Skip to contentSkip to conversation

Patches in a sake of checkpoint/restore, procfs and prctls

PATCH In review Cyrill Gorcunov 5 participants 13 messages Dec 12, 2011

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover Patches in a sake of checkpoint/restore, procfs and prctls

Hi,

while /proc/pid/children patch is still under review/rework other patches were fixed and I hope in good shape. The first one introduces CONFIG_CHECKPOINT_RESTORE Kconfig symbol which should be a lever to turn on/off all c/r related features for those who not need it. In particular new prctl codes are covered by.

Eric pointed that /proc/<pid>/stat enhancement might be a dangerous one from user-space point of view, so while I've successfuly tested it, it doesn't mean I've covered every single user-space utility which might use this file, and if it's still considered to be pretty harmful change -- I'll hapily move to /proc/pid/statm or whatever. The only reason it was introduced in "stat" -- we already have mm->start_code and etc there, so I wanted to have them in one place, not sprinkled over several files.

Any complains are welcome as usual.

Cyrill

#1 lore raw

PATCH 1/3 Kconfig: Introduce CHECKPOINT_RESTORE symbol

In a sake of checkpoint/restore we need auxilary features being compiled into the kernel, such as additional prctl codes, /proc/<pid>/map_files and etc...
but same time these features are not mandatory for a regular kernel so CHECKPOINT_RESTORE config symbol should bring a way to disable them all at once if one wish to get rid of additional functionality.

---
 init/Kconfig |   11 +++++++++++
 1 file changed, 11 insertions(+)
init/Kconfig +11 -0
Index: linux-2.6.git/init/Kconfig
===================================================================
@@ -773,6 +773,17 @@ config DEBUG_BLK_CGROUP
 
 endif # CGROUPS
 
+config CHECKPOINT_RESTORE
+	bool "Checkpoint/restore support" if EXPERT
+	default n
+	help
+	  Enables additional kernel features in a sake of checkpoint/restore.
+	  In particular it adds auxiliary prctl codes to setup process text,
+	  data and heap segment sizes, and a few additional /proc filesystem
+	  entries.
+
+	  If unsure, say N here.
+
 menuconfig NAMESPACES
 	bool "Namespaces support" if EXPERT
 	default !EXPERT
Reviewed-by on v1 1/3
#2 lore raw

Re: [patch 1/3] Kconfig: Introduce CHECKPOINT_RESTORE symbol

Quoted 9 lines · Cyrill Gorcunov

In a sake of checkpoint/restore we need auxilary features being compiled into the kernel, such as additional prctl codes, /proc/<pid>/map_files and etc...
but same time these features are not mandatory for a regular kernel so CHECKPOINT_RESTORE config symbol should bring a way to disable them all at once if one wish to get rid of additional functionality.

Signature · 2 lines
Kees Cook
ChromeOS Security
#3 lore raw

PATCH 2/3 fs, proc: Add start_data, end_data, start_brk members to /proc/$pid/stat v4

In reply to Cyrill Gorcunov

The mm->start_code/end_code, mm->start_data/end_data, mm->start_brk are involved into calculation of program text/data segment sizes (which might be seen in /proc/<pid>/statm) and into brk() call final address.

In a sake of restore we need to know all these values. While mm->start_code/end_code already present in /proc/$pid/stat, the rest members are not, so this patch brings them in.

The restore procedure of these members is addressed in another patch and uses prctl facility.

v2:
- Kees and Alexey pointed out that "1" hack is unnecessary, so make it to have plain (mm && permitted) ? mm->member : 0 form.

v3:
- Andrew pointed that documentation update is missed. Chagelog is updated as well.

v4:
- Update proc.txt to address concerns from Kees and Kame.

---
 Documentation/filesystems/proc.txt |    3 +++
 fs/proc/array.c                    |    7 +++++--
 2 files changed, 8 insertions(+), 2 deletions(-)
Documentation/filesystems/proc.txt +3 -0
Index: linux-2.6.git/Documentation/filesystems/proc.txt
===================================================================
@@ -305,6 +305,9 @@ Table 1-4: Contents of the stat files (a
   blkio_ticks   time spent waiting for block IO
   gtime         guest time of the task in jiffies
   cgtime        guest time of the task children in jiffies
+  start_data    address above which program data+bss is placed
+  end_data      address below which program data+bss is placed
+  start_brk     address above which program heap can be expaned with brk() call
 ..............................................................................
 
 The /proc/PID/maps file containing the currently mapped memory regions and
fs/proc/array.c +5 -2
Index: linux-2.6.git/fs/proc/array.c
===================================================================
@@ -464,7 +464,7 @@ static int do_task_stat(struct seq_file
 
 	seq_printf(m, "%d (%s) %c %d %d %d %d %d %u %lu \
 %lu %lu %lu %lu %lu %ld %ld %ld %ld %d 0 %llu %lu %ld %lu %lu %lu %lu %lu \
-%lu %lu %lu %lu %lu %lu %lu %lu %d %d %u %u %llu %lu %ld\n",
+%lu %lu %lu %lu %lu %lu %lu %lu %d %d %u %u %llu %lu %ld %lu %lu %lu\n",
 		pid_nr_ns(pid, ns),
 		tcomm,
 		state,
@@ -511,7 +511,10 @@ static int do_task_stat(struct seq_file
 		task->policy,
 		(unsigned long long)delayacct_blkio_ticks(task),
 		cputime_to_clock_t(gtime),
-		cputime_to_clock_t(cgtime));
+		cputime_to_clock_t(cgtime),
+		(mm && permitted) ? mm->start_data : 0,
+		(mm && permitted) ? mm->end_data : 0,
+		(mm && permitted) ? mm->start_brk : 0);
 	if (mm)
 		mmput(mm);
 	return 0;
#4 lore raw

PATCH 3/3 prctl: Add PR_SET_MM codes to set up mm_struct entires v3

In reply to Cyrill Gorcunov

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.

While most of them have a statistical nature (their values are involved into calculation of /proc/<pid>/statm output) the start_brk and brk values are used to compute an allowed size of program data segment expansion. Which means an arbitrary changes of this values might be dangerous operation. So to restrict access the following requirements applied to prctl calls:

- The process has to have CAP_SYS_ADMIN capability granted.
- For all opcodes except start_brk/brk members an appropriate VMA area must exist and should fit certain VMA flags, such as:
- code segment must be executable but not writable;
- data segment must not be executable.

start_brk/brk values must not intersect with data segment and must not exceed RLIMIT_DATA resource limit.

Still the main guard is CAP_SYS_ADMIN capability check.

Note the kernel should be compiled with CONFIG_CHECKPOINT_RESTORE support otherwise these prctl calls will return -EINVAL.

v2:
- Add a check for vma start address, testing for vma ending address is not enough. From Kees Cook.
- Add some sanity tests for assigned addresses.
v3:
- Make code CONFIG_CHECKPOINT_RESTORE dependant.
- Drop get_task_mm call since "current" is known to be running and we have control over it (from Andrew Morton).

Cc Andrew Morton <[email protected]>
Cc Kees Cook <[email protected]>
Cc Tejun Heo <[email protected]>
Cc Andrew Vagin <[email protected]>
Cc Serge Hallyn <[email protected]>
Cc Pavel Emelyanov <[email protected]>
Cc Vasiliy Kulikov <[email protected]>
Cc KAMEZAWA Hiroyuki <[email protected]>
Cc Michael Kerrisk <[email protected]>
---
 include/linux/prctl.h |   12 +++++
 kernel/sys.c          |  120 ++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 132 insertions(+)
include/linux/prctl.h +12 -0
Index: linux-2.6.git/include/linux/prctl.h
===================================================================
@@ -102,4 +102,16 @@
 
 #define PR_MCE_KILL_GET 34
 
+/*
+ * Tune up process memory map specifics.
+ */
+#define PR_SET_MM		35
+# define PR_SET_MM_START_CODE		1
+# define PR_SET_MM_END_CODE		2
+# define PR_SET_MM_START_DATA		3
+# define PR_SET_MM_END_DATA		4
+# define PR_SET_MM_START_STACK		5
+# define PR_SET_MM_START_BRK		6
+# define PR_SET_MM_BRK			7
+
 #endif /* _LINUX_PRCTL_H */
kernel/sys.c +120 -0

132 lines not on this page · raw message

Reviewed-by on v1 3/3
#5 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

Quoted 5 lines · Cyrill Gorcunov

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.
...

Looks good; I like having this wrapped in CONFIG_CHECKPOINT_RESTORE.

kernel/sys.c SYSCALL_DEFINE1(umask, int, mask)

Quoted 6 lines
kernel/sys.c +5 -0
+#ifdef CONFIG_CHECKPOINT_RESTORE
+static int prctl_set_mm(int opt, unsigned long addr,
+                       unsigned long arg4, unsigned long arg5)
...
+#else /* CONFIG_CHECKPOINT_RESTORE */
+static int prctl_set_mm(int opt, unsigned long addr)

These need to have matching argument lists.

-Kees

Signature · 2 lines
Kees Cook
ChromeOS Security
Reviewed-by Kees Cook
#6 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

Quoted 18 lines · Kees Cook
Quoted 5 lines · Cyrill Gorcunov

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.
...

Looks good; I like having this wrapped in CONFIG_CHECKPOINT_RESTORE.

Review

Quoted 6 lines
(quoted diff) +5 -0
+#ifdef CONFIG_CHECKPOINT_RESTORE
+static int prctl_set_mm(int opt, unsigned long addr,
+                       unsigned long arg4, unsigned long arg5)
...
+#else /* CONFIG_CHECKPOINT_RESTORE */
+static int prctl_set_mm(int opt, unsigned long addr)

These need to have matching argument lists.

Quoted 2 lines

Ups! Thanks Kees! Will update in a minute.

Cyrill
---
From: Cyrill Gorcunov <[email protected]>
Subject: [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v4

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.

While most of them have a statistical nature (their values are involved into calculation of /proc/<pid>/statm output) the start_brk and brk values are used to compute an allowed size of program data segment expansion. Which means an arbitrary changes of this values might be dangerous operation. So to restrict access the following requirements applied to prctl calls:

- The process has to have CAP_SYS_ADMIN capability granted.
- For all opcodes except start_brk/brk members an appropriate VMA area must exist and should fit certain VMA flags, such as:
- code segment must be executable but not writable;
- data segment must not be executable.

start_brk/brk values must not intersect with data segment and must not exceed RLIMIT_DATA resource limit.

Still the main guard is CAP_SYS_ADMIN capability check.

Note the kernel should be compiled with CONFIG_CHECKPOINT_RESTORE support otherwise these prctl calls will return -EINVAL.

v2:
- Add a check for vma start address, testing for vma ending address is not enough. From Kees Cook.
- Add some sanity tests for assigned addresses.
v3:
- Make code CONFIG_CHECKPOINT_RESTORE dependant.
- Drop get_task_mm call since "current" is known to be running and we have control over it (from Andrew Morton).

v4:
- Arguments matching for function declaration
from Kees Cook.

Cc Andrew Morton <[email protected]>
Cc Tejun Heo <[email protected]>
Cc Andrew Vagin <[email protected]>
Cc Serge Hallyn <[email protected]>
Cc Pavel Emelyanov <[email protected]>
Cc Vasiliy Kulikov <[email protected]>
Cc KAMEZAWA Hiroyuki <[email protected]>
Cc Michael Kerrisk <[email protected]>
---
 include/linux/prctl.h |   12 +++++
 kernel/sys.c          |  120 ++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 132 insertions(+)
include/linux/prctl.h +12 -0
Index: linux-2.6.git/include/linux/prctl.h
===================================================================
@@ -102,4 +102,16 @@
 
 #define PR_MCE_KILL_GET 34
 
+/*
+ * Tune up process memory map specifics.
+ */
+#define PR_SET_MM		35
+# define PR_SET_MM_START_CODE		1
+# define PR_SET_MM_END_CODE		2
+# define PR_SET_MM_START_DATA		3
+# define PR_SET_MM_END_DATA		4
+# define PR_SET_MM_START_STACK		5
+# define PR_SET_MM_START_BRK		6
+# define PR_SET_MM_BRK			7
+
 #endif /* _LINUX_PRCTL_H */
kernel/sys.c +120 -0

132 lines not on this page · raw message

#7 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

Quoted 39 lines · On Tue, 13 Dec 2011 00:51:14 +0400 Cyrill Gorcunov <[email protected]> wrote:

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.

While most of them have a statistical nature (their values are involved into calculation of /proc/<pid>/statm output) the start_brk and brk values are used to compute an allowed size of program data segment expansion. Which means an arbitrary changes of this values might be dangerous operation. So to restrict access the following requirements applied to prctl calls:

- The process has to have CAP_SYS_ADMIN capability granted.
- For all opcodes except start_brk/brk members an appropriate VMA area must exist and should fit certain VMA flags, such as:
- code segment must be executable but not writable;
- data segment must not be executable.

start_brk/brk values must not intersect with data segment and must not exceed RLIMIT_DATA resource limit.

Still the main guard is CAP_SYS_ADMIN capability check.

Note the kernel should be compiled with CONFIG_CHECKPOINT_RESTORE support otherwise these prctl calls will return -EINVAL.

...

(quoted diff) +10 -0
+#ifdef CONFIG_CHECKPOINT_RESTORE
+static int prctl_set_mm(int opt, unsigned long addr,
+			unsigned long arg4, unsigned long arg5)
+{
+	unsigned long rlim = rlimit(RLIMIT_DATA);
+	unsigned long vm_req_flags;
+	unsigned long vm_bad_flags;
+	struct vm_area_struct *vma;
+	int error = 0;
+

Review

Quoted 10 lines
(quoted diff) +10 -0
+	if (arg4 | arg5)
+		return -EINVAL;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (addr >= TASK_SIZE)
+		return -EINVAL;
+
+	down_read(&current->mm->mmap_sem);

This may not be true of all compiler versions, but when I cache current->mm in a local, the code size is reduced rather a lot:

akpm:/usr/src/25> size kernel/sys.o
text data bss dec hex filename

  22685   14376    7616   44677    ae85 kernel/sys.o
  22489   14376    7616   44481    adc1 kernel/sys.o
kernel/sys.c renamed from kernel/sys.c~c-r-prctl-add-pr_set_mm-codes-to-set-up-mm_struct-entries-fix +17 -16
diff -puN kernel/sys.c~c-r-prctl-add-pr_set_mm-codes-to-set-up-mm_struct-entries-fix kernel/sys.c
@@ -1701,6 +1701,7 @@ static int prctl_set_mm(int opt, unsigne
 	unsigned long vm_bad_flags;
 	struct vm_area_struct *vma;
 	int error = 0;
+	struct mm_struct *mm = current->mm;
 
 	if (arg4 | arg5)
 		return -EINVAL;
@@ -1711,8 +1712,8 @@ static int prctl_set_mm(int opt, unsigne
 	if (addr >= TASK_SIZE)
 		return -EINVAL;
 
-	down_read(&current->mm->mmap_sem);
-	vma = find_vma(current->mm, addr);
+	down_read(&mm->mmap_sem);
+	vma = find_vma(mm, addr);
 
 	if (opt != PR_SET_MM_START_BRK && opt != PR_SET_MM_BRK) {
 		/* It must be existing VMA */
@@ -1732,9 +1733,9 @@ static int prctl_set_mm(int opt, unsigne
 			goto out;
 
 		if (opt == PR_SET_MM_START_CODE)
-			current->mm->start_code = addr;
+			mm->start_code = addr;
 		else
-			current->mm->end_code = addr;
+			mm->end_code = addr;
 		break;
 
 	case PR_SET_MM_START_DATA:
@@ -1747,9 +1748,9 @@ static int prctl_set_mm(int opt, unsigne
 			goto out;
 
 		if (opt == PR_SET_MM_START_DATA)
-			current->mm->start_data = addr;
+			mm->start_data = addr;
 		else
-			current->mm->end_data = addr;
+			mm->end_data = addr;
 		break;
 
 	case PR_SET_MM_START_STACK:
@@ -1762,31 +1763,31 @@ static int prctl_set_mm(int opt, unsigne
 		if ((vma->vm_flags & vm_req_flags) != vm_req_flags)
 			goto out;
 
-		current->mm->start_stack = addr;
+		mm->start_stack = addr;
 		break;
 
 	case PR_SET_MM_START_BRK:
-		if (addr <= current->mm->end_data)
+		if (addr <= mm->end_data)
 			goto out;
 
 		if (rlim < RLIM_INFINITY &&
-		    (current->mm->brk - addr) +
-		    (current->mm->end_data - current->mm->start_data) > rlim)
+		    (mm->brk - addr) +
+		    (mm->end_data - mm->start_data) > rlim)
 			goto out;
 
-		current->mm->start_brk = addr;
+		mm->start_brk = addr;
 		break;
 
 	case PR_SET_MM_BRK:
-		if (addr <= current->mm->end_data)
+		if (addr <= mm->end_data)
 			goto out;
 
 		if (rlim < RLIM_INFINITY &&
-		    (addr - current->mm->start_brk) +
-		    (current->mm->end_data - current->mm->start_data) > rlim)
+		    (addr - mm->start_brk) +
+		    (mm->end_data - mm->start_data) > rlim)
 			goto out;
 
-		current->mm->brk = addr;
+		mm->brk = addr;
 		break;
 
 	default:
@@ -1797,7 +1798,7 @@ static int prctl_set_mm(int opt, unsigne
 	error = 0;
 
 out:
-	up_read(&current->mm->mmap_sem);
+	up_read(&mm->mmap_sem);
 
 	return error;
 }

_

#8 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

On Mon, Dec 12, 2011 at 01:53:23PM -0800, Andrew Morton wrote:
...

Quoted 25 lines
Quoted 11 lines
(quoted diff) +10 -0
+#ifdef CONFIG_CHECKPOINT_RESTORE
+static int prctl_set_mm(int opt, unsigned long addr,
+			unsigned long arg4, unsigned long arg5)
+{
+	unsigned long rlim = rlimit(RLIMIT_DATA);
+	unsigned long vm_req_flags;
+	unsigned long vm_bad_flags;
+	struct vm_area_struct *vma;
+	int error = 0;
+

Review

Quoted 10 lines
(quoted diff) +10 -0
+	if (arg4 | arg5)
+		return -EINVAL;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (addr >= TASK_SIZE)
+		return -EINVAL;
+
+	down_read(&current->mm->mmap_sem);

This may not be true of all compiler versions, but when I cache current->mm in a local, the code size is reduced rather a lot:

Quoted 5 lines

akpm:/usr/src/25> size kernel/sys.o
text data bss dec hex filename

  22685   14376    7616   44677    ae85 kernel/sys.o
  22489   14376    7616   44481    adc1 kernel/sys.o

Hmm, this is great and weird. Letme try...

Cyrill

#9 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

On Tue, Dec 13, 2011 at 02:01:32AM +0400, Cyrill Gorcunov wrote:
...

Quoted 4 lines
Quoted 4 lines

This may not be true of all compiler versions, but when I cache current->mm in a local, the code size is reduced rather a lot:

Quoted 8 lines
Quoted 5 lines

akpm:/usr/src/25> size kernel/sys.o
text data bss dec hex filename

  22685   14376    7616   44677    ae85 kernel/sys.o
  22489   14376    7616   44481    adc1 kernel/sys.o

Hmm, this is great and weird. Letme try...

Same here (gcc version 4.6.2 20111027)

text data bss dec hex filename

  20829	  14376	   5736	  40941	   9fed	kernel/sys.o
  20682	  14376	   5736	  40794	   9f5a	kernel/sys.o

thanks Andrew!

Cyrill

#10 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

In reply to Cyrill Gorcunov

Hi

Quoted 4 lines

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.

Quoted 8 lines

While most of them have a statistical nature (their values are involved into calculation of /proc/<pid>/statm output) the start_brk and brk values are used to compute an allowed size of program data segment expansion. Which means an arbitrary changes of this values might be dangerous operation. So to restrict access the following requirements applied to prctl calls:

- The process has to have CAP_SYS_ADMIN capability granted.

This is very dangerous feature and useless from regular admins. Moreover, CAP_SYS_ADMIN has a pretty overweight meanings and we can't disable it on practical. So, I have a question. Why don't you make new capability for checkpoint?

#11 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

Quoted 13 lines · KOSAKI Motohiro

Hi

Quoted 11 lines

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.

While most of them have a statistical nature (their values are involved into calculation of /proc/<pid>/statm output) the start_brk and brk values are used to compute an allowed size of program data segment expansion. Which means an arbitrary changes of this values might be dangerous operation. So to restrict access the following requirements applied to prctl calls:

Quoted 3 lines
Quoted 1 line

- The process has to have CAP_SYS_ADMIN capability granted.

This is very dangerous feature and useless from regular admins.

Except brk() call I don't see where it might be extremelly dangerous at moment but indeed it might become very dangerous once code grows. Still if evil minded person got CAP_SYS_ADMIN these prctls are least thing one should carry about.

Quoted 4 lines

Moreover, CAP_SYS_ADMIN has a pretty overweight meanings and we can't disable it on practical. So, I have a question. Why don't you make new capability for checkpoint?

It's not a problem to introduce CAP_CHECKPOINT_RESTORE, but would it be accepted? I mean, are we fine with new capability introduction? If yes -- I'll add new one and rebase the patch.

Cyrill

#12 lore raw

Re: [patch 3/3] [PATCH] prctl: Add PR_SET_MM codes to set up mm_struct entires v3

Quoted 14 lines · Cyrill Gorcunov
Quoted 13 lines · KOSAKI Motohiro

Hi

Quoted 11 lines

When we restore a task we need to set up text, data and data heap sizes from userspace to the values a task had at checkpoint time. This patch adds auxilary prctl codes for that.

While most of them have a statistical nature (their values are involved into calculation of /proc/<pid>/statm output) the start_brk and brk values are used to compute an allowed size of program data segment expansion. Which means an arbitrary changes of this values might be dangerous operation. So to restrict access the following requirements applied to prctl calls:

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

- The process has to have CAP_SYS_ADMIN capability granted.

This is very dangerous feature and useless from regular admins.

Except brk() call I don't see where it might be extremelly dangerous at moment but indeed it might become very dangerous once code grows. Still if evil minded person got CAP_SYS_ADMIN these prctls are least thing one should carry about.

I'm sorry, I misunderstood your code. Your code only allow to change their own process attribute. So, it's enough harmless. Please ignore my last mail.

Quoted 8 lines
Quoted 4 lines

Moreover, CAP_SYS_ADMIN has a pretty overweight meanings and we can't disable it on practical. So, I have a question. Why don't you make new capability for checkpoint?

It's not a problem to introduce CAP_CHECKPOINT_RESTORE, but would it be accepted? I mean, are we fine with new capability introduction? If yes -- I'll add new one and rebase the patch.