* [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct
@ 2018-03-24 0:36 Yang Shi
2018-03-24 4:30 ` Matthew Wilcox
0 siblings, 1 reply; 6+ messages in thread
From: Yang Shi @ 2018-03-24 0:36 UTC (permalink / raw)
To: adobriyan, mhocko, akpm; +Cc: yang.shi, linux-mm, linux-kernel
mmap_sem is on the hot path of kernel, and it very contended, but it is
abused too. It is used to protect arg_start|end and evn_start|end when
reading /proc/$PID/cmdline and /proc/$PID/environ, but it doesn't make
sense since those proc files just expect to read 4 values atomically and
not related to VM, they could be set to arbitrary values by C/R.
And, the mmap_sem contention may cause unexpected issue like below:
INFO: task ps:14018 blocked for more than 120 seconds.
Tainted: G E 4.9.79-009.ali3000.alios7.x86_64 #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this
message.
ps D 0 14018 1 0x00000004
ffff885582f84000 ffff885e8682f000 ffff880972943000 ffff885ebf499bc0
ffff8828ee120000 ffffc900349bfca8 ffffffff817154d0 0000000000000040
00ffffff812f872a ffff885ebf499bc0 024000d000948300 ffff880972943000
Call Trace:
[<ffffffff817154d0>] ? __schedule+0x250/0x730
[<ffffffff817159e6>] schedule+0x36/0x80
[<ffffffff81718560>] rwsem_down_read_failed+0xf0/0x150
[<ffffffff81390a28>] call_rwsem_down_read_failed+0x18/0x30
[<ffffffff81717db0>] down_read+0x20/0x40
[<ffffffff812b9439>] proc_pid_cmdline_read+0xd9/0x4e0
[<ffffffff81253c95>] ? do_filp_open+0xa5/0x100
[<ffffffff81241d87>] __vfs_read+0x37/0x150
[<ffffffff812f824b>] ? security_file_permission+0x9b/0xc0
[<ffffffff81242266>] vfs_read+0x96/0x130
[<ffffffff812437b5>] SyS_read+0x55/0xc0
[<ffffffff8171a6da>] entry_SYSCALL_64_fastpath+0x1a/0xc5
Both Alexey Dobriyan and Michal Hocko suggested to use dedicated lock
for them to mitigate the abuse of mmap_sem.
So, introduce a new rwlock in mm_struct to protect the concurrent access
to arg_start|end and env_start|end.
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Michal Hocko <mhocko@kernel.org>
---
fs/proc/base.c | 8 ++++----
include/linux/mm_types.h | 2 ++
kernel/fork.c | 1 +
kernel/sys.c | 6 ++++++
mm/init-mm.c | 1 +
5 files changed, 14 insertions(+), 4 deletions(-)
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 28fa852..0bc3107 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -239,12 +239,12 @@ static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
goto out_mmput;
}
- down_read(&mm->mmap_sem);
+ read_lock(&mm->arg_lock);
arg_start = mm->arg_start;
arg_end = mm->arg_end;
env_start = mm->env_start;
env_end = mm->env_end;
- up_read(&mm->mmap_sem);
+ read_unlock(&mm->arg_lock);
BUG_ON(arg_start > arg_end);
BUG_ON(env_start > env_end);
@@ -926,10 +926,10 @@ static ssize_t environ_read(struct file *file, char __user *buf,
if (!mmget_not_zero(mm))
goto free;
- down_read(&mm->mmap_sem);
+ read_lock(&mm->arg_lock);
env_start = mm->env_start;
env_end = mm->env_end;
- up_read(&mm->mmap_sem);
+ read_unlock(&mm->arg_lock);
while (count > 0) {
size_t this_len, max_len;
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index cfd0ac4..92bd9cc 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -419,6 +419,8 @@ struct mm_struct {
unsigned long def_flags;
unsigned long start_code, end_code, start_data, end_data;
unsigned long start_brk, brk, start_stack;
+
+ rwlock_t arg_lock; /* protect concurrent access to arg_* and env_* */
unsigned long arg_start, arg_end, env_start, env_end;
unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
diff --git a/kernel/fork.c b/kernel/fork.c
index 432eadf..94cdf28 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -823,6 +823,7 @@ static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
mm->pinned_vm = 0;
memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
spin_lock_init(&mm->page_table_lock);
+ rwlock_init(&mm->arg_lock);
mm_init_cpumask(mm);
mm_init_aio(mm);
mm_init_owner(mm, p);
diff --git a/kernel/sys.c b/kernel/sys.c
index 83ffd7d..de357e1 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -1980,10 +1980,13 @@ static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data
mm->start_brk = prctl_map.start_brk;
mm->brk = prctl_map.brk;
mm->start_stack = prctl_map.start_stack;
+
+ write_lock(&mm->arg_lock);
mm->arg_start = prctl_map.arg_start;
mm->arg_end = prctl_map.arg_end;
mm->env_start = prctl_map.env_start;
mm->env_end = prctl_map.env_end;
+ write_unlock(&mm->arg_lock);
/*
* Note this update of @saved_auxv is lockless thus
@@ -2149,10 +2152,13 @@ static int prctl_set_mm(int opt, unsigned long addr,
mm->start_brk = prctl_map.start_brk;
mm->brk = prctl_map.brk;
mm->start_stack = prctl_map.start_stack;
+
+ write_lock(&mm->arg_lock);
mm->arg_start = prctl_map.arg_start;
mm->arg_end = prctl_map.arg_end;
mm->env_start = prctl_map.env_start;
mm->env_end = prctl_map.env_end;
+ write_unlock(&mm->arg_lock);
error = 0;
out:
diff --git a/mm/init-mm.c b/mm/init-mm.c
index f94d5d1..5b2a044 100644
--- a/mm/init-mm.c
+++ b/mm/init-mm.c
@@ -23,6 +23,7 @@ struct mm_struct init_mm = {
.mmap_sem = __RWSEM_INITIALIZER(init_mm.mmap_sem),
.page_table_lock = __SPIN_LOCK_UNLOCKED(init_mm.page_table_lock),
.mmlist = LIST_HEAD_INIT(init_mm.mmlist),
+ .arg_lock = __RW_LOCK_UNLOCKED(init_mm.arg_lock),
.user_ns = &init_user_ns,
INIT_MM_CONTEXT(init_mm)
};
--
1.8.3.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct
2018-03-24 0:36 [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct Yang Shi
@ 2018-03-24 4:30 ` Matthew Wilcox
2018-03-25 0:32 ` Yang Shi
0 siblings, 1 reply; 6+ messages in thread
From: Matthew Wilcox @ 2018-03-24 4:30 UTC (permalink / raw)
To: Yang Shi; +Cc: adobriyan, mhocko, akpm, linux-mm, linux-kernel
> So, introduce a new rwlock in mm_struct to protect the concurrent access
> to arg_start|end and env_start|end.
I don't think an rwlock makes much sense here. There is almost no
concurrency on the read side, and an rwlock is more expensive than
a spinlock. Just use a spinlock.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct
2018-03-24 4:30 ` Matthew Wilcox
@ 2018-03-25 0:32 ` Yang Shi
2018-03-26 14:49 ` Tetsuo Handa
0 siblings, 1 reply; 6+ messages in thread
From: Yang Shi @ 2018-03-25 0:32 UTC (permalink / raw)
To: Matthew Wilcox; +Cc: adobriyan, mhocko, akpm, linux-mm, linux-kernel
On 3/23/18 9:30 PM, Matthew Wilcox wrote:
>> So, introduce a new rwlock in mm_struct to protect the concurrent access
>> to arg_start|end and env_start|end.
> I don't think an rwlock makes much sense here. There is almost no
> concurrency on the read side, and an rwlock is more expensive than
> a spinlock. Just use a spinlock.
Thanks for the comment. Yes, actually there is not concurrency on the
read side, will change to regular spin lock.
Yang
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct
2018-03-25 0:32 ` Yang Shi
@ 2018-03-26 14:49 ` Tetsuo Handa
2018-03-26 14:55 ` Michal Hocko
2018-03-26 15:31 ` Yang Shi
0 siblings, 2 replies; 6+ messages in thread
From: Tetsuo Handa @ 2018-03-26 14:49 UTC (permalink / raw)
To: Yang Shi, Matthew Wilcox; +Cc: adobriyan, mhocko, akpm, linux-mm, linux-kernel
On 2018/03/24 9:36, Yang Shi wrote:
> And, the mmap_sem contention may cause unexpected issue like below:
>
> INFO: task ps:14018 blocked for more than 120 seconds.
> Tainted: G E 4.9.79-009.ali3000.alios7.x86_64 #1
> "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this
> message.
> ps D 0 14018 1 0x00000004
> ffff885582f84000 ffff885e8682f000 ffff880972943000 ffff885ebf499bc0
> ffff8828ee120000 ffffc900349bfca8 ffffffff817154d0 0000000000000040
> 00ffffff812f872a ffff885ebf499bc0 024000d000948300 ffff880972943000
> Call Trace:
> [<ffffffff817154d0>] ? __schedule+0x250/0x730
> [<ffffffff817159e6>] schedule+0x36/0x80
> [<ffffffff81718560>] rwsem_down_read_failed+0xf0/0x150
> [<ffffffff81390a28>] call_rwsem_down_read_failed+0x18/0x30
> [<ffffffff81717db0>] down_read+0x20/0x40
> [<ffffffff812b9439>] proc_pid_cmdline_read+0xd9/0x4e0
> [<ffffffff81253c95>] ? do_filp_open+0xa5/0x100
> [<ffffffff81241d87>] __vfs_read+0x37/0x150
> [<ffffffff812f824b>] ? security_file_permission+0x9b/0xc0
> [<ffffffff81242266>] vfs_read+0x96/0x130
> [<ffffffff812437b5>] SyS_read+0x55/0xc0
> [<ffffffff8171a6da>] entry_SYSCALL_64_fastpath+0x1a/0xc5
Yes, but
>
> Both Alexey Dobriyan and Michal Hocko suggested to use dedicated lock
> for them to mitigate the abuse of mmap_sem.
>
> So, introduce a new rwlock in mm_struct to protect the concurrent access
> to arg_start|end and env_start|end.
does arg_lock really help?
I wonder whether per "struct mm_struct" granularity is needed if arg_lock
protects only a few atomic reads. A global lock would be sufficient.
Also, even if we succeeded to avoid mmap_sem contention at that location,
won't we after all get mmap_sem contention messages a bit later, for
access_remote_vm() holds mmap_sem which would lead to traces like above
if mmap_sem is already contended?
>
> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
> Cc: Alexey Dobriyan <adobriyan@gmail.com>
> Cc: Michal Hocko <mhocko@kernel.org>
> ---
> fs/proc/base.c | 8 ++++----
> include/linux/mm_types.h | 2 ++
> kernel/fork.c | 1 +
> kernel/sys.c | 6 ++++++
> mm/init-mm.c | 1 +
> 5 files changed, 14 insertions(+), 4 deletions(-)
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct
2018-03-26 14:49 ` Tetsuo Handa
@ 2018-03-26 14:55 ` Michal Hocko
2018-03-26 15:31 ` Yang Shi
1 sibling, 0 replies; 6+ messages in thread
From: Michal Hocko @ 2018-03-26 14:55 UTC (permalink / raw)
To: Tetsuo Handa
Cc: Yang Shi, Matthew Wilcox, adobriyan, akpm, linux-mm, linux-kernel
On Mon 26-03-18 23:49:18, Tetsuo Handa wrote:
[...]
> Also, even if we succeeded to avoid mmap_sem contention at that location,
> won't we after all get mmap_sem contention messages a bit later, for
> access_remote_vm() holds mmap_sem which would lead to traces like above
> if mmap_sem is already contended?
Yes, but at least we get rid of the mmap_sem for something that can use
a more fine grained locking.
Maybe we can get a finer grained range locking for mmap_sem one day and
not having the full range locked section will be a plus.
--
Michal Hocko
SUSE Labs
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct
2018-03-26 14:49 ` Tetsuo Handa
2018-03-26 14:55 ` Michal Hocko
@ 2018-03-26 15:31 ` Yang Shi
1 sibling, 0 replies; 6+ messages in thread
From: Yang Shi @ 2018-03-26 15:31 UTC (permalink / raw)
To: Tetsuo Handa, Matthew Wilcox
Cc: adobriyan, mhocko, akpm, linux-mm, linux-kernel
On 3/26/18 10:49 AM, Tetsuo Handa wrote:
> On 2018/03/24 9:36, Yang Shi wrote:
>> And, the mmap_sem contention may cause unexpected issue like below:
>>
>> INFO: task ps:14018 blocked for more than 120 seconds.
>> Tainted: G E 4.9.79-009.ali3000.alios7.x86_64 #1
>> "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this
>> message.
>> ps D 0 14018 1 0x00000004
>> ffff885582f84000 ffff885e8682f000 ffff880972943000 ffff885ebf499bc0
>> ffff8828ee120000 ffffc900349bfca8 ffffffff817154d0 0000000000000040
>> 00ffffff812f872a ffff885ebf499bc0 024000d000948300 ffff880972943000
>> Call Trace:
>> [<ffffffff817154d0>] ? __schedule+0x250/0x730
>> [<ffffffff817159e6>] schedule+0x36/0x80
>> [<ffffffff81718560>] rwsem_down_read_failed+0xf0/0x150
>> [<ffffffff81390a28>] call_rwsem_down_read_failed+0x18/0x30
>> [<ffffffff81717db0>] down_read+0x20/0x40
>> [<ffffffff812b9439>] proc_pid_cmdline_read+0xd9/0x4e0
>> [<ffffffff81253c95>] ? do_filp_open+0xa5/0x100
>> [<ffffffff81241d87>] __vfs_read+0x37/0x150
>> [<ffffffff812f824b>] ? security_file_permission+0x9b/0xc0
>> [<ffffffff81242266>] vfs_read+0x96/0x130
>> [<ffffffff812437b5>] SyS_read+0x55/0xc0
>> [<ffffffff8171a6da>] entry_SYSCALL_64_fastpath+0x1a/0xc5
> Yes, but
>
>> Both Alexey Dobriyan and Michal Hocko suggested to use dedicated lock
>> for them to mitigate the abuse of mmap_sem.
>>
>> So, introduce a new rwlock in mm_struct to protect the concurrent access
>> to arg_start|end and env_start|end.
> does arg_lock really help?
>
> I wonder whether per "struct mm_struct" granularity is needed if arg_lock
> protects only a few atomic reads. A global lock would be sufficient.
However, a global lock might be hard to know what it is used for, and
might be abused again.
And, it may introduce unexpected contention for parallel reading for /proc
>
> Also, even if we succeeded to avoid mmap_sem contention at that location,
> won't we after all get mmap_sem contention messages a bit later, for
> access_remote_vm() holds mmap_sem which would lead to traces like above
> if mmap_sem is already contended?
Yes, definitely, this patch is aimed to remove the abuse to mmap_sem.
The mmap_sem contention will be addressed separately.
Yang
>
>> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
>> Cc: Alexey Dobriyan <adobriyan@gmail.com>
>> Cc: Michal Hocko <mhocko@kernel.org>
>> ---
>> fs/proc/base.c | 8 ++++----
>> include/linux/mm_types.h | 2 ++
>> kernel/fork.c | 1 +
>> kernel/sys.c | 6 ++++++
>> mm/init-mm.c | 1 +
>> 5 files changed, 14 insertions(+), 4 deletions(-)
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2018-03-26 15:31 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-03-24 0:36 [PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct Yang Shi
2018-03-24 4:30 ` Matthew Wilcox
2018-03-25 0:32 ` Yang Shi
2018-03-26 14:49 ` Tetsuo Handa
2018-03-26 14:55 ` Michal Hocko
2018-03-26 15:31 ` Yang Shi
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).