Re: [PATCH] drm: restore is_master upon failure in drm_new_set_master()

From: Daniel Vetter
Date: Wed Nov 21 2018 - 04:55:20 EST


On Sun, Nov 18, 2018 at 08:57:20PM -0300, Sergio Correia wrote:
> When drm_new_set_master() fails, we restore the old master, however we may
> have changed the is_master flag to 1, before failing, and it may be the
> case it was 0 previously. Restore also this flag to its original state, in
> case of failure.
>
> Here is a problematic flow: we check is_master in drm_is_current_master(),
> then proceed to call drm_lease_owner() passing master. If we do not restore
> is_master status when drm_new_set_master() fails, we may have a situation
> in which is_master will be 1 and master itself, NULL, leading to the deref
> of a NULL pointer in drm_lease_owner().
>
> This fixes the following OOPS, observed on an ArchLinux running a 4.19.2
> kernel:
>
> [ 97.804282] BUG: unable to handle kernel NULL pointer dereference at 0000000000000080
> [ 97.807224] PGD 0 P4D 0
> [ 97.807224] Oops: 0000 [#1] PREEMPT SMP NOPTI
> [ 97.807224] CPU: 0 PID: 1348 Comm: xfwm4 Tainted: P OE 4.19.2-arch1-1-ARCH #1
> [ 97.807224] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./AB350 Pro4, BIOS P5.10 10/16/2018
> [ 97.807224] RIP: 0010:drm_lease_owner+0xd/0x20 [drm]
> [ 97.807224] Code: 83 c4 18 5b 5d c3 b8 ea ff ff ff eb e2 b8 ed ff ff ff eb db e8 b4 ca 68 fb 0f 1f 40 00 0f 1f 44 00 00 48 89 f8 eb 03 48 89 d0 <48> 8b 90 80 00 00 00 48 85 d2 75 f1 c3 66 0f 1f 44 00 00 0f 1f 44
> [ 97.807224] RSP: 0018:ffffb8cf08e07bb0 EFLAGS: 00010202
> [ 97.807224] RAX: 0000000000000000 RBX: ffff9cf0f2586c00 RCX: ffff9cf0f2586c88
> [ 97.807224] RDX: ffff9cf0ddbd8000 RSI: 0000000000000000 RDI: 0000000000000000
> [ 97.807224] RBP: ffff9cf1040e9800 R08: 0000000000000000 R09: 0000000000000000
> [ 97.807224] R10: ffffdeb30fd5d680 R11: ffffdeb30f5d6808 R12: ffff9cf1040e9888
> [ 97.807224] R13: 0000000000000000 R14: dead000000000200 R15: ffff9cf0f2586cc8
> [ 97.807224] FS: 00007f4145513180(0000) GS:ffff9cf10ea00000(0000) knlGS:0000000000000000
> [ 97.807224] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 97.807224] CR2: 0000000000000080 CR3: 00000003d7548000 CR4: 00000000003406f0
> [ 97.807224] Call Trace:
> [ 97.807224] drm_is_current_master+0x1a/0x30 [drm]
> [ 97.807224] drm_master_release+0x3e/0x130 [drm]
> [ 97.807224] drm_file_free.part.0+0x2be/0x2d0 [drm]
> [ 97.807224] drm_open+0x1ba/0x1e0 [drm]
> [ 97.807224] drm_stub_open+0xaf/0xe0 [drm]
> [ 97.807224] chrdev_open+0xa3/0x1b0
> [ 97.807224] ? cdev_put.part.0+0x20/0x20
> [ 97.807224] do_dentry_open+0x132/0x340
> [ 97.807224] path_openat+0x2d1/0x14e0
> [ 97.807224] ? mem_cgroup_commit_charge+0x7a/0x520
> [ 97.807224] do_filp_open+0x93/0x100
> [ 97.807224] ? __check_object_size+0x102/0x189
> [ 97.807224] ? _raw_spin_unlock+0x16/0x30
> [ 97.807224] do_sys_open+0x186/0x210
> [ 97.807224] do_syscall_64+0x5b/0x170
> [ 97.807224] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> [ 97.807224] RIP: 0033:0x7f4147b07976
> [ 97.807224] Code: 89 54 24 08 e8 7b f4 ff ff 8b 74 24 0c 48 8b 3c 24 41 89 c0 44 8b 54 24 08 b8 01 01 00 00 89 f2 48 89 fe bf 9c ff ff ff 0f 05 <48> 3d 00 f0 ff ff 77 30 44 89 c7 89 44 24 08 e8 a6 f4 ff ff 8b 44
> [ 97.807224] RSP: 002b:00007ffcced96ca0 EFLAGS: 00000293 ORIG_RAX: 0000000000000101
> [ 97.807224] RAX: ffffffffffffffda RBX: 00005619d5037f80 RCX: 00007f4147b07976
> [ 97.807224] RDX: 0000000000000002 RSI: 00005619d46b969c RDI: 00000000ffffff9c
> [ 98.040039] RBP: 0000000000000024 R08: 0000000000000000 R09: 0000000000000000
> [ 98.040039] R10: 0000000000000000 R11: 0000000000000293 R12: 0000000000000024
> [ 98.040039] R13: 0000000000000012 R14: 00005619d5035950 R15: 0000000000000012
> [ 98.040039] Modules linked in: nct6775 hwmon_vid algif_skcipher af_alg nls_iso8859_1 nls_cp437 vfat fat uvcvideo videobuf2_vmalloc videobuf2_memops videobuf2_v4l2 videobuf2_common arc4 videodev media snd_usb_audio snd_hda_codec_hdmi snd_usbmidi_lib snd_rawmidi snd_seq_device mousedev input_leds iwlmvm mac80211 snd_hda_codec_realtek snd_hda_codec_generic snd_hda_intel snd_hda_codec edac_mce_amd kvm_amd snd_hda_core kvm iwlwifi snd_hwdep r8169 wmi_bmof cfg80211 snd_pcm irqbypass snd_timer snd libphy soundcore pinctrl_amd rfkill pcspkr sp5100_tco evdev gpio_amdpt k10temp mac_hid i2c_piix4 wmi pcc_cpufreq acpi_cpufreq vboxnetflt(OE) vboxnetadp(OE) vboxpci(OE) vboxdrv(OE) msr sg crypto_user ip_tables x_tables ext4 crc32c_generic crc16 mbcache jbd2 fscrypto uas usb_storage dm_crypt hid_generic usbhid hid
> [ 98.040039] dm_mod raid1 md_mod sd_mod crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel pcbc ahci libahci aesni_intel aes_x86_64 libata crypto_simd cryptd glue_helper ccp xhci_pci rng_core scsi_mod xhci_hcd nvidia_drm(POE) drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops drm agpgart nvidia_uvm(POE) nvidia_modeset(POE) nvidia(POE) ipmi_devintf ipmi_msghandler
> [ 98.040039] CR2: 0000000000000080
> [ 98.040039] ---[ end trace 3b65093b6fe62b2f ]---
> [ 98.040039] RIP: 0010:drm_lease_owner+0xd/0x20 [drm]
> [ 98.040039] Code: 83 c4 18 5b 5d c3 b8 ea ff ff ff eb e2 b8 ed ff ff ff eb db e8 b4 ca 68 fb 0f 1f 40 00 0f 1f 44 00 00 48 89 f8 eb 03 48 89 d0 <48> 8b 90 80 00 00 00 48 85 d2 75 f1 c3 66 0f 1f 44 00 00 0f 1f 44
> [ 98.040039] RSP: 0018:ffffb8cf08e07bb0 EFLAGS: 00010202
> [ 98.040039] RAX: 0000000000000000 RBX: ffff9cf0f2586c00 RCX: ffff9cf0f2586c88
> [ 98.040039] RDX: ffff9cf0ddbd8000 RSI: 0000000000000000 RDI: 0000000000000000
> [ 98.040039] RBP: ffff9cf1040e9800 R08: 0000000000000000 R09: 0000000000000000
> [ 98.040039] R10: ffffdeb30fd5d680 R11: ffffdeb30f5d6808 R12: ffff9cf1040e9888
> [ 98.040039] R13: 0000000000000000 R14: dead000000000200 R15: ffff9cf0f2586cc8
> [ 98.040039] FS: 00007f4145513180(0000) GS:ffff9cf10ea00000(0000) knlGS:0000000000000000
> [ 98.040039] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 98.040039] CR2: 0000000000000080 CR3: 00000003d7548000 CR4: 00000000003406f0
>
> Signed-off-by: Sergio Correia <sergio@xxxxxxxxxx>

Nice catch! I guess this is with the vmwgfx driver running? Only that has
a master_set function, and I don't see anything else that could fail.

> ---
> drivers/gpu/drm/drm_auth.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/gpu/drm/drm_auth.c b/drivers/gpu/drm/drm_auth.c
> index d9c0f7573905..959a1ba5a3c6 100644
> --- a/drivers/gpu/drm/drm_auth.c
> +++ b/drivers/gpu/drm/drm_auth.c
> @@ -138,11 +138,13 @@ static int drm_set_master(struct drm_device *dev, struct drm_file *fpriv,
> static int drm_new_set_master(struct drm_device *dev, struct drm_file *fpriv)
> {
> struct drm_master *old_master;
> + int old_master_status;
> int ret;
>
> lockdep_assert_held_once(&dev->master_mutex);
>
> old_master = fpriv->master;
> + old_master_status = fpriv->is_master;

We have the invariant here that fpriv->is_master == 0, so no need to
restore anything, you can just unconditionally set it to 0 at the end. If
you want, put a WARN_ON(fpriv->is_master) here.

Can you pls respin? I'll apply it right away then.

Thanks, Daniel

> fpriv->master = drm_master_create(dev);
> if (!fpriv->master) {
> fpriv->master = old_master;
> @@ -170,6 +172,7 @@ static int drm_new_set_master(struct drm_device *dev, struct drm_file *fpriv)
> /* drop references and restore old master on failure */
> drm_master_put(&fpriv->master);
> fpriv->master = old_master;
> + fpriv->is_master = old_master_status;
>
> return ret;
> }
> --
> 2.19.1
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch