fs: fix use-after-free in __fput() when a chardev is removed but a file is still open
Subsystems (5)
PATCH 1/1 fs: fix use-after-free in __fput() when a chardev is removed but a file is still open
In a case when a chardev file (like /dev/ptp0) is open but an underlying device is removed, closing this file leads to a use-after-free. This reproduces easily in a KVM virtual machine:
# cat openptp0.c
int main() { ... fp = fopen("/dev/ptp0", "r"); ... sleep(10); }# uname -r 5.4.0-219d5433 # cat /proc/cmdline ... slub_debug=FZP # modprobe ptp_kvm # ./openptp0 & [1] 670 opened /dev/ptp0, sleeping 10s... # rmmod ptp_kvm # ls /dev/ptp* ls: cannot access '/dev/ptp*': No such file or directory # ...woken up [ 102.375849] general protection fault: 0000 [#1] SMP [ 102.377372] CPU: 1 PID: 670 Comm: openptp0 Not tainted 5.4.0-219d5433 #1 [ 102.379163] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), ... [ 102.381129] RIP: 0010:module_put.part.0+0x7/0x80 [ 102.383019] RSP: 0018:ffff9ba440687e00 EFLAGS: 00010202 [ 102.383451] RAX: 0000000000002000 RBX: 6b6b6b6b6b6b6b6b RCX: ffff91e736800ad0 [ 102.384030] RDX: ffffcf6408bc2808 RSI: 0000000000000247 RDI: 6b6b6b6b6b6b6b6b [ 102.386032] ... ^^^ a slub poison [ 102.389866] Call Trace: [ 102.390086] __fput+0x21f/0x240 [ 102.390363] task_work_run+0x79/0x90 [ 102.390671] do_exit+0x2c9/0xad0 [ 102.390931] ? vfs_write+0x16a/0x190 [ 102.391241] do_group_exit+0x35/0x90 [ 102.391549] __x64_sys_exit_group+0xf/0x10 [ 102.391898] do_syscall_64+0x3d/0x110 [ 102.392240] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 102.392695] RIP: 0033:0x7f0fa7016246 [ 102.396615] ... [ 102.397225] Modules linked in: [last unloaded: ptp_kvm] [ 102.410323] Fixing recursive fault but reboot is needed!
This happens in:
static void __fput(struct file *file)
{ ...
if (file->f_op->release)
file->f_op->release(inode, file); <<< cdev is kfree'd here
if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
!(mode & FMODE_PATH))) {
cdev_put(inode->i_cdev); <<< cdev fields are accessed herebecause of:
__fput()
posix_clock_release()
kref_put(&clk->kref, delete_clock) <<< the last reference
delete_clock()
delete_ptp_clock()
kfree(ptp) <<< cdev is embedded in ptp
cdev_put
module_put(p->owner) <<< *p is kfree'd
The fix is to call cdev_put() before file->f_op->release(). This fix the class of bugs when a chardev device is removed when its file is open, for example:
# lspci 00:09.0 System peripheral: Intel Corporation 6300ESB Watchdog Timer # ./openwdog0 & [1] 672 opened /dev/watchdog0, sleeping 10s... # echo 1 > /sys/devices/pci0000:00/0000:00:09.0/remove # ls /dev/watch* ls: cannot access '/dev/watch*': No such file or directory # ...woken up [ 63.500271] general protection fault: 0000 [#1] SMP [ 63.501757] CPU: 1 PID: 672 Comm: openwdog0 Not tainted 5.4.0-219d5433 #4 [ 63.503605] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), ... [ 63.507064] RIP: 0010:module_put.part.0+0x7/0x80 [ 63.513841] RSP: 0018:ffffb96b00667e00 EFLAGS: 00010202 [ 63.515376] RAX: 0000000000002000 RBX: 6b6b6b6b6b6b6b6b RCX: 0000000000150013 [ 63.517478] RDX: 0000000000000246 RSI: 0000000000000000 RDI: 6b6b6b6b6b6b6b6b
--- fs/file_table.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
fs/file_table.c +2 -2
@@ -276,12 +276,12 @@ static void __fput(struct file *file)if (file->f_op->fasync) file->f_op->fasync (-1 , file,0 ); } -if (file->f_op->release) - file->f_op->release (inode, file);if (unlikely (S_ISCHR (inode->i_mode) && inode->i_cdev !=NULL && !(mode &FMODE_PATH ))) {cdev_put (inode->i_cdev); } +if (file->f_op->release) + file->f_op->release (inode, file);fops_put (file->f_op);put_pid (file->f_owner.pid);if ((mode & (FMODE_READ |FMODE_WRITE )) ==FMODE_READ )
Signature · 1 line
2.20.1
Al Viro
maintainerQuoted 25 lines · Vladis Dronov
In a case when a chardev file (like
/dev/ptp0) is open but an underlying device is removed, closing this file leads to a use-after-free. This reproduces easily in a KVM virtual machine:# cat openptp0.c int main() { ... fp = fopen("/dev/ptp0", "r"); ... sleep(10); }static void
__fput(struct file *file)
{ ...if (file->f_op->release) file->f_op->release(inode, file); <<< cdev is kfree'd here if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && !(mode & FMODE_PATH))) { cdev_put(inode->i_cdev); <<< cdev fields are accessed herebecause of:
__fput()
posix_clock_release()
kref_put(&clk->kref, delete_clock)<<< the last reference
delete_clock()
delete_ptp_clock()
kfree(ptp) <<< cdev is embedded in ptp
cdev_put
Quoted 5 lines
module_put(p->owner) <<< *p is kfree'dThe fix is to call
cdev_put()beforefile->f_op->release(). This fix the class of bugs when a chardev device is removed when its file is open, for example:
And what's to prevent rmmod coming and freeing ->release code right as you are executing it?
Al Viro
maintainerQuoted 28 lines · Al Viro
Quoted 27 lines · Vladis Dronov
In a case when a chardev file (like
/dev/ptp0) is open but an underlying device is removed, closing this file leads to a use-after-free. This reproduces easily in a KVM virtual machine:# cat openptp0.c int main() { ... fp = fopen("/dev/ptp0", "r"); ... sleep(10); }static void
__fput(struct file *file)
{ ...if (file->f_op->release) file->f_op->release(inode, file); <<< cdev is kfree'd here if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && !(mode & FMODE_PATH))) { cdev_put(inode->i_cdev); <<< cdev fields are accessed herebecause of:
__fput()
posix_clock_release()
kref_put(&clk->kref, delete_clock)<<< the last reference
delete_clock()
delete_ptp_clock()
kfree(ptp) <<< cdev is embedded in ptp
cdev_putmodule_put(p->owner) <<< *p is kfree'd
Quoted 6 lines
Quoted 3 lines
The fix is to call
cdev_put()beforefile->f_op->release(). This fix the class of bugs when a chardev device is removed when its file is open, for example:And what's to prevent rmmod coming and freeing ->release code right as you are executing it?
FWIW, the bug here seems to be that the lifetime rules of cdev are fucked - if it can get freed while its ->kobj is still alive, we have something very wrong there. IOW, you have ptp lifetime controlled by *TWO* refcounts - that of clk and that of of cdev->kobj. That's doesn't work. Replace that kfree() with dropping a kobject reference, perhaps, so that freeing would've been done by its release callback?
PATCH v2 1/1 ptp: fix the race between the release of ptp_clock and cdev
In a case when a ptp chardev (like /dev/ptp0) is open but an underlying device is removed, closing this file leads to a race. This reproduces easily in a kvm virtual machine:
ts# cat openptp0.c
int main() { ... fp = fopen("/dev/ptp0", "r"); ... sleep(10); } ts# uname -r
5.5.0-rc3-46cf053e
ts# cat /proc/cmdline
... slub_debug=FZP
ts# modprobe ptp_kvm
ts# ./openptp0 &
[1] 670
opened /dev/ptp0, sleeping 10s...
ts# rmmod ptp_kvm
ts# ls /dev/ptp*
ls: cannot access '/dev/ptp*': No such file or directory ts# ...woken up
[ 48.010809] general protection fault: 0000 [#1] SMP [ 48.012502] CPU: 6 PID: 658 Comm: openptp0 Not tainted 5.5.0-rc3-46cf053e #25 [ 48.014624] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), ... [ 48.016270] RIP: 0010:module_put.part.0+0x7/0x80 [ 48.017939] RSP: 0018:ffffb3850073be00 EFLAGS: 00010202 [ 48.018339] RAX: 000000006b6b6b6b RBX: 6b6b6b6b6b6b6b6b RCX: ffff89a476c00ad0 [ 48.018936] RDX: fffff65a08d3ea08 RSI: 0000000000000247 RDI: 6b6b6b6b6b6b6b6b [ 48.019470] ... ^^^ a slub poison [ 48.023854] Call Trace: [ 48.024050] __fput+0x21f/0x240 [ 48.024288] task_work_run+0x79/0x90 [ 48.024555] do_exit+0x2af/0xab0 [ 48.024799] ? vfs_write+0x16a/0x190 [ 48.025082] do_group_exit+0x35/0x90 [ 48.025387] __x64_sys_exit_group+0xf/0x10 [ 48.025737] do_syscall_64+0x3d/0x130 [ 48.026056] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 48.026479] RIP: 0033:0x7f53b12082f6 [ 48.026792] ... [ 48.030945] Modules linked in: ptp i6300esb watchdog [last unloaded: ptp_kvm] [ 48.045001] Fixing recursive fault but reboot is needed!
This happens in:
static void __fput(struct file *file)
{ ...
if (file->f_op->release)
file->f_op->release(inode, file); <<< cdev is kfree'd here
if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
!(mode & FMODE_PATH))) {
cdev_put(inode->i_cdev); <<< cdev fields are accessed hereNamely:
__fput()
posix_clock_release()
kref_put(&clk->kref, delete_clock) <<< the last reference
delete_clock()
delete_ptp_clock()
kfree(ptp) <<< cdev is embedded in ptp
cdev_put
module_put(p->owner) <<< *p is kfree'd, bang!
Here cdev is embedded in posix_clock which is embedded in ptp_clock. The race happens because ptp_clock's lifetime is controlled by two
refcounts: kref and cdev.kobj in posix_clock. This is wrong.
Make ptp_clock's sysfs device a parent of cdev with cdev_device_add() created especially for such cases. This way the parent device with its ptp_clock is not released until all references to the cdev are released. This adds a requirement that an initialized but not exposed struct device should be provided to posix_clock_register() by a caller instead of a simple dev_t.
This approach was adopted from the commit 72139dfa2464 ("watchdog: Fix the race between the release of watchdog_core_data and cdev"). See details of the implementation in the commit 233ed09d7fda ("chardev: add helper function to register char devs with a struct device").
--- drivers/ptp/ptp_clock.c | 31 ++++++++++++++----------------- drivers/ptp/ptp_private.h | 2 +- include/linux/posix-clock.h | 19 +++++++++++-------- kernel/time/posix-clock.c | 31 +++++++++++++------------------ 4 files changed, 39 insertions(+), 44 deletions(-)
drivers/ptp/ptp_clock.c +14 -17
@@ -166,9 +166,9 @@ static struct posix_clock_operations ptp_clock_ops = { .read = ptp_read, }; -static void delete_ptp_clock (struct posix_clock *pc) +static void ptp_clock_release (struct device *dev) { - struct ptp_clock *ptp = container_of(pc, struct ptp_clock, clock); + struct ptp_clock *ptp = container_of(dev, struct ptp_clock, dev);mutex_destroy (&ptp->tsevq_mux);mutex_destroy (&ptp->pincfg_mux);
@@ -213,7 +213,6 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info, } ptp->clock.ops = ptp_clock_ops; - ptp->clock.release = delete_ptp_clock; ptp->info = info; ptp->devid =MKDEV (major, index); ptp->index = index;
@@ -236,15 +235,6 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,if (err) goto no_pin_groups; -/* Create a new device in our class. */ - ptp->dev =device_create_with_groups (ptp_class, parent, ptp->devid, - ptp, ptp->pin_attr_groups, -"ptp%d" , ptp->index); -if (IS_ERR (ptp->dev)) { - err =PTR_ERR (ptp->dev); - goto no_device; - } -/* Register a new PPS source. */ if (info->pps) {struct pps_source_info pps;
@@ -260,8 +250,18 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info, } } -/* Create a posix clock. */ - err =posix_clock_register (&ptp->clock, ptp->devid); +/* Initialize a new device of our class in our clock structure. */ +device_initialize (&ptp->dev); + ptp->dev.devt = ptp->devid; + ptp->dev.class = ptp_class; + ptp->dev.parent = parent; + ptp->dev.groups = ptp->pin_attr_groups; + ptp->dev.release = ptp_clock_release; +dev_set_drvdata (&ptp->dev, ptp); +dev_set_name (&ptp->dev,"ptp%d" , ptp->index); + +/* Create a posix clock and link it to the device. */ + err =posix_clock_register (&ptp->clock, &ptp->dev);if (err) {pr_err ("failed to create posix clock\n" ); goto no_clock;
@@ -273,8 +273,6 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,if (ptp->pps_source)pps_unregister_source (ptp->pps_source); no_pps: -device_destroy (ptp_class, ptp->devid); -no_device:ptp_cleanup_pin_groups (ptp); no_pin_groups:if (ptp->kworker)
@@ -304,7 +302,6 @@ int ptp_clock_unregister(struct ptp_clock *ptp)if (ptp->pps_source)pps_unregister_source (ptp->pps_source); -device_destroy (ptp_class, ptp->devid);ptp_cleanup_pin_groups (ptp);posix_clock_unregister (&ptp->clock);
drivers/ptp/ptp_private.h +1 -1
@@ -28,7 +28,7 @@ struct timestamp_event_queue {struct ptp_clock {struct posix_clock clock; -struct device *dev; +struct device dev;struct ptp_clock_info *info;dev_t devid;int index;/* index into clocks.map */
include/linux/posix-clock.h +11 -8
@@ -69,29 +69,32 @@ struct posix_clock_operations {* * @ops: Functional interface to the clock * @cdev: Character device instance for this clock -* @kref: Reference count. +* @dev: Pointer to the clock's device. * @rwsem: Protects the 'zombie' field from concurrent access. * @zombie: If 'zombie' is true, then the hardware has disappeared. -* @release: A function to free the structure when the reference count reaches -* zero. May be NULL if structure is statically allocated. * * Drivers should embed their struct posix_clock within a private * structure, obtaining a reference to it during callbacks using * container_of(). +* +* Drivers should supply an initialized but not exposed struct device +* to posix_clock_register(). It is used to manage lifetime of the +* driver's private structure. It's 'release' field should be set to +* a release function for this private structure. */ struct posix_clock {struct posix_clock_operations ops;struct cdev cdev; -struct kref kref; +struct device *dev;struct rw_semaphore rwsem;bool zombie; -void (*release)(struct posix_clock *clk); };/** * posix_clock_register() - register a new clock -* @clk: Pointer to the clock. Caller must provide 'ops' and 'release' -* @devid: Allocated device id +* @clk: Pointer to the clock. Caller must provide 'ops' field +* @dev: Pointer to the initialized device. Caller must provide +* 'release' filed * * A clock driver calls this function to register itself with the * clock device subsystem. If 'clk' points to dynamically allocated
@@ -100,7 +103,7 @@ struct posix_clock {* * Returns zero on success, non-zero otherwise. */ -int posix_clock_register (struct posix_clock *clk,dev_t devid); +int posix_clock_register (struct posix_clock *clk,struct device *dev);/** * posix_clock_unregister() - unregister a clock
kernel/time/posix-clock.c +13 -18
@@ -14,8 +14,6 @@#include "posix-timers.h" -static void delete_clock (struct kref *kref); -/* * Returns NULL if the posix_clock instance attached to 'fp' is old and stale. */
@@ -125,7 +123,7 @@ static int posix_clock_open(struct inode *inode, struct file *fp) err =0 ;if (!err) { -kref_get (&clk->kref); +get_device (clk->dev); fp->private_data = clk; } out:
@@ -141,7 +139,7 @@ static int posix_clock_release(struct inode *inode, struct file *fp)if (clk->ops.release) err = clk->ops.release (clk); -kref_put (&clk->kref, delete_clock); +put_device (clk->dev); fp->private_data =NULL ;
@@ -161,38 +159,35 @@ static const struct file_operations posix_clock_file_operations = {#endif }; -int posix_clock_register (struct posix_clock *clk,dev_t devid) +int posix_clock_register (struct posix_clock *clk,struct device *dev) {int err; -kref_init (&clk->kref);init_rwsem (&clk->rwsem);cdev_init (&clk->cdev, &posix_clock_file_operations); + err =cdev_device_add (&clk->cdev, dev); +if (err) { +pr_err ("%s unable to add device %d:%d\n" , +dev_name (dev),MAJOR (dev->devt),MINOR (dev->devt)); +return err; + } clk->cdev.owner = clk->ops.owner; - err =cdev_add (&clk->cdev, devid,1 ); + clk->dev = dev; -return err; +return 0 ; }EXPORT_SYMBOL_GPL (posix_clock_register); -static void delete_clock (struct kref *kref) -{ - struct posix_clock *clk = container_of(kref, struct posix_clock, kref); - -if (clk->release) - clk->release (clk); -} -void posix_clock_unregister (struct posix_clock *clk) { -cdev_del (&clk->cdev); +cdev_device_del (&clk->cdev, clk->dev);down_write (&clk->rwsem); clk->zombie = true;up_write (&clk->rwsem); -kref_put (&clk->kref, delete_clock); +put_device (clk->dev); }EXPORT_SYMBOL_GPL (posix_clock_unregister);
Signature · 1 line
2.20.1
Re: [PATCH v2] ptp: fix the race between the release of ptp_clock and cdev
Quoted 11 lines · Vladis Dronov
Here cdev is embedded in
posix_clockwhich is embedded inptp_clock. The race happens because ptp_clock's lifetime is controlled by two
refcounts: kref and cdev.kobj inposix_clock. This is wrong.Make ptp_clock's sysfs device a parent of cdev with
cdev_device_add()created especially for such cases. This way the parent device with itsptp_clockis not released until all references to the cdev are released. This adds a requirement that an initialized but not exposed struct device should be provided toposix_clock_register()by a caller instead of a simpledev_t.
Quoted 4 lines
This approach was adopted from the commit
72139dfa2464("watchdog: Fix the race between the release ofwatchdog_core_dataand cdev"). See details of the implementation in the commit233ed09d7fda("chardev: add helper function to register char devs with a struct device").
Thanks for digging into this!
Quoted 7 lines
/**
*posix_clock_register()- register a new clock
- * @clk: Pointer to the clock. Caller must provide 'ops' and 'release'
- * @devid: Allocated device id+ * @clk: Pointer to the clock. Caller must provide 'ops' field + * @dev: Pointer to the initialized device. Caller must provide + * 'release' filed
field
Thanks,
Richard
Re: [PATCH v2] ptp: fix the race between the release of ptp_clock and cdev
Hello, Richard,
Thank you for the review!
Quoted 4 lines
Quoted 2 lines
+ * @dev: Pointer to the initialized device. Caller must provide
+ * 'release' filedfield
Indeed. *sigh* Nothing is ideal. Let's hope a maintainer could fix it if this is approved.
Best regards,
Vladis Dronov | Red Hat, Inc. | The Core Kernel | Senior Software Engineer
----- Original Message -----
Quoted 40 lines
From: "Richard Cochran" <[email protected]>
To: "Vladis Dronov" <[email protected]>
Cc: [email protected], "Alexander Viro" <[email protected]>, "Al Viro" <[email protected]>, [email protected], [email protected]
Sent: Friday, December 27, 2019 4:02:19 PM
Subject: Re: [PATCH v2] ptp: fix the race between the release ofptp_clockand cdevQuoted 11 lines · Vladis Dronov
Here cdev is embedded in
posix_clockwhich is embedded inptp_clock. The race happens because ptp_clock's lifetime is controlled by two
refcounts: kref and cdev.kobj inposix_clock. This is wrong.Make ptp_clock's sysfs device a parent of cdev with
cdev_device_add()created especially for such cases. This way the parent device with itsptp_clockis not released until all references to the cdev are released. This adds a requirement that an initialized but not exposed struct device should be provided toposix_clock_register()by a caller instead of a simpledev_t.Quoted 4 lines
This approach was adopted from the commit
72139dfa2464("watchdog: Fix the race between the release ofwatchdog_core_dataand cdev"). See details of the implementation in the commit233ed09d7fda("chardev: add helper function to register char devs with a struct device").Thanks for digging into this!
Acked-by Richard Cochran <[email protected]>Quoted 7 lines
/**
*posix_clock_register()- register a new clock
- * @clk: Pointer to the clock. Caller must provide 'ops' and 'release'
- * @devid: Allocated device id+ * @clk: Pointer to the clock. Caller must provide 'ops' field + * @dev: Pointer to the initialized device. Caller must provide + * 'release' filedfield
Thanks,
Richard
Re: [PATCH v2] ptp: fix the race between the release of ptp_clock and cdev
In reply to Vladis DronovFrom: Vladis Dronov <[email protected]>
Date: Fri, 27 Dec 2019 03:26:27 +0100
Quoted 3 lines
In a case when a ptp chardev (like
/dev/ptp0) is open but an underlying device is removed, closing this file leads to a race. This reproduces easily in a kvm virtual machine:
. ..
Quoted 37 lines
This happens in:
static void
__fput(struct file *file)
{ ...if (file->f_op->release) file->f_op->release(inode, file); <<< cdev is kfree'd here if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && !(mode & FMODE_PATH))) { cdev_put(inode->i_cdev); <<< cdev fields are accessed hereNamely:
__fput()
posix_clock_release()
kref_put(&clk->kref, delete_clock)<<< the last reference
delete_clock()
delete_ptp_clock()
kfree(ptp) <<< cdev is embedded in ptp
cdev_putmodule_put(p->owner) <<< *p is kfree'd, bang!Here cdev is embedded in
posix_clockwhich is embedded inptp_clock. The race happens because ptp_clock's lifetime is controlled by two
refcounts: kref and cdev.kobj inposix_clock. This is wrong.Make ptp_clock's sysfs device a parent of cdev with
cdev_device_add()created especially for such cases. This way the parent device with itsptp_clockis not released until all references to the cdev are released. This adds a requirement that an initialized but not exposed struct device should be provided toposix_clock_register()by a caller instead of a simpledev_t.This approach was adopted from the commit
72139dfa2464("watchdog: Fix the race between the release ofwatchdog_core_dataand cdev"). See details of the implementation in the commit233ed09d7fda("chardev: add helper function to register char devs with a struct device").
Quoted 4 lines
Analyzed-by Stephen Johnston <[email protected]>Analyzed-by Vern Lovejoy <[email protected]>Signed-off-by Vladis Dronov <[email protected]>
Applied, thanks.