pstore: Correctly initialize spinlock and flags
authorKees Cook <keescook@chromium.org>
Thu, 9 Feb 2017 23:43:44 +0000 (15:43 -0800)
committerKees Cook <keescook@chromium.org>
Mon, 13 Feb 2017 18:25:52 +0000 (10:25 -0800)
commit76d5692a58031696e282384cbd893832bc92bd76
tree95d2c1eed40eef5adbce92e1026139e530f97d69
parent7089db84e356562f8ba737c29e472cc42d530dbc
pstore: Correctly initialize spinlock and flags

The ram backend wasn't always initializing its spinlock correctly. Since
it was coming from kzalloc memory, though, it was harmless on
architectures that initialize unlocked spinlocks to 0 (at least x86 and
ARM). This also fixes a possibly ignored flag setting too.

When running under CONFIG_DEBUG_SPINLOCK, the following Oops was visible:

[    0.760836] persistent_ram: found existing buffer, size 29988, start 29988
[    0.765112] persistent_ram: found existing buffer, size 30105, start 30105
[    0.769435] persistent_ram: found existing buffer, size 118542, start 118542
[    0.785960] persistent_ram: found existing buffer, size 0, start 0
[    0.786098] persistent_ram: found existing buffer, size 0, start 0
[    0.786131] pstore: using zlib compression
[    0.790716] BUG: spinlock bad magic on CPU#0, swapper/0/1
[    0.790729]  lock: 0xffffffc0d1ca9bb0, .magic: 00000000, .owner: <none>/-1, .owner_cpu: 0
[    0.790742] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.10.0-rc2+ #913
[    0.790747] Hardware name: Google Kevin (DT)
[    0.790750] Call trace:
[    0.790768] [<ffffff900808ae88>] dump_backtrace+0x0/0x2bc
[    0.790780] [<ffffff900808b164>] show_stack+0x20/0x28
[    0.790794] [<ffffff9008460ee0>] dump_stack+0xa4/0xcc
[    0.790809] [<ffffff9008113cfc>] spin_dump+0xe0/0xf0
[    0.790821] [<ffffff9008113d3c>] spin_bug+0x30/0x3c
[    0.790834] [<ffffff9008113e28>] do_raw_spin_lock+0x50/0x1b8
[    0.790846] [<ffffff9008a2d2ec>] _raw_spin_lock_irqsave+0x54/0x6c
[    0.790862] [<ffffff90083ac3b4>] buffer_size_add+0x48/0xcc
[    0.790875] [<ffffff90083acb34>] persistent_ram_write+0x60/0x11c
[    0.790888] [<ffffff90083aab1c>] ramoops_pstore_write_buf+0xd4/0x2a4
[    0.790900] [<ffffff90083a9d3c>] pstore_console_write+0xf0/0x134
[    0.790912] [<ffffff900811c304>] console_unlock+0x48c/0x5e8
[    0.790923] [<ffffff900811da18>] register_console+0x3b0/0x4d4
[    0.790935] [<ffffff90083aa7d0>] pstore_register+0x1a8/0x234
[    0.790947] [<ffffff90083ac250>] ramoops_probe+0x6b8/0x7d4
[    0.790961] [<ffffff90085ca548>] platform_drv_probe+0x7c/0xd0
[    0.790972] [<ffffff90085c76ac>] driver_probe_device+0x1b4/0x3bc
[    0.790982] [<ffffff90085c7ac8>] __device_attach_driver+0xc8/0xf4
[    0.790996] [<ffffff90085c4bfc>] bus_for_each_drv+0xb4/0xe4
[    0.791006] [<ffffff90085c7414>] __device_attach+0xd0/0x158
[    0.791016] [<ffffff90085c7b18>] device_initial_probe+0x24/0x30
[    0.791026] [<ffffff90085c648c>] bus_probe_device+0x50/0xe4
[    0.791038] [<ffffff90085c35b8>] device_add+0x3a4/0x76c
[    0.791051] [<ffffff90087d0e84>] of_device_add+0x74/0x84
[    0.791062] [<ffffff90087d19b8>] of_platform_device_create_pdata+0xc0/0x100
[    0.791073] [<ffffff90087d1a2c>] of_platform_device_create+0x34/0x40
[    0.791086] [<ffffff900903c910>] of_platform_default_populate_init+0x58/0x78
[    0.791097] [<ffffff90080831fc>] do_one_initcall+0x88/0x160
[    0.791109] [<ffffff90090010ac>] kernel_init_freeable+0x264/0x31c
[    0.791123] [<ffffff9008a25bd0>] kernel_init+0x18/0x11c
[    0.791133] [<ffffff9008082ec0>] ret_from_fork+0x10/0x50
[    0.793717] console [pstore-1] enabled
[    0.797845] pstore: Registered ramoops as persistent store backend
[    0.804647] ramoops: attached 0x100000@0xf7edc000, ecc: 0/0

Fixes: 663deb47880f ("pstore: Allow prz to control need for locking")
Fixes: 109704492ef6 ("pstore: Make spinlock per zone instead of global")
Reported-by: Brian Norris <briannorris@chromium.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
fs/pstore/ram_core.c