1/*
2 * VMware VMCI Driver
3 *
4 * Copyright (C) 2012 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
12 * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
13 * for more details.
14 */
15
16#include <linux/vmw_vmci_defs.h>
17#include <linux/vmw_vmci_api.h>
18#include <linux/moduleparam.h>
19#include <linux/miscdevice.h>
20#include <linux/interrupt.h>
21#include <linux/highmem.h>
22#include <linux/atomic.h>
23#include <linux/kernel.h>
24#include <linux/module.h>
25#include <linux/mutex.h>
26#include <linux/sched.h>
27#include <linux/slab.h>
28#include <linux/file.h>
29#include <linux/init.h>
30#include <linux/poll.h>
31#include <linux/pci.h>
32#include <linux/smp.h>
33#include <linux/fs.h>
34#include <linux/io.h>
35
36#include "vmci_handle_array.h"
37#include "vmci_queue_pair.h"
38#include "vmci_datagram.h"
39#include "vmci_doorbell.h"
40#include "vmci_resource.h"
41#include "vmci_context.h"
42#include "vmci_driver.h"
43#include "vmci_event.h"
44
45#define VMCI_UTIL_NUM_RESOURCES 1
46
47enum {
48	VMCI_NOTIFY_RESOURCE_QUEUE_PAIR = 0,
49	VMCI_NOTIFY_RESOURCE_DOOR_BELL = 1,
50};
51
52enum {
53	VMCI_NOTIFY_RESOURCE_ACTION_NOTIFY = 0,
54	VMCI_NOTIFY_RESOURCE_ACTION_CREATE = 1,
55	VMCI_NOTIFY_RESOURCE_ACTION_DESTROY = 2,
56};
57
58/*
59 * VMCI driver initialization. This block can also be used to
60 * pass initial group membership etc.
61 */
62struct vmci_init_blk {
63	u32 cid;
64	u32 flags;
65};
66
67/* VMCIqueue_pairAllocInfo_VMToVM */
68struct vmci_qp_alloc_info_vmvm {
69	struct vmci_handle handle;
70	u32 peer;
71	u32 flags;
72	u64 produce_size;
73	u64 consume_size;
74	u64 produce_page_file;	  /* User VA. */
75	u64 consume_page_file;	  /* User VA. */
76	u64 produce_page_file_size;  /* Size of the file name array. */
77	u64 consume_page_file_size;  /* Size of the file name array. */
78	s32 result;
79	u32 _pad;
80};
81
82/* VMCISetNotifyInfo: Used to pass notify flag's address to the host driver. */
83struct vmci_set_notify_info {
84	u64 notify_uva;
85	s32 result;
86	u32 _pad;
87};
88
89/*
90 * Per-instance host state
91 */
92struct vmci_host_dev {
93	struct vmci_ctx *context;
94	int user_version;
95	enum vmci_obj_type ct_type;
96	struct mutex lock;  /* Mutex lock for vmci context access */
97};
98
99static struct vmci_ctx *host_context;
100static bool vmci_host_device_initialized;
101static atomic_t vmci_host_active_users = ATOMIC_INIT(0);
102
103/*
104 * Determines whether the VMCI host personality is
105 * available. Since the core functionality of the host driver is
106 * always present, all guests could possibly use the host
107 * personality. However, to minimize the deviation from the
108 * pre-unified driver state of affairs, we only consider the host
109 * device active if there is no active guest device or if there
110 * are VMX'en with active VMCI contexts using the host device.
111 */
112bool vmci_host_code_active(void)
113{
114	return vmci_host_device_initialized &&
115	    (!vmci_guest_code_active() ||
116	     atomic_read(&vmci_host_active_users) > 0);
117}
118
119/*
120 * Called on open of /dev/vmci.
121 */
122static int vmci_host_open(struct inode *inode, struct file *filp)
123{
124	struct vmci_host_dev *vmci_host_dev;
125
126	vmci_host_dev = kzalloc(sizeof(struct vmci_host_dev), GFP_KERNEL);
127	if (vmci_host_dev == NULL)
128		return -ENOMEM;
129
130	vmci_host_dev->ct_type = VMCIOBJ_NOT_SET;
131	mutex_init(&vmci_host_dev->lock);
132	filp->private_data = vmci_host_dev;
133
134	return 0;
135}
136
137/*
138 * Called on close of /dev/vmci, most often when the process
139 * exits.
140 */
141static int vmci_host_close(struct inode *inode, struct file *filp)
142{
143	struct vmci_host_dev *vmci_host_dev = filp->private_data;
144
145	if (vmci_host_dev->ct_type == VMCIOBJ_CONTEXT) {
146		vmci_ctx_destroy(vmci_host_dev->context);
147		vmci_host_dev->context = NULL;
148
149		/*
150		 * The number of active contexts is used to track whether any
151		 * VMX'en are using the host personality. It is incremented when
152		 * a context is created through the IOCTL_VMCI_INIT_CONTEXT
153		 * ioctl.
154		 */
155		atomic_dec(&vmci_host_active_users);
156	}
157	vmci_host_dev->ct_type = VMCIOBJ_NOT_SET;
158
159	kfree(vmci_host_dev);
160	filp->private_data = NULL;
161	return 0;
162}
163
164/*
165 * This is used to wake up the VMX when a VMCI call arrives, or
166 * to wake up select() or poll() at the next clock tick.
167 */
168static unsigned int vmci_host_poll(struct file *filp, poll_table *wait)
169{
170	struct vmci_host_dev *vmci_host_dev = filp->private_data;
171	struct vmci_ctx *context = vmci_host_dev->context;
172	unsigned int mask = 0;
173
174	if (vmci_host_dev->ct_type == VMCIOBJ_CONTEXT) {
175		/* Check for VMCI calls to this VM context. */
176		if (wait)
177			poll_wait(filp, &context->host_context.wait_queue,
178				  wait);
179
180		spin_lock(&context->lock);
181		if (context->pending_datagrams > 0 ||
182		    vmci_handle_arr_get_size(
183				context->pending_doorbell_array) > 0) {
184			mask = POLLIN;
185		}
186		spin_unlock(&context->lock);
187	}
188	return mask;
189}
190
191/*
192 * Copies the handles of a handle array into a user buffer, and
193 * returns the new length in userBufferSize. If the copy to the
194 * user buffer fails, the functions still returns VMCI_SUCCESS,
195 * but retval != 0.
196 */
197static int drv_cp_harray_to_user(void __user *user_buf_uva,
198				 u64 *user_buf_size,
199				 struct vmci_handle_arr *handle_array,
200				 int *retval)
201{
202	u32 array_size = 0;
203	struct vmci_handle *handles;
204
205	if (handle_array)
206		array_size = vmci_handle_arr_get_size(handle_array);
207
208	if (array_size * sizeof(*handles) > *user_buf_size)
209		return VMCI_ERROR_MORE_DATA;
210
211	*user_buf_size = array_size * sizeof(*handles);
212	if (*user_buf_size)
213		*retval = copy_to_user(user_buf_uva,
214				       vmci_handle_arr_get_handles
215				       (handle_array), *user_buf_size);
216
217	return VMCI_SUCCESS;
218}
219
220/*
221 * Sets up a given context for notify to work.  Calls drv_map_bool_ptr()
222 * which maps the notify boolean in user VA in kernel space.
223 */
224static int vmci_host_setup_notify(struct vmci_ctx *context,
225				  unsigned long uva)
226{
227	struct page *page;
228	int retval;
229
230	if (context->notify_page) {
231		pr_devel("%s: Notify mechanism is already set up\n", __func__);
232		return VMCI_ERROR_DUPLICATE_ENTRY;
233	}
234
235	/*
236	 * We are using 'bool' internally, but let's make sure we explicit
237	 * about the size.
238	 */
239	BUILD_BUG_ON(sizeof(bool) != sizeof(u8));
240	if (!access_ok(VERIFY_WRITE, (void __user *)uva, sizeof(u8)))
241		return VMCI_ERROR_GENERIC;
242
243	/*
244	 * Lock physical page backing a given user VA.
245	 */
246	retval = get_user_pages_fast(PAGE_ALIGN(uva), 1, 1, &page);
247	if (retval != 1)
248		return VMCI_ERROR_GENERIC;
249
250	/*
251	 * Map the locked page and set up notify pointer.
252	 */
253	context->notify = kmap(page) + (uva & (PAGE_SIZE - 1));
254	vmci_ctx_check_signal_notify(context);
255
256	return VMCI_SUCCESS;
257}
258
259static int vmci_host_get_version(struct vmci_host_dev *vmci_host_dev,
260				 unsigned int cmd, void __user *uptr)
261{
262	if (cmd == IOCTL_VMCI_VERSION2) {
263		int __user *vptr = uptr;
264		if (get_user(vmci_host_dev->user_version, vptr))
265			return -EFAULT;
266	}
267
268	/*
269	 * The basic logic here is:
270	 *
271	 * If the user sends in a version of 0 tell it our version.
272	 * If the user didn't send in a version, tell it our version.
273	 * If the user sent in an old version, tell it -its- version.
274	 * If the user sent in an newer version, tell it our version.
275	 *
276	 * The rationale behind telling the caller its version is that
277	 * Workstation 6.5 required that VMX and VMCI kernel module were
278	 * version sync'd.  All new VMX users will be programmed to
279	 * handle the VMCI kernel module version.
280	 */
281
282	if (vmci_host_dev->user_version > 0 &&
283	    vmci_host_dev->user_version < VMCI_VERSION_HOSTQP) {
284		return vmci_host_dev->user_version;
285	}
286
287	return VMCI_VERSION;
288}
289
290#define vmci_ioctl_err(fmt, ...)	\
291	pr_devel("%s: " fmt, ioctl_name, ##__VA_ARGS__)
292
293static int vmci_host_do_init_context(struct vmci_host_dev *vmci_host_dev,
294				     const char *ioctl_name,
295				     void __user *uptr)
296{
297	struct vmci_init_blk init_block;
298	const struct cred *cred;
299	int retval;
300
301	if (copy_from_user(&init_block, uptr, sizeof(init_block))) {
302		vmci_ioctl_err("error reading init block\n");
303		return -EFAULT;
304	}
305
306	mutex_lock(&vmci_host_dev->lock);
307
308	if (vmci_host_dev->ct_type != VMCIOBJ_NOT_SET) {
309		vmci_ioctl_err("received VMCI init on initialized handle\n");
310		retval = -EINVAL;
311		goto out;
312	}
313
314	if (init_block.flags & ~VMCI_PRIVILEGE_FLAG_RESTRICTED) {
315		vmci_ioctl_err("unsupported VMCI restriction flag\n");
316		retval = -EINVAL;
317		goto out;
318	}
319
320	cred = get_current_cred();
321	vmci_host_dev->context = vmci_ctx_create(init_block.cid,
322						 init_block.flags, 0,
323						 vmci_host_dev->user_version,
324						 cred);
325	put_cred(cred);
326	if (IS_ERR(vmci_host_dev->context)) {
327		retval = PTR_ERR(vmci_host_dev->context);
328		vmci_ioctl_err("error initializing context\n");
329		goto out;
330	}
331
332	/*
333	 * Copy cid to userlevel, we do this to allow the VMX
334	 * to enforce its policy on cid generation.
335	 */
336	init_block.cid = vmci_ctx_get_id(vmci_host_dev->context);
337	if (copy_to_user(uptr, &init_block, sizeof(init_block))) {
338		vmci_ctx_destroy(vmci_host_dev->context);
339		vmci_host_dev->context = NULL;
340		vmci_ioctl_err("error writing init block\n");
341		retval = -EFAULT;
342		goto out;
343	}
344
345	vmci_host_dev->ct_type = VMCIOBJ_CONTEXT;
346	atomic_inc(&vmci_host_active_users);
347
348	retval = 0;
349
350out:
351	mutex_unlock(&vmci_host_dev->lock);
352	return retval;
353}
354
355static int vmci_host_do_send_datagram(struct vmci_host_dev *vmci_host_dev,
356				      const char *ioctl_name,
357				      void __user *uptr)
358{
359	struct vmci_datagram_snd_rcv_info send_info;
360	struct vmci_datagram *dg = NULL;
361	u32 cid;
362
363	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
364		vmci_ioctl_err("only valid for contexts\n");
365		return -EINVAL;
366	}
367
368	if (copy_from_user(&send_info, uptr, sizeof(send_info)))
369		return -EFAULT;
370
371	if (send_info.len > VMCI_MAX_DG_SIZE) {
372		vmci_ioctl_err("datagram is too big (size=%d)\n",
373			       send_info.len);
374		return -EINVAL;
375	}
376
377	if (send_info.len < sizeof(*dg)) {
378		vmci_ioctl_err("datagram is too small (size=%d)\n",
379			       send_info.len);
380		return -EINVAL;
381	}
382
383	dg = kmalloc(send_info.len, GFP_KERNEL);
384	if (!dg) {
385		vmci_ioctl_err(
386			"cannot allocate memory to dispatch datagram\n");
387		return -ENOMEM;
388	}
389
390	if (copy_from_user(dg, (void __user *)(uintptr_t)send_info.addr,
391			   send_info.len)) {
392		vmci_ioctl_err("error getting datagram\n");
393		kfree(dg);
394		return -EFAULT;
395	}
396
397	pr_devel("Datagram dst (handle=0x%x:0x%x) src (handle=0x%x:0x%x), payload (size=%llu bytes)\n",
398		 dg->dst.context, dg->dst.resource,
399		 dg->src.context, dg->src.resource,
400		 (unsigned long long)dg->payload_size);
401
402	/* Get source context id. */
403	cid = vmci_ctx_get_id(vmci_host_dev->context);
404	send_info.result = vmci_datagram_dispatch(cid, dg, true);
405	kfree(dg);
406
407	return copy_to_user(uptr, &send_info, sizeof(send_info)) ? -EFAULT : 0;
408}
409
410static int vmci_host_do_receive_datagram(struct vmci_host_dev *vmci_host_dev,
411					 const char *ioctl_name,
412					 void __user *uptr)
413{
414	struct vmci_datagram_snd_rcv_info recv_info;
415	struct vmci_datagram *dg = NULL;
416	int retval;
417	size_t size;
418
419	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
420		vmci_ioctl_err("only valid for contexts\n");
421		return -EINVAL;
422	}
423
424	if (copy_from_user(&recv_info, uptr, sizeof(recv_info)))
425		return -EFAULT;
426
427	size = recv_info.len;
428	recv_info.result = vmci_ctx_dequeue_datagram(vmci_host_dev->context,
429						     &size, &dg);
430
431	if (recv_info.result >= VMCI_SUCCESS) {
432		void __user *ubuf = (void __user *)(uintptr_t)recv_info.addr;
433		retval = copy_to_user(ubuf, dg, VMCI_DG_SIZE(dg));
434		kfree(dg);
435		if (retval != 0)
436			return -EFAULT;
437	}
438
439	return copy_to_user(uptr, &recv_info, sizeof(recv_info)) ? -EFAULT : 0;
440}
441
442static int vmci_host_do_alloc_queuepair(struct vmci_host_dev *vmci_host_dev,
443					const char *ioctl_name,
444					void __user *uptr)
445{
446	struct vmci_handle handle;
447	int vmci_status;
448	int __user *retptr;
449	u32 cid;
450
451	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
452		vmci_ioctl_err("only valid for contexts\n");
453		return -EINVAL;
454	}
455
456	cid = vmci_ctx_get_id(vmci_host_dev->context);
457
458	if (vmci_host_dev->user_version < VMCI_VERSION_NOVMVM) {
459		struct vmci_qp_alloc_info_vmvm alloc_info;
460		struct vmci_qp_alloc_info_vmvm __user *info = uptr;
461
462		if (copy_from_user(&alloc_info, uptr, sizeof(alloc_info)))
463			return -EFAULT;
464
465		handle = alloc_info.handle;
466		retptr = &info->result;
467
468		vmci_status = vmci_qp_broker_alloc(alloc_info.handle,
469						alloc_info.peer,
470						alloc_info.flags,
471						VMCI_NO_PRIVILEGE_FLAGS,
472						alloc_info.produce_size,
473						alloc_info.consume_size,
474						NULL,
475						vmci_host_dev->context);
476
477		if (vmci_status == VMCI_SUCCESS)
478			vmci_status = VMCI_SUCCESS_QUEUEPAIR_CREATE;
479	} else {
480		struct vmci_qp_alloc_info alloc_info;
481		struct vmci_qp_alloc_info __user *info = uptr;
482		struct vmci_qp_page_store page_store;
483
484		if (copy_from_user(&alloc_info, uptr, sizeof(alloc_info)))
485			return -EFAULT;
486
487		handle = alloc_info.handle;
488		retptr = &info->result;
489
490		page_store.pages = alloc_info.ppn_va;
491		page_store.len = alloc_info.num_ppns;
492
493		vmci_status = vmci_qp_broker_alloc(alloc_info.handle,
494						alloc_info.peer,
495						alloc_info.flags,
496						VMCI_NO_PRIVILEGE_FLAGS,
497						alloc_info.produce_size,
498						alloc_info.consume_size,
499						&page_store,
500						vmci_host_dev->context);
501	}
502
503	if (put_user(vmci_status, retptr)) {
504		if (vmci_status >= VMCI_SUCCESS) {
505			vmci_status = vmci_qp_broker_detach(handle,
506							vmci_host_dev->context);
507		}
508		return -EFAULT;
509	}
510
511	return 0;
512}
513
514static int vmci_host_do_queuepair_setva(struct vmci_host_dev *vmci_host_dev,
515					const char *ioctl_name,
516					void __user *uptr)
517{
518	struct vmci_qp_set_va_info set_va_info;
519	struct vmci_qp_set_va_info __user *info = uptr;
520	s32 result;
521
522	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
523		vmci_ioctl_err("only valid for contexts\n");
524		return -EINVAL;
525	}
526
527	if (vmci_host_dev->user_version < VMCI_VERSION_NOVMVM) {
528		vmci_ioctl_err("is not allowed\n");
529		return -EINVAL;
530	}
531
532	if (copy_from_user(&set_va_info, uptr, sizeof(set_va_info)))
533		return -EFAULT;
534
535	if (set_va_info.va) {
536		/*
537		 * VMX is passing down a new VA for the queue
538		 * pair mapping.
539		 */
540		result = vmci_qp_broker_map(set_va_info.handle,
541					    vmci_host_dev->context,
542					    set_va_info.va);
543	} else {
544		/*
545		 * The queue pair is about to be unmapped by
546		 * the VMX.
547		 */
548		result = vmci_qp_broker_unmap(set_va_info.handle,
549					 vmci_host_dev->context, 0);
550	}
551
552	return put_user(result, &info->result) ? -EFAULT : 0;
553}
554
555static int vmci_host_do_queuepair_setpf(struct vmci_host_dev *vmci_host_dev,
556					const char *ioctl_name,
557					void __user *uptr)
558{
559	struct vmci_qp_page_file_info page_file_info;
560	struct vmci_qp_page_file_info __user *info = uptr;
561	s32 result;
562
563	if (vmci_host_dev->user_version < VMCI_VERSION_HOSTQP ||
564	    vmci_host_dev->user_version >= VMCI_VERSION_NOVMVM) {
565		vmci_ioctl_err("not supported on this VMX (version=%d)\n",
566			       vmci_host_dev->user_version);
567		return -EINVAL;
568	}
569
570	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
571		vmci_ioctl_err("only valid for contexts\n");
572		return -EINVAL;
573	}
574
575	if (copy_from_user(&page_file_info, uptr, sizeof(*info)))
576		return -EFAULT;
577
578	/*
579	 * Communicate success pre-emptively to the caller.  Note that the
580	 * basic premise is that it is incumbent upon the caller not to look at
581	 * the info.result field until after the ioctl() returns.  And then,
582	 * only if the ioctl() result indicates no error.  We send up the
583	 * SUCCESS status before calling SetPageStore() store because failing
584	 * to copy up the result code means unwinding the SetPageStore().
585	 *
586	 * It turns out the logic to unwind a SetPageStore() opens a can of
587	 * worms.  For example, if a host had created the queue_pair and a
588	 * guest attaches and SetPageStore() is successful but writing success
589	 * fails, then ... the host has to be stopped from writing (anymore)
590	 * data into the queue_pair.  That means an additional test in the
591	 * VMCI_Enqueue() code path.  Ugh.
592	 */
593
594	if (put_user(VMCI_SUCCESS, &info->result)) {
595		/*
596		 * In this case, we can't write a result field of the
597		 * caller's info block.  So, we don't even try to
598		 * SetPageStore().
599		 */
600		return -EFAULT;
601	}
602
603	result = vmci_qp_broker_set_page_store(page_file_info.handle,
604						page_file_info.produce_va,
605						page_file_info.consume_va,
606						vmci_host_dev->context);
607	if (result < VMCI_SUCCESS) {
608		if (put_user(result, &info->result)) {
609			/*
610			 * Note that in this case the SetPageStore()
611			 * call failed but we were unable to
612			 * communicate that to the caller (because the
613			 * copy_to_user() call failed).  So, if we
614			 * simply return an error (in this case
615			 * -EFAULT) then the caller will know that the
616			 *  SetPageStore failed even though we couldn't
617			 *  put the result code in the result field and
618			 *  indicate exactly why it failed.
619			 *
620			 * That says nothing about the issue where we
621			 * were once able to write to the caller's info
622			 * memory and now can't.  Something more
623			 * serious is probably going on than the fact
624			 * that SetPageStore() didn't work.
625			 */
626			return -EFAULT;
627		}
628	}
629
630	return 0;
631}
632
633static int vmci_host_do_qp_detach(struct vmci_host_dev *vmci_host_dev,
634				  const char *ioctl_name,
635				  void __user *uptr)
636{
637	struct vmci_qp_dtch_info detach_info;
638	struct vmci_qp_dtch_info __user *info = uptr;
639	s32 result;
640
641	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
642		vmci_ioctl_err("only valid for contexts\n");
643		return -EINVAL;
644	}
645
646	if (copy_from_user(&detach_info, uptr, sizeof(detach_info)))
647		return -EFAULT;
648
649	result = vmci_qp_broker_detach(detach_info.handle,
650				       vmci_host_dev->context);
651	if (result == VMCI_SUCCESS &&
652	    vmci_host_dev->user_version < VMCI_VERSION_NOVMVM) {
653		result = VMCI_SUCCESS_LAST_DETACH;
654	}
655
656	return put_user(result, &info->result) ? -EFAULT : 0;
657}
658
659static int vmci_host_do_ctx_add_notify(struct vmci_host_dev *vmci_host_dev,
660				       const char *ioctl_name,
661				       void __user *uptr)
662{
663	struct vmci_ctx_info ar_info;
664	struct vmci_ctx_info __user *info = uptr;
665	s32 result;
666	u32 cid;
667
668	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
669		vmci_ioctl_err("only valid for contexts\n");
670		return -EINVAL;
671	}
672
673	if (copy_from_user(&ar_info, uptr, sizeof(ar_info)))
674		return -EFAULT;
675
676	cid = vmci_ctx_get_id(vmci_host_dev->context);
677	result = vmci_ctx_add_notification(cid, ar_info.remote_cid);
678
679	return put_user(result, &info->result) ? -EFAULT : 0;
680}
681
682static int vmci_host_do_ctx_remove_notify(struct vmci_host_dev *vmci_host_dev,
683					  const char *ioctl_name,
684					  void __user *uptr)
685{
686	struct vmci_ctx_info ar_info;
687	struct vmci_ctx_info __user *info = uptr;
688	u32 cid;
689	int result;
690
691	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
692		vmci_ioctl_err("only valid for contexts\n");
693		return -EINVAL;
694	}
695
696	if (copy_from_user(&ar_info, uptr, sizeof(ar_info)))
697		return -EFAULT;
698
699	cid = vmci_ctx_get_id(vmci_host_dev->context);
700	result = vmci_ctx_remove_notification(cid,
701					      ar_info.remote_cid);
702
703	return put_user(result, &info->result) ? -EFAULT : 0;
704}
705
706static int vmci_host_do_ctx_get_cpt_state(struct vmci_host_dev *vmci_host_dev,
707					  const char *ioctl_name,
708					  void __user *uptr)
709{
710	struct vmci_ctx_chkpt_buf_info get_info;
711	u32 cid;
712	void *cpt_buf;
713	int retval;
714
715	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
716		vmci_ioctl_err("only valid for contexts\n");
717		return -EINVAL;
718	}
719
720	if (copy_from_user(&get_info, uptr, sizeof(get_info)))
721		return -EFAULT;
722
723	cid = vmci_ctx_get_id(vmci_host_dev->context);
724	get_info.result = vmci_ctx_get_chkpt_state(cid, get_info.cpt_type,
725						&get_info.buf_size, &cpt_buf);
726	if (get_info.result == VMCI_SUCCESS && get_info.buf_size) {
727		void __user *ubuf = (void __user *)(uintptr_t)get_info.cpt_buf;
728		retval = copy_to_user(ubuf, cpt_buf, get_info.buf_size);
729		kfree(cpt_buf);
730
731		if (retval)
732			return -EFAULT;
733	}
734
735	return copy_to_user(uptr, &get_info, sizeof(get_info)) ? -EFAULT : 0;
736}
737
738static int vmci_host_do_ctx_set_cpt_state(struct vmci_host_dev *vmci_host_dev,
739					  const char *ioctl_name,
740					  void __user *uptr)
741{
742	struct vmci_ctx_chkpt_buf_info set_info;
743	u32 cid;
744	void *cpt_buf;
745	int retval;
746
747	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
748		vmci_ioctl_err("only valid for contexts\n");
749		return -EINVAL;
750	}
751
752	if (copy_from_user(&set_info, uptr, sizeof(set_info)))
753		return -EFAULT;
754
755	cpt_buf = kmalloc(set_info.buf_size, GFP_KERNEL);
756	if (!cpt_buf) {
757		vmci_ioctl_err(
758			"cannot allocate memory to set cpt state (type=%d)\n",
759			set_info.cpt_type);
760		return -ENOMEM;
761	}
762
763	if (copy_from_user(cpt_buf, (void __user *)(uintptr_t)set_info.cpt_buf,
764			   set_info.buf_size)) {
765		retval = -EFAULT;
766		goto out;
767	}
768
769	cid = vmci_ctx_get_id(vmci_host_dev->context);
770	set_info.result = vmci_ctx_set_chkpt_state(cid, set_info.cpt_type,
771						   set_info.buf_size, cpt_buf);
772
773	retval = copy_to_user(uptr, &set_info, sizeof(set_info)) ? -EFAULT : 0;
774
775out:
776	kfree(cpt_buf);
777	return retval;
778}
779
780static int vmci_host_do_get_context_id(struct vmci_host_dev *vmci_host_dev,
781				       const char *ioctl_name,
782				       void __user *uptr)
783{
784	u32 __user *u32ptr = uptr;
785
786	return put_user(VMCI_HOST_CONTEXT_ID, u32ptr) ? -EFAULT : 0;
787}
788
789static int vmci_host_do_set_notify(struct vmci_host_dev *vmci_host_dev,
790				   const char *ioctl_name,
791				   void __user *uptr)
792{
793	struct vmci_set_notify_info notify_info;
794
795	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
796		vmci_ioctl_err("only valid for contexts\n");
797		return -EINVAL;
798	}
799
800	if (copy_from_user(&notify_info, uptr, sizeof(notify_info)))
801		return -EFAULT;
802
803	if (notify_info.notify_uva) {
804		notify_info.result =
805			vmci_host_setup_notify(vmci_host_dev->context,
806					       notify_info.notify_uva);
807	} else {
808		vmci_ctx_unset_notify(vmci_host_dev->context);
809		notify_info.result = VMCI_SUCCESS;
810	}
811
812	return copy_to_user(uptr, &notify_info, sizeof(notify_info)) ?
813		-EFAULT : 0;
814}
815
816static int vmci_host_do_notify_resource(struct vmci_host_dev *vmci_host_dev,
817					const char *ioctl_name,
818					void __user *uptr)
819{
820	struct vmci_dbell_notify_resource_info info;
821	u32 cid;
822
823	if (vmci_host_dev->user_version < VMCI_VERSION_NOTIFY) {
824		vmci_ioctl_err("invalid for current VMX versions\n");
825		return -EINVAL;
826	}
827
828	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
829		vmci_ioctl_err("only valid for contexts\n");
830		return -EINVAL;
831	}
832
833	if (copy_from_user(&info, uptr, sizeof(info)))
834		return -EFAULT;
835
836	cid = vmci_ctx_get_id(vmci_host_dev->context);
837
838	switch (info.action) {
839	case VMCI_NOTIFY_RESOURCE_ACTION_NOTIFY:
840		if (info.resource == VMCI_NOTIFY_RESOURCE_DOOR_BELL) {
841			u32 flags = VMCI_NO_PRIVILEGE_FLAGS;
842			info.result = vmci_ctx_notify_dbell(cid, info.handle,
843							    flags);
844		} else {
845			info.result = VMCI_ERROR_UNAVAILABLE;
846		}
847		break;
848
849	case VMCI_NOTIFY_RESOURCE_ACTION_CREATE:
850		info.result = vmci_ctx_dbell_create(cid, info.handle);
851		break;
852
853	case VMCI_NOTIFY_RESOURCE_ACTION_DESTROY:
854		info.result = vmci_ctx_dbell_destroy(cid, info.handle);
855		break;
856
857	default:
858		vmci_ioctl_err("got unknown action (action=%d)\n",
859			       info.action);
860		info.result = VMCI_ERROR_INVALID_ARGS;
861	}
862
863	return copy_to_user(uptr, &info, sizeof(info)) ? -EFAULT : 0;
864}
865
866static int vmci_host_do_recv_notifications(struct vmci_host_dev *vmci_host_dev,
867					   const char *ioctl_name,
868					   void __user *uptr)
869{
870	struct vmci_ctx_notify_recv_info info;
871	struct vmci_handle_arr *db_handle_array;
872	struct vmci_handle_arr *qp_handle_array;
873	void __user *ubuf;
874	u32 cid;
875	int retval = 0;
876
877	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
878		vmci_ioctl_err("only valid for contexts\n");
879		return -EINVAL;
880	}
881
882	if (vmci_host_dev->user_version < VMCI_VERSION_NOTIFY) {
883		vmci_ioctl_err("not supported for the current vmx version\n");
884		return -EINVAL;
885	}
886
887	if (copy_from_user(&info, uptr, sizeof(info)))
888		return -EFAULT;
889
890	if ((info.db_handle_buf_size && !info.db_handle_buf_uva) ||
891	    (info.qp_handle_buf_size && !info.qp_handle_buf_uva)) {
892		return -EINVAL;
893	}
894
895	cid = vmci_ctx_get_id(vmci_host_dev->context);
896
897	info.result = vmci_ctx_rcv_notifications_get(cid,
898				&db_handle_array, &qp_handle_array);
899	if (info.result != VMCI_SUCCESS)
900		return copy_to_user(uptr, &info, sizeof(info)) ? -EFAULT : 0;
901
902	ubuf = (void __user *)(uintptr_t)info.db_handle_buf_uva;
903	info.result = drv_cp_harray_to_user(ubuf, &info.db_handle_buf_size,
904					    db_handle_array, &retval);
905	if (info.result == VMCI_SUCCESS && !retval) {
906		ubuf = (void __user *)(uintptr_t)info.qp_handle_buf_uva;
907		info.result = drv_cp_harray_to_user(ubuf,
908						    &info.qp_handle_buf_size,
909						    qp_handle_array, &retval);
910	}
911
912	if (!retval && copy_to_user(uptr, &info, sizeof(info)))
913		retval = -EFAULT;
914
915	vmci_ctx_rcv_notifications_release(cid,
916				db_handle_array, qp_handle_array,
917				info.result == VMCI_SUCCESS && !retval);
918
919	return retval;
920}
921
922static long vmci_host_unlocked_ioctl(struct file *filp,
923				     unsigned int iocmd, unsigned long ioarg)
924{
925#define VMCI_DO_IOCTL(ioctl_name, ioctl_fn) do {			\
926		char *name = __stringify(IOCTL_VMCI_ ## ioctl_name);	\
927		return vmci_host_do_ ## ioctl_fn(			\
928			vmci_host_dev, name, uptr);			\
929	} while (0)
930
931	struct vmci_host_dev *vmci_host_dev = filp->private_data;
932	void __user *uptr = (void __user *)ioarg;
933
934	switch (iocmd) {
935	case IOCTL_VMCI_INIT_CONTEXT:
936		VMCI_DO_IOCTL(INIT_CONTEXT, init_context);
937	case IOCTL_VMCI_DATAGRAM_SEND:
938		VMCI_DO_IOCTL(DATAGRAM_SEND, send_datagram);
939	case IOCTL_VMCI_DATAGRAM_RECEIVE:
940		VMCI_DO_IOCTL(DATAGRAM_RECEIVE, receive_datagram);
941	case IOCTL_VMCI_QUEUEPAIR_ALLOC:
942		VMCI_DO_IOCTL(QUEUEPAIR_ALLOC, alloc_queuepair);
943	case IOCTL_VMCI_QUEUEPAIR_SETVA:
944		VMCI_DO_IOCTL(QUEUEPAIR_SETVA, queuepair_setva);
945	case IOCTL_VMCI_QUEUEPAIR_SETPAGEFILE:
946		VMCI_DO_IOCTL(QUEUEPAIR_SETPAGEFILE, queuepair_setpf);
947	case IOCTL_VMCI_QUEUEPAIR_DETACH:
948		VMCI_DO_IOCTL(QUEUEPAIR_DETACH, qp_detach);
949	case IOCTL_VMCI_CTX_ADD_NOTIFICATION:
950		VMCI_DO_IOCTL(CTX_ADD_NOTIFICATION, ctx_add_notify);
951	case IOCTL_VMCI_CTX_REMOVE_NOTIFICATION:
952		VMCI_DO_IOCTL(CTX_REMOVE_NOTIFICATION, ctx_remove_notify);
953	case IOCTL_VMCI_CTX_GET_CPT_STATE:
954		VMCI_DO_IOCTL(CTX_GET_CPT_STATE, ctx_get_cpt_state);
955	case IOCTL_VMCI_CTX_SET_CPT_STATE:
956		VMCI_DO_IOCTL(CTX_SET_CPT_STATE, ctx_set_cpt_state);
957	case IOCTL_VMCI_GET_CONTEXT_ID:
958		VMCI_DO_IOCTL(GET_CONTEXT_ID, get_context_id);
959	case IOCTL_VMCI_SET_NOTIFY:
960		VMCI_DO_IOCTL(SET_NOTIFY, set_notify);
961	case IOCTL_VMCI_NOTIFY_RESOURCE:
962		VMCI_DO_IOCTL(NOTIFY_RESOURCE, notify_resource);
963	case IOCTL_VMCI_NOTIFICATIONS_RECEIVE:
964		VMCI_DO_IOCTL(NOTIFICATIONS_RECEIVE, recv_notifications);
965
966	case IOCTL_VMCI_VERSION:
967	case IOCTL_VMCI_VERSION2:
968		return vmci_host_get_version(vmci_host_dev, iocmd, uptr);
969
970	default:
971		pr_devel("%s: Unknown ioctl (iocmd=%d)\n", __func__, iocmd);
972		return -EINVAL;
973	}
974
975#undef VMCI_DO_IOCTL
976}
977
978static const struct file_operations vmuser_fops = {
979	.owner		= THIS_MODULE,
980	.open		= vmci_host_open,
981	.release	= vmci_host_close,
982	.poll		= vmci_host_poll,
983	.unlocked_ioctl	= vmci_host_unlocked_ioctl,
984	.compat_ioctl	= vmci_host_unlocked_ioctl,
985};
986
987static struct miscdevice vmci_host_miscdev = {
988	 .name = "vmci",
989	 .minor = MISC_DYNAMIC_MINOR,
990	 .fops = &vmuser_fops,
991};
992
993int __init vmci_host_init(void)
994{
995	int error;
996
997	host_context = vmci_ctx_create(VMCI_HOST_CONTEXT_ID,
998					VMCI_DEFAULT_PROC_PRIVILEGE_FLAGS,
999					-1, VMCI_VERSION, NULL);
1000	if (IS_ERR(host_context)) {
1001		error = PTR_ERR(host_context);
1002		pr_warn("Failed to initialize VMCIContext (error%d)\n",
1003			error);
1004		return error;
1005	}
1006
1007	error = misc_register(&vmci_host_miscdev);
1008	if (error) {
1009		pr_warn("Module registration error (name=%s, major=%d, minor=%d, err=%d)\n",
1010			vmci_host_miscdev.name,
1011			MISC_MAJOR, vmci_host_miscdev.minor,
1012			error);
1013		pr_warn("Unable to initialize host personality\n");
1014		vmci_ctx_destroy(host_context);
1015		return error;
1016	}
1017
1018	pr_info("VMCI host device registered (name=%s, major=%d, minor=%d)\n",
1019		vmci_host_miscdev.name, MISC_MAJOR, vmci_host_miscdev.minor);
1020
1021	vmci_host_device_initialized = true;
1022	return 0;
1023}
1024
1025void __exit vmci_host_exit(void)
1026{
1027	int error;
1028
1029	vmci_host_device_initialized = false;
1030
1031	error = misc_deregister(&vmci_host_miscdev);
1032	if (error)
1033		pr_warn("Error unregistering character device: %d\n", error);
1034
1035	vmci_ctx_destroy(host_context);
1036	vmci_qp_broker_exit();
1037
1038	pr_debug("VMCI host driver module unloaded\n");
1039}
1040