Re: [PATCH V3] fbcon -- fix race between open and removal offramebuffers

From: Bruno PrÃmont
Date: Tue May 10 2011 - 17:44:47 EST

On Tue, 10 May 2011 Tim Gardner <tim.gardner@xxxxxxxxxxxxx> wrote:
> From ca3ef33e2235c88856a6257c0be63192ab56c678 Mon Sep 17 00:00:00 2001
> From: Andy Whitcroft <apw@xxxxxxxxxxxxx>
> Date: Thu, 29 Jul 2010 16:48:20 +0100
> Subject: [PATCH] fbcon -- fix race between open and removal of framebuffers
>
> Currently there is no locking for updates to the registered_fb list.
> This allows an open through /dev/fbN to pick up a registered framebuffer
> pointer in parallel with it being released, as happens when a conflicting
> framebuffer is ejected or on module unload. There is also no reference
> counting on the framebuffer descriptor which is referenced from all open
> files, leading to references to released or reused memory to persist on
> these open files.
>
> This patch adds a reference count to the framebuffer descriptor to prevent
> it from being released until after all pending opens are closed. This
> allows the pending opens to detect the closed status and unmap themselves.
> It also adds locking to the framebuffer lookup path, locking it against
> the removal path such that it is possible to atomically lookup and take a
> reference to the descriptor. It also adds locking to the read and write
> paths which currently could access the framebuffer descriptor after it
> has been freed. Finally it moves the device to FBINFO_STATE_REMOVED to
> indicate that all access should be errored for this device.

What framebuffer drivers was this patch tested with? Just x86 with
mainstream GPU (intel/amd/nvidia KMS) in compination with vgafb/vesafb or
did it see some testing with other framebuffers like those from embedded
world?

Otherwise a much smaller (memory leaking) patch would be to just change
vesafb/vgafb to not free their fb_info after unregistration as was suggested
by Alan Cox.

> Signed-off-by: Andy Whitcroft <apw@xxxxxxxxxxxxx>
> Acked-by: Stefan Bader <stefan.bader@xxxxxxxxxxxxx>
> Signed-off-by: Leann Ogasawara <leann.ogasawara@xxxxxxxxxxxxx>
> Signed-off-by: Tim Gardner <tim.gardner@xxxxxxxxxxxxx>
> ---
> drivers/video/fbmem.c | 136 ++++++++++++++++++++++++++++++++++++++-----------
> include/linux/fb.h | 2 +
> 2 files changed, 108 insertions(+), 30 deletions(-)
>
> diff --git a/drivers/video/fbmem.c b/drivers/video/fbmem.c
> index e0c2284..1afe435 100644
> --- a/drivers/video/fbmem.c
> +++ b/drivers/video/fbmem.c
> @@ -42,6 +42,8 @@
>
> #define FBPIXMAPSIZE (1024 * 8)
>
> +/* Protects the registered framebuffer list and count. */
> +static DEFINE_SPINLOCK(registered_lock);

This only partially protects the list and count as two concurrent
framebuffer registrations do still race against each other.
For the issue addressed by this patch I don't think it makes sense to
have this spinlock at all as it's only used in get_framebuffer_info()
and in put_framebuffer_info() and put_framebuffer_info() doesn't even
look at registered_fb or num_registered_fb.
Such a spinlock makes sense in a separate patch that really protects
all access to registered_fb or num_registered_fb, be it during framebuffer
(un)registration or during access from fbcon.

I'm working on a more complete set of patches to get proper locking,
refcount (using kref) and release for struct fb_info but auditing the
various framebuffer drivers will take some time (some of the
open/release counting of drivers can probably get removed once fb_info
is ref-counted). Hopefully I can get some of it out for review on
Thursday evening so others can also look at their respective drivers.

> struct fb_info *registered_fb[FB_MAX] __read_mostly;
> int num_registered_fb __read_mostly;
>
> @@ -694,9 +696,7 @@ static ssize_t
> fb_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
> {
> unsigned long p = *ppos;
> - struct inode *inode = file->f_path.dentry->d_inode;
> - int fbidx = iminor(inode);
> - struct fb_info *info = registered_fb[fbidx];
> + struct fb_info *info = file->private_data;
> u8 *buffer, *dst;
> u8 __iomem *src;
> int c, cnt = 0, err = 0;
> @@ -705,19 +705,28 @@ fb_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
> if (!info || ! info->screen_base)
> return -ENODEV;
>
> - if (info->state != FBINFO_STATE_RUNNING)
> - return -EPERM;
> + if (!lock_fb_info(info))
> + return -ENODEV;
> +
> + if (info->state != FBINFO_STATE_RUNNING) {
> + err = -EPERM;
> + goto out_fb_info;
> + }
>
> - if (info->fbops->fb_read)
> - return info->fbops->fb_read(info, buf, count, ppos);
> + if (info->fbops->fb_read) {
> + err = info->fbops->fb_read(info, buf, count, ppos);
> + goto out_fb_info;
> + }
>
> total_size = info->screen_size;
>
> if (total_size == 0)
> total_size = info->fix.smem_len;
>
> - if (p >= total_size)
> - return 0;
> + if (p >= total_size) {
> + err = 0;
> + goto out_fb_info;
> + }
>
> if (count >= total_size)
> count = total_size;
> @@ -727,8 +736,10 @@ fb_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
>
> buffer = kmalloc((count > PAGE_SIZE) ? PAGE_SIZE : count,
> GFP_KERNEL);
> - if (!buffer)
> - return -ENOMEM;
> + if (!buffer) {
> + err = -ENOMEM;
> + goto out_fb_info;
> + }
>
> src = (u8 __iomem *) (info->screen_base + p);
>
> @@ -751,19 +762,21 @@ fb_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
> cnt += c;
> count -= c;
> }
> + if (!err)
> + err = cnt;
>
> kfree(buffer);
> +out_fb_info:
> + unlock_fb_info(info);
>
> - return (err) ? err : cnt;
> + return err;
> }
>
> static ssize_t
> fb_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
> {
> unsigned long p = *ppos;
> - struct inode *inode = file->f_path.dentry->d_inode;
> - int fbidx = iminor(inode);
> - struct fb_info *info = registered_fb[fbidx];
> + struct fb_info *info = file->private_data;
> u8 *buffer, *src;
> u8 __iomem *dst;
> int c, cnt = 0, err = 0;
> @@ -772,8 +785,13 @@ fb_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
> if (!info || !info->screen_base)
> return -ENODEV;
>
> - if (info->state != FBINFO_STATE_RUNNING)
> - return -EPERM;
> + if (!lock_fb_info(info))
> + return -ENODEV;
> +
> + if (info->state != FBINFO_STATE_RUNNING) {
> + err = -EPERM;
> + goto out_fb_info;
> + }
>
> if (info->fbops->fb_write)
> return info->fbops->fb_write(info, buf, count, ppos);
> @@ -783,8 +801,10 @@ fb_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
> if (total_size == 0)
> total_size = info->fix.smem_len;
>
> - if (p > total_size)
> - return -EFBIG;
> + if (p > total_size) {
> + err = -EFBIG;
> + goto out_fb_info;
> + }
>
> if (count > total_size) {
> err = -EFBIG;
> @@ -800,8 +820,10 @@ fb_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
>
> buffer = kmalloc((count > PAGE_SIZE) ? PAGE_SIZE : count,
> GFP_KERNEL);
> - if (!buffer)
> - return -ENOMEM;
> + if (!buffer) {
> + err = -ENOMEM;
> + goto out_fb_info;
> + }
>
> dst = (u8 __iomem *) (info->screen_base + p);
>
> @@ -825,10 +847,14 @@ fb_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
> cnt += c;
> count -= c;
> }
> + if (cnt)
> + err = cnt;
>
> kfree(buffer);
> +out_fb_info:
> + unlock_fb_info(info);
>
> - return (cnt) ? cnt : err;
> + return err;
> }
>
> int
> @@ -1303,8 +1329,7 @@ static long fb_compat_ioctl(struct file *file, unsigned int cmd,
> static int
> fb_mmap(struct file *file, struct vm_area_struct * vma)
> {
> - int fbidx = iminor(file->f_path.dentry->d_inode);
> - struct fb_info *info = registered_fb[fbidx];
> + struct fb_info * const info = file->private_data;
> struct fb_ops *fb = info->fbops;
> unsigned long off;
> unsigned long start;
> @@ -1316,6 +1341,11 @@ fb_mmap(struct file *file, struct vm_area_struct * vma)
> if (!fb)
> return -ENODEV;
> mutex_lock(&info->mm_lock);
> + if (info->state == FBINFO_STATE_REMOVED) {
> + mutex_unlock(&info->mm_lock);
> + return -ENODEV;
> + }
> +
> if (fb->fb_mmap) {
> int res;
> res = fb->fb_mmap(info, vma);
> @@ -1352,6 +1382,35 @@ fb_mmap(struct file *file, struct vm_area_struct * vma)
> return 0;
> }
>
> +static struct fb_info *get_framebuffer_info(int idx)
> +__acquires(&registered_lock)
> +__releases(&registered_lock)
> +{
> + struct fb_info *fb_info;
> +
> + spin_lock(&registered_lock);
> + fb_info = registered_fb[idx];
> + if (fb_info)
> + fb_info->ref_count++;
> + spin_unlock(&registered_lock);
> +
> + return fb_info;
> +}
> +
> +static void put_framebuffer_info(struct fb_info *fb_info)
> +__acquires(&registered_lock)
> +__releases(&registered_lock)
> +{
> + int keep;
> +
> + spin_lock(&registered_lock);
> + keep = --fb_info->ref_count;
> + spin_unlock(&registered_lock);
> +
> + if (!keep && fb_info->fbops->fb_destroy)
> + fb_info->fbops->fb_destroy(fb_info);

What happens in case fbops->fb_destroy is NULL, we just leak
the framebuffer struct? (I didn't audit frambuffer drivers yet)
Maybe calling framebuffer_release(fb_info) would be a first step.

> +}
> +
> static int
> fb_open(struct inode *inode, struct file *file)
> __acquires(&info->lock)
> @@ -1363,13 +1422,18 @@ __releases(&info->lock)
>
> if (fbidx >= FB_MAX)
> return -ENODEV;
> - info = registered_fb[fbidx];
> - if (!info)
> + info = get_framebuffer_info(fbidx);
> + if (!info) {
> request_module("fb%d", fbidx);
> - info = registered_fb[fbidx];
> + info = get_framebuffer_info(fbidx);
> + }
> if (!info)
> return -ENODEV;
> mutex_lock(&info->lock);
> + if (info->state == FBINFO_STATE_REMOVED) {
> + res = -ENODEV;
> + goto out;
> + }
> if (!try_module_get(info->fbops->owner)) {
> res = -ENODEV;
> goto out;
> @@ -1386,6 +1450,8 @@ __releases(&info->lock)
> #endif
> out:
> mutex_unlock(&info->lock);
> + if (res)
> + put_framebuffer_info(info);
> return res;
> }
>
> @@ -1401,6 +1467,7 @@ __releases(&info->lock)
> info->fbops->fb_release(info,1);
> module_put(info->fbops->owner);
> mutex_unlock(&info->lock);
> + put_framebuffer_info(info);
> return 0;
> }
>
> @@ -1549,6 +1616,7 @@ register_framebuffer(struct fb_info *fb_info)
> fb_info->node = i;
> mutex_init(&fb_info->lock);
> mutex_init(&fb_info->mm_lock);
> + fb_info->ref_count = 1;
>
> fb_info->dev = device_create(fb_class, fb_info->device,
> MKDEV(FB_MAJOR, i), NULL, "fb%d", i);
> @@ -1592,7 +1660,6 @@ register_framebuffer(struct fb_info *fb_info)
> return 0;
> }
>
> -
> /**
> * unregister_framebuffer - releases a frame buffer device
> * @fb_info: frame buffer info structure
> @@ -1627,6 +1694,16 @@ unregister_framebuffer(struct fb_info *fb_info)
> return -ENODEV;
> event.info = fb_info;
> ret = fb_notifier_call_chain(FB_EVENT_FB_UNBIND, &event);
> + if (!ret) {
> + mutex_lock(&fb_info->mm_lock);
> + /*
> + * We must prevent any operations for this transition, we
> + * already have info->lock so grab the info->mm_lock to hold
> + * the remainder.
> + */
> + fb_info->state = FBINFO_STATE_REMOVED;
> + mutex_unlock(&fb_info->mm_lock);
> + }
> unlock_fb_info(fb_info);
>
> if (ret) {
> @@ -1646,8 +1723,7 @@ unregister_framebuffer(struct fb_info *fb_info)
> fb_notifier_call_chain(FB_EVENT_FB_UNREGISTERED, &event);
>
> /* this may free fb info */
> - if (fb_info->fbops->fb_destroy)
> - fb_info->fbops->fb_destroy(fb_info);
> + put_framebuffer_info(fb_info);
> done:
> return ret;
> }
> diff --git a/include/linux/fb.h b/include/linux/fb.h
> index df728c1..60de3fa 100644
> --- a/include/linux/fb.h
> +++ b/include/linux/fb.h
> @@ -834,6 +834,7 @@ struct fb_tile_ops {
> struct fb_info {
> int node;
> int flags;
> + int ref_count;
> struct mutex lock; /* Lock for open/release/ioctl funcs */
> struct mutex mm_lock; /* Lock for fb_mmap and smem_* fields */
> struct fb_var_screeninfo var; /* Current var */
> @@ -873,6 +874,7 @@ struct fb_info {
> void *pseudo_palette; /* Fake palette of 16 colors */
> #define FBINFO_STATE_RUNNING 0
> #define FBINFO_STATE_SUSPENDED 1
> +#define FBINFO_STATE_REMOVED 2
> u32 state; /* Hardware state i.e suspend */
> void *fbcon_par; /* fbcon use-only private area */
> /* From here on everything is device dependent */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/