Implement clone2() system call

=== NEW CLONE() SYSTEM CALL:

To support application checkpoint/restart, a task must have the same pid it had when it was checkpointed. When containers are nested, the tasks within the containers exist in multiple pid namespaces and hence have multiple pids to specify during restart.

This patchset implements a new system call, clone2() that lets a process specify the pids of the child process.

Patches 1 through 6 are helper patches, needed for choosing a pid for the child process.

Patch 8 defines a prototype of the new system call. Patch 9 adds some documentation on the new system call, some/all of which will eventually go into a man page.

Changelog[v7]:
- [Peter Zijlstra, Arnd Bergmann]
Rename clone_with_pids() to clone2(). Also group the arguments to clone2() into a 'struct clone_struct' to workaround the issue of exceeding 6 arguments to the system call. Also define clone-flags as u64 to allow additional clone-flags.

Changelog[v6]:
- [Nathan Lynch, Arnd Bergmann, H. Peter Anvin, Linus Torvalds] Change 'pid_set.pids' to 'pid_t pids[]' so sizeof(struct pid_set) is constant across architectures (Patches 7, 8).
- (Nathan Lynch) Change pid_set.num_pids to unsigned and remove 'unum_pids < 0' check (Patches 7,8)
- (Pavel Machek) New patch (Patch 9) to add some documentation.

Changelog[v5]:
- Make 'pid_max' a property of pid_ns (Integrated Serge Hallyn's patch into this set)
- (Eric Biederman): Avoid the new function, set_pidmap() - added couple of checks on 'target_pid' in alloc_pidmap() itself.

=== IMPORTANT TODO:

clone() system call has another limitation - all available bits in clone-flags are in use and any new clone-flag will need a variant of the clone() system call.

It appears to make sense to try and extend this new system call to address this limitation as well. The requirements of a new clone system call could then be summarized as:

- do everything clone() does today, and
- give application an ability to choose pids for the child process in all ancestor pid namespaces, and
- allow more clone_flags

Contstraints:

- system-calls are restricted to 6 parameters and clone() already takes 5 parameters, any extension to clone() interface would require one or more copy_from_user(). (Not sure if copy_from_user() of ~40 bytes would have a significant impact on performance of clone() on any architecture).

Based on these requirements and constraints, we explored a couple of system call interfaces (in earlier versions of this patchset) and currently define the system call as:

	struct clone_struct {
		u64 flags;
		u64 child_stack;
		u32 nr_pids;
		u32 parent_tid;
		u32 child_tid;
		u32 reserved1;
		u64 reserved2;
	};

	sys_clone2(struct clone_struct __user *cs, pid_t __user *pids)

Subject: [RFC][v7][PATCH 1/9]: Factor out code to allocate pidmap page

To simplify alloc_pidmap(), move code to allocate a pid map page to a separate function.

Changelog[v3]:
- Earlier version of patchset called alloc_pidmap_page() from two places. But now its called from only one place. Even so, moving this code out into a separate function simplifies alloc_pidmap().
Changelog[v2]:
- (Matt Helsley, Dave Hansen) Have alloc_pidmap_page() return -ENOMEM on error instead of -1.

---
 kernel/pid.c |   45 ++++++++++++++++++++++++++++++---------------
 1 file changed, 30 insertions(+), 15 deletions(-)

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -122,9 +122,35 @@ static void free_pidmap(struct upid *upi
 	atomic_inc(&map->nr_free);
 }
 
+static int alloc_pidmap_page(struct pidmap *map)
+{
+	void *page;
+
+	if (likely(map->page))
+		return 0;
+
+	page = kzalloc(PAGE_SIZE, GFP_KERNEL);
+
+	/*
+	 * Free the page if someone raced with us installing it:
+	 */
+	spin_lock_irq(&pidmap_lock);
+	if (map->page)
+		kfree(page);
+	else
+		map->page = page;
+	spin_unlock_irq(&pidmap_lock);
+
+	if (unlikely(!map->page))
+		return -ENOMEM;
+
+	return 0;
+}
+
 static int alloc_pidmap(struct pid_namespace *pid_ns)
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
+	int rc;
 	struct pidmap *map;
 
 	pid = last + 1;

@@ -134,21 +160,10 @@ static int alloc_pidmap(struct pid_names
 	map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
 	max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
 	for (i = 0; i <= max_scan; ++i) {
-		if (unlikely(!map->page)) {
-			void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
-			/*
-			 * Free the page if someone raced with us
-			 * installing it:
-			 */
-			spin_lock_irq(&pidmap_lock);
-			if (map->page)
-				kfree(page);
-			else
-				map->page = page;
-			spin_unlock_irq(&pidmap_lock);
-			if (unlikely(!map->page))
-				break;
-		}
+		rc = alloc_pidmap_page(map);
+		if (rc)
+			break;
+
 		if (likely(atomic_read(&map->nr_free))) {
 			do {
 				if (!test_and_set_bit(offset, map->page)) {

Subject: [RFC][v7][PATCH 2/9]: Have alloc_pidmap() return actual error code

alloc_pidmap() can fail either because all pid numbers are in use or because memory allocation failed. With support for setting a specific pid number, alloc_pidmap() would also fail if either the given pid number is invalid or in use.

Rather than have callers assume -ENOMEM, have alloc_pidmap() return the actual error.

---
 kernel/fork.c |    5 +++--
 kernel/pid.c  |   14 +++++++++-----
 2 files changed, 12 insertions(+), 7 deletions(-)

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -1110,10 +1110,11 @@ static struct task_struct *copy_process(
 		goto bad_fork_cleanup_io;
 
 	if (pid != &init_struct_pid) {
-		retval = -ENOMEM;
 		pid = alloc_pid(p->nsproxy->pid_ns);
-		if (!pid)
+		if (IS_ERR(pid)) {
+			retval = PTR_ERR(pid);
 			goto bad_fork_cleanup_io;
+		}
 
 		if (clone_flags & CLONE_NEWPID) {
 			retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -150,7 +150,7 @@ static int alloc_pidmap_page(struct pidm
 static int alloc_pidmap(struct pid_namespace *pid_ns)
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
-	int rc;
+	int rc = -EAGAIN;
 	struct pidmap *map;
 
 	pid = last + 1;

@@ -189,12 +189,14 @@ static int alloc_pidmap(struct pid_names
 		} else {
 			map = &pid_ns->pidmap[0];
 			offset = RESERVED_PIDS;
-			if (unlikely(last == offset))
+			if (unlikely(last == offset)) {
+				rc = -EAGAIN;
 				break;
+			}
 		}
 		pid = mk_pid(pid_ns, map, offset);
 	}
-	return -1;
+	return rc;
 }
 
 int next_pidmap(struct pid_namespace *pid_ns, int last)

@@ -263,8 +265,10 @@ struct pid *alloc_pid(struct pid_namespa
 	struct upid *upid;
 
 	pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
-	if (!pid)
+	if (!pid) {
+		pid = ERR_PTR(-ENOMEM);
 		goto out;
+	}
 
 	tmp = ns;
 	for (i = ns->level; i >= 0; i--) {

@@ -299,7 +303,7 @@ out_free:
 		free_pidmap(pid->numbers + i);
 
 	kmem_cache_free(ns->pid_cachep, pid);
-	pid = NULL;
+	pid = ERR_PTR(nr);
 	goto out;
 }

From: Serge Hallyn <serue-r/[email protected]>
Subject: [RFC][v7][PATCH 3/9] Make pid_max a pid_ns property

Remove the pid_max global, and make it a property of the pid_namespace. When a pid_ns is created, it inherits the parent's pid_ns.

Fixing up sysctl (trivial akin to ipc version, but potentially tedious to get right for all CONFIG* combinations) is left for later.

Changelog[v2]:
- Port to newer kernel
- Make pid_max a local variable in alloc_pidmap() to simplify code/patch

---
 include/linux/pid_namespace.h |    1 +
 kernel/pid.c                  |    4 ++--
 kernel/pid_namespace.c        |    1 +
 kernel/sysctl.c               |    4 ++--
 4 files changed, 6 insertions(+), 4 deletions(-)

Index: linux-2.6/include/linux/pid_namespace.h
===================================================================
@@ -30,6 +30,7 @@ struct pid_namespace {
 #ifdef CONFIG_BSD_PROCESS_ACCT
 	struct bsd_acct_struct *bacct;
 #endif
+	int pid_max;
 };
 
 extern struct pid_namespace init_pid_ns;

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -43,8 +43,6 @@ static struct hlist_head *pid_hash;
 static int pidhash_shift;
 struct pid init_struct_pid = INIT_STRUCT_PID;
 
-int pid_max = PID_MAX_DEFAULT;
-
 #define RESERVED_PIDS		300
 
 int pid_max_min = RESERVED_PIDS + 1;

@@ -78,6 +76,7 @@ struct pid_namespace init_pid_ns = {
 	.last_pid = 0,
 	.level = 0,
 	.child_reaper = &init_task,
+	.pid_max = PID_MAX_DEFAULT,
 };
 EXPORT_SYMBOL_GPL(init_pid_ns);
 

@@ -151,6 +150,7 @@ static int alloc_pidmap(struct pid_names
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
 	int rc = -EAGAIN;
+	int pid_max = pid_ns->pid_max;
 	struct pidmap *map;
 
 	pid = last + 1;

Index: linux-2.6/kernel/pid_namespace.c
===================================================================
@@ -87,6 +87,7 @@ static struct pid_namespace *create_pid_
 
 	kref_init(&ns->kref);
 	ns->level = level;
+	ns->pid_max = parent_pid_ns->pid_max;
 	ns->parent = get_pid_ns(parent_pid_ns);
 
 	set_bit(0, ns->pidmap[0].page);

Index: linux-2.6/kernel/sysctl.c
===================================================================
@@ -55,6 +55,7 @@
 
 #include <asm/uaccess.h>
 #include <asm/processor.h>
+#include <linux/pid_namespace.h>
 
 #ifdef CONFIG_X86
 #include <asm/nmi.h>

@@ -78,7 +79,6 @@ extern int max_threads;
 extern int core_uses_pid;
 extern int suid_dumpable;
 extern char core_pattern[];
-extern int pid_max;
 extern int min_free_kbytes;
 extern int pid_max_min, pid_max_max;
 extern int sysctl_drop_caches;

@@ -670,7 +670,7 @@ static struct ctl_table kern_table[] = {
 	{
 		.ctl_name	= KERN_PIDMAX,
 		.procname	= "pid_max",
-		.data		= &pid_max,
+		.data		= &init_pid_ns.pid_max,
 		.maxlen		= sizeof (int),
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec_minmax,

Subject: [RFC][v7][PATCH 4/9]: Add target_pid parameter to alloc_pidmap()

With support for setting a specific pid number for a process, alloc_pidmap() will need a 'target_pid' parameter.

Changelog[v6]:
- Separate target_pid > 0 case to minimize the number of checks needed.
Changelog[v3]:
- (Eric Biederman): Avoid set_pidmap() function. Added couple of checks for target_pid in alloc_pidmap() itself.
Changelog[v2]:
- (Serge Hallyn) Check for 'pid < 0' in set_pidmap().(Code actually checks for 'pid <= 0' for completeness).

---
 kernel/pid.c |   26 +++++++++++++++++++++-----
 1 file changed, 21 insertions(+), 5 deletions(-)

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -146,16 +146,22 @@ static int alloc_pidmap_page(struct pidm
 	return 0;
 }
 
-static int alloc_pidmap(struct pid_namespace *pid_ns)
+static int alloc_pidmap(struct pid_namespace *pid_ns, int target_pid)
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
 	int rc = -EAGAIN;
 	int pid_max = pid_ns->pid_max;
 	struct pidmap *map;
 
-	pid = last + 1;
-	if (pid >= pid_max)
-		pid = RESERVED_PIDS;
+	if (target_pid) {
+		pid = target_pid;
+		if (pid < 0 || pid >= pid_max)
+			return -EINVAL;
+	} else {
+		pid = last + 1;
+		if (pid >= pid_max)
+			pid = RESERVED_PIDS;
+	}
 	offset = pid & BITS_PER_PAGE_MASK;
 	map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
 	max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;

@@ -164,6 +170,15 @@ static int alloc_pidmap(struct pid_names
 		if (rc)
 			break;
 
+		if (target_pid) {
+			rc = -EBUSY;
+			if (!test_and_set_bit(offset, map->page)) {
+				atomic_dec(&map->nr_free);
+				rc = pid;
+			}
+			return rc;
+		}
+
 		if (likely(atomic_read(&map->nr_free))) {
 			do {
 				if (!test_and_set_bit(offset, map->page)) {

@@ -196,6 +211,7 @@ static int alloc_pidmap(struct pid_names
 		}
 		pid = mk_pid(pid_ns, map, offset);
 	}
+
 	return rc;
 }
 

@@ -272,7 +288,7 @@ struct pid *alloc_pid(struct pid_namespa
 
 	tmp = ns;
 	for (i = ns->level; i >= 0; i--) {
-		nr = alloc_pidmap(tmp);
+		nr = alloc_pidmap(tmp, 0);
 		if (nr < 0)
 			goto out_free;

Subject: [RFC][v7][PATCH 5/9]: Add target_pids parameter to alloc_pid()

This parameter is currently NULL, but will be used in a follow-on patch.

---
 include/linux/pid.h |    2 +-
 kernel/fork.c       |    3 ++-
 kernel/pid.c        |    9 +++++++--
 3 files changed, 10 insertions(+), 4 deletions(-)

Index: linux-2.6/include/linux/pid.h
===================================================================
@@ -119,7 +119,7 @@ extern struct pid *find_get_pid(int nr);
 extern struct pid *find_ge_pid(int nr, struct pid_namespace *);
 int next_pidmap(struct pid_namespace *pid_ns, int last);
 
-extern struct pid *alloc_pid(struct pid_namespace *ns);
+extern struct pid *alloc_pid(struct pid_namespace *ns, pid_t *target_pids);
 extern void free_pid(struct pid *pid);
 
 /*

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -940,6 +940,7 @@ static struct task_struct *copy_process(
 	int retval;
 	struct task_struct *p;
 	int cgroup_callbacks_done = 0;
+	pid_t *target_pids = NULL;
 
 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 		return ERR_PTR(-EINVAL);

@@ -1110,7 +1111,7 @@ static struct task_struct *copy_process(
 		goto bad_fork_cleanup_io;
 
 	if (pid != &init_struct_pid) {
-		pid = alloc_pid(p->nsproxy->pid_ns);
+		pid = alloc_pid(p->nsproxy->pid_ns, target_pids);
 		if (IS_ERR(pid)) {
 			retval = PTR_ERR(pid);
 			goto bad_fork_cleanup_io;

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -272,13 +272,14 @@ void free_pid(struct pid *pid)
 	call_rcu(&pid->rcu, delayed_put_pid);
 }
 
-struct pid *alloc_pid(struct pid_namespace *ns)
+struct pid *alloc_pid(struct pid_namespace *ns, pid_t *target_pids)
 {
 	struct pid *pid;
 	enum pid_type type;
 	int i, nr;
 	struct pid_namespace *tmp;
 	struct upid *upid;
+	int tpid;
 
 	pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
 	if (!pid) {

@@ -288,7 +289,11 @@ struct pid *alloc_pid(struct pid_namespa
 
 	tmp = ns;
 	for (i = ns->level; i >= 0; i--) {
-		nr = alloc_pidmap(tmp, 0);
+		tpid = 0;
+		if (target_pids)
+			tpid = target_pids[i];
+
+		nr = alloc_pidmap(tmp, tpid);
 		if (nr < 0)
 			goto out_free;

Subject: [RFC][v7][PATCH 6/9]: Add target_pids parameter to copy_process()

Add a 'target_pids' parameter to copy_process(). The new parameter will be used in a follow-on patch when clone_with_pids() is implemented.

---
 kernel/fork.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -935,12 +935,12 @@ static struct task_struct *copy_process(
 					unsigned long stack_size,
 					int __user *child_tidptr,
 					struct pid *pid,
+					pid_t *target_pids,
 					int trace)
 {
 	int retval;
 	struct task_struct *p;
 	int cgroup_callbacks_done = 0;
-	pid_t *target_pids = NULL;
 
 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 		return ERR_PTR(-EINVAL);

@@ -1319,7 +1319,7 @@ struct task_struct * __cpuinit fork_idle
 	struct pt_regs regs;
 
 	task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
-			    &init_struct_pid, 0);
+			    &init_struct_pid, NULL, 0);
 	if (!IS_ERR(task))
 		init_idle(task, cpu);
 

@@ -1342,6 +1342,7 @@ long do_fork(unsigned long clone_flags,
 	struct task_struct *p;
 	int trace = 0;
 	long nr;
+	pid_t *target_pids = NULL;
 
 	/*
 	 * Do some preliminary argument and permissions checking before we

@@ -1382,7 +1383,7 @@ long do_fork(unsigned long clone_flags,
 		trace = tracehook_prepare_clone(clone_flags);
 
 	p = copy_process(clone_flags, stack_start, regs, stack_size,
-			 child_tidptr, NULL, trace);
+			 child_tidptr, NULL, target_pids, trace);
 	/*
 	 * Do this prior waking up the new thread - the thread pointer
 	 * might get invalid after that point, if the thread exits quickly.

Subject: [RFC][v7][PATCH 7/9]: Define do_fork_with_pids()

do_fork_with_pids() is same as do_fork(), except that it takes an additional, 'pid_set', parameter. This parameter, currently unused, specifies the set of target pids of the process in each of its pid namespaces.

Changelog[v7]:
- Drop 'struct pid_set' object and pass in 'pid_t *target_pids' instead of 'struct pid_set *'.

Changelog[v6]:
- (Nathan Lynch, Arnd Bergmann, H. Peter Anvin, Linus Torvalds) Change 'pid_set.pids' to a 'pid_t pids[]' so size of 'struct pid_set' is constant across architectures.
- (Nathan Lynch) Change 'pid_set.num_pids' to 'unsigned int'.

Changelog[v4]:
- Rename 'struct target_pid_set' to 'struct pid_set' since it may be useful in other contexts.

Changelog[v3]:
- Fix "long-line" warning from checkpatch.pl

Changelog[v2]:
- To facilitate moving architecture-inpdendent code to kernel/fork.c pass in 'struct target_pid_set __user *' to do_fork_with_pids() rather than 'pid_t *' (next patch moves the arch-independent code to kernel/fork.c)

---
 include/linux/sched.h |    3 +++
 kernel/fork.c         |   17 +++++++++++++++--
 2 files changed, 18 insertions(+), 2 deletions(-)

Index: linux-2.6/include/linux/sched.h
===================================================================
@@ -2054,6 +2054,9 @@ extern int disallow_signal(int);
 
 extern int do_execve(char *, char __user * __user *, char __user * __user *, struct pt_regs *);
 extern long do_fork(unsigned long, unsigned long, struct pt_regs *, unsigned long, int __user *, int __user *);
+extern long do_fork_with_pids(unsigned long, unsigned long, struct pt_regs *,
+				unsigned long, int __user *, int __user *,
+				unsigned int, pid_t __user *);
 struct task_struct *fork_idle(int);
 
 extern void set_task_comm(struct task_struct *tsk, char *from);

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -1332,12 +1332,14 @@ struct task_struct * __cpuinit fork_idle
  * It copies the process, and if successful kick-starts
  * it and waits for it to finish using the VM if required.
  */
-long do_fork(unsigned long clone_flags,
+long do_fork_with_pids(unsigned long clone_flags,
 	      unsigned long stack_start,
 	      struct pt_regs *regs,
 	      unsigned long stack_size,
 	      int __user *parent_tidptr,
-	      int __user *child_tidptr)
+	      int __user *child_tidptr,
+	      unsigned int num_pids,
+	      pid_t __user *upids)
 {
 	struct task_struct *p;
 	int trace = 0;

@@ -1440,6 +1442,17 @@ long do_fork(unsigned long clone_flags,
 	return nr;
 }
 
+long do_fork(unsigned long clone_flags,
+	      unsigned long stack_start,
+	      struct pt_regs *regs,
+	      unsigned long stack_size,
+	      int __user *parent_tidptr,
+	      int __user *child_tidptr)
+{
+	return do_fork_with_pids(clone_flags, stack_start, regs, stack_size,
+			parent_tidptr, child_tidptr, 0, NULL);
+}
+
 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
 #define ARCH_MIN_MMSTRUCT_ALIGN 0
 #endif

Subject: [RFC][v7][PATCH 8/9]: Define clone2() syscall

Container restart requires that a task have the same pid it had when it was checkpointed. When containers are nested the tasks within the containers exist in multiple pid namespaces and hence have multiple pids to specify during restart.

clone2(), intended for use during restart, is the same as clone(), except that it takes a 'pids' paramter. This parameter lets caller choose specific pid numbers for the child process, in the process's active and ancestor pid namespaces. (Descendant pid namespaces in general don't matter since processes don't have pids in them anyway, but see comments in copy_target_pids() regarding CLONE_NEWPID).

Clone2() system call also attempts to address a second limitation of the clone() system call. clone() is restricted to 32 clone flags and most (all ?) of these are in use. If a new clone flag is needed, we will be forced to define a new variant of the clone() system call.

To prevent unprivileged processes from misusing this interface, clone2() currently needs CAP_SYS_ADMIN, when the 'pids' parameter is non-NULL.

See Documentation/clone2 in next patch for more details of clone2() and an example of its usage.

NOTE:
System calls are restricted to 6 parameters and the number and sizes of parameters needed for sys_clone2() exceed 6 integers. The new prototype works around this restriction while providing some flexibility if clone2() needs to be further extended in the future.
TODO:
- May need additional sanity checks in do_fork_with_pids().

Changelog[v7]:
- [Peter Zijlstra, Arnd Bergmann] Rename system call to clone2() and group parameters into a new 'struct clone_struct' object.

Changelog[v6]:
- (Nathan Lynch, Arnd Bergmann, H. Peter Anvin, Linus Torvalds) Change 'pid_set.pids' to a 'pid_t pids[]' so size of 'struct pid_set' is constant across architectures.
- (Nathan Lynch) Change pid_set.num_pids to unsigned and remove 'unum_pids < 0' check.

Changelog[v4]:
- (Oren Laadan) rename 'struct target_pid_set' to 'struct pid_set'

Changelog[v3]:
- (Oren Laadan) Allow CLONE_NEWPID flag (by allocating an extra pid in the target_pids[] list and setting it 0. See copy_target_pids()).
- (Oren Laadan) Specified target pids should apply only to youngest pid-namespaces (see copy_target_pids())
- (Matt Helsley) Update patch description.

Changelog[v2]:
- Remove unnecessary printk and add a note to callers of copy_target_pids() to free target_pids.
- (Serge Hallyn) Mention CAP_SYS_ADMIN restriction in patch description.
- (Oren Laadan) Add checks for 'num_pids < 0' (return -EINVAL) and 'num_pids == 0' (fall back to normal clone()).
- Move arch-independent code (sanity checks and copy-in of target-pids) into kernel/fork.c and simplify sys_clone_with_pids()

Changelog[v1]:
- Fixed some compile errors (had fixed these errors earlier in my git tree but had not refreshed patches before emailing them)

---
 arch/x86/include/asm/syscalls.h    |    1 
 arch/x86/include/asm/unistd_32.h   |    1 
 arch/x86/kernel/process_32.c       |   31 +++++++++++
 arch/x86/kernel/syscall_table_32.S |    1 
 include/linux/types.h              |   10 +++
 kernel/fork.c                      |   96 ++++++++++++++++++++++++++++++++++++-
 6 files changed, 139 insertions(+), 1 deletion(-)

Index: linux-2.6/arch/x86/include/asm/syscalls.h
===================================================================
@@ -55,6 +55,7 @@ struct sel_arg_struct;
 struct oldold_utsname;
 struct old_utsname;
 
+asmlinkage long sys_clone2(struct clone_struct __user *cs, pid_t *pids);
 asmlinkage long sys_mmap2(unsigned long, unsigned long, unsigned long,
 			  unsigned long, unsigned long, unsigned long);
 asmlinkage int old_mmap(struct mmap_arg_struct __user *);

Index: linux-2.6/arch/x86/include/asm/unistd_32.h
===================================================================
@@ -342,6 +342,7 @@
 #define __NR_pwritev		334
 #define __NR_rt_tgsigqueueinfo	335
 #define __NR_perf_counter_open	336
+#define __NR_clone2		337
 
 #ifdef __KERNEL__
 

Index: linux-2.6/arch/x86/kernel/process_32.c
===================================================================
@@ -443,6 +443,37 @@ int sys_clone(struct pt_regs *regs)
 	return do_fork(clone_flags, newsp, regs, 0, parent_tidptr, child_tidptr);
 }
 
+asmlinkage long
+sys_clone2(struct clone_struct __user *ucs, pid_t __user *pids)
+{
+	int rc;
+	struct clone_struct kcs;
+	unsigned long clone_flags;
+	int __user *parent_tid_ptr;
+	int __user *child_tid_ptr;
+	unsigned long __user child_stack_base;
+	struct pt_regs *regs;
+
+	rc = copy_from_user(&ucs, &kcs, sizeof(kcs));
+	if (rc)
+		return -EFAULT;
+
+	/*
+	 * TODO: Convert clone_flags to 64-bit
+	 */
+	clone_flags = (unsigned long)kcs.flags;
+	child_stack_base = (unsigned long)kcs.child_stack;
+	parent_tid_ptr = &ucs.parent_tid;
+	child_tid_ptr =  &ucs.child_tid;
+	regs = task_pt_regs(current);
+
+	if (!child_stack_base)
+		child_stack_base = user_stack_pointer(regs);
+
+	return do_fork_with_pids(clone_flags, child_stack_base, regs, 0,
+			parent_tid_ptr, child_tid_ptr, kcs.nr_pids, pids);
+}
+
 /*
  * sys_execve() executes a new program.
  */

Index: linux-2.6/arch/x86/kernel/syscall_table_32.S
===================================================================
@@ -336,3 +336,4 @@ ENTRY(sys_call_table)
 	.long sys_pwritev
 	.long sys_rt_tgsigqueueinfo	/* 335 */
 	.long sys_perf_counter_open
+	.long sys_clone2

Index: linux-2.6/include/linux/types.h
===================================================================
@@ -204,6 +204,16 @@ struct ustat {
 	char			f_fpack[6];
 };
 
+struct clone_struct {
+	u64 flags;
+	u64 child_stack;
+	u32 nr_pids;
+	u32 parent_tid;
+	u32 child_tid;
+	u32 reserved1;
+	u64 reserved2;
+};
+
 #endif	/* __KERNEL__ */
 #endif /*  __ASSEMBLY__ */
 #endif /* _LINUX_TYPES_H */

Subject: [RFC][v7][PATCH 9/9]: Document clone2() syscall

This gives a brief overview of the clone2() system call. We should eventually describe more details in existing clone(2) man page or in a new man page.

Changelog[v7]:
- Rename clone_with_pids() to clone2()
- Changes to reflect new prototype of clone2() (using clone_struct).

---
 Documentation/clone2 |   85 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 85 insertions(+)

Index: linux-2.6/Documentation/clone2
===================================================================
@@ -0,0 +1,85 @@
+
+struct clone_struct {
+	u64 flags;
+	u64 child_stack;
+	u32 nr_pids;
+	u32 parent_tid;
+	u32 child_tid;
+	u32 reserved1;
+	u64 reserved2;
+};
+
+clone2(struct clone_struct * __user clone_args, pid_t * __user pids)
+
+	In addition to doing everything that clone() system call does,
+	the clone2() system call:
+
+		- allows additional clone flags (all 32 bits in the flags
+		  parameter to clone() are in use)
+
+		- allows user to specify a pid for the child process in its
+		  active and ancestor pid name spaces.
+
+	This system call is meant to be used when restarting an application
+	from a checkpoint.  Such restart requires that the processes in the
+	application have the same pids they had when the application was
+	checkpointed. When containers are nested, the processes within the
+	containers exist in multiple pid namespaces and hence have multiple
+	pids to specify during restart.
+
+	The @pids defines the set of pids that should be assigned to the child
+	process in its active and ancestor pid name spaces. The descendant pid
+	namespaces do not matter since a process does not have a pid in
+	descendant namespaces, unless the process is in a new pid namespace
+	in which case the process is a container-init (and must have the pid 1
+	in that namespace).
+
+	See CLONE_NEWPID section of clone(2) man page for details about pid
+	namespaces.
+
+	The order pids in @pids corresponds to the nesting order of pid-
+	namespaces, with @pids[0] corresponding to the init_pid_ns.
+
+	If a pid in the @pids list is 0, the kernel will assign the next
+	available pid in the pid namespace, for the process.
+
+	If a pid in the @pids list is non-zero, the kernel tries to assign
+	the specified pid in that namespace.  If that pid is already in use
+	by another process, the system call fails with -EBUSY.
+
+	On success, the system call returns the pid of the child process in
+	the parent's active pid namespace.
+
+	On failure, clone2() returns -1 and sets 'errno' to one of following
+	values (the child process is not created).
+
+	EPERM	Caller does not have the SYS_ADMIN privilege needed to excute
+		this call.
+
+	EINVAL	The number of pids specified in 'clone_args.nr_pids' exceeds
+		the current nesting level of parent process
+
+	EBUSY	A requested pid is in use by another process in that name space.
+
+Example:
+
+	pid_t pids[] = { 77, 99 };
+	struct clone_struct cs;
+
+	cs.flags = (u64) SIGCHLD;
+	cs.child_stack = (u64) setup_child_stack();
+	cs.nr_pids = 2;
+	cs.parent_tid = 0;
+	cs.child_tid = 0;
+
+	rc = syscall(__NR_clone2, &cs, pids);
+
+	if (rc < 0) {
+		perror("clone2()");
+		exit(1);
+	} else if (rc) {
+		/* Parent */
+	} else {
+		/* Child */
+	}
+

[...]

So @parent_tid and @child_tid are pointers to userspace memory and require 'u64' (and it won't hurt to make @reserved1 a 'u64' as well).

Also, for forward/backward compatibility, explicitly state in the documentation, and enforce in the kernel, that flags which are not defined must not be set, and that reserved{1,2} must remain 0.

Otherwise, looks great.

Oren.

I don't like this even more.

Pid namespaces are hierarchical _and_ anonymous, so simply set of numbers doesn't describe the final object.

struct pid isn't special, it's just another invariant if you like as far as C/R is concerned, but system call is made special wrt pids.

What will be in an image? I hope "struct kstate_image_pid" with several numbers and with references to such object from other places, so it seems natural to do alloc_pid() with needed numbers and that attach new shiny pid to where needed. But this clone_pid is only for task_struct's pids.

Subject: [RFC][v7][PATCH 1/9]: Factor out code to allocate pidmap page

To simplify alloc_pidmap(), move code to allocate a pid map page to a separate function.

Changelog[v3]:
- Earlier version of patchset called alloc_pidmap_page() from two places. But now its called from only one place. Even so, moving this code out into a separate function simplifies alloc_pidmap().
Changelog[v2]:
- (Matt Helsley, Dave Hansen) Have alloc_pidmap_page() return -ENOMEM on error instead of -1.

---
 kernel/pid.c |   45 ++++++++++++++++++++++++++++++---------------
 1 file changed, 30 insertions(+), 15 deletions(-)

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -122,9 +122,35 @@ static void free_pidmap(struct upid *upi
 	atomic_inc(&map->nr_free);
 }
 
+static int alloc_pidmap_page(struct pidmap *map)
+{
+	void *page;
+
+	if (likely(map->page))
+		return 0;
+
+	page = kzalloc(PAGE_SIZE, GFP_KERNEL);
+
+	/*
+	 * Free the page if someone raced with us installing it:
+	 */
+	spin_lock_irq(&pidmap_lock);
+	if (map->page)
+		kfree(page);
+	else
+		map->page = page;
+	spin_unlock_irq(&pidmap_lock);
+
+	if (unlikely(!map->page))
+		return -ENOMEM;
+
+	return 0;
+}
+
 static int alloc_pidmap(struct pid_namespace *pid_ns)
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
+	int rc;
 	struct pidmap *map;
 
 	pid = last + 1;

@@ -134,21 +160,10 @@ static int alloc_pidmap(struct pid_names
 	map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
 	max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
 	for (i = 0; i <= max_scan; ++i) {
-		if (unlikely(!map->page)) {
-			void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
-			/*
-			 * Free the page if someone raced with us
-			 * installing it:
-			 */
-			spin_lock_irq(&pidmap_lock);
-			if (map->page)
-				kfree(page);
-			else
-				map->page = page;
-			spin_unlock_irq(&pidmap_lock);
-			if (unlikely(!map->page))
-				break;
-		}
+		rc = alloc_pidmap_page(map);
+		if (rc)
+			break;
+
 		if (likely(atomic_read(&map->nr_free))) {
 			do {
 				if (!test_and_set_bit(offset, map->page)) {

Subject: [RFC][v7][PATCH 2/9]: Have alloc_pidmap() return actual error code

alloc_pidmap() can fail either because all pid numbers are in use or because memory allocation failed. With support for setting a specific pid number, alloc_pidmap() would also fail if either the given pid number is invalid or in use.

Rather than have callers assume -ENOMEM, have alloc_pidmap() return the actual error.

---
 kernel/fork.c |    5 +++--
 kernel/pid.c  |   14 +++++++++-----
 2 files changed, 12 insertions(+), 7 deletions(-)

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -1110,10 +1110,11 @@ static struct task_struct *copy_process(
 		goto bad_fork_cleanup_io;
 
 	if (pid != &init_struct_pid) {
-		retval = -ENOMEM;
 		pid = alloc_pid(p->nsproxy->pid_ns);
-		if (!pid)
+		if (IS_ERR(pid)) {
+			retval = PTR_ERR(pid);
 			goto bad_fork_cleanup_io;
+		}
 
 		if (clone_flags & CLONE_NEWPID) {
 			retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -150,7 +150,7 @@ static int alloc_pidmap_page(struct pidm
 static int alloc_pidmap(struct pid_namespace *pid_ns)
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
-	int rc;
+	int rc = -EAGAIN;
 	struct pidmap *map;
 
 	pid = last + 1;

@@ -189,12 +189,14 @@ static int alloc_pidmap(struct pid_names
 		} else {
 			map = &pid_ns->pidmap[0];
 			offset = RESERVED_PIDS;
-			if (unlikely(last == offset))
+			if (unlikely(last == offset)) {
+				rc = -EAGAIN;
 				break;
+			}
 		}
 		pid = mk_pid(pid_ns, map, offset);
 	}
-	return -1;
+	return rc;
 }
 
 int next_pidmap(struct pid_namespace *pid_ns, int last)

@@ -263,8 +265,10 @@ struct pid *alloc_pid(struct pid_namespa
 	struct upid *upid;
 
 	pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
-	if (!pid)
+	if (!pid) {
+		pid = ERR_PTR(-ENOMEM);
 		goto out;
+	}
 
 	tmp = ns;
 	for (i = ns->level; i >= 0; i--) {

@@ -299,7 +303,7 @@ out_free:
 		free_pidmap(pid->numbers + i);
 
 	kmem_cache_free(ns->pid_cachep, pid);
-	pid = NULL;
+	pid = ERR_PTR(nr);
 	goto out;
 }
 

From: Serge Hallyn <[email protected]>
Subject: [RFC][v7][PATCH 3/9] Make pid_max a pid_ns property

Remove the pid_max global, and make it a property of the pid_namespace. When a pid_ns is created, it inherits the parent's pid_ns.

Fixing up sysctl (trivial akin to ipc version, but potentially tedious to get right for all CONFIG* combinations) is left for later.

Changelog[v2]:
- Port to newer kernel
- Make pid_max a local variable in alloc_pidmap() to simplify code/patch

---
 include/linux/pid_namespace.h |    1 +
 kernel/pid.c                  |    4 ++--
 kernel/pid_namespace.c        |    1 +
 kernel/sysctl.c               |    4 ++--
 4 files changed, 6 insertions(+), 4 deletions(-)

Index: linux-2.6/include/linux/pid_namespace.h
===================================================================
@@ -30,6 +30,7 @@ struct pid_namespace {
 #ifdef CONFIG_BSD_PROCESS_ACCT
 	struct bsd_acct_struct *bacct;
 #endif
+	int pid_max;
 };
 
 extern struct pid_namespace init_pid_ns;

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -43,8 +43,6 @@ static struct hlist_head *pid_hash;
 static int pidhash_shift;
 struct pid init_struct_pid = INIT_STRUCT_PID;
 
-int pid_max = PID_MAX_DEFAULT;
-
 #define RESERVED_PIDS		300
 
 int pid_max_min = RESERVED_PIDS + 1;

@@ -78,6 +76,7 @@ struct pid_namespace init_pid_ns = {
 	.last_pid = 0,
 	.level = 0,
 	.child_reaper = &init_task,
+	.pid_max = PID_MAX_DEFAULT,
 };
 EXPORT_SYMBOL_GPL(init_pid_ns);
 

@@ -151,6 +150,7 @@ static int alloc_pidmap(struct pid_names
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
 	int rc = -EAGAIN;
+	int pid_max = pid_ns->pid_max;
 	struct pidmap *map;
 
 	pid = last + 1;

Index: linux-2.6/kernel/pid_namespace.c
===================================================================
@@ -87,6 +87,7 @@ static struct pid_namespace *create_pid_
 
 	kref_init(&ns->kref);
 	ns->level = level;
+	ns->pid_max = parent_pid_ns->pid_max;
 	ns->parent = get_pid_ns(parent_pid_ns);
 
 	set_bit(0, ns->pidmap[0].page);

Index: linux-2.6/kernel/sysctl.c
===================================================================
@@ -55,6 +55,7 @@
 
 #include <asm/uaccess.h>
 #include <asm/processor.h>
+#include <linux/pid_namespace.h>
 
 #ifdef CONFIG_X86
 #include <asm/nmi.h>

@@ -78,7 +79,6 @@ extern int max_threads;
 extern int core_uses_pid;
 extern int suid_dumpable;
 extern char core_pattern[];
-extern int pid_max;
 extern int min_free_kbytes;
 extern int pid_max_min, pid_max_max;
 extern int sysctl_drop_caches;

@@ -670,7 +670,7 @@ static struct ctl_table kern_table[] = {
 	{
 		.ctl_name	= KERN_PIDMAX,
 		.procname	= "pid_max",
-		.data		= &pid_max,
+		.data		= &init_pid_ns.pid_max,
 		.maxlen		= sizeof (int),
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec_minmax,

Subject: [RFC][v7][PATCH 4/9]: Add target_pid parameter to alloc_pidmap()

With support for setting a specific pid number for a process, alloc_pidmap() will need a 'target_pid' parameter.

Changelog[v6]:
- Separate target_pid > 0 case to minimize the number of checks needed.
Changelog[v3]:
- (Eric Biederman): Avoid set_pidmap() function. Added couple of checks for target_pid in alloc_pidmap() itself.
Changelog[v2]:
- (Serge Hallyn) Check for 'pid < 0' in set_pidmap().(Code actually checks for 'pid <= 0' for completeness).

---
 kernel/pid.c |   26 +++++++++++++++++++++-----
 1 file changed, 21 insertions(+), 5 deletions(-)

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -146,16 +146,22 @@ static int alloc_pidmap_page(struct pidm
 	return 0;
 }
 
-static int alloc_pidmap(struct pid_namespace *pid_ns)
+static int alloc_pidmap(struct pid_namespace *pid_ns, int target_pid)
 {
 	int i, offset, max_scan, pid, last = pid_ns->last_pid;
 	int rc = -EAGAIN;
 	int pid_max = pid_ns->pid_max;
 	struct pidmap *map;
 
-	pid = last + 1;
-	if (pid >= pid_max)
-		pid = RESERVED_PIDS;
+	if (target_pid) {
+		pid = target_pid;
+		if (pid < 0 || pid >= pid_max)
+			return -EINVAL;
+	} else {
+		pid = last + 1;
+		if (pid >= pid_max)
+			pid = RESERVED_PIDS;
+	}
 	offset = pid & BITS_PER_PAGE_MASK;
 	map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
 	max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;

@@ -164,6 +170,15 @@ static int alloc_pidmap(struct pid_names
 		if (rc)
 			break;
 
+		if (target_pid) {
+			rc = -EBUSY;
+			if (!test_and_set_bit(offset, map->page)) {
+				atomic_dec(&map->nr_free);
+				rc = pid;
+			}
+			return rc;
+		}
+
 		if (likely(atomic_read(&map->nr_free))) {
 			do {
 				if (!test_and_set_bit(offset, map->page)) {

@@ -196,6 +211,7 @@ static int alloc_pidmap(struct pid_names
 		}
 		pid = mk_pid(pid_ns, map, offset);
 	}
+
 	return rc;
 }
 

@@ -272,7 +288,7 @@ struct pid *alloc_pid(struct pid_namespa
 
 	tmp = ns;
 	for (i = ns->level; i >= 0; i--) {
-		nr = alloc_pidmap(tmp);
+		nr = alloc_pidmap(tmp, 0);
 		if (nr < 0)
 			goto out_free;
 

Subject: [RFC][v7][PATCH 5/9]: Add target_pids parameter to alloc_pid()

This parameter is currently NULL, but will be used in a follow-on patch.

---
 include/linux/pid.h |    2 +-
 kernel/fork.c       |    3 ++-
 kernel/pid.c        |    9 +++++++--
 3 files changed, 10 insertions(+), 4 deletions(-)

Index: linux-2.6/include/linux/pid.h
===================================================================
@@ -119,7 +119,7 @@ extern struct pid *find_get_pid(int nr);
 extern struct pid *find_ge_pid(int nr, struct pid_namespace *);
 int next_pidmap(struct pid_namespace *pid_ns, int last);
 
-extern struct pid *alloc_pid(struct pid_namespace *ns);
+extern struct pid *alloc_pid(struct pid_namespace *ns, pid_t *target_pids);
 extern void free_pid(struct pid *pid);
 
 /*

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -940,6 +940,7 @@ static struct task_struct *copy_process(
 	int retval;
 	struct task_struct *p;
 	int cgroup_callbacks_done = 0;
+	pid_t *target_pids = NULL;
 
 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 		return ERR_PTR(-EINVAL);

@@ -1110,7 +1111,7 @@ static struct task_struct *copy_process(
 		goto bad_fork_cleanup_io;
 
 	if (pid != &init_struct_pid) {
-		pid = alloc_pid(p->nsproxy->pid_ns);
+		pid = alloc_pid(p->nsproxy->pid_ns, target_pids);
 		if (IS_ERR(pid)) {
 			retval = PTR_ERR(pid);
 			goto bad_fork_cleanup_io;

Index: linux-2.6/kernel/pid.c
===================================================================
@@ -272,13 +272,14 @@ void free_pid(struct pid *pid)
 	call_rcu(&pid->rcu, delayed_put_pid);
 }
 
-struct pid *alloc_pid(struct pid_namespace *ns)
+struct pid *alloc_pid(struct pid_namespace *ns, pid_t *target_pids)
 {
 	struct pid *pid;
 	enum pid_type type;
 	int i, nr;
 	struct pid_namespace *tmp;
 	struct upid *upid;
+	int tpid;
 
 	pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
 	if (!pid) {

@@ -288,7 +289,11 @@ struct pid *alloc_pid(struct pid_namespa
 
 	tmp = ns;
 	for (i = ns->level; i >= 0; i--) {
-		nr = alloc_pidmap(tmp, 0);
+		tpid = 0;
+		if (target_pids)
+			tpid = target_pids[i];
+
+		nr = alloc_pidmap(tmp, tpid);
 		if (nr < 0)
 			goto out_free;
 

Subject: [RFC][v7][PATCH 6/9]: Add target_pids parameter to copy_process()

Add a 'target_pids' parameter to copy_process(). The new parameter will be used in a follow-on patch when clone_with_pids() is implemented.

---
 kernel/fork.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -935,12 +935,12 @@ static struct task_struct *copy_process(
 					unsigned long stack_size,
 					int __user *child_tidptr,
 					struct pid *pid,
+					pid_t *target_pids,
 					int trace)
 {
 	int retval;
 	struct task_struct *p;
 	int cgroup_callbacks_done = 0;
-	pid_t *target_pids = NULL;
 
 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 		return ERR_PTR(-EINVAL);

@@ -1319,7 +1319,7 @@ struct task_struct * __cpuinit fork_idle
 	struct pt_regs regs;
 
 	task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
-			    &init_struct_pid, 0);
+			    &init_struct_pid, NULL, 0);
 	if (!IS_ERR(task))
 		init_idle(task, cpu);
 

@@ -1342,6 +1342,7 @@ long do_fork(unsigned long clone_flags,
 	struct task_struct *p;
 	int trace = 0;
 	long nr;
+	pid_t *target_pids = NULL;
 
 	/*
 	 * Do some preliminary argument and permissions checking before we

@@ -1382,7 +1383,7 @@ long do_fork(unsigned long clone_flags,
 		trace = tracehook_prepare_clone(clone_flags);
 
 	p = copy_process(clone_flags, stack_start, regs, stack_size,
-			 child_tidptr, NULL, trace);
+			 child_tidptr, NULL, target_pids, trace);
 	/*
 	 * Do this prior waking up the new thread - the thread pointer
 	 * might get invalid after that point, if the thread exits quickly.

Subject: [RFC][v7][PATCH 7/9]: Define do_fork_with_pids()

do_fork_with_pids() is same as do_fork(), except that it takes an additional, 'pid_set', parameter. This parameter, currently unused, specifies the set of target pids of the process in each of its pid namespaces.

Changelog[v7]:
- Drop 'struct pid_set' object and pass in 'pid_t *target_pids' instead of 'struct pid_set *'.

Changelog[v6]:
- (Nathan Lynch, Arnd Bergmann, H. Peter Anvin, Linus Torvalds) Change 'pid_set.pids' to a 'pid_t pids[]' so size of 'struct pid_set' is constant across architectures.
- (Nathan Lynch) Change 'pid_set.num_pids' to 'unsigned int'.

Changelog[v4]:
- Rename 'struct target_pid_set' to 'struct pid_set' since it may be useful in other contexts.

Changelog[v3]:
- Fix "long-line" warning from checkpatch.pl

Changelog[v2]:
- To facilitate moving architecture-inpdendent code to kernel/fork.c pass in 'struct target_pid_set __user *' to do_fork_with_pids() rather than 'pid_t *' (next patch moves the arch-independent code to kernel/fork.c)

---
 include/linux/sched.h |    3 +++
 kernel/fork.c         |   17 +++++++++++++++--
 2 files changed, 18 insertions(+), 2 deletions(-)

Index: linux-2.6/include/linux/sched.h
===================================================================
@@ -2054,6 +2054,9 @@ extern int disallow_signal(int);
 
 extern int do_execve(char *, char __user * __user *, char __user * __user *, struct pt_regs *);
 extern long do_fork(unsigned long, unsigned long, struct pt_regs *, unsigned long, int __user *, int __user *);
+extern long do_fork_with_pids(unsigned long, unsigned long, struct pt_regs *,
+				unsigned long, int __user *, int __user *,
+				unsigned int, pid_t __user *);
 struct task_struct *fork_idle(int);
 
 extern void set_task_comm(struct task_struct *tsk, char *from);

Index: linux-2.6/kernel/fork.c
===================================================================
@@ -1332,12 +1332,14 @@ struct task_struct * __cpuinit fork_idle
  * It copies the process, and if successful kick-starts
  * it and waits for it to finish using the VM if required.
  */
-long do_fork(unsigned long clone_flags,
+long do_fork_with_pids(unsigned long clone_flags,
 	      unsigned long stack_start,
 	      struct pt_regs *regs,
 	      unsigned long stack_size,
 	      int __user *parent_tidptr,
-	      int __user *child_tidptr)
+	      int __user *child_tidptr,
+	      unsigned int num_pids,
+	      pid_t __user *upids)
 {
 	struct task_struct *p;
 	int trace = 0;

@@ -1440,6 +1442,17 @@ long do_fork(unsigned long clone_flags,
 	return nr;
 }
 
+long do_fork(unsigned long clone_flags,
+	      unsigned long stack_start,
+	      struct pt_regs *regs,
+	      unsigned long stack_size,
+	      int __user *parent_tidptr,
+	      int __user *child_tidptr)
+{
+	return do_fork_with_pids(clone_flags, stack_start, regs, stack_size,
+			parent_tidptr, child_tidptr, 0, NULL);
+}
+
 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
 #define ARCH_MIN_MMSTRUCT_ALIGN 0
 #endif

Subject: [RFC][v7][PATCH 8/9]: Define clone2() syscall

Container restart requires that a task have the same pid it had when it was checkpointed. When containers are nested the tasks within the containers exist in multiple pid namespaces and hence have multiple pids to specify during restart.

clone2(), intended for use during restart, is the same as clone(), except that it takes a 'pids' paramter. This parameter lets caller choose specific pid numbers for the child process, in the process's active and ancestor pid namespaces. (Descendant pid namespaces in general don't matter since processes don't have pids in them anyway, but see comments in copy_target_pids() regarding CLONE_NEWPID).

Clone2() system call also attempts to address a second limitation of the clone() system call. clone() is restricted to 32 clone flags and most (all ?) of these are in use. If a new clone flag is needed, we will be forced to define a new variant of the clone() system call.

To prevent unprivileged processes from misusing this interface, clone2() currently needs CAP_SYS_ADMIN, when the 'pids' parameter is non-NULL.

See Documentation/clone2 in next patch for more details of clone2() and an example of its usage.

NOTE:
System calls are restricted to 6 parameters and the number and sizes of parameters needed for sys_clone2() exceed 6 integers. The new prototype works around this restriction while providing some flexibility if clone2() needs to be further extended in the future.
TODO:
- May need additional sanity checks in do_fork_with_pids().

Changelog[v7]:
- [Peter Zijlstra, Arnd Bergmann] Rename system call to clone2() and group parameters into a new 'struct clone_struct' object.

Changelog[v6]:
- (Nathan Lynch, Arnd Bergmann, H. Peter Anvin, Linus Torvalds) Change 'pid_set.pids' to a 'pid_t pids[]' so size of 'struct pid_set' is constant across architectures.
- (Nathan Lynch) Change pid_set.num_pids to unsigned and remove 'unum_pids < 0' check.

Changelog[v4]:
- (Oren Laadan) rename 'struct target_pid_set' to 'struct pid_set'

Changelog[v3]:
- (Oren Laadan) Allow CLONE_NEWPID flag (by allocating an extra pid in the target_pids[] list and setting it 0. See copy_target_pids()).
- (Oren Laadan) Specified target pids should apply only to youngest pid-namespaces (see copy_target_pids())
- (Matt Helsley) Update patch description.

Changelog[v2]:
- Remove unnecessary printk and add a note to callers of copy_target_pids() to free target_pids.
- (Serge Hallyn) Mention CAP_SYS_ADMIN restriction in patch description.
- (Oren Laadan) Add checks for 'num_pids < 0' (return -EINVAL) and 'num_pids == 0' (fall back to normal clone()).
- Move arch-independent code (sanity checks and copy-in of target-pids) into kernel/fork.c and simplify sys_clone_with_pids()

Changelog[v1]:
- Fixed some compile errors (had fixed these errors earlier in my git tree but had not refreshed patches before emailing them)

---
 arch/x86/include/asm/syscalls.h    |    1 
 arch/x86/include/asm/unistd_32.h   |    1 
 arch/x86/kernel/process_32.c       |   31 +++++++++++
 arch/x86/kernel/syscall_table_32.S |    1 
 include/linux/types.h              |   10 +++
 kernel/fork.c                      |   96 ++++++++++++++++++++++++++++++++++++-
 6 files changed, 139 insertions(+), 1 deletion(-)

Index: linux-2.6/arch/x86/include/asm/syscalls.h
===================================================================
@@ -55,6 +55,7 @@ struct sel_arg_struct;
 struct oldold_utsname;
 struct old_utsname;
 
+asmlinkage long sys_clone2(struct clone_struct __user *cs, pid_t *pids);
 asmlinkage long sys_mmap2(unsigned long, unsigned long, unsigned long,
 			  unsigned long, unsigned long, unsigned long);
 asmlinkage int old_mmap(struct mmap_arg_struct __user *);

Index: linux-2.6/arch/x86/include/asm/unistd_32.h
===================================================================
@@ -342,6 +342,7 @@
 #define __NR_pwritev		334
 #define __NR_rt_tgsigqueueinfo	335
 #define __NR_perf_counter_open	336
+#define __NR_clone2		337
 
 #ifdef __KERNEL__
 

Index: linux-2.6/arch/x86/kernel/process_32.c
===================================================================
@@ -443,6 +443,37 @@ int sys_clone(struct pt_regs *regs)
 	return do_fork(clone_flags, newsp, regs, 0, parent_tidptr, child_tidptr);
 }
 
+asmlinkage long
+sys_clone2(struct clone_struct __user *ucs, pid_t __user *pids)
+{
+	int rc;
+	struct clone_struct kcs;
+	unsigned long clone_flags;
+	int __user *parent_tid_ptr;
+	int __user *child_tid_ptr;
+	unsigned long __user child_stack_base;
+	struct pt_regs *regs;
+
+	rc = copy_from_user(&ucs, &kcs, sizeof(kcs));
+	if (rc)
+		return -EFAULT;
+
+	/*
+	 * TODO: Convert clone_flags to 64-bit
+	 */
+	clone_flags = (unsigned long)kcs.flags;
+	child_stack_base = (unsigned long)kcs.child_stack;
+	parent_tid_ptr = &ucs.parent_tid;
+	child_tid_ptr =  &ucs.child_tid;
+	regs = task_pt_regs(current);
+
+	if (!child_stack_base)
+		child_stack_base = user_stack_pointer(regs);
+
+	return do_fork_with_pids(clone_flags, child_stack_base, regs, 0,
+			parent_tid_ptr, child_tid_ptr, kcs.nr_pids, pids);
+}
+
 /*
  * sys_execve() executes a new program.
  */

Index: linux-2.6/arch/x86/kernel/syscall_table_32.S
===================================================================
@@ -336,3 +336,4 @@ ENTRY(sys_call_table)
 	.long sys_pwritev
 	.long sys_rt_tgsigqueueinfo	/* 335 */
 	.long sys_perf_counter_open
+	.long sys_clone2

Index: linux-2.6/include/linux/types.h
===================================================================
@@ -204,6 +204,16 @@ struct ustat {
 	char			f_fpack[6];
 };
 
+struct clone_struct {
+	u64 flags;
+	u64 child_stack;
+	u32 nr_pids;
+	u32 parent_tid;
+	u32 child_tid;
+	u32 reserved1;
+	u64 reserved2;
+};
+
 #endif	/* __KERNEL__ */
 #endif /*  __ASSEMBLY__ */
 #endif /* _LINUX_TYPES_H */

It will very likely break ia64, which defines CONFIG_HAVE_ARCH_TRACEHOOK and already has sys_clone2().

Louis

Dr Louis Rilling			Kerlabs
Skype: louis.rilling			Batiment Germanium
Phone: (+33|0) 6 80 89 08 23		80 avenue des Buttes de Coesmes
http://www.kerlabs.com/			35700 Rennes

_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linux-foundation.org/mailman/listinfo/containers

It will very likely break ia64, which defines CONFIG_HAVE_ARCH_TRACEHOOK and already has sys_clone2().

Louis

Dr Louis Rilling			Kerlabs
Skype: louis.rilling			Batiment Germanium
Phone: (+33|0) 6 80 89 08 23		80 avenue des Buttes de Coesmes
http://www.kerlabs.com/			35700 Rennes

-> sys_clone_ext() ?

Louis

Dr Louis Rilling			Kerlabs
Skype: louis.rilling			Batiment Germanium
Phone: (+33|0) 6 80 89 08 23		80 avenue des Buttes de Coesmes
http://www.kerlabs.com/			35700 Rennes

_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linux-foundation.org/mailman/listinfo/containers

-> sys_clone_ext() ?

Louis

Dr Louis Rilling			Kerlabs
Skype: louis.rilling			Batiment Germanium
Phone: (+33|0) 6 80 89 08 23		80 avenue des Buttes de Coesmes
http://www.kerlabs.com/			35700 Rennes

Ccing [email protected] and linux-api on this thread.

Louis Rilling [[email protected]] wrote:

| > It will very likely break ia64, which defines CONFIG_HAVE_ARCH_TRACEHOOK and
| > already has sys_clone2().
| 
| -> sys_clone_ext() ?
| 
| Louis

How about spelling out extended and calling it clone_extended() ?

The other options I can think of are clone_with_pids() and clone3().

Thanks for your feedback.

Sukadev
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Why add a new syscall at all instead of just using a new CLONE_* flag to indicate that the argument layout is different?
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

What an absolutely atrociously bad idea.

We already have a syscall layer which is painful to thunk in places, and this would make it much worse.

-hpa

--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

syscalls are cheap as well.
cheaper than decades of dealing with such multiplexer mess ;/

Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Well, I'd agree, except the clone flags really _are_ about multiplexer issues, and the new flag woudln't really change anything.

If the new system call actually had appreciably separate code-paths, I'd buy the "multiplexer" argument. But it doesn't really. It's going to call down to the same basic clone functionality, and the core clone code ends up de-multiplexing the cases anyway.

So this would not at all be like the socket calls (to pick the traditional Linux system call multiplexing example) in that sense.

Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

That's not the main issue here, though. The main issue is that the prototype of the function now depends on one of its arguments, which is absolute hell for anything that needs to thunk arguments in a systematic way, which we have to do on several architectures, and which would be useful to be able to do for others, too.

-hpa

--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Ok, I agree with that. The kernel side is easy (we have magic calling conventions there and need to turn registers into arguments anyway before you get to the shared code), but your point about the user side prototype is valid.

However, that could easily be handled by just having a extended_clone() prototype that then sets the CLONE_EXTINFO (or whatever) bit in the flags. I think most of the time the clone() stuff needs special user-level wrappers anyway to handle the stack setup etc, no?

In other words, what I'd suggest we could do is

- the kernel "do_fork()" interface would be made to have the "extended" format by default - so the _kernel_ never has two formats in its generic logic.

- the "sys_clone()" system call, that already needs to munge the user mode registers into the "do_fork()" format, would be the one that recognizes the new flag and copies the extended data from user mode memory to the extended info mode.

Then each architecture would need to update it's "sys_clone()" function to take advantage of the new extended format, but that's something that the new system call would have had to do anyway, so that's not an added burden in any way.

Hmm?

I don't feel horribly strongly about this, and as far as I'm concerned it's fine to also do it as a new system call too (we already have 'fork()' and 'vfork()' as special case interfaces to do_fork() - the new 'extended clone' would be no different).

I just think that Roland is correct that if the new extended fork handles the "no new info" case itself _anyway_, then there is no upside to making it a new system call, since the complexity is the same as just extending the old one.

Linus

I think it would also apply to kernel-side munging. It's quite possibly you're right in that clone is such a special case anyway, but it seems pointless to make it more special in the short bus sort of way even if it is possible.

Let's just make it another system call. It doesn't have any downside that I can see, might prevent problems, and avoids setting a bad precedent that someone can misinterpret.

-hpa

--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

One more argument for this is that the new code is architecture independent using user_stack_pointer(), while the original sys_clone is highly architecture specific, which is a source for bugs when trying to extend it.

	Arnd <><

--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Umm. I don't think that is possible.

You need architecture-specific code to even get access to all registers to copy and get a signal-handler-compatible stack frame. See for example arch/alpha/kernel/entry.S with the switch-stack thing etc. I don't think there is any way to make that even remotely architecture-neutral.

Linus

Right, you still need to save all the registers from the entry code. I was under the wrong assumption that task_pt_regs(current) would give the full register set on all architectures.

However, I'd still hope that a new system call can be defined in a way that you only need to have an assembly wrapper to save the full pt_regs, but no arch specific code to get the syscall arguments out of that again. In do_clone(), you need a pointer to pt_regs and the user stack pointer, but that can be generated from user_stack_pointer(regs).

Does task_pt_regs(current) give the right pointer on all architectures or do we also need to pass the regs into the syscall?

	Arnd <><

I don't think it can. You don't know what the system call stack layout is.

I do not believe that it gives the right pointer in general. In fact, I can guarantee it doesn't. Even on x86 it only works for certain contexts (non-vm86 mode at a minimum), and on architectures like alpha it's not at all sufficient, because even if you can locate the 'pt_regs' structure, you _also_ need the extra guarantees of the pt_regs being next to the extended signal state register structure - and that only happens for magic sequences like signal handling and explicit setups like fork/clone.

So I do repeat: if you think you can do all of this in generic code, then you're sadly and totally mistaken. Don't even try. It may work on some architectures, but it's simply fundamentally _wrong_.

Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

I don't think it can. You don't know what the system call stack layout is.

I do not believe that it gives the right pointer in general. In fact, I can guarantee it doesn't. Even on x86 it only works for certain contexts (non-vm86 mode at a minimum), and on architectures like alpha it's not at all sufficient, because even if you can locate the 'pt_regs' structure, you _also_ need the extra guarantees of the pt_regs being next to the extended signal state register structure - and that only happens for magic sequences like signal handling and explicit setups like fork/clone.

So I do repeat: if you think you can do all of this in generic code, then you're sadly and totally mistaken. Don't even try. It may work on some architectures, but it's simply fundamentally _wrong_.

Linus

One more argument for this is that the new code is architecture independent using user_stack_pointer(), while the original sys_clone is highly architecture specific, which is a source for bugs when trying to extend it.

	Arnd <><

I think it would also apply to kernel-side munging. It's quite possibly you're right in that clone is such a special case anyway, but it seems pointless to make it more special in the short bus sort of way even if it is possible.

Let's just make it another system call. It doesn't have any downside that I can see, might prevent problems, and avoids setting a bad precedent that someone can misinterpret.

-hpa

The glibc prototype for clone uses ... and is not a direct map to the syscall args anyway. So that would not change for adding more optional args enabled by certain flags, as it did not change to add the tid pointer arguments before. But indeed the library function would have to change to pass on additional or different args to the existing syscall.

Thanks,
Roland

The glibc prototype for clone uses ... and is not a direct map to the syscall args anyway. So that would not change for adding more optional args enabled by certain flags, as it did not change to add the tid pointer arguments before. But indeed the library function would have to change to pass on additional or different args to the existing syscall.

Thanks,
Roland
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

That's not the main issue here, though. The main issue is that the prototype of the function now depends on one of its arguments, which is absolute hell for anything that needs to thunk arguments in a systematic way, which we have to do on several architectures, and which would be useful to be able to do for others, too.

-hpa

Well, I'd agree, except the clone flags really _are_ about multiplexer issues, and the new flag woudln't really change anything.

If the new system call actually had appreciably separate code-paths, I'd buy the "multiplexer" argument. But it doesn't really. It's going to call down to the same basic clone functionality, and the core clone code ends up de-multiplexing the cases anyway.

So this would not at all be like the socket calls (to pick the traditional Linux system call multiplexing example) in that sense.

Linus

It really comes down to wanting all the dispatch to happen in one central place.

-hpa

It really comes down to wanting all the dispatch to happen in one central place.

-hpa

What an absolutely atrociously bad idea.

We already have a syscall layer which is painful to thunk in places, and this would make it much worse.

-hpa

Why add a new syscall at all instead of just using a new CLONE_* flag to indicate that the argument layout is different?

I like clone3(), or clone_new() ?

or even better -- how about xerox() :p

Oren.

I like clone3(), or clone_new() ?

or even better -- how about xerox() :p

Oren.

Ccing [email protected] and linux-api on this thread.

Louis Rilling [[email protected]] wrote:

| > It will very likely break ia64, which defines CONFIG_HAVE_ARCH_TRACEHOOK and
| > already has sys_clone2().
| 
| -> sys_clone_ext() ?
| 
| Louis

How about spelling out extended and calling it clone_extended() ?

The other options I can think of are clone_with_pids() and clone3().

Thanks for your feedback.

Sukadev

Subject: [RFC][v7][PATCH 9/9]: Document clone2() syscall

This gives a brief overview of the clone2() system call. We should eventually describe more details in existing clone(2) man page or in a new man page.

Changelog[v7]:
- Rename clone_with_pids() to clone2()
- Changes to reflect new prototype of clone2() (using clone_struct).

---
 Documentation/clone2 |   85 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 85 insertions(+)

Index: linux-2.6/Documentation/clone2
===================================================================
@@ -0,0 +1,85 @@
+
+struct clone_struct {
+	u64 flags;
+	u64 child_stack;
+	u32 nr_pids;
+	u32 parent_tid;
+	u32 child_tid;
+	u32 reserved1;
+	u64 reserved2;
+};
+
+clone2(struct clone_struct * __user clone_args, pid_t * __user pids)
+
+	In addition to doing everything that clone() system call does,
+	the clone2() system call:
+
+		- allows additional clone flags (all 32 bits in the flags
+		  parameter to clone() are in use)
+
+		- allows user to specify a pid for the child process in its
+		  active and ancestor pid name spaces.
+
+	This system call is meant to be used when restarting an application
+	from a checkpoint.  Such restart requires that the processes in the
+	application have the same pids they had when the application was
+	checkpointed. When containers are nested, the processes within the
+	containers exist in multiple pid namespaces and hence have multiple
+	pids to specify during restart.
+
+	The @pids defines the set of pids that should be assigned to the child
+	process in its active and ancestor pid name spaces. The descendant pid
+	namespaces do not matter since a process does not have a pid in
+	descendant namespaces, unless the process is in a new pid namespace
+	in which case the process is a container-init (and must have the pid 1
+	in that namespace).
+
+	See CLONE_NEWPID section of clone(2) man page for details about pid
+	namespaces.
+
+	The order pids in @pids corresponds to the nesting order of pid-
+	namespaces, with @pids[0] corresponding to the init_pid_ns.
+
+	If a pid in the @pids list is 0, the kernel will assign the next
+	available pid in the pid namespace, for the process.
+
+	If a pid in the @pids list is non-zero, the kernel tries to assign
+	the specified pid in that namespace.  If that pid is already in use
+	by another process, the system call fails with -EBUSY.
+
+	On success, the system call returns the pid of the child process in
+	the parent's active pid namespace.
+
+	On failure, clone2() returns -1 and sets 'errno' to one of following
+	values (the child process is not created).
+
+	EPERM	Caller does not have the SYS_ADMIN privilege needed to excute
+		this call.
+
+	EINVAL	The number of pids specified in 'clone_args.nr_pids' exceeds
+		the current nesting level of parent process
+
+	EBUSY	A requested pid is in use by another process in that name space.
+
+Example:
+
+	pid_t pids[] = { 77, 99 };
+	struct clone_struct cs;
+
+	cs.flags = (u64) SIGCHLD;
+	cs.child_stack = (u64) setup_child_stack();
+	cs.nr_pids = 2;
+	cs.parent_tid = 0;
+	cs.child_tid = 0;
+
+	rc = syscall(__NR_clone2, &cs, pids);
+
+	if (rc < 0) {
+		perror("clone2()");
+		exit(1);
+	} else if (rc) {
+		/* Parent */
+	} else {
+		/* Child */
+	}
+

Hi,

We have a separate mailing list ([email protected]) where new kernel APIs are (or were?) meant to be discussed/checked/tested.

Maybe Michael Kerrisk would care (or would have cared?) about this.

I don't see [email protected] listed in MAINTAINERS, but it is referred to in Documentation/HOWTO and Documentation/SubmitChecklist. Does it need to be listed in MAINTAINERS?
(oh, you didn't read Documentation/SubmitChecklist ??)

Anyway, please cc: [email protected] on future patches like this series.

AFAIK, ia64 already have clone2() systemcall and it is imcompatible this.

Hi,

We have a separate mailing list ([email protected]) where new kernel APIs are (or were?) meant to be discussed/checked/tested.

Maybe Michael Kerrisk would care (or would have cared?) about this.

I don't see [email protected] listed in MAINTAINERS,
but it is referred to in Documentation/HOWTO and Documentation/SubmitChecklist. Does it need to be listed in MAINTAINERS?
(oh, you didn't read Documentation/SubmitChecklist ??)

Anyway, please cc: [email protected] on future patches like this series.

AFAIK, ia64 already have clone2() systemcall and it is imcompatible this.

[...]

So @parent_tid and @child_tid are pointers to userspace memory and require 'u64' (and it won't hurt to make @reserved1 a 'u64' as well).

Also, for forward/backward compatibility, explicitly state in the documentation, and enforce in the kernel, that flags which are not defined must not be set, and that reserved{1,2} must remain 0.

Otherwise, looks great.

Oren.

Oren Laadan [[email protected]] wrote:

| 
| 
| Sukadev Bhattiprolu wrote:
| > Based on these requirements and constraints, we explored a couple of system
| > call interfaces (in earlier versions of this patchset) and currently define
| > the system call as:
| > 
| > 	struct clone_struct {
| > 		u64 flags;
| > 		u64 child_stack;
| > 		u32 nr_pids;
| > 		u32 parent_tid;
| > 		u32 child_tid;
| 
| So @parent_tid and @child_tid are pointers to userspace memory and
| require 'u64' (and it won't hurt to make @reserved1 a 'u64' as well).

No, as Arnd pointed out, we already pass in a pointer to 'struct clone_struct' and the kernel can use that pointer to copy the parent and child tids.

| 
| > 		u32 reserved1;
| > 		u64 reserved2;
| > 	};
| > 
| 
| Also, for forward/backward compatibility, explicitly state in the
| documentation, and enforce in the kernel, that flags which are not
| defined must not be set, and that reserved{1,2} must remain 0.

Good idea. Will do.

Thanks,

Sukadev

Oren Laadan [[email protected]] wrote:

| 
| 
| Sukadev Bhattiprolu wrote:
| > === NEW CLONE() SYSTEM CALL:
| > 
| > To support application checkpoint/restart, a task must have the same pid it
| > had when it was checkpointed.  When containers are nested, the tasks within
| > the containers exist in multiple pid namespaces and hence have multiple pids
| > to specify during restart.
| > 
| > This patchset implements a new system call, clone2() that lets a process
| > specify the pids of the child process.
| > 
| > Patches 1 through 6 are helper patches, needed for choosing a pid for the
| > child process.
| > 
| > Patch 8 defines a prototype of the new system call. Patch 9 adds some
| > documentation on the new system call, some/all of which will eventually
| > go into a man page.
| > 
| 
| [...]
| 
| > 
| > Based on these requirements and constraints, we explored a couple of system
| > call interfaces (in earlier versions of this patchset) and currently define
| > the system call as:
| > 
| > 	struct clone_struct {
| > 		u64 flags;
| > 		u64 child_stack;
| > 		u32 nr_pids;
| > 		u32 parent_tid;
| > 		u32 child_tid;
| 
| So @parent_tid and @child_tid are pointers to userspace memory and
| require 'u64' (and it won't hurt to make @reserved1 a 'u64' as well).

Well, if we make parent_tid and child_tid u64, we could move reserved1 after ->nr_pids and leave it as a 32-bit value.

| 
| > 		u32 reserved1;
| > 		u64 reserved2;
| > 	};
| > 
| 
| Also, for forward/backward compatibility, explicitly state in the
| documentation, and enforce in the kernel, that flags which are not
| defined must not be set, and that reserved{1,2} must remain 0.

Agree with checking for reserved1 and reserved2.

We currently don't check for invalid clone_flags - we just ignore them. Adding checks like

if (fls(kcs.flags) > fls(CLONE_LAST_FLAG))

would assume we always use bits in order (while it seems to make sense, to use them in order, we don't seem to have done so in the past).

Alternatively we could define a CLONE_FLAG_MASK of valid flags and update the mask when each new clone flag is added.

But do we really need to check for invalid flags ?

Sukadev

Oren Laadan [[email protected]] wrote:

| 
| 
| Sukadev Bhattiprolu wrote:
| > Based on these requirements and constraints, we explored a couple of system
| > call interfaces (in earlier versions of this patchset) and currently define
| > the system call as:
| > 
| > 	struct clone_struct {
| > 		u64 flags;
| > 		u64 child_stack;
| > 		u32 nr_pids;
| > 		u32 parent_tid;
| > 		u32 child_tid;
| 
| So @parent_tid and @child_tid are pointers to userspace memory and
| require 'u64' (and it won't hurt to make @reserved1 a 'u64' as well).

No, as Arnd pointed out, we already pass in a pointer to 'struct clone_struct' and the kernel can use that pointer to copy the parent and child tids.

| 
| > 		u32 reserved1;
| > 		u64 reserved2;
| > 	};
| > 
| 
| Also, for forward/backward compatibility, explicitly state in the
| documentation, and enforce in the kernel, that flags which are not
| defined must not be set, and that reserved{1,2} must remain 0.

Good idea. Will do.

Thanks,

Sukadev

In this form, you place a constraints on where userspace may place the {parent,child}_tid variable, and require that this particular clone_struct remain valid memory in the parent until the child terminates. This may break existing programs that use this (threads libraries ?)

Oren.

In this form, you place a constraints on where userspace may place the {parent,child}_tid variable, and require that this particular clone_struct remain valid memory in the parent until the child terminates. This may break existing programs that use this (threads libraries ?)

Oren.

No existing program uses sys_clone2, and the kernel function may well differ from the user space calling conventions, which are not bound by the six-argument limitation.

So a clone2 library call could set up the structure with the arguments to the real syscall, call into the kernel and copy the output data back into the pointers it was given by the user.

	Arnd <><

No existing program uses sys_clone2, and the kernel function may well differ from the user space calling conventions, which are not bound by the six-argument limitation.

So a clone2 library call could set up the structure with the arguments to the real syscall, call into the kernel and copy the output data back into the pointers it was given by the user.

	Arnd <><

That may work well for parent_tid, however child_tid is also kept on the task_struct and written to when the child exits, and there is no explicit user-space wrapper on that.

Also, I may be mistaken, but I thought that the idea of these was that the kernel writes them to user space, so other threads may see them quickly, _before_ the parent returns to userspace; otherwise, the parent (at the library) could himself copy the return value.

Oren.

That may work well for parent_tid, however child_tid is also kept on the task_struct and written to when the child exits, and there is no explicit user-space wrapper on that.

Also, I may be mistaken, but I thought that the idea of these was that the kernel writes them to user space, so other threads may see them quickly, _before_ the parent returns to userspace; otherwise, the parent (at the library) could himself copy the return value.

Oren.

Oren Laadan [[email protected]] wrote:

| 
| 
| Sukadev Bhattiprolu wrote:
| > === NEW CLONE() SYSTEM CALL:
| > 
| > To support application checkpoint/restart, a task must have the same pid it
| > had when it was checkpointed.  When containers are nested, the tasks within
| > the containers exist in multiple pid namespaces and hence have multiple pids
| > to specify during restart.
| > 
| > This patchset implements a new system call, clone2() that lets a process
| > specify the pids of the child process.
| > 
| > Patches 1 through 6 are helper patches, needed for choosing a pid for the
| > child process.
| > 
| > Patch 8 defines a prototype of the new system call. Patch 9 adds some
| > documentation on the new system call, some/all of which will eventually
| > go into a man page.
| > 
| 
| [...]
| 
| > 
| > Based on these requirements and constraints, we explored a couple of system
| > call interfaces (in earlier versions of this patchset) and currently define
| > the system call as:
| > 
| > 	struct clone_struct {
| > 		u64 flags;
| > 		u64 child_stack;
| > 		u32 nr_pids;
| > 		u32 parent_tid;
| > 		u32 child_tid;
| 
| So @parent_tid and @child_tid are pointers to userspace memory and
| require 'u64' (and it won't hurt to make @reserved1 a 'u64' as well).

Well, if we make parent_tid and child_tid u64, we could move reserved1 after ->nr_pids and leave it as a 32-bit value.

| 
| > 		u32 reserved1;
| > 		u64 reserved2;
| > 	};
| > 
| 
| Also, for forward/backward compatibility, explicitly state in the
| documentation, and enforce in the kernel, that flags which are not
| defined must not be set, and that reserved{1,2} must remain 0.

Agree with checking for reserved1 and reserved2.

We currently don't check for invalid clone_flags - we just ignore them. Adding checks like

if (fls(kcs.flags) > fls(CLONE_LAST_FLAG))

would assume we always use bits in order (while it seems to make sense, to use them in order, we don't seem to have done so in the past).

Alternatively we could define a CLONE_FLAG_MASK of valid flags and update the mask when each new clone flag is added.

But do we really need to check for invalid flags ?

Sukadev

Sure. In any case, won't hurt to leave large reserved space - someone may be thankful for it in the future ;)

I'd go for a a mask.

The idea is that we want to educate userspace to _not_ use unused flags now. For if userspace sets an unused flag now and we let it be, the application will break when we give meaning to that flag.

Oren.

Sure. In any case, won't hurt to leave large reserved space - someone may be thankful for it in the future ;)

I'd go for a a mask.

The idea is that we want to educate userspace to _not_ use unused flags now. For if userspace sets an unused flag now and we let it be, the application will break when we give meaning to that flag.

Oren.

I don't like this even more.

Pid namespaces are hierarchical _and_ anonymous, so simply set of numbers doesn't describe the final object.

struct pid isn't special, it's just another invariant if you like as far as C/R is concerned, but system call is made special wrt pids.

What will be in an image? I hope "struct kstate_image_pid" with several numbers and with references to such object from other places, so it seems natural to do alloc_pid() with needed numbers and that attach new shiny pid to where needed. But this clone_pid is only for task_struct's pids.

Sure pid namespaces are anonymous, but we will give each an 'objref' valid only for a checkpoint image, and store the relationship between pid namespaces based on those objrefs. Basically the same way that user structs and hierarchical user namespaces are handled right now.

Sure pid namespaces are anonymous, but we will give each an 'objref' valid only for a checkpoint image, and store the relationship between pid namespaces based on those objrefs. Basically the same way that user structs and hierarchical user namespaces are handled right now.

OK, that's certainly doable.

You're commiting yourself to creation of tasks in userspace if this goes in. :-\
Which can let you into putting wrong kind of relations into image. IIRC, clone_flags were in image (still?), but tomorrow kernel will get new way to acquire, say, uts_ns, which, in theory, can't be described by a set of consecutive clones, so, you'll have to fixup something in kernel.

OK, that's certainly doable.

You're commiting yourself to creation of tasks in userspace if this goes in. :-\
Which can let you into putting wrong kind of relations into image. IIRC, clone_flags were in image (still?), but tomorrow kernel will get new way to acquire, say, uts_ns, which, in theory, can't be described by a set of consecutive clones, so, you'll have to fixup something in kernel.

A malicious user can put "wrong" king of relations into the image, regardless of whether the tasks are created in the kernel or in userspace. As long as the creation follows the "instructions" in the image, the result would be the same.

The only thing enforced in user space is task relations, threads and (as a by-product) session id's. The rest are refined in the kernel. This includes uts_ns, for example.

(FWIW, _any_ clone relationships can be described by a set of clones. In particular because that's how they were constructed in the first place).

Oren.

A malicious user can put "wrong" king of relations into the image, regardless of whether the tasks are created in the kernel or in userspace. As long as the creation follows the "instructions" in the image, the result would be the same.

The only thing enforced in user space is task relations, threads and (as a by-product) session id's. The rest are refined in the kernel. This includes uts_ns, for example.

(FWIW, _any_ clone relationships can be described by a set of clones. In particular because that's how they were constructed in the first place).

Oren.

Wrong as in "fundamentally wrong", not malicious.
In case of uts_ns, the correct data to put into image is "task uses this uts_ns", not "at this point do clone(CLONE_NEWUTS)".

BTW, now I'm convinced that nsproxy should not even be mentioned be in an image, it's irrelevant technical detail, not future-proof at all.

Wrong as in "fundamentally wrong", not malicious.
In case of uts_ns, the correct data to put into image is "task uses this uts_ns", not "at this point do clone(CLONE_NEWUTS)".

BTW, now I'm convinced that nsproxy should not even be mentioned be in an image, it's irrelevant technical detail, not future-proof at all.

So we are in total agreement: that's how it is done now.

Only task creation per-se, including pid-ns (future work) is done in userspace. Network namespaces will probably be created in userspace but attached to tasks in the kernel. Remaining namespaces are covered in the kernel the way you described.

It's helpful (as is more efficient) to keep it now. We can always decide to ignore it in the future.

Thanks,

Oren.

So we are in total agreement: that's how it is done now.

Only task creation per-se, including pid-ns (future work) is done in userspace. Network namespaces will probably be created in userspace but attached to tasks in the kernel. Remaining namespaces are covered in the kernel the way you described.

It's helpful (as is more efficient) to keep it now. We can always decide to ignore it in the future.

Thanks,

Oren.