[Intel-gfx] [PATCH] drm/i915: Prefault before locking pages in shmem_pwrite

[Chris Wilson]

If the user passes in a pointer to a GGTT mmaping of the same buffer
being written to, we can hit a deadlock in acquiring the shmemfs page
(once as the write destination and then as the read source).

[<0>] io_schedule+0xd/0x30
[<0>] __lock_page+0x105/0x1b0
[<0>] find_lock_entry+0x55/0x90
[<0>] shmem_getpage_gfp+0xbb/0x800
[<0>] shmem_read_mapping_page_gfp+0x2d/0x50
[<0>] shmem_get_pages+0x158/0x5d0 [i915]
[<0>] ____i915_gem_object_get_pages+0x17/0x90 [i915]
[<0>] __i915_gem_object_get_pages+0x57/0x70 [i915]
[<0>] i915_gem_fault+0x1b4/0x5c0 [i915]
[<0>] __do_fault+0x2d/0x80
[<0>] __handle_mm_fault+0xad4/0xfb0
[<0>] handle_mm_fault+0xe6/0x1f0
[<0>] __do_page_fault+0x18f/0x3f0
[<0>] page_fault+0x1b/0x20
[<0>] copy_user_enhanced_fast_string+0x7/0x10
[<0>] _copy_from_user+0x37/0x60
[<0>] shmem_pwrite+0xf0/0x160 [i915]
[<0>] i915_gem_pwrite_ioctl+0x14e/0x520 [i915]
[<0>] drm_ioctl_kernel+0x81/0xd0
[<0>] drm_ioctl+0x1a7/0x310
[<0>] do_vfs_ioctl+0x88/0x5d0
[<0>] ksys_ioctl+0x35/0x70
[<0>] __x64_sys_ioctl+0x11/0x20
[<0>] do_syscall_64+0x39/0xe0
[<0>] entry_SYSCALL_64_after_hwframe+0x44/0xa9

We can reduce (but not eliminate!) the chance of this happening by
faulting the user_data before we take the page lock in
pagecache_write_begin(). One way to eliminate the potential recursion
here is by disabling pagefaults for the copy, and handling the fallback
to use an alternative method -- so convert to use kmap_atomic (which
should disable preemption and pagefaulting for the copy) and report
ENODEV instead of EFAULT so that our caller tries again with a different
copy mechanism -- we already check that the page should have been
faultable so a false negative should be rare.

Testcase: igt/gem_pwrite/self
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Cc: Matthew Auld <matthew.william.auld at gmail.com>
---
 drivers/gpu/drm/i915/i915_gem.c | 30 ++++++++++++++++++++++++------
 1 file changed, 24 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index c3b4ec52e1b7..bf594a5e88bc 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2788,7 +2788,11 @@ i915_gem_object_pwrite_gtt(struct drm_i915_gem_object *obj,
 	u64 remain, offset;
 	unsigned int pg;
 
-	/* Before we instantiate/pin the backing store for our use, we
+	/* Caller already validated user args */
+	GEM_BUG_ON(!access_ok(user_data, arg->size));
+
+	/*
+	 * Before we instantiate/pin the backing store for our use, we
 	 * can prepopulate the shmemfs filp efficiently using a write into
 	 * the pagecache. We avoid the penalty of instantiating all the
 	 * pages, important if the user is just writing to a few and never
@@ -2802,7 +2806,8 @@ i915_gem_object_pwrite_gtt(struct drm_i915_gem_object *obj,
 	if (obj->mm.madv != I915_MADV_WILLNEED)
 		return -EFAULT;
 
-	/* Before the pages are instantiated the object is treated as being
+	/*
+	 * Before the pages are instantiated the object is treated as being
 	 * in the CPU domain. The pages will be clflushed as required before
 	 * use, and we can freely write into the pages directly. If userspace
 	 * races pwrite with any other operation; corruption will ensue -
@@ -2818,20 +2823,32 @@ i915_gem_object_pwrite_gtt(struct drm_i915_gem_object *obj,
 		struct page *page;
 		void *data, *vaddr;
 		int err;
+		char c;
 
 		len = PAGE_SIZE - pg;
 		if (len > remain)
 			len = remain;
 
+		/* Prefault the user page to reduce potential recursion */
+		err = __get_user(c, user_data);
+		if (err)
+			return err;
+
+		err = __get_user(c, user_data + len - 1);
+		if (err)
+			return err;
+
 		err = pagecache_write_begin(obj->base.filp, mapping,
 					    offset, len, 0,
 					    &page, &data);
 		if (err < 0)
 			return err;
 
-		vaddr = kmap(page);
-		unwritten = copy_from_user(vaddr + pg, user_data, len);
-		kunmap(page);
+		vaddr = kmap_atomic(page);
+		unwritten = __copy_from_user_inatomic(vaddr + pg,
+						      user_data,
+						      len);
+		kunmap_atomic(vaddr);
 
 		err = pagecache_write_end(obj->base.filp, mapping,
 					  offset, len, len - unwritten,
@@ -2839,8 +2856,9 @@ i915_gem_object_pwrite_gtt(struct drm_i915_gem_object *obj,
 		if (err < 0)
 			return err;
 
+		/* We don't handle -EFAULT, leave it to the caller to check */
 		if (unwritten)
-			return -EFAULT;
+			return -ENODEV;
 
 		remain -= len;
 		user_data += len;
-- 
2.20.1