = fget_raw(data);
> +		ret = -EBADF;
> +
> +		if (!file)
> +			break;
> +
> +		task_lock(child);
> +		if (!child->files)
> +			goto out_getfd;
> +
> +		ret = alloc_task_fd(child, 0, 0);

alloc_fdtable() does kmalloc(GFP_KERNEL)/vmalloc(), not good under
spin_lock().

> +		if (ret < 0) {
> +			fput(file);

We can't do this under task_lock() too.

A tracer's sub-thread (or any CLONE_FILES task) can close this file,
it is possible that we have the last reference.


Probably, instead of the new *_task_* helpers, you can add
alloc_file_fd/fd_files_install which take the result of
get_files_struct() as the additional argument. This way you can
avoid task_lock().

Oleg.