sys/prctl: expose TASK_SIZE value to userspace
v3 cover sys/prctl: expose TASK_SIZE value to userspace
In the mainline kernel, there is no quick mechanism to get the virtual memory size of the current process from userspace.
Despite the current state of affairs, this information is available to the user through several means, one being a linear search of the entire address space. This is an inefficient use of cpu cycles.
A component of the libhugetlb kernel test does exactly this, and as systems' address spaces increase beyond 32-bits, this method becomes exceedingly tedious.
For example, on a ppc64le system with a 47-bit address space, the linear search causes the test to hang for some unknown amount of time. I couldn't give you an exact number because I just ran it for about 10-20 minutes and went to go do something else, probably to get coffee or something, and when I came back, I just killed the test and patched it to use this new mechanism. I re-ran my new version of the test using a kernel with this patch, and of course it passed through the previously bottlenecking codepath nearly instantaneously.
As such, I propose that the prctl syscall be extended to include the option to retrieve TASK_SIZE from the kernel.
This patch will allow us to upgrade an O(n) codepath to O(1) in an architecture-independent manner, and provide a mechanism for future generations to do the same.
Changes from v2:
We now account for the case of 32-bit compat userspace on a 64-bit kernel More detail about the nature of TASK_SIZE in documentation
Joel Savitz(2):
sys/prctl: add PR_GET_TASK_SIZE option to prctl(2) prctl.2: Document the new PR_GET_TASK_SIZE option
include/uapi/linux/prctl.h | 3 +++ kernel/sys.c | 23 +++++++++++++++++++++++ 2 files changed, 26 insertions(+)
man2/prctl.2 | 10 ++++++++++ 1 file changed, 10 insertions(+)
--
2.18.1
PATCH v3 1/2 kernel/sys: add PR_GET_TASK_SIZE option to prctl(2)
When PR_GET_TASK_SIZE is passed to prctl, the kernel will attempt to copy the value of TASK_SIZE to the userspace address in arg2.
It is important that we account for the case of the userspace task running in 32-bit compat mode on a 64-bit kernel. As such, we must be careful to copy the correct number of bytes to userspace to avoid stack corruption.
--- include/uapi/linux/prctl.h | 3 +++ kernel/sys.c | 23 +++++++++++++++++++++++ 2 files changed, 26 insertions(+)
include/uapi/linux/prctl.h +3 -0
@@ -229,4 +229,7 @@ struct prctl_mm_map {# define PR_PAC_APDBKEY (1UL << 3)# define PR_PAC_APGAKEY (1UL << 4) +/* Get the process virtual memory size (i.e. the highest usable VM address) */ +#define PR_GET_TASK_SIZE 55 +#endif /* _LINUX_PRCTL_H */
kernel/sys.c +22 -0
@@ -2252,6 +2252,26 @@ static int propagate_has_child_subreaper(struct task_struct *p, void *data)return 1 ; } +static int prctl_get_tasksize (void __user *uaddr) +{ +unsigned long current_task_size, current_word_size; + + current_task_size =TASK_SIZE ; + current_word_size =sizeof (unsigned long ); + +#ifdef CONFIG_64BIT +/* On 64-bit architecture, we must check whether the current thread +* is running in 32-bit compat mode. If it is, we can simply cut +* the size in half. This avoids corruption of the userspace stack. +*/ +if (test_thread_flag (TIF_ADDR32 )) + current_word_size >>=1 ; +#endif + +return copy_to_user (uaddr, ¤t_task_size, current_word_size) ? -EFAULT :0 ; +} +int __weak arch_prctl_spec_ctrl_get (struct task_struct *t,unsigned long which) {return -EINVAL ;
@@ -2486,6 +2506,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,return -EINVAL ; error =PAC_RESET_KEYS (me, arg2);break ; +case PR_GET_TASK_SIZE : + error =prctl_get_tasksize ((void *)arg2); +break ;default : error = -EINVAL ;break ;
--
2.18.1
Yury Norov
NAKRe: [PATCH v3 1/2] kernel/sys: add PR_GET_TASK_SIZE option to prctl(2)
Quoted 9 lines · Joel Savitz
When
PR_GET_TASK_SIZEis passed to prctl, the kernel will attempt to copy the value ofTASK_SIZEto the userspace address in arg2.It is important that we account for the case of the userspace task running in 32-bit compat mode on a 64-bit kernel. As such, we must be careful to copy the correct number of bytes to userspace to avoid stack corruption.
Suggested-by Yuri Norov <[email protected]>
I actually didn't suggest that. If you _really_ need TASK_SIZE to be exposed, I would suggest to expose it in kernel headers. TASK_SIZE is a compile-time information, and it may available for userspace at compile time as well.
Quoted 30 lines
Suggested-by Alexey Dobriyan <[email protected]>Signed-off-by Joel Savitz <[email protected]>--- include/uapi/linux/prctl.h | 3 +++ kernel/sys.c | 23 +++++++++++++++++++++++ 2 files changed, 26 insertions(+)include/uapi/linux/prctl.h +3
-0@@ -229,4 +229,7 @@ struct prctl_mm_map {# define PR_PAC_APDBKEY (1UL << 3)# define PR_PAC_APGAKEY (1UL << 4) +/* Get the process virtual memory size (i.e. the highest usable VM address) */ +#define PR_GET_TASK_SIZE 55 +#endif /* _LINUX_PRCTL_H */ kernel/sys.c +3
-0@@ -2252,6 +2252,26 @@ static int propagate_has_child_subreaper(struct task_struct *p, void *data)return 1 ; } +static int prctl_get_tasksize (void __user *uaddr) +{ +unsigned long current_task_size, current_word_size;
kernel/sys.c static int propagate_has_child_subreaper(struct task_struct *p, void *data) · line 2267
Quoted 10 lines
kernel/sys.c +10
-0+ + current_task_size =TASK_SIZE ; + current_word_size =sizeof (unsigned long ); + +#ifdef CONFIG_64BIT +/* On 64-bit architecture, we must check whether the current thread +* is running in 32-bit compat mode. If it is, we can simply cut +* the size in half. This avoids corruption of the userspace stack. +*/ +if (test_thread_flag (TIF_ADDR32 ))
It breaks build for all architectures except x86 since TIF_ADDR32 is defined for x86 only.
In comment to v2 I suggested you to stick to fixed-size data type to avoid exactly this problem.
NACK
Yury
Quoted 21 lines
kernel/sys.c +9
-1+ current_word_size >>=1 ; +#endif + +return copy_to_user (uaddr, ¤t_task_size, current_word_size) ? -EFAULT :0 ; +} +int __weak arch_prctl_spec_ctrl_get (struct task_struct *t,unsigned long which) {return -EINVAL ;@@ -2486,6 +2506,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,return -EINVAL ; error =PAC_RESET_KEYS (me, arg2);break ; +case PR_GET_TASK_SIZE : + error =prctl_get_tasksize ((void *)arg2); +break ;default : error = -EINVAL ;break ; --2.18.1
Re: [PATCH v3 1/2] kernel/sys: add PR_GET_TASK_SIZE option to prctl(2)
Quoted 9 lines · Yury Norov
Quoted 8 lines · Joel Savitz
When
PR_GET_TASK_SIZEis passed to prctl, the kernel will attempt to copy the value ofTASK_SIZEto the userspace address in arg2.It is important that we account for the case of the userspace task running in 32-bit compat mode on a 64-bit kernel. As such, we must be careful to copy the correct number of bytes to userspace to avoid stack corruption.
Quoted 7 lines
Quoted 1 line
Suggested-by Yuri Norov <[email protected]>I actually didn't suggest that. If you _really_ need
TASK_SIZEto be exposed, I would suggest to expose it in kernel headers.TASK_SIZEis a compile-time information, and it may available for userspace at compile time as well.
TASK_SIZE is a runtime resolved macro, dependent on the thread currently running on the CPU. It's not a compile time constant.
Anyways, it's proven that going prctl(2), although interesting, as suggested by Alexey, wasn't worth the hassle as it poses more issues than it can possibly solve.
A better way to get this value exposed to userspace is really through /proc/<pid>/status, where one can utilize TASK_SIZE_OF(mm->owner), or simply mm->task_size, which seems to be properly assigned for each arch
Quoted 72 lines
Quoted 30 lines
Suggested-by Alexey Dobriyan <[email protected]>Signed-off-by Joel Savitz <[email protected]>--- include/uapi/linux/prctl.h | 3 +++ kernel/sys.c | 23 +++++++++++++++++++++++ 2 files changed, 26 insertions(+)include/uapi/linux/prctl.h +3
-0@@ -229,4 +229,7 @@ struct prctl_mm_map {# define PR_PAC_APDBKEY (1UL << 3)# define PR_PAC_APGAKEY (1UL << 4) +/* Get the process virtual memory size (i.e. the highest usable VM address) */ +#define PR_GET_TASK_SIZE 55 +#endif /* _LINUX_PRCTL_H */ kernel/sys.c +3
-0@@ -2252,6 +2252,26 @@ static int propagate_has_child_subreaper(struct task_struct *p, void *data)return 1 ; } +static int prctl_get_tasksize (void __user *uaddr) +{ +unsigned long current_task_size, current_word_size;kernel/sys.c static int propagate_has_child_subreaper(struct task_struct *p, void *data) · line 2267
Quoted 10 lines
kernel/sys.c +10
-0+ + current_task_size =TASK_SIZE ; + current_word_size =sizeof (unsigned long ); + +#ifdef CONFIG_64BIT +/* On 64-bit architecture, we must check whether the current thread +* is running in 32-bit compat mode. If it is, we can simply cut +* the size in half. This avoids corruption of the userspace stack. +*/ +if (test_thread_flag (TIF_ADDR32 ))It breaks build for all architectures except x86 since
TIF_ADDR32is defined for x86 only.In comment to v2 I suggested you to stick to fixed-size data type to avoid exactly this problem.
NACK
Yury
Quoted 21 lines
kernel/sys.c +9
-1+ current_word_size >>=1 ; +#endif + +return copy_to_user (uaddr, ¤t_task_size, current_word_size) ? -EFAULT :0 ; +} +int __weak arch_prctl_spec_ctrl_get (struct task_struct *t,unsigned long which) {return -EINVAL ;@@ -2486,6 +2506,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,return -EINVAL ; error =PAC_RESET_KEYS (me, arg2);break ; +case PR_GET_TASK_SIZE : + error =prctl_get_tasksize ((void *)arg2); +break ;default : error = -EINVAL ;break ; --2.18.1
Re: [PATCH v3 1/2] kernel/sys: add PR_GET_TASK_SIZE option to prctl(2)
In reply to Yury NorovQuoted 43 lines · Yury Norov
Quoted 9 lines · Joel Savitz
When
PR_GET_TASK_SIZEis passed to prctl, the kernel will attempt to copy the value ofTASK_SIZEto the userspace address in arg2.It is important that we account for the case of the userspace task running in 32-bit compat mode on a 64-bit kernel. As such, we must be careful to copy the correct number of bytes to userspace to avoid stack corruption.
Suggested-by Yuri Norov <[email protected]>I actually didn't suggest that. If you _really_ need
TASK_SIZEto be exposed, I would suggest to expose it in kernel headers.TASK_SIZEis a compile-time information, and it may available for userspace at compile time as well.Quoted 27 lines
Suggested-by Alexey Dobriyan <[email protected]>Signed-off-by Joel Savitz <[email protected]>--- include/uapi/linux/prctl.h | 3 +++ kernel/sys.c | 23 +++++++++++++++++++++++ 2 files changed, 26 insertions(+)include/uapi/linux/prctl.h +3
-0@@ -229,4 +229,7 @@ struct prctl_mm_map {# define PR_PAC_APDBKEY (1UL << 3)# define PR_PAC_APGAKEY (1UL << 4) +/* Get the process virtual memory size (i.e. the highest usable VM address) */ +#define PR_GET_TASK_SIZE 55 +#endif /* _LINUX_PRCTL_H */ kernel/sys.c +0
-0@@ -2252,6 +2252,26 @@ static int propagate_has_child_subreaper(struct task_struct *p, void *data)return 1 ; }
Quoted 16 lines
kernel/sys.c static int propagate_has_child_subreaper(struct task_struct *p, void *data) · line 2267
Quoted 13 lines
kernel/sys.c +13
-0+static int prctl_get_tasksize (void __user *uaddr) +{ +unsigned long current_task_size, current_word_size; + + current_task_size =TASK_SIZE ; + current_word_size =sizeof (unsigned long ); + +#ifdef CONFIG_64BIT +/* On 64-bit architecture, we must check whether the current thread +* is running in 32-bit compat mode. If it is, we can simply cut +* the size in half. This avoids corruption of the userspace stack. +*/ +if (test_thread_flag (TIF_ADDR32 ))It breaks build for all architectures except x86 since
TIF_ADDR32is defined for x86 only.
Or we could get TIF_32BIT also defined for x86 (same value of
TIF_ADDR32) and check for it instead. i.e.
...
#if defined(CONFIG_64BIT) && defined(TIF_32BIT) if (test_thread_flag(TIF_32BIT))
...
which is also uglier and keeps adding unecessary complexity to a very simple task. At this point, I think we just should give up on trying this via prctl(2) and do it via /proc/<pid>/status instead.
Quoted 29 lines
In comment to v2 I suggested you to stick to fixed-size data type to avoid exactly this problem.
NACK
Yury
Quoted 21 lines
kernel/sys.c +9
-1+ current_word_size >>=1 ; +#endif + +return copy_to_user (uaddr, ¤t_task_size, current_word_size) ? -EFAULT :0 ; +} +int __weak arch_prctl_spec_ctrl_get (struct task_struct *t,unsigned long which) {return -EINVAL ;@@ -2486,6 +2506,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,return -EINVAL ; error =PAC_RESET_KEYS (me, arg2);break ; +case PR_GET_TASK_SIZE : + error =prctl_get_tasksize ((void *)arg2); +break ;default : error = -EINVAL ;break ; --2.18.1
Re: [PATCH v3 1/2] kernel/sys: add PR_GET_TASK_SIZE option to prctl(2)
In reply to Joel SavitzQuoted 2 lines · Joel Savitz
When
PR_GET_TASK_SIZEis passed to prctl, the kernel will attempt to copy the value ofTASK_SIZEto the userspace address in arg2.
A commit message shouldn't just describe what you're doing, but also why you're doing it. Is this intended for processes that are running on X86-64 and want to determine whether the system supports 5-level paging, or something like that?
Quoted 8 lines
kernel/sys.c +8
-0+static int prctl_get_tasksize (void __user *uaddr) +{ +unsigned long current_task_size, current_word_size; + + current_task_size =TASK_SIZE ; + current_word_size =sizeof (unsigned long ); + +#ifdef CONFIG_64BIT
kernel/sys.c static int propagate_has_child_subreaper(struct task_struct *p, void *data) · line 2272
Quoted 10 lines
kernel/sys.c +10
-0+/* On 64-bit architecture, we must check whether the current thread +* is running in 32-bit compat mode. If it is, we can simply cut +* the size in half. This avoids corruption of the userspace stack. +*/ +if (test_thread_flag (TIF_ADDR32 )) + current_word_size >>=1 ; +#endif + +return copy_to_user (uaddr, ¤t_task_size, current_word_size) ? -EFAULT :0 ; +}
This function looks completely wrong; in particular, you're assuming that the architecture is little-endian.
Make the value a u64, and you won't have these problems:
static int prctl_get_tasksize(u64 __user *uaddr)
{
return put_user(TASK_SIZE, uaddr) ? -EFAULT : 0;
}A bunch of other new pieces of userspace API already use "u64" to store userspace pointers and lengths to avoid compat issues.
kernel/sys.c SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, · line 2510
Quoted 6 lines
kernel/sys.c +2
-0@@ -2486,6 +2506,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,return -EINVAL ; error =PAC_RESET_KEYS (me, arg2);break ; +case PR_GET_TASK_SIZE : + error =prctl_get_tasksize ((void *)arg2);
s/void */void __user */
Re: [PATCH v3 1/2] kernel/sys: add PR_GET_TASK_SIZE option to prctl(2)
In reply to Joel Savitzinclude/uapi/linux/prctl.h struct prctl_mm_map { · line 233
Quoted 2 lines · Joel Savitz
include/uapi/linux/prctl.h +2
-0+/* Get the process virtual memory size (i.e. the highest usable VM address) */ +#define PR_GET_TASK_SIZE 55
TASK_SIZE is in fact the lowest _un_usable address. :^)
PATCH v3 2/2 prctl.2: Document the new PR_GET_TASK_SIZE option
In reply to Joel SavitzAdd a short explanation of the new PR_GET_TASK_SIZE option for the benefit of future generations.
--- man2/prctl.2 | 10 ++++++++++ 1 file changed, 10 insertions(+)
man2/prctl.2 +10 -0
@@ -49,6 +49,7 @@ .\" 2013-01-10 Kees Cook, document PR_SET_PTRACER .\" 2012-02-04 Michael Kerrisk, document PR_{SET,GET}_CHILD_SUBREAPER .\" 2014-11-10 Dave Hansen, document PR_MPX_{EN,DIS}ABLE_MANAGEMENT +.\" 2019-05-03 Joel Savitz, document PR_GET_TASK_SIZE .\" .\" .TH PRCTL 2 2019-03-06 "Linux" "Linux Programmer's Manual"
@@ -1375,6 +1376,15 @@ system call on Tru64). for information on versions and architectures) Return unaligned access control bits, in the location pointed to by .IR "(unsigned int\ *) arg2" . +.TP +.B PR_GET_TASK_SIZE +Copy the value of TASK_SIZE to the userspace address in +.IR "(unsigned long\ *) arg2" . +This value represents the highest virtual address available +to the current process. Return +.B EFAULT +if this operation fails. + .SH RETURN VALUE On success, .BR PR_GET_DUMPABLE ,
--
2.18.1
Quoted 20 lines · Joel Savitz
In the mainline kernel, there is no quick mechanism to get the virtual memory size of the current process from userspace.
Despite the current state of affairs, this information is available to the user through several means, one being a linear search of the entire address space. This is an inefficient use of cpu cycles.
A component of the libhugetlb kernel test does exactly this, and as systems' address spaces increase beyond 32-bits, this method becomes exceedingly tedious.
For example, on a ppc64le system with a 47-bit address space, the linear search causes the test to hang for some unknown amount of time. I couldn't give you an exact number because I just ran it for about 10-20 minutes and went to go do something else, probably to get coffee or something, and when I came back, I just killed the test and patched it to use this new mechanism. I re-ran my new version of the test using a kernel with this patch, and of course it passed through the previously bottlenecking codepath nearly instantaneously.
Quoted 6 lines
As such, I propose that the prctl syscall be extended to include the option to retrieve
TASK_SIZEfrom the kernel.This patch will allow us to upgrade an O(n) codepath to O(1) in an architecture-independent manner, and provide a mechanism for future generations to do the same.
So the only reason for the new API is boosting some random poorly written userspace test? Why don't you introduce binary search instead?
Look at /proc/<pid>/maps. It may help to reduce the memory area to be checked.
Quoted 16 lines
Changes from v2:
We now account for the case of 32-bit compat userspace on a 64-bit kernel More detail about the nature ofTASK_SIZEin documentationJoel Savitz(2):
sys/prctl: addPR_GET_TASK_SIZEoption to prctl(2) prctl.2: Document the newPR_GET_TASK_SIZEoptioninclude/uapi/linux/prctl.h | 3 +++ kernel/sys.c | 23 +++++++++++++++++++++++ 2 files changed, 26 insertions(+)man2/prctl.2 | 10 ++++++++++ 1 file changed, 10 insertions(+)--
2.18.1
Quoted 24 lines · Yury Norov
Quoted 23 lines · Joel Savitz
In the mainline kernel, there is no quick mechanism to get the virtual memory size of the current process from userspace.
Despite the current state of affairs, this information is available to the user through several means, one being a linear search of the entire address space. This is an inefficient use of cpu cycles.
A component of the libhugetlb kernel test does exactly this, and as systems' address spaces increase beyond 32-bits, this method becomes exceedingly tedious.
For example, on a ppc64le system with a 47-bit address space, the linear search causes the test to hang for some unknown amount of time. I couldn't give you an exact number because I just ran it for about 10-20 minutes and went to go do something else, probably to get coffee or something, and when I came back, I just killed the test and patched it to use this new mechanism. I re-ran my new version of the test using a kernel with this patch, and of course it passed through the previously bottlenecking codepath nearly instantaneously.
As such, I propose that the prctl syscall be extended to include the option to retrieve
TASK_SIZEfrom the kernel.
Quoted 7 lines
Quoted 3 lines
This patch will allow us to upgrade an O(n) codepath to O(1) in an architecture-independent manner, and provide a mechanism for future generations to do the same.
So the only reason for the new API is boosting some random poorly written userspace test? Why don't you introduce binary search instead?
there's no real cost in exposing the value that is known to the kernel, anyways, as long as it's not a freaking hassle (like trying to go with this prctl(2) stunt). We just need to get it properly exported alongside other task's VM-related values at /proc/<pid>/status.
Quoted 19 lines
Look at /proc/<pid>/maps. It may help to reduce the memory area to be checked.
Quoted 16 lines
Changes from v2:
We now account for the case of 32-bit compat userspace on a 64-bit kernel More detail about the nature ofTASK_SIZEin documentationJoel Savitz(2):
sys/prctl: addPR_GET_TASK_SIZEoption to prctl(2) prctl.2: Document the newPR_GET_TASK_SIZEoptioninclude/uapi/linux/prctl.h | 3 +++ kernel/sys.c | 23 +++++++++++++++++++++++ 2 files changed, 26 insertions(+)man2/prctl.2 | 10 ++++++++++ 1 file changed, 10 insertions(+)--
2.18.1
Quoted 29 lines · Rafael Aquini
Quoted 28 lines · Yury Norov
Quoted 26 lines · Joel Savitz
In the mainline kernel, there is no quick mechanism to get the virtual memory size of the current process from userspace.
Despite the current state of affairs, this information is available to the user through several means, one being a linear search of the entire address space. This is an inefficient use of cpu cycles.
A component of the libhugetlb kernel test does exactly this, and as systems' address spaces increase beyond 32-bits, this method becomes exceedingly tedious.
For example, on a ppc64le system with a 47-bit address space, the linear search causes the test to hang for some unknown amount of time. I couldn't give you an exact number because I just ran it for about 10-20 minutes and went to go do something else, probably to get coffee or something, and when I came back, I just killed the test and patched it to use this new mechanism. I re-ran my new version of the test using a kernel with this patch, and of course it passed through the previously bottlenecking codepath nearly instantaneously.
As such, I propose that the prctl syscall be extended to include the option to retrieve
TASK_SIZEfrom the kernel.This patch will allow us to upgrade an O(n) codepath to O(1) in an architecture-independent manner, and provide a mechanism for future generations to do the same.
Quoted 5 lines
Quoted 3 lines
So the only reason for the new API is boosting some random poorly written userspace test? Why don't you introduce binary search instead?
there's no real cost in exposing the value that is known to the kernel,
Really? We all here used to think that kernel programming is one of the most difficult professions in the world. There is huge cost of proper implementation of a feature, careful review, spread testing on various platforms and long-term maintenance and support.
In this specific example of exposing TASK_SIZE your team made too much things wrong to realize it, I hope.
Quoted 3 lines
anyways, as long as it's not a freaking hassle (like trying to go with this prctl(2) stunt). We just need to get it properly exported alongside other task's VM-related values at /proc/<pid>/status.
I found this thread thrilling. Please keep me in CC with your /proc/<pid>/status effort.
Yury
Quoted 2 lines
Quoted 2 lines
Look at /proc/<pid>/maps. It may help to reduce the memory area to be checked.