blob: f866a4baecb5407cb952e2ff6e55f4b671ed63a4 [file] [log] [blame]
George Zhang28d66922013-01-08 15:52:59 -08001/*
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/highmem.h>
19#include <linux/kernel.h>
20#include <linux/module.h>
21#include <linux/sched.h>
22#include <linux/slab.h>
23
24#include "vmci_queue_pair.h"
25#include "vmci_datagram.h"
26#include "vmci_doorbell.h"
27#include "vmci_context.h"
28#include "vmci_driver.h"
29#include "vmci_event.h"
30
31/*
32 * List of current VMCI contexts. Contexts can be added by
33 * vmci_ctx_create() and removed via vmci_ctx_destroy().
34 * These, along with context lookup, are protected by the
35 * list structure's lock.
36 */
37static struct {
38 struct list_head head;
39 spinlock_t lock; /* Spinlock for context list operations */
40} ctx_list = {
41 .head = LIST_HEAD_INIT(ctx_list.head),
42 .lock = __SPIN_LOCK_UNLOCKED(ctx_list.lock),
43};
44
45/* Used by contexts that did not set up notify flag pointers */
46static bool ctx_dummy_notify;
47
48static void ctx_signal_notify(struct vmci_ctx *context)
49{
50 *context->notify = true;
51}
52
53static void ctx_clear_notify(struct vmci_ctx *context)
54{
55 *context->notify = false;
56}
57
58/*
59 * If nothing requires the attention of the guest, clears both
60 * notify flag and call.
61 */
62static void ctx_clear_notify_call(struct vmci_ctx *context)
63{
64 if (context->pending_datagrams == 0 &&
65 vmci_handle_arr_get_size(context->pending_doorbell_array) == 0)
66 ctx_clear_notify(context);
67}
68
69/*
70 * Sets the context's notify flag iff datagrams are pending for this
71 * context. Called from vmci_setup_notify().
72 */
73void vmci_ctx_check_signal_notify(struct vmci_ctx *context)
74{
75 spin_lock(&context->lock);
76 if (context->pending_datagrams)
77 ctx_signal_notify(context);
78 spin_unlock(&context->lock);
79}
80
81/*
82 * Allocates and initializes a VMCI context.
83 */
84struct vmci_ctx *vmci_ctx_create(u32 cid, u32 priv_flags,
85 uintptr_t event_hnd,
86 int user_version,
87 const struct cred *cred)
88{
89 struct vmci_ctx *context;
90 int error;
91
92 if (cid == VMCI_INVALID_ID) {
93 pr_devel("Invalid context ID for VMCI context\n");
94 error = -EINVAL;
95 goto err_out;
96 }
97
98 if (priv_flags & ~VMCI_PRIVILEGE_ALL_FLAGS) {
99 pr_devel("Invalid flag (flags=0x%x) for VMCI context\n",
100 priv_flags);
101 error = -EINVAL;
102 goto err_out;
103 }
104
105 if (user_version == 0) {
106 pr_devel("Invalid suer_version %d\n", user_version);
107 error = -EINVAL;
108 goto err_out;
109 }
110
111 context = kzalloc(sizeof(*context), GFP_KERNEL);
112 if (!context) {
113 pr_warn("Failed to allocate memory for VMCI context\n");
114 error = -EINVAL;
115 goto err_out;
116 }
117
118 kref_init(&context->kref);
119 spin_lock_init(&context->lock);
120 INIT_LIST_HEAD(&context->list_item);
121 INIT_LIST_HEAD(&context->datagram_queue);
122 INIT_LIST_HEAD(&context->notifier_list);
123
124 /* Initialize host-specific VMCI context. */
125 init_waitqueue_head(&context->host_context.wait_queue);
126
127 context->queue_pair_array = vmci_handle_arr_create(0);
128 if (!context->queue_pair_array) {
129 error = -ENOMEM;
130 goto err_free_ctx;
131 }
132
133 context->doorbell_array = vmci_handle_arr_create(0);
134 if (!context->doorbell_array) {
135 error = -ENOMEM;
136 goto err_free_qp_array;
137 }
138
139 context->pending_doorbell_array = vmci_handle_arr_create(0);
140 if (!context->pending_doorbell_array) {
141 error = -ENOMEM;
142 goto err_free_db_array;
143 }
144
145 context->user_version = user_version;
146
147 context->priv_flags = priv_flags;
148
149 if (cred)
150 context->cred = get_cred(cred);
151
152 context->notify = &ctx_dummy_notify;
153 context->notify_page = NULL;
154
155 /*
156 * If we collide with an existing context we generate a new
157 * and use it instead. The VMX will determine if regeneration
158 * is okay. Since there isn't 4B - 16 VMs running on a given
159 * host, the below loop will terminate.
160 */
161 spin_lock(&ctx_list.lock);
162
163 while (vmci_ctx_exists(cid)) {
164 /* We reserve the lowest 16 ids for fixed contexts. */
165 cid = max(cid, VMCI_RESERVED_CID_LIMIT - 1) + 1;
166 if (cid == VMCI_INVALID_ID)
167 cid = VMCI_RESERVED_CID_LIMIT;
168 }
169 context->cid = cid;
170
171 list_add_tail_rcu(&context->list_item, &ctx_list.head);
172 spin_unlock(&ctx_list.lock);
173
174 return context;
175
176 err_free_db_array:
177 vmci_handle_arr_destroy(context->doorbell_array);
178 err_free_qp_array:
179 vmci_handle_arr_destroy(context->queue_pair_array);
180 err_free_ctx:
181 kfree(context);
182 err_out:
183 return ERR_PTR(error);
184}
185
186/*
187 * Destroy VMCI context.
188 */
189void vmci_ctx_destroy(struct vmci_ctx *context)
190{
191 spin_lock(&ctx_list.lock);
192 list_del_rcu(&context->list_item);
193 spin_unlock(&ctx_list.lock);
194 synchronize_rcu();
195
196 vmci_ctx_put(context);
197}
198
199/*
200 * Fire notification for all contexts interested in given cid.
201 */
202static int ctx_fire_notification(u32 context_id, u32 priv_flags)
203{
204 u32 i, array_size;
205 struct vmci_ctx *sub_ctx;
206 struct vmci_handle_arr *subscriber_array;
207 struct vmci_handle context_handle =
208 vmci_make_handle(context_id, VMCI_EVENT_HANDLER);
209
210 /*
211 * We create an array to hold the subscribers we find when
212 * scanning through all contexts.
213 */
214 subscriber_array = vmci_handle_arr_create(0);
215 if (subscriber_array == NULL)
216 return VMCI_ERROR_NO_MEM;
217
218 /*
219 * Scan all contexts to find who is interested in being
220 * notified about given contextID.
221 */
222 rcu_read_lock();
223 list_for_each_entry_rcu(sub_ctx, &ctx_list.head, list_item) {
224 struct vmci_handle_list *node;
225
226 /*
227 * We only deliver notifications of the removal of
228 * contexts, if the two contexts are allowed to
229 * interact.
230 */
231 if (vmci_deny_interaction(priv_flags, sub_ctx->priv_flags))
232 continue;
233
234 list_for_each_entry_rcu(node, &sub_ctx->notifier_list, node) {
235 if (!vmci_handle_is_equal(node->handle, context_handle))
236 continue;
237
238 vmci_handle_arr_append_entry(&subscriber_array,
239 vmci_make_handle(sub_ctx->cid,
240 VMCI_EVENT_HANDLER));
241 }
242 }
243 rcu_read_unlock();
244
245 /* Fire event to all subscribers. */
246 array_size = vmci_handle_arr_get_size(subscriber_array);
247 for (i = 0; i < array_size; i++) {
248 int result;
249 struct vmci_event_ctx ev;
250
251 ev.msg.hdr.dst = vmci_handle_arr_get_entry(subscriber_array, i);
252 ev.msg.hdr.src = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID,
253 VMCI_CONTEXT_RESOURCE_ID);
254 ev.msg.hdr.payload_size = sizeof(ev) - sizeof(ev.msg.hdr);
255 ev.msg.event_data.event = VMCI_EVENT_CTX_REMOVED;
256 ev.payload.context_id = context_id;
257
258 result = vmci_datagram_dispatch(VMCI_HYPERVISOR_CONTEXT_ID,
259 &ev.msg.hdr, false);
260 if (result < VMCI_SUCCESS) {
261 pr_devel("Failed to enqueue event datagram (type=%d) for context (ID=0x%x)\n",
262 ev.msg.event_data.event,
263 ev.msg.hdr.dst.context);
264 /* We continue to enqueue on next subscriber. */
265 }
266 }
267 vmci_handle_arr_destroy(subscriber_array);
268
269 return VMCI_SUCCESS;
270}
271
272/*
273 * Returns the current number of pending datagrams. The call may
274 * also serve as a synchronization point for the datagram queue,
275 * as no enqueue operations can occur concurrently.
276 */
277int vmci_ctx_pending_datagrams(u32 cid, u32 *pending)
278{
279 struct vmci_ctx *context;
280
281 context = vmci_ctx_get(cid);
282 if (context == NULL)
283 return VMCI_ERROR_INVALID_ARGS;
284
285 spin_lock(&context->lock);
286 if (pending)
287 *pending = context->pending_datagrams;
288 spin_unlock(&context->lock);
289 vmci_ctx_put(context);
290
291 return VMCI_SUCCESS;
292}
293
294/*
295 * Queues a VMCI datagram for the appropriate target VM context.
296 */
297int vmci_ctx_enqueue_datagram(u32 cid, struct vmci_datagram *dg)
298{
299 struct vmci_datagram_queue_entry *dq_entry;
300 struct vmci_ctx *context;
301 struct vmci_handle dg_src;
302 size_t vmci_dg_size;
303
304 vmci_dg_size = VMCI_DG_SIZE(dg);
305 if (vmci_dg_size > VMCI_MAX_DG_SIZE) {
306 pr_devel("Datagram too large (bytes=%Zu)\n", vmci_dg_size);
307 return VMCI_ERROR_INVALID_ARGS;
308 }
309
310 /* Get the target VM's VMCI context. */
311 context = vmci_ctx_get(cid);
312 if (!context) {
313 pr_devel("Invalid context (ID=0x%x)\n", cid);
314 return VMCI_ERROR_INVALID_ARGS;
315 }
316
317 /* Allocate guest call entry and add it to the target VM's queue. */
318 dq_entry = kmalloc(sizeof(*dq_entry), GFP_KERNEL);
319 if (dq_entry == NULL) {
320 pr_warn("Failed to allocate memory for datagram\n");
321 vmci_ctx_put(context);
322 return VMCI_ERROR_NO_MEM;
323 }
324 dq_entry->dg = dg;
325 dq_entry->dg_size = vmci_dg_size;
326 dg_src = dg->src;
327 INIT_LIST_HEAD(&dq_entry->list_item);
328
329 spin_lock(&context->lock);
330
331 /*
332 * We put a higher limit on datagrams from the hypervisor. If
333 * the pending datagram is not from hypervisor, then we check
334 * if enqueueing it would exceed the
335 * VMCI_MAX_DATAGRAM_QUEUE_SIZE limit on the destination. If
336 * the pending datagram is from hypervisor, we allow it to be
337 * queued at the destination side provided we don't reach the
338 * VMCI_MAX_DATAGRAM_AND_EVENT_QUEUE_SIZE limit.
339 */
340 if (context->datagram_queue_size + vmci_dg_size >=
341 VMCI_MAX_DATAGRAM_QUEUE_SIZE &&
342 (!vmci_handle_is_equal(dg_src,
343 vmci_make_handle
344 (VMCI_HYPERVISOR_CONTEXT_ID,
345 VMCI_CONTEXT_RESOURCE_ID)) ||
346 context->datagram_queue_size + vmci_dg_size >=
347 VMCI_MAX_DATAGRAM_AND_EVENT_QUEUE_SIZE)) {
348 spin_unlock(&context->lock);
349 vmci_ctx_put(context);
350 kfree(dq_entry);
351 pr_devel("Context (ID=0x%x) receive queue is full\n", cid);
352 return VMCI_ERROR_NO_RESOURCES;
353 }
354
355 list_add(&dq_entry->list_item, &context->datagram_queue);
356 context->pending_datagrams++;
357 context->datagram_queue_size += vmci_dg_size;
358 ctx_signal_notify(context);
359 wake_up(&context->host_context.wait_queue);
360 spin_unlock(&context->lock);
361 vmci_ctx_put(context);
362
363 return vmci_dg_size;
364}
365
366/*
367 * Verifies whether a context with the specified context ID exists.
368 * FIXME: utility is dubious as no decisions can be reliably made
369 * using this data as context can appear and disappear at any time.
370 */
371bool vmci_ctx_exists(u32 cid)
372{
373 struct vmci_ctx *context;
374 bool exists = false;
375
376 rcu_read_lock();
377
378 list_for_each_entry_rcu(context, &ctx_list.head, list_item) {
379 if (context->cid == cid) {
380 exists = true;
381 break;
382 }
383 }
384
385 rcu_read_unlock();
386 return exists;
387}
388
389/*
390 * Retrieves VMCI context corresponding to the given cid.
391 */
392struct vmci_ctx *vmci_ctx_get(u32 cid)
393{
394 struct vmci_ctx *c, *context = NULL;
395
396 if (cid == VMCI_INVALID_ID)
397 return NULL;
398
399 rcu_read_lock();
400 list_for_each_entry_rcu(c, &ctx_list.head, list_item) {
401 if (c->cid == cid) {
402 /*
403 * The context owner drops its own reference to the
404 * context only after removing it from the list and
405 * waiting for RCU grace period to expire. This
406 * means that we are not about to increase the
407 * reference count of something that is in the
408 * process of being destroyed.
409 */
410 context = c;
411 kref_get(&context->kref);
412 break;
413 }
414 }
415 rcu_read_unlock();
416
417 return context;
418}
419
420/*
421 * Deallocates all parts of a context data structure. This
422 * function doesn't lock the context, because it assumes that
423 * the caller was holding the last reference to context.
424 */
425static void ctx_free_ctx(struct kref *kref)
426{
427 struct vmci_ctx *context = container_of(kref, struct vmci_ctx, kref);
428 struct vmci_datagram_queue_entry *dq_entry, *dq_entry_tmp;
429 struct vmci_handle temp_handle;
430 struct vmci_handle_list *notifier, *tmp;
431
432 /*
433 * Fire event to all contexts interested in knowing this
434 * context is dying.
435 */
436 ctx_fire_notification(context->cid, context->priv_flags);
437
438 /*
439 * Cleanup all queue pair resources attached to context. If
440 * the VM dies without cleaning up, this code will make sure
441 * that no resources are leaked.
442 */
443 temp_handle = vmci_handle_arr_get_entry(context->queue_pair_array, 0);
444 while (!vmci_handle_is_equal(temp_handle, VMCI_INVALID_HANDLE)) {
445 if (vmci_qp_broker_detach(temp_handle,
446 context) < VMCI_SUCCESS) {
447 /*
448 * When vmci_qp_broker_detach() succeeds it
449 * removes the handle from the array. If
450 * detach fails, we must remove the handle
451 * ourselves.
452 */
453 vmci_handle_arr_remove_entry(context->queue_pair_array,
454 temp_handle);
455 }
456 temp_handle =
457 vmci_handle_arr_get_entry(context->queue_pair_array, 0);
458 }
459
460 /*
461 * It is fine to destroy this without locking the callQueue, as
462 * this is the only thread having a reference to the context.
463 */
464 list_for_each_entry_safe(dq_entry, dq_entry_tmp,
465 &context->datagram_queue, list_item) {
466 WARN_ON(dq_entry->dg_size != VMCI_DG_SIZE(dq_entry->dg));
467 list_del(&dq_entry->list_item);
468 kfree(dq_entry->dg);
469 kfree(dq_entry);
470 }
471
472 list_for_each_entry_safe(notifier, tmp,
473 &context->notifier_list, node) {
474 list_del(&notifier->node);
475 kfree(notifier);
476 }
477
478 vmci_handle_arr_destroy(context->queue_pair_array);
479 vmci_handle_arr_destroy(context->doorbell_array);
480 vmci_handle_arr_destroy(context->pending_doorbell_array);
481 vmci_ctx_unset_notify(context);
482 if (context->cred)
483 put_cred(context->cred);
484 kfree(context);
485}
486
487/*
488 * Drops reference to VMCI context. If this is the last reference to
489 * the context it will be deallocated. A context is created with
490 * a reference count of one, and on destroy, it is removed from
491 * the context list before its reference count is decremented. Thus,
492 * if we reach zero, we are sure that nobody else are about to increment
493 * it (they need the entry in the context list for that), and so there
494 * is no need for locking.
495 */
496void vmci_ctx_put(struct vmci_ctx *context)
497{
498 kref_put(&context->kref, ctx_free_ctx);
499}
500
501/*
502 * Dequeues the next datagram and returns it to caller.
503 * The caller passes in a pointer to the max size datagram
504 * it can handle and the datagram is only unqueued if the
505 * size is less than max_size. If larger max_size is set to
506 * the size of the datagram to give the caller a chance to
507 * set up a larger buffer for the guestcall.
508 */
509int vmci_ctx_dequeue_datagram(struct vmci_ctx *context,
510 size_t *max_size,
511 struct vmci_datagram **dg)
512{
513 struct vmci_datagram_queue_entry *dq_entry;
514 struct list_head *list_item;
515 int rv;
516
517 /* Dequeue the next datagram entry. */
518 spin_lock(&context->lock);
519 if (context->pending_datagrams == 0) {
520 ctx_clear_notify_call(context);
521 spin_unlock(&context->lock);
522 pr_devel("No datagrams pending\n");
523 return VMCI_ERROR_NO_MORE_DATAGRAMS;
524 }
525
526 list_item = context->datagram_queue.next;
527
528 dq_entry =
529 list_entry(list_item, struct vmci_datagram_queue_entry, list_item);
530
531 /* Check size of caller's buffer. */
532 if (*max_size < dq_entry->dg_size) {
533 *max_size = dq_entry->dg_size;
534 spin_unlock(&context->lock);
535 pr_devel("Caller's buffer should be at least (size=%u bytes)\n",
536 (u32) *max_size);
537 return VMCI_ERROR_NO_MEM;
538 }
539
540 list_del(list_item);
541 context->pending_datagrams--;
542 context->datagram_queue_size -= dq_entry->dg_size;
543 if (context->pending_datagrams == 0) {
544 ctx_clear_notify_call(context);
545 rv = VMCI_SUCCESS;
546 } else {
547 /*
548 * Return the size of the next datagram.
549 */
550 struct vmci_datagram_queue_entry *next_entry;
551
552 list_item = context->datagram_queue.next;
553 next_entry =
554 list_entry(list_item, struct vmci_datagram_queue_entry,
555 list_item);
556
557 /*
558 * The following size_t -> int truncation is fine as
559 * the maximum size of a (routable) datagram is 68KB.
560 */
561 rv = (int)next_entry->dg_size;
562 }
563 spin_unlock(&context->lock);
564
565 /* Caller must free datagram. */
566 *dg = dq_entry->dg;
567 dq_entry->dg = NULL;
568 kfree(dq_entry);
569
570 return rv;
571}
572
573/*
574 * Reverts actions set up by vmci_setup_notify(). Unmaps and unlocks the
575 * page mapped/locked by vmci_setup_notify().
576 */
577void vmci_ctx_unset_notify(struct vmci_ctx *context)
578{
579 struct page *notify_page;
580
581 spin_lock(&context->lock);
582
583 notify_page = context->notify_page;
584 context->notify = &ctx_dummy_notify;
585 context->notify_page = NULL;
586
587 spin_unlock(&context->lock);
588
589 if (notify_page) {
590 kunmap(notify_page);
591 put_page(notify_page);
592 }
593}
594
595/*
596 * Add remote_cid to list of contexts current contexts wants
597 * notifications from/about.
598 */
599int vmci_ctx_add_notification(u32 context_id, u32 remote_cid)
600{
601 struct vmci_ctx *context;
602 struct vmci_handle_list *notifier, *n;
603 int result;
604 bool exists = false;
605
606 context = vmci_ctx_get(context_id);
607 if (!context)
608 return VMCI_ERROR_NOT_FOUND;
609
610 if (VMCI_CONTEXT_IS_VM(context_id) && VMCI_CONTEXT_IS_VM(remote_cid)) {
611 pr_devel("Context removed notifications for other VMs not supported (src=0x%x, remote=0x%x)\n",
612 context_id, remote_cid);
613 result = VMCI_ERROR_DST_UNREACHABLE;
614 goto out;
615 }
616
617 if (context->priv_flags & VMCI_PRIVILEGE_FLAG_RESTRICTED) {
618 result = VMCI_ERROR_NO_ACCESS;
619 goto out;
620 }
621
622 notifier = kmalloc(sizeof(struct vmci_handle_list), GFP_KERNEL);
623 if (!notifier) {
624 result = VMCI_ERROR_NO_MEM;
625 goto out;
626 }
627
628 INIT_LIST_HEAD(&notifier->node);
629 notifier->handle = vmci_make_handle(remote_cid, VMCI_EVENT_HANDLER);
630
631 spin_lock(&context->lock);
632
633 list_for_each_entry(n, &context->notifier_list, node) {
634 if (vmci_handle_is_equal(n->handle, notifier->handle)) {
635 exists = true;
636 break;
637 }
638 }
639
640 if (exists) {
641 kfree(notifier);
642 result = VMCI_ERROR_ALREADY_EXISTS;
643 } else {
644 list_add_tail_rcu(&notifier->node, &context->notifier_list);
645 context->n_notifiers++;
646 result = VMCI_SUCCESS;
647 }
648
649 spin_unlock(&context->lock);
650
651 out:
652 vmci_ctx_put(context);
653 return result;
654}
655
656/*
657 * Remove remote_cid from current context's list of contexts it is
658 * interested in getting notifications from/about.
659 */
660int vmci_ctx_remove_notification(u32 context_id, u32 remote_cid)
661{
662 struct vmci_ctx *context;
663 struct vmci_handle_list *notifier, *tmp;
664 struct vmci_handle handle;
665 bool found = false;
666
667 context = vmci_ctx_get(context_id);
668 if (!context)
669 return VMCI_ERROR_NOT_FOUND;
670
671 handle = vmci_make_handle(remote_cid, VMCI_EVENT_HANDLER);
672
673 spin_lock(&context->lock);
674 list_for_each_entry_safe(notifier, tmp,
675 &context->notifier_list, node) {
676 if (vmci_handle_is_equal(notifier->handle, handle)) {
677 list_del_rcu(&notifier->node);
678 context->n_notifiers--;
679 found = true;
680 break;
681 }
682 }
683 spin_unlock(&context->lock);
684
685 if (found) {
686 synchronize_rcu();
687 kfree(notifier);
688 }
689
690 vmci_ctx_put(context);
691
692 return found ? VMCI_SUCCESS : VMCI_ERROR_NOT_FOUND;
693}
694
695static int vmci_ctx_get_chkpt_notifiers(struct vmci_ctx *context,
696 u32 *buf_size, void **pbuf)
697{
698 u32 *notifiers;
699 size_t data_size;
700 struct vmci_handle_list *entry;
701 int i = 0;
702
703 if (context->n_notifiers == 0) {
704 *buf_size = 0;
705 *pbuf = NULL;
706 return VMCI_SUCCESS;
707 }
708
709 data_size = context->n_notifiers * sizeof(*notifiers);
710 if (*buf_size < data_size) {
711 *buf_size = data_size;
712 return VMCI_ERROR_MORE_DATA;
713 }
714
715 notifiers = kmalloc(data_size, GFP_ATOMIC); /* FIXME: want GFP_KERNEL */
716 if (!notifiers)
717 return VMCI_ERROR_NO_MEM;
718
719 list_for_each_entry(entry, &context->notifier_list, node)
720 notifiers[i++] = entry->handle.context;
721
722 *buf_size = data_size;
723 *pbuf = notifiers;
724 return VMCI_SUCCESS;
725}
726
727static int vmci_ctx_get_chkpt_doorbells(struct vmci_ctx *context,
728 u32 *buf_size, void **pbuf)
729{
730 struct dbell_cpt_state *dbells;
731 size_t n_doorbells;
732 int i;
733
734 n_doorbells = vmci_handle_arr_get_size(context->doorbell_array);
735 if (n_doorbells > 0) {
736 size_t data_size = n_doorbells * sizeof(*dbells);
737 if (*buf_size < data_size) {
738 *buf_size = data_size;
739 return VMCI_ERROR_MORE_DATA;
740 }
741
742 dbells = kmalloc(data_size, GFP_ATOMIC);
743 if (!dbells)
744 return VMCI_ERROR_NO_MEM;
745
746 for (i = 0; i < n_doorbells; i++)
747 dbells[i].handle = vmci_handle_arr_get_entry(
748 context->doorbell_array, i);
749
750 *buf_size = data_size;
751 *pbuf = dbells;
752 } else {
753 *buf_size = 0;
754 *pbuf = NULL;
755 }
756
757 return VMCI_SUCCESS;
758}
759
760/*
761 * Get current context's checkpoint state of given type.
762 */
763int vmci_ctx_get_chkpt_state(u32 context_id,
764 u32 cpt_type,
765 u32 *buf_size,
766 void **pbuf)
767{
768 struct vmci_ctx *context;
769 int result;
770
771 context = vmci_ctx_get(context_id);
772 if (!context)
773 return VMCI_ERROR_NOT_FOUND;
774
775 spin_lock(&context->lock);
776
777 switch (cpt_type) {
778 case VMCI_NOTIFICATION_CPT_STATE:
779 result = vmci_ctx_get_chkpt_notifiers(context, buf_size, pbuf);
780 break;
781
782 case VMCI_WELLKNOWN_CPT_STATE:
783 /*
784 * For compatibility with VMX'en with VM to VM communication, we
785 * always return zero wellknown handles.
786 */
787
788 *buf_size = 0;
789 *pbuf = NULL;
790 result = VMCI_SUCCESS;
791 break;
792
793 case VMCI_DOORBELL_CPT_STATE:
794 result = vmci_ctx_get_chkpt_doorbells(context, buf_size, pbuf);
795 break;
796
797 default:
798 pr_devel("Invalid cpt state (type=%d)\n", cpt_type);
799 result = VMCI_ERROR_INVALID_ARGS;
800 break;
801 }
802
803 spin_unlock(&context->lock);
804 vmci_ctx_put(context);
805
806 return result;
807}
808
809/*
810 * Set current context's checkpoint state of given type.
811 */
812int vmci_ctx_set_chkpt_state(u32 context_id,
813 u32 cpt_type,
814 u32 buf_size,
815 void *cpt_buf)
816{
817 u32 i;
818 u32 current_id;
819 int result = VMCI_SUCCESS;
820 u32 num_ids = buf_size / sizeof(u32);
821
822 if (cpt_type == VMCI_WELLKNOWN_CPT_STATE && num_ids > 0) {
823 /*
824 * We would end up here if VMX with VM to VM communication
825 * attempts to restore a checkpoint with wellknown handles.
826 */
827 pr_warn("Attempt to restore checkpoint with obsolete wellknown handles\n");
828 return VMCI_ERROR_OBSOLETE;
829 }
830
831 if (cpt_type != VMCI_NOTIFICATION_CPT_STATE) {
832 pr_devel("Invalid cpt state (type=%d)\n", cpt_type);
833 return VMCI_ERROR_INVALID_ARGS;
834 }
835
836 for (i = 0; i < num_ids && result == VMCI_SUCCESS; i++) {
837 current_id = ((u32 *)cpt_buf)[i];
838 result = vmci_ctx_add_notification(context_id, current_id);
839 if (result != VMCI_SUCCESS)
840 break;
841 }
842 if (result != VMCI_SUCCESS)
843 pr_devel("Failed to set cpt state (type=%d) (error=%d)\n",
844 cpt_type, result);
845
846 return result;
847}
848
849/*
850 * Retrieves the specified context's pending notifications in the
851 * form of a handle array. The handle arrays returned are the
852 * actual data - not a copy and should not be modified by the
853 * caller. They must be released using
854 * vmci_ctx_rcv_notifications_release.
855 */
856int vmci_ctx_rcv_notifications_get(u32 context_id,
857 struct vmci_handle_arr **db_handle_array,
858 struct vmci_handle_arr **qp_handle_array)
859{
860 struct vmci_ctx *context;
861 int result = VMCI_SUCCESS;
862
863 context = vmci_ctx_get(context_id);
864 if (context == NULL)
865 return VMCI_ERROR_NOT_FOUND;
866
867 spin_lock(&context->lock);
868
869 *db_handle_array = context->pending_doorbell_array;
870 context->pending_doorbell_array = vmci_handle_arr_create(0);
871 if (!context->pending_doorbell_array) {
872 context->pending_doorbell_array = *db_handle_array;
873 *db_handle_array = NULL;
874 result = VMCI_ERROR_NO_MEM;
875 }
876 *qp_handle_array = NULL;
877
878 spin_unlock(&context->lock);
879 vmci_ctx_put(context);
880
881 return result;
882}
883
884/*
885 * Releases handle arrays with pending notifications previously
886 * retrieved using vmci_ctx_rcv_notifications_get. If the
887 * notifications were not successfully handed over to the guest,
888 * success must be false.
889 */
890void vmci_ctx_rcv_notifications_release(u32 context_id,
891 struct vmci_handle_arr *db_handle_array,
892 struct vmci_handle_arr *qp_handle_array,
893 bool success)
894{
895 struct vmci_ctx *context = vmci_ctx_get(context_id);
896
897 spin_lock(&context->lock);
898 if (!success) {
899 struct vmci_handle handle;
900
901 /*
902 * New notifications may have been added while we were not
903 * holding the context lock, so we transfer any new pending
904 * doorbell notifications to the old array, and reinstate the
905 * old array.
906 */
907
908 handle = vmci_handle_arr_remove_tail(
909 context->pending_doorbell_array);
910 while (!vmci_handle_is_invalid(handle)) {
911 if (!vmci_handle_arr_has_entry(db_handle_array,
912 handle)) {
913 vmci_handle_arr_append_entry(
914 &db_handle_array, handle);
915 }
916 handle = vmci_handle_arr_remove_tail(
917 context->pending_doorbell_array);
918 }
919 vmci_handle_arr_destroy(context->pending_doorbell_array);
920 context->pending_doorbell_array = db_handle_array;
921 db_handle_array = NULL;
922 } else {
923 ctx_clear_notify_call(context);
924 }
925 spin_unlock(&context->lock);
926 vmci_ctx_put(context);
927
928 if (db_handle_array)
929 vmci_handle_arr_destroy(db_handle_array);
930
931 if (qp_handle_array)
932 vmci_handle_arr_destroy(qp_handle_array);
933}
934
935/*
936 * Registers that a new doorbell handle has been allocated by the
937 * context. Only doorbell handles registered can be notified.
938 */
939int vmci_ctx_dbell_create(u32 context_id, struct vmci_handle handle)
940{
941 struct vmci_ctx *context;
942 int result;
943
944 if (context_id == VMCI_INVALID_ID || vmci_handle_is_invalid(handle))
945 return VMCI_ERROR_INVALID_ARGS;
946
947 context = vmci_ctx_get(context_id);
948 if (context == NULL)
949 return VMCI_ERROR_NOT_FOUND;
950
951 spin_lock(&context->lock);
952 if (!vmci_handle_arr_has_entry(context->doorbell_array, handle)) {
953 vmci_handle_arr_append_entry(&context->doorbell_array, handle);
954 result = VMCI_SUCCESS;
955 } else {
956 result = VMCI_ERROR_DUPLICATE_ENTRY;
957 }
958
959 spin_unlock(&context->lock);
960 vmci_ctx_put(context);
961
962 return result;
963}
964
965/*
966 * Unregisters a doorbell handle that was previously registered
967 * with vmci_ctx_dbell_create.
968 */
969int vmci_ctx_dbell_destroy(u32 context_id, struct vmci_handle handle)
970{
971 struct vmci_ctx *context;
972 struct vmci_handle removed_handle;
973
974 if (context_id == VMCI_INVALID_ID || vmci_handle_is_invalid(handle))
975 return VMCI_ERROR_INVALID_ARGS;
976
977 context = vmci_ctx_get(context_id);
978 if (context == NULL)
979 return VMCI_ERROR_NOT_FOUND;
980
981 spin_lock(&context->lock);
982 removed_handle =
983 vmci_handle_arr_remove_entry(context->doorbell_array, handle);
984 vmci_handle_arr_remove_entry(context->pending_doorbell_array, handle);
985 spin_unlock(&context->lock);
986
987 vmci_ctx_put(context);
988
989 return vmci_handle_is_invalid(removed_handle) ?
990 VMCI_ERROR_NOT_FOUND : VMCI_SUCCESS;
991}
992
993/*
994 * Unregisters all doorbell handles that were previously
995 * registered with vmci_ctx_dbell_create.
996 */
997int vmci_ctx_dbell_destroy_all(u32 context_id)
998{
999 struct vmci_ctx *context;
1000 struct vmci_handle handle;
1001
1002 if (context_id == VMCI_INVALID_ID)
1003 return VMCI_ERROR_INVALID_ARGS;
1004
1005 context = vmci_ctx_get(context_id);
1006 if (context == NULL)
1007 return VMCI_ERROR_NOT_FOUND;
1008
1009 spin_lock(&context->lock);
1010 do {
1011 struct vmci_handle_arr *arr = context->doorbell_array;
1012 handle = vmci_handle_arr_remove_tail(arr);
1013 } while (!vmci_handle_is_invalid(handle));
1014 do {
1015 struct vmci_handle_arr *arr = context->pending_doorbell_array;
1016 handle = vmci_handle_arr_remove_tail(arr);
1017 } while (!vmci_handle_is_invalid(handle));
1018 spin_unlock(&context->lock);
1019
1020 vmci_ctx_put(context);
1021
1022 return VMCI_SUCCESS;
1023}
1024
1025/*
1026 * Registers a notification of a doorbell handle initiated by the
1027 * specified source context. The notification of doorbells are
1028 * subject to the same isolation rules as datagram delivery. To
1029 * allow host side senders of notifications a finer granularity
1030 * of sender rights than those assigned to the sending context
1031 * itself, the host context is required to specify a different
1032 * set of privilege flags that will override the privileges of
1033 * the source context.
1034 */
1035int vmci_ctx_notify_dbell(u32 src_cid,
1036 struct vmci_handle handle,
1037 u32 src_priv_flags)
1038{
1039 struct vmci_ctx *dst_context;
1040 int result;
1041
1042 if (vmci_handle_is_invalid(handle))
1043 return VMCI_ERROR_INVALID_ARGS;
1044
1045 /* Get the target VM's VMCI context. */
1046 dst_context = vmci_ctx_get(handle.context);
1047 if (!dst_context) {
1048 pr_devel("Invalid context (ID=0x%x)\n", handle.context);
1049 return VMCI_ERROR_NOT_FOUND;
1050 }
1051
1052 if (src_cid != handle.context) {
1053 u32 dst_priv_flags;
1054
1055 if (VMCI_CONTEXT_IS_VM(src_cid) &&
1056 VMCI_CONTEXT_IS_VM(handle.context)) {
1057 pr_devel("Doorbell notification from VM to VM not supported (src=0x%x, dst=0x%x)\n",
1058 src_cid, handle.context);
1059 result = VMCI_ERROR_DST_UNREACHABLE;
1060 goto out;
1061 }
1062
1063 result = vmci_dbell_get_priv_flags(handle, &dst_priv_flags);
1064 if (result < VMCI_SUCCESS) {
1065 pr_warn("Failed to get privilege flags for destination (handle=0x%x:0x%x)\n",
1066 handle.context, handle.resource);
1067 goto out;
1068 }
1069
1070 if (src_cid != VMCI_HOST_CONTEXT_ID ||
1071 src_priv_flags == VMCI_NO_PRIVILEGE_FLAGS) {
1072 src_priv_flags = vmci_context_get_priv_flags(src_cid);
1073 }
1074
1075 if (vmci_deny_interaction(src_priv_flags, dst_priv_flags)) {
1076 result = VMCI_ERROR_NO_ACCESS;
1077 goto out;
1078 }
1079 }
1080
1081 if (handle.context == VMCI_HOST_CONTEXT_ID) {
1082 result = vmci_dbell_host_context_notify(src_cid, handle);
1083 } else {
1084 spin_lock(&dst_context->lock);
1085
1086 if (!vmci_handle_arr_has_entry(dst_context->doorbell_array,
1087 handle)) {
1088 result = VMCI_ERROR_NOT_FOUND;
1089 } else {
1090 if (!vmci_handle_arr_has_entry(
1091 dst_context->pending_doorbell_array,
1092 handle)) {
1093 vmci_handle_arr_append_entry(
1094 &dst_context->pending_doorbell_array,
1095 handle);
1096
1097 ctx_signal_notify(dst_context);
1098 wake_up(&dst_context->host_context.wait_queue);
1099
1100 }
1101 result = VMCI_SUCCESS;
1102 }
1103 spin_unlock(&dst_context->lock);
1104 }
1105
1106 out:
1107 vmci_ctx_put(dst_context);
1108
1109 return result;
1110}
1111
1112bool vmci_ctx_supports_host_qp(struct vmci_ctx *context)
1113{
1114 return context && context->user_version >= VMCI_VERSION_HOSTQP;
1115}
1116
1117/*
1118 * Registers that a new queue pair handle has been allocated by
1119 * the context.
1120 */
1121int vmci_ctx_qp_create(struct vmci_ctx *context, struct vmci_handle handle)
1122{
1123 int result;
1124
1125 if (context == NULL || vmci_handle_is_invalid(handle))
1126 return VMCI_ERROR_INVALID_ARGS;
1127
1128 if (!vmci_handle_arr_has_entry(context->queue_pair_array, handle)) {
1129 vmci_handle_arr_append_entry(&context->queue_pair_array,
1130 handle);
1131 result = VMCI_SUCCESS;
1132 } else {
1133 result = VMCI_ERROR_DUPLICATE_ENTRY;
1134 }
1135
1136 return result;
1137}
1138
1139/*
1140 * Unregisters a queue pair handle that was previously registered
1141 * with vmci_ctx_qp_create.
1142 */
1143int vmci_ctx_qp_destroy(struct vmci_ctx *context, struct vmci_handle handle)
1144{
1145 struct vmci_handle hndl;
1146
1147 if (context == NULL || vmci_handle_is_invalid(handle))
1148 return VMCI_ERROR_INVALID_ARGS;
1149
1150 hndl = vmci_handle_arr_remove_entry(context->queue_pair_array, handle);
1151
1152 return vmci_handle_is_invalid(hndl) ?
1153 VMCI_ERROR_NOT_FOUND : VMCI_SUCCESS;
1154}
1155
1156/*
1157 * Determines whether a given queue pair handle is registered
1158 * with the given context.
1159 */
1160bool vmci_ctx_qp_exists(struct vmci_ctx *context, struct vmci_handle handle)
1161{
1162 if (context == NULL || vmci_handle_is_invalid(handle))
1163 return false;
1164
1165 return vmci_handle_arr_has_entry(context->queue_pair_array, handle);
1166}
1167
1168/*
1169 * vmci_context_get_priv_flags() - Retrieve privilege flags.
1170 * @context_id: The context ID of the VMCI context.
1171 *
1172 * Retrieves privilege flags of the given VMCI context ID.
1173 */
1174u32 vmci_context_get_priv_flags(u32 context_id)
1175{
1176 if (vmci_host_code_active()) {
1177 u32 flags;
1178 struct vmci_ctx *context;
1179
1180 context = vmci_ctx_get(context_id);
1181 if (!context)
1182 return VMCI_LEAST_PRIVILEGE_FLAGS;
1183
1184 flags = context->priv_flags;
1185 vmci_ctx_put(context);
1186 return flags;
1187 }
1188 return VMCI_NO_PRIVILEGE_FLAGS;
1189}
1190EXPORT_SYMBOL_GPL(vmci_context_get_priv_flags);
1191
1192/*
1193 * vmci_is_context_owner() - Determimnes if user is the context owner
1194 * @context_id: The context ID of the VMCI context.
1195 * @uid: The host user id (real kernel value).
1196 *
1197 * Determines whether a given UID is the owner of given VMCI context.
1198 */
1199bool vmci_is_context_owner(u32 context_id, kuid_t uid)
1200{
1201 bool is_owner = false;
1202
1203 if (vmci_host_code_active()) {
1204 struct vmci_ctx *context = vmci_ctx_get(context_id);
1205 if (context) {
1206 if (context->cred)
1207 is_owner = uid_eq(context->cred->uid, uid);
1208 vmci_ctx_put(context);
1209 }
1210 }
1211
1212 return is_owner;
1213}
1214EXPORT_SYMBOL_GPL(vmci_is_context_owner);