]> git.karo-electronics.de Git - linux-beck.git/commitdiff
kernfs: Move faulting copy_user operations outside of the mutex
authorChris Wilson <chris@chris-wilson.co.uk>
Thu, 31 Mar 2016 10:45:06 +0000 (11:45 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 30 Apr 2016 17:05:05 +0000 (10:05 -0700)
A fault in a user provided buffer may lead anywhere, and lockdep warns
that we have a potential deadlock between the mm->mmap_sem and the
kernfs file mutex:

[   82.811702] ======================================================
[   82.811705] [ INFO: possible circular locking dependency detected ]
[   82.811709] 4.5.0-rc4-gfxbench+ #1 Not tainted
[   82.811711] -------------------------------------------------------
[   82.811714] kms_setmode/5859 is trying to acquire lock:
[   82.811717]  (&dev->struct_mutex){+.+.+.}, at: [<ffffffff8150d9c1>] drm_gem_mmap+0x1a1/0x270
[   82.811731]
but task is already holding lock:
[   82.811734]  (&mm->mmap_sem){++++++}, at: [<ffffffff8117b364>] vm_mmap_pgoff+0x44/0xa0
[   82.811745]
which lock already depends on the new lock.

[   82.811749]
the existing dependency chain (in reverse order) is:
[   82.811752]
-> #3 (&mm->mmap_sem){++++++}:
[   82.811761]        [<ffffffff810cc883>] lock_acquire+0xc3/0x1d0
[   82.811766]        [<ffffffff8118bc65>] __might_fault+0x75/0xa0
[   82.811771]        [<ffffffff8124da4a>] kernfs_fop_write+0x8a/0x180
[   82.811787]        [<ffffffff811d1023>] __vfs_write+0x23/0xe0
[   82.811792]        [<ffffffff811d1d74>] vfs_write+0xa4/0x190
[   82.811797]        [<ffffffff811d2c14>] SyS_write+0x44/0xb0
[   82.811801]        [<ffffffff817bb81b>] entry_SYSCALL_64_fastpath+0x16/0x73
[   82.811807]
-> #2 (s_active#6){++++.+}:
[   82.811814]        [<ffffffff810cc883>] lock_acquire+0xc3/0x1d0
[   82.811819]        [<ffffffff8124c070>] __kernfs_remove+0x210/0x2f0
[   82.811823]        [<ffffffff8124d040>] kernfs_remove_by_name_ns+0x40/0xa0
[   82.811828]        [<ffffffff8124e9e0>] sysfs_remove_file_ns+0x10/0x20
[   82.811832]        [<ffffffff815318d4>] device_del+0x124/0x250
[   82.811837]        [<ffffffff81531a19>] device_unregister+0x19/0x60
[   82.811841]        [<ffffffff8153c051>] cpu_cache_sysfs_exit+0x51/0xb0
[   82.811846]        [<ffffffff8153c628>] cacheinfo_cpu_callback+0x38/0x70
[   82.811851]        [<ffffffff8109ae89>] notifier_call_chain+0x39/0xa0
[   82.811856]        [<ffffffff8109aef9>] __raw_notifier_call_chain+0x9/0x10
[   82.811860]        [<ffffffff810786de>] cpu_notify+0x1e/0x40
[   82.811865]        [<ffffffff81078779>] cpu_notify_nofail+0x9/0x20
[   82.811869]        [<ffffffff81078ac3>] _cpu_down+0x233/0x340
[   82.811874]        [<ffffffff81079019>] disable_nonboot_cpus+0xc9/0x350
[   82.811878]        [<ffffffff810d2e11>] suspend_devices_and_enter+0x5a1/0xb50
[   82.811883]        [<ffffffff810d3903>] pm_suspend+0x543/0x8d0
[   82.811888]        [<ffffffff810d1b77>] state_store+0x77/0xe0
[   82.811892]        [<ffffffff813fa68f>] kobj_attr_store+0xf/0x20
[   82.811897]        [<ffffffff8124e740>] sysfs_kf_write+0x40/0x50
[   82.811902]        [<ffffffff8124dafc>] kernfs_fop_write+0x13c/0x180
[   82.811906]        [<ffffffff811d1023>] __vfs_write+0x23/0xe0
[   82.811910]        [<ffffffff811d1d74>] vfs_write+0xa4/0x190
[   82.811914]        [<ffffffff811d2c14>] SyS_write+0x44/0xb0
[   82.811918]        [<ffffffff817bb81b>] entry_SYSCALL_64_fastpath+0x16/0x73
[   82.811923]
-> #1 (cpu_hotplug.lock){+.+.+.}:
[   82.811929]        [<ffffffff810cc883>] lock_acquire+0xc3/0x1d0
[   82.811933]        [<ffffffff817b6f72>] mutex_lock_nested+0x62/0x3b0
[   82.811940]        [<ffffffff810784c1>] get_online_cpus+0x61/0x80
[   82.811944]        [<ffffffff811170eb>] stop_machine+0x1b/0xe0
[   82.811949]        [<ffffffffa0178edd>] gen8_ggtt_insert_entries__BKL+0x2d/0x30 [i915]
[   82.812009]        [<ffffffffa017d3a6>] ggtt_bind_vma+0x46/0x70 [i915]
[   82.812045]        [<ffffffffa017eb70>] i915_vma_bind+0x140/0x290 [i915]
[   82.812081]        [<ffffffffa01862b9>] i915_gem_object_do_pin+0x899/0xb00 [i915]
[   82.812117]        [<ffffffffa0186555>] i915_gem_object_pin+0x35/0x40 [i915]
[   82.812154]        [<ffffffffa019a23e>] intel_init_pipe_control+0xbe/0x210 [i915]
[   82.812192]        [<ffffffffa0197312>] intel_logical_rings_init+0xe2/0xde0 [i915]
[   82.812232]        [<ffffffffa0186fe3>] i915_gem_init+0xf3/0x130 [i915]
[   82.812278]        [<ffffffffa02097ed>] i915_driver_load+0xf2d/0x1770 [i915]
[   82.812318]        [<ffffffff81512474>] drm_dev_register+0xa4/0xb0
[   82.812323]        [<ffffffff8151467e>] drm_get_pci_dev+0xce/0x1e0
[   82.812328]        [<ffffffffa01472cf>] i915_pci_probe+0x2f/0x50 [i915]
[   82.812360]        [<ffffffff8143f907>] pci_device_probe+0x87/0xf0
[   82.812366]        [<ffffffff81535f89>] driver_probe_device+0x229/0x450
[   82.812371]        [<ffffffff81536233>] __driver_attach+0x83/0x90
[   82.812375]        [<ffffffff81533c61>] bus_for_each_dev+0x61/0xa0
[   82.812380]        [<ffffffff81535879>] driver_attach+0x19/0x20
[   82.812384]        [<ffffffff8153535f>] bus_add_driver+0x1ef/0x290
[   82.812388]        [<ffffffff81536e9b>] driver_register+0x5b/0xe0
[   82.812393]        [<ffffffff8143e83b>] __pci_register_driver+0x5b/0x60
[   82.812398]        [<ffffffff81514866>] drm_pci_init+0xd6/0x100
[   82.812402]        [<ffffffffa027c094>] 0xffffffffa027c094
[   82.812406]        [<ffffffff810003de>] do_one_initcall+0xae/0x1d0
[   82.812412]        [<ffffffff811595a0>] do_init_module+0x5b/0x1cb
[   82.812417]        [<ffffffff81106160>] load_module+0x1c20/0x2480
[   82.812422]        [<ffffffff81106bae>] SyS_finit_module+0x7e/0xa0
[   82.812428]        [<ffffffff817bb81b>] entry_SYSCALL_64_fastpath+0x16/0x73
[   82.812433]
-> #0 (&dev->struct_mutex){+.+.+.}:
[   82.812439]        [<ffffffff810cbe59>] __lock_acquire+0x1fc9/0x20f0
[   82.812443]        [<ffffffff810cc883>] lock_acquire+0xc3/0x1d0
[   82.812456]        [<ffffffff8150d9e7>] drm_gem_mmap+0x1c7/0x270
[   82.812460]        [<ffffffff81196a14>] mmap_region+0x334/0x580
[   82.812466]        [<ffffffff81196fc4>] do_mmap+0x364/0x410
[   82.812470]        [<ffffffff8117b38d>] vm_mmap_pgoff+0x6d/0xa0
[   82.812474]        [<ffffffff811950f4>] SyS_mmap_pgoff+0x184/0x220
[   82.812479]        [<ffffffff8100a0fd>] SyS_mmap+0x1d/0x20
[   82.812484]        [<ffffffff817bb81b>] entry_SYSCALL_64_fastpath+0x16/0x73
[   82.812489]
other info that might help us debug this:

[   82.812493] Chain exists of:
  &dev->struct_mutex --> s_active#6 --> &mm->mmap_sem

[   82.812502]  Possible unsafe locking scenario:

[   82.812506]        CPU0                    CPU1
[   82.812508]        ----                    ----
[   82.812510]   lock(&mm->mmap_sem);
[   82.812514]                                lock(s_active#6);
[   82.812519]                                lock(&mm->mmap_sem);
[   82.812522]   lock(&dev->struct_mutex);
[   82.812526]
 *** DEADLOCK ***

[   82.812531] 1 lock held by kms_setmode/5859:
[   82.812533]  #0:  (&mm->mmap_sem){++++++}, at: [<ffffffff8117b364>] vm_mmap_pgoff+0x44/0xa0
[   82.812541]
stack backtrace:
[   82.812547] CPU: 0 PID: 5859 Comm: kms_setmode Not tainted 4.5.0-rc4-gfxbench+ #1
[   82.812550] Hardware name:                  /NUC5CPYB, BIOS PYBSWCEL.86A.0040.2015.0814.1353 08/14/2015
[   82.812553]  0000000000000000 ffff880079407bf0 ffffffff813f8505 ffffffff825fb270
[   82.812560]  ffffffff825c4190 ffff880079407c30 ffffffff810c84ac ffff880079407c90
[   82.812566]  ffff8800797ed328 ffff8800797ecb00 0000000000000001 ffff8800797ed350
[   82.812573] Call Trace:
[   82.812578]  [<ffffffff813f8505>] dump_stack+0x67/0x92
[   82.812582]  [<ffffffff810c84ac>] print_circular_bug+0x1fc/0x310
[   82.812586]  [<ffffffff810cbe59>] __lock_acquire+0x1fc9/0x20f0
[   82.812590]  [<ffffffff810cc883>] lock_acquire+0xc3/0x1d0
[   82.812594]  [<ffffffff8150d9c1>] ? drm_gem_mmap+0x1a1/0x270
[   82.812599]  [<ffffffff8150d9e7>] drm_gem_mmap+0x1c7/0x270
[   82.812603]  [<ffffffff8150d9c1>] ? drm_gem_mmap+0x1a1/0x270
[   82.812608]  [<ffffffff81196a14>] mmap_region+0x334/0x580
[   82.812612]  [<ffffffff81196fc4>] do_mmap+0x364/0x410
[   82.812616]  [<ffffffff8117b38d>] vm_mmap_pgoff+0x6d/0xa0
[   82.812629]  [<ffffffff811950f4>] SyS_mmap_pgoff+0x184/0x220
[   82.812633]  [<ffffffff8100a0fd>] SyS_mmap+0x1d/0x20
[   82.812637]  [<ffffffff817bb81b>] entry_SYSCALL_64_fastpath+0x16/0x73

Highly unlikely though this scenario is, we can avoid the issue entirely
by moving the copy operation from out under the kernfs_get_active()
tracking by assigning the preallocated buffer its own mutex. The
temporary buffer allocation doesn't require mutex locking as it is
entirely local.

The locked section was extended by the addition of the preallocated buf
to speed up md user operations in

commit 2b75869bba676c248d8d25ae6d2bd9221dfffdb6
Author: NeilBrown <neilb@suse.de>
Date:   Mon Oct 13 16:41:28 2014 +1100

    sysfs/kernfs: allow attributes to request write buffer be pre-allocated.

Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94350
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: NeilBrown <neilb@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/kernfs/file.c
include/linux/kernfs.h

index 7247252ee9b1beffa3c624cb54db676a707fffe2..e1574008adc9e4607e9deb2c1898f5d25be528a9 100644 (file)
@@ -190,15 +190,16 @@ static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of,
        char *buf;
 
        buf = of->prealloc_buf;
-       if (!buf)
+       if (buf)
+               mutex_lock(&of->prealloc_mutex);
+       else
                buf = kmalloc(len, GFP_KERNEL);
        if (!buf)
                return -ENOMEM;
 
        /*
         * @of->mutex nests outside active ref and is used both to ensure that
-        * the ops aren't called concurrently for the same open file, and
-        * to provide exclusive access to ->prealloc_buf (when that exists).
+        * the ops aren't called concurrently for the same open file.
         */
        mutex_lock(&of->mutex);
        if (!kernfs_get_active(of->kn)) {
@@ -214,21 +215,23 @@ static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of,
        else
                len = -EINVAL;
 
+       kernfs_put_active(of->kn);
+       mutex_unlock(&of->mutex);
+
        if (len < 0)
-               goto out_unlock;
+               goto out_free;
 
        if (copy_to_user(user_buf, buf, len)) {
                len = -EFAULT;
-               goto out_unlock;
+               goto out_free;
        }
 
        *ppos += len;
 
- out_unlock:
-       kernfs_put_active(of->kn);
-       mutex_unlock(&of->mutex);
  out_free:
-       if (buf != of->prealloc_buf)
+       if (buf == of->prealloc_buf)
+               mutex_unlock(&of->prealloc_mutex);
+       else
                kfree(buf);
        return len;
 }
@@ -284,15 +287,22 @@ static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf,
        }
 
        buf = of->prealloc_buf;
-       if (!buf)
+       if (buf)
+               mutex_lock(&of->prealloc_mutex);
+       else
                buf = kmalloc(len + 1, GFP_KERNEL);
        if (!buf)
                return -ENOMEM;
 
+       if (copy_from_user(buf, user_buf, len)) {
+               len = -EFAULT;
+               goto out_free;
+       }
+       buf[len] = '\0';        /* guarantee string termination */
+
        /*
         * @of->mutex nests outside active ref and is used both to ensure that
-        * the ops aren't called concurrently for the same open file, and
-        * to provide exclusive access to ->prealloc_buf (when that exists).
+        * the ops aren't called concurrently for the same open file.
         */
        mutex_lock(&of->mutex);
        if (!kernfs_get_active(of->kn)) {
@@ -301,26 +311,22 @@ static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf,
                goto out_free;
        }
 
-       if (copy_from_user(buf, user_buf, len)) {
-               len = -EFAULT;
-               goto out_unlock;
-       }
-       buf[len] = '\0';        /* guarantee string termination */
-
        ops = kernfs_ops(of->kn);
        if (ops->write)
                len = ops->write(of, buf, len, *ppos);
        else
                len = -EINVAL;
 
+       kernfs_put_active(of->kn);
+       mutex_unlock(&of->mutex);
+
        if (len > 0)
                *ppos += len;
 
-out_unlock:
-       kernfs_put_active(of->kn);
-       mutex_unlock(&of->mutex);
 out_free:
-       if (buf != of->prealloc_buf)
+       if (buf == of->prealloc_buf)
+               mutex_unlock(&of->prealloc_mutex);
+       else
                kfree(buf);
        return len;
 }
@@ -687,6 +693,7 @@ static int kernfs_fop_open(struct inode *inode, struct file *file)
                error = -ENOMEM;
                if (!of->prealloc_buf)
                        goto err_free;
+               mutex_init(&of->prealloc_mutex);
        }
 
        /*
index c06c44242f3993ae05a8454f844425a0bd14d525..d306e282bb1d1b30cc94aec802a2af4bbd076464 100644 (file)
@@ -177,6 +177,7 @@ struct kernfs_open_file {
 
        /* private fields, do not use outside kernfs proper */
        struct mutex            mutex;
+       struct mutex            prealloc_mutex;
        int                     event;
        struct list_head        list;
        char                    *prealloc_buf;