Re: [PATCH v9 01/12] vfio: Mediated device Core driver

From: Kirti Wankhede
Date: Wed Oct 19 2016 - 15:17:57 EST




On 10/19/2016 4:46 AM, Alex Williamson wrote:
> On Tue, 18 Oct 2016 02:52:01 +0530
> Kirti Wankhede <kwankhede@xxxxxxxxxx> wrote:
>
...
>> +static struct mdev_device *__find_mdev_device(struct parent_device *parent,
>> + uuid_le uuid)
>> +{
>> + struct device *dev;
>> +
>> + dev = device_find_child(parent->dev, &uuid, _find_mdev_device);
>> + if (!dev)
>> + return NULL;
>> +
>> + put_device(dev);
>> +
>> + return to_mdev_device(dev);
>> +}
>
> This function is only used by mdev_device_create() for the purpose of
> checking whether a given uuid for a parent already exists, so the
> returned device is not actually used. However, at the point where
> we're using to_mdev_device() here, we don't actually hold a reference to
> the device, so that function call and any possible use of the returned
> pointer by the callee is invalid. I would either turn this into a
> "get" function where the callee has a device reference and needs to do
> a "put" on it or change this to a "exists" test where true/false is
> returned and the function cannot be later mis-used to do a device
> lookup where the reference isn't actually valid.
>

I'll change it to return 0 if not found and -EEXIST if found.


>> +int mdev_device_create(struct kobject *kobj, struct device *dev, uuid_le uuid)
>> +{
>> + int ret;
>> + struct mdev_device *mdev;
>> + struct parent_device *parent;
>> + struct mdev_type *type = to_mdev_type(kobj);
>> +
>> + parent = mdev_get_parent(type->parent);
>> + if (!parent)
>> + return -EINVAL;
>> +
>> + /* Check for duplicate */
>> + mdev = __find_mdev_device(parent, uuid);
>> + if (mdev) {
>> + ret = -EEXIST;
>> + goto create_err;
>> + }
>
> We check here whether the {parent,uuid} already exists, but what
> prevents us racing with another create call with the same uuid? ie.
> neither exists at this point. Will device_register() fail if the
> device name already exists? If so, should we just rely on the error
> there and skip this duplicate check? If not, we need a mutex to avoid
> the race.
>

Yes, device_register() fails if device exists already with below
warning. Is it ok to dump such warning? I think, this should be fine,
right? then we can remove duplicate check.

If we want to avoid such warning, we should have duplication check.

[ 610.847958] ------------[ cut here ]------------
[ 610.855377] WARNING: CPU: 15 PID: 19839 at fs/sysfs/dir.c:31
sysfs_warn_dup+0x64/0x80
[ 610.865798] sysfs: cannot create duplicate filename
'/devices/pci0000:80/0000:80:02.0/0000:83:00.0/0000:84:08.0/0000:85:00.0/83b8f4f2-509f-382f-3c1e-e6bfe0fa1234'
[ 610.885101] Modules linked in:[ 610.888039] nvidia(POE)
vfio_iommu_type1 vfio_mdev mdev vfio nfsv4 dns_resolver nfs fscache
sb_edac edac_core x86_pkg_temp_thermal intel_powerclamp coretemp
kvm_intel kvm irqbypass crct10dif_pclmul crc32_pclmul crc32c_intel
ghash_clmulni_intel aesni_intel glue_helper lrw gf128mul ablk_helper
cryptd nfsd auth_rpcgss nfs_acl lockd mei_me grace iTCO_wdt
iTCO_vendor_support mei ipmi_si pcspkr ioatdma i2c_i801 lpc_ich shpchp
i2c_smbus mfd_core ipmi_msghandler acpi_pad uinput sunrpc xfs libcrc32c
sd_mod mgag200 drm_kms_helper syscopyarea sysfillrect sysimgblt
fb_sys_fops ttm drm igb ahci libahci ptp libata pps_core dca
i2c_algo_bit i2c_core dm_mirror dm_region_hash dm_log dm_mod [last
unloaded: mdev]
[ 610.963835] CPU: 15 PID: 19839 Comm: bash Tainted: P OE
4.8.0-next-20161013+ #0
[ 610.973779] Hardware name: Supermicro
SYS-2027GR-5204A-NC024/X9DRG-HF, BIOS 1.0c 02/28/2013
[ 610.983769] ffffc90009323ae0 ffffffff813568bf ffffc90009323b30
0000000000000000
[ 610.992867] ffffc90009323b20 ffffffff81085511 0000001f00001000
ffff8808839ef000
[ 611.001954] ffff88108b30f900 ffff88109ae368e8 ffff88109ae580b0
ffff881099cc0818
[ 611.011055] Call Trace:
[ 611.015087] [<ffffffff813568bf>] dump_stack+0x63/0x84
[ 611.021784] [<ffffffff81085511>] __warn+0xd1/0xf0
[ 611.028115] [<ffffffff8108558f>] warn_slowpath_fmt+0x5f/0x80
[ 611.035379] [<ffffffff812a4e80>] ? kernfs_path_from_node+0x50/0x60
[ 611.043148] [<ffffffff812a86c4>] sysfs_warn_dup+0x64/0x80
[ 611.050109] [<ffffffff812a87ae>] sysfs_create_dir_ns+0x7e/0x90
[ 611.057481] [<ffffffff81359891>] kobject_add_internal+0xc1/0x340
[ 611.065018] [<ffffffff81359d45>] kobject_add+0x75/0xd0
[ 611.071635] [<ffffffff81483829>] device_add+0x119/0x610
[ 611.078314] [<ffffffff81483d3a>] device_register+0x1a/0x20
[ 611.085261] [<ffffffffa03c748d>] mdev_device_create+0xdd/0x200 [mdev]
[ 611.093143] [<ffffffffa03c7768>] create_store+0xa8/0xe0 [mdev]
[ 611.100385] [<ffffffffa03c76ab>] mdev_type_attr_store+0x1b/0x30 [mdev]
[ 611.108309] [<ffffffff812a7d8a>] sysfs_kf_write+0x3a/0x50
[ 611.115096] [<ffffffff812a78bb>] kernfs_fop_write+0x10b/0x190
[ 611.122231] [<ffffffff81224e97>] __vfs_write+0x37/0x140
[ 611.128817] [<ffffffff811cea84>] ? handle_mm_fault+0x724/0xd80
[ 611.135976] [<ffffffff81225da2>] vfs_write+0xb2/0x1b0
[ 611.142354] [<ffffffff81003510>] ? syscall_trace_enter+0x1d0/0x2b0
[ 611.149836] [<ffffffff812271f5>] SyS_write+0x55/0xc0
[ 611.156065] [<ffffffff81003a47>] do_syscall_64+0x67/0x180
[ 611.162734] [<ffffffff816d41eb>] entry_SYSCALL64_slow_path+0x25/0x25
[ 611.170345] ---[ end trace b05a73599da2ba3f ]---
[ 611.175940] ------------[ cut here ]------------



>> +static ssize_t create_store(struct kobject *kobj, struct device *dev,
>> + const char *buf, size_t count)
>> +{
>> + char *str;
>> + uuid_le uuid;
>> + int ret;
>> +
>> + if (count < UUID_STRING_LEN)
>> + return -EINVAL;
>
>
> Can't we also test for something unreasonably large?
>

Ok. I'll add that check.

>
>> +
>> + str = kstrndup(buf, count, GFP_KERNEL);
>> + if (!str)
>> + return -ENOMEM;
>> +
>> + ret = uuid_le_to_bin(str, &uuid);
>
> nit, we can kfree(str) here regardless of the return.
>
>> + if (!ret) {
>> +
>> + ret = mdev_device_create(kobj, dev, uuid);
>> + if (ret)
>> + pr_err("mdev_create: Failed to create mdev device\n");
>
> What value does this pr_err add? It doesn't tell us why it failed and
> the user will already know if failed by the return value of their write.
>

Ok, will remove it.

>> + else
>> + ret = count;
>> + }
>> +
>> + kfree(str);
>> + return ret;
>> +}

...

>> +static inline struct mdev_driver *to_mdev_driver(struct device_driver *drv)
>> +{
>> + return drv ? container_of(drv, struct mdev_driver, driver) : NULL;
>> +}
>> +
>> +static inline struct mdev_device *to_mdev_device(struct device *dev)
>> +{
>> + return dev ? container_of(dev, struct mdev_device, dev) : NULL;
>
>
> Do we really need this NULL dev/drv behavior? I don't see that any of
> the callers can pass NULL into these. The PCI equivalents don't
> support this behavior and it doesn't seem they need to. Thanks,
>

Ok, I'll update that.

Kirti