1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <kvm/iodev.h>
17 
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
23 #include <linux/mm.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
51 #include <linux/io.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
55 
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 
60 #include "coalesced_mmio.h"
61 #include "async_pf.h"
62 #include "kvm_mm.h"
63 #include "vfio.h"
64 
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
67 
68 #include <linux/kvm_dirty_ring.h>
69 
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
72 
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
75 
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
78 module_param(halt_poll_ns, uint, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns);
80 
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow = 2;
83 module_param(halt_poll_ns_grow, uint, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
85 
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start, uint, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
90 
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink;
93 module_param(halt_poll_ns_shrink, uint, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 
96 /*
97  * Ordering of locks:
98  *
99  *	kvm->lock --> kvm->slots_lock --> kvm->irq_lock
100  */
101 
102 DEFINE_MUTEX(kvm_lock);
103 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
104 LIST_HEAD(vm_list);
105 
106 static cpumask_var_t cpus_hardware_enabled;
107 static int kvm_usage_count;
108 static atomic_t hardware_enable_failed;
109 
110 static struct kmem_cache *kvm_vcpu_cache;
111 
112 static __read_mostly struct preempt_ops kvm_preempt_ops;
113 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
114 
115 struct dentry *kvm_debugfs_dir;
116 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
117 
118 static const struct file_operations stat_fops_per_vm;
119 
120 static struct file_operations kvm_chardev_ops;
121 
122 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
123 			   unsigned long arg);
124 #ifdef CONFIG_KVM_COMPAT
125 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
126 				  unsigned long arg);
127 #define KVM_COMPAT(c)	.compat_ioctl	= (c)
128 #else
129 /*
130  * For architectures that don't implement a compat infrastructure,
131  * adopt a double line of defense:
132  * - Prevent a compat task from opening /dev/kvm
133  * - If the open has been done by a 64bit task, and the KVM fd
134  *   passed to a compat task, let the ioctls fail.
135  */
kvm_no_compat_ioctl(struct file * file,unsigned int ioctl,unsigned long arg)136 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
137 				unsigned long arg) { return -EINVAL; }
138 
kvm_no_compat_open(struct inode * inode,struct file * file)139 static int kvm_no_compat_open(struct inode *inode, struct file *file)
140 {
141 	return is_compat_task() ? -ENODEV : 0;
142 }
143 #define KVM_COMPAT(c)	.compat_ioctl	= kvm_no_compat_ioctl,	\
144 			.open		= kvm_no_compat_open
145 #endif
146 static int hardware_enable_all(void);
147 static void hardware_disable_all(void);
148 
149 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
150 
151 __visible bool kvm_rebooting;
152 EXPORT_SYMBOL_GPL(kvm_rebooting);
153 
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
159 
160 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
161 
kvm_arch_mmu_notifier_invalidate_range(struct kvm * kvm,unsigned long start,unsigned long end)162 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
163 						   unsigned long start, unsigned long end)
164 {
165 }
166 
kvm_arch_guest_memory_reclaimed(struct kvm * kvm)167 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
168 {
169 }
170 
kvm_is_zone_device_page(struct page * page)171 bool kvm_is_zone_device_page(struct page *page)
172 {
173 	/*
174 	 * The metadata used by is_zone_device_page() to determine whether or
175 	 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
176 	 * the device has been pinned, e.g. by get_user_pages().  WARN if the
177 	 * page_count() is zero to help detect bad usage of this helper.
178 	 */
179 	if (WARN_ON_ONCE(!page_count(page)))
180 		return false;
181 
182 	return is_zone_device_page(page);
183 }
184 
185 /*
186  * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
187  * page, NULL otherwise.  Note, the list of refcounted PG_reserved page types
188  * is likely incomplete, it has been compiled purely through people wanting to
189  * back guest with a certain type of memory and encountering issues.
190  */
kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)191 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
192 {
193 	struct page *page;
194 
195 	if (!pfn_valid(pfn))
196 		return NULL;
197 
198 	page = pfn_to_page(pfn);
199 	if (!PageReserved(page))
200 		return page;
201 
202 	/* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
203 	if (is_zero_pfn(pfn))
204 		return page;
205 
206 	/*
207 	 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
208 	 * perspective they are "normal" pages, albeit with slightly different
209 	 * usage rules.
210 	 */
211 	if (kvm_is_zone_device_page(page))
212 		return page;
213 
214 	return NULL;
215 }
216 
217 /*
218  * Switches to specified vcpu, until a matching vcpu_put()
219  */
vcpu_load(struct kvm_vcpu * vcpu)220 void vcpu_load(struct kvm_vcpu *vcpu)
221 {
222 	int cpu = get_cpu();
223 
224 	__this_cpu_write(kvm_running_vcpu, vcpu);
225 	preempt_notifier_register(&vcpu->preempt_notifier);
226 	kvm_arch_vcpu_load(vcpu, cpu);
227 	put_cpu();
228 }
229 EXPORT_SYMBOL_GPL(vcpu_load);
230 
vcpu_put(struct kvm_vcpu * vcpu)231 void vcpu_put(struct kvm_vcpu *vcpu)
232 {
233 	preempt_disable();
234 	kvm_arch_vcpu_put(vcpu);
235 	preempt_notifier_unregister(&vcpu->preempt_notifier);
236 	__this_cpu_write(kvm_running_vcpu, NULL);
237 	preempt_enable();
238 }
239 EXPORT_SYMBOL_GPL(vcpu_put);
240 
241 /* TODO: merge with kvm_arch_vcpu_should_kick */
kvm_request_needs_ipi(struct kvm_vcpu * vcpu,unsigned req)242 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
243 {
244 	int mode = kvm_vcpu_exiting_guest_mode(vcpu);
245 
246 	/*
247 	 * We need to wait for the VCPU to reenable interrupts and get out of
248 	 * READING_SHADOW_PAGE_TABLES mode.
249 	 */
250 	if (req & KVM_REQUEST_WAIT)
251 		return mode != OUTSIDE_GUEST_MODE;
252 
253 	/*
254 	 * Need to kick a running VCPU, but otherwise there is nothing to do.
255 	 */
256 	return mode == IN_GUEST_MODE;
257 }
258 
ack_kick(void * _completed)259 static void ack_kick(void *_completed)
260 {
261 }
262 
kvm_kick_many_cpus(struct cpumask * cpus,bool wait)263 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
264 {
265 	if (cpumask_empty(cpus))
266 		return false;
267 
268 	smp_call_function_many(cpus, ack_kick, NULL, wait);
269 	return true;
270 }
271 
kvm_make_vcpu_request(struct kvm_vcpu * vcpu,unsigned int req,struct cpumask * tmp,int current_cpu)272 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
273 				  struct cpumask *tmp, int current_cpu)
274 {
275 	int cpu;
276 
277 	if (likely(!(req & KVM_REQUEST_NO_ACTION)))
278 		__kvm_make_request(req, vcpu);
279 
280 	if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 		return;
282 
283 	/*
284 	 * Note, the vCPU could get migrated to a different pCPU at any point
285 	 * after kvm_request_needs_ipi(), which could result in sending an IPI
286 	 * to the previous pCPU.  But, that's OK because the purpose of the IPI
287 	 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
288 	 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
289 	 * after this point is also OK, as the requirement is only that KVM wait
290 	 * for vCPUs that were reading SPTEs _before_ any changes were
291 	 * finalized. See kvm_vcpu_kick() for more details on handling requests.
292 	 */
293 	if (kvm_request_needs_ipi(vcpu, req)) {
294 		cpu = READ_ONCE(vcpu->cpu);
295 		if (cpu != -1 && cpu != current_cpu)
296 			__cpumask_set_cpu(cpu, tmp);
297 	}
298 }
299 
kvm_make_vcpus_request_mask(struct kvm * kvm,unsigned int req,unsigned long * vcpu_bitmap)300 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
301 				 unsigned long *vcpu_bitmap)
302 {
303 	struct kvm_vcpu *vcpu;
304 	struct cpumask *cpus;
305 	int i, me;
306 	bool called;
307 
308 	me = get_cpu();
309 
310 	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
311 	cpumask_clear(cpus);
312 
313 	for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
314 		vcpu = kvm_get_vcpu(kvm, i);
315 		if (!vcpu)
316 			continue;
317 		kvm_make_vcpu_request(vcpu, req, cpus, me);
318 	}
319 
320 	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
321 	put_cpu();
322 
323 	return called;
324 }
325 
kvm_make_all_cpus_request_except(struct kvm * kvm,unsigned int req,struct kvm_vcpu * except)326 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
327 				      struct kvm_vcpu *except)
328 {
329 	struct kvm_vcpu *vcpu;
330 	struct cpumask *cpus;
331 	unsigned long i;
332 	bool called;
333 	int me;
334 
335 	me = get_cpu();
336 
337 	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
338 	cpumask_clear(cpus);
339 
340 	kvm_for_each_vcpu(i, vcpu, kvm) {
341 		if (vcpu == except)
342 			continue;
343 		kvm_make_vcpu_request(vcpu, req, cpus, me);
344 	}
345 
346 	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
347 	put_cpu();
348 
349 	return called;
350 }
351 
kvm_make_all_cpus_request(struct kvm * kvm,unsigned int req)352 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
353 {
354 	return kvm_make_all_cpus_request_except(kvm, req, NULL);
355 }
356 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
357 
358 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
kvm_flush_remote_tlbs(struct kvm * kvm)359 void kvm_flush_remote_tlbs(struct kvm *kvm)
360 {
361 	++kvm->stat.generic.remote_tlb_flush_requests;
362 
363 	/*
364 	 * We want to publish modifications to the page tables before reading
365 	 * mode. Pairs with a memory barrier in arch-specific code.
366 	 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
367 	 * and smp_mb in walk_shadow_page_lockless_begin/end.
368 	 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
369 	 *
370 	 * There is already an smp_mb__after_atomic() before
371 	 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
372 	 * barrier here.
373 	 */
374 	if (!kvm_arch_flush_remote_tlb(kvm)
375 	    || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
376 		++kvm->stat.generic.remote_tlb_flush;
377 }
378 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
379 #endif
380 
kvm_flush_shadow_all(struct kvm * kvm)381 static void kvm_flush_shadow_all(struct kvm *kvm)
382 {
383 	kvm_arch_flush_shadow_all(kvm);
384 	kvm_arch_guest_memory_reclaimed(kvm);
385 }
386 
387 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache * mc,gfp_t gfp_flags)388 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
389 					       gfp_t gfp_flags)
390 {
391 	gfp_flags |= mc->gfp_zero;
392 
393 	if (mc->kmem_cache)
394 		return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
395 	else
396 		return (void *)__get_free_page(gfp_flags);
397 }
398 
__kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache * mc,int capacity,int min)399 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
400 {
401 	gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
402 	void *obj;
403 
404 	if (mc->nobjs >= min)
405 		return 0;
406 
407 	if (unlikely(!mc->objects)) {
408 		if (WARN_ON_ONCE(!capacity))
409 			return -EIO;
410 
411 		mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
412 		if (!mc->objects)
413 			return -ENOMEM;
414 
415 		mc->capacity = capacity;
416 	}
417 
418 	/* It is illegal to request a different capacity across topups. */
419 	if (WARN_ON_ONCE(mc->capacity != capacity))
420 		return -EIO;
421 
422 	while (mc->nobjs < mc->capacity) {
423 		obj = mmu_memory_cache_alloc_obj(mc, gfp);
424 		if (!obj)
425 			return mc->nobjs >= min ? 0 : -ENOMEM;
426 		mc->objects[mc->nobjs++] = obj;
427 	}
428 	return 0;
429 }
430 
kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache * mc,int min)431 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
432 {
433 	return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
434 }
435 
kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache * mc)436 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
437 {
438 	return mc->nobjs;
439 }
440 
kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache * mc)441 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
442 {
443 	while (mc->nobjs) {
444 		if (mc->kmem_cache)
445 			kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
446 		else
447 			free_page((unsigned long)mc->objects[--mc->nobjs]);
448 	}
449 
450 	kvfree(mc->objects);
451 
452 	mc->objects = NULL;
453 	mc->capacity = 0;
454 }
455 
kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache * mc)456 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
457 {
458 	void *p;
459 
460 	if (WARN_ON(!mc->nobjs))
461 		p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
462 	else
463 		p = mc->objects[--mc->nobjs];
464 	BUG_ON(!p);
465 	return p;
466 }
467 #endif
468 
kvm_vcpu_init(struct kvm_vcpu * vcpu,struct kvm * kvm,unsigned id)469 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
470 {
471 	mutex_init(&vcpu->mutex);
472 	vcpu->cpu = -1;
473 	vcpu->kvm = kvm;
474 	vcpu->vcpu_id = id;
475 	vcpu->pid = NULL;
476 #ifndef __KVM_HAVE_ARCH_WQP
477 	rcuwait_init(&vcpu->wait);
478 #endif
479 	kvm_async_pf_vcpu_init(vcpu);
480 
481 	kvm_vcpu_set_in_spin_loop(vcpu, false);
482 	kvm_vcpu_set_dy_eligible(vcpu, false);
483 	vcpu->preempted = false;
484 	vcpu->ready = false;
485 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
486 	vcpu->last_used_slot = NULL;
487 
488 	/* Fill the stats id string for the vcpu */
489 	snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
490 		 task_pid_nr(current), id);
491 }
492 
kvm_vcpu_destroy(struct kvm_vcpu * vcpu)493 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
494 {
495 	kvm_arch_vcpu_destroy(vcpu);
496 	kvm_dirty_ring_free(&vcpu->dirty_ring);
497 
498 	/*
499 	 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
500 	 * the vcpu->pid pointer, and at destruction time all file descriptors
501 	 * are already gone.
502 	 */
503 	put_pid(rcu_dereference_protected(vcpu->pid, 1));
504 
505 	free_page((unsigned long)vcpu->run);
506 	kmem_cache_free(kvm_vcpu_cache, vcpu);
507 }
508 
kvm_destroy_vcpus(struct kvm * kvm)509 void kvm_destroy_vcpus(struct kvm *kvm)
510 {
511 	unsigned long i;
512 	struct kvm_vcpu *vcpu;
513 
514 	kvm_for_each_vcpu(i, vcpu, kvm) {
515 		kvm_vcpu_destroy(vcpu);
516 		xa_erase(&kvm->vcpu_array, i);
517 	}
518 
519 	atomic_set(&kvm->online_vcpus, 0);
520 }
521 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
522 
523 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
mmu_notifier_to_kvm(struct mmu_notifier * mn)524 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
525 {
526 	return container_of(mn, struct kvm, mmu_notifier);
527 }
528 
kvm_mmu_notifier_invalidate_range(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long start,unsigned long end)529 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
530 					      struct mm_struct *mm,
531 					      unsigned long start, unsigned long end)
532 {
533 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
534 	int idx;
535 
536 	idx = srcu_read_lock(&kvm->srcu);
537 	kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
538 	srcu_read_unlock(&kvm->srcu, idx);
539 }
540 
541 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
542 
543 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
544 			     unsigned long end);
545 
546 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
547 
548 struct kvm_hva_range {
549 	unsigned long start;
550 	unsigned long end;
551 	pte_t pte;
552 	hva_handler_t handler;
553 	on_lock_fn_t on_lock;
554 	on_unlock_fn_t on_unlock;
555 	bool flush_on_ret;
556 	bool may_block;
557 };
558 
559 /*
560  * Use a dedicated stub instead of NULL to indicate that there is no callback
561  * function/handler.  The compiler technically can't guarantee that a real
562  * function will have a non-zero address, and so it will generate code to
563  * check for !NULL, whereas comparing against a stub will be elided at compile
564  * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
565  */
kvm_null_fn(void)566 static void kvm_null_fn(void)
567 {
568 
569 }
570 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
571 
572 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
573 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last)	     \
574 	for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
575 	     node;							     \
576 	     node = interval_tree_iter_next(node, start, last))	     \
577 
__kvm_handle_hva_range(struct kvm * kvm,const struct kvm_hva_range * range)578 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
579 						  const struct kvm_hva_range *range)
580 {
581 	bool ret = false, locked = false;
582 	struct kvm_gfn_range gfn_range;
583 	struct kvm_memory_slot *slot;
584 	struct kvm_memslots *slots;
585 	int i, idx;
586 
587 	if (WARN_ON_ONCE(range->end <= range->start))
588 		return 0;
589 
590 	/* A null handler is allowed if and only if on_lock() is provided. */
591 	if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
592 			 IS_KVM_NULL_FN(range->handler)))
593 		return 0;
594 
595 	idx = srcu_read_lock(&kvm->srcu);
596 
597 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
598 		struct interval_tree_node *node;
599 
600 		slots = __kvm_memslots(kvm, i);
601 		kvm_for_each_memslot_in_hva_range(node, slots,
602 						  range->start, range->end - 1) {
603 			unsigned long hva_start, hva_end;
604 
605 			slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
606 			hva_start = max(range->start, slot->userspace_addr);
607 			hva_end = min(range->end, slot->userspace_addr +
608 						  (slot->npages << PAGE_SHIFT));
609 
610 			/*
611 			 * To optimize for the likely case where the address
612 			 * range is covered by zero or one memslots, don't
613 			 * bother making these conditional (to avoid writes on
614 			 * the second or later invocation of the handler).
615 			 */
616 			gfn_range.pte = range->pte;
617 			gfn_range.may_block = range->may_block;
618 
619 			/*
620 			 * {gfn(page) | page intersects with [hva_start, hva_end)} =
621 			 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
622 			 */
623 			gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
624 			gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
625 			gfn_range.slot = slot;
626 
627 			if (!locked) {
628 				locked = true;
629 				KVM_MMU_LOCK(kvm);
630 				if (!IS_KVM_NULL_FN(range->on_lock))
631 					range->on_lock(kvm, range->start, range->end);
632 				if (IS_KVM_NULL_FN(range->handler))
633 					break;
634 			}
635 			ret |= range->handler(kvm, &gfn_range);
636 		}
637 	}
638 
639 	if (range->flush_on_ret && ret)
640 		kvm_flush_remote_tlbs(kvm);
641 
642 	if (locked) {
643 		KVM_MMU_UNLOCK(kvm);
644 		if (!IS_KVM_NULL_FN(range->on_unlock))
645 			range->on_unlock(kvm);
646 	}
647 
648 	srcu_read_unlock(&kvm->srcu, idx);
649 
650 	/* The notifiers are averse to booleans. :-( */
651 	return (int)ret;
652 }
653 
kvm_handle_hva_range(struct mmu_notifier * mn,unsigned long start,unsigned long end,pte_t pte,hva_handler_t handler)654 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
655 						unsigned long start,
656 						unsigned long end,
657 						pte_t pte,
658 						hva_handler_t handler)
659 {
660 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
661 	const struct kvm_hva_range range = {
662 		.start		= start,
663 		.end		= end,
664 		.pte		= pte,
665 		.handler	= handler,
666 		.on_lock	= (void *)kvm_null_fn,
667 		.on_unlock	= (void *)kvm_null_fn,
668 		.flush_on_ret	= true,
669 		.may_block	= false,
670 	};
671 
672 	return __kvm_handle_hva_range(kvm, &range);
673 }
674 
kvm_handle_hva_range_no_flush(struct mmu_notifier * mn,unsigned long start,unsigned long end,hva_handler_t handler)675 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
676 							 unsigned long start,
677 							 unsigned long end,
678 							 hva_handler_t handler)
679 {
680 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
681 	const struct kvm_hva_range range = {
682 		.start		= start,
683 		.end		= end,
684 		.pte		= __pte(0),
685 		.handler	= handler,
686 		.on_lock	= (void *)kvm_null_fn,
687 		.on_unlock	= (void *)kvm_null_fn,
688 		.flush_on_ret	= false,
689 		.may_block	= false,
690 	};
691 
692 	return __kvm_handle_hva_range(kvm, &range);
693 }
kvm_mmu_notifier_change_pte(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long address,pte_t pte)694 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
695 					struct mm_struct *mm,
696 					unsigned long address,
697 					pte_t pte)
698 {
699 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
700 
701 	trace_kvm_set_spte_hva(address);
702 
703 	/*
704 	 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
705 	 * If mmu_invalidate_in_progress is zero, then no in-progress
706 	 * invalidations, including this one, found a relevant memslot at
707 	 * start(); rechecking memslots here is unnecessary.  Note, a false
708 	 * positive (count elevated by a different invalidation) is sub-optimal
709 	 * but functionally ok.
710 	 */
711 	WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
712 	if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
713 		return;
714 
715 	kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
716 }
717 
kvm_mmu_invalidate_begin(struct kvm * kvm,unsigned long start,unsigned long end)718 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
719 			      unsigned long end)
720 {
721 	/*
722 	 * The count increase must become visible at unlock time as no
723 	 * spte can be established without taking the mmu_lock and
724 	 * count is also read inside the mmu_lock critical section.
725 	 */
726 	kvm->mmu_invalidate_in_progress++;
727 	if (likely(kvm->mmu_invalidate_in_progress == 1)) {
728 		kvm->mmu_invalidate_range_start = start;
729 		kvm->mmu_invalidate_range_end = end;
730 	} else {
731 		/*
732 		 * Fully tracking multiple concurrent ranges has diminishing
733 		 * returns. Keep things simple and just find the minimal range
734 		 * which includes the current and new ranges. As there won't be
735 		 * enough information to subtract a range after its invalidate
736 		 * completes, any ranges invalidated concurrently will
737 		 * accumulate and persist until all outstanding invalidates
738 		 * complete.
739 		 */
740 		kvm->mmu_invalidate_range_start =
741 			min(kvm->mmu_invalidate_range_start, start);
742 		kvm->mmu_invalidate_range_end =
743 			max(kvm->mmu_invalidate_range_end, end);
744 	}
745 }
746 
kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier * mn,const struct mmu_notifier_range * range)747 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
748 					const struct mmu_notifier_range *range)
749 {
750 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
751 	const struct kvm_hva_range hva_range = {
752 		.start		= range->start,
753 		.end		= range->end,
754 		.pte		= __pte(0),
755 		.handler	= kvm_unmap_gfn_range,
756 		.on_lock	= kvm_mmu_invalidate_begin,
757 		.on_unlock	= kvm_arch_guest_memory_reclaimed,
758 		.flush_on_ret	= true,
759 		.may_block	= mmu_notifier_range_blockable(range),
760 	};
761 
762 	trace_kvm_unmap_hva_range(range->start, range->end);
763 
764 	/*
765 	 * Prevent memslot modification between range_start() and range_end()
766 	 * so that conditionally locking provides the same result in both
767 	 * functions.  Without that guarantee, the mmu_invalidate_in_progress
768 	 * adjustments will be imbalanced.
769 	 *
770 	 * Pairs with the decrement in range_end().
771 	 */
772 	spin_lock(&kvm->mn_invalidate_lock);
773 	kvm->mn_active_invalidate_count++;
774 	spin_unlock(&kvm->mn_invalidate_lock);
775 
776 	/*
777 	 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
778 	 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
779 	 * each cache's lock.  There are relatively few caches in existence at
780 	 * any given time, and the caches themselves can check for hva overlap,
781 	 * i.e. don't need to rely on memslot overlap checks for performance.
782 	 * Because this runs without holding mmu_lock, the pfn caches must use
783 	 * mn_active_invalidate_count (see above) instead of
784 	 * mmu_invalidate_in_progress.
785 	 */
786 	gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
787 					  hva_range.may_block);
788 
789 	__kvm_handle_hva_range(kvm, &hva_range);
790 
791 	return 0;
792 }
793 
kvm_mmu_invalidate_end(struct kvm * kvm,unsigned long start,unsigned long end)794 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
795 			    unsigned long end)
796 {
797 	/*
798 	 * This sequence increase will notify the kvm page fault that
799 	 * the page that is going to be mapped in the spte could have
800 	 * been freed.
801 	 */
802 	kvm->mmu_invalidate_seq++;
803 	smp_wmb();
804 	/*
805 	 * The above sequence increase must be visible before the
806 	 * below count decrease, which is ensured by the smp_wmb above
807 	 * in conjunction with the smp_rmb in mmu_invalidate_retry().
808 	 */
809 	kvm->mmu_invalidate_in_progress--;
810 }
811 
kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier * mn,const struct mmu_notifier_range * range)812 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
813 					const struct mmu_notifier_range *range)
814 {
815 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
816 	const struct kvm_hva_range hva_range = {
817 		.start		= range->start,
818 		.end		= range->end,
819 		.pte		= __pte(0),
820 		.handler	= (void *)kvm_null_fn,
821 		.on_lock	= kvm_mmu_invalidate_end,
822 		.on_unlock	= (void *)kvm_null_fn,
823 		.flush_on_ret	= false,
824 		.may_block	= mmu_notifier_range_blockable(range),
825 	};
826 	bool wake;
827 
828 	__kvm_handle_hva_range(kvm, &hva_range);
829 
830 	/* Pairs with the increment in range_start(). */
831 	spin_lock(&kvm->mn_invalidate_lock);
832 	wake = (--kvm->mn_active_invalidate_count == 0);
833 	spin_unlock(&kvm->mn_invalidate_lock);
834 
835 	/*
836 	 * There can only be one waiter, since the wait happens under
837 	 * slots_lock.
838 	 */
839 	if (wake)
840 		rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
841 
842 	BUG_ON(kvm->mmu_invalidate_in_progress < 0);
843 }
844 
kvm_mmu_notifier_clear_flush_young(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long start,unsigned long end)845 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
846 					      struct mm_struct *mm,
847 					      unsigned long start,
848 					      unsigned long end)
849 {
850 	trace_kvm_age_hva(start, end);
851 
852 	return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
853 }
854 
kvm_mmu_notifier_clear_young(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long start,unsigned long end)855 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
856 					struct mm_struct *mm,
857 					unsigned long start,
858 					unsigned long end)
859 {
860 	trace_kvm_age_hva(start, end);
861 
862 	/*
863 	 * Even though we do not flush TLB, this will still adversely
864 	 * affect performance on pre-Haswell Intel EPT, where there is
865 	 * no EPT Access Bit to clear so that we have to tear down EPT
866 	 * tables instead. If we find this unacceptable, we can always
867 	 * add a parameter to kvm_age_hva so that it effectively doesn't
868 	 * do anything on clear_young.
869 	 *
870 	 * Also note that currently we never issue secondary TLB flushes
871 	 * from clear_young, leaving this job up to the regular system
872 	 * cadence. If we find this inaccurate, we might come up with a
873 	 * more sophisticated heuristic later.
874 	 */
875 	return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
876 }
877 
kvm_mmu_notifier_test_young(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long address)878 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
879 				       struct mm_struct *mm,
880 				       unsigned long address)
881 {
882 	trace_kvm_test_age_hva(address);
883 
884 	return kvm_handle_hva_range_no_flush(mn, address, address + 1,
885 					     kvm_test_age_gfn);
886 }
887 
kvm_mmu_notifier_release(struct mmu_notifier * mn,struct mm_struct * mm)888 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
889 				     struct mm_struct *mm)
890 {
891 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
892 	int idx;
893 
894 	idx = srcu_read_lock(&kvm->srcu);
895 	kvm_flush_shadow_all(kvm);
896 	srcu_read_unlock(&kvm->srcu, idx);
897 }
898 
899 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
900 	.invalidate_range	= kvm_mmu_notifier_invalidate_range,
901 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
902 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
903 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
904 	.clear_young		= kvm_mmu_notifier_clear_young,
905 	.test_young		= kvm_mmu_notifier_test_young,
906 	.change_pte		= kvm_mmu_notifier_change_pte,
907 	.release		= kvm_mmu_notifier_release,
908 };
909 
kvm_init_mmu_notifier(struct kvm * kvm)910 static int kvm_init_mmu_notifier(struct kvm *kvm)
911 {
912 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
913 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
914 }
915 
916 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
917 
kvm_init_mmu_notifier(struct kvm * kvm)918 static int kvm_init_mmu_notifier(struct kvm *kvm)
919 {
920 	return 0;
921 }
922 
923 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
924 
925 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
kvm_pm_notifier_call(struct notifier_block * bl,unsigned long state,void * unused)926 static int kvm_pm_notifier_call(struct notifier_block *bl,
927 				unsigned long state,
928 				void *unused)
929 {
930 	struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
931 
932 	return kvm_arch_pm_notifier(kvm, state);
933 }
934 
kvm_init_pm_notifier(struct kvm * kvm)935 static void kvm_init_pm_notifier(struct kvm *kvm)
936 {
937 	kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
938 	/* Suspend KVM before we suspend ftrace, RCU, etc. */
939 	kvm->pm_notifier.priority = INT_MAX;
940 	register_pm_notifier(&kvm->pm_notifier);
941 }
942 
kvm_destroy_pm_notifier(struct kvm * kvm)943 static void kvm_destroy_pm_notifier(struct kvm *kvm)
944 {
945 	unregister_pm_notifier(&kvm->pm_notifier);
946 }
947 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
kvm_init_pm_notifier(struct kvm * kvm)948 static void kvm_init_pm_notifier(struct kvm *kvm)
949 {
950 }
951 
kvm_destroy_pm_notifier(struct kvm * kvm)952 static void kvm_destroy_pm_notifier(struct kvm *kvm)
953 {
954 }
955 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
956 
kvm_destroy_dirty_bitmap(struct kvm_memory_slot * memslot)957 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
958 {
959 	if (!memslot->dirty_bitmap)
960 		return;
961 
962 	kvfree(memslot->dirty_bitmap);
963 	memslot->dirty_bitmap = NULL;
964 }
965 
966 /* This does not remove the slot from struct kvm_memslots data structures */
kvm_free_memslot(struct kvm * kvm,struct kvm_memory_slot * slot)967 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
968 {
969 	kvm_destroy_dirty_bitmap(slot);
970 
971 	kvm_arch_free_memslot(kvm, slot);
972 
973 	kfree(slot);
974 }
975 
kvm_free_memslots(struct kvm * kvm,struct kvm_memslots * slots)976 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
977 {
978 	struct hlist_node *idnode;
979 	struct kvm_memory_slot *memslot;
980 	int bkt;
981 
982 	/*
983 	 * The same memslot objects live in both active and inactive sets,
984 	 * arbitrarily free using index '1' so the second invocation of this
985 	 * function isn't operating over a structure with dangling pointers
986 	 * (even though this function isn't actually touching them).
987 	 */
988 	if (!slots->node_idx)
989 		return;
990 
991 	hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
992 		kvm_free_memslot(kvm, memslot);
993 }
994 
kvm_stats_debugfs_mode(const struct _kvm_stats_desc * pdesc)995 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
996 {
997 	switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
998 	case KVM_STATS_TYPE_INSTANT:
999 		return 0444;
1000 	case KVM_STATS_TYPE_CUMULATIVE:
1001 	case KVM_STATS_TYPE_PEAK:
1002 	default:
1003 		return 0644;
1004 	}
1005 }
1006 
1007 
kvm_destroy_vm_debugfs(struct kvm * kvm)1008 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1009 {
1010 	int i;
1011 	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1012 				      kvm_vcpu_stats_header.num_desc;
1013 
1014 	if (IS_ERR(kvm->debugfs_dentry))
1015 		return;
1016 
1017 	debugfs_remove_recursive(kvm->debugfs_dentry);
1018 
1019 	if (kvm->debugfs_stat_data) {
1020 		for (i = 0; i < kvm_debugfs_num_entries; i++)
1021 			kfree(kvm->debugfs_stat_data[i]);
1022 		kfree(kvm->debugfs_stat_data);
1023 	}
1024 }
1025 
kvm_create_vm_debugfs(struct kvm * kvm,const char * fdname)1026 static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1027 {
1028 	static DEFINE_MUTEX(kvm_debugfs_lock);
1029 	struct dentry *dent;
1030 	char dir_name[ITOA_MAX_LEN * 2];
1031 	struct kvm_stat_data *stat_data;
1032 	const struct _kvm_stats_desc *pdesc;
1033 	int i, ret = -ENOMEM;
1034 	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1035 				      kvm_vcpu_stats_header.num_desc;
1036 
1037 	if (!debugfs_initialized())
1038 		return 0;
1039 
1040 	snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1041 	mutex_lock(&kvm_debugfs_lock);
1042 	dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1043 	if (dent) {
1044 		pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1045 		dput(dent);
1046 		mutex_unlock(&kvm_debugfs_lock);
1047 		return 0;
1048 	}
1049 	dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1050 	mutex_unlock(&kvm_debugfs_lock);
1051 	if (IS_ERR(dent))
1052 		return 0;
1053 
1054 	kvm->debugfs_dentry = dent;
1055 	kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1056 					 sizeof(*kvm->debugfs_stat_data),
1057 					 GFP_KERNEL_ACCOUNT);
1058 	if (!kvm->debugfs_stat_data)
1059 		goto out_err;
1060 
1061 	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1062 		pdesc = &kvm_vm_stats_desc[i];
1063 		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1064 		if (!stat_data)
1065 			goto out_err;
1066 
1067 		stat_data->kvm = kvm;
1068 		stat_data->desc = pdesc;
1069 		stat_data->kind = KVM_STAT_VM;
1070 		kvm->debugfs_stat_data[i] = stat_data;
1071 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1072 				    kvm->debugfs_dentry, stat_data,
1073 				    &stat_fops_per_vm);
1074 	}
1075 
1076 	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1077 		pdesc = &kvm_vcpu_stats_desc[i];
1078 		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1079 		if (!stat_data)
1080 			goto out_err;
1081 
1082 		stat_data->kvm = kvm;
1083 		stat_data->desc = pdesc;
1084 		stat_data->kind = KVM_STAT_VCPU;
1085 		kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1086 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1087 				    kvm->debugfs_dentry, stat_data,
1088 				    &stat_fops_per_vm);
1089 	}
1090 
1091 	ret = kvm_arch_create_vm_debugfs(kvm);
1092 	if (ret)
1093 		goto out_err;
1094 
1095 	return 0;
1096 out_err:
1097 	kvm_destroy_vm_debugfs(kvm);
1098 	return ret;
1099 }
1100 
1101 /*
1102  * Called after the VM is otherwise initialized, but just before adding it to
1103  * the vm_list.
1104  */
kvm_arch_post_init_vm(struct kvm * kvm)1105 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1106 {
1107 	return 0;
1108 }
1109 
1110 /*
1111  * Called just after removing the VM from the vm_list, but before doing any
1112  * other destruction.
1113  */
kvm_arch_pre_destroy_vm(struct kvm * kvm)1114 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1115 {
1116 }
1117 
1118 /*
1119  * Called after per-vm debugfs created.  When called kvm->debugfs_dentry should
1120  * be setup already, so we can create arch-specific debugfs entries under it.
1121  * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1122  * a per-arch destroy interface is not needed.
1123  */
kvm_arch_create_vm_debugfs(struct kvm * kvm)1124 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1125 {
1126 	return 0;
1127 }
1128 
kvm_create_vm(unsigned long type,const char * fdname)1129 static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1130 {
1131 	struct kvm *kvm = kvm_arch_alloc_vm();
1132 	struct kvm_memslots *slots;
1133 	int r = -ENOMEM;
1134 	int i, j;
1135 
1136 	if (!kvm)
1137 		return ERR_PTR(-ENOMEM);
1138 
1139 	/* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1140 	__module_get(kvm_chardev_ops.owner);
1141 
1142 	KVM_MMU_LOCK_INIT(kvm);
1143 	mmgrab(current->mm);
1144 	kvm->mm = current->mm;
1145 	kvm_eventfd_init(kvm);
1146 	mutex_init(&kvm->lock);
1147 	mutex_init(&kvm->irq_lock);
1148 	mutex_init(&kvm->slots_lock);
1149 	mutex_init(&kvm->slots_arch_lock);
1150 	spin_lock_init(&kvm->mn_invalidate_lock);
1151 	rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1152 	xa_init(&kvm->vcpu_array);
1153 
1154 	INIT_LIST_HEAD(&kvm->gpc_list);
1155 	spin_lock_init(&kvm->gpc_lock);
1156 
1157 	INIT_LIST_HEAD(&kvm->devices);
1158 	kvm->max_vcpus = KVM_MAX_VCPUS;
1159 
1160 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1161 
1162 	/*
1163 	 * Force subsequent debugfs file creations to fail if the VM directory
1164 	 * is not created (by kvm_create_vm_debugfs()).
1165 	 */
1166 	kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1167 
1168 	snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1169 		 task_pid_nr(current));
1170 
1171 	if (init_srcu_struct(&kvm->srcu))
1172 		goto out_err_no_srcu;
1173 	if (init_srcu_struct(&kvm->irq_srcu))
1174 		goto out_err_no_irq_srcu;
1175 
1176 	refcount_set(&kvm->users_count, 1);
1177 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1178 		for (j = 0; j < 2; j++) {
1179 			slots = &kvm->__memslots[i][j];
1180 
1181 			atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1182 			slots->hva_tree = RB_ROOT_CACHED;
1183 			slots->gfn_tree = RB_ROOT;
1184 			hash_init(slots->id_hash);
1185 			slots->node_idx = j;
1186 
1187 			/* Generations must be different for each address space. */
1188 			slots->generation = i;
1189 		}
1190 
1191 		rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1192 	}
1193 
1194 	for (i = 0; i < KVM_NR_BUSES; i++) {
1195 		rcu_assign_pointer(kvm->buses[i],
1196 			kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1197 		if (!kvm->buses[i])
1198 			goto out_err_no_arch_destroy_vm;
1199 	}
1200 
1201 	r = kvm_arch_init_vm(kvm, type);
1202 	if (r)
1203 		goto out_err_no_arch_destroy_vm;
1204 
1205 	r = hardware_enable_all();
1206 	if (r)
1207 		goto out_err_no_disable;
1208 
1209 #ifdef CONFIG_HAVE_KVM_IRQFD
1210 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1211 #endif
1212 
1213 	r = kvm_init_mmu_notifier(kvm);
1214 	if (r)
1215 		goto out_err_no_mmu_notifier;
1216 
1217 	r = kvm_coalesced_mmio_init(kvm);
1218 	if (r < 0)
1219 		goto out_no_coalesced_mmio;
1220 
1221 	r = kvm_create_vm_debugfs(kvm, fdname);
1222 	if (r)
1223 		goto out_err_no_debugfs;
1224 
1225 	r = kvm_arch_post_init_vm(kvm);
1226 	if (r)
1227 		goto out_err;
1228 
1229 	mutex_lock(&kvm_lock);
1230 	list_add(&kvm->vm_list, &vm_list);
1231 	mutex_unlock(&kvm_lock);
1232 
1233 	preempt_notifier_inc();
1234 	kvm_init_pm_notifier(kvm);
1235 
1236 	return kvm;
1237 
1238 out_err:
1239 	kvm_destroy_vm_debugfs(kvm);
1240 out_err_no_debugfs:
1241 	kvm_coalesced_mmio_free(kvm);
1242 out_no_coalesced_mmio:
1243 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1244 	if (kvm->mmu_notifier.ops)
1245 		mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1246 #endif
1247 out_err_no_mmu_notifier:
1248 	hardware_disable_all();
1249 out_err_no_disable:
1250 	kvm_arch_destroy_vm(kvm);
1251 out_err_no_arch_destroy_vm:
1252 	WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1253 	for (i = 0; i < KVM_NR_BUSES; i++)
1254 		kfree(kvm_get_bus(kvm, i));
1255 	cleanup_srcu_struct(&kvm->irq_srcu);
1256 out_err_no_irq_srcu:
1257 	cleanup_srcu_struct(&kvm->srcu);
1258 out_err_no_srcu:
1259 	kvm_arch_free_vm(kvm);
1260 	mmdrop(current->mm);
1261 	module_put(kvm_chardev_ops.owner);
1262 	return ERR_PTR(r);
1263 }
1264 
kvm_destroy_devices(struct kvm * kvm)1265 static void kvm_destroy_devices(struct kvm *kvm)
1266 {
1267 	struct kvm_device *dev, *tmp;
1268 
1269 	/*
1270 	 * We do not need to take the kvm->lock here, because nobody else
1271 	 * has a reference to the struct kvm at this point and therefore
1272 	 * cannot access the devices list anyhow.
1273 	 */
1274 	list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1275 		list_del(&dev->vm_node);
1276 		dev->ops->destroy(dev);
1277 	}
1278 }
1279 
kvm_destroy_vm(struct kvm * kvm)1280 static void kvm_destroy_vm(struct kvm *kvm)
1281 {
1282 	int i;
1283 	struct mm_struct *mm = kvm->mm;
1284 
1285 	kvm_destroy_pm_notifier(kvm);
1286 	kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1287 	kvm_destroy_vm_debugfs(kvm);
1288 	kvm_arch_sync_events(kvm);
1289 	mutex_lock(&kvm_lock);
1290 	list_del(&kvm->vm_list);
1291 	mutex_unlock(&kvm_lock);
1292 	kvm_arch_pre_destroy_vm(kvm);
1293 
1294 	kvm_free_irq_routing(kvm);
1295 	for (i = 0; i < KVM_NR_BUSES; i++) {
1296 		struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1297 
1298 		if (bus)
1299 			kvm_io_bus_destroy(bus);
1300 		kvm->buses[i] = NULL;
1301 	}
1302 	kvm_coalesced_mmio_free(kvm);
1303 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1304 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1305 	/*
1306 	 * At this point, pending calls to invalidate_range_start()
1307 	 * have completed but no more MMU notifiers will run, so
1308 	 * mn_active_invalidate_count may remain unbalanced.
1309 	 * No threads can be waiting in install_new_memslots as the
1310 	 * last reference on KVM has been dropped, but freeing
1311 	 * memslots would deadlock without this manual intervention.
1312 	 */
1313 	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1314 	kvm->mn_active_invalidate_count = 0;
1315 #else
1316 	kvm_flush_shadow_all(kvm);
1317 #endif
1318 	kvm_arch_destroy_vm(kvm);
1319 	kvm_destroy_devices(kvm);
1320 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1321 		kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1322 		kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1323 	}
1324 	cleanup_srcu_struct(&kvm->irq_srcu);
1325 	cleanup_srcu_struct(&kvm->srcu);
1326 	kvm_arch_free_vm(kvm);
1327 	preempt_notifier_dec();
1328 	hardware_disable_all();
1329 	mmdrop(mm);
1330 	module_put(kvm_chardev_ops.owner);
1331 }
1332 
kvm_get_kvm(struct kvm * kvm)1333 void kvm_get_kvm(struct kvm *kvm)
1334 {
1335 	refcount_inc(&kvm->users_count);
1336 }
1337 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1338 
1339 /*
1340  * Make sure the vm is not during destruction, which is a safe version of
1341  * kvm_get_kvm().  Return true if kvm referenced successfully, false otherwise.
1342  */
kvm_get_kvm_safe(struct kvm * kvm)1343 bool kvm_get_kvm_safe(struct kvm *kvm)
1344 {
1345 	return refcount_inc_not_zero(&kvm->users_count);
1346 }
1347 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1348 
kvm_put_kvm(struct kvm * kvm)1349 void kvm_put_kvm(struct kvm *kvm)
1350 {
1351 	if (refcount_dec_and_test(&kvm->users_count))
1352 		kvm_destroy_vm(kvm);
1353 }
1354 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1355 
1356 /*
1357  * Used to put a reference that was taken on behalf of an object associated
1358  * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1359  * of the new file descriptor fails and the reference cannot be transferred to
1360  * its final owner.  In such cases, the caller is still actively using @kvm and
1361  * will fail miserably if the refcount unexpectedly hits zero.
1362  */
kvm_put_kvm_no_destroy(struct kvm * kvm)1363 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1364 {
1365 	WARN_ON(refcount_dec_and_test(&kvm->users_count));
1366 }
1367 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1368 
kvm_vm_release(struct inode * inode,struct file * filp)1369 static int kvm_vm_release(struct inode *inode, struct file *filp)
1370 {
1371 	struct kvm *kvm = filp->private_data;
1372 
1373 	kvm_irqfd_release(kvm);
1374 
1375 	kvm_put_kvm(kvm);
1376 	return 0;
1377 }
1378 
1379 /*
1380  * Allocation size is twice as large as the actual dirty bitmap size.
1381  * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1382  */
kvm_alloc_dirty_bitmap(struct kvm_memory_slot * memslot)1383 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1384 {
1385 	unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1386 
1387 	memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1388 	if (!memslot->dirty_bitmap)
1389 		return -ENOMEM;
1390 
1391 	return 0;
1392 }
1393 
kvm_get_inactive_memslots(struct kvm * kvm,int as_id)1394 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1395 {
1396 	struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1397 	int node_idx_inactive = active->node_idx ^ 1;
1398 
1399 	return &kvm->__memslots[as_id][node_idx_inactive];
1400 }
1401 
1402 /*
1403  * Helper to get the address space ID when one of memslot pointers may be NULL.
1404  * This also serves as a sanity that at least one of the pointers is non-NULL,
1405  * and that their address space IDs don't diverge.
1406  */
kvm_memslots_get_as_id(struct kvm_memory_slot * a,struct kvm_memory_slot * b)1407 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1408 				  struct kvm_memory_slot *b)
1409 {
1410 	if (WARN_ON_ONCE(!a && !b))
1411 		return 0;
1412 
1413 	if (!a)
1414 		return b->as_id;
1415 	if (!b)
1416 		return a->as_id;
1417 
1418 	WARN_ON_ONCE(a->as_id != b->as_id);
1419 	return a->as_id;
1420 }
1421 
kvm_insert_gfn_node(struct kvm_memslots * slots,struct kvm_memory_slot * slot)1422 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1423 				struct kvm_memory_slot *slot)
1424 {
1425 	struct rb_root *gfn_tree = &slots->gfn_tree;
1426 	struct rb_node **node, *parent;
1427 	int idx = slots->node_idx;
1428 
1429 	parent = NULL;
1430 	for (node = &gfn_tree->rb_node; *node; ) {
1431 		struct kvm_memory_slot *tmp;
1432 
1433 		tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1434 		parent = *node;
1435 		if (slot->base_gfn < tmp->base_gfn)
1436 			node = &(*node)->rb_left;
1437 		else if (slot->base_gfn > tmp->base_gfn)
1438 			node = &(*node)->rb_right;
1439 		else
1440 			BUG();
1441 	}
1442 
1443 	rb_link_node(&slot->gfn_node[idx], parent, node);
1444 	rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1445 }
1446 
kvm_erase_gfn_node(struct kvm_memslots * slots,struct kvm_memory_slot * slot)1447 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1448 			       struct kvm_memory_slot *slot)
1449 {
1450 	rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1451 }
1452 
kvm_replace_gfn_node(struct kvm_memslots * slots,struct kvm_memory_slot * old,struct kvm_memory_slot * new)1453 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1454 				 struct kvm_memory_slot *old,
1455 				 struct kvm_memory_slot *new)
1456 {
1457 	int idx = slots->node_idx;
1458 
1459 	WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1460 
1461 	rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1462 			&slots->gfn_tree);
1463 }
1464 
1465 /*
1466  * Replace @old with @new in the inactive memslots.
1467  *
1468  * With NULL @old this simply adds @new.
1469  * With NULL @new this simply removes @old.
1470  *
1471  * If @new is non-NULL its hva_node[slots_idx] range has to be set
1472  * appropriately.
1473  */
kvm_replace_memslot(struct kvm * kvm,struct kvm_memory_slot * old,struct kvm_memory_slot * new)1474 static void kvm_replace_memslot(struct kvm *kvm,
1475 				struct kvm_memory_slot *old,
1476 				struct kvm_memory_slot *new)
1477 {
1478 	int as_id = kvm_memslots_get_as_id(old, new);
1479 	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1480 	int idx = slots->node_idx;
1481 
1482 	if (old) {
1483 		hash_del(&old->id_node[idx]);
1484 		interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1485 
1486 		if ((long)old == atomic_long_read(&slots->last_used_slot))
1487 			atomic_long_set(&slots->last_used_slot, (long)new);
1488 
1489 		if (!new) {
1490 			kvm_erase_gfn_node(slots, old);
1491 			return;
1492 		}
1493 	}
1494 
1495 	/*
1496 	 * Initialize @new's hva range.  Do this even when replacing an @old
1497 	 * slot, kvm_copy_memslot() deliberately does not touch node data.
1498 	 */
1499 	new->hva_node[idx].start = new->userspace_addr;
1500 	new->hva_node[idx].last = new->userspace_addr +
1501 				  (new->npages << PAGE_SHIFT) - 1;
1502 
1503 	/*
1504 	 * (Re)Add the new memslot.  There is no O(1) interval_tree_replace(),
1505 	 * hva_node needs to be swapped with remove+insert even though hva can't
1506 	 * change when replacing an existing slot.
1507 	 */
1508 	hash_add(slots->id_hash, &new->id_node[idx], new->id);
1509 	interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1510 
1511 	/*
1512 	 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1513 	 * switch the node in the gfn tree instead of removing the old and
1514 	 * inserting the new as two separate operations. Replacement is a
1515 	 * single O(1) operation versus two O(log(n)) operations for
1516 	 * remove+insert.
1517 	 */
1518 	if (old && old->base_gfn == new->base_gfn) {
1519 		kvm_replace_gfn_node(slots, old, new);
1520 	} else {
1521 		if (old)
1522 			kvm_erase_gfn_node(slots, old);
1523 		kvm_insert_gfn_node(slots, new);
1524 	}
1525 }
1526 
check_memory_region_flags(const struct kvm_userspace_memory_region * mem)1527 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1528 {
1529 	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1530 
1531 #ifdef __KVM_HAVE_READONLY_MEM
1532 	valid_flags |= KVM_MEM_READONLY;
1533 #endif
1534 
1535 	if (mem->flags & ~valid_flags)
1536 		return -EINVAL;
1537 
1538 	return 0;
1539 }
1540 
kvm_swap_active_memslots(struct kvm * kvm,int as_id)1541 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1542 {
1543 	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1544 
1545 	/* Grab the generation from the activate memslots. */
1546 	u64 gen = __kvm_memslots(kvm, as_id)->generation;
1547 
1548 	WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1549 	slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1550 
1551 	/*
1552 	 * Do not store the new memslots while there are invalidations in
1553 	 * progress, otherwise the locking in invalidate_range_start and
1554 	 * invalidate_range_end will be unbalanced.
1555 	 */
1556 	spin_lock(&kvm->mn_invalidate_lock);
1557 	prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1558 	while (kvm->mn_active_invalidate_count) {
1559 		set_current_state(TASK_UNINTERRUPTIBLE);
1560 		spin_unlock(&kvm->mn_invalidate_lock);
1561 		schedule();
1562 		spin_lock(&kvm->mn_invalidate_lock);
1563 	}
1564 	finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1565 	rcu_assign_pointer(kvm->memslots[as_id], slots);
1566 	spin_unlock(&kvm->mn_invalidate_lock);
1567 
1568 	/*
1569 	 * Acquired in kvm_set_memslot. Must be released before synchronize
1570 	 * SRCU below in order to avoid deadlock with another thread
1571 	 * acquiring the slots_arch_lock in an srcu critical section.
1572 	 */
1573 	mutex_unlock(&kvm->slots_arch_lock);
1574 
1575 	synchronize_srcu_expedited(&kvm->srcu);
1576 
1577 	/*
1578 	 * Increment the new memslot generation a second time, dropping the
1579 	 * update in-progress flag and incrementing the generation based on
1580 	 * the number of address spaces.  This provides a unique and easily
1581 	 * identifiable generation number while the memslots are in flux.
1582 	 */
1583 	gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1584 
1585 	/*
1586 	 * Generations must be unique even across address spaces.  We do not need
1587 	 * a global counter for that, instead the generation space is evenly split
1588 	 * across address spaces.  For example, with two address spaces, address
1589 	 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1590 	 * use generations 1, 3, 5, ...
1591 	 */
1592 	gen += KVM_ADDRESS_SPACE_NUM;
1593 
1594 	kvm_arch_memslots_updated(kvm, gen);
1595 
1596 	slots->generation = gen;
1597 }
1598 
kvm_prepare_memory_region(struct kvm * kvm,const struct kvm_memory_slot * old,struct kvm_memory_slot * new,enum kvm_mr_change change)1599 static int kvm_prepare_memory_region(struct kvm *kvm,
1600 				     const struct kvm_memory_slot *old,
1601 				     struct kvm_memory_slot *new,
1602 				     enum kvm_mr_change change)
1603 {
1604 	int r;
1605 
1606 	/*
1607 	 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1608 	 * will be freed on "commit".  If logging is enabled in both old and
1609 	 * new, reuse the existing bitmap.  If logging is enabled only in the
1610 	 * new and KVM isn't using a ring buffer, allocate and initialize a
1611 	 * new bitmap.
1612 	 */
1613 	if (change != KVM_MR_DELETE) {
1614 		if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1615 			new->dirty_bitmap = NULL;
1616 		else if (old && old->dirty_bitmap)
1617 			new->dirty_bitmap = old->dirty_bitmap;
1618 		else if (!kvm->dirty_ring_size) {
1619 			r = kvm_alloc_dirty_bitmap(new);
1620 			if (r)
1621 				return r;
1622 
1623 			if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1624 				bitmap_set(new->dirty_bitmap, 0, new->npages);
1625 		}
1626 	}
1627 
1628 	r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1629 
1630 	/* Free the bitmap on failure if it was allocated above. */
1631 	if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1632 		kvm_destroy_dirty_bitmap(new);
1633 
1634 	return r;
1635 }
1636 
kvm_commit_memory_region(struct kvm * kvm,struct kvm_memory_slot * old,const struct kvm_memory_slot * new,enum kvm_mr_change change)1637 static void kvm_commit_memory_region(struct kvm *kvm,
1638 				     struct kvm_memory_slot *old,
1639 				     const struct kvm_memory_slot *new,
1640 				     enum kvm_mr_change change)
1641 {
1642 	/*
1643 	 * Update the total number of memslot pages before calling the arch
1644 	 * hook so that architectures can consume the result directly.
1645 	 */
1646 	if (change == KVM_MR_DELETE)
1647 		kvm->nr_memslot_pages -= old->npages;
1648 	else if (change == KVM_MR_CREATE)
1649 		kvm->nr_memslot_pages += new->npages;
1650 
1651 	kvm_arch_commit_memory_region(kvm, old, new, change);
1652 
1653 	switch (change) {
1654 	case KVM_MR_CREATE:
1655 		/* Nothing more to do. */
1656 		break;
1657 	case KVM_MR_DELETE:
1658 		/* Free the old memslot and all its metadata. */
1659 		kvm_free_memslot(kvm, old);
1660 		break;
1661 	case KVM_MR_MOVE:
1662 	case KVM_MR_FLAGS_ONLY:
1663 		/*
1664 		 * Free the dirty bitmap as needed; the below check encompasses
1665 		 * both the flags and whether a ring buffer is being used)
1666 		 */
1667 		if (old->dirty_bitmap && !new->dirty_bitmap)
1668 			kvm_destroy_dirty_bitmap(old);
1669 
1670 		/*
1671 		 * The final quirk.  Free the detached, old slot, but only its
1672 		 * memory, not any metadata.  Metadata, including arch specific
1673 		 * data, may be reused by @new.
1674 		 */
1675 		kfree(old);
1676 		break;
1677 	default:
1678 		BUG();
1679 	}
1680 }
1681 
1682 /*
1683  * Activate @new, which must be installed in the inactive slots by the caller,
1684  * by swapping the active slots and then propagating @new to @old once @old is
1685  * unreachable and can be safely modified.
1686  *
1687  * With NULL @old this simply adds @new to @active (while swapping the sets).
1688  * With NULL @new this simply removes @old from @active and frees it
1689  * (while also swapping the sets).
1690  */
kvm_activate_memslot(struct kvm * kvm,struct kvm_memory_slot * old,struct kvm_memory_slot * new)1691 static void kvm_activate_memslot(struct kvm *kvm,
1692 				 struct kvm_memory_slot *old,
1693 				 struct kvm_memory_slot *new)
1694 {
1695 	int as_id = kvm_memslots_get_as_id(old, new);
1696 
1697 	kvm_swap_active_memslots(kvm, as_id);
1698 
1699 	/* Propagate the new memslot to the now inactive memslots. */
1700 	kvm_replace_memslot(kvm, old, new);
1701 }
1702 
kvm_copy_memslot(struct kvm_memory_slot * dest,const struct kvm_memory_slot * src)1703 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1704 			     const struct kvm_memory_slot *src)
1705 {
1706 	dest->base_gfn = src->base_gfn;
1707 	dest->npages = src->npages;
1708 	dest->dirty_bitmap = src->dirty_bitmap;
1709 	dest->arch = src->arch;
1710 	dest->userspace_addr = src->userspace_addr;
1711 	dest->flags = src->flags;
1712 	dest->id = src->id;
1713 	dest->as_id = src->as_id;
1714 }
1715 
kvm_invalidate_memslot(struct kvm * kvm,struct kvm_memory_slot * old,struct kvm_memory_slot * invalid_slot)1716 static void kvm_invalidate_memslot(struct kvm *kvm,
1717 				   struct kvm_memory_slot *old,
1718 				   struct kvm_memory_slot *invalid_slot)
1719 {
1720 	/*
1721 	 * Mark the current slot INVALID.  As with all memslot modifications,
1722 	 * this must be done on an unreachable slot to avoid modifying the
1723 	 * current slot in the active tree.
1724 	 */
1725 	kvm_copy_memslot(invalid_slot, old);
1726 	invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1727 	kvm_replace_memslot(kvm, old, invalid_slot);
1728 
1729 	/*
1730 	 * Activate the slot that is now marked INVALID, but don't propagate
1731 	 * the slot to the now inactive slots. The slot is either going to be
1732 	 * deleted or recreated as a new slot.
1733 	 */
1734 	kvm_swap_active_memslots(kvm, old->as_id);
1735 
1736 	/*
1737 	 * From this point no new shadow pages pointing to a deleted, or moved,
1738 	 * memslot will be created.  Validation of sp->gfn happens in:
1739 	 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1740 	 *	- kvm_is_visible_gfn (mmu_check_root)
1741 	 */
1742 	kvm_arch_flush_shadow_memslot(kvm, old);
1743 	kvm_arch_guest_memory_reclaimed(kvm);
1744 
1745 	/* Was released by kvm_swap_active_memslots, reacquire. */
1746 	mutex_lock(&kvm->slots_arch_lock);
1747 
1748 	/*
1749 	 * Copy the arch-specific field of the newly-installed slot back to the
1750 	 * old slot as the arch data could have changed between releasing
1751 	 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1752 	 * above.  Writers are required to retrieve memslots *after* acquiring
1753 	 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1754 	 */
1755 	old->arch = invalid_slot->arch;
1756 }
1757 
kvm_create_memslot(struct kvm * kvm,struct kvm_memory_slot * new)1758 static void kvm_create_memslot(struct kvm *kvm,
1759 			       struct kvm_memory_slot *new)
1760 {
1761 	/* Add the new memslot to the inactive set and activate. */
1762 	kvm_replace_memslot(kvm, NULL, new);
1763 	kvm_activate_memslot(kvm, NULL, new);
1764 }
1765 
kvm_delete_memslot(struct kvm * kvm,struct kvm_memory_slot * old,struct kvm_memory_slot * invalid_slot)1766 static void kvm_delete_memslot(struct kvm *kvm,
1767 			       struct kvm_memory_slot *old,
1768 			       struct kvm_memory_slot *invalid_slot)
1769 {
1770 	/*
1771 	 * Remove the old memslot (in the inactive memslots) by passing NULL as
1772 	 * the "new" slot, and for the invalid version in the active slots.
1773 	 */
1774 	kvm_replace_memslot(kvm, old, NULL);
1775 	kvm_activate_memslot(kvm, invalid_slot, NULL);
1776 }
1777 
kvm_move_memslot(struct kvm * kvm,struct kvm_memory_slot * old,struct kvm_memory_slot * new,struct kvm_memory_slot * invalid_slot)1778 static void kvm_move_memslot(struct kvm *kvm,
1779 			     struct kvm_memory_slot *old,
1780 			     struct kvm_memory_slot *new,
1781 			     struct kvm_memory_slot *invalid_slot)
1782 {
1783 	/*
1784 	 * Replace the old memslot in the inactive slots, and then swap slots
1785 	 * and replace the current INVALID with the new as well.
1786 	 */
1787 	kvm_replace_memslot(kvm, old, new);
1788 	kvm_activate_memslot(kvm, invalid_slot, new);
1789 }
1790 
kvm_update_flags_memslot(struct kvm * kvm,struct kvm_memory_slot * old,struct kvm_memory_slot * new)1791 static void kvm_update_flags_memslot(struct kvm *kvm,
1792 				     struct kvm_memory_slot *old,
1793 				     struct kvm_memory_slot *new)
1794 {
1795 	/*
1796 	 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1797 	 * an intermediate step. Instead, the old memslot is simply replaced
1798 	 * with a new, updated copy in both memslot sets.
1799 	 */
1800 	kvm_replace_memslot(kvm, old, new);
1801 	kvm_activate_memslot(kvm, old, new);
1802 }
1803 
kvm_set_memslot(struct kvm * kvm,struct kvm_memory_slot * old,struct kvm_memory_slot * new,enum kvm_mr_change change)1804 static int kvm_set_memslot(struct kvm *kvm,
1805 			   struct kvm_memory_slot *old,
1806 			   struct kvm_memory_slot *new,
1807 			   enum kvm_mr_change change)
1808 {
1809 	struct kvm_memory_slot *invalid_slot;
1810 	int r;
1811 
1812 	/*
1813 	 * Released in kvm_swap_active_memslots.
1814 	 *
1815 	 * Must be held from before the current memslots are copied until
1816 	 * after the new memslots are installed with rcu_assign_pointer,
1817 	 * then released before the synchronize srcu in kvm_swap_active_memslots.
1818 	 *
1819 	 * When modifying memslots outside of the slots_lock, must be held
1820 	 * before reading the pointer to the current memslots until after all
1821 	 * changes to those memslots are complete.
1822 	 *
1823 	 * These rules ensure that installing new memslots does not lose
1824 	 * changes made to the previous memslots.
1825 	 */
1826 	mutex_lock(&kvm->slots_arch_lock);
1827 
1828 	/*
1829 	 * Invalidate the old slot if it's being deleted or moved.  This is
1830 	 * done prior to actually deleting/moving the memslot to allow vCPUs to
1831 	 * continue running by ensuring there are no mappings or shadow pages
1832 	 * for the memslot when it is deleted/moved.  Without pre-invalidation
1833 	 * (and without a lock), a window would exist between effecting the
1834 	 * delete/move and committing the changes in arch code where KVM or a
1835 	 * guest could access a non-existent memslot.
1836 	 *
1837 	 * Modifications are done on a temporary, unreachable slot.  The old
1838 	 * slot needs to be preserved in case a later step fails and the
1839 	 * invalidation needs to be reverted.
1840 	 */
1841 	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1842 		invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1843 		if (!invalid_slot) {
1844 			mutex_unlock(&kvm->slots_arch_lock);
1845 			return -ENOMEM;
1846 		}
1847 		kvm_invalidate_memslot(kvm, old, invalid_slot);
1848 	}
1849 
1850 	r = kvm_prepare_memory_region(kvm, old, new, change);
1851 	if (r) {
1852 		/*
1853 		 * For DELETE/MOVE, revert the above INVALID change.  No
1854 		 * modifications required since the original slot was preserved
1855 		 * in the inactive slots.  Changing the active memslots also
1856 		 * release slots_arch_lock.
1857 		 */
1858 		if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1859 			kvm_activate_memslot(kvm, invalid_slot, old);
1860 			kfree(invalid_slot);
1861 		} else {
1862 			mutex_unlock(&kvm->slots_arch_lock);
1863 		}
1864 		return r;
1865 	}
1866 
1867 	/*
1868 	 * For DELETE and MOVE, the working slot is now active as the INVALID
1869 	 * version of the old slot.  MOVE is particularly special as it reuses
1870 	 * the old slot and returns a copy of the old slot (in working_slot).
1871 	 * For CREATE, there is no old slot.  For DELETE and FLAGS_ONLY, the
1872 	 * old slot is detached but otherwise preserved.
1873 	 */
1874 	if (change == KVM_MR_CREATE)
1875 		kvm_create_memslot(kvm, new);
1876 	else if (change == KVM_MR_DELETE)
1877 		kvm_delete_memslot(kvm, old, invalid_slot);
1878 	else if (change == KVM_MR_MOVE)
1879 		kvm_move_memslot(kvm, old, new, invalid_slot);
1880 	else if (change == KVM_MR_FLAGS_ONLY)
1881 		kvm_update_flags_memslot(kvm, old, new);
1882 	else
1883 		BUG();
1884 
1885 	/* Free the temporary INVALID slot used for DELETE and MOVE. */
1886 	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1887 		kfree(invalid_slot);
1888 
1889 	/*
1890 	 * No need to refresh new->arch, changes after dropping slots_arch_lock
1891 	 * will directly hit the final, active memslot.  Architectures are
1892 	 * responsible for knowing that new->arch may be stale.
1893 	 */
1894 	kvm_commit_memory_region(kvm, old, new, change);
1895 
1896 	return 0;
1897 }
1898 
kvm_check_memslot_overlap(struct kvm_memslots * slots,int id,gfn_t start,gfn_t end)1899 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1900 				      gfn_t start, gfn_t end)
1901 {
1902 	struct kvm_memslot_iter iter;
1903 
1904 	kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1905 		if (iter.slot->id != id)
1906 			return true;
1907 	}
1908 
1909 	return false;
1910 }
1911 
1912 /*
1913  * Allocate some memory and give it an address in the guest physical address
1914  * space.
1915  *
1916  * Discontiguous memory is allowed, mostly for framebuffers.
1917  *
1918  * Must be called holding kvm->slots_lock for write.
1919  */
__kvm_set_memory_region(struct kvm * kvm,const struct kvm_userspace_memory_region * mem)1920 int __kvm_set_memory_region(struct kvm *kvm,
1921 			    const struct kvm_userspace_memory_region *mem)
1922 {
1923 	struct kvm_memory_slot *old, *new;
1924 	struct kvm_memslots *slots;
1925 	enum kvm_mr_change change;
1926 	unsigned long npages;
1927 	gfn_t base_gfn;
1928 	int as_id, id;
1929 	int r;
1930 
1931 	r = check_memory_region_flags(mem);
1932 	if (r)
1933 		return r;
1934 
1935 	as_id = mem->slot >> 16;
1936 	id = (u16)mem->slot;
1937 
1938 	/* General sanity checks */
1939 	if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1940 	    (mem->memory_size != (unsigned long)mem->memory_size))
1941 		return -EINVAL;
1942 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1943 		return -EINVAL;
1944 	/* We can read the guest memory with __xxx_user() later on. */
1945 	if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1946 	    (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1947 	     !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1948 			mem->memory_size))
1949 		return -EINVAL;
1950 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1951 		return -EINVAL;
1952 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1953 		return -EINVAL;
1954 	if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1955 		return -EINVAL;
1956 
1957 	slots = __kvm_memslots(kvm, as_id);
1958 
1959 	/*
1960 	 * Note, the old memslot (and the pointer itself!) may be invalidated
1961 	 * and/or destroyed by kvm_set_memslot().
1962 	 */
1963 	old = id_to_memslot(slots, id);
1964 
1965 	if (!mem->memory_size) {
1966 		if (!old || !old->npages)
1967 			return -EINVAL;
1968 
1969 		if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1970 			return -EIO;
1971 
1972 		return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1973 	}
1974 
1975 	base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1976 	npages = (mem->memory_size >> PAGE_SHIFT);
1977 
1978 	if (!old || !old->npages) {
1979 		change = KVM_MR_CREATE;
1980 
1981 		/*
1982 		 * To simplify KVM internals, the total number of pages across
1983 		 * all memslots must fit in an unsigned long.
1984 		 */
1985 		if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1986 			return -EINVAL;
1987 	} else { /* Modify an existing slot. */
1988 		if ((mem->userspace_addr != old->userspace_addr) ||
1989 		    (npages != old->npages) ||
1990 		    ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1991 			return -EINVAL;
1992 
1993 		if (base_gfn != old->base_gfn)
1994 			change = KVM_MR_MOVE;
1995 		else if (mem->flags != old->flags)
1996 			change = KVM_MR_FLAGS_ONLY;
1997 		else /* Nothing to change. */
1998 			return 0;
1999 	}
2000 
2001 	if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2002 	    kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2003 		return -EEXIST;
2004 
2005 	/* Allocate a slot that will persist in the memslot. */
2006 	new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2007 	if (!new)
2008 		return -ENOMEM;
2009 
2010 	new->as_id = as_id;
2011 	new->id = id;
2012 	new->base_gfn = base_gfn;
2013 	new->npages = npages;
2014 	new->flags = mem->flags;
2015 	new->userspace_addr = mem->userspace_addr;
2016 
2017 	r = kvm_set_memslot(kvm, old, new, change);
2018 	if (r)
2019 		kfree(new);
2020 	return r;
2021 }
2022 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2023 
kvm_set_memory_region(struct kvm * kvm,const struct kvm_userspace_memory_region * mem)2024 int kvm_set_memory_region(struct kvm *kvm,
2025 			  const struct kvm_userspace_memory_region *mem)
2026 {
2027 	int r;
2028 
2029 	mutex_lock(&kvm->slots_lock);
2030 	r = __kvm_set_memory_region(kvm, mem);
2031 	mutex_unlock(&kvm->slots_lock);
2032 	return r;
2033 }
2034 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2035 
kvm_vm_ioctl_set_memory_region(struct kvm * kvm,struct kvm_userspace_memory_region * mem)2036 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2037 					  struct kvm_userspace_memory_region *mem)
2038 {
2039 	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2040 		return -EINVAL;
2041 
2042 	return kvm_set_memory_region(kvm, mem);
2043 }
2044 
2045 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2046 /**
2047  * kvm_get_dirty_log - get a snapshot of dirty pages
2048  * @kvm:	pointer to kvm instance
2049  * @log:	slot id and address to which we copy the log
2050  * @is_dirty:	set to '1' if any dirty pages were found
2051  * @memslot:	set to the associated memslot, always valid on success
2052  */
kvm_get_dirty_log(struct kvm * kvm,struct kvm_dirty_log * log,int * is_dirty,struct kvm_memory_slot ** memslot)2053 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2054 		      int *is_dirty, struct kvm_memory_slot **memslot)
2055 {
2056 	struct kvm_memslots *slots;
2057 	int i, as_id, id;
2058 	unsigned long n;
2059 	unsigned long any = 0;
2060 
2061 	/* Dirty ring tracking is exclusive to dirty log tracking */
2062 	if (kvm->dirty_ring_size)
2063 		return -ENXIO;
2064 
2065 	*memslot = NULL;
2066 	*is_dirty = 0;
2067 
2068 	as_id = log->slot >> 16;
2069 	id = (u16)log->slot;
2070 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2071 		return -EINVAL;
2072 
2073 	slots = __kvm_memslots(kvm, as_id);
2074 	*memslot = id_to_memslot(slots, id);
2075 	if (!(*memslot) || !(*memslot)->dirty_bitmap)
2076 		return -ENOENT;
2077 
2078 	kvm_arch_sync_dirty_log(kvm, *memslot);
2079 
2080 	n = kvm_dirty_bitmap_bytes(*memslot);
2081 
2082 	for (i = 0; !any && i < n/sizeof(long); ++i)
2083 		any = (*memslot)->dirty_bitmap[i];
2084 
2085 	if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2086 		return -EFAULT;
2087 
2088 	if (any)
2089 		*is_dirty = 1;
2090 	return 0;
2091 }
2092 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2093 
2094 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2095 /**
2096  * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2097  *	and reenable dirty page tracking for the corresponding pages.
2098  * @kvm:	pointer to kvm instance
2099  * @log:	slot id and address to which we copy the log
2100  *
2101  * We need to keep it in mind that VCPU threads can write to the bitmap
2102  * concurrently. So, to avoid losing track of dirty pages we keep the
2103  * following order:
2104  *
2105  *    1. Take a snapshot of the bit and clear it if needed.
2106  *    2. Write protect the corresponding page.
2107  *    3. Copy the snapshot to the userspace.
2108  *    4. Upon return caller flushes TLB's if needed.
2109  *
2110  * Between 2 and 4, the guest may write to the page using the remaining TLB
2111  * entry.  This is not a problem because the page is reported dirty using
2112  * the snapshot taken before and step 4 ensures that writes done after
2113  * exiting to userspace will be logged for the next call.
2114  *
2115  */
kvm_get_dirty_log_protect(struct kvm * kvm,struct kvm_dirty_log * log)2116 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2117 {
2118 	struct kvm_memslots *slots;
2119 	struct kvm_memory_slot *memslot;
2120 	int i, as_id, id;
2121 	unsigned long n;
2122 	unsigned long *dirty_bitmap;
2123 	unsigned long *dirty_bitmap_buffer;
2124 	bool flush;
2125 
2126 	/* Dirty ring tracking is exclusive to dirty log tracking */
2127 	if (kvm->dirty_ring_size)
2128 		return -ENXIO;
2129 
2130 	as_id = log->slot >> 16;
2131 	id = (u16)log->slot;
2132 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2133 		return -EINVAL;
2134 
2135 	slots = __kvm_memslots(kvm, as_id);
2136 	memslot = id_to_memslot(slots, id);
2137 	if (!memslot || !memslot->dirty_bitmap)
2138 		return -ENOENT;
2139 
2140 	dirty_bitmap = memslot->dirty_bitmap;
2141 
2142 	kvm_arch_sync_dirty_log(kvm, memslot);
2143 
2144 	n = kvm_dirty_bitmap_bytes(memslot);
2145 	flush = false;
2146 	if (kvm->manual_dirty_log_protect) {
2147 		/*
2148 		 * Unlike kvm_get_dirty_log, we always return false in *flush,
2149 		 * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
2150 		 * is some code duplication between this function and
2151 		 * kvm_get_dirty_log, but hopefully all architecture
2152 		 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2153 		 * can be eliminated.
2154 		 */
2155 		dirty_bitmap_buffer = dirty_bitmap;
2156 	} else {
2157 		dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2158 		memset(dirty_bitmap_buffer, 0, n);
2159 
2160 		KVM_MMU_LOCK(kvm);
2161 		for (i = 0; i < n / sizeof(long); i++) {
2162 			unsigned long mask;
2163 			gfn_t offset;
2164 
2165 			if (!dirty_bitmap[i])
2166 				continue;
2167 
2168 			flush = true;
2169 			mask = xchg(&dirty_bitmap[i], 0);
2170 			dirty_bitmap_buffer[i] = mask;
2171 
2172 			offset = i * BITS_PER_LONG;
2173 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2174 								offset, mask);
2175 		}
2176 		KVM_MMU_UNLOCK(kvm);
2177 	}
2178 
2179 	if (flush)
2180 		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2181 
2182 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2183 		return -EFAULT;
2184 	return 0;
2185 }
2186 
2187 
2188 /**
2189  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2190  * @kvm: kvm instance
2191  * @log: slot id and address to which we copy the log
2192  *
2193  * Steps 1-4 below provide general overview of dirty page logging. See
2194  * kvm_get_dirty_log_protect() function description for additional details.
2195  *
2196  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2197  * always flush the TLB (step 4) even if previous step failed  and the dirty
2198  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2199  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2200  * writes will be marked dirty for next log read.
2201  *
2202  *   1. Take a snapshot of the bit and clear it if needed.
2203  *   2. Write protect the corresponding page.
2204  *   3. Copy the snapshot to the userspace.
2205  *   4. Flush TLB's if needed.
2206  */
kvm_vm_ioctl_get_dirty_log(struct kvm * kvm,struct kvm_dirty_log * log)2207 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2208 				      struct kvm_dirty_log *log)
2209 {
2210 	int r;
2211 
2212 	mutex_lock(&kvm->slots_lock);
2213 
2214 	r = kvm_get_dirty_log_protect(kvm, log);
2215 
2216 	mutex_unlock(&kvm->slots_lock);
2217 	return r;
2218 }
2219 
2220 /**
2221  * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2222  *	and reenable dirty page tracking for the corresponding pages.
2223  * @kvm:	pointer to kvm instance
2224  * @log:	slot id and address from which to fetch the bitmap of dirty pages
2225  */
kvm_clear_dirty_log_protect(struct kvm * kvm,struct kvm_clear_dirty_log * log)2226 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2227 				       struct kvm_clear_dirty_log *log)
2228 {
2229 	struct kvm_memslots *slots;
2230 	struct kvm_memory_slot *memslot;
2231 	int as_id, id;
2232 	gfn_t offset;
2233 	unsigned long i, n;
2234 	unsigned long *dirty_bitmap;
2235 	unsigned long *dirty_bitmap_buffer;
2236 	bool flush;
2237 
2238 	/* Dirty ring tracking is exclusive to dirty log tracking */
2239 	if (kvm->dirty_ring_size)
2240 		return -ENXIO;
2241 
2242 	as_id = log->slot >> 16;
2243 	id = (u16)log->slot;
2244 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2245 		return -EINVAL;
2246 
2247 	if (log->first_page & 63)
2248 		return -EINVAL;
2249 
2250 	slots = __kvm_memslots(kvm, as_id);
2251 	memslot = id_to_memslot(slots, id);
2252 	if (!memslot || !memslot->dirty_bitmap)
2253 		return -ENOENT;
2254 
2255 	dirty_bitmap = memslot->dirty_bitmap;
2256 
2257 	n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2258 
2259 	if (log->first_page > memslot->npages ||
2260 	    log->num_pages > memslot->npages - log->first_page ||
2261 	    (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2262 	    return -EINVAL;
2263 
2264 	kvm_arch_sync_dirty_log(kvm, memslot);
2265 
2266 	flush = false;
2267 	dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2268 	if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2269 		return -EFAULT;
2270 
2271 	KVM_MMU_LOCK(kvm);
2272 	for (offset = log->first_page, i = offset / BITS_PER_LONG,
2273 		 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2274 	     i++, offset += BITS_PER_LONG) {
2275 		unsigned long mask = *dirty_bitmap_buffer++;
2276 		atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2277 		if (!mask)
2278 			continue;
2279 
2280 		mask &= atomic_long_fetch_andnot(mask, p);
2281 
2282 		/*
2283 		 * mask contains the bits that really have been cleared.  This
2284 		 * never includes any bits beyond the length of the memslot (if
2285 		 * the length is not aligned to 64 pages), therefore it is not
2286 		 * a problem if userspace sets them in log->dirty_bitmap.
2287 		*/
2288 		if (mask) {
2289 			flush = true;
2290 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2291 								offset, mask);
2292 		}
2293 	}
2294 	KVM_MMU_UNLOCK(kvm);
2295 
2296 	if (flush)
2297 		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2298 
2299 	return 0;
2300 }
2301 
kvm_vm_ioctl_clear_dirty_log(struct kvm * kvm,struct kvm_clear_dirty_log * log)2302 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2303 					struct kvm_clear_dirty_log *log)
2304 {
2305 	int r;
2306 
2307 	mutex_lock(&kvm->slots_lock);
2308 
2309 	r = kvm_clear_dirty_log_protect(kvm, log);
2310 
2311 	mutex_unlock(&kvm->slots_lock);
2312 	return r;
2313 }
2314 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2315 
gfn_to_memslot(struct kvm * kvm,gfn_t gfn)2316 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2317 {
2318 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2319 }
2320 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2321 
kvm_vcpu_gfn_to_memslot(struct kvm_vcpu * vcpu,gfn_t gfn)2322 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2323 {
2324 	struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2325 	u64 gen = slots->generation;
2326 	struct kvm_memory_slot *slot;
2327 
2328 	/*
2329 	 * This also protects against using a memslot from a different address space,
2330 	 * since different address spaces have different generation numbers.
2331 	 */
2332 	if (unlikely(gen != vcpu->last_used_slot_gen)) {
2333 		vcpu->last_used_slot = NULL;
2334 		vcpu->last_used_slot_gen = gen;
2335 	}
2336 
2337 	slot = try_get_memslot(vcpu->last_used_slot, gfn);
2338 	if (slot)
2339 		return slot;
2340 
2341 	/*
2342 	 * Fall back to searching all memslots. We purposely use
2343 	 * search_memslots() instead of __gfn_to_memslot() to avoid
2344 	 * thrashing the VM-wide last_used_slot in kvm_memslots.
2345 	 */
2346 	slot = search_memslots(slots, gfn, false);
2347 	if (slot) {
2348 		vcpu->last_used_slot = slot;
2349 		return slot;
2350 	}
2351 
2352 	return NULL;
2353 }
2354 
kvm_is_visible_gfn(struct kvm * kvm,gfn_t gfn)2355 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2356 {
2357 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2358 
2359 	return kvm_is_visible_memslot(memslot);
2360 }
2361 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2362 
kvm_vcpu_is_visible_gfn(struct kvm_vcpu * vcpu,gfn_t gfn)2363 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2364 {
2365 	struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2366 
2367 	return kvm_is_visible_memslot(memslot);
2368 }
2369 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2370 
kvm_host_page_size(struct kvm_vcpu * vcpu,gfn_t gfn)2371 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2372 {
2373 	struct vm_area_struct *vma;
2374 	unsigned long addr, size;
2375 
2376 	size = PAGE_SIZE;
2377 
2378 	addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2379 	if (kvm_is_error_hva(addr))
2380 		return PAGE_SIZE;
2381 
2382 	mmap_read_lock(current->mm);
2383 	vma = find_vma(current->mm, addr);
2384 	if (!vma)
2385 		goto out;
2386 
2387 	size = vma_kernel_pagesize(vma);
2388 
2389 out:
2390 	mmap_read_unlock(current->mm);
2391 
2392 	return size;
2393 }
2394 
memslot_is_readonly(const struct kvm_memory_slot * slot)2395 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2396 {
2397 	return slot->flags & KVM_MEM_READONLY;
2398 }
2399 
__gfn_to_hva_many(const struct kvm_memory_slot * slot,gfn_t gfn,gfn_t * nr_pages,bool write)2400 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2401 				       gfn_t *nr_pages, bool write)
2402 {
2403 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2404 		return KVM_HVA_ERR_BAD;
2405 
2406 	if (memslot_is_readonly(slot) && write)
2407 		return KVM_HVA_ERR_RO_BAD;
2408 
2409 	if (nr_pages)
2410 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
2411 
2412 	return __gfn_to_hva_memslot(slot, gfn);
2413 }
2414 
gfn_to_hva_many(struct kvm_memory_slot * slot,gfn_t gfn,gfn_t * nr_pages)2415 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2416 				     gfn_t *nr_pages)
2417 {
2418 	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2419 }
2420 
gfn_to_hva_memslot(struct kvm_memory_slot * slot,gfn_t gfn)2421 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2422 					gfn_t gfn)
2423 {
2424 	return gfn_to_hva_many(slot, gfn, NULL);
2425 }
2426 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2427 
gfn_to_hva(struct kvm * kvm,gfn_t gfn)2428 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2429 {
2430 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2431 }
2432 EXPORT_SYMBOL_GPL(gfn_to_hva);
2433 
kvm_vcpu_gfn_to_hva(struct kvm_vcpu * vcpu,gfn_t gfn)2434 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2435 {
2436 	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2437 }
2438 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2439 
2440 /*
2441  * Return the hva of a @gfn and the R/W attribute if possible.
2442  *
2443  * @slot: the kvm_memory_slot which contains @gfn
2444  * @gfn: the gfn to be translated
2445  * @writable: used to return the read/write attribute of the @slot if the hva
2446  * is valid and @writable is not NULL
2447  */
gfn_to_hva_memslot_prot(struct kvm_memory_slot * slot,gfn_t gfn,bool * writable)2448 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2449 				      gfn_t gfn, bool *writable)
2450 {
2451 	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2452 
2453 	if (!kvm_is_error_hva(hva) && writable)
2454 		*writable = !memslot_is_readonly(slot);
2455 
2456 	return hva;
2457 }
2458 
gfn_to_hva_prot(struct kvm * kvm,gfn_t gfn,bool * writable)2459 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2460 {
2461 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2462 
2463 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2464 }
2465 
kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu * vcpu,gfn_t gfn,bool * writable)2466 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2467 {
2468 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2469 
2470 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2471 }
2472 
check_user_page_hwpoison(unsigned long addr)2473 static inline int check_user_page_hwpoison(unsigned long addr)
2474 {
2475 	int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2476 
2477 	rc = get_user_pages(addr, 1, flags, NULL, NULL);
2478 	return rc == -EHWPOISON;
2479 }
2480 
2481 /*
2482  * The fast path to get the writable pfn which will be stored in @pfn,
2483  * true indicates success, otherwise false is returned.  It's also the
2484  * only part that runs if we can in atomic context.
2485  */
hva_to_pfn_fast(unsigned long addr,bool write_fault,bool * writable,kvm_pfn_t * pfn)2486 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2487 			    bool *writable, kvm_pfn_t *pfn)
2488 {
2489 	struct page *page[1];
2490 
2491 	/*
2492 	 * Fast pin a writable pfn only if it is a write fault request
2493 	 * or the caller allows to map a writable pfn for a read fault
2494 	 * request.
2495 	 */
2496 	if (!(write_fault || writable))
2497 		return false;
2498 
2499 	if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2500 		*pfn = page_to_pfn(page[0]);
2501 
2502 		if (writable)
2503 			*writable = true;
2504 		return true;
2505 	}
2506 
2507 	return false;
2508 }
2509 
2510 /*
2511  * The slow path to get the pfn of the specified host virtual address,
2512  * 1 indicates success, -errno is returned if error is detected.
2513  */
hva_to_pfn_slow(unsigned long addr,bool * async,bool write_fault,bool * writable,kvm_pfn_t * pfn)2514 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2515 			   bool *writable, kvm_pfn_t *pfn)
2516 {
2517 	unsigned int flags = FOLL_HWPOISON;
2518 	struct page *page;
2519 	int npages;
2520 
2521 	might_sleep();
2522 
2523 	if (writable)
2524 		*writable = write_fault;
2525 
2526 	if (write_fault)
2527 		flags |= FOLL_WRITE;
2528 	if (async)
2529 		flags |= FOLL_NOWAIT;
2530 
2531 	npages = get_user_pages_unlocked(addr, 1, &page, flags);
2532 	if (npages != 1)
2533 		return npages;
2534 
2535 	/* map read fault as writable if possible */
2536 	if (unlikely(!write_fault) && writable) {
2537 		struct page *wpage;
2538 
2539 		if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2540 			*writable = true;
2541 			put_page(page);
2542 			page = wpage;
2543 		}
2544 	}
2545 	*pfn = page_to_pfn(page);
2546 	return npages;
2547 }
2548 
vma_is_valid(struct vm_area_struct * vma,bool write_fault)2549 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2550 {
2551 	if (unlikely(!(vma->vm_flags & VM_READ)))
2552 		return false;
2553 
2554 	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2555 		return false;
2556 
2557 	return true;
2558 }
2559 
kvm_try_get_pfn(kvm_pfn_t pfn)2560 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2561 {
2562 	struct page *page = kvm_pfn_to_refcounted_page(pfn);
2563 
2564 	if (!page)
2565 		return 1;
2566 
2567 	return get_page_unless_zero(page);
2568 }
2569 
hva_to_pfn_remapped(struct vm_area_struct * vma,unsigned long addr,bool write_fault,bool * writable,kvm_pfn_t * p_pfn)2570 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2571 			       unsigned long addr, bool write_fault,
2572 			       bool *writable, kvm_pfn_t *p_pfn)
2573 {
2574 	kvm_pfn_t pfn;
2575 	pte_t *ptep;
2576 	spinlock_t *ptl;
2577 	int r;
2578 
2579 	r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2580 	if (r) {
2581 		/*
2582 		 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2583 		 * not call the fault handler, so do it here.
2584 		 */
2585 		bool unlocked = false;
2586 		r = fixup_user_fault(current->mm, addr,
2587 				     (write_fault ? FAULT_FLAG_WRITE : 0),
2588 				     &unlocked);
2589 		if (unlocked)
2590 			return -EAGAIN;
2591 		if (r)
2592 			return r;
2593 
2594 		r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2595 		if (r)
2596 			return r;
2597 	}
2598 
2599 	if (write_fault && !pte_write(*ptep)) {
2600 		pfn = KVM_PFN_ERR_RO_FAULT;
2601 		goto out;
2602 	}
2603 
2604 	if (writable)
2605 		*writable = pte_write(*ptep);
2606 	pfn = pte_pfn(*ptep);
2607 
2608 	/*
2609 	 * Get a reference here because callers of *hva_to_pfn* and
2610 	 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2611 	 * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
2612 	 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2613 	 * simply do nothing for reserved pfns.
2614 	 *
2615 	 * Whoever called remap_pfn_range is also going to call e.g.
2616 	 * unmap_mapping_range before the underlying pages are freed,
2617 	 * causing a call to our MMU notifier.
2618 	 *
2619 	 * Certain IO or PFNMAP mappings can be backed with valid
2620 	 * struct pages, but be allocated without refcounting e.g.,
2621 	 * tail pages of non-compound higher order allocations, which
2622 	 * would then underflow the refcount when the caller does the
2623 	 * required put_page. Don't allow those pages here.
2624 	 */
2625 	if (!kvm_try_get_pfn(pfn))
2626 		r = -EFAULT;
2627 
2628 out:
2629 	pte_unmap_unlock(ptep, ptl);
2630 	*p_pfn = pfn;
2631 
2632 	return r;
2633 }
2634 
2635 /*
2636  * Pin guest page in memory and return its pfn.
2637  * @addr: host virtual address which maps memory to the guest
2638  * @atomic: whether this function can sleep
2639  * @async: whether this function need to wait IO complete if the
2640  *         host page is not in the memory
2641  * @write_fault: whether we should get a writable host page
2642  * @writable: whether it allows to map a writable host page for !@write_fault
2643  *
2644  * The function will map a writable host page for these two cases:
2645  * 1): @write_fault = true
2646  * 2): @write_fault = false && @writable, @writable will tell the caller
2647  *     whether the mapping is writable.
2648  */
hva_to_pfn(unsigned long addr,bool atomic,bool * async,bool write_fault,bool * writable)2649 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2650 		     bool write_fault, bool *writable)
2651 {
2652 	struct vm_area_struct *vma;
2653 	kvm_pfn_t pfn;
2654 	int npages, r;
2655 
2656 	/* we can do it either atomically or asynchronously, not both */
2657 	BUG_ON(atomic && async);
2658 
2659 	if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2660 		return pfn;
2661 
2662 	if (atomic)
2663 		return KVM_PFN_ERR_FAULT;
2664 
2665 	npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2666 	if (npages == 1)
2667 		return pfn;
2668 
2669 	mmap_read_lock(current->mm);
2670 	if (npages == -EHWPOISON ||
2671 	      (!async && check_user_page_hwpoison(addr))) {
2672 		pfn = KVM_PFN_ERR_HWPOISON;
2673 		goto exit;
2674 	}
2675 
2676 retry:
2677 	vma = vma_lookup(current->mm, addr);
2678 
2679 	if (vma == NULL)
2680 		pfn = KVM_PFN_ERR_FAULT;
2681 	else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2682 		r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2683 		if (r == -EAGAIN)
2684 			goto retry;
2685 		if (r < 0)
2686 			pfn = KVM_PFN_ERR_FAULT;
2687 	} else {
2688 		if (async && vma_is_valid(vma, write_fault))
2689 			*async = true;
2690 		pfn = KVM_PFN_ERR_FAULT;
2691 	}
2692 exit:
2693 	mmap_read_unlock(current->mm);
2694 	return pfn;
2695 }
2696 
__gfn_to_pfn_memslot(const struct kvm_memory_slot * slot,gfn_t gfn,bool atomic,bool * async,bool write_fault,bool * writable,hva_t * hva)2697 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2698 			       bool atomic, bool *async, bool write_fault,
2699 			       bool *writable, hva_t *hva)
2700 {
2701 	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2702 
2703 	if (hva)
2704 		*hva = addr;
2705 
2706 	if (addr == KVM_HVA_ERR_RO_BAD) {
2707 		if (writable)
2708 			*writable = false;
2709 		return KVM_PFN_ERR_RO_FAULT;
2710 	}
2711 
2712 	if (kvm_is_error_hva(addr)) {
2713 		if (writable)
2714 			*writable = false;
2715 		return KVM_PFN_NOSLOT;
2716 	}
2717 
2718 	/* Do not map writable pfn in the readonly memslot. */
2719 	if (writable && memslot_is_readonly(slot)) {
2720 		*writable = false;
2721 		writable = NULL;
2722 	}
2723 
2724 	return hva_to_pfn(addr, atomic, async, write_fault,
2725 			  writable);
2726 }
2727 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2728 
gfn_to_pfn_prot(struct kvm * kvm,gfn_t gfn,bool write_fault,bool * writable)2729 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2730 		      bool *writable)
2731 {
2732 	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2733 				    write_fault, writable, NULL);
2734 }
2735 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2736 
gfn_to_pfn_memslot(const struct kvm_memory_slot * slot,gfn_t gfn)2737 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2738 {
2739 	return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2740 }
2741 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2742 
gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot * slot,gfn_t gfn)2743 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2744 {
2745 	return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2746 }
2747 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2748 
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu * vcpu,gfn_t gfn)2749 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2750 {
2751 	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2752 }
2753 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2754 
gfn_to_pfn(struct kvm * kvm,gfn_t gfn)2755 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2756 {
2757 	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2758 }
2759 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2760 
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu * vcpu,gfn_t gfn)2761 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2762 {
2763 	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2764 }
2765 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2766 
gfn_to_page_many_atomic(struct kvm_memory_slot * slot,gfn_t gfn,struct page ** pages,int nr_pages)2767 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2768 			    struct page **pages, int nr_pages)
2769 {
2770 	unsigned long addr;
2771 	gfn_t entry = 0;
2772 
2773 	addr = gfn_to_hva_many(slot, gfn, &entry);
2774 	if (kvm_is_error_hva(addr))
2775 		return -1;
2776 
2777 	if (entry < nr_pages)
2778 		return 0;
2779 
2780 	return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2781 }
2782 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2783 
2784 /*
2785  * Do not use this helper unless you are absolutely certain the gfn _must_ be
2786  * backed by 'struct page'.  A valid example is if the backing memslot is
2787  * controlled by KVM.  Note, if the returned page is valid, it's refcount has
2788  * been elevated by gfn_to_pfn().
2789  */
gfn_to_page(struct kvm * kvm,gfn_t gfn)2790 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2791 {
2792 	struct page *page;
2793 	kvm_pfn_t pfn;
2794 
2795 	pfn = gfn_to_pfn(kvm, gfn);
2796 
2797 	if (is_error_noslot_pfn(pfn))
2798 		return KVM_ERR_PTR_BAD_PAGE;
2799 
2800 	page = kvm_pfn_to_refcounted_page(pfn);
2801 	if (!page)
2802 		return KVM_ERR_PTR_BAD_PAGE;
2803 
2804 	return page;
2805 }
2806 EXPORT_SYMBOL_GPL(gfn_to_page);
2807 
kvm_release_pfn(kvm_pfn_t pfn,bool dirty)2808 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2809 {
2810 	if (dirty)
2811 		kvm_release_pfn_dirty(pfn);
2812 	else
2813 		kvm_release_pfn_clean(pfn);
2814 }
2815 
kvm_vcpu_map(struct kvm_vcpu * vcpu,gfn_t gfn,struct kvm_host_map * map)2816 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2817 {
2818 	kvm_pfn_t pfn;
2819 	void *hva = NULL;
2820 	struct page *page = KVM_UNMAPPED_PAGE;
2821 
2822 	if (!map)
2823 		return -EINVAL;
2824 
2825 	pfn = gfn_to_pfn(vcpu->kvm, gfn);
2826 	if (is_error_noslot_pfn(pfn))
2827 		return -EINVAL;
2828 
2829 	if (pfn_valid(pfn)) {
2830 		page = pfn_to_page(pfn);
2831 		hva = kmap(page);
2832 #ifdef CONFIG_HAS_IOMEM
2833 	} else {
2834 		hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2835 #endif
2836 	}
2837 
2838 	if (!hva)
2839 		return -EFAULT;
2840 
2841 	map->page = page;
2842 	map->hva = hva;
2843 	map->pfn = pfn;
2844 	map->gfn = gfn;
2845 
2846 	return 0;
2847 }
2848 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2849 
kvm_vcpu_unmap(struct kvm_vcpu * vcpu,struct kvm_host_map * map,bool dirty)2850 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2851 {
2852 	if (!map)
2853 		return;
2854 
2855 	if (!map->hva)
2856 		return;
2857 
2858 	if (map->page != KVM_UNMAPPED_PAGE)
2859 		kunmap(map->page);
2860 #ifdef CONFIG_HAS_IOMEM
2861 	else
2862 		memunmap(map->hva);
2863 #endif
2864 
2865 	if (dirty)
2866 		kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2867 
2868 	kvm_release_pfn(map->pfn, dirty);
2869 
2870 	map->hva = NULL;
2871 	map->page = NULL;
2872 }
2873 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2874 
kvm_is_ad_tracked_page(struct page * page)2875 static bool kvm_is_ad_tracked_page(struct page *page)
2876 {
2877 	/*
2878 	 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2879 	 * touched (e.g. set dirty) except by its owner".
2880 	 */
2881 	return !PageReserved(page);
2882 }
2883 
kvm_set_page_dirty(struct page * page)2884 static void kvm_set_page_dirty(struct page *page)
2885 {
2886 	if (kvm_is_ad_tracked_page(page))
2887 		SetPageDirty(page);
2888 }
2889 
kvm_set_page_accessed(struct page * page)2890 static void kvm_set_page_accessed(struct page *page)
2891 {
2892 	if (kvm_is_ad_tracked_page(page))
2893 		mark_page_accessed(page);
2894 }
2895 
kvm_release_page_clean(struct page * page)2896 void kvm_release_page_clean(struct page *page)
2897 {
2898 	WARN_ON(is_error_page(page));
2899 
2900 	kvm_set_page_accessed(page);
2901 	put_page(page);
2902 }
2903 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2904 
kvm_release_pfn_clean(kvm_pfn_t pfn)2905 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2906 {
2907 	struct page *page;
2908 
2909 	if (is_error_noslot_pfn(pfn))
2910 		return;
2911 
2912 	page = kvm_pfn_to_refcounted_page(pfn);
2913 	if (!page)
2914 		return;
2915 
2916 	kvm_release_page_clean(page);
2917 }
2918 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2919 
kvm_release_page_dirty(struct page * page)2920 void kvm_release_page_dirty(struct page *page)
2921 {
2922 	WARN_ON(is_error_page(page));
2923 
2924 	kvm_set_page_dirty(page);
2925 	kvm_release_page_clean(page);
2926 }
2927 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2928 
kvm_release_pfn_dirty(kvm_pfn_t pfn)2929 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2930 {
2931 	struct page *page;
2932 
2933 	if (is_error_noslot_pfn(pfn))
2934 		return;
2935 
2936 	page = kvm_pfn_to_refcounted_page(pfn);
2937 	if (!page)
2938 		return;
2939 
2940 	kvm_release_page_dirty(page);
2941 }
2942 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2943 
2944 /*
2945  * Note, checking for an error/noslot pfn is the caller's responsibility when
2946  * directly marking a page dirty/accessed.  Unlike the "release" helpers, the
2947  * "set" helpers are not to be used when the pfn might point at garbage.
2948  */
kvm_set_pfn_dirty(kvm_pfn_t pfn)2949 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2950 {
2951 	if (WARN_ON(is_error_noslot_pfn(pfn)))
2952 		return;
2953 
2954 	if (pfn_valid(pfn))
2955 		kvm_set_page_dirty(pfn_to_page(pfn));
2956 }
2957 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2958 
kvm_set_pfn_accessed(kvm_pfn_t pfn)2959 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2960 {
2961 	if (WARN_ON(is_error_noslot_pfn(pfn)))
2962 		return;
2963 
2964 	if (pfn_valid(pfn))
2965 		kvm_set_page_accessed(pfn_to_page(pfn));
2966 }
2967 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2968 
next_segment(unsigned long len,int offset)2969 static int next_segment(unsigned long len, int offset)
2970 {
2971 	if (len > PAGE_SIZE - offset)
2972 		return PAGE_SIZE - offset;
2973 	else
2974 		return len;
2975 }
2976 
__kvm_read_guest_page(struct kvm_memory_slot * slot,gfn_t gfn,void * data,int offset,int len)2977 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2978 				 void *data, int offset, int len)
2979 {
2980 	int r;
2981 	unsigned long addr;
2982 
2983 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2984 	if (kvm_is_error_hva(addr))
2985 		return -EFAULT;
2986 	r = __copy_from_user(data, (void __user *)addr + offset, len);
2987 	if (r)
2988 		return -EFAULT;
2989 	return 0;
2990 }
2991 
kvm_read_guest_page(struct kvm * kvm,gfn_t gfn,void * data,int offset,int len)2992 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2993 			int len)
2994 {
2995 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2996 
2997 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
2998 }
2999 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3000 
kvm_vcpu_read_guest_page(struct kvm_vcpu * vcpu,gfn_t gfn,void * data,int offset,int len)3001 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3002 			     int offset, int len)
3003 {
3004 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3005 
3006 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
3007 }
3008 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3009 
kvm_read_guest(struct kvm * kvm,gpa_t gpa,void * data,unsigned long len)3010 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3011 {
3012 	gfn_t gfn = gpa >> PAGE_SHIFT;
3013 	int seg;
3014 	int offset = offset_in_page(gpa);
3015 	int ret;
3016 
3017 	while ((seg = next_segment(len, offset)) != 0) {
3018 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3019 		if (ret < 0)
3020 			return ret;
3021 		offset = 0;
3022 		len -= seg;
3023 		data += seg;
3024 		++gfn;
3025 	}
3026 	return 0;
3027 }
3028 EXPORT_SYMBOL_GPL(kvm_read_guest);
3029 
kvm_vcpu_read_guest(struct kvm_vcpu * vcpu,gpa_t gpa,void * data,unsigned long len)3030 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3031 {
3032 	gfn_t gfn = gpa >> PAGE_SHIFT;
3033 	int seg;
3034 	int offset = offset_in_page(gpa);
3035 	int ret;
3036 
3037 	while ((seg = next_segment(len, offset)) != 0) {
3038 		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3039 		if (ret < 0)
3040 			return ret;
3041 		offset = 0;
3042 		len -= seg;
3043 		data += seg;
3044 		++gfn;
3045 	}
3046 	return 0;
3047 }
3048 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3049 
__kvm_read_guest_atomic(struct kvm_memory_slot * slot,gfn_t gfn,void * data,int offset,unsigned long len)3050 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3051 			           void *data, int offset, unsigned long len)
3052 {
3053 	int r;
3054 	unsigned long addr;
3055 
3056 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3057 	if (kvm_is_error_hva(addr))
3058 		return -EFAULT;
3059 	pagefault_disable();
3060 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3061 	pagefault_enable();
3062 	if (r)
3063 		return -EFAULT;
3064 	return 0;
3065 }
3066 
kvm_vcpu_read_guest_atomic(struct kvm_vcpu * vcpu,gpa_t gpa,void * data,unsigned long len)3067 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3068 			       void *data, unsigned long len)
3069 {
3070 	gfn_t gfn = gpa >> PAGE_SHIFT;
3071 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3072 	int offset = offset_in_page(gpa);
3073 
3074 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3075 }
3076 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3077 
__kvm_write_guest_page(struct kvm * kvm,struct kvm_memory_slot * memslot,gfn_t gfn,const void * data,int offset,int len)3078 static int __kvm_write_guest_page(struct kvm *kvm,
3079 				  struct kvm_memory_slot *memslot, gfn_t gfn,
3080 			          const void *data, int offset, int len)
3081 {
3082 	int r;
3083 	unsigned long addr;
3084 
3085 	addr = gfn_to_hva_memslot(memslot, gfn);
3086 	if (kvm_is_error_hva(addr))
3087 		return -EFAULT;
3088 	r = __copy_to_user((void __user *)addr + offset, data, len);
3089 	if (r)
3090 		return -EFAULT;
3091 	mark_page_dirty_in_slot(kvm, memslot, gfn);
3092 	return 0;
3093 }
3094 
kvm_write_guest_page(struct kvm * kvm,gfn_t gfn,const void * data,int offset,int len)3095 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3096 			 const void *data, int offset, int len)
3097 {
3098 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3099 
3100 	return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3101 }
3102 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3103 
kvm_vcpu_write_guest_page(struct kvm_vcpu * vcpu,gfn_t gfn,const void * data,int offset,int len)3104 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3105 			      const void *data, int offset, int len)
3106 {
3107 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3108 
3109 	return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3110 }
3111 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3112 
kvm_write_guest(struct kvm * kvm,gpa_t gpa,const void * data,unsigned long len)3113 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3114 		    unsigned long len)
3115 {
3116 	gfn_t gfn = gpa >> PAGE_SHIFT;
3117 	int seg;
3118 	int offset = offset_in_page(gpa);
3119 	int ret;
3120 
3121 	while ((seg = next_segment(len, offset)) != 0) {
3122 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3123 		if (ret < 0)
3124 			return ret;
3125 		offset = 0;
3126 		len -= seg;
3127 		data += seg;
3128 		++gfn;
3129 	}
3130 	return 0;
3131 }
3132 EXPORT_SYMBOL_GPL(kvm_write_guest);
3133 
kvm_vcpu_write_guest(struct kvm_vcpu * vcpu,gpa_t gpa,const void * data,unsigned long len)3134 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3135 		         unsigned long len)
3136 {
3137 	gfn_t gfn = gpa >> PAGE_SHIFT;
3138 	int seg;
3139 	int offset = offset_in_page(gpa);
3140 	int ret;
3141 
3142 	while ((seg = next_segment(len, offset)) != 0) {
3143 		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3144 		if (ret < 0)
3145 			return ret;
3146 		offset = 0;
3147 		len -= seg;
3148 		data += seg;
3149 		++gfn;
3150 	}
3151 	return 0;
3152 }
3153 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3154 
__kvm_gfn_to_hva_cache_init(struct kvm_memslots * slots,struct gfn_to_hva_cache * ghc,gpa_t gpa,unsigned long len)3155 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3156 				       struct gfn_to_hva_cache *ghc,
3157 				       gpa_t gpa, unsigned long len)
3158 {
3159 	int offset = offset_in_page(gpa);
3160 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
3161 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3162 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3163 	gfn_t nr_pages_avail;
3164 
3165 	/* Update ghc->generation before performing any error checks. */
3166 	ghc->generation = slots->generation;
3167 
3168 	if (start_gfn > end_gfn) {
3169 		ghc->hva = KVM_HVA_ERR_BAD;
3170 		return -EINVAL;
3171 	}
3172 
3173 	/*
3174 	 * If the requested region crosses two memslots, we still
3175 	 * verify that the entire region is valid here.
3176 	 */
3177 	for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3178 		ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3179 		ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3180 					   &nr_pages_avail);
3181 		if (kvm_is_error_hva(ghc->hva))
3182 			return -EFAULT;
3183 	}
3184 
3185 	/* Use the slow path for cross page reads and writes. */
3186 	if (nr_pages_needed == 1)
3187 		ghc->hva += offset;
3188 	else
3189 		ghc->memslot = NULL;
3190 
3191 	ghc->gpa = gpa;
3192 	ghc->len = len;
3193 	return 0;
3194 }
3195 
kvm_gfn_to_hva_cache_init(struct kvm * kvm,struct gfn_to_hva_cache * ghc,gpa_t gpa,unsigned long len)3196 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3197 			      gpa_t gpa, unsigned long len)
3198 {
3199 	struct kvm_memslots *slots = kvm_memslots(kvm);
3200 	return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3201 }
3202 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3203 
kvm_write_guest_offset_cached(struct kvm * kvm,struct gfn_to_hva_cache * ghc,void * data,unsigned int offset,unsigned long len)3204 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3205 				  void *data, unsigned int offset,
3206 				  unsigned long len)
3207 {
3208 	struct kvm_memslots *slots = kvm_memslots(kvm);
3209 	int r;
3210 	gpa_t gpa = ghc->gpa + offset;
3211 
3212 	if (WARN_ON_ONCE(len + offset > ghc->len))
3213 		return -EINVAL;
3214 
3215 	if (slots->generation != ghc->generation) {
3216 		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3217 			return -EFAULT;
3218 	}
3219 
3220 	if (kvm_is_error_hva(ghc->hva))
3221 		return -EFAULT;
3222 
3223 	if (unlikely(!ghc->memslot))
3224 		return kvm_write_guest(kvm, gpa, data, len);
3225 
3226 	r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3227 	if (r)
3228 		return -EFAULT;
3229 	mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3230 
3231 	return 0;
3232 }
3233 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3234 
kvm_write_guest_cached(struct kvm * kvm,struct gfn_to_hva_cache * ghc,void * data,unsigned long len)3235 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3236 			   void *data, unsigned long len)
3237 {
3238 	return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3239 }
3240 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3241 
kvm_read_guest_offset_cached(struct kvm * kvm,struct gfn_to_hva_cache * ghc,void * data,unsigned int offset,unsigned long len)3242 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3243 				 void *data, unsigned int offset,
3244 				 unsigned long len)
3245 {
3246 	struct kvm_memslots *slots = kvm_memslots(kvm);
3247 	int r;
3248 	gpa_t gpa = ghc->gpa + offset;
3249 
3250 	if (WARN_ON_ONCE(len + offset > ghc->len))
3251 		return -EINVAL;
3252 
3253 	if (slots->generation != ghc->generation) {
3254 		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3255 			return -EFAULT;
3256 	}
3257 
3258 	if (kvm_is_error_hva(ghc->hva))
3259 		return -EFAULT;
3260 
3261 	if (unlikely(!ghc->memslot))
3262 		return kvm_read_guest(kvm, gpa, data, len);
3263 
3264 	r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3265 	if (r)
3266 		return -EFAULT;
3267 
3268 	return 0;
3269 }
3270 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3271 
kvm_read_guest_cached(struct kvm * kvm,struct gfn_to_hva_cache * ghc,void * data,unsigned long len)3272 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3273 			  void *data, unsigned long len)
3274 {
3275 	return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3276 }
3277 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3278 
kvm_clear_guest(struct kvm * kvm,gpa_t gpa,unsigned long len)3279 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3280 {
3281 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3282 	gfn_t gfn = gpa >> PAGE_SHIFT;
3283 	int seg;
3284 	int offset = offset_in_page(gpa);
3285 	int ret;
3286 
3287 	while ((seg = next_segment(len, offset)) != 0) {
3288 		ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3289 		if (ret < 0)
3290 			return ret;
3291 		offset = 0;
3292 		len -= seg;
3293 		++gfn;
3294 	}
3295 	return 0;
3296 }
3297 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3298 
mark_page_dirty_in_slot(struct kvm * kvm,const struct kvm_memory_slot * memslot,gfn_t gfn)3299 void mark_page_dirty_in_slot(struct kvm *kvm,
3300 			     const struct kvm_memory_slot *memslot,
3301 		 	     gfn_t gfn)
3302 {
3303 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3304 
3305 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3306 	if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3307 		return;
3308 #endif
3309 
3310 	if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3311 		unsigned long rel_gfn = gfn - memslot->base_gfn;
3312 		u32 slot = (memslot->as_id << 16) | memslot->id;
3313 
3314 		if (kvm->dirty_ring_size)
3315 			kvm_dirty_ring_push(&vcpu->dirty_ring,
3316 					    slot, rel_gfn);
3317 		else
3318 			set_bit_le(rel_gfn, memslot->dirty_bitmap);
3319 	}
3320 }
3321 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3322 
mark_page_dirty(struct kvm * kvm,gfn_t gfn)3323 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3324 {
3325 	struct kvm_memory_slot *memslot;
3326 
3327 	memslot = gfn_to_memslot(kvm, gfn);
3328 	mark_page_dirty_in_slot(kvm, memslot, gfn);
3329 }
3330 EXPORT_SYMBOL_GPL(mark_page_dirty);
3331 
kvm_vcpu_mark_page_dirty(struct kvm_vcpu * vcpu,gfn_t gfn)3332 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3333 {
3334 	struct kvm_memory_slot *memslot;
3335 
3336 	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3337 	mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3338 }
3339 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3340 
kvm_sigset_activate(struct kvm_vcpu * vcpu)3341 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3342 {
3343 	if (!vcpu->sigset_active)
3344 		return;
3345 
3346 	/*
3347 	 * This does a lockless modification of ->real_blocked, which is fine
3348 	 * because, only current can change ->real_blocked and all readers of
3349 	 * ->real_blocked don't care as long ->real_blocked is always a subset
3350 	 * of ->blocked.
3351 	 */
3352 	sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
3353 }
3354 
kvm_sigset_deactivate(struct kvm_vcpu * vcpu)3355 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3356 {
3357 	if (!vcpu->sigset_active)
3358 		return;
3359 
3360 	sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
3361 	sigemptyset(&current->real_blocked);
3362 }
3363 
grow_halt_poll_ns(struct kvm_vcpu * vcpu)3364 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3365 {
3366 	unsigned int old, val, grow, grow_start;
3367 
3368 	old = val = vcpu->halt_poll_ns;
3369 	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3370 	grow = READ_ONCE(halt_poll_ns_grow);
3371 	if (!grow)
3372 		goto out;
3373 
3374 	val *= grow;
3375 	if (val < grow_start)
3376 		val = grow_start;
3377 
3378 	vcpu->halt_poll_ns = val;
3379 out:
3380 	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3381 }
3382 
shrink_halt_poll_ns(struct kvm_vcpu * vcpu)3383 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3384 {
3385 	unsigned int old, val, shrink, grow_start;
3386 
3387 	old = val = vcpu->halt_poll_ns;
3388 	shrink = READ_ONCE(halt_poll_ns_shrink);
3389 	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3390 	if (shrink == 0)
3391 		val = 0;
3392 	else
3393 		val /= shrink;
3394 
3395 	if (val < grow_start)
3396 		val = 0;
3397 
3398 	vcpu->halt_poll_ns = val;
3399 	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3400 }
3401 
kvm_vcpu_check_block(struct kvm_vcpu * vcpu)3402 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3403 {
3404 	int ret = -EINTR;
3405 	int idx = srcu_read_lock(&vcpu->kvm->srcu);
3406 
3407 	if (kvm_arch_vcpu_runnable(vcpu))
3408 		goto out;
3409 	if (kvm_cpu_has_pending_timer(vcpu))
3410 		goto out;
3411 	if (signal_pending(current))
3412 		goto out;
3413 	if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3414 		goto out;
3415 
3416 	ret = 0;
3417 out:
3418 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3419 	return ret;
3420 }
3421 
3422 /*
3423  * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3424  * pending.  This is mostly used when halting a vCPU, but may also be used
3425  * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3426  */
kvm_vcpu_block(struct kvm_vcpu * vcpu)3427 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3428 {
3429 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3430 	bool waited = false;
3431 
3432 	vcpu->stat.generic.blocking = 1;
3433 
3434 	preempt_disable();
3435 	kvm_arch_vcpu_blocking(vcpu);
3436 	prepare_to_rcuwait(wait);
3437 	preempt_enable();
3438 
3439 	for (;;) {
3440 		set_current_state(TASK_INTERRUPTIBLE);
3441 
3442 		if (kvm_vcpu_check_block(vcpu) < 0)
3443 			break;
3444 
3445 		waited = true;
3446 		schedule();
3447 	}
3448 
3449 	preempt_disable();
3450 	finish_rcuwait(wait);
3451 	kvm_arch_vcpu_unblocking(vcpu);
3452 	preempt_enable();
3453 
3454 	vcpu->stat.generic.blocking = 0;
3455 
3456 	return waited;
3457 }
3458 
update_halt_poll_stats(struct kvm_vcpu * vcpu,ktime_t start,ktime_t end,bool success)3459 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3460 					  ktime_t end, bool success)
3461 {
3462 	struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3463 	u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3464 
3465 	++vcpu->stat.generic.halt_attempted_poll;
3466 
3467 	if (success) {
3468 		++vcpu->stat.generic.halt_successful_poll;
3469 
3470 		if (!vcpu_valid_wakeup(vcpu))
3471 			++vcpu->stat.generic.halt_poll_invalid;
3472 
3473 		stats->halt_poll_success_ns += poll_ns;
3474 		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3475 	} else {
3476 		stats->halt_poll_fail_ns += poll_ns;
3477 		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3478 	}
3479 }
3480 
kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu * vcpu)3481 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3482 {
3483 	struct kvm *kvm = vcpu->kvm;
3484 
3485 	if (kvm->override_halt_poll_ns) {
3486 		/*
3487 		 * Ensure kvm->max_halt_poll_ns is not read before
3488 		 * kvm->override_halt_poll_ns.
3489 		 *
3490 		 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3491 		 */
3492 		smp_rmb();
3493 		return READ_ONCE(kvm->max_halt_poll_ns);
3494 	}
3495 
3496 	return READ_ONCE(halt_poll_ns);
3497 }
3498 
3499 /*
3500  * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc...  If halt
3501  * polling is enabled, busy wait for a short time before blocking to avoid the
3502  * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3503  * is halted.
3504  */
kvm_vcpu_halt(struct kvm_vcpu * vcpu)3505 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3506 {
3507 	unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3508 	bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3509 	ktime_t start, cur, poll_end;
3510 	bool waited = false;
3511 	bool do_halt_poll;
3512 	u64 halt_ns;
3513 
3514 	if (vcpu->halt_poll_ns > max_halt_poll_ns)
3515 		vcpu->halt_poll_ns = max_halt_poll_ns;
3516 
3517 	do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3518 
3519 	start = cur = poll_end = ktime_get();
3520 	if (do_halt_poll) {
3521 		ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3522 
3523 		do {
3524 			/*
3525 			 * This sets KVM_REQ_UNHALT if an interrupt
3526 			 * arrives.
3527 			 */
3528 			if (kvm_vcpu_check_block(vcpu) < 0)
3529 				goto out;
3530 			cpu_relax();
3531 			poll_end = cur = ktime_get();
3532 		} while (kvm_vcpu_can_poll(cur, stop));
3533 	}
3534 
3535 	waited = kvm_vcpu_block(vcpu);
3536 
3537 	cur = ktime_get();
3538 	if (waited) {
3539 		vcpu->stat.generic.halt_wait_ns +=
3540 			ktime_to_ns(cur) - ktime_to_ns(poll_end);
3541 		KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3542 				ktime_to_ns(cur) - ktime_to_ns(poll_end));
3543 	}
3544 out:
3545 	/* The total time the vCPU was "halted", including polling time. */
3546 	halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3547 
3548 	/*
3549 	 * Note, halt-polling is considered successful so long as the vCPU was
3550 	 * never actually scheduled out, i.e. even if the wake event arrived
3551 	 * after of the halt-polling loop itself, but before the full wait.
3552 	 */
3553 	if (do_halt_poll)
3554 		update_halt_poll_stats(vcpu, start, poll_end, !waited);
3555 
3556 	if (halt_poll_allowed) {
3557 		/* Recompute the max halt poll time in case it changed. */
3558 		max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3559 
3560 		if (!vcpu_valid_wakeup(vcpu)) {
3561 			shrink_halt_poll_ns(vcpu);
3562 		} else if (max_halt_poll_ns) {
3563 			if (halt_ns <= vcpu->halt_poll_ns)
3564 				;
3565 			/* we had a long block, shrink polling */
3566 			else if (vcpu->halt_poll_ns &&
3567 				 halt_ns > max_halt_poll_ns)
3568 				shrink_halt_poll_ns(vcpu);
3569 			/* we had a short halt and our poll time is too small */
3570 			else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3571 				 halt_ns < max_halt_poll_ns)
3572 				grow_halt_poll_ns(vcpu);
3573 		} else {
3574 			vcpu->halt_poll_ns = 0;
3575 		}
3576 	}
3577 
3578 	trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3579 }
3580 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3581 
kvm_vcpu_wake_up(struct kvm_vcpu * vcpu)3582 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3583 {
3584 	if (__kvm_vcpu_wake_up(vcpu)) {
3585 		WRITE_ONCE(vcpu->ready, true);
3586 		++vcpu->stat.generic.halt_wakeup;
3587 		return true;
3588 	}
3589 
3590 	return false;
3591 }
3592 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3593 
3594 #ifndef CONFIG_S390
3595 /*
3596  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3597  */
kvm_vcpu_kick(struct kvm_vcpu * vcpu)3598 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3599 {
3600 	int me, cpu;
3601 
3602 	if (kvm_vcpu_wake_up(vcpu))
3603 		return;
3604 
3605 	me = get_cpu();
3606 	/*
3607 	 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3608 	 * to EXITING_GUEST_MODE.  Therefore the moderately expensive "should
3609 	 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3610 	 * within the vCPU thread itself.
3611 	 */
3612 	if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3613 		if (vcpu->mode == IN_GUEST_MODE)
3614 			WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3615 		goto out;
3616 	}
3617 
3618 	/*
3619 	 * Note, the vCPU could get migrated to a different pCPU at any point
3620 	 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3621 	 * IPI to the previous pCPU.  But, that's ok because the purpose of the
3622 	 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3623 	 * vCPU also requires it to leave IN_GUEST_MODE.
3624 	 */
3625 	if (kvm_arch_vcpu_should_kick(vcpu)) {
3626 		cpu = READ_ONCE(vcpu->cpu);
3627 		if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3628 			smp_send_reschedule(cpu);
3629 	}
3630 out:
3631 	put_cpu();
3632 }
3633 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3634 #endif /* !CONFIG_S390 */
3635 
kvm_vcpu_yield_to(struct kvm_vcpu * target)3636 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3637 {
3638 	struct pid *pid;
3639 	struct task_struct *task = NULL;
3640 	int ret = 0;
3641 
3642 	rcu_read_lock();
3643 	pid = rcu_dereference(target->pid);
3644 	if (pid)
3645 		task = get_pid_task(pid, PIDTYPE_PID);
3646 	rcu_read_unlock();
3647 	if (!task)
3648 		return ret;
3649 	ret = yield_to(task, 1);
3650 	put_task_struct(task);
3651 
3652 	return ret;
3653 }
3654 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3655 
3656 /*
3657  * Helper that checks whether a VCPU is eligible for directed yield.
3658  * Most eligible candidate to yield is decided by following heuristics:
3659  *
3660  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3661  *  (preempted lock holder), indicated by @in_spin_loop.
3662  *  Set at the beginning and cleared at the end of interception/PLE handler.
3663  *
3664  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3665  *  chance last time (mostly it has become eligible now since we have probably
3666  *  yielded to lockholder in last iteration. This is done by toggling
3667  *  @dy_eligible each time a VCPU checked for eligibility.)
3668  *
3669  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3670  *  to preempted lock-holder could result in wrong VCPU selection and CPU
3671  *  burning. Giving priority for a potential lock-holder increases lock
3672  *  progress.
3673  *
3674  *  Since algorithm is based on heuristics, accessing another VCPU data without
3675  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
3676  *  and continue with next VCPU and so on.
3677  */
kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu * vcpu)3678 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3679 {
3680 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3681 	bool eligible;
3682 
3683 	eligible = !vcpu->spin_loop.in_spin_loop ||
3684 		    vcpu->spin_loop.dy_eligible;
3685 
3686 	if (vcpu->spin_loop.in_spin_loop)
3687 		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3688 
3689 	return eligible;
3690 #else
3691 	return true;
3692 #endif
3693 }
3694 
3695 /*
3696  * Unlike kvm_arch_vcpu_runnable, this function is called outside
3697  * a vcpu_load/vcpu_put pair.  However, for most architectures
3698  * kvm_arch_vcpu_runnable does not require vcpu_load.
3699  */
kvm_arch_dy_runnable(struct kvm_vcpu * vcpu)3700 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3701 {
3702 	return kvm_arch_vcpu_runnable(vcpu);
3703 }
3704 
vcpu_dy_runnable(struct kvm_vcpu * vcpu)3705 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3706 {
3707 	if (kvm_arch_dy_runnable(vcpu))
3708 		return true;
3709 
3710 #ifdef CONFIG_KVM_ASYNC_PF
3711 	if (!list_empty_careful(&vcpu->async_pf.done))
3712 		return true;
3713 #endif
3714 
3715 	return false;
3716 }
3717 
kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu * vcpu)3718 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3719 {
3720 	return false;
3721 }
3722 
kvm_vcpu_on_spin(struct kvm_vcpu * me,bool yield_to_kernel_mode)3723 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3724 {
3725 	struct kvm *kvm = me->kvm;
3726 	struct kvm_vcpu *vcpu;
3727 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3728 	unsigned long i;
3729 	int yielded = 0;
3730 	int try = 3;
3731 	int pass;
3732 
3733 	kvm_vcpu_set_in_spin_loop(me, true);
3734 	/*
3735 	 * We boost the priority of a VCPU that is runnable but not
3736 	 * currently running, because it got preempted by something
3737 	 * else and called schedule in __vcpu_run.  Hopefully that
3738 	 * VCPU is holding the lock that we need and will release it.
3739 	 * We approximate round-robin by starting at the last boosted VCPU.
3740 	 */
3741 	for (pass = 0; pass < 2 && !yielded && try; pass++) {
3742 		kvm_for_each_vcpu(i, vcpu, kvm) {
3743 			if (!pass && i <= last_boosted_vcpu) {
3744 				i = last_boosted_vcpu;
3745 				continue;
3746 			} else if (pass && i > last_boosted_vcpu)
3747 				break;
3748 			if (!READ_ONCE(vcpu->ready))
3749 				continue;
3750 			if (vcpu == me)
3751 				continue;
3752 			if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3753 				continue;
3754 			if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3755 			    !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3756 			    !kvm_arch_vcpu_in_kernel(vcpu))
3757 				continue;
3758 			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3759 				continue;
3760 
3761 			yielded = kvm_vcpu_yield_to(vcpu);
3762 			if (yielded > 0) {
3763 				kvm->last_boosted_vcpu = i;
3764 				break;
3765 			} else if (yielded < 0) {
3766 				try--;
3767 				if (!try)
3768 					break;
3769 			}
3770 		}
3771 	}
3772 	kvm_vcpu_set_in_spin_loop(me, false);
3773 
3774 	/* Ensure vcpu is not eligible during next spinloop */
3775 	kvm_vcpu_set_dy_eligible(me, false);
3776 }
3777 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3778 
kvm_page_in_dirty_ring(struct kvm * kvm,unsigned long pgoff)3779 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3780 {
3781 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3782 	return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3783 	    (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3784 	     kvm->dirty_ring_size / PAGE_SIZE);
3785 #else
3786 	return false;
3787 #endif
3788 }
3789 
kvm_vcpu_fault(struct vm_fault * vmf)3790 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3791 {
3792 	struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3793 	struct page *page;
3794 
3795 	if (vmf->pgoff == 0)
3796 		page = virt_to_page(vcpu->run);
3797 #ifdef CONFIG_X86
3798 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3799 		page = virt_to_page(vcpu->arch.pio_data);
3800 #endif
3801 #ifdef CONFIG_KVM_MMIO
3802 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3803 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3804 #endif
3805 	else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3806 		page = kvm_dirty_ring_get_page(
3807 		    &vcpu->dirty_ring,
3808 		    vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3809 	else
3810 		return kvm_arch_vcpu_fault(vcpu, vmf);
3811 	get_page(page);
3812 	vmf->page = page;
3813 	return 0;
3814 }
3815 
3816 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3817 	.fault = kvm_vcpu_fault,
3818 };
3819 
kvm_vcpu_mmap(struct file * file,struct vm_area_struct * vma)3820 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3821 {
3822 	struct kvm_vcpu *vcpu = file->private_data;
3823 	unsigned long pages = vma_pages(vma);
3824 
3825 	if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3826 	     kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3827 	    ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3828 		return -EINVAL;
3829 
3830 	vma->vm_ops = &kvm_vcpu_vm_ops;
3831 	return 0;
3832 }
3833 
kvm_vcpu_release(struct inode * inode,struct file * filp)3834 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3835 {
3836 	struct kvm_vcpu *vcpu = filp->private_data;
3837 
3838 	kvm_put_kvm(vcpu->kvm);
3839 	return 0;
3840 }
3841 
3842 static const struct file_operations kvm_vcpu_fops = {
3843 	.release        = kvm_vcpu_release,
3844 	.unlocked_ioctl = kvm_vcpu_ioctl,
3845 	.mmap           = kvm_vcpu_mmap,
3846 	.llseek		= noop_llseek,
3847 	KVM_COMPAT(kvm_vcpu_compat_ioctl),
3848 };
3849 
3850 /*
3851  * Allocates an inode for the vcpu.
3852  */
create_vcpu_fd(struct kvm_vcpu * vcpu)3853 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3854 {
3855 	char name[8 + 1 + ITOA_MAX_LEN + 1];
3856 
3857 	snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3858 	return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3859 }
3860 
3861 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
vcpu_get_pid(void * data,u64 * val)3862 static int vcpu_get_pid(void *data, u64 *val)
3863 {
3864 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data;
3865 	*val = pid_nr(rcu_access_pointer(vcpu->pid));
3866 	return 0;
3867 }
3868 
3869 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3870 
kvm_create_vcpu_debugfs(struct kvm_vcpu * vcpu)3871 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3872 {
3873 	struct dentry *debugfs_dentry;
3874 	char dir_name[ITOA_MAX_LEN * 2];
3875 
3876 	if (!debugfs_initialized())
3877 		return;
3878 
3879 	snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3880 	debugfs_dentry = debugfs_create_dir(dir_name,
3881 					    vcpu->kvm->debugfs_dentry);
3882 	debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3883 			    &vcpu_get_pid_fops);
3884 
3885 	kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3886 }
3887 #endif
3888 
3889 /*
3890  * Creates some virtual cpus.  Good luck creating more than one.
3891  */
kvm_vm_ioctl_create_vcpu(struct kvm * kvm,u32 id)3892 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3893 {
3894 	int r;
3895 	struct kvm_vcpu *vcpu;
3896 	struct page *page;
3897 
3898 	if (id >= KVM_MAX_VCPU_IDS)
3899 		return -EINVAL;
3900 
3901 	mutex_lock(&kvm->lock);
3902 	if (kvm->created_vcpus >= kvm->max_vcpus) {
3903 		mutex_unlock(&kvm->lock);
3904 		return -EINVAL;
3905 	}
3906 
3907 	r = kvm_arch_vcpu_precreate(kvm, id);
3908 	if (r) {
3909 		mutex_unlock(&kvm->lock);
3910 		return r;
3911 	}
3912 
3913 	kvm->created_vcpus++;
3914 	mutex_unlock(&kvm->lock);
3915 
3916 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3917 	if (!vcpu) {
3918 		r = -ENOMEM;
3919 		goto vcpu_decrement;
3920 	}
3921 
3922 	BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3923 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3924 	if (!page) {
3925 		r = -ENOMEM;
3926 		goto vcpu_free;
3927 	}
3928 	vcpu->run = page_address(page);
3929 
3930 	kvm_vcpu_init(vcpu, kvm, id);
3931 
3932 	r = kvm_arch_vcpu_create(vcpu);
3933 	if (r)
3934 		goto vcpu_free_run_page;
3935 
3936 	if (kvm->dirty_ring_size) {
3937 		r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3938 					 id, kvm->dirty_ring_size);
3939 		if (r)
3940 			goto arch_vcpu_destroy;
3941 	}
3942 
3943 	mutex_lock(&kvm->lock);
3944 	if (kvm_get_vcpu_by_id(kvm, id)) {
3945 		r = -EEXIST;
3946 		goto unlock_vcpu_destroy;
3947 	}
3948 
3949 	vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3950 	r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3951 	BUG_ON(r == -EBUSY);
3952 	if (r)
3953 		goto unlock_vcpu_destroy;
3954 
3955 	/* Now it's all set up, let userspace reach it */
3956 	kvm_get_kvm(kvm);
3957 	r = create_vcpu_fd(vcpu);
3958 	if (r < 0) {
3959 		xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3960 		kvm_put_kvm_no_destroy(kvm);
3961 		goto unlock_vcpu_destroy;
3962 	}
3963 
3964 	/*
3965 	 * Pairs with smp_rmb() in kvm_get_vcpu.  Store the vcpu
3966 	 * pointer before kvm->online_vcpu's incremented value.
3967 	 */
3968 	smp_wmb();
3969 	atomic_inc(&kvm->online_vcpus);
3970 
3971 	mutex_unlock(&kvm->lock);
3972 	kvm_arch_vcpu_postcreate(vcpu);
3973 	kvm_create_vcpu_debugfs(vcpu);
3974 	return r;
3975 
3976 unlock_vcpu_destroy:
3977 	mutex_unlock(&kvm->lock);
3978 	kvm_dirty_ring_free(&vcpu->dirty_ring);
3979 arch_vcpu_destroy:
3980 	kvm_arch_vcpu_destroy(vcpu);
3981 vcpu_free_run_page:
3982 	free_page((unsigned long)vcpu->run);
3983 vcpu_free:
3984 	kmem_cache_free(kvm_vcpu_cache, vcpu);
3985 vcpu_decrement:
3986 	mutex_lock(&kvm->lock);
3987 	kvm->created_vcpus--;
3988 	mutex_unlock(&kvm->lock);
3989 	return r;
3990 }
3991 
kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu * vcpu,sigset_t * sigset)3992 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3993 {
3994 	if (sigset) {
3995 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3996 		vcpu->sigset_active = 1;
3997 		vcpu->sigset = *sigset;
3998 	} else
3999 		vcpu->sigset_active = 0;
4000 	return 0;
4001 }
4002 
kvm_vcpu_stats_read(struct file * file,char __user * user_buffer,size_t size,loff_t * offset)4003 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4004 			      size_t size, loff_t *offset)
4005 {
4006 	struct kvm_vcpu *vcpu = file->private_data;
4007 
4008 	return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4009 			&kvm_vcpu_stats_desc[0], &vcpu->stat,
4010 			sizeof(vcpu->stat), user_buffer, size, offset);
4011 }
4012 
4013 static const struct file_operations kvm_vcpu_stats_fops = {
4014 	.read = kvm_vcpu_stats_read,
4015 	.llseek = noop_llseek,
4016 };
4017 
kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu * vcpu)4018 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4019 {
4020 	int fd;
4021 	struct file *file;
4022 	char name[15 + ITOA_MAX_LEN + 1];
4023 
4024 	snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4025 
4026 	fd = get_unused_fd_flags(O_CLOEXEC);
4027 	if (fd < 0)
4028 		return fd;
4029 
4030 	file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4031 	if (IS_ERR(file)) {
4032 		put_unused_fd(fd);
4033 		return PTR_ERR(file);
4034 	}
4035 	file->f_mode |= FMODE_PREAD;
4036 	fd_install(fd, file);
4037 
4038 	return fd;
4039 }
4040 
kvm_vcpu_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)4041 static long kvm_vcpu_ioctl(struct file *filp,
4042 			   unsigned int ioctl, unsigned long arg)
4043 {
4044 	struct kvm_vcpu *vcpu = filp->private_data;
4045 	void __user *argp = (void __user *)arg;
4046 	int r;
4047 	struct kvm_fpu *fpu = NULL;
4048 	struct kvm_sregs *kvm_sregs = NULL;
4049 
4050 	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4051 		return -EIO;
4052 
4053 	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4054 		return -EINVAL;
4055 
4056 	/*
4057 	 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4058 	 * execution; mutex_lock() would break them.
4059 	 */
4060 	r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4061 	if (r != -ENOIOCTLCMD)
4062 		return r;
4063 
4064 	if (mutex_lock_killable(&vcpu->mutex))
4065 		return -EINTR;
4066 	switch (ioctl) {
4067 	case KVM_RUN: {
4068 		struct pid *oldpid;
4069 		r = -EINVAL;
4070 		if (arg)
4071 			goto out;
4072 		oldpid = rcu_access_pointer(vcpu->pid);
4073 		if (unlikely(oldpid != task_pid(current))) {
4074 			/* The thread running this VCPU changed. */
4075 			struct pid *newpid;
4076 
4077 			r = kvm_arch_vcpu_run_pid_change(vcpu);
4078 			if (r)
4079 				break;
4080 
4081 			newpid = get_task_pid(current, PIDTYPE_PID);
4082 			rcu_assign_pointer(vcpu->pid, newpid);
4083 			if (oldpid)
4084 				synchronize_rcu();
4085 			put_pid(oldpid);
4086 		}
4087 		r = kvm_arch_vcpu_ioctl_run(vcpu);
4088 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4089 		break;
4090 	}
4091 	case KVM_GET_REGS: {
4092 		struct kvm_regs *kvm_regs;
4093 
4094 		r = -ENOMEM;
4095 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4096 		if (!kvm_regs)
4097 			goto out;
4098 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4099 		if (r)
4100 			goto out_free1;
4101 		r = -EFAULT;
4102 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4103 			goto out_free1;
4104 		r = 0;
4105 out_free1:
4106 		kfree(kvm_regs);
4107 		break;
4108 	}
4109 	case KVM_SET_REGS: {
4110 		struct kvm_regs *kvm_regs;
4111 
4112 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4113 		if (IS_ERR(kvm_regs)) {
4114 			r = PTR_ERR(kvm_regs);
4115 			goto out;
4116 		}
4117 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4118 		kfree(kvm_regs);
4119 		break;
4120 	}
4121 	case KVM_GET_SREGS: {
4122 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4123 				    GFP_KERNEL_ACCOUNT);
4124 		r = -ENOMEM;
4125 		if (!kvm_sregs)
4126 			goto out;
4127 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4128 		if (r)
4129 			goto out;
4130 		r = -EFAULT;
4131 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4132 			goto out;
4133 		r = 0;
4134 		break;
4135 	}
4136 	case KVM_SET_SREGS: {
4137 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4138 		if (IS_ERR(kvm_sregs)) {
4139 			r = PTR_ERR(kvm_sregs);
4140 			kvm_sregs = NULL;
4141 			goto out;
4142 		}
4143 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4144 		break;
4145 	}
4146 	case KVM_GET_MP_STATE: {
4147 		struct kvm_mp_state mp_state;
4148 
4149 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4150 		if (r)
4151 			goto out;
4152 		r = -EFAULT;
4153 		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4154 			goto out;
4155 		r = 0;
4156 		break;
4157 	}
4158 	case KVM_SET_MP_STATE: {
4159 		struct kvm_mp_state mp_state;
4160 
4161 		r = -EFAULT;
4162 		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4163 			goto out;
4164 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4165 		break;
4166 	}
4167 	case KVM_TRANSLATE: {
4168 		struct kvm_translation tr;
4169 
4170 		r = -EFAULT;
4171 		if (copy_from_user(&tr, argp, sizeof(tr)))
4172 			goto out;
4173 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4174 		if (r)
4175 			goto out;
4176 		r = -EFAULT;
4177 		if (copy_to_user(argp, &tr, sizeof(tr)))
4178 			goto out;
4179 		r = 0;
4180 		break;
4181 	}
4182 	case KVM_SET_GUEST_DEBUG: {
4183 		struct kvm_guest_debug dbg;
4184 
4185 		r = -EFAULT;
4186 		if (copy_from_user(&dbg, argp, sizeof(dbg)))
4187 			goto out;
4188 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4189 		break;
4190 	}
4191 	case KVM_SET_SIGNAL_MASK: {
4192 		struct kvm_signal_mask __user *sigmask_arg = argp;
4193 		struct kvm_signal_mask kvm_sigmask;
4194 		sigset_t sigset, *p;
4195 
4196 		p = NULL;
4197 		if (argp) {
4198 			r = -EFAULT;
4199 			if (copy_from_user(&kvm_sigmask, argp,
4200 					   sizeof(kvm_sigmask)))
4201 				goto out;
4202 			r = -EINVAL;
4203 			if (kvm_sigmask.len != sizeof(sigset))
4204 				goto out;
4205 			r = -EFAULT;
4206 			if (copy_from_user(&sigset, sigmask_arg->sigset,
4207 					   sizeof(sigset)))
4208 				goto out;
4209 			p = &sigset;
4210 		}
4211 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4212 		break;
4213 	}
4214 	case KVM_GET_FPU: {
4215 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4216 		r = -ENOMEM;
4217 		if (!fpu)
4218 			goto out;
4219 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4220 		if (r)
4221 			goto out;
4222 		r = -EFAULT;
4223 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4224 			goto out;
4225 		r = 0;
4226 		break;
4227 	}
4228 	case KVM_SET_FPU: {
4229 		fpu = memdup_user(argp, sizeof(*fpu));
4230 		if (IS_ERR(fpu)) {
4231 			r = PTR_ERR(fpu);
4232 			fpu = NULL;
4233 			goto out;
4234 		}
4235 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4236 		break;
4237 	}
4238 	case KVM_GET_STATS_FD: {
4239 		r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4240 		break;
4241 	}
4242 	default:
4243 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4244 	}
4245 out:
4246 	mutex_unlock(&vcpu->mutex);
4247 	kfree(fpu);
4248 	kfree(kvm_sregs);
4249 	return r;
4250 }
4251 
4252 #ifdef CONFIG_KVM_COMPAT
kvm_vcpu_compat_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)4253 static long kvm_vcpu_compat_ioctl(struct file *filp,
4254 				  unsigned int ioctl, unsigned long arg)
4255 {
4256 	struct kvm_vcpu *vcpu = filp->private_data;
4257 	void __user *argp = compat_ptr(arg);
4258 	int r;
4259 
4260 	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4261 		return -EIO;
4262 
4263 	switch (ioctl) {
4264 	case KVM_SET_SIGNAL_MASK: {
4265 		struct kvm_signal_mask __user *sigmask_arg = argp;
4266 		struct kvm_signal_mask kvm_sigmask;
4267 		sigset_t sigset;
4268 
4269 		if (argp) {
4270 			r = -EFAULT;
4271 			if (copy_from_user(&kvm_sigmask, argp,
4272 					   sizeof(kvm_sigmask)))
4273 				goto out;
4274 			r = -EINVAL;
4275 			if (kvm_sigmask.len != sizeof(compat_sigset_t))
4276 				goto out;
4277 			r = -EFAULT;
4278 			if (get_compat_sigset(&sigset,
4279 					      (compat_sigset_t __user *)sigmask_arg->sigset))
4280 				goto out;
4281 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4282 		} else
4283 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4284 		break;
4285 	}
4286 	default:
4287 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
4288 	}
4289 
4290 out:
4291 	return r;
4292 }
4293 #endif
4294 
kvm_device_mmap(struct file * filp,struct vm_area_struct * vma)4295 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4296 {
4297 	struct kvm_device *dev = filp->private_data;
4298 
4299 	if (dev->ops->mmap)
4300 		return dev->ops->mmap(dev, vma);
4301 
4302 	return -ENODEV;
4303 }
4304 
kvm_device_ioctl_attr(struct kvm_device * dev,int (* accessor)(struct kvm_device * dev,struct kvm_device_attr * attr),unsigned long arg)4305 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4306 				 int (*accessor)(struct kvm_device *dev,
4307 						 struct kvm_device_attr *attr),
4308 				 unsigned long arg)
4309 {
4310 	struct kvm_device_attr attr;
4311 
4312 	if (!accessor)
4313 		return -EPERM;
4314 
4315 	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4316 		return -EFAULT;
4317 
4318 	return accessor(dev, &attr);
4319 }
4320 
kvm_device_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)4321 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4322 			     unsigned long arg)
4323 {
4324 	struct kvm_device *dev = filp->private_data;
4325 
4326 	if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4327 		return -EIO;
4328 
4329 	switch (ioctl) {
4330 	case KVM_SET_DEVICE_ATTR:
4331 		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4332 	case KVM_GET_DEVICE_ATTR:
4333 		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4334 	case KVM_HAS_DEVICE_ATTR:
4335 		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4336 	default:
4337 		if (dev->ops->ioctl)
4338 			return dev->ops->ioctl(dev, ioctl, arg);
4339 
4340 		return -ENOTTY;
4341 	}
4342 }
4343 
kvm_device_release(struct inode * inode,struct file * filp)4344 static int kvm_device_release(struct inode *inode, struct file *filp)
4345 {
4346 	struct kvm_device *dev = filp->private_data;
4347 	struct kvm *kvm = dev->kvm;
4348 
4349 	if (dev->ops->release) {
4350 		mutex_lock(&kvm->lock);
4351 		list_del(&dev->vm_node);
4352 		dev->ops->release(dev);
4353 		mutex_unlock(&kvm->lock);
4354 	}
4355 
4356 	kvm_put_kvm(kvm);
4357 	return 0;
4358 }
4359 
4360 static const struct file_operations kvm_device_fops = {
4361 	.unlocked_ioctl = kvm_device_ioctl,
4362 	.release = kvm_device_release,
4363 	KVM_COMPAT(kvm_device_ioctl),
4364 	.mmap = kvm_device_mmap,
4365 };
4366 
kvm_device_from_filp(struct file * filp)4367 struct kvm_device *kvm_device_from_filp(struct file *filp)
4368 {
4369 	if (filp->f_op != &kvm_device_fops)
4370 		return NULL;
4371 
4372 	return filp->private_data;
4373 }
4374 
4375 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4376 #ifdef CONFIG_KVM_MPIC
4377 	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
4378 	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
4379 #endif
4380 };
4381 
kvm_register_device_ops(const struct kvm_device_ops * ops,u32 type)4382 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4383 {
4384 	if (type >= ARRAY_SIZE(kvm_device_ops_table))
4385 		return -ENOSPC;
4386 
4387 	if (kvm_device_ops_table[type] != NULL)
4388 		return -EEXIST;
4389 
4390 	kvm_device_ops_table[type] = ops;
4391 	return 0;
4392 }
4393 
kvm_unregister_device_ops(u32 type)4394 void kvm_unregister_device_ops(u32 type)
4395 {
4396 	if (kvm_device_ops_table[type] != NULL)
4397 		kvm_device_ops_table[type] = NULL;
4398 }
4399 
kvm_ioctl_create_device(struct kvm * kvm,struct kvm_create_device * cd)4400 static int kvm_ioctl_create_device(struct kvm *kvm,
4401 				   struct kvm_create_device *cd)
4402 {
4403 	const struct kvm_device_ops *ops;
4404 	struct kvm_device *dev;
4405 	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4406 	int type;
4407 	int ret;
4408 
4409 	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4410 		return -ENODEV;
4411 
4412 	type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4413 	ops = kvm_device_ops_table[type];
4414 	if (ops == NULL)
4415 		return -ENODEV;
4416 
4417 	if (test)
4418 		return 0;
4419 
4420 	dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4421 	if (!dev)
4422 		return -ENOMEM;
4423 
4424 	dev->ops = ops;
4425 	dev->kvm = kvm;
4426 
4427 	mutex_lock(&kvm->lock);
4428 	ret = ops->create(dev, type);
4429 	if (ret < 0) {
4430 		mutex_unlock(&kvm->lock);
4431 		kfree(dev);
4432 		return ret;
4433 	}
4434 	list_add(&dev->vm_node, &kvm->devices);
4435 	mutex_unlock(&kvm->lock);
4436 
4437 	if (ops->init)
4438 		ops->init(dev);
4439 
4440 	kvm_get_kvm(kvm);
4441 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4442 	if (ret < 0) {
4443 		kvm_put_kvm_no_destroy(kvm);
4444 		mutex_lock(&kvm->lock);
4445 		list_del(&dev->vm_node);
4446 		if (ops->release)
4447 			ops->release(dev);
4448 		mutex_unlock(&kvm->lock);
4449 		if (ops->destroy)
4450 			ops->destroy(dev);
4451 		return ret;
4452 	}
4453 
4454 	cd->fd = ret;
4455 	return 0;
4456 }
4457 
kvm_vm_ioctl_check_extension_generic(struct kvm * kvm,long arg)4458 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4459 {
4460 	switch (arg) {
4461 	case KVM_CAP_USER_MEMORY:
4462 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4463 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4464 	case KVM_CAP_INTERNAL_ERROR_DATA:
4465 #ifdef CONFIG_HAVE_KVM_MSI
4466 	case KVM_CAP_SIGNAL_MSI:
4467 #endif
4468 #ifdef CONFIG_HAVE_KVM_IRQFD
4469 	case KVM_CAP_IRQFD:
4470 	case KVM_CAP_IRQFD_RESAMPLE:
4471 #endif
4472 	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4473 	case KVM_CAP_CHECK_EXTENSION_VM:
4474 	case KVM_CAP_ENABLE_CAP_VM:
4475 	case KVM_CAP_HALT_POLL:
4476 		return 1;
4477 #ifdef CONFIG_KVM_MMIO
4478 	case KVM_CAP_COALESCED_MMIO:
4479 		return KVM_COALESCED_MMIO_PAGE_OFFSET;
4480 	case KVM_CAP_COALESCED_PIO:
4481 		return 1;
4482 #endif
4483 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4484 	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4485 		return KVM_DIRTY_LOG_MANUAL_CAPS;
4486 #endif
4487 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4488 	case KVM_CAP_IRQ_ROUTING:
4489 		return KVM_MAX_IRQ_ROUTES;
4490 #endif
4491 #if KVM_ADDRESS_SPACE_NUM > 1
4492 	case KVM_CAP_MULTI_ADDRESS_SPACE:
4493 		return KVM_ADDRESS_SPACE_NUM;
4494 #endif
4495 	case KVM_CAP_NR_MEMSLOTS:
4496 		return KVM_USER_MEM_SLOTS;
4497 	case KVM_CAP_DIRTY_LOG_RING:
4498 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4499 		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4500 #else
4501 		return 0;
4502 #endif
4503 	case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4504 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4505 		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4506 #else
4507 		return 0;
4508 #endif
4509 	case KVM_CAP_BINARY_STATS_FD:
4510 	case KVM_CAP_SYSTEM_EVENT_DATA:
4511 		return 1;
4512 	default:
4513 		break;
4514 	}
4515 	return kvm_vm_ioctl_check_extension(kvm, arg);
4516 }
4517 
kvm_vm_ioctl_enable_dirty_log_ring(struct kvm * kvm,u32 size)4518 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4519 {
4520 	int r;
4521 
4522 	if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4523 		return -EINVAL;
4524 
4525 	/* the size should be power of 2 */
4526 	if (!size || (size & (size - 1)))
4527 		return -EINVAL;
4528 
4529 	/* Should be bigger to keep the reserved entries, or a page */
4530 	if (size < kvm_dirty_ring_get_rsvd_entries() *
4531 	    sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4532 		return -EINVAL;
4533 
4534 	if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4535 	    sizeof(struct kvm_dirty_gfn))
4536 		return -E2BIG;
4537 
4538 	/* We only allow it to set once */
4539 	if (kvm->dirty_ring_size)
4540 		return -EINVAL;
4541 
4542 	mutex_lock(&kvm->lock);
4543 
4544 	if (kvm->created_vcpus) {
4545 		/* We don't allow to change this value after vcpu created */
4546 		r = -EINVAL;
4547 	} else {
4548 		kvm->dirty_ring_size = size;
4549 		r = 0;
4550 	}
4551 
4552 	mutex_unlock(&kvm->lock);
4553 	return r;
4554 }
4555 
kvm_vm_ioctl_reset_dirty_pages(struct kvm * kvm)4556 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4557 {
4558 	unsigned long i;
4559 	struct kvm_vcpu *vcpu;
4560 	int cleared = 0;
4561 
4562 	if (!kvm->dirty_ring_size)
4563 		return -EINVAL;
4564 
4565 	mutex_lock(&kvm->slots_lock);
4566 
4567 	kvm_for_each_vcpu(i, vcpu, kvm)
4568 		cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4569 
4570 	mutex_unlock(&kvm->slots_lock);
4571 
4572 	if (cleared)
4573 		kvm_flush_remote_tlbs(kvm);
4574 
4575 	return cleared;
4576 }
4577 
kvm_vm_ioctl_enable_cap(struct kvm * kvm,struct kvm_enable_cap * cap)4578 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4579 						  struct kvm_enable_cap *cap)
4580 {
4581 	return -EINVAL;
4582 }
4583 
kvm_vm_ioctl_enable_cap_generic(struct kvm * kvm,struct kvm_enable_cap * cap)4584 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4585 					   struct kvm_enable_cap *cap)
4586 {
4587 	switch (cap->cap) {
4588 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4589 	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4590 		u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4591 
4592 		if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4593 			allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4594 
4595 		if (cap->flags || (cap->args[0] & ~allowed_options))
4596 			return -EINVAL;
4597 		kvm->manual_dirty_log_protect = cap->args[0];
4598 		return 0;
4599 	}
4600 #endif
4601 	case KVM_CAP_HALT_POLL: {
4602 		if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4603 			return -EINVAL;
4604 
4605 		kvm->max_halt_poll_ns = cap->args[0];
4606 
4607 		/*
4608 		 * Ensure kvm->override_halt_poll_ns does not become visible
4609 		 * before kvm->max_halt_poll_ns.
4610 		 *
4611 		 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4612 		 */
4613 		smp_wmb();
4614 		kvm->override_halt_poll_ns = true;
4615 
4616 		return 0;
4617 	}
4618 	case KVM_CAP_DIRTY_LOG_RING:
4619 	case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4620 		if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
4621 			return -EINVAL;
4622 
4623 		return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4624 	default:
4625 		return kvm_vm_ioctl_enable_cap(kvm, cap);
4626 	}
4627 }
4628 
kvm_vm_stats_read(struct file * file,char __user * user_buffer,size_t size,loff_t * offset)4629 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4630 			      size_t size, loff_t *offset)
4631 {
4632 	struct kvm *kvm = file->private_data;
4633 
4634 	return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4635 				&kvm_vm_stats_desc[0], &kvm->stat,
4636 				sizeof(kvm->stat), user_buffer, size, offset);
4637 }
4638 
4639 static const struct file_operations kvm_vm_stats_fops = {
4640 	.read = kvm_vm_stats_read,
4641 	.llseek = noop_llseek,
4642 };
4643 
kvm_vm_ioctl_get_stats_fd(struct kvm * kvm)4644 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4645 {
4646 	int fd;
4647 	struct file *file;
4648 
4649 	fd = get_unused_fd_flags(O_CLOEXEC);
4650 	if (fd < 0)
4651 		return fd;
4652 
4653 	file = anon_inode_getfile("kvm-vm-stats",
4654 			&kvm_vm_stats_fops, kvm, O_RDONLY);
4655 	if (IS_ERR(file)) {
4656 		put_unused_fd(fd);
4657 		return PTR_ERR(file);
4658 	}
4659 	file->f_mode |= FMODE_PREAD;
4660 	fd_install(fd, file);
4661 
4662 	return fd;
4663 }
4664 
kvm_vm_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)4665 static long kvm_vm_ioctl(struct file *filp,
4666 			   unsigned int ioctl, unsigned long arg)
4667 {
4668 	struct kvm *kvm = filp->private_data;
4669 	void __user *argp = (void __user *)arg;
4670 	int r;
4671 
4672 	if (kvm->mm != current->mm || kvm->vm_dead)
4673 		return -EIO;
4674 	switch (ioctl) {
4675 	case KVM_CREATE_VCPU:
4676 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4677 		break;
4678 	case KVM_ENABLE_CAP: {
4679 		struct kvm_enable_cap cap;
4680 
4681 		r = -EFAULT;
4682 		if (copy_from_user(&cap, argp, sizeof(cap)))
4683 			goto out;
4684 		r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4685 		break;
4686 	}
4687 	case KVM_SET_USER_MEMORY_REGION: {
4688 		struct kvm_userspace_memory_region kvm_userspace_mem;
4689 
4690 		r = -EFAULT;
4691 		if (copy_from_user(&kvm_userspace_mem, argp,
4692 						sizeof(kvm_userspace_mem)))
4693 			goto out;
4694 
4695 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4696 		break;
4697 	}
4698 	case KVM_GET_DIRTY_LOG: {
4699 		struct kvm_dirty_log log;
4700 
4701 		r = -EFAULT;
4702 		if (copy_from_user(&log, argp, sizeof(log)))
4703 			goto out;
4704 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4705 		break;
4706 	}
4707 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4708 	case KVM_CLEAR_DIRTY_LOG: {
4709 		struct kvm_clear_dirty_log log;
4710 
4711 		r = -EFAULT;
4712 		if (copy_from_user(&log, argp, sizeof(log)))
4713 			goto out;
4714 		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4715 		break;
4716 	}
4717 #endif
4718 #ifdef CONFIG_KVM_MMIO
4719 	case KVM_REGISTER_COALESCED_MMIO: {
4720 		struct kvm_coalesced_mmio_zone zone;
4721 
4722 		r = -EFAULT;
4723 		if (copy_from_user(&zone, argp, sizeof(zone)))
4724 			goto out;
4725 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4726 		break;
4727 	}
4728 	case KVM_UNREGISTER_COALESCED_MMIO: {
4729 		struct kvm_coalesced_mmio_zone zone;
4730 
4731 		r = -EFAULT;
4732 		if (copy_from_user(&zone, argp, sizeof(zone)))
4733 			goto out;
4734 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4735 		break;
4736 	}
4737 #endif
4738 	case KVM_IRQFD: {
4739 		struct kvm_irqfd data;
4740 
4741 		r = -EFAULT;
4742 		if (copy_from_user(&data, argp, sizeof(data)))
4743 			goto out;
4744 		r = kvm_irqfd(kvm, &data);
4745 		break;
4746 	}
4747 	case KVM_IOEVENTFD: {
4748 		struct kvm_ioeventfd data;
4749 
4750 		r = -EFAULT;
4751 		if (copy_from_user(&data, argp, sizeof(data)))
4752 			goto out;
4753 		r = kvm_ioeventfd(kvm, &data);
4754 		break;
4755 	}
4756 #ifdef CONFIG_HAVE_KVM_MSI
4757 	case KVM_SIGNAL_MSI: {
4758 		struct kvm_msi msi;
4759 
4760 		r = -EFAULT;
4761 		if (copy_from_user(&msi, argp, sizeof(msi)))
4762 			goto out;
4763 		r = kvm_send_userspace_msi(kvm, &msi);
4764 		break;
4765 	}
4766 #endif
4767 #ifdef __KVM_HAVE_IRQ_LINE
4768 	case KVM_IRQ_LINE_STATUS:
4769 	case KVM_IRQ_LINE: {
4770 		struct kvm_irq_level irq_event;
4771 
4772 		r = -EFAULT;
4773 		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4774 			goto out;
4775 
4776 		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4777 					ioctl == KVM_IRQ_LINE_STATUS);
4778 		if (r)
4779 			goto out;
4780 
4781 		r = -EFAULT;
4782 		if (ioctl == KVM_IRQ_LINE_STATUS) {
4783 			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4784 				goto out;
4785 		}
4786 
4787 		r = 0;
4788 		break;
4789 	}
4790 #endif
4791 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4792 	case KVM_SET_GSI_ROUTING: {
4793 		struct kvm_irq_routing routing;
4794 		struct kvm_irq_routing __user *urouting;
4795 		struct kvm_irq_routing_entry *entries = NULL;
4796 
4797 		r = -EFAULT;
4798 		if (copy_from_user(&routing, argp, sizeof(routing)))
4799 			goto out;
4800 		r = -EINVAL;
4801 		if (!kvm_arch_can_set_irq_routing(kvm))
4802 			goto out;
4803 		if (routing.nr > KVM_MAX_IRQ_ROUTES)
4804 			goto out;
4805 		if (routing.flags)
4806 			goto out;
4807 		if (routing.nr) {
4808 			urouting = argp;
4809 			entries = vmemdup_user(urouting->entries,
4810 					       array_size(sizeof(*entries),
4811 							  routing.nr));
4812 			if (IS_ERR(entries)) {
4813 				r = PTR_ERR(entries);
4814 				goto out;
4815 			}
4816 		}
4817 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
4818 					routing.flags);
4819 		kvfree(entries);
4820 		break;
4821 	}
4822 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4823 	case KVM_CREATE_DEVICE: {
4824 		struct kvm_create_device cd;
4825 
4826 		r = -EFAULT;
4827 		if (copy_from_user(&cd, argp, sizeof(cd)))
4828 			goto out;
4829 
4830 		r = kvm_ioctl_create_device(kvm, &cd);
4831 		if (r)
4832 			goto out;
4833 
4834 		r = -EFAULT;
4835 		if (copy_to_user(argp, &cd, sizeof(cd)))
4836 			goto out;
4837 
4838 		r = 0;
4839 		break;
4840 	}
4841 	case KVM_CHECK_EXTENSION:
4842 		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4843 		break;
4844 	case KVM_RESET_DIRTY_RINGS:
4845 		r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4846 		break;
4847 	case KVM_GET_STATS_FD:
4848 		r = kvm_vm_ioctl_get_stats_fd(kvm);
4849 		break;
4850 	default:
4851 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4852 	}
4853 out:
4854 	return r;
4855 }
4856 
4857 #ifdef CONFIG_KVM_COMPAT
4858 struct compat_kvm_dirty_log {
4859 	__u32 slot;
4860 	__u32 padding1;
4861 	union {
4862 		compat_uptr_t dirty_bitmap; /* one bit per page */
4863 		__u64 padding2;
4864 	};
4865 };
4866 
4867 struct compat_kvm_clear_dirty_log {
4868 	__u32 slot;
4869 	__u32 num_pages;
4870 	__u64 first_page;
4871 	union {
4872 		compat_uptr_t dirty_bitmap; /* one bit per page */
4873 		__u64 padding2;
4874 	};
4875 };
4876 
kvm_arch_vm_compat_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)4877 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4878 				     unsigned long arg)
4879 {
4880 	return -ENOTTY;
4881 }
4882 
kvm_vm_compat_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)4883 static long kvm_vm_compat_ioctl(struct file *filp,
4884 			   unsigned int ioctl, unsigned long arg)
4885 {
4886 	struct kvm *kvm = filp->private_data;
4887 	int r;
4888 
4889 	if (kvm->mm != current->mm || kvm->vm_dead)
4890 		return -EIO;
4891 
4892 	r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
4893 	if (r != -ENOTTY)
4894 		return r;
4895 
4896 	switch (ioctl) {
4897 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4898 	case KVM_CLEAR_DIRTY_LOG: {
4899 		struct compat_kvm_clear_dirty_log compat_log;
4900 		struct kvm_clear_dirty_log log;
4901 
4902 		if (copy_from_user(&compat_log, (void __user *)arg,
4903 				   sizeof(compat_log)))
4904 			return -EFAULT;
4905 		log.slot	 = compat_log.slot;
4906 		log.num_pages	 = compat_log.num_pages;
4907 		log.first_page	 = compat_log.first_page;
4908 		log.padding2	 = compat_log.padding2;
4909 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4910 
4911 		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4912 		break;
4913 	}
4914 #endif
4915 	case KVM_GET_DIRTY_LOG: {
4916 		struct compat_kvm_dirty_log compat_log;
4917 		struct kvm_dirty_log log;
4918 
4919 		if (copy_from_user(&compat_log, (void __user *)arg,
4920 				   sizeof(compat_log)))
4921 			return -EFAULT;
4922 		log.slot	 = compat_log.slot;
4923 		log.padding1	 = compat_log.padding1;
4924 		log.padding2	 = compat_log.padding2;
4925 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4926 
4927 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4928 		break;
4929 	}
4930 	default:
4931 		r = kvm_vm_ioctl(filp, ioctl, arg);
4932 	}
4933 	return r;
4934 }
4935 #endif
4936 
4937 static const struct file_operations kvm_vm_fops = {
4938 	.release        = kvm_vm_release,
4939 	.unlocked_ioctl = kvm_vm_ioctl,
4940 	.llseek		= noop_llseek,
4941 	KVM_COMPAT(kvm_vm_compat_ioctl),
4942 };
4943 
file_is_kvm(struct file * file)4944 bool file_is_kvm(struct file *file)
4945 {
4946 	return file && file->f_op == &kvm_vm_fops;
4947 }
4948 EXPORT_SYMBOL_GPL(file_is_kvm);
4949 
kvm_dev_ioctl_create_vm(unsigned long type)4950 static int kvm_dev_ioctl_create_vm(unsigned long type)
4951 {
4952 	char fdname[ITOA_MAX_LEN + 1];
4953 	int r, fd;
4954 	struct kvm *kvm;
4955 	struct file *file;
4956 
4957 	fd = get_unused_fd_flags(O_CLOEXEC);
4958 	if (fd < 0)
4959 		return fd;
4960 
4961 	snprintf(fdname, sizeof(fdname), "%d", fd);
4962 
4963 	kvm = kvm_create_vm(type, fdname);
4964 	if (IS_ERR(kvm)) {
4965 		r = PTR_ERR(kvm);
4966 		goto put_fd;
4967 	}
4968 
4969 	file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4970 	if (IS_ERR(file)) {
4971 		r = PTR_ERR(file);
4972 		goto put_kvm;
4973 	}
4974 
4975 	/*
4976 	 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4977 	 * already set, with ->release() being kvm_vm_release().  In error
4978 	 * cases it will be called by the final fput(file) and will take
4979 	 * care of doing kvm_put_kvm(kvm).
4980 	 */
4981 	kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4982 
4983 	fd_install(fd, file);
4984 	return fd;
4985 
4986 put_kvm:
4987 	kvm_put_kvm(kvm);
4988 put_fd:
4989 	put_unused_fd(fd);
4990 	return r;
4991 }
4992 
kvm_dev_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)4993 static long kvm_dev_ioctl(struct file *filp,
4994 			  unsigned int ioctl, unsigned long arg)
4995 {
4996 	long r = -EINVAL;
4997 
4998 	switch (ioctl) {
4999 	case KVM_GET_API_VERSION:
5000 		if (arg)
5001 			goto out;
5002 		r = KVM_API_VERSION;
5003 		break;
5004 	case KVM_CREATE_VM:
5005 		r = kvm_dev_ioctl_create_vm(arg);
5006 		break;
5007 	case KVM_CHECK_EXTENSION:
5008 		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5009 		break;
5010 	case KVM_GET_VCPU_MMAP_SIZE:
5011 		if (arg)
5012 			goto out;
5013 		r = PAGE_SIZE;     /* struct kvm_run */
5014 #ifdef CONFIG_X86
5015 		r += PAGE_SIZE;    /* pio data page */
5016 #endif
5017 #ifdef CONFIG_KVM_MMIO
5018 		r += PAGE_SIZE;    /* coalesced mmio ring page */
5019 #endif
5020 		break;
5021 	case KVM_TRACE_ENABLE:
5022 	case KVM_TRACE_PAUSE:
5023 	case KVM_TRACE_DISABLE:
5024 		r = -EOPNOTSUPP;
5025 		break;
5026 	default:
5027 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
5028 	}
5029 out:
5030 	return r;
5031 }
5032 
5033 static struct file_operations kvm_chardev_ops = {
5034 	.unlocked_ioctl = kvm_dev_ioctl,
5035 	.llseek		= noop_llseek,
5036 	KVM_COMPAT(kvm_dev_ioctl),
5037 };
5038 
5039 static struct miscdevice kvm_dev = {
5040 	KVM_MINOR,
5041 	"kvm",
5042 	&kvm_chardev_ops,
5043 };
5044 
hardware_enable_nolock(void * junk)5045 static void hardware_enable_nolock(void *junk)
5046 {
5047 	int cpu = raw_smp_processor_id();
5048 	int r;
5049 
5050 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
5051 		return;
5052 
5053 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
5054 
5055 	r = kvm_arch_hardware_enable();
5056 
5057 	if (r) {
5058 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5059 		atomic_inc(&hardware_enable_failed);
5060 		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
5061 	}
5062 }
5063 
kvm_starting_cpu(unsigned int cpu)5064 static int kvm_starting_cpu(unsigned int cpu)
5065 {
5066 	raw_spin_lock(&kvm_count_lock);
5067 	if (kvm_usage_count)
5068 		hardware_enable_nolock(NULL);
5069 	raw_spin_unlock(&kvm_count_lock);
5070 	return 0;
5071 }
5072 
hardware_disable_nolock(void * junk)5073 static void hardware_disable_nolock(void *junk)
5074 {
5075 	int cpu = raw_smp_processor_id();
5076 
5077 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
5078 		return;
5079 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5080 	kvm_arch_hardware_disable();
5081 }
5082 
kvm_dying_cpu(unsigned int cpu)5083 static int kvm_dying_cpu(unsigned int cpu)
5084 {
5085 	raw_spin_lock(&kvm_count_lock);
5086 	if (kvm_usage_count)
5087 		hardware_disable_nolock(NULL);
5088 	raw_spin_unlock(&kvm_count_lock);
5089 	return 0;
5090 }
5091 
hardware_disable_all_nolock(void)5092 static void hardware_disable_all_nolock(void)
5093 {
5094 	BUG_ON(!kvm_usage_count);
5095 
5096 	kvm_usage_count--;
5097 	if (!kvm_usage_count)
5098 		on_each_cpu(hardware_disable_nolock, NULL, 1);
5099 }
5100 
hardware_disable_all(void)5101 static void hardware_disable_all(void)
5102 {
5103 	raw_spin_lock(&kvm_count_lock);
5104 	hardware_disable_all_nolock();
5105 	raw_spin_unlock(&kvm_count_lock);
5106 }
5107 
hardware_enable_all(void)5108 static int hardware_enable_all(void)
5109 {
5110 	int r = 0;
5111 
5112 	raw_spin_lock(&kvm_count_lock);
5113 
5114 	kvm_usage_count++;
5115 	if (kvm_usage_count == 1) {
5116 		atomic_set(&hardware_enable_failed, 0);
5117 		on_each_cpu(hardware_enable_nolock, NULL, 1);
5118 
5119 		if (atomic_read(&hardware_enable_failed)) {
5120 			hardware_disable_all_nolock();
5121 			r = -EBUSY;
5122 		}
5123 	}
5124 
5125 	raw_spin_unlock(&kvm_count_lock);
5126 
5127 	return r;
5128 }
5129 
kvm_reboot(struct notifier_block * notifier,unsigned long val,void * v)5130 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
5131 		      void *v)
5132 {
5133 	/*
5134 	 * Some (well, at least mine) BIOSes hang on reboot if
5135 	 * in vmx root mode.
5136 	 *
5137 	 * And Intel TXT required VMX off for all cpu when system shutdown.
5138 	 */
5139 	pr_info("kvm: exiting hardware virtualization\n");
5140 	kvm_rebooting = true;
5141 	on_each_cpu(hardware_disable_nolock, NULL, 1);
5142 	return NOTIFY_OK;
5143 }
5144 
5145 static struct notifier_block kvm_reboot_notifier = {
5146 	.notifier_call = kvm_reboot,
5147 	.priority = 0,
5148 };
5149 
kvm_io_bus_destroy(struct kvm_io_bus * bus)5150 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5151 {
5152 	int i;
5153 
5154 	for (i = 0; i < bus->dev_count; i++) {
5155 		struct kvm_io_device *pos = bus->range[i].dev;
5156 
5157 		kvm_iodevice_destructor(pos);
5158 	}
5159 	kfree(bus);
5160 }
5161 
kvm_io_bus_cmp(const struct kvm_io_range * r1,const struct kvm_io_range * r2)5162 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5163 				 const struct kvm_io_range *r2)
5164 {
5165 	gpa_t addr1 = r1->addr;
5166 	gpa_t addr2 = r2->addr;
5167 
5168 	if (addr1 < addr2)
5169 		return -1;
5170 
5171 	/* If r2->len == 0, match the exact address.  If r2->len != 0,
5172 	 * accept any overlapping write.  Any order is acceptable for
5173 	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5174 	 * we process all of them.
5175 	 */
5176 	if (r2->len) {
5177 		addr1 += r1->len;
5178 		addr2 += r2->len;
5179 	}
5180 
5181 	if (addr1 > addr2)
5182 		return 1;
5183 
5184 	return 0;
5185 }
5186 
kvm_io_bus_sort_cmp(const void * p1,const void * p2)5187 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5188 {
5189 	return kvm_io_bus_cmp(p1, p2);
5190 }
5191 
kvm_io_bus_get_first_dev(struct kvm_io_bus * bus,gpa_t addr,int len)5192 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5193 			     gpa_t addr, int len)
5194 {
5195 	struct kvm_io_range *range, key;
5196 	int off;
5197 
5198 	key = (struct kvm_io_range) {
5199 		.addr = addr,
5200 		.len = len,
5201 	};
5202 
5203 	range = bsearch(&key, bus->range, bus->dev_count,
5204 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5205 	if (range == NULL)
5206 		return -ENOENT;
5207 
5208 	off = range - bus->range;
5209 
5210 	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5211 		off--;
5212 
5213 	return off;
5214 }
5215 
__kvm_io_bus_write(struct kvm_vcpu * vcpu,struct kvm_io_bus * bus,struct kvm_io_range * range,const void * val)5216 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5217 			      struct kvm_io_range *range, const void *val)
5218 {
5219 	int idx;
5220 
5221 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5222 	if (idx < 0)
5223 		return -EOPNOTSUPP;
5224 
5225 	while (idx < bus->dev_count &&
5226 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5227 		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5228 					range->len, val))
5229 			return idx;
5230 		idx++;
5231 	}
5232 
5233 	return -EOPNOTSUPP;
5234 }
5235 
5236 /* kvm_io_bus_write - called under kvm->slots_lock */
kvm_io_bus_write(struct kvm_vcpu * vcpu,enum kvm_bus bus_idx,gpa_t addr,int len,const void * val)5237 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5238 		     int len, const void *val)
5239 {
5240 	struct kvm_io_bus *bus;
5241 	struct kvm_io_range range;
5242 	int r;
5243 
5244 	range = (struct kvm_io_range) {
5245 		.addr = addr,
5246 		.len = len,
5247 	};
5248 
5249 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5250 	if (!bus)
5251 		return -ENOMEM;
5252 	r = __kvm_io_bus_write(vcpu, bus, &range, val);
5253 	return r < 0 ? r : 0;
5254 }
5255 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5256 
5257 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
kvm_io_bus_write_cookie(struct kvm_vcpu * vcpu,enum kvm_bus bus_idx,gpa_t addr,int len,const void * val,long cookie)5258 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5259 			    gpa_t addr, int len, const void *val, long cookie)
5260 {
5261 	struct kvm_io_bus *bus;
5262 	struct kvm_io_range range;
5263 
5264 	range = (struct kvm_io_range) {
5265 		.addr = addr,
5266 		.len = len,
5267 	};
5268 
5269 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5270 	if (!bus)
5271 		return -ENOMEM;
5272 
5273 	/* First try the device referenced by cookie. */
5274 	if ((cookie >= 0) && (cookie < bus->dev_count) &&
5275 	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5276 		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5277 					val))
5278 			return cookie;
5279 
5280 	/*
5281 	 * cookie contained garbage; fall back to search and return the
5282 	 * correct cookie value.
5283 	 */
5284 	return __kvm_io_bus_write(vcpu, bus, &range, val);
5285 }
5286 
__kvm_io_bus_read(struct kvm_vcpu * vcpu,struct kvm_io_bus * bus,struct kvm_io_range * range,void * val)5287 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5288 			     struct kvm_io_range *range, void *val)
5289 {
5290 	int idx;
5291 
5292 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5293 	if (idx < 0)
5294 		return -EOPNOTSUPP;
5295 
5296 	while (idx < bus->dev_count &&
5297 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5298 		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5299 				       range->len, val))
5300 			return idx;
5301 		idx++;
5302 	}
5303 
5304 	return -EOPNOTSUPP;
5305 }
5306 
5307 /* kvm_io_bus_read - called under kvm->slots_lock */
kvm_io_bus_read(struct kvm_vcpu * vcpu,enum kvm_bus bus_idx,gpa_t addr,int len,void * val)5308 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5309 		    int len, void *val)
5310 {
5311 	struct kvm_io_bus *bus;
5312 	struct kvm_io_range range;
5313 	int r;
5314 
5315 	range = (struct kvm_io_range) {
5316 		.addr = addr,
5317 		.len = len,
5318 	};
5319 
5320 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5321 	if (!bus)
5322 		return -ENOMEM;
5323 	r = __kvm_io_bus_read(vcpu, bus, &range, val);
5324 	return r < 0 ? r : 0;
5325 }
5326 
5327 /* Caller must hold slots_lock. */
kvm_io_bus_register_dev(struct kvm * kvm,enum kvm_bus bus_idx,gpa_t addr,int len,struct kvm_io_device * dev)5328 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5329 			    int len, struct kvm_io_device *dev)
5330 {
5331 	int i;
5332 	struct kvm_io_bus *new_bus, *bus;
5333 	struct kvm_io_range range;
5334 
5335 	bus = kvm_get_bus(kvm, bus_idx);
5336 	if (!bus)
5337 		return -ENOMEM;
5338 
5339 	/* exclude ioeventfd which is limited by maximum fd */
5340 	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5341 		return -ENOSPC;
5342 
5343 	new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5344 			  GFP_KERNEL_ACCOUNT);
5345 	if (!new_bus)
5346 		return -ENOMEM;
5347 
5348 	range = (struct kvm_io_range) {
5349 		.addr = addr,
5350 		.len = len,
5351 		.dev = dev,
5352 	};
5353 
5354 	for (i = 0; i < bus->dev_count; i++)
5355 		if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5356 			break;
5357 
5358 	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5359 	new_bus->dev_count++;
5360 	new_bus->range[i] = range;
5361 	memcpy(new_bus->range + i + 1, bus->range + i,
5362 		(bus->dev_count - i) * sizeof(struct kvm_io_range));
5363 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5364 	synchronize_srcu_expedited(&kvm->srcu);
5365 	kfree(bus);
5366 
5367 	return 0;
5368 }
5369 
kvm_io_bus_unregister_dev(struct kvm * kvm,enum kvm_bus bus_idx,struct kvm_io_device * dev)5370 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5371 			      struct kvm_io_device *dev)
5372 {
5373 	int i, j;
5374 	struct kvm_io_bus *new_bus, *bus;
5375 
5376 	lockdep_assert_held(&kvm->slots_lock);
5377 
5378 	bus = kvm_get_bus(kvm, bus_idx);
5379 	if (!bus)
5380 		return 0;
5381 
5382 	for (i = 0; i < bus->dev_count; i++) {
5383 		if (bus->range[i].dev == dev) {
5384 			break;
5385 		}
5386 	}
5387 
5388 	if (i == bus->dev_count)
5389 		return 0;
5390 
5391 	new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5392 			  GFP_KERNEL_ACCOUNT);
5393 	if (new_bus) {
5394 		memcpy(new_bus, bus, struct_size(bus, range, i));
5395 		new_bus->dev_count--;
5396 		memcpy(new_bus->range + i, bus->range + i + 1,
5397 				flex_array_size(new_bus, range, new_bus->dev_count - i));
5398 	}
5399 
5400 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5401 	synchronize_srcu_expedited(&kvm->srcu);
5402 
5403 	/* Destroy the old bus _after_ installing the (null) bus. */
5404 	if (!new_bus) {
5405 		pr_err("kvm: failed to shrink bus, removing it completely\n");
5406 		for (j = 0; j < bus->dev_count; j++) {
5407 			if (j == i)
5408 				continue;
5409 			kvm_iodevice_destructor(bus->range[j].dev);
5410 		}
5411 	}
5412 
5413 	kfree(bus);
5414 	return new_bus ? 0 : -ENOMEM;
5415 }
5416 
kvm_io_bus_get_dev(struct kvm * kvm,enum kvm_bus bus_idx,gpa_t addr)5417 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5418 					 gpa_t addr)
5419 {
5420 	struct kvm_io_bus *bus;
5421 	int dev_idx, srcu_idx;
5422 	struct kvm_io_device *iodev = NULL;
5423 
5424 	srcu_idx = srcu_read_lock(&kvm->srcu);
5425 
5426 	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5427 	if (!bus)
5428 		goto out_unlock;
5429 
5430 	dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5431 	if (dev_idx < 0)
5432 		goto out_unlock;
5433 
5434 	iodev = bus->range[dev_idx].dev;
5435 
5436 out_unlock:
5437 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5438 
5439 	return iodev;
5440 }
5441 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5442 
kvm_debugfs_open(struct inode * inode,struct file * file,int (* get)(void *,u64 *),int (* set)(void *,u64),const char * fmt)5443 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5444 			   int (*get)(void *, u64 *), int (*set)(void *, u64),
5445 			   const char *fmt)
5446 {
5447 	int ret;
5448 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5449 					  inode->i_private;
5450 
5451 	/*
5452 	 * The debugfs files are a reference to the kvm struct which
5453         * is still valid when kvm_destroy_vm is called.  kvm_get_kvm_safe
5454         * avoids the race between open and the removal of the debugfs directory.
5455 	 */
5456 	if (!kvm_get_kvm_safe(stat_data->kvm))
5457 		return -ENOENT;
5458 
5459 	ret = simple_attr_open(inode, file, get,
5460 			       kvm_stats_debugfs_mode(stat_data->desc) & 0222
5461 			       ? set : NULL, fmt);
5462 	if (ret)
5463 		kvm_put_kvm(stat_data->kvm);
5464 
5465 	return ret;
5466 }
5467 
kvm_debugfs_release(struct inode * inode,struct file * file)5468 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5469 {
5470 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5471 					  inode->i_private;
5472 
5473 	simple_attr_release(inode, file);
5474 	kvm_put_kvm(stat_data->kvm);
5475 
5476 	return 0;
5477 }
5478 
kvm_get_stat_per_vm(struct kvm * kvm,size_t offset,u64 * val)5479 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5480 {
5481 	*val = *(u64 *)((void *)(&kvm->stat) + offset);
5482 
5483 	return 0;
5484 }
5485 
kvm_clear_stat_per_vm(struct kvm * kvm,size_t offset)5486 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5487 {
5488 	*(u64 *)((void *)(&kvm->stat) + offset) = 0;
5489 
5490 	return 0;
5491 }
5492 
kvm_get_stat_per_vcpu(struct kvm * kvm,size_t offset,u64 * val)5493 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5494 {
5495 	unsigned long i;
5496 	struct kvm_vcpu *vcpu;
5497 
5498 	*val = 0;
5499 
5500 	kvm_for_each_vcpu(i, vcpu, kvm)
5501 		*val += *(u64 *)((void *)(&vcpu->stat) + offset);
5502 
5503 	return 0;
5504 }
5505 
kvm_clear_stat_per_vcpu(struct kvm * kvm,size_t offset)5506 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5507 {
5508 	unsigned long i;
5509 	struct kvm_vcpu *vcpu;
5510 
5511 	kvm_for_each_vcpu(i, vcpu, kvm)
5512 		*(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5513 
5514 	return 0;
5515 }
5516 
kvm_stat_data_get(void * data,u64 * val)5517 static int kvm_stat_data_get(void *data, u64 *val)
5518 {
5519 	int r = -EFAULT;
5520 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5521 
5522 	switch (stat_data->kind) {
5523 	case KVM_STAT_VM:
5524 		r = kvm_get_stat_per_vm(stat_data->kvm,
5525 					stat_data->desc->desc.offset, val);
5526 		break;
5527 	case KVM_STAT_VCPU:
5528 		r = kvm_get_stat_per_vcpu(stat_data->kvm,
5529 					  stat_data->desc->desc.offset, val);
5530 		break;
5531 	}
5532 
5533 	return r;
5534 }
5535 
kvm_stat_data_clear(void * data,u64 val)5536 static int kvm_stat_data_clear(void *data, u64 val)
5537 {
5538 	int r = -EFAULT;
5539 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5540 
5541 	if (val)
5542 		return -EINVAL;
5543 
5544 	switch (stat_data->kind) {
5545 	case KVM_STAT_VM:
5546 		r = kvm_clear_stat_per_vm(stat_data->kvm,
5547 					  stat_data->desc->desc.offset);
5548 		break;
5549 	case KVM_STAT_VCPU:
5550 		r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5551 					    stat_data->desc->desc.offset);
5552 		break;
5553 	}
5554 
5555 	return r;
5556 }
5557 
kvm_stat_data_open(struct inode * inode,struct file * file)5558 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5559 {
5560 	__simple_attr_check_format("%llu\n", 0ull);
5561 	return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5562 				kvm_stat_data_clear, "%llu\n");
5563 }
5564 
5565 static const struct file_operations stat_fops_per_vm = {
5566 	.owner = THIS_MODULE,
5567 	.open = kvm_stat_data_open,
5568 	.release = kvm_debugfs_release,
5569 	.read = simple_attr_read,
5570 	.write = simple_attr_write,
5571 	.llseek = no_llseek,
5572 };
5573 
vm_stat_get(void * _offset,u64 * val)5574 static int vm_stat_get(void *_offset, u64 *val)
5575 {
5576 	unsigned offset = (long)_offset;
5577 	struct kvm *kvm;
5578 	u64 tmp_val;
5579 
5580 	*val = 0;
5581 	mutex_lock(&kvm_lock);
5582 	list_for_each_entry(kvm, &vm_list, vm_list) {
5583 		kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5584 		*val += tmp_val;
5585 	}
5586 	mutex_unlock(&kvm_lock);
5587 	return 0;
5588 }
5589 
vm_stat_clear(void * _offset,u64 val)5590 static int vm_stat_clear(void *_offset, u64 val)
5591 {
5592 	unsigned offset = (long)_offset;
5593 	struct kvm *kvm;
5594 
5595 	if (val)
5596 		return -EINVAL;
5597 
5598 	mutex_lock(&kvm_lock);
5599 	list_for_each_entry(kvm, &vm_list, vm_list) {
5600 		kvm_clear_stat_per_vm(kvm, offset);
5601 	}
5602 	mutex_unlock(&kvm_lock);
5603 
5604 	return 0;
5605 }
5606 
5607 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5608 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5609 
vcpu_stat_get(void * _offset,u64 * val)5610 static int vcpu_stat_get(void *_offset, u64 *val)
5611 {
5612 	unsigned offset = (long)_offset;
5613 	struct kvm *kvm;
5614 	u64 tmp_val;
5615 
5616 	*val = 0;
5617 	mutex_lock(&kvm_lock);
5618 	list_for_each_entry(kvm, &vm_list, vm_list) {
5619 		kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5620 		*val += tmp_val;
5621 	}
5622 	mutex_unlock(&kvm_lock);
5623 	return 0;
5624 }
5625 
vcpu_stat_clear(void * _offset,u64 val)5626 static int vcpu_stat_clear(void *_offset, u64 val)
5627 {
5628 	unsigned offset = (long)_offset;
5629 	struct kvm *kvm;
5630 
5631 	if (val)
5632 		return -EINVAL;
5633 
5634 	mutex_lock(&kvm_lock);
5635 	list_for_each_entry(kvm, &vm_list, vm_list) {
5636 		kvm_clear_stat_per_vcpu(kvm, offset);
5637 	}
5638 	mutex_unlock(&kvm_lock);
5639 
5640 	return 0;
5641 }
5642 
5643 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5644 			"%llu\n");
5645 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5646 
kvm_uevent_notify_change(unsigned int type,struct kvm * kvm)5647 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5648 {
5649 	struct kobj_uevent_env *env;
5650 	unsigned long long created, active;
5651 
5652 	if (!kvm_dev.this_device || !kvm)
5653 		return;
5654 
5655 	mutex_lock(&kvm_lock);
5656 	if (type == KVM_EVENT_CREATE_VM) {
5657 		kvm_createvm_count++;
5658 		kvm_active_vms++;
5659 	} else if (type == KVM_EVENT_DESTROY_VM) {
5660 		kvm_active_vms--;
5661 	}
5662 	created = kvm_createvm_count;
5663 	active = kvm_active_vms;
5664 	mutex_unlock(&kvm_lock);
5665 
5666 	env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5667 	if (!env)
5668 		return;
5669 
5670 	add_uevent_var(env, "CREATED=%llu", created);
5671 	add_uevent_var(env, "COUNT=%llu", active);
5672 
5673 	if (type == KVM_EVENT_CREATE_VM) {
5674 		add_uevent_var(env, "EVENT=create");
5675 		kvm->userspace_pid = task_pid_nr(current);
5676 	} else if (type == KVM_EVENT_DESTROY_VM) {
5677 		add_uevent_var(env, "EVENT=destroy");
5678 	}
5679 	add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5680 
5681 	if (!IS_ERR(kvm->debugfs_dentry)) {
5682 		char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5683 
5684 		if (p) {
5685 			tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5686 			if (!IS_ERR(tmp))
5687 				add_uevent_var(env, "STATS_PATH=%s", tmp);
5688 			kfree(p);
5689 		}
5690 	}
5691 	/* no need for checks, since we are adding at most only 5 keys */
5692 	env->envp[env->envp_idx++] = NULL;
5693 	kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5694 	kfree(env);
5695 }
5696 
kvm_init_debug(void)5697 static void kvm_init_debug(void)
5698 {
5699 	const struct file_operations *fops;
5700 	const struct _kvm_stats_desc *pdesc;
5701 	int i;
5702 
5703 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5704 
5705 	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5706 		pdesc = &kvm_vm_stats_desc[i];
5707 		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5708 			fops = &vm_stat_fops;
5709 		else
5710 			fops = &vm_stat_readonly_fops;
5711 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5712 				kvm_debugfs_dir,
5713 				(void *)(long)pdesc->desc.offset, fops);
5714 	}
5715 
5716 	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5717 		pdesc = &kvm_vcpu_stats_desc[i];
5718 		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5719 			fops = &vcpu_stat_fops;
5720 		else
5721 			fops = &vcpu_stat_readonly_fops;
5722 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5723 				kvm_debugfs_dir,
5724 				(void *)(long)pdesc->desc.offset, fops);
5725 	}
5726 }
5727 
kvm_suspend(void)5728 static int kvm_suspend(void)
5729 {
5730 	if (kvm_usage_count)
5731 		hardware_disable_nolock(NULL);
5732 	return 0;
5733 }
5734 
kvm_resume(void)5735 static void kvm_resume(void)
5736 {
5737 	if (kvm_usage_count) {
5738 		lockdep_assert_not_held(&kvm_count_lock);
5739 		hardware_enable_nolock(NULL);
5740 	}
5741 }
5742 
5743 static struct syscore_ops kvm_syscore_ops = {
5744 	.suspend = kvm_suspend,
5745 	.resume = kvm_resume,
5746 };
5747 
5748 static inline
preempt_notifier_to_vcpu(struct preempt_notifier * pn)5749 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5750 {
5751 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
5752 }
5753 
kvm_sched_in(struct preempt_notifier * pn,int cpu)5754 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5755 {
5756 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5757 
5758 	WRITE_ONCE(vcpu->preempted, false);
5759 	WRITE_ONCE(vcpu->ready, false);
5760 
5761 	__this_cpu_write(kvm_running_vcpu, vcpu);
5762 	kvm_arch_sched_in(vcpu, cpu);
5763 	kvm_arch_vcpu_load(vcpu, cpu);
5764 }
5765 
kvm_sched_out(struct preempt_notifier * pn,struct task_struct * next)5766 static void kvm_sched_out(struct preempt_notifier *pn,
5767 			  struct task_struct *next)
5768 {
5769 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5770 
5771 	if (current->on_rq) {
5772 		WRITE_ONCE(vcpu->preempted, true);
5773 		WRITE_ONCE(vcpu->ready, true);
5774 	}
5775 	kvm_arch_vcpu_put(vcpu);
5776 	__this_cpu_write(kvm_running_vcpu, NULL);
5777 }
5778 
5779 /**
5780  * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5781  *
5782  * We can disable preemption locally around accessing the per-CPU variable,
5783  * and use the resolved vcpu pointer after enabling preemption again,
5784  * because even if the current thread is migrated to another CPU, reading
5785  * the per-CPU value later will give us the same value as we update the
5786  * per-CPU variable in the preempt notifier handlers.
5787  */
kvm_get_running_vcpu(void)5788 struct kvm_vcpu *kvm_get_running_vcpu(void)
5789 {
5790 	struct kvm_vcpu *vcpu;
5791 
5792 	preempt_disable();
5793 	vcpu = __this_cpu_read(kvm_running_vcpu);
5794 	preempt_enable();
5795 
5796 	return vcpu;
5797 }
5798 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5799 
5800 /**
5801  * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5802  */
kvm_get_running_vcpus(void)5803 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5804 {
5805         return &kvm_running_vcpu;
5806 }
5807 
5808 #ifdef CONFIG_GUEST_PERF_EVENTS
kvm_guest_state(void)5809 static unsigned int kvm_guest_state(void)
5810 {
5811 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5812 	unsigned int state;
5813 
5814 	if (!kvm_arch_pmi_in_guest(vcpu))
5815 		return 0;
5816 
5817 	state = PERF_GUEST_ACTIVE;
5818 	if (!kvm_arch_vcpu_in_kernel(vcpu))
5819 		state |= PERF_GUEST_USER;
5820 
5821 	return state;
5822 }
5823 
kvm_guest_get_ip(void)5824 static unsigned long kvm_guest_get_ip(void)
5825 {
5826 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5827 
5828 	/* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5829 	if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5830 		return 0;
5831 
5832 	return kvm_arch_vcpu_get_ip(vcpu);
5833 }
5834 
5835 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5836 	.state			= kvm_guest_state,
5837 	.get_ip			= kvm_guest_get_ip,
5838 	.handle_intel_pt_intr	= NULL,
5839 };
5840 
kvm_register_perf_callbacks(unsigned int (* pt_intr_handler)(void))5841 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5842 {
5843 	kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5844 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5845 }
kvm_unregister_perf_callbacks(void)5846 void kvm_unregister_perf_callbacks(void)
5847 {
5848 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5849 }
5850 #endif
5851 
5852 struct kvm_cpu_compat_check {
5853 	void *opaque;
5854 	int *ret;
5855 };
5856 
check_processor_compat(void * data)5857 static void check_processor_compat(void *data)
5858 {
5859 	struct kvm_cpu_compat_check *c = data;
5860 
5861 	*c->ret = kvm_arch_check_processor_compat(c->opaque);
5862 }
5863 
kvm_init(void * opaque,unsigned vcpu_size,unsigned vcpu_align,struct module * module)5864 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5865 		  struct module *module)
5866 {
5867 	struct kvm_cpu_compat_check c;
5868 	int r;
5869 	int cpu;
5870 
5871 	r = kvm_arch_init(opaque);
5872 	if (r)
5873 		goto out_fail;
5874 
5875 	/*
5876 	 * kvm_arch_init makes sure there's at most one caller
5877 	 * for architectures that support multiple implementations,
5878 	 * like intel and amd on x86.
5879 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5880 	 * conflicts in case kvm is already setup for another implementation.
5881 	 */
5882 	r = kvm_irqfd_init();
5883 	if (r)
5884 		goto out_irqfd;
5885 
5886 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5887 		r = -ENOMEM;
5888 		goto out_free_0;
5889 	}
5890 
5891 	r = kvm_arch_hardware_setup(opaque);
5892 	if (r < 0)
5893 		goto out_free_1;
5894 
5895 	c.ret = &r;
5896 	c.opaque = opaque;
5897 	for_each_online_cpu(cpu) {
5898 		smp_call_function_single(cpu, check_processor_compat, &c, 1);
5899 		if (r < 0)
5900 			goto out_free_2;
5901 	}
5902 
5903 	r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5904 				      kvm_starting_cpu, kvm_dying_cpu);
5905 	if (r)
5906 		goto out_free_2;
5907 	register_reboot_notifier(&kvm_reboot_notifier);
5908 
5909 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
5910 	if (!vcpu_align)
5911 		vcpu_align = __alignof__(struct kvm_vcpu);
5912 	kvm_vcpu_cache =
5913 		kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5914 					   SLAB_ACCOUNT,
5915 					   offsetof(struct kvm_vcpu, arch),
5916 					   offsetofend(struct kvm_vcpu, stats_id)
5917 					   - offsetof(struct kvm_vcpu, arch),
5918 					   NULL);
5919 	if (!kvm_vcpu_cache) {
5920 		r = -ENOMEM;
5921 		goto out_free_3;
5922 	}
5923 
5924 	for_each_possible_cpu(cpu) {
5925 		if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5926 					    GFP_KERNEL, cpu_to_node(cpu))) {
5927 			r = -ENOMEM;
5928 			goto out_free_4;
5929 		}
5930 	}
5931 
5932 	r = kvm_async_pf_init();
5933 	if (r)
5934 		goto out_free_4;
5935 
5936 	kvm_chardev_ops.owner = module;
5937 
5938 	r = misc_register(&kvm_dev);
5939 	if (r) {
5940 		pr_err("kvm: misc device register failed\n");
5941 		goto out_unreg;
5942 	}
5943 
5944 	register_syscore_ops(&kvm_syscore_ops);
5945 
5946 	kvm_preempt_ops.sched_in = kvm_sched_in;
5947 	kvm_preempt_ops.sched_out = kvm_sched_out;
5948 
5949 	kvm_init_debug();
5950 
5951 	r = kvm_vfio_ops_init();
5952 	WARN_ON(r);
5953 
5954 	return 0;
5955 
5956 out_unreg:
5957 	kvm_async_pf_deinit();
5958 out_free_4:
5959 	for_each_possible_cpu(cpu)
5960 		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5961 	kmem_cache_destroy(kvm_vcpu_cache);
5962 out_free_3:
5963 	unregister_reboot_notifier(&kvm_reboot_notifier);
5964 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5965 out_free_2:
5966 	kvm_arch_hardware_unsetup();
5967 out_free_1:
5968 	free_cpumask_var(cpus_hardware_enabled);
5969 out_free_0:
5970 	kvm_irqfd_exit();
5971 out_irqfd:
5972 	kvm_arch_exit();
5973 out_fail:
5974 	return r;
5975 }
5976 EXPORT_SYMBOL_GPL(kvm_init);
5977 
kvm_exit(void)5978 void kvm_exit(void)
5979 {
5980 	int cpu;
5981 
5982 	debugfs_remove_recursive(kvm_debugfs_dir);
5983 	misc_deregister(&kvm_dev);
5984 	for_each_possible_cpu(cpu)
5985 		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5986 	kmem_cache_destroy(kvm_vcpu_cache);
5987 	kvm_async_pf_deinit();
5988 	unregister_syscore_ops(&kvm_syscore_ops);
5989 	unregister_reboot_notifier(&kvm_reboot_notifier);
5990 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5991 	on_each_cpu(hardware_disable_nolock, NULL, 1);
5992 	kvm_arch_hardware_unsetup();
5993 	kvm_arch_exit();
5994 	kvm_irqfd_exit();
5995 	free_cpumask_var(cpus_hardware_enabled);
5996 	kvm_vfio_ops_exit();
5997 }
5998 EXPORT_SYMBOL_GPL(kvm_exit);
5999 
6000 struct kvm_vm_worker_thread_context {
6001 	struct kvm *kvm;
6002 	struct task_struct *parent;
6003 	struct completion init_done;
6004 	kvm_vm_thread_fn_t thread_fn;
6005 	uintptr_t data;
6006 	int err;
6007 };
6008 
kvm_vm_worker_thread(void * context)6009 static int kvm_vm_worker_thread(void *context)
6010 {
6011 	/*
6012 	 * The init_context is allocated on the stack of the parent thread, so
6013 	 * we have to locally copy anything that is needed beyond initialization
6014 	 */
6015 	struct kvm_vm_worker_thread_context *init_context = context;
6016 	struct task_struct *parent;
6017 	struct kvm *kvm = init_context->kvm;
6018 	kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6019 	uintptr_t data = init_context->data;
6020 	int err;
6021 
6022 	err = kthread_park(current);
6023 	/* kthread_park(current) is never supposed to return an error */
6024 	WARN_ON(err != 0);
6025 	if (err)
6026 		goto init_complete;
6027 
6028 	err = cgroup_attach_task_all(init_context->parent, current);
6029 	if (err) {
6030 		kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6031 			__func__, err);
6032 		goto init_complete;
6033 	}
6034 
6035 	set_user_nice(current, task_nice(init_context->parent));
6036 
6037 init_complete:
6038 	init_context->err = err;
6039 	complete(&init_context->init_done);
6040 	init_context = NULL;
6041 
6042 	if (err)
6043 		goto out;
6044 
6045 	/* Wait to be woken up by the spawner before proceeding. */
6046 	kthread_parkme();
6047 
6048 	if (!kthread_should_stop())
6049 		err = thread_fn(kvm, data);
6050 
6051 out:
6052 	/*
6053 	 * Move kthread back to its original cgroup to prevent it lingering in
6054 	 * the cgroup of the VM process, after the latter finishes its
6055 	 * execution.
6056 	 *
6057 	 * kthread_stop() waits on the 'exited' completion condition which is
6058 	 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6059 	 * kthread is removed from the cgroup in the cgroup_exit() which is
6060 	 * called after the exit_mm(). This causes the kthread_stop() to return
6061 	 * before the kthread actually quits the cgroup.
6062 	 */
6063 	rcu_read_lock();
6064 	parent = rcu_dereference(current->real_parent);
6065 	get_task_struct(parent);
6066 	rcu_read_unlock();
6067 	cgroup_attach_task_all(parent, current);
6068 	put_task_struct(parent);
6069 
6070 	return err;
6071 }
6072 
kvm_vm_create_worker_thread(struct kvm * kvm,kvm_vm_thread_fn_t thread_fn,uintptr_t data,const char * name,struct task_struct ** thread_ptr)6073 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6074 				uintptr_t data, const char *name,
6075 				struct task_struct **thread_ptr)
6076 {
6077 	struct kvm_vm_worker_thread_context init_context = {};
6078 	struct task_struct *thread;
6079 
6080 	*thread_ptr = NULL;
6081 	init_context.kvm = kvm;
6082 	init_context.parent = current;
6083 	init_context.thread_fn = thread_fn;
6084 	init_context.data = data;
6085 	init_completion(&init_context.init_done);
6086 
6087 	thread = kthread_run(kvm_vm_worker_thread, &init_context,
6088 			     "%s-%d", name, task_pid_nr(current));
6089 	if (IS_ERR(thread))
6090 		return PTR_ERR(thread);
6091 
6092 	/* kthread_run is never supposed to return NULL */
6093 	WARN_ON(thread == NULL);
6094 
6095 	wait_for_completion(&init_context.init_done);
6096 
6097 	if (!init_context.err)
6098 		*thread_ptr = thread;
6099 
6100 	return init_context.err;
6101 }
6102