Skip to contentSkip to conversation

mm, dax: fix DAX deadlocks (COW fault)

PATCH In review Yigal Korman 4 participants 4 messages Nov 16, 2015 – Nov 17, 2015

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm, dax: fix DAX deadlocks (COW fault)

DAX handling of COW faults has wrong locking sequence:
dax_fault does i_mmap_lock_read do_cow_fault does i_mmap_unlock_write

Ross's commit[1] missed a fix[2] that Kirill added to Matthew's commit[3].

Original COW locking logic was introduced by Matthew here[4].

This should be applied to v4.3 as well.

[1] 0f90cc6609c7 mm, dax: fix DAX deadlocks
[2] 52a2b53ffde6 mm, dax: use i_mmap_unlock_write() in do_cow_fault()
[3] 843172978bb9 dax: fix race between simultaneous faults
[4] 2e4cdab0584f mm: allow page fault handlers to perform the COW

Cc Stable Tree <[email protected]>
Cc Boaz Harrosh <[email protected]>
Cc Ross Zwisler <[email protected]>
Cc Alexander Viro <[email protected]>
Cc Dan Williams <[email protected]>
Cc Dave Chinner <[email protected]>
Cc Jan Kara <[email protected]>
Cc "Kirill A. Shutemov" <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 mm/memory.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
mm/memory.c +4 -4
index c716913..e5071af 100644
@@ -3015,9 +3015,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 		} else {
 			/*
 			 * The fault handler has no page to lock, so it holds
-			 * i_mmap_lock for write to protect against truncate.
+			 * i_mmap_lock for read to protect against truncate.
 			 */
-			i_mmap_unlock_write(vma->vm_file->f_mapping);
+			i_mmap_unlock_read(vma->vm_file->f_mapping);
 		}
 		goto uncharge_out;
 	}
@@ -3031,9 +3031,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 	} else {
 		/*
 		 * The fault handler has no page to lock, so it holds
-		 * i_mmap_lock for write to protect against truncate.
+		 * i_mmap_lock for read to protect against truncate.
 		 */
-		i_mmap_unlock_write(vma->vm_file->f_mapping);
+		i_mmap_unlock_read(vma->vm_file->f_mapping);
 	}
 	return ret;
 uncharge_out:
Signature · 7 lines
1.9.3

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#1 lore raw
Quoted 49 lines · Yigal Korman

DAX handling of COW faults has wrong locking sequence:
dax_fault does i_mmap_lock_read do_cow_fault does i_mmap_unlock_write

Ross's commit[1] missed a fix[2] that Kirill added to Matthew's commit[3].

Original COW locking logic was introduced by Matthew here[4].

This should be applied to v4.3 as well.

[1] 0f90cc6609c7 mm, dax: fix DAX deadlocks
[2] 52a2b53ffde6 mm, dax: use i_mmap_unlock_write() in do_cow_fault()
[3] 843172978bb9 dax: fix race between simultaneous faults
[4] 2e4cdab0584f mm: allow page fault handlers to perform the COW

Cc Stable Tree <[email protected]>
Cc Boaz Harrosh <[email protected]>
Cc Ross Zwisler <[email protected]>
Cc Alexander Viro <[email protected]>
Cc Dan Williams <[email protected]>
Cc Dave Chinner <[email protected]>
Cc Jan Kara <[email protected]>
Cc "Kirill A. Shutemov" <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 mm/memory.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
mm/memory.c +2 -2
index c716913..e5071af 100644
@@ -3015,9 +3015,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
                } else {
                        /*
                         * The fault handler has no page to lock, so it holds
-                        * i_mmap_lock for write to protect against truncate.
+                        * i_mmap_lock for read to protect against truncate.
                         */
-                       i_mmap_unlock_write(vma->vm_file->f_mapping);
+                       i_mmap_unlock_read(vma->vm_file->f_mapping);
                }
                goto uncharge_out;
        }
@@ -3031,9 +3031,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        } else {

mm/memory.c static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, · line 3039

Quoted 10 lines
mm/memory.c +2 -2
                /*
                 * The fault handler has no page to lock, so it holds
-                * i_mmap_lock for write to protect against truncate.
+                * i_mmap_lock for read to protect against truncate.
                 */
-               i_mmap_unlock_write(vma->vm_file->f_mapping);
+               i_mmap_unlock_read(vma->vm_file->f_mapping);
        }
        return ret;
 uncharge_out:

Looks good to me. I'll include this with some other DAX fixes I have pending.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#2 lore raw
Quoted 54 lines · Dan Williams
Quoted 53 lines · Yigal Korman

DAX handling of COW faults has wrong locking sequence:
dax_fault does i_mmap_lock_read do_cow_fault does i_mmap_unlock_write

Ross's commit[1] missed a fix[2] that Kirill added to Matthew's commit[3].

Original COW locking logic was introduced by Matthew here[4].

This should be applied to v4.3 as well.

[1] 0f90cc6609c7 mm, dax: fix DAX deadlocks
[2] 52a2b53ffde6 mm, dax: use i_mmap_unlock_write() in do_cow_fault()
[3] 843172978bb9 dax: fix race between simultaneous faults
[4] 2e4cdab0584f mm: allow page fault handlers to perform the COW

Cc Stable Tree <[email protected]>
Cc Boaz Harrosh <[email protected]>
Cc Ross Zwisler <[email protected]>
Cc Alexander Viro <[email protected]>
Cc Dan Williams <[email protected]>
Cc Dave Chinner <[email protected]>
Cc Jan Kara <[email protected]>
Cc "Kirill A. Shutemov" <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 mm/memory.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
mm/memory.c +3 -3
index c716913..e5071af 100644
@@ -3015,9 +3015,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
                } else {
                        /*
                         * The fault handler has no page to lock, so it holds
-                        * i_mmap_lock for write to protect against truncate.
+                        * i_mmap_lock for read to protect against truncate.
                         */
-                       i_mmap_unlock_write(vma->vm_file->f_mapping);
+                       i_mmap_unlock_read(vma->vm_file->f_mapping);
                }
                goto uncharge_out;
        }
@@ -3031,9 +3031,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        } else {
                /*
                 * The fault handler has no page to lock, so it holds
-                * i_mmap_lock for write to protect against truncate.
+                * i_mmap_lock for read to protect against truncate.
Quoted 8 lines

mm/memory.c static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, · line 3039

Quoted 6 lines
mm/memory.c +1 -1
                 */
-               i_mmap_unlock_write(vma->vm_file->f_mapping);
+               i_mmap_unlock_read(vma->vm_file->f_mapping);
        }
        return ret;
 uncharge_out:

Looks good to me. I'll include this with some other DAX fixes I have pending.

Looks good to me as well. Thanks for catching this.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#3 lore raw
Quoted 64 lines · Ross Zwisler
Quoted 54 lines · Dan Williams
Quoted 53 lines · Yigal Korman

DAX handling of COW faults has wrong locking sequence:
dax_fault does i_mmap_lock_read do_cow_fault does i_mmap_unlock_write

Ross's commit[1] missed a fix[2] that Kirill added to Matthew's commit[3].

Original COW locking logic was introduced by Matthew here[4].

This should be applied to v4.3 as well.

[1] 0f90cc6609c7 mm, dax: fix DAX deadlocks
[2] 52a2b53ffde6 mm, dax: use i_mmap_unlock_write() in do_cow_fault()
[3] 843172978bb9 dax: fix race between simultaneous faults
[4] 2e4cdab0584f mm: allow page fault handlers to perform the COW

Cc Stable Tree <[email protected]>
Cc Boaz Harrosh <[email protected]>
Cc Ross Zwisler <[email protected]>
Cc Alexander Viro <[email protected]>
Cc Dan Williams <[email protected]>
Cc Dave Chinner <[email protected]>
Cc Jan Kara <[email protected]>
Cc "Kirill A. Shutemov" <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 mm/memory.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
mm/memory.c +3 -3
index c716913..e5071af 100644
@@ -3015,9 +3015,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
                } else {
                        /*
                         * The fault handler has no page to lock, so it holds
-                        * i_mmap_lock for write to protect against truncate.
+                        * i_mmap_lock for read to protect against truncate.
                         */
-                       i_mmap_unlock_write(vma->vm_file->f_mapping);
+                       i_mmap_unlock_read(vma->vm_file->f_mapping);
                }
                goto uncharge_out;
        }
@@ -3031,9 +3031,9 @@ static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        } else {
                /*
                 * The fault handler has no page to lock, so it holds
-                * i_mmap_lock for write to protect against truncate.
+                * i_mmap_lock for read to protect against truncate.
Quoted 8 lines

mm/memory.c static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, · line 3039

Quoted 6 lines
mm/memory.c +1 -1
                 */
-               i_mmap_unlock_write(vma->vm_file->f_mapping);
+               i_mmap_unlock_read(vma->vm_file->f_mapping);
        }
        return ret;
 uncharge_out:

Looks good to me. I'll include this with some other DAX fixes I have pending.

Quoted 2 lines

Looks good to me as well. Thanks for catching this.

Yes. None of the xfstests catch this. It needs a private-mapping mmap in some combination of other activity on the file at the same time.

Which the linker of gcc does. We have a test of a git clone Kernel-tree and make. which catches this in the make phase. For some reason on ext4 it is reliable to crash but on xfs 1/2 the runs go through, go figure.

Thanks Yigal for the fast fix
Boaz

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>