Misplaced driver_sysfs_remove in really_probe?

[Jiri Slaby]

Hi,

since commit 1901fb2604fbcd53201f38725182ea807581159e
Author: Kay Sievers <kay.sievers@xxxxxxxxxx>
Date:   Sat Oct 7 21:55:55 2006 +0200

    Driver core: fix "driver" symlink timing

driver_sysfs_remove seems to be misplaced in the fail path of
really_probe. When driver_sysfs_add fails (or anything which is
currently above it in dd.c -- be it pinctrl_bind_pins or
dev->bus->dma_configure), driver_sysfs_remove is called. Given
dev->driver is set, attempt to remove sysfs device and driver links is
performed, but it is supposed to fail, as the links do not exist yet.

I am dealing with a Syzkaller WARNING from SLE15-SP1 (4.12) which
corresponds to the described scenario. Perhaps Syzkaller fault-injected
a kzalloc failure to pinctrl_bind_pins as I cannot reproduce the report
at all:
> WARNING: CPU: 1 PID: 2091 at ../fs/kernfs/dir.c:1481
kernfs_remove_by_name_ns+0xfa/0x120 fs/kernfs/dir.c:1480
> Supported: No, Unreleased kernel
> CPU: 1 PID: 2091 Comm: systemd-udevd Not tainted 4.12.14-396-default
#1 SLE15-SP1 (unreleased)
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
rel-1.12.0-0-ga698c89-prebuilt.qemu.org 04/01/2014
> task: ffff880053794fc0 task.stack: ffff880040ea0000
> RIP: 0010:kernfs_remove_by_name_ns+0xfa/0x120 fs/kernfs/dir.c:1480
> RSP: 0018:ffff880040ea7488 EFLAGS: 00010282
> RAX: 000000000000002d RBX: 0000000000000000 RCX: 0000000000000000
> RDX: 000000000000002d RSI: 1ffff100081d4e48 RDI: ffffed00081d4e85
> RBP: ffffffffa772a380 R08: 0000000000000000 R09: 0000000000000000
> R10: ffffffffaa1872c4 R11: 0000000000000000 R12: 0000000000000000
> R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000037
> FS:  00007f0d24dc0f80(0000) GS:ffff88005e080000(0000)
knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 000000000045c061 CR3: 000000003b878006 CR4: 0000000000060ee0
> Call Trace:
>  driver_sysfs_remove+0xb0/0x110 drivers/base/dd.c:290
>  really_probe drivers/base/dd.c:433 [inline]
>  driver_probe_device+0x2b3/0x1200 drivers/base/dd.c:530
>  __driver_attach+0x1dc/0x280 drivers/base/dd.c:763
>  bus_for_each_dev+0x146/0x1e0 drivers/base/bus.c:316
>  bus_add_driver+0x40f/0x850 drivers/base/bus.c:710
>  driver_register+0x1c9/0x410 drivers/base/driver.c:168
>  __hid_register_driver+0x1e0/0x2d0 drivers/hid/hid-core.c:2974
>  ? 0xffffffffc1510000
>  do_one_initcall+0xb7/0x300 init/main.c:808
>  do_init_module+0x23e/0x641 kernel/module.c:3515
>  load_module+0x47d6/0x60b0 kernel/module.c:3867
>  SYSC_finit_module+0x239/0x2a0 kernel/module.c:3980
>  do_syscall_64+0x26c/0x6e0 arch/x86/entry/common.c:284
>  entry_SYSCALL_64_after_hwframe+0x3d/0xa2

I believe it survived from 2.6.20 to the current tree.

Does it look correct to you? This should help IMO and if you agree I
will send a proper patch:
--- a/drivers/base/dd.c
+++ b/drivers/base/dd.c
@@ -496,7 +496,7 @@ static int really_probe(struct device *dev, struct
device_driver *drv)
        if (driver_sysfs_add(dev)) {
                printk(KERN_ERR "%s: driver_sysfs_add(%s) failed\n",
                        __func__, dev_name(dev));
-               goto probe_failed;
+               goto sysfs_failed;
        }

        if (dev->pm_domain && dev->pm_domain->activate) {
@@ -546,6 +546,8 @@ static int really_probe(struct device *dev, struct
device_driver *drv)
        goto done;

 probe_failed:
+       driver_sysfs_remove(dev);
+sysfs_failed:
        arch_teardown_dma_ops(dev);
 dma_failed:
        if (dev->bus)
@@ -554,7 +556,6 @@ static int really_probe(struct device *dev, struct
device_driver *drv)
 pinctrl_bind_failed:
        device_links_no_driver(dev);
        devres_release_all(dev);
-       driver_sysfs_remove(dev);
        dev->driver = NULL;
        dev_set_drvdata(dev, NULL);
        if (dev->pm_domain && dev->pm_domain->dismiss)




thanks,
-- 
js
suse labs