| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20240712121800.3049-2-spasswolf@web.de>
Date: Fri, 12 Jul 2024 14:17:59 +0200
From: Bert Karwatzki <spasswolf@....de>
To: "Liam R . Howlett" <Liam.Howlett@...cle.com>
Cc: Bert Karwatzki <spasswolf@....de>,
Andrew Morton <akpm@...ux-foundation.org>,
Jiri Olsa <olsajiri@...il.com>,
linux-mm@...ck.org,
linux-kernel@...r.kernel.org,
linux-next@...r.kernel.org
Subject: Re: commit 1c29a32ce65f4cd0f1c causes Bad rss-counter state and firefox-esr crash in linux-next-20240613
I did some experiments on the rss counter bug. The next patch is made for linux-next-20240613
with commit 1c29a32ce65f4cd0f1c0f9 reverted. Then I simply inlined the code of do_vmi_unmap()
and do_vmi_align_munmap() into mmap_region(). This version of the code works fine and does not
show the rss counter bug.
diff --git a/mm/mmap.c b/mm/mmap.c
index f95af72ddc9f..0f020c535c83 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -733,7 +733,6 @@ 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:
@@ -2911,6 +2910,8 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
struct vm_area_struct *next, *prev, *merge;
pgoff_t pglen = len >> PAGE_SHIFT;
unsigned long charged = 0;
+ struct vma_munmap_struct vms;
+ struct ma_state mas_detach;
unsigned long end = addr + len;
unsigned long merge_start = addr, merge_end = end;
bool writable_file_mapping = false;
@@ -2933,12 +2934,46 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
return -ENOMEM;
}
- /* Unmap any existing mapping in the area */
- error = do_vmi_munmap(&vmi, mm, addr, len, uf, false);
- if (error == -EPERM)
- return error;
- else if (error)
- return -ENOMEM;
+ /* Find the first overlapping VMA */
+ vma = vma_find(&vmi, end);
+ if (vma) {
+ struct maple_tree mt_detach;
+
+ /*
+ * Check if memory is sealed before arch_unmap.
+ * Prevent unmapping a sealed VMA.
+ * can_modify_mm assumes we have acquired the lock on MM.
+ */
+ if (unlikely(!can_modify_mm(mm, addr, end))) {
+ return -EPERM;
+ }
+
+ /* arch_unmap() might do unmaps itself. */
+ arch_unmap(mm, addr, end);
+
+ mt_init_flags(&mt_detach, vmi.mas.tree->ma_flags & MT_FLAGS_LOCK_MASK);
+ mt_on_stack(mt_detach);
+ mas_init(&mas_detach, &mt_detach, 0);
+
+ init_vma_munmap(&vms, &vmi, vma, addr, end, uf, false);
+ error = vms_gather_munmap_vmas(&vms, &mas_detach);
+ if (error) {
+ validate_mm(mm);
+ return -ENOMEM;
+ }
+
+ vma = NULL;
+ error = vma_iter_clear_gfp(&vmi, addr, end, GFP_KERNEL);
+ if (error) {
+ abort_munmap_vmas(&mas_detach);
+ return -ENOMEM;
+ }
+
+ /* Point of no return */
+ vms_complete_munmap_vmas(&vms, &mas_detach);
+ } else {
+ // TODO
+ }
/*
* Private writable mapping: check memory availability
The next patch now moves the call to vms_complete_munmap_vmas() towards the end of
mmap_region(). This code is also free of the rss counter bug.
commit a4b24bb18dde627792297455befcc465e45be66d
Author: Bert Karwatzki <spasswolf@....de>
Date: Thu Jun 20 17:02:08 2024 +0200
mm: mmap: push back vms_complete_munmap_vmas()
In order to to debug the rss counter bug we're going to push back
vms_complete_munmap_vmas() in mmap_region.
Signed-off-by: Bert Karwatzki <spasswolf@....de>
diff --git a/mm/mmap.c b/mm/mmap.c
index 0f020c535c83..4fb9dd2e6d6e 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2970,9 +2970,9 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
}
/* Point of no return */
- vms_complete_munmap_vmas(&vms, &mas_detach);
} else {
- // TODO
+ vms.end = 0;
+ vms.nr_pages = 0;
}
/*
@@ -3016,6 +3016,11 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
vma_iter_next_range(&vmi);
}
+ if (vms.end) {
+ vms_complete_munmap_vmas(&vms, &mas_detach);
+ vms.end = 0; // avoid double unmap below
+ }
+
/* Actually expand, if possible */
if (vma &&
!vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) {
@@ -3026,7 +3031,8 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
if (vma == prev)
vma_iter_set(&vmi, addr);
cannot_expand:
-
+ if (vms.end)
+ vms_complete_munmap_vmas(&vms, &mas_detach);
/*
* Determine the object being mapped and call the appropriate
* specific mapper. the address has already been validated, but
The next patch move vms_complete_munmap_vmas() a little further beyond the
call to vma_expand(). This code contain the rss counter bug.
commit 02d6be2410fa503d008f4cc8dcd1518ca56f8793
Author: Bert Karwatzki <spasswolf@....de>
Date: Thu Jun 20 20:07:13 2024 +0200
mm: mmap: push back vms_complete_munmap_vmas()
This commit actually show the rss counter bug, while the previus does
not!
Signed-off-by: Bert Karwatzki <spasswolf@....de>
diff --git a/mm/mmap.c b/mm/mmap.c
index 4fb9dd2e6d6e..c5f4b4b6fb84 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -3016,14 +3016,12 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
vma_iter_next_range(&vmi);
}
- if (vms.end) {
- vms_complete_munmap_vmas(&vms, &mas_detach);
- vms.end = 0; // avoid double unmap below
- }
-
/* Actually expand, if possible */
if (vma &&
!vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) {
+ if (vms.end) {
+ vms_complete_munmap_vmas(&vms, &mas_detach);
+ }
khugepaged_enter_vma(vma, vm_flags);
goto expanded;
}
So there might be some unwanted interaction between vms_complete_munmap_vmas though
I've no yet figured out what exactly is happening. Hope this will be helpful in
solving the problem.
Bert Karwatzki
Powered by blists - more mailing lists