memory leak of auth_key.response in multichannel establishment
Subsystems (2)
Mounting an SMB share with multichannel enabled appears to leak all but one allocations of ses->auth_key.response, at least with ntlmssp auth. This was on (roughly) bf1028a41ea:
unreferenced object 0xffff88802cefc500 (size 256):
comm "mount.cifs", pid 680, jiffies 4294907522 (age 381.360s) hex dump (first 32 bytes):
19 68 bf 70 7b a3 b8 00 43 63 8c fa e1 00 90 fb .h.p{...Cc......
1d d8 39 3c 25 d1 d3 f0 a2 4e e4 99 4f 35 9b 2e ..9<%....N..O5..
backtrace:
[<000000006c8d6e81>] setup_ntlmv2_rsp+0x2f4/0x2190
[<000000006cf0d77f>] build_ntlmssp_auth_blob+0x2b/0x10d0
[<00000000300414f8>] SMB2_sess_auth_rawntlmssp_authenticate+0x316/0x7c0
[<000000002ee76096>] SMB2_sess_setup+0x377/0x7b0
[<0000000076fded43>] cifs_setup_session+0x359/0x770
[<00000000f6a70b79>] cifs_get_smb_ses+0xfe1/0x2ae0
[<0000000030763a8d>] mount_get_conns+0x20d/0xec0
[<000000003deb8295>] cifs_mount+0xbe/0x1d00
[<00000000d18bac6c>] cifs_smb3_do_mount+0x26e/0x1330
[<000000000952c786>] legacy_get_tree+0x101/0x1f0
[<00000000f07e0c29>] vfs_get_tree+0x8a/0x2d0
[<00000000edbc514a>] do_mount+0xed7/0x16f0
[<00000000ee452092>] __x64_sys_mount+0x12c/0x1a0
[<00000000a6ccd160>] do_syscall_64+0x56/0xa0
[<000000003143088f>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
unreferenced object 0xffff88802a33e500 (size 256):
comm "mount.cifs", pid 680, jiffies 4294907547 (age 381.260s)
hex dump (first 32 bytes):
5f ff 39 9f c0 e9 1e 3b 0a f1 06 9d c7 53 c2 74 _.9....;.....S.t
03 7e 5f 2f 27 f7 a8 84 70 e8 58 e3 91 f6 76 d6 .~_/'...p.X...v.
backtrace:
[<000000006c8d6e81>] setup_ntlmv2_rsp+0x2f4/0x2190
[<000000006cf0d77f>] build_ntlmssp_auth_blob+0x2b/0x10d0
[<00000000300414f8>] SMB2_sess_auth_rawntlmssp_authenticate+0x316/0x7c0
[<000000002ee76096>] SMB2_sess_setup+0x377/0x7b0
[<0000000076fded43>] cifs_setup_session+0x359/0x770
[<000000006bb9ce4c>] cifs_ses_add_channel+0x9bc/0x10a0
[<000000004034ab3b>] cifs_try_adding_channels+0x1f0/0x580
[<000000009ac39b0b>] cifs_mount+0xed1/0x1d00
[<00000000d18bac6c>] cifs_smb3_do_mount+0x26e/0x1330
[<000000000952c786>] legacy_get_tree+0x101/0x1f0
[<00000000f07e0c29>] vfs_get_tree+0x8a/0x2d0
[<00000000edbc514a>] do_mount+0xed7/0x16f0
[<00000000ee452092>] __x64_sys_mount+0x12c/0x1a0
[<00000000a6ccd160>] do_syscall_64+0x56/0xa0
[<000000003143088f>] entry_SYSCALL_64_after_hwframe+0x44/0xa9Regards,
~Paul
Hi Paul,
These stack traces are where printed when the object is lost right?
Cheers,
Signature · 4 lines
Aurélien Aptel / SUSE Labs Samba Team GPG: 1839 CB5F 9F5B FB9B AA97 8C99 03C8 A49B 521B D5D3 SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg, DE GF: Felix Imendörffer, Mary Higgins, Sri Rasiah HRB 247165 (AG München)
They're the stack trace of where the allocations occurred, for objects that are lost. (I won't pretend to understand kmemleak's algorithm for determining lost objects.)
In this case, I was mounting a server with max_channels=4, which generates three leak reports (one for the initial session, and two with the same stack trace, cifs_try_adding_channels -> cifs_ses_add_channel). (With max_channels=8, I get 7 leak reports)
With
CONFIG_HAVE_DEBUG_KMEMLEAK=y
CONFIG_DEBUG_KMEMLEAK=y
CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE=16000
CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN=yTo reproduce:
- mount -t cifs -o multichannel,max_channels=4,... ...
- echo scan > /sys/kernel/debug/kmemleak
or wait long enough for the kmemleak periodic scan to pick it up
- cat /sys/kernel/debug/kmemleak
~Paul
Quoted 11 lines · Aurélien Aptel
Hi Paul,
These stack traces are where printed when the object is lost right?
Cheers,
Signature · 4 lines
Aurélien Aptel / SUSE Labs Samba Team GPG: 1839 CB5F 9F5B FB9B AA97 8C99 03C8 A49B 521B D5D3 SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg, DE GF: Felix Imendörffer, Mary Higgins, Sri Rasiah HRB 247165 (AG München)