validate_mm() calls are too spread out and duplicated in numerous
locations. Also, now that the stack write is done under the write lock,
it is not necessary to validate the mm prior to write operations.
Add a validate_mm() to the stack expansions, and to vma_complete() so
that numerous others may be dropped.
Note that vma_link() (and also insert_vm_struct() by call path) already
call validate_mm().
vma_merge() also had an unnecessary call to vma_iter_free() since the
logic change to abort earlier if no merging is necessary.
Drop extra validate_mm() calls at the start of functions and error paths
which won't write to the tree.
Relocate the validate_mm() call in the do_brk_flags() to avoid
re-running the same test when vma_complete() is used.
The call within the error path of mmap_region() is left intentionally
because of the complexity of the function and the potential of drivers
modifying the tree.
Cc: Linus Torvalds <[email protected]>
Cc: Oliver Sang <[email protected]>
Signed-off-by: Liam R. Howlett <[email protected]>
---
mm/mmap.c | 24 ++++--------------------
1 file changed, 4 insertions(+), 20 deletions(-)
diff --git a/mm/mmap.c b/mm/mmap.c
index 204ddcd52625..964a8aa59297 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -597,6 +597,7 @@ static inline void vma_complete(struct vma_prepare *vp,
}
if (vp->insert && vp->file)
uprobe_mmap(vp->insert);
+ validate_mm(mm);
}
/*
@@ -678,7 +679,6 @@ int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
vma_iter_store(vmi, vma);
vma_complete(&vp, vmi, vma->vm_mm);
- validate_mm(vma->vm_mm);
return 0;
nomem:
@@ -718,7 +718,6 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
vma->vm_end = end;
vma->vm_pgoff = pgoff;
vma_complete(&vp, vmi, vma->vm_mm);
- validate_mm(vma->vm_mm);
return 0;
}
@@ -891,7 +890,6 @@ struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm,
pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
long adj_start = 0;
- validate_mm(mm);
/*
* We later require that vma->vm_flags == vm_flags,
* so this tests vma->vm_flags & VM_SPECIAL, too.
@@ -1018,10 +1016,7 @@ struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm,
}
vma_complete(&vp, vmi, mm);
- vma_iter_free(vmi);
- validate_mm(mm);
khugepaged_enter_vma(res, vm_flags);
-
return res;
}
@@ -1196,7 +1191,6 @@ unsigned long do_mmap(struct file *file, unsigned long addr,
vm_flags_t vm_flags;
int pkey = 0;
- validate_mm(mm);
*populate = 0;
if (!len)
@@ -2023,6 +2017,7 @@ static int expand_upwards(struct vm_area_struct *vma, unsigned long address)
anon_vma_unlock_write(vma->anon_vma);
khugepaged_enter_vma(vma, vma->vm_flags);
mas_destroy(&mas);
+ validate_mm(mm);
return error;
}
#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
@@ -2111,6 +2106,7 @@ int expand_downwards(struct vm_area_struct *vma, unsigned long address)
anon_vma_unlock_write(vma->anon_vma);
khugepaged_enter_vma(vma, vma->vm_flags);
mas_destroy(&mas);
+ validate_mm(mm);
return error;
}
@@ -2288,7 +2284,6 @@ static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas)
remove_vma(vma, false);
}
vm_unacct_memory(nr_accounted);
- validate_mm(mm);
}
/*
@@ -2325,8 +2320,6 @@ int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
struct vm_area_struct *new;
int err;
- validate_mm(vma->vm_mm);
-
WARN_ON(vma->vm_start >= addr);
WARN_ON(vma->vm_end <= addr);
@@ -2383,7 +2376,6 @@ int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
/* Success. */
if (new_below)
vma_next(vmi);
- validate_mm(vma->vm_mm);
return 0;
out_free_mpol:
@@ -2392,7 +2384,6 @@ int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
vma_iter_free(vmi);
out_free_vma:
vm_area_free(new);
- validate_mm(vma->vm_mm);
return err;
}
@@ -3043,7 +3034,6 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
struct mm_struct *mm = current->mm;
struct vma_prepare vp;
- validate_mm(mm);
/*
* Check against address space limits by the changed size
* Note: This happens *after* clearing old mappings in some code paths.
@@ -3095,6 +3085,7 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
goto mas_store_fail;
mm->map_count++;
+ validate_mm(mm);
ksm_add_vma(vma);
out:
perf_event_mmap(vma);
@@ -3103,7 +3094,6 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
if (flags & VM_LOCKED)
mm->locked_vm += (len >> PAGE_SHIFT);
vm_flags_set(vma, VM_SOFTDIRTY);
- validate_mm(mm);
return 0;
mas_store_fail:
@@ -3284,7 +3274,6 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
bool faulted_in_anon_vma = true;
VMA_ITERATOR(vmi, mm, addr);
- validate_mm(mm);
/*
* If anonymous vma has not yet been faulted, update new pgoff
* to match new location, to increase its chance of merging.
@@ -3343,7 +3332,6 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
goto out_vma_link;
*need_rmap_locks = false;
}
- validate_mm(mm);
return new_vma;
out_vma_link:
@@ -3359,7 +3347,6 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
out_free_vma:
vm_area_free(new_vma);
out:
- validate_mm(mm);
return NULL;
}
@@ -3496,7 +3483,6 @@ static struct vm_area_struct *__install_special_mapping(
int ret;
struct vm_area_struct *vma;
- validate_mm(mm);
vma = vm_area_alloc(mm);
if (unlikely(vma == NULL))
return ERR_PTR(-ENOMEM);
@@ -3519,12 +3505,10 @@ static struct vm_area_struct *__install_special_mapping(
perf_event_mmap(vma);
- validate_mm(mm);
return vma;
out:
vm_area_free(vma);
- validate_mm(mm);
return ERR_PTR(ret);
}
--
2.39.2
On Tue, 4 Jul 2023 at 11:25, Liam R. Howlett <[email protected]> wrote:
>
> validate_mm() calls are too spread out and duplicated in numerous
> locations. Also, now that the stack write is done under the write lock,
> it is not necessary to validate the mm prior to write operations.
So while I applied the fixes directly since I was doing all the
write-locking stuff (and asked for the locking cleanup), I'm hoping
these kinds of cleanups will now go back to normal and go through
Andrew.
I do have a question related to the write locking: now that we should
always hold the mmap lock for writing when doing any modifications,
can the "lock_is_held()" assertions be tightened?
Right now it's "any locking", but for actual modification it should
probably be using
lockdep_is_held_type(mt->ma_external_lock, 1)
but there's just one 'mt_lock_is_held()' function (presumably because
the internal lock is always just a spinlock that doesn't have the
reader/writer distinction).
Linus
* Linus Torvalds <[email protected]> [230704 14:36]:
> On Tue, 4 Jul 2023 at 11:25, Liam R. Howlett <[email protected]> wrote:
> >
> > validate_mm() calls are too spread out and duplicated in numerous
> > locations. Also, now that the stack write is done under the write lock,
> > it is not necessary to validate the mm prior to write operations.
>
> So while I applied the fixes directly since I was doing all the
> write-locking stuff (and asked for the locking cleanup), I'm hoping
> these kinds of cleanups will now go back to normal and go through
> Andrew.
>
> I do have a question related to the write locking: now that we should
> always hold the mmap lock for writing when doing any modifications,
> can the "lock_is_held()" assertions be tightened?
>
> Right now it's "any locking", but for actual modification it should
> probably be using
>
> lockdep_is_held_type(mt->ma_external_lock, 1)
>
> but there's just one 'mt_lock_is_held()' function (presumably because
> the internal lock is always just a spinlock that doesn't have the
> reader/writer distinction).
Ah, yes. I was trying to do just that, but ran into an issue and backed
out of fully fixing this portion up until later.
The issue arises with the use of the same external lock for the munmap()
case where I'm using the second tree to track the VMAs. Using the
spinlock causes issues with the potential sleeping within allocations.
So, I'm still working out a way to do what you (and willy) asked here.
Thanks,
Liam
* Liam R. Howlett <[email protected]> [230704 14:47]:
> * Linus Torvalds <[email protected]> [230704 14:36]:
> > On Tue, 4 Jul 2023 at 11:25, Liam R. Howlett <[email protected]> wrote:
> > >
> > > validate_mm() calls are too spread out and duplicated in numerous
> > > locations. Also, now that the stack write is done under the write lock,
> > > it is not necessary to validate the mm prior to write operations.
> >
> > So while I applied the fixes directly since I was doing all the
> > write-locking stuff (and asked for the locking cleanup), I'm hoping
> > these kinds of cleanups will now go back to normal and go through
> > Andrew.
> >
> > I do have a question related to the write locking: now that we should
> > always hold the mmap lock for writing when doing any modifications,
> > can the "lock_is_held()" assertions be tightened?
> >
> > Right now it's "any locking", but for actual modification it should
> > probably be using
> >
> > lockdep_is_held_type(mt->ma_external_lock, 1)
For completeness of the email tread; it turns out we want 0 as the last
parameter.
(include/linux/lockdep.h)
/*
* Acquire a lock.
*
* Values for "read":
*
* 0: exclusive (write) acquire
* 1: read-acquire (no recursion allowed)
* 2: read-acquire with same-instance recursion allowed
*
* Values for check:
*
* 0: simple checks (freeing, held-at-exit-time, etc.)
* 1: full validation
*/
...
/*
* Same "read" as for lock_acquire(), except -1 means any.
*/
extern int lock_is_held_type(const struct lockdep_map *lock, int read);
> >
> > but there's just one 'mt_lock_is_held()' function (presumably because
> > the internal lock is always just a spinlock that doesn't have the
> > reader/writer distinction).
>
> Ah, yes. I was trying to do just that, but ran into an issue and backed
> out of fully fixing this portion up until later.
>
Here are two patches to increase the strictness of the maple tree
locking. I've boot tested them on x86_64 with the bots config and
ensured the lockdep problem was resolved.
The first introduces the new mt_write_locked() function, which ensures
the lock type is for writing.
The second updates the munmap path to avoid triggering the warnings
associated with dropping the mmap_lock prior to freeing the VMAs.
Thanks,
Liam