Re: [PATCH v5 01/19] KVM: x86: Allocate new rmap and large page tracking when moving memslot

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, Feb 05, 2020 at 09:00:31PM -0500, Peter Xu wrote:
> On Wed, Feb 05, 2020 at 03:55:33PM -0800, Sean Christopherson wrote:
> > On Wed, Feb 05, 2020 at 04:49:52PM -0500, Peter Xu wrote:
> > > On Tue, Jan 21, 2020 at 02:31:39PM -0800, Sean Christopherson wrote:
> > > > Reallocate a rmap array and recalcuate large page compatibility when
> > > > moving an existing memslot to correctly handle the alignment properties
> > > > of the new memslot.  The number of rmap entries required at each level
> > > > is dependent on the alignment of the memslot's base gfn with respect to
> > > > that level, e.g. moving a large-page aligned memslot so that it becomes
> > > > unaligned will increase the number of rmap entries needed at the now
> > > > unaligned level.
> > > > 
> > > > Not updating the rmap array is the most obvious bug, as KVM accesses
> > > > garbage data beyond the end of the rmap.  KVM interprets the bad data as
> > > > pointers, leading to non-canonical #GPs, unexpected #PFs, etc...
> > > > 
> > > >   general protection fault: 0000 [#1] SMP
> > > >   CPU: 0 PID: 1909 Comm: move_memory_reg Not tainted 5.4.0-rc7+ #139
> > > >   Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
> > > >   RIP: 0010:rmap_get_first+0x37/0x50 [kvm]
> > > >   Code: <48> 8b 3b 48 85 ff 74 ec e8 6c f4 ff ff 85 c0 74 e3 48 89 d8 5b c3
> > > >   RSP: 0018:ffffc9000021bbc8 EFLAGS: 00010246
> > > >   RAX: ffff00617461642e RBX: ffff00617461642e RCX: 0000000000000012
> > > >   RDX: ffff88827400f568 RSI: ffffc9000021bbe0 RDI: ffff88827400f570
> > > >   RBP: 0010000000000000 R08: ffffc9000021bd00 R09: ffffc9000021bda8
> > > >   R10: ffffc9000021bc48 R11: 0000000000000000 R12: 0030000000000000
> > > >   R13: 0000000000000000 R14: ffff88827427d700 R15: ffffc9000021bce8
> > > >   FS:  00007f7eda014700(0000) GS:ffff888277a00000(0000) knlGS:0000000000000000
> > > >   CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > >   CR2: 00007f7ed9216ff8 CR3: 0000000274391003 CR4: 0000000000162eb0
> > > >   Call Trace:
> > > >    kvm_mmu_slot_set_dirty+0xa1/0x150 [kvm]
> > > >    __kvm_set_memory_region.part.64+0x559/0x960 [kvm]
> > > >    kvm_set_memory_region+0x45/0x60 [kvm]
> > > >    kvm_vm_ioctl+0x30f/0x920 [kvm]
> > > >    do_vfs_ioctl+0xa1/0x620
> > > >    ksys_ioctl+0x66/0x70
> > > >    __x64_sys_ioctl+0x16/0x20
> > > >    do_syscall_64+0x4c/0x170
> > > >    entry_SYSCALL_64_after_hwframe+0x44/0xa9
> > > >   RIP: 0033:0x7f7ed9911f47
> > > >   Code: <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 21 6f 2c 00 f7 d8 64 89 01 48
> > > >   RSP: 002b:00007ffc00937498 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
> > > >   RAX: ffffffffffffffda RBX: 0000000001ab0010 RCX: 00007f7ed9911f47
> > > >   RDX: 0000000001ab1350 RSI: 000000004020ae46 RDI: 0000000000000004
> > > >   RBP: 000000000000000a R08: 0000000000000000 R09: 00007f7ed9214700
> > > >   R10: 00007f7ed92149d0 R11: 0000000000000246 R12: 00000000bffff000
> > > >   R13: 0000000000000003 R14: 00007f7ed9215000 R15: 0000000000000000
> > > >   Modules linked in: kvm_intel kvm irqbypass
> > > >   ---[ end trace 0c5f570b3358ca89 ]---
> > > > 
> > > > The disallow_lpage tracking is more subtle.  Failure to update results
> > > > in KVM creating large pages when it shouldn't, either due to stale data
> > > > or again due to indexing beyond the end of the metadata arrays, which
> > > > can lead to memory corruption and/or leaking data to guest/userspace.
> > > > 
> > > > Note, the arrays for the old memslot are freed by the unconditional call
> > > > to kvm_free_memslot() in __kvm_set_memory_region().
> > > 
> > > If __kvm_set_memory_region() failed, I think the old memslot will be
> > > kept and the new memslot will be freed instead?
> > 
> > This is referring to a successful MOVE operation to note that zeroing @arch
> > in kvm_arch_create_memslot() won't leak memory.
> > 
> > > > 
> > > > Fixes: 05da45583de9b ("KVM: MMU: large page support")
> > > > Cc: stable@xxxxxxxxxxxxxxx
> > > > Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> > > > ---
> > > >  arch/x86/kvm/x86.c | 11 +++++++++++
> > > >  1 file changed, 11 insertions(+)
> > > > 
> > > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > > > index 4c30ebe74e5d..1953c71c52f2 100644
> > > > --- a/arch/x86/kvm/x86.c
> > > > +++ b/arch/x86/kvm/x86.c
> > > > @@ -9793,6 +9793,13 @@ int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
> > > >  {
> > > >  	int i;
> > > >  
> > > > +	/*
> > > > +	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
> > > > +	 * old arrays will be freed by __kvm_set_memory_region() if installing
> > > > +	 * the new memslot is successful.
> > > > +	 */
> > > > +	memset(&slot->arch, 0, sizeof(slot->arch));
> > > 
> > > I actually gave r-b on this patch but it was lost... And then when I
> > > read it again I start to confuse on why we need to set these to zeros.
> > > Even if they're not zeros, iiuc kvm_free_memslot() will compare each
> > > of the array pointer and it will only free the changed pointers, then
> > > it looks fine even without zeroing?
> > 
> > It's for the failure path, the out_free label, which blindy calls kvfree()
> > and relies on un-allocated pointers being NULL.  If @arch isn't zeroed, the
> > failure path will free metadata from the previous memslot.
> 
> IMHO it won't, because kvm_free_memslot() will only free metadata if
> the pointer changed.  So:
> 
>   - For succeeded kvcalloc(), the pointer will change in the new slot,
>     so kvm_free_memslot() will free it,
> 
>   - For failed kvcalloc(), the pointer will be NULL, so
>     kvm_free_memslot() will skip it,

No.  The out_free path iterates over all possible entries and would free
pointers from the old memslot.  It's still be wrong even if the very last
kcalloc() failed as that allocation is captured in a local variable and
only propagated to lpage_info on success.

out_free:
	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
		kvfree(slot->arch.rmap[i]);
		slot->arch.rmap[i] = NULL;
		if (i == 0)
			continue;

		kvfree(slot->arch.lpage_info[i - 1]);
		slot->arch.lpage_info[i - 1] = NULL;
	}
	return -ENOMEM;

>   - For untouched pointer, it'll be the same as the old, so
>     kvm_free_memslot() will skip it as well.
> 
> > 
> > > > +
> > > >  	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
> > > >  		struct kvm_lpage_info *linfo;
> > > >  		unsigned long ugfn;
> > > > @@ -9867,6 +9874,10 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
> > > >  				const struct kvm_userspace_memory_region *mem,
> > > >  				enum kvm_mr_change change)
> > > >  {
> > > > +	if (change == KVM_MR_MOVE)
> > > > +		return kvm_arch_create_memslot(kvm, memslot,
> > > > +					       mem->memory_size >> PAGE_SHIFT);
> > > > +
> > > 
> > > Instead of calling kvm_arch_create_memslot() explicitly again here,
> > > can it be replaced by below?
> > > 
> > > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> > > index 72b45f491692..85a7b02fd752 100644
> > > --- a/virt/kvm/kvm_main.c
> > > +++ b/virt/kvm/kvm_main.c
> > > @@ -1144,7 +1144,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
> > >                 new.dirty_bitmap = NULL;
> > >  
> > >         r = -ENOMEM;
> > > -       if (change == KVM_MR_CREATE) {
> > > +       if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) {
> > >                 new.userspace_addr = mem->userspace_addr;
> > >  
> > >                 if (kvm_arch_create_memslot(kvm, &new, npages))
> > 
> > No, because other architectures don't need to re-allocate new metadata on
> > MOVE and rely on __kvm_set_memory_region() to copy @arch from old to new,
> > e.g. see kvmppc_core_create_memslot_hv().
> 
> Yes it's only required in x86, but iiuc it also will still work for
> ppc?  Say, in that case ppc won't copy @arch from old to new, and
> kvmppc_core_free_memslot_hv() will free the old, however it should
> still work.

No, calling kvm_arch_create_memslot() for MOVE will result in PPC leaking
memory due to overwriting slot->arch.rmap with a new allocation.

> > 
> > That being said, that's effectively what the x86 code looks like once
> > kvm_arch_create_memslot() gets merged into kvm_arch_prepare_memory_region().
> 
> Right.  I don't have strong opinion on this, but if my above analysis
> is correct, it's still slightly cleaner, imho, to have this patch as a
> oneliner as I provided, then in the other patch move the whole
> CREATE|MOVE into prepare_memory_region().  The final code should be
> the same.
> 
> -- 
> Peter Xu
> 
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm



[Index of Archives]     [Linux KVM]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux