BUG: Securityfs and bind mounts (2.6.34)
Hi guys,
If I overlay a file in securityfs using mount --bind with a file from a regular filesystem, should I be allowed to rmmod the module which registered the overlaid securityfs file? I was able to do that, then I unmounted the bind mount, and then when attempting to unmount securityfs hit a BUG at fs/dcache.c:676 (see below). It would have made more sense to first umount the overlay file and then remove the module which registered with securityfs, nevertheless should kernel crash in this case?
Jul 8 10:33:46 deuteros kernel: [2486427.077574] BUG: Dentry ffff880102684f00{i=cb6e217,n=talpa} still in use (2) [unmount of securityfs
securityfs]
Jul 8 10:33:46 deuteros kernel: [2486427.077610] ------------[ cut here ]------------
Jul 8 10:33:46 deuteros kernel: [2486427.077618] kernel BUG at fs/dcache.c:676!
Jul 8 10:33:46 deuteros kernel: [2486427.077625] invalid opcode: 0000 [#1] PREEMPT SMP
Jul 8 10:33:46 deuteros kernel: [2486427.077634] last sysfs file: /sys/devices/system/cpu/cpu1/cache/index2/shared_cpu_map
Jul 8 10:33:46 deuteros kernel: [2486427.077642] CPU 0
Jul 8 10:33:46 deuteros kernel: [2486427.077645] Modules linked in: cpufreq_stats snd_seq_dummy sysprof_module nls_utf8 cifs snd_pcm_osssnd_mixer_oss snd_seq snd_seq_device edd n
fs nfsd fscache lockd nfs_acl sunrpc exportfs af_packet radeon ttm drm_kms_helper drm i2c_algo_bit bridge stp llc cpufreq_conservative cpufreq_userspace cpufreq_powersave acpi_cpuf
req fuse loop snd_hda_codec_realtek snd_hda_intel snd_hda_codec snd_hwdep snd_pcm dcdbas iTCO_wdt iTCO_vendor_support kvm_intel snd_timer tg3 snd kvm intel_agp snd_page_alloc broad
com sr_mod cdrom button libphy pcspkr sg ext4 jbd2 crc16 linear raid456 async_raid6_recov async_pq raid6_pq async_xor xor async_memcpy async_tx raid1 raid0 dm_snapshot dm_mod fan p
rocessor thermal thermal_sys [last unloaded: talpa_syscallhook]
Jul 8 10:33:46 deuteros kernel: [2486427.077756] Jul 8 10:33:46 deuteros kernel: [2486427.077763] Pid: 5328, comm: umount Not tainted 2.6.34 #3 0F0TGN/OptiPlex 380 Jul 8 10:33:46 deuteros kernel: [2486427.077771] RIP: 0010:[<ffffffff811085c8>] [<ffffffff811085c8>] shrink_dcache_for_umount_subtree+0x268/0x270 Jul 8 10:33:46 deuteros kernel: [2486427.077787] RSP: 0018:ffff8800140f7de8 EFLAGS: 00010296 Jul 8 10:33:46 deuteros kernel: [2486427.077794] RAX: 000000000000007b RBX: ffff880102684f00 RCX: 0000000000000000 Jul 8 10:33:46 deuteros kernel: [2486427.077802] RDX: ffff880001800000 RSI: 0000000000000046 RDI: ffffffff81661e90 Jul 8 10:33:46 deuteros kernel: [2486427.077810] RBP: ffff8800140f7e18 R08: 0000000000000000 R09: 0000000000000005 Jul 8 10:33:46 deuteros kernel: [2486427.077818] R10: 0000000000000000 R11: 00000000000000be R12: 0000000000000002 Jul 8 10:33:46 deuteros kernel: [2486427.077826] R13: ffff880102684f00 R14: ffff8800173678a0 R15: ffff880129aa2be0 Jul 8 10:33:46 deuteros kernel: [2486427.077835] FS: 00007fdaa0dd8730(0000) GS:ffff880001800000(0000) knlGS:0000000000000000 Jul 8 10:33:46 deuteros kernel: [2486427.077843] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 Jul 8 10:33:46 deuteros kernel: [2486427.077851] CR2: 00007fdaa06b3170 CR3: 0000000011fa9000 CR4: 00000000000406f0 Jul 8 10:33:46 deuteros kernel: [2486427.077859] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 Jul 8 10:33:46 deuteros kernel: [2486427.077867] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 Jul 8 10:33:46 deuteros kernel: [2486427.077876] Process umount (pid: 5328, threadinfo ffff8800140f6000, task ffff880129aa2be0) Jul 8 10:33:46 deuteros kernel: [2486427.077884] Stack: Jul 8 10:33:46 deuteros kernel: [2486427.077889] ffff880129072e68 ffffffff810ece32 ffff880129072c00 ffffffff81406940 Jul 8 10:33:46 deuteros kernel: [2486427.077897] <0> ffffffff815cea40 ffff880129072c00 ffff8800140f7e38 ffffffff81108601 Jul 8 10:33:46 deuteros kernel: [2486427.077907] <0> 0000000000000000 ffff880129072c00 ffff8800140f7e58 ffffffff810f72ba Jul 8 10:33:46 deuteros kernel: [2486427.077921] Call Trace: Jul 8 10:33:46 deuteros kernel: [2486427.077930] [<ffffffff810ece32>] ? add_partial+0x52/0x80 Jul 8 10:33:46 deuteros kernel: [2486427.077938] [<ffffffff81108601>] shrink_dcache_for_umount+0x31/0x50 Jul 8 10:33:46 deuteros kernel: [2486427.077948] [<ffffffff810f72ba>] generic_shutdown_super+0x1a/0x100 Jul 8 10:33:46 deuteros kernel: [2486427.077958] [<ffffffff810f7401>] kill_anon_super+0x11/0x60 Jul 8 10:33:46 deuteros kernel: [2486427.077967] [<ffffffff810f7472>] kill_litter_super+0x22/0x30 Jul 8 10:33:46 deuteros kernel: [2486427.077975] [<ffffffff810f8730>] deactivate_super+0x70/0xa0 Jul 8 10:33:46 deuteros kernel: [2486427.077985] [<ffffffff81110726>] mntput_no_expire+0xa6/0xf0 Jul 8 10:33:46 deuteros kernel: [2486427.077994] [<ffffffff81110a63>] sys_umount+0x73/0x3b0 Jul 8 10:33:46 deuteros kernel: [2486427.078003] [<ffffffff810026eb>] system_call_fastpath+0x16/0x1b Jul 8 10:33:46 deuteros kernel: [2486427.078003] Code: 0a 48 8b 4b 38 31 d2 48 85 f6 74 04 48 8b 56 40 48 05 68 02 00 00 48 89 de 48 89 04 24 48 c7 c7 f8 f1 50 81 31 c0 e8 8e 6e 2d 00 <0f> 0b eb fe 0f 0b eb fe 55 48 89 e5 53 48 89 fb 48 83 ec 08 48 Jul 8 10:33:46 deuteros kernel: [2486427.078003] RIP [<ffffffff811085c8>] shrink_dcache_for_umount_subtree+0x268/0x270 Jul 8 10:33:46 deuteros kernel: [2486427.078003] RSP <ffff8800140f7de8> Jul 8 10:33:46 deuteros kernel: [2486427.078504] ---[ end trace 289722301befe423 ]---
Following that I could not sync are reboot because it looks some mutex was left locked:
Jul 8 10:36:02 deuteros kernel: [2486562.304208] SysRq : Show Blocked State Jul 8 10:36:02 deuteros kernel: [2486562.304220] task PC stack pid father Jul 8 10:36:02 deuteros kernel: [2486562.304247] shutdown D 00000001943089cb 0 5475 2001 0x00000000 Jul 8 10:36:02 deuteros kernel: [2486562.304251] ffff88012a341e30 0000000000000086 ffff88012b824bb0 ffffffff00000000 Jul 8 10:36:02 deuteros kernel: [2486562.304251] ffff88012a340000 ffff88012a341fd8 0000000000013500 ffff88012a341fd8 Jul 8 10:36:02 deuteros kernel: [2486562.304251] ffff88012a341fd8 ffff8800140c0000 ffff88012a340000 ffff88012a340000 Jul 8 10:36:02 deuteros kernel: [2486562.304251] Call Trace: Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff810396f4>] ? try_to_wake_up+0x334/0x420 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff813e1f45>] rwsem_down_failed_common+0x95/0x200 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff813e2106>] rwsem_down_read_failed+0x26/0x30 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff811c8574>] call_rwsem_down_read_failed+0x14/0x30 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff813e12c2>] ? down_read+0x12/0x20 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff8111b450>] sync_filesystems+0xb0/0x130 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff8111b522>] sys_sync+0x12/0x40 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff810026eb>] system_call_fastpath+0x16/0x1b Jul 8 10:36:02 deuteros kernel: [2486562.304251] sync D 0000000194310777 0 5534 5493 0x00000000 Jul 8 10:36:02 deuteros kernel: [2486562.304251] ffff88012a5ebea8 0000000000000082 ffff88012a5ebe38 ffffffff00000000 Jul 8 10:36:02 deuteros kernel: [2486562.304251] ffff88012a5ea000 ffff88012a5ebfd8 0000000000013500 ffff88012a5ebfd8 Jul 8 10:36:02 deuteros kernel: [2486562.304251] ffff88012a5ebfd8 ffff88011fe5d7c0 ffff88012a5ea000 ffff88012a5ea000 Jul 8 10:36:02 deuteros kernel: [2486562.304251] Call Trace: Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff813e0cd7>] __mutex_lock_slowpath+0x127/0x1e0 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff813e082e>] mutex_lock+0x1e/0x40 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff8111b3bc>] sync_filesystems+0x1c/0x130 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff8111b522>] sys_sync+0x12/0x40 Jul 8 10:36:02 deuteros kernel: [2486562.304251] [<ffffffff810026eb>] system_call_fastpath+0x16/0x1b
Sophos Plc, The Pentagon, Abingdon Science Park, Abingdon, OX14 3YP, United Kingdom. Company Reg No 2096520. VAT Reg No GB 348 3873 20.
Quoted 6 lines · Tvrtko Ursulin
Hi guys,
If I overlay a file in securityfs using mount --bind with a file from a regular filesystem, should I be allowed to rmmod the module which registered the overlaid securityfs file?
Why would you want to overlay securityfs in the first place?
And you might be able to rmmod the module, but I didn't think that security modules were able to be unloaded anymore.
Quoted 6 lines
I was able to do that, then I
unmounted the bind mount, and then when attempting to unmount securityfs hit a BUG atfs/dcache.c:676(see below). It would have made more sense to first umount the overlay file and then remove the module which registered with securityfs, nevertheless should kernel crash in this case?
Probably not, but then again, you did something that you shouldn't have, so perhaps it is telling you not to do such a thing in the future :)
thanks,
greg k-h
Quoted 8 lines · On Thursday 08 Jul 2010 15:43:17 Greg KH wrote:
Quoted 5 lines · Tvrtko Ursulin
Hi guys,
If I overlay a file in securityfs using mount --bind with a file from a regular filesystem, should I be allowed to rmmod the module which registered the overlaid securityfs file?
Why would you want to overlay securityfs in the first place?
For testing, more precisely faking some data exposed in securityfs module in order to provoke userspace reaction. It was convenient to leave the majority of real data and just overlay one file.
Quoted 2 lines
And you might be able to rmmod the module, but I didn't think that security modules were able to be unloaded anymore.
Perhaps it is not a security module in the way you think about it, just a module which happens to register some directories and files under securityfs.
Quoted 9 lines
Quoted 6 lines
I was able to do that, then I
unmounted the bind mount, and then when attempting to unmount securityfs hit a BUG atfs/dcache.c:676(see below). It would have made more sense to first umount the overlay file and then remove the module which registered with securityfs, nevertheless should kernel crash in this case?Probably not, but then again, you did something that you shouldn't have, so perhaps it is telling you not to do such a thing in the future :)
:) Well I do not know, but, it kind of smelled like a bug in the vfs/mount handling/securityfs area so I thought to let experts know. I _think_ I did nothing that much wrong. Just used the exposed API (securityfs_remove) and some bind mount shuffling from userspace.
Tvrtko
Sophos Plc, The Pentagon, Abingdon Science Park, Abingdon, OX14 3YP, United Kingdom. Company Reg No 2096520. VAT Reg No GB 348 3873 20.
Quoted 14 lines · Tvrtko Ursulin
Quoted 8 lines · On Thursday 08 Jul 2010 15:43:17 Greg KH wrote:
Quoted 5 lines · Tvrtko Ursulin
Hi guys,
If I overlay a file in securityfs using mount --bind with a file from a regular filesystem, should I be allowed to rmmod the module which registered the overlaid securityfs file?
Why would you want to overlay securityfs in the first place?
For testing, more precisely faking some data exposed in securityfs module in order to provoke userspace reaction. It was convenient to leave the majority of real data and just overlay one file.
Quoted 5 lines
Quoted 2 lines
And you might be able to rmmod the module, but I didn't think that security modules were able to be unloaded anymore.
Perhaps it is not a security module in the way you think about it, just a module which happens to register some directories and files under securityfs.
Ick, don't do that then :)
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
I was able to do that, then I
unmounted the bind mount, and then when attempting to unmount securityfs hit a BUG atfs/dcache.c:676(see below). It would have made more sense to first umount the overlay file and then remove the module which registered with securityfs, nevertheless should kernel crash in this case?
Quoted 7 lines
Quoted 2 lines
Probably not, but then again, you did something that you shouldn't have, so perhaps it is telling you not to do such a thing in the future :)
:) Well I do not know, but, it kind of smelled like a bug in the vfs/mount handling/securityfs area so I thought to let experts know. I _think_ I did nothing that much wrong. Just used the exposed API (
securityfs_remove) and some bind mount shuffling from userspace.
securitfs just uses libfs underneath it, and really doesn't have any bindings for module ownerships, so I wouldn't recommend doing what you just did.
thanks,
greg k-h
Quoted 10 lines · On Thursday 08 Jul 2010 16:20:59 Greg KH wrote:
Quoted 6 lines
:) Well I do not know, but, it kind of smelled like a bug in the : vfs/mount
handling/securityfs area so I thought to let experts know. I _think_ I did nothing that much wrong. Just used the exposed API (
securityfs_remove) and some bind mount shuffling from userspace.securitfs just uses libfs underneath it, and really doesn't have any bindings for module ownerships, so I wouldn't recommend doing what you just did.
Just do double check what you are saying, securityfs is not safe for use from modules? If so I would then recommend removing the exports otherwise it is an invitation to shoot yourself into the foot. Also, in-three TPM driver can be built as a module so how does that work?
Tvrtko
Sophos Plc, The Pentagon, Abingdon Science Park, Abingdon, OX14 3YP, United Kingdom. Company Reg No 2096520. VAT Reg No GB 348 3873 20.
Quoted 8 lines · Tvrtko Ursulin
Quoted 7 lines · On Thursday 08 Jul 2010 16:20:59 Greg KH wrote:
Quoted 6 lines
:) Well I do not know, but, it kind of smelled like a bug in the : vfs/mount
handling/securityfs area so I thought to let experts know. I _think_ I did nothing that much wrong. Just used the exposed API (
securityfs_remove) and some bind mount shuffling from userspace.
Quoted 7 lines
Quoted 3 lines
securitfs just uses libfs underneath it, and really doesn't have any bindings for module ownerships, so I wouldn't recommend doing what you just did.
Just do double check what you are saying, securityfs is not safe for use from modules? If so I would then recommend removing the exports otherwise it is an invitation to shoot yourself into the foot.
Hm, did you properly set the module owner of the file_operations that you passed to securityfs? That should protect if you have an open file, but I doubt anyone thought you would do crazy things like bind mounts on top of a ramfs and then think it was safe to unload a lower module :)
Quoted 2 lines
Also, in-three TPM driver can be built as a module so how does that work?
You have to be root to unload modules, and if you are that, you can do worse things than this.
thanks,
greg k-h
Quoted 13 lines · On Thursday 08 Jul 2010 16:46:48 Greg KH wrote:
Quoted 12 lines · Tvrtko Ursulin
Quoted 10 lines · On Thursday 08 Jul 2010 16:20:59 Greg KH wrote:
Quoted 6 lines
:) Well I do not know, but, it kind of smelled like a bug in the : vfs/mount
handling/securityfs area so I thought to let experts know. I _think_ I did nothing that much wrong. Just used the exposed API (
securityfs_remove) and some bind mount shuffling from userspace.securitfs just uses libfs underneath it, and really doesn't have any bindings for module ownerships, so I wouldn't recommend doing what you just did.
Quoted 6 lines
Quoted 3 lines
Just do double check what you are saying, securityfs is not safe for use from modules? If so I would then recommend removing the exports otherwise it is an invitation to shoot yourself into the foot.
Hm, did you properly set the module owner of the
file_operationsthat you passed to securityfs? That should protect if you have an open file,
I do not have it set (did not think it is strictly needed for filesystems (before mention TPM also does not set it)), but I am also pretty sure no one had a file open there at the time. I will try to repeat the test with it set and get back to you.
Quoted 8 lines
but I doubt anyone thought you would do crazy things like bind mounts on top of a ramfs and then think it was safe to unload a lower module :)
Quoted 2 lines
Also, in-three TPM driver can be built as a module so how does that work?
You have to be root to unload modules, and if you are that, you can do worse things than this.
That is beside the point, but I will let you off hook :) until I test the situation with the owner field set. Thanks for the discussion so far!
Tvrtko
Sophos Plc, The Pentagon, Abingdon Science Park, Abingdon, OX14 3YP, United Kingdom. Company Reg No 2096520. VAT Reg No GB 348 3873 20.
Quoted 13 lines · On Thursday 08 Jul 2010 16:46:48 Greg KH wrote:
Quoted 12 lines · Tvrtko Ursulin
Quoted 10 lines · On Thursday 08 Jul 2010 16:20:59 Greg KH wrote:
Quoted 6 lines
:) Well I do not know, but, it kind of smelled like a bug in the : vfs/mount
handling/securityfs area so I thought to let experts know. I _think_ I did nothing that much wrong. Just used the exposed API (
securityfs_remove) and some bind mount shuffling from userspace.securitfs just uses libfs underneath it, and really doesn't have any bindings for module ownerships, so I wouldn't recommend doing what you just did.
Quoted 8 lines
Quoted 3 lines
Just do double check what you are saying, securityfs is not safe for use from modules? If so I would then recommend removing the exports otherwise it is an invitation to shoot yourself into the foot.
Hm, did you properly set the module owner of the
file_operationsthat you passed to securityfs? That should protect if you have an open file, but I doubt anyone thought you would do crazy things like bind mounts on top of a ramfs and then think it was safe to unload a lower module :)
Hi again,
Here is the reproducer. Build the attached test module and put test.ko in the cwd. Then run:
umount /sys/kernel/security
mount none -t securityfs /sys/kernel/security/
insmod test.ko
mount file --bind /sys/kernel/security/test/file
rmmod test
insmod test.ko
umount /sys/kernel/security/test/file
umount /sys/kernel/security # <-------- BANG:
It is a bit different than what I initially said. It looks like reference counting on a dentry gets messed up once securityfs_create_dir() in second insmod of test.ko fails with -EEXIST.
HTH,
Tvrtko
Sophos Plc, The Pentagon, Abingdon Science Park, Abingdon, OX14 3YP, United Kingdom. Company Reg No 2096520. VAT Reg No GB 348 3873 20.
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/fs.h>
#include <linux/security.h>
static struct dentry *root;
static struct dentry *file;
static int test_open(struct inode *inode, struct file *file)
{
return -1;
}static struct file_operations test_fops = {
.owner = THIS_MODULE,
.open = test_open,
};
static int __init test_init(void)
{
root = securityfs_create_dir("test", NULL);
if (IS_ERR(root)) {
printk("error creating root %d\n", PTR_ERR(root));
return PTR_ERR(root);
}
file = securityfs_create_file("file", S_IRUSR, root, NULL, &test_fops);
if (IS_ERR(file)) {
printk("error creating file %d\n", PTR_ERR(file));
securityfs_remove(root);
return PTR_ERR(file);
}
return 0;
}static void __exit test_exit(void)
{
securityfs_remove(file);
securityfs_remove(root);
}MODULE_LICENSE("GPL");
module_init(test_init);module_exit(test_exit);