mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Vasily Averin <vvs@virtuozzo.com>
To: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: linux-kernel@vger.kernel.org,
	Hari Bathini <hbathini@linux.vnet.ibm.com>,
	Namhyung Kim <namhyung@kernel.org>, Jiri Olsa <jolsa@redhat.com>,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	Ingo Molnar <mingo@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>
Subject: Re: [PATCH] lost path_put in perf_fill_ns_link_info
Date: Wed, 15 Nov 2017 08:48:38 +0300	[thread overview]
Message-ID: <9e5342b4-ab66-dadb-4cff-0ec73502cd67@virtuozzo.com> (raw)
In-Reply-To: <2d0707fa-7265-8b82-23a0-b3a938111767@virtuozzo.com>

On 2017-11-08 16:04, Vasily Averin wrote:
> On 2017-11-08 15:09, Alexander Shishkin wrote:
>> On Mon, Nov 06, 2017 at 09:22:18AM +0300, Vasily Averin wrote:
>>> Fixes: commit e422267322cd ("perf: Add PERF_RECORD_NAMESPACES to include namespaces related info")
>>> Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
>>
>> The change description is missing. One needs to open the source code and
>> look for proof of correctness for this patch.
> 
> perf_fill_ns_link_info() calls ns_get_path()
> it returns ns_path with increased mnt and dentry counters.
> 
> Problem is that nodody decrement these counters.
> 
> You can call ./perf record --namespaces unshare -m
> and look how grows mount counter on nsfs_mnt. 

Situation is even worse, leaked dentry does not allow to free related namespaces.

[root@localhost ~]# uname -a
Linux localhost.localdomain 4.14.0+ #10 SMP Wed Nov 15 00:31:34 MSK 2017 x86_64 x86_64 x86_64 GNU/Linux

VvS:  without --namespace perf works correctly

[root@localhost ~]# grep namespace /proc/slabinfo
pid_namespace          0      0   2568   12    8 : tunables    0    0    0 : slabdata      0      0      0
user_namespace         0      0    824   39    8 : tunables    0    0    0 : slabdata      0      0      0
net_namespace          0      5   6272    5    8 : tunables    0    0    0 : slabdata      1      1      0
[root@localhost ~]# perf record  -q  unshare -n -U  -p --fork true 
[root@localhost ~]# grep namespace /proc/slabinfo
pid_namespace          0     12   2568   12    8 : tunables    0    0    0 : slabdata      1      1      0
user_namespace         0     39    824   39    8 : tunables    0    0    0 : slabdata      1      1      0
net_namespace          0      5   6272    5    8 : tunables    0    0    0 : slabdata      1      1      0

VvS: with --namespace perf leaks namespaces

[root@localhost ~]# perf record  -q  --namespace unshare -n -U  -p --fork true
[root@localhost ~]# grep namespace /proc/slabinfo
pid_namespace          1     12   2568   12    8 : tunables    0    0    0 : slabdata      1      1      0
user_namespace         1     39    824   39    8 : tunables    0    0    0 : slabdata      1      1      0
net_namespace          1      5   6272    5    8 : tunables    0    0    0 : slabdata      1      1      0

VvS: ... and once again, to be sure

[root@localhost ~]# perf record  -q  --namespace unshare -n -U  -p --fork true
[root@localhost ~]# grep namespace /proc/slabinfo
pid_namespace          2     12   2568   12    8 : tunables    0    0    0 : slabdata      1      1      0
user_namespace         2     39    824   39    8 : tunables    0    0    0 : slabdata      1      1      0
net_namespace          2      5   6272    5    8 : tunables    0    0    0 : slabdata      1      1      0


kmemleak also detects leaks dentry, inode, related structures and namespaces

unreferenced object 0xffff998008e76738 (size 192): <<< dentry
  comm "unshare", pid 1436, jiffies 4294786539 (age 509.114s)
    [<ffffffffbb29f495>] __ns_get_path+0xf5/0x160
    [<ffffffffbb29f5a8>] ns_get_path+0x28/0x50
    [<ffffffffbb1adb00>] perf_fill_ns_link_info+0x20/0x80
    [<ffffffffbb1b1ba7>] perf_event_namespaces.part.101+0xd7/0x120
    [<ffffffffbb09e85d>] copy_process.part.34+0x171d/0x1ae0
    [<ffffffffbb09eddc>] _do_fork+0xcc/0x390
    [<ffffffffbb003a11>] do_syscall_64+0x61/0x170
    [<ffffffffbb86e6fb>] return_from_SYSCALL_64+0x0/0x65
    [<ffffffffffffffff>] 0xffffffffffffffff

unreferenced object 0xffff998008e7c928 (size 600): <<< inode
  comm "unshare", pid 1436, jiffies 4294786539 (age 509.114s)
    [<ffffffffbb286f4e>] new_inode_pseudo+0xe/0x60
    [<ffffffffbb29f3e2>] __ns_get_path+0x42/0x160
    [<ffffffffbb29f5a8>] ns_get_path+0x28/0x50
     ...

unreferenced object 0xffff99800cacd320 (size 40): <<< inode_security_struct
  comm "unshare", pid 1436, jiffies 4294786539 (age 509.114s)
    [<ffffffffbb383876>] security_inode_alloc+0x36/0x50
    [<ffffffffbb284bd5>] inode_init_always+0xf5/0x1d0
    [<ffffffffbb28505b>] alloc_inode+0x2b/0x80
    [<ffffffffbb286f4e>] new_inode_pseudo+0xe/0x60
    [<ffffffffbb29f3e2>] __ns_get_path+0x42/0x160
     ...

unreferenced object 0xffff99800cb88a10 (size 2232): <<< pid_namespace
  comm "unshare", pid 1436, jiffies 4294786439 (age 509.214s)
    [<ffffffffbb0c0104>] create_new_namespaces+0xd4/0x1b0
    [<ffffffffbb0c0359>] unshare_nsproxy_namespaces+0x59/0xb0
    [<ffffffffbb09f415>] SyS_unshare+0x1e5/0x370
    [<ffffffffbb86e673>] entry_SYSCALL_64_fastpath+0x1a/0x7d
    [<ffffffffffffffff>] 0xffffffffffffffff

I've resend the patch as  "[PATCH] memory leaks triggered by perf --namespace"

Thank you,
	Vasily Averin

      reply	other threads:[~2017-11-15  5:48 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-11-06  6:22 Vasily Averin
2017-11-06  6:34 ` Vasily Averin
2017-11-07 19:03 ` Andrei Vagin
2017-11-07 19:51   ` Andrei Vagin
     [not found] ` <20171108120915.n7i46j2yly6fb7bj@ukko.fi.intel.com>
2017-11-08 13:04   ` [PATCH] " Vasily Averin
2017-11-15  5:48     ` Vasily Averin [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=9e5342b4-ab66-dadb-4cff-0ec73502cd67@virtuozzo.com \
    --to=vvs@virtuozzo.com \
    --cc=acme@kernel.org \
    --cc=alexander.shishkin@linux.intel.com \
    --cc=hbathini@linux.vnet.ibm.com \
    --cc=jolsa@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@redhat.com \
    --cc=namhyung@kernel.org \
    --cc=peterz@infradead.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®