KVM: remove the usage of the mmap_sem for the protection of the memory slots.

This patch replaces the mmap_sem lock for the memory slots with a new
kvm private lock, it is needed beacuse untill now there were cases where
kvm accesses user memory while holding the mmap semaphore.

Signed-off-by: Izik Eidus <izike@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>

6 files changed, 75 insertions, 40 deletions

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 8efdcdb..2603710 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -876,11 +876,18 @@ static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
 
 struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
 {
+	struct page *page;
+
 	gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
 
 	if (gpa == UNMAPPED_GVA)
 		return NULL;
-	return gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
+
+	down_read(&current->mm->mmap_sem);
+	page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
+	up_read(&current->mm->mmap_sem);
+
+	return page;
 }
 
 static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
@@ -1020,15 +1027,18 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 
 	struct page *page;
 
+	down_read(&vcpu->kvm->slots_lock);
+
 	down_read(&current->mm->mmap_sem);
 	page = gfn_to_page(vcpu->kvm, gfn);
+	up_read(&current->mm->mmap_sem);
 
 	spin_lock(&vcpu->kvm->mmu_lock);
 	kvm_mmu_free_some_pages(vcpu);
 	r = __nonpaging_map(vcpu, v, write, gfn, page);
 	spin_unlock(&vcpu->kvm->mmu_lock);
 
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 
 	return r;
 }
@@ -1362,6 +1372,7 @@ static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 	gfn_t gfn;
 	int r;
 	u64 gpte = 0;
+	struct page *page;
 
 	if (bytes != 4 && bytes != 8)
 		return;
@@ -1389,6 +1400,11 @@ static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 	if (!is_present_pte(gpte))
 		return;
 	gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
+
+	down_read(&current->mm->mmap_sem);
+	page = gfn_to_page(vcpu->kvm, gfn);
+	up_read(&current->mm->mmap_sem);
+
 	vcpu->arch.update_pte.gfn = gfn;
 	vcpu->arch.update_pte.page = gfn_to_page(vcpu->kvm, gfn);
 }
@@ -1496,9 +1512,9 @@ int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
 	gpa_t gpa;
 	int r;
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 
 	spin_lock(&vcpu->kvm->mmu_lock);
 	r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
index 03ba860..2009c6e 100644
--- a/arch/x86/kvm/paging_tmpl.h
+++ b/arch/x86/kvm/paging_tmpl.h
@@ -91,7 +91,10 @@ static bool FNAME(cmpxchg_gpte)(struct kvm *kvm,
 	pt_element_t *table;
 	struct page *page;
 
+	down_read(&current->mm->mmap_sem);
 	page = gfn_to_page(kvm, table_gfn);
+	up_read(&current->mm->mmap_sem);
+
 	table = kmap_atomic(page, KM_USER0);
 
 	ret = CMPXCHG(&table[index], orig_pte, new_pte);
@@ -378,7 +381,7 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	if (r)
 		return r;
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	/*
 	 * Look up the shadow pte for the faulting address.
 	 */
@@ -392,11 +395,13 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 		pgprintk("%s: guest page fault\n", __FUNCTION__);
 		inject_page_fault(vcpu, addr, walker.error_code);
 		vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
-		up_read(&current->mm->mmap_sem);
+		up_read(&vcpu->kvm->slots_lock);
 		return 0;
 	}
 
+	down_read(&current->mm->mmap_sem);
 	page = gfn_to_page(vcpu->kvm, walker.gfn);
+	up_read(&current->mm->mmap_sem);
 
 	spin_lock(&vcpu->kvm->mmu_lock);
 	kvm_mmu_free_some_pages(vcpu);
@@ -413,14 +418,14 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	 */
 	if (shadow_pte && is_io_pte(*shadow_pte)) {
 		spin_unlock(&vcpu->kvm->mmu_lock);
-		up_read(&current->mm->mmap_sem);
+		up_read(&vcpu->kvm->slots_lock);
 		return 1;
 	}
 
 	++vcpu->stat.pf_fixed;
 	kvm_mmu_audit(vcpu, "post page fault (fixed)");
 	spin_unlock(&vcpu->kvm->mmu_lock);
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 
 	return write_pt;
 }
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index ad36447..86f5bf1 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -1477,7 +1477,7 @@ static int alloc_apic_access_page(struct kvm *kvm)
 	struct kvm_userspace_memory_region kvm_userspace_mem;
 	int r = 0;
 
-	down_write(&current->mm->mmap_sem);
+	down_write(&kvm->slots_lock);
 	if (kvm->arch.apic_access_page)
 		goto out;
 	kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
@@ -1487,9 +1487,12 @@ static int alloc_apic_access_page(struct kvm *kvm)
 	r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
 	if (r)
 		goto out;
+
+	down_read(&current->mm->mmap_sem);
 	kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
+	up_read(&current->mm->mmap_sem);
 out:
-	up_write(&current->mm->mmap_sem);
+	up_write(&kvm->slots_lock);
 	return r;
 }
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 338764fa..6b01552 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -184,7 +184,7 @@ int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 	int ret;
 	u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 				  offset * sizeof(u64), sizeof(pdpte));
 	if (ret < 0) {
@@ -201,7 +201,7 @@ int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 
 	memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 out:
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 
 	return ret;
 }
@@ -215,13 +215,13 @@ static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 	if (is_long_mode(vcpu) || !is_pae(vcpu))
 		return false;
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 	if (r < 0)
 		goto out;
 	changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 out:
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 
 	return changed;
 }
@@ -359,7 +359,7 @@ void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 		 */
 	}
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	/*
 	 * Does the new cr3 value map to physical memory? (Note, we
 	 * catch an invalid cr3 even in real-mode, because it would
@@ -375,7 +375,7 @@ void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 		vcpu->arch.cr3 = cr3;
 		vcpu->arch.mmu.new_cr3(vcpu);
 	}
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 }
 EXPORT_SYMBOL_GPL(set_cr3);
 
@@ -1232,12 +1232,12 @@ static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 		return -EINVAL;
 
-	down_write(&current->mm->mmap_sem);
+	down_write(&kvm->slots_lock);
 
 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
 
-	up_write(&current->mm->mmap_sem);
+	up_write(&kvm->slots_lock);
 	return 0;
 }
 
@@ -1286,7 +1286,7 @@ static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 	    < alias->target_phys_addr)
 		goto out;
 
-	down_write(&current->mm->mmap_sem);
+	down_write(&kvm->slots_lock);
 
 	p = &kvm->arch.aliases[alias->slot];
 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
@@ -1300,7 +1300,7 @@ static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 
 	kvm_mmu_zap_all(kvm);
 
-	up_write(&current->mm->mmap_sem);
+	up_write(&kvm->slots_lock);
 
 	return 0;
 
@@ -1376,7 +1376,7 @@ int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 	struct kvm_memory_slot *memslot;
 	int is_dirty = 0;
 
-	down_write(&current->mm->mmap_sem);
+	down_write(&kvm->slots_lock);
 
 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
 	if (r)
@@ -1392,7 +1392,7 @@ int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 	}
 	r = 0;
 out:
-	up_write(&current->mm->mmap_sem);
+	up_write(&kvm->slots_lock);
 	return r;
 }
 
@@ -1570,7 +1570,7 @@ int emulator_read_std(unsigned long addr,
 	void *data = val;
 	int r = X86EMUL_CONTINUE;
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	while (bytes) {
 		gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
 		unsigned offset = addr & (PAGE_SIZE-1);
@@ -1592,7 +1592,7 @@ int emulator_read_std(unsigned long addr,
 		addr += tocopy;
 	}
 out:
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 	return r;
 }
 EXPORT_SYMBOL_GPL(emulator_read_std);
@@ -1611,9 +1611,9 @@ static int emulator_read_emulated(unsigned long addr,
 		return X86EMUL_CONTINUE;
 	}
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 
 	/* For APIC access vmexit */
 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
@@ -1651,14 +1651,14 @@ static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 {
 	int ret;
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
 	if (ret < 0) {
-		up_read(&current->mm->mmap_sem);
+		up_read(&vcpu->kvm->slots_lock);
 		return 0;
 	}
 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 	return 1;
 }
 
@@ -1670,9 +1670,9 @@ static int emulator_write_emulated_onepage(unsigned long addr,
 	struct kvm_io_device *mmio_dev;
 	gpa_t                 gpa;
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 
 	if (gpa == UNMAPPED_GVA) {
 		kvm_inject_page_fault(vcpu, addr, 2);
@@ -1749,7 +1749,7 @@ static int emulator_cmpxchg_emulated(unsigned long addr,
 		char *kaddr;
 		u64 val;
 
-		down_read(&current->mm->mmap_sem);
+		down_read(&vcpu->kvm->slots_lock);
 		gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
 
 		if (gpa == UNMAPPED_GVA ||
@@ -1760,13 +1760,17 @@ static int emulator_cmpxchg_emulated(unsigned long addr,
 			goto emul_write;
 
 		val = *(u64 *)new;
+
+		down_read(&current->mm->mmap_sem);
 		page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
+		up_read(&current->mm->mmap_sem);
+
 		kaddr = kmap_atomic(page, KM_USER0);
 		set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
 		kunmap_atomic(kaddr, KM_USER0);
 		kvm_release_page_dirty(page);
 	emul_write:
-		up_read(&current->mm->mmap_sem);
+		up_read(&vcpu->kvm->slots_lock);
 	}
 #endif
 
@@ -2159,10 +2163,10 @@ int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
 		kvm_x86_ops->skip_emulated_instruction(vcpu);
 
 	for (i = 0; i < nr_pages; ++i) {
-		down_read(&current->mm->mmap_sem);
+		down_read(&vcpu->kvm->slots_lock);
 		page = gva_to_page(vcpu, address + i * PAGE_SIZE);
 		vcpu->arch.pio.guest_pages[i] = page;
-		up_read(&current->mm->mmap_sem);
+		up_read(&vcpu->kvm->slots_lock);
 		if (!page) {
 			kvm_inject_gp(vcpu, 0);
 			free_pio_guest_pages(vcpu);
@@ -2485,8 +2489,9 @@ static void vapic_enter(struct kvm_vcpu *vcpu)
 
 	down_read(&current->mm->mmap_sem);
 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
-	vcpu->arch.apic->vapic_page = page;
 	up_read(&current->mm->mmap_sem);
+
+	vcpu->arch.apic->vapic_page = page;
 }
 
 static void vapic_exit(struct kvm_vcpu *vcpu)
@@ -2959,9 +2964,9 @@ int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 	gpa_t gpa;
 
 	vcpu_load(vcpu);
-	down_read(&current->mm->mmap_sem);
+	down_read(&vcpu->kvm->slots_lock);
 	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
-	up_read(&current->mm->mmap_sem);
+	up_read(&vcpu->kvm->slots_lock);
 	tr->physical_address = gpa;
 	tr->valid = gpa != UNMAPPED_GVA;
 	tr->writeable = 1;
@@ -3234,11 +3239,13 @@ int kvm_arch_set_memory_region(struct kvm *kvm,
 	 */
 	if (!user_alloc) {
 		if (npages && !old.rmap) {
+			down_write(&current->mm->mmap_sem);
 			memslot->userspace_addr = do_mmap(NULL, 0,
 						     npages * PAGE_SIZE,
 						     PROT_READ | PROT_WRITE,
 						     MAP_SHARED | MAP_ANONYMOUS,
 						     0);
+			up_write(&current->mm->mmap_sem);
 
 			if (IS_ERR((void *)memslot->userspace_addr))
 				return PTR_ERR((void *)memslot->userspace_addr);
@@ -3246,8 +3253,10 @@ int kvm_arch_set_memory_region(struct kvm *kvm,
 			if (!old.user_alloc && old.rmap) {
 				int ret;
 
+				down_write(&current->mm->mmap_sem);
 				ret = do_munmap(current->mm, old.userspace_addr,
 						old.npages * PAGE_SIZE);
+				up_write(&current->mm->mmap_sem);
 				if (ret < 0)
 					printk(KERN_WARNING
 				       "kvm_vm_ioctl_set_memory_region: "
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index ea4764b..928b0d59 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -107,6 +107,7 @@ struct kvm_memory_slot {
 struct kvm {
 	struct mutex lock; /* protects the vcpus array and APIC accesses */
 	spinlock_t mmu_lock;
+	struct rw_semaphore slots_lock;
 	struct mm_struct *mm; /* userspace tied to this vm */
 	int nmemslots;
 	struct kvm_memory_slot memslots[KVM_MEMORY_SLOTS +
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 32fbf80..b2e1289 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -169,6 +169,7 @@ static struct kvm *kvm_create_vm(void)
 	kvm_io_bus_init(&kvm->pio_bus);
 	mutex_init(&kvm->lock);
 	kvm_io_bus_init(&kvm->mmio_bus);
+	init_rwsem(&kvm->slots_lock);
 	spin_lock(&kvm_lock);
 	list_add(&kvm->vm_list, &vm_list);
 	spin_unlock(&kvm_lock);
@@ -339,9 +340,9 @@ int kvm_set_memory_region(struct kvm *kvm,
 {
 	int r;
 
-	down_write(&current->mm->mmap_sem);
+	down_write(&kvm->slots_lock);
 	r = __kvm_set_memory_region(kvm, mem, user_alloc);
-	up_write(&current->mm->mmap_sem);
+	up_write(&kvm->slots_lock);
 	return r;
 }
 EXPORT_SYMBOL_GPL(kvm_set_memory_region);