On Thu, 15 Oct 2015, Andrea Arcangeli wrote:
> The ksm_test_exit() run after down_read(&mm->mmap_sem) assumed it
> could serialize against ksm_exit() and prevent exit_mmap() to run
> until the up_read(&mm->mmap_sem). That is true when the rmap_item->mm
> is the one associated with the ksm_scan.mm_slot, as ksm_exit() would
> take the !easy_to_free_path.
>
> The problem is that when merging the current rmap_item (the one whose
> ->mm pointer is always associated ksm_scan.mm_slot) with a
> tree_rmap_item in the unstable tree, the unstable tree
> tree_rmap_item->mm can be any random mm. The locking technique
> described above is a noop if the rmap_item->mm is not the one
> associated with the ksm_scan.mm_slot.
I was convinced for an hour, though puzzled how this had survived
six years without being noticed: I'd certainly found the need for
the special ksm_exit()/ksm_test_exit() stuff when testing originally,
that wasn't hard, and why would this race be so very much harder?
Now, after looking again at ksm_exit(), I just don't see the point
you're making. As I read it (and I certainly accept that I may be
wrong on all this), it will do the down_write,up_write on any mm
that is registered with it, and that has a chain of rmap_items -
the easy_to_free route is only for those that have no rmap_items
(and are not being worked on at present); and those that have no
rmap_items, have no rmap_items in the unstable or the stable tree.
Please explain what I'm missing before I look again harder. One
nit below. It looked very reasonable and nicely implemented to me,
but I didn't complete checking it before I lost sight of what it's
fixing. (And incrementing mm_users always makes me worry a bit
about what happens under OOM, so I prefer not to do it.)
I've seen no problems at all in running rc5-mm1 and rc6-mm1
(with "allksm" of course) on this series.
Hugh
> In turn the tree_rmap_item when
> converted to a stable tree rmap_item and added to the
> stable_node->hlist, can have a &rmap_item->anon_vma that points to
> already freed memory. The find_vma and other vma operations to reach
> the anon_vma also run on potentially already freed memory. The
> get_anon_vma atomic_inc itself could corrupt memory randomly in
> already re-used memory.
>
> The result are oopses like below:
>
> general protection fault: 0000 [#1] SMP
> last sysfs file: /sys/kernel/mm/ksm/sleep_millisecs
> CPU 14
> Modules linked in: netconsole nfs nfs_acl auth_rpcgss fscache lockd sunrpc msr binfmt_misc sr_mod
>
> Pid: 904, comm: ksmd Not tainted 2.6.32 #21 Supermicro X8DTN/X8DTN
Time for an update, maybe :-? Though, in seriousness, I don't
think we have fixed anything of this nature in KSM since then.
Have probably fixed some anon_vma lifetime issues, though.
> RIP: 0010:[<ffffffff81138300>] [<ffffffff81138300>] drop_anon_vma+0x0/0xe0
> RSP: 0000:ffff88023b94bd28 EFLAGS: 00010206
> RAX: 0000000000000000 RBX: ffff880231e64d50 RCX: 0000000000000017
> RDX: ffff88023b94bfd8 RSI: 0000000000000216 RDI: 80000002192c6067
> RBP: ffff88023b94bd40 R08: 0000000000000001 R09: 0000000000000000
> R10: 0000000000000001 R11: 0000000000000001 R12: ffff880217fa4198
> R13: ffff880217fa4198 R14: 000000000021916f R15: ffff880217fa419b
> FS: 0000000000000000(0000) GS:ffff880030600000(0000) knlGS:0000000000000000
> CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b
> CR2: 0000000000d116a0 CR3: 000000023aa35000 CR4: 00000000000007e0
> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> Process ksmd (pid: 904, threadinfo ffff88023b948000, task ffff88023b946500)
> Stack:
> ffffffff8114d049 ffffea000757d048 ffffea0000000000 ffff88023b94bda0
> <d> ffffffff8114d153 ffff88023b946500 ffff88023afaab40 ffffffff8114e85d
> <d> 0100000000000038 ffff88023b94bdb0 ffff880231d33560 ffff880217fa4198
> Call Trace:
> [<ffffffff8114d049>] ? remove_node_from_stable_tree+0x29/0x80
> [<ffffffff8114d153>] get_ksm_page+0xb3/0x1e0
> [<ffffffff8114e85d>] ? ksm_scan_thread+0x60d/0x1130
> [<ffffffff8114d499>] remove_rmap_item_from_tree+0x99/0x130
> [<ffffffff8114ed19>] ksm_scan_thread+0xac9/0x1130
> [<ffffffff81095ce0>] ? autoremove_wake_function+0x0/0x40
> [<ffffffff8114e250>] ? ksm_scan_thread+0x0/0x1130
> [<ffffffff8109508b>] kthread+0x8b/0xb0
> [<ffffffff8100c0ea>] child_rip+0xa/0x20
> [<ffffffff8100b910>] ? restore_args+0x0/0x30
> [<ffffffff81095000>] ? kthread+0x0/0xb0
> [<ffffffff8100c0e0>] ? child_rip+0x0/0x20
> Code: 01 75 10 e8 d3 f7 ff ff 5d c3 90 0f 0b eb fe 0f 1f 40 00 e8 73 f6 ff ff 5d c3 e8 4c 77 01 00 5d c3 66 2e 0f 1f 84 00 00 00 00 00 <8b> 47 48 85 c0 0f 8e c5 00 00 00 55 48 89 e5 41 56 41 55 41 54
> RIP [<ffffffff81138300>] drop_anon_vma+0x0/0xe0
> RSP <ffff88023b94bd28>
> ---[ end trace b1f69fd4c12ce1ce ]---
>
> rmap_item->anon_vma was set to the RDI value 0x80000002192c6067.
>
> 0xffffffff81138300 <+0>: mov 0x48(%rdi),%eax
> 0xffffffff81138303 <+3>: test %eax,%eax
> 0xffffffff81138305 <+5>: jle 0xffffffff811383d0 <drop_anon_vma+208>
> 0xffffffff8113830b <+11>: push %rbp
>
> Other oopses are more random and harder to debug side effects of
> memory corruption. In this case the anon_vma was a dangling pointer
> because when try_to_merge_with_ksm_page did rmap_item->anon_vma =
> vma->anon_vma, the vma already was already freed and reused memory. At
> other times the oopses materialize with an vma->anon_vma pointer that
> looks legit but it points to an already freed and reused anon_vma.
>
> Signed-off-by: Andrea Arcangeli <aarcange@xxxxxxxxxx>
> ---
> mm/ksm.c | 55 +++++++++++++++++++++++++++++++++++++++++++++++++++----
> 1 file changed, 51 insertions(+), 4 deletions(-)
>
> diff --git a/mm/ksm.c b/mm/ksm.c
> index 7ee101e..8fc6793 100644
> --- a/mm/ksm.c
> +++ b/mm/ksm.c
> @@ -350,6 +350,24 @@ static inline bool ksm_test_exit(struct mm_struct *mm)
> }
>
> /*
> + * If the mm isn't the one associated with the current
> + * ksm_scan.mm_slot ksm_exit() will not down_write();up_write() and in
> + * turn the ksm_test_exit() check run inside a mm->mmap_sem critical
> + * section, will not prevent exit_mmap() to run from under us. In
> + * turn, in those cases where we could work with an "mm" that isn't
> + * guaranteed to be associated with the current ksm_scan.mm_slot,
> + * ksm_get_mm() is needed instead of the ksm_test_exit() run inside
> + * the mmap_sem. Return true if the mm_users was incremented or false
> + * if it we failed at taking the mm because it was freed from under
> + * us. If it returns 1, the caller must take care of calling mmput()
> + * after it finishes using the mm.
> + */
> +static __always_inline bool ksm_get_mm(struct mm_struct *mm)
The nit was to ask, what's behind that __always_inline?
If the compiler chooses oddly not to inline it, that's okay, isn't it?
> +{
> + return likely(atomic_inc_not_zero(&mm->mm_users));
> +}
> +
> +/*
> * We use break_ksm to break COW on a ksm page: it's a stripped down
> *
> * if (get_user_pages(current, mm, addr, 1, 1, 1, &page, NULL) == 1)
> @@ -412,8 +430,6 @@ static struct vm_area_struct *find_mergeable_vma(struct mm_struct *mm,
> unsigned long addr)
> {
> struct vm_area_struct *vma;
> - if (ksm_test_exit(mm))
> - return NULL;
> vma = find_vma(mm, addr);
> if (!vma || vma->vm_start > addr)
> return NULL;
> @@ -434,11 +450,21 @@ static void break_cow(struct rmap_item *rmap_item)
> */
> put_anon_vma(rmap_item->anon_vma);
>
> + /*
> + * The "mm" of the unstable tree rmap_item isn't necessairly
> + * associated with the current ksm_scan.mm_slot, it could be
> + * any random mm. So we need ksm_get_mm here to prevent the
> + * exit_mmap to run from under us in mmput().
> + */
> + if (!ksm_get_mm(mm))
> + return;
> +
> down_read(&mm->mmap_sem);
> vma = find_mergeable_vma(mm, addr);
> if (vma)
> break_ksm(vma, addr);
> up_read(&mm->mmap_sem);
> + mmput(mm);
> }
>
> static struct page *page_trans_compound_anon(struct page *page)
> @@ -462,6 +488,15 @@ static struct page *get_mergeable_page(struct rmap_item *rmap_item)
> struct vm_area_struct *vma;
> struct page *page;
>
> + /*
> + * The "mm" of the unstable tree rmap_item isn't necessairly
> + * associated with the current ksm_scan.mm_slot, it could be
> + * any random mm. So we need ksm_get_mm here to prevent the
> + * exit_mmap to run from under us in mmput().
> + */
> + if (!ksm_get_mm(mm))
> + return NULL;
> +
> down_read(&mm->mmap_sem);
> vma = find_mergeable_vma(mm, addr);
> if (!vma)
> @@ -478,6 +513,7 @@ static struct page *get_mergeable_page(struct rmap_item *rmap_item)
> out: page = NULL;
> }
> up_read(&mm->mmap_sem);
> + mmput(mm);
> return page;
> }
>
> @@ -1086,9 +1122,19 @@ static int try_to_merge_with_ksm_page(struct rmap_item *rmap_item,
> struct vm_area_struct *vma;
> int err = -EFAULT;
>
> + /*
> + * The "mm" of the unstable tree rmap_item isn't necessairly
> + * associated with the current ksm_scan.mm_slot, it could be
> + * any random mm. So we need ksm_get_mm() here to prevent the
> + * exit_mmap to run from under us in mmput(). Otherwise
> + * rmap_item->anon_vma could point to an anon_vma that has
> + * already been freed (i.e. get_anon_vma() below would run too
> + * late).
> + */
> + if (!ksm_get_mm(mm))
> + return err;
> +
> down_read(&mm->mmap_sem);
> - if (ksm_test_exit(mm))
> - goto out;
> vma = find_vma(mm, rmap_item->address);
> if (!vma || vma->vm_start > rmap_item->address)
> goto out;
> @@ -1105,6 +1151,7 @@ static int try_to_merge_with_ksm_page(struct rmap_item *rmap_item,
> get_anon_vma(vma->anon_vma);
> out:
> up_read(&mm->mmap_sem);
> + mmput(mm);
> return err;
> }
>
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>