Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * VMware VMCI Driver
0004  *
0005  * Copyright (C) 2012 VMware, Inc. All rights reserved.
0006  */
0007 
0008 #include <linux/vmw_vmci_defs.h>
0009 #include <linux/vmw_vmci_api.h>
0010 #include <linux/highmem.h>
0011 #include <linux/kernel.h>
0012 #include <linux/mm.h>
0013 #include <linux/module.h>
0014 #include <linux/mutex.h>
0015 #include <linux/pagemap.h>
0016 #include <linux/pci.h>
0017 #include <linux/sched.h>
0018 #include <linux/slab.h>
0019 #include <linux/uio.h>
0020 #include <linux/wait.h>
0021 #include <linux/vmalloc.h>
0022 #include <linux/skbuff.h>
0023 
0024 #include "vmci_handle_array.h"
0025 #include "vmci_queue_pair.h"
0026 #include "vmci_datagram.h"
0027 #include "vmci_resource.h"
0028 #include "vmci_context.h"
0029 #include "vmci_driver.h"
0030 #include "vmci_event.h"
0031 #include "vmci_route.h"
0032 
0033 /*
0034  * In the following, we will distinguish between two kinds of VMX processes -
0035  * the ones with versions lower than VMCI_VERSION_NOVMVM that use specialized
0036  * VMCI page files in the VMX and supporting VM to VM communication and the
0037  * newer ones that use the guest memory directly. We will in the following
0038  * refer to the older VMX versions as old-style VMX'en, and the newer ones as
0039  * new-style VMX'en.
0040  *
0041  * The state transition datagram is as follows (the VMCIQPB_ prefix has been
0042  * removed for readability) - see below for more details on the transtions:
0043  *
0044  *            --------------  NEW  -------------
0045  *            |                                |
0046  *           \_/                              \_/
0047  *     CREATED_NO_MEM <-----------------> CREATED_MEM
0048  *            |    |                           |
0049  *            |    o-----------------------o   |
0050  *            |                            |   |
0051  *           \_/                          \_/ \_/
0052  *     ATTACHED_NO_MEM <----------------> ATTACHED_MEM
0053  *            |                            |   |
0054  *            |     o----------------------o   |
0055  *            |     |                          |
0056  *           \_/   \_/                        \_/
0057  *     SHUTDOWN_NO_MEM <----------------> SHUTDOWN_MEM
0058  *            |                                |
0059  *            |                                |
0060  *            -------------> gone <-------------
0061  *
0062  * In more detail. When a VMCI queue pair is first created, it will be in the
0063  * VMCIQPB_NEW state. It will then move into one of the following states:
0064  *
0065  * - VMCIQPB_CREATED_NO_MEM: this state indicates that either:
0066  *
0067  *     - the created was performed by a host endpoint, in which case there is
0068  *       no backing memory yet.
0069  *
0070  *     - the create was initiated by an old-style VMX, that uses
0071  *       vmci_qp_broker_set_page_store to specify the UVAs of the queue pair at
0072  *       a later point in time. This state can be distinguished from the one
0073  *       above by the context ID of the creator. A host side is not allowed to
0074  *       attach until the page store has been set.
0075  *
0076  * - VMCIQPB_CREATED_MEM: this state is the result when the queue pair
0077  *     is created by a VMX using the queue pair device backend that
0078  *     sets the UVAs of the queue pair immediately and stores the
0079  *     information for later attachers. At this point, it is ready for
0080  *     the host side to attach to it.
0081  *
0082  * Once the queue pair is in one of the created states (with the exception of
0083  * the case mentioned for older VMX'en above), it is possible to attach to the
0084  * queue pair. Again we have two new states possible:
0085  *
0086  * - VMCIQPB_ATTACHED_MEM: this state can be reached through the following
0087  *   paths:
0088  *
0089  *     - from VMCIQPB_CREATED_NO_MEM when a new-style VMX allocates a queue
0090  *       pair, and attaches to a queue pair previously created by the host side.
0091  *
0092  *     - from VMCIQPB_CREATED_MEM when the host side attaches to a queue pair
0093  *       already created by a guest.
0094  *
0095  *     - from VMCIQPB_ATTACHED_NO_MEM, when an old-style VMX calls
0096  *       vmci_qp_broker_set_page_store (see below).
0097  *
0098  * - VMCIQPB_ATTACHED_NO_MEM: If the queue pair already was in the
0099  *     VMCIQPB_CREATED_NO_MEM due to a host side create, an old-style VMX will
0100  *     bring the queue pair into this state. Once vmci_qp_broker_set_page_store
0101  *     is called to register the user memory, the VMCIQPB_ATTACH_MEM state
0102  *     will be entered.
0103  *
0104  * From the attached queue pair, the queue pair can enter the shutdown states
0105  * when either side of the queue pair detaches. If the guest side detaches
0106  * first, the queue pair will enter the VMCIQPB_SHUTDOWN_NO_MEM state, where
0107  * the content of the queue pair will no longer be available. If the host
0108  * side detaches first, the queue pair will either enter the
0109  * VMCIQPB_SHUTDOWN_MEM, if the guest memory is currently mapped, or
0110  * VMCIQPB_SHUTDOWN_NO_MEM, if the guest memory is not mapped
0111  * (e.g., the host detaches while a guest is stunned).
0112  *
0113  * New-style VMX'en will also unmap guest memory, if the guest is
0114  * quiesced, e.g., during a snapshot operation. In that case, the guest
0115  * memory will no longer be available, and the queue pair will transition from
0116  * *_MEM state to a *_NO_MEM state. The VMX may later map the memory once more,
0117  * in which case the queue pair will transition from the *_NO_MEM state at that
0118  * point back to the *_MEM state. Note that the *_NO_MEM state may have changed,
0119  * since the peer may have either attached or detached in the meantime. The
0120  * values are laid out such that ++ on a state will move from a *_NO_MEM to a
0121  * *_MEM state, and vice versa.
0122  */
0123 
0124 /* The Kernel specific component of the struct vmci_queue structure. */
0125 struct vmci_queue_kern_if {
0126     struct mutex __mutex;   /* Protects the queue. */
0127     struct mutex *mutex;    /* Shared by producer and consumer queues. */
0128     size_t num_pages;   /* Number of pages incl. header. */
0129     bool host;      /* Host or guest? */
0130     union {
0131         struct {
0132             dma_addr_t *pas;
0133             void **vas;
0134         } g;        /* Used by the guest. */
0135         struct {
0136             struct page **page;
0137             struct page **header_page;
0138         } h;        /* Used by the host. */
0139     } u;
0140 };
0141 
0142 /*
0143  * This structure is opaque to the clients.
0144  */
0145 struct vmci_qp {
0146     struct vmci_handle handle;
0147     struct vmci_queue *produce_q;
0148     struct vmci_queue *consume_q;
0149     u64 produce_q_size;
0150     u64 consume_q_size;
0151     u32 peer;
0152     u32 flags;
0153     u32 priv_flags;
0154     bool guest_endpoint;
0155     unsigned int blocked;
0156     unsigned int generation;
0157     wait_queue_head_t event;
0158 };
0159 
0160 enum qp_broker_state {
0161     VMCIQPB_NEW,
0162     VMCIQPB_CREATED_NO_MEM,
0163     VMCIQPB_CREATED_MEM,
0164     VMCIQPB_ATTACHED_NO_MEM,
0165     VMCIQPB_ATTACHED_MEM,
0166     VMCIQPB_SHUTDOWN_NO_MEM,
0167     VMCIQPB_SHUTDOWN_MEM,
0168     VMCIQPB_GONE
0169 };
0170 
0171 #define QPBROKERSTATE_HAS_MEM(_qpb) (_qpb->state == VMCIQPB_CREATED_MEM || \
0172                      _qpb->state == VMCIQPB_ATTACHED_MEM || \
0173                      _qpb->state == VMCIQPB_SHUTDOWN_MEM)
0174 
0175 /*
0176  * In the queue pair broker, we always use the guest point of view for
0177  * the produce and consume queue values and references, e.g., the
0178  * produce queue size stored is the guests produce queue size. The
0179  * host endpoint will need to swap these around. The only exception is
0180  * the local queue pairs on the host, in which case the host endpoint
0181  * that creates the queue pair will have the right orientation, and
0182  * the attaching host endpoint will need to swap.
0183  */
0184 struct qp_entry {
0185     struct list_head list_item;
0186     struct vmci_handle handle;
0187     u32 peer;
0188     u32 flags;
0189     u64 produce_size;
0190     u64 consume_size;
0191     u32 ref_count;
0192 };
0193 
0194 struct qp_broker_entry {
0195     struct vmci_resource resource;
0196     struct qp_entry qp;
0197     u32 create_id;
0198     u32 attach_id;
0199     enum qp_broker_state state;
0200     bool require_trusted_attach;
0201     bool created_by_trusted;
0202     bool vmci_page_files;   /* Created by VMX using VMCI page files */
0203     struct vmci_queue *produce_q;
0204     struct vmci_queue *consume_q;
0205     struct vmci_queue_header saved_produce_q;
0206     struct vmci_queue_header saved_consume_q;
0207     vmci_event_release_cb wakeup_cb;
0208     void *client_data;
0209     void *local_mem;    /* Kernel memory for local queue pair */
0210 };
0211 
0212 struct qp_guest_endpoint {
0213     struct vmci_resource resource;
0214     struct qp_entry qp;
0215     u64 num_ppns;
0216     void *produce_q;
0217     void *consume_q;
0218     struct ppn_set ppn_set;
0219 };
0220 
0221 struct qp_list {
0222     struct list_head head;
0223     struct mutex mutex; /* Protect queue list. */
0224 };
0225 
0226 static struct qp_list qp_broker_list = {
0227     .head = LIST_HEAD_INIT(qp_broker_list.head),
0228     .mutex = __MUTEX_INITIALIZER(qp_broker_list.mutex),
0229 };
0230 
0231 static struct qp_list qp_guest_endpoints = {
0232     .head = LIST_HEAD_INIT(qp_guest_endpoints.head),
0233     .mutex = __MUTEX_INITIALIZER(qp_guest_endpoints.mutex),
0234 };
0235 
0236 #define INVALID_VMCI_GUEST_MEM_ID  0
0237 #define QPE_NUM_PAGES(_QPE) ((u32) \
0238                  (DIV_ROUND_UP(_QPE.produce_size, PAGE_SIZE) + \
0239                   DIV_ROUND_UP(_QPE.consume_size, PAGE_SIZE) + 2))
0240 #define QP_SIZES_ARE_VALID(_prod_qsize, _cons_qsize) \
0241     ((_prod_qsize) + (_cons_qsize) >= max(_prod_qsize, _cons_qsize) && \
0242      (_prod_qsize) + (_cons_qsize) <= VMCI_MAX_GUEST_QP_MEMORY)
0243 
0244 /*
0245  * Frees kernel VA space for a given queue and its queue header, and
0246  * frees physical data pages.
0247  */
0248 static void qp_free_queue(void *q, u64 size)
0249 {
0250     struct vmci_queue *queue = q;
0251 
0252     if (queue) {
0253         u64 i;
0254 
0255         /* Given size does not include header, so add in a page here. */
0256         for (i = 0; i < DIV_ROUND_UP(size, PAGE_SIZE) + 1; i++) {
0257             dma_free_coherent(&vmci_pdev->dev, PAGE_SIZE,
0258                       queue->kernel_if->u.g.vas[i],
0259                       queue->kernel_if->u.g.pas[i]);
0260         }
0261 
0262         vfree(queue);
0263     }
0264 }
0265 
0266 /*
0267  * Allocates kernel queue pages of specified size with IOMMU mappings,
0268  * plus space for the queue structure/kernel interface and the queue
0269  * header.
0270  */
0271 static void *qp_alloc_queue(u64 size, u32 flags)
0272 {
0273     u64 i;
0274     struct vmci_queue *queue;
0275     size_t pas_size;
0276     size_t vas_size;
0277     size_t queue_size = sizeof(*queue) + sizeof(*queue->kernel_if);
0278     u64 num_pages;
0279 
0280     if (size > SIZE_MAX - PAGE_SIZE)
0281         return NULL;
0282     num_pages = DIV_ROUND_UP(size, PAGE_SIZE) + 1;
0283     if (num_pages >
0284          (SIZE_MAX - queue_size) /
0285          (sizeof(*queue->kernel_if->u.g.pas) +
0286           sizeof(*queue->kernel_if->u.g.vas)))
0287         return NULL;
0288 
0289     pas_size = num_pages * sizeof(*queue->kernel_if->u.g.pas);
0290     vas_size = num_pages * sizeof(*queue->kernel_if->u.g.vas);
0291     queue_size += pas_size + vas_size;
0292 
0293     queue = vmalloc(queue_size);
0294     if (!queue)
0295         return NULL;
0296 
0297     queue->q_header = NULL;
0298     queue->saved_header = NULL;
0299     queue->kernel_if = (struct vmci_queue_kern_if *)(queue + 1);
0300     queue->kernel_if->mutex = NULL;
0301     queue->kernel_if->num_pages = num_pages;
0302     queue->kernel_if->u.g.pas = (dma_addr_t *)(queue->kernel_if + 1);
0303     queue->kernel_if->u.g.vas =
0304         (void **)((u8 *)queue->kernel_if->u.g.pas + pas_size);
0305     queue->kernel_if->host = false;
0306 
0307     for (i = 0; i < num_pages; i++) {
0308         queue->kernel_if->u.g.vas[i] =
0309             dma_alloc_coherent(&vmci_pdev->dev, PAGE_SIZE,
0310                        &queue->kernel_if->u.g.pas[i],
0311                        GFP_KERNEL);
0312         if (!queue->kernel_if->u.g.vas[i]) {
0313             /* Size excl. the header. */
0314             qp_free_queue(queue, i * PAGE_SIZE);
0315             return NULL;
0316         }
0317     }
0318 
0319     /* Queue header is the first page. */
0320     queue->q_header = queue->kernel_if->u.g.vas[0];
0321 
0322     return queue;
0323 }
0324 
0325 /*
0326  * Copies from a given buffer or iovector to a VMCI Queue.  Uses
0327  * kmap()/kunmap() to dynamically map/unmap required portions of the queue
0328  * by traversing the offset -> page translation structure for the queue.
0329  * Assumes that offset + size does not wrap around in the queue.
0330  */
0331 static int qp_memcpy_to_queue_iter(struct vmci_queue *queue,
0332                   u64 queue_offset,
0333                   struct iov_iter *from,
0334                   size_t size)
0335 {
0336     struct vmci_queue_kern_if *kernel_if = queue->kernel_if;
0337     size_t bytes_copied = 0;
0338 
0339     while (bytes_copied < size) {
0340         const u64 page_index =
0341             (queue_offset + bytes_copied) / PAGE_SIZE;
0342         const size_t page_offset =
0343             (queue_offset + bytes_copied) & (PAGE_SIZE - 1);
0344         void *va;
0345         size_t to_copy;
0346 
0347         if (kernel_if->host)
0348             va = kmap(kernel_if->u.h.page[page_index]);
0349         else
0350             va = kernel_if->u.g.vas[page_index + 1];
0351             /* Skip header. */
0352 
0353         if (size - bytes_copied > PAGE_SIZE - page_offset)
0354             /* Enough payload to fill up from this page. */
0355             to_copy = PAGE_SIZE - page_offset;
0356         else
0357             to_copy = size - bytes_copied;
0358 
0359         if (!copy_from_iter_full((u8 *)va + page_offset, to_copy,
0360                      from)) {
0361             if (kernel_if->host)
0362                 kunmap(kernel_if->u.h.page[page_index]);
0363             return VMCI_ERROR_INVALID_ARGS;
0364         }
0365         bytes_copied += to_copy;
0366         if (kernel_if->host)
0367             kunmap(kernel_if->u.h.page[page_index]);
0368     }
0369 
0370     return VMCI_SUCCESS;
0371 }
0372 
0373 /*
0374  * Copies to a given buffer or iovector from a VMCI Queue.  Uses
0375  * kmap()/kunmap() to dynamically map/unmap required portions of the queue
0376  * by traversing the offset -> page translation structure for the queue.
0377  * Assumes that offset + size does not wrap around in the queue.
0378  */
0379 static int qp_memcpy_from_queue_iter(struct iov_iter *to,
0380                     const struct vmci_queue *queue,
0381                     u64 queue_offset, size_t size)
0382 {
0383     struct vmci_queue_kern_if *kernel_if = queue->kernel_if;
0384     size_t bytes_copied = 0;
0385 
0386     while (bytes_copied < size) {
0387         const u64 page_index =
0388             (queue_offset + bytes_copied) / PAGE_SIZE;
0389         const size_t page_offset =
0390             (queue_offset + bytes_copied) & (PAGE_SIZE - 1);
0391         void *va;
0392         size_t to_copy;
0393         int err;
0394 
0395         if (kernel_if->host)
0396             va = kmap(kernel_if->u.h.page[page_index]);
0397         else
0398             va = kernel_if->u.g.vas[page_index + 1];
0399             /* Skip header. */
0400 
0401         if (size - bytes_copied > PAGE_SIZE - page_offset)
0402             /* Enough payload to fill up this page. */
0403             to_copy = PAGE_SIZE - page_offset;
0404         else
0405             to_copy = size - bytes_copied;
0406 
0407         err = copy_to_iter((u8 *)va + page_offset, to_copy, to);
0408         if (err != to_copy) {
0409             if (kernel_if->host)
0410                 kunmap(kernel_if->u.h.page[page_index]);
0411             return VMCI_ERROR_INVALID_ARGS;
0412         }
0413         bytes_copied += to_copy;
0414         if (kernel_if->host)
0415             kunmap(kernel_if->u.h.page[page_index]);
0416     }
0417 
0418     return VMCI_SUCCESS;
0419 }
0420 
0421 /*
0422  * Allocates two list of PPNs --- one for the pages in the produce queue,
0423  * and the other for the pages in the consume queue. Intializes the list
0424  * of PPNs with the page frame numbers of the KVA for the two queues (and
0425  * the queue headers).
0426  */
0427 static int qp_alloc_ppn_set(void *prod_q,
0428                 u64 num_produce_pages,
0429                 void *cons_q,
0430                 u64 num_consume_pages, struct ppn_set *ppn_set)
0431 {
0432     u64 *produce_ppns;
0433     u64 *consume_ppns;
0434     struct vmci_queue *produce_q = prod_q;
0435     struct vmci_queue *consume_q = cons_q;
0436     u64 i;
0437 
0438     if (!produce_q || !num_produce_pages || !consume_q ||
0439         !num_consume_pages || !ppn_set)
0440         return VMCI_ERROR_INVALID_ARGS;
0441 
0442     if (ppn_set->initialized)
0443         return VMCI_ERROR_ALREADY_EXISTS;
0444 
0445     produce_ppns =
0446         kmalloc_array(num_produce_pages, sizeof(*produce_ppns),
0447               GFP_KERNEL);
0448     if (!produce_ppns)
0449         return VMCI_ERROR_NO_MEM;
0450 
0451     consume_ppns =
0452         kmalloc_array(num_consume_pages, sizeof(*consume_ppns),
0453               GFP_KERNEL);
0454     if (!consume_ppns) {
0455         kfree(produce_ppns);
0456         return VMCI_ERROR_NO_MEM;
0457     }
0458 
0459     for (i = 0; i < num_produce_pages; i++)
0460         produce_ppns[i] =
0461             produce_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT;
0462 
0463     for (i = 0; i < num_consume_pages; i++)
0464         consume_ppns[i] =
0465             consume_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT;
0466 
0467     ppn_set->num_produce_pages = num_produce_pages;
0468     ppn_set->num_consume_pages = num_consume_pages;
0469     ppn_set->produce_ppns = produce_ppns;
0470     ppn_set->consume_ppns = consume_ppns;
0471     ppn_set->initialized = true;
0472     return VMCI_SUCCESS;
0473 }
0474 
0475 /*
0476  * Frees the two list of PPNs for a queue pair.
0477  */
0478 static void qp_free_ppn_set(struct ppn_set *ppn_set)
0479 {
0480     if (ppn_set->initialized) {
0481         /* Do not call these functions on NULL inputs. */
0482         kfree(ppn_set->produce_ppns);
0483         kfree(ppn_set->consume_ppns);
0484     }
0485     memset(ppn_set, 0, sizeof(*ppn_set));
0486 }
0487 
0488 /*
0489  * Populates the list of PPNs in the hypercall structure with the PPNS
0490  * of the produce queue and the consume queue.
0491  */
0492 static int qp_populate_ppn_set(u8 *call_buf, const struct ppn_set *ppn_set)
0493 {
0494     if (vmci_use_ppn64()) {
0495         memcpy(call_buf, ppn_set->produce_ppns,
0496                ppn_set->num_produce_pages *
0497                sizeof(*ppn_set->produce_ppns));
0498         memcpy(call_buf +
0499                ppn_set->num_produce_pages *
0500                sizeof(*ppn_set->produce_ppns),
0501                ppn_set->consume_ppns,
0502                ppn_set->num_consume_pages *
0503                sizeof(*ppn_set->consume_ppns));
0504     } else {
0505         int i;
0506         u32 *ppns = (u32 *) call_buf;
0507 
0508         for (i = 0; i < ppn_set->num_produce_pages; i++)
0509             ppns[i] = (u32) ppn_set->produce_ppns[i];
0510 
0511         ppns = &ppns[ppn_set->num_produce_pages];
0512 
0513         for (i = 0; i < ppn_set->num_consume_pages; i++)
0514             ppns[i] = (u32) ppn_set->consume_ppns[i];
0515     }
0516 
0517     return VMCI_SUCCESS;
0518 }
0519 
0520 /*
0521  * Allocates kernel VA space of specified size plus space for the queue
0522  * and kernel interface.  This is different from the guest queue allocator,
0523  * because we do not allocate our own queue header/data pages here but
0524  * share those of the guest.
0525  */
0526 static struct vmci_queue *qp_host_alloc_queue(u64 size)
0527 {
0528     struct vmci_queue *queue;
0529     size_t queue_page_size;
0530     u64 num_pages;
0531     const size_t queue_size = sizeof(*queue) + sizeof(*(queue->kernel_if));
0532 
0533     if (size > min_t(size_t, VMCI_MAX_GUEST_QP_MEMORY, SIZE_MAX - PAGE_SIZE))
0534         return NULL;
0535     num_pages = DIV_ROUND_UP(size, PAGE_SIZE) + 1;
0536     if (num_pages > (SIZE_MAX - queue_size) /
0537          sizeof(*queue->kernel_if->u.h.page))
0538         return NULL;
0539 
0540     queue_page_size = num_pages * sizeof(*queue->kernel_if->u.h.page);
0541 
0542     if (queue_size + queue_page_size > KMALLOC_MAX_SIZE)
0543         return NULL;
0544 
0545     queue = kzalloc(queue_size + queue_page_size, GFP_KERNEL);
0546     if (queue) {
0547         queue->q_header = NULL;
0548         queue->saved_header = NULL;
0549         queue->kernel_if = (struct vmci_queue_kern_if *)(queue + 1);
0550         queue->kernel_if->host = true;
0551         queue->kernel_if->mutex = NULL;
0552         queue->kernel_if->num_pages = num_pages;
0553         queue->kernel_if->u.h.header_page =
0554             (struct page **)((u8 *)queue + queue_size);
0555         queue->kernel_if->u.h.page =
0556             &queue->kernel_if->u.h.header_page[1];
0557     }
0558 
0559     return queue;
0560 }
0561 
0562 /*
0563  * Frees kernel memory for a given queue (header plus translation
0564  * structure).
0565  */
0566 static void qp_host_free_queue(struct vmci_queue *queue, u64 queue_size)
0567 {
0568     kfree(queue);
0569 }
0570 
0571 /*
0572  * Initialize the mutex for the pair of queues.  This mutex is used to
0573  * protect the q_header and the buffer from changing out from under any
0574  * users of either queue.  Of course, it's only any good if the mutexes
0575  * are actually acquired.  Queue structure must lie on non-paged memory
0576  * or we cannot guarantee access to the mutex.
0577  */
0578 static void qp_init_queue_mutex(struct vmci_queue *produce_q,
0579                 struct vmci_queue *consume_q)
0580 {
0581     /*
0582      * Only the host queue has shared state - the guest queues do not
0583      * need to synchronize access using a queue mutex.
0584      */
0585 
0586     if (produce_q->kernel_if->host) {
0587         produce_q->kernel_if->mutex = &produce_q->kernel_if->__mutex;
0588         consume_q->kernel_if->mutex = &produce_q->kernel_if->__mutex;
0589         mutex_init(produce_q->kernel_if->mutex);
0590     }
0591 }
0592 
0593 /*
0594  * Cleans up the mutex for the pair of queues.
0595  */
0596 static void qp_cleanup_queue_mutex(struct vmci_queue *produce_q,
0597                    struct vmci_queue *consume_q)
0598 {
0599     if (produce_q->kernel_if->host) {
0600         produce_q->kernel_if->mutex = NULL;
0601         consume_q->kernel_if->mutex = NULL;
0602     }
0603 }
0604 
0605 /*
0606  * Acquire the mutex for the queue.  Note that the produce_q and
0607  * the consume_q share a mutex.  So, only one of the two need to
0608  * be passed in to this routine.  Either will work just fine.
0609  */
0610 static void qp_acquire_queue_mutex(struct vmci_queue *queue)
0611 {
0612     if (queue->kernel_if->host)
0613         mutex_lock(queue->kernel_if->mutex);
0614 }
0615 
0616 /*
0617  * Release the mutex for the queue.  Note that the produce_q and
0618  * the consume_q share a mutex.  So, only one of the two need to
0619  * be passed in to this routine.  Either will work just fine.
0620  */
0621 static void qp_release_queue_mutex(struct vmci_queue *queue)
0622 {
0623     if (queue->kernel_if->host)
0624         mutex_unlock(queue->kernel_if->mutex);
0625 }
0626 
0627 /*
0628  * Helper function to release pages in the PageStoreAttachInfo
0629  * previously obtained using get_user_pages.
0630  */
0631 static void qp_release_pages(struct page **pages,
0632                  u64 num_pages, bool dirty)
0633 {
0634     int i;
0635 
0636     for (i = 0; i < num_pages; i++) {
0637         if (dirty)
0638             set_page_dirty_lock(pages[i]);
0639 
0640         put_page(pages[i]);
0641         pages[i] = NULL;
0642     }
0643 }
0644 
0645 /*
0646  * Lock the user pages referenced by the {produce,consume}Buffer
0647  * struct into memory and populate the {produce,consume}Pages
0648  * arrays in the attach structure with them.
0649  */
0650 static int qp_host_get_user_memory(u64 produce_uva,
0651                    u64 consume_uva,
0652                    struct vmci_queue *produce_q,
0653                    struct vmci_queue *consume_q)
0654 {
0655     int retval;
0656     int err = VMCI_SUCCESS;
0657 
0658     retval = get_user_pages_fast((uintptr_t) produce_uva,
0659                      produce_q->kernel_if->num_pages,
0660                      FOLL_WRITE,
0661                      produce_q->kernel_if->u.h.header_page);
0662     if (retval < (int)produce_q->kernel_if->num_pages) {
0663         pr_debug("get_user_pages_fast(produce) failed (retval=%d)",
0664             retval);
0665         if (retval > 0)
0666             qp_release_pages(produce_q->kernel_if->u.h.header_page,
0667                     retval, false);
0668         err = VMCI_ERROR_NO_MEM;
0669         goto out;
0670     }
0671 
0672     retval = get_user_pages_fast((uintptr_t) consume_uva,
0673                      consume_q->kernel_if->num_pages,
0674                      FOLL_WRITE,
0675                      consume_q->kernel_if->u.h.header_page);
0676     if (retval < (int)consume_q->kernel_if->num_pages) {
0677         pr_debug("get_user_pages_fast(consume) failed (retval=%d)",
0678             retval);
0679         if (retval > 0)
0680             qp_release_pages(consume_q->kernel_if->u.h.header_page,
0681                     retval, false);
0682         qp_release_pages(produce_q->kernel_if->u.h.header_page,
0683                  produce_q->kernel_if->num_pages, false);
0684         err = VMCI_ERROR_NO_MEM;
0685     }
0686 
0687  out:
0688     return err;
0689 }
0690 
0691 /*
0692  * Registers the specification of the user pages used for backing a queue
0693  * pair. Enough information to map in pages is stored in the OS specific
0694  * part of the struct vmci_queue structure.
0695  */
0696 static int qp_host_register_user_memory(struct vmci_qp_page_store *page_store,
0697                     struct vmci_queue *produce_q,
0698                     struct vmci_queue *consume_q)
0699 {
0700     u64 produce_uva;
0701     u64 consume_uva;
0702 
0703     /*
0704      * The new style and the old style mapping only differs in
0705      * that we either get a single or two UVAs, so we split the
0706      * single UVA range at the appropriate spot.
0707      */
0708     produce_uva = page_store->pages;
0709     consume_uva = page_store->pages +
0710         produce_q->kernel_if->num_pages * PAGE_SIZE;
0711     return qp_host_get_user_memory(produce_uva, consume_uva, produce_q,
0712                        consume_q);
0713 }
0714 
0715 /*
0716  * Releases and removes the references to user pages stored in the attach
0717  * struct.  Pages are released from the page cache and may become
0718  * swappable again.
0719  */
0720 static void qp_host_unregister_user_memory(struct vmci_queue *produce_q,
0721                        struct vmci_queue *consume_q)
0722 {
0723     qp_release_pages(produce_q->kernel_if->u.h.header_page,
0724              produce_q->kernel_if->num_pages, true);
0725     memset(produce_q->kernel_if->u.h.header_page, 0,
0726            sizeof(*produce_q->kernel_if->u.h.header_page) *
0727            produce_q->kernel_if->num_pages);
0728     qp_release_pages(consume_q->kernel_if->u.h.header_page,
0729              consume_q->kernel_if->num_pages, true);
0730     memset(consume_q->kernel_if->u.h.header_page, 0,
0731            sizeof(*consume_q->kernel_if->u.h.header_page) *
0732            consume_q->kernel_if->num_pages);
0733 }
0734 
0735 /*
0736  * Once qp_host_register_user_memory has been performed on a
0737  * queue, the queue pair headers can be mapped into the
0738  * kernel. Once mapped, they must be unmapped with
0739  * qp_host_unmap_queues prior to calling
0740  * qp_host_unregister_user_memory.
0741  * Pages are pinned.
0742  */
0743 static int qp_host_map_queues(struct vmci_queue *produce_q,
0744                   struct vmci_queue *consume_q)
0745 {
0746     int result;
0747 
0748     if (!produce_q->q_header || !consume_q->q_header) {
0749         struct page *headers[2];
0750 
0751         if (produce_q->q_header != consume_q->q_header)
0752             return VMCI_ERROR_QUEUEPAIR_MISMATCH;
0753 
0754         if (produce_q->kernel_if->u.h.header_page == NULL ||
0755             *produce_q->kernel_if->u.h.header_page == NULL)
0756             return VMCI_ERROR_UNAVAILABLE;
0757 
0758         headers[0] = *produce_q->kernel_if->u.h.header_page;
0759         headers[1] = *consume_q->kernel_if->u.h.header_page;
0760 
0761         produce_q->q_header = vmap(headers, 2, VM_MAP, PAGE_KERNEL);
0762         if (produce_q->q_header != NULL) {
0763             consume_q->q_header =
0764                 (struct vmci_queue_header *)((u8 *)
0765                              produce_q->q_header +
0766                              PAGE_SIZE);
0767             result = VMCI_SUCCESS;
0768         } else {
0769             pr_warn("vmap failed\n");
0770             result = VMCI_ERROR_NO_MEM;
0771         }
0772     } else {
0773         result = VMCI_SUCCESS;
0774     }
0775 
0776     return result;
0777 }
0778 
0779 /*
0780  * Unmaps previously mapped queue pair headers from the kernel.
0781  * Pages are unpinned.
0782  */
0783 static int qp_host_unmap_queues(u32 gid,
0784                 struct vmci_queue *produce_q,
0785                 struct vmci_queue *consume_q)
0786 {
0787     if (produce_q->q_header) {
0788         if (produce_q->q_header < consume_q->q_header)
0789             vunmap(produce_q->q_header);
0790         else
0791             vunmap(consume_q->q_header);
0792 
0793         produce_q->q_header = NULL;
0794         consume_q->q_header = NULL;
0795     }
0796 
0797     return VMCI_SUCCESS;
0798 }
0799 
0800 /*
0801  * Finds the entry in the list corresponding to a given handle. Assumes
0802  * that the list is locked.
0803  */
0804 static struct qp_entry *qp_list_find(struct qp_list *qp_list,
0805                      struct vmci_handle handle)
0806 {
0807     struct qp_entry *entry;
0808 
0809     if (vmci_handle_is_invalid(handle))
0810         return NULL;
0811 
0812     list_for_each_entry(entry, &qp_list->head, list_item) {
0813         if (vmci_handle_is_equal(entry->handle, handle))
0814             return entry;
0815     }
0816 
0817     return NULL;
0818 }
0819 
0820 /*
0821  * Finds the entry in the list corresponding to a given handle.
0822  */
0823 static struct qp_guest_endpoint *
0824 qp_guest_handle_to_entry(struct vmci_handle handle)
0825 {
0826     struct qp_guest_endpoint *entry;
0827     struct qp_entry *qp = qp_list_find(&qp_guest_endpoints, handle);
0828 
0829     entry = qp ? container_of(
0830         qp, struct qp_guest_endpoint, qp) : NULL;
0831     return entry;
0832 }
0833 
0834 /*
0835  * Finds the entry in the list corresponding to a given handle.
0836  */
0837 static struct qp_broker_entry *
0838 qp_broker_handle_to_entry(struct vmci_handle handle)
0839 {
0840     struct qp_broker_entry *entry;
0841     struct qp_entry *qp = qp_list_find(&qp_broker_list, handle);
0842 
0843     entry = qp ? container_of(
0844         qp, struct qp_broker_entry, qp) : NULL;
0845     return entry;
0846 }
0847 
0848 /*
0849  * Dispatches a queue pair event message directly into the local event
0850  * queue.
0851  */
0852 static int qp_notify_peer_local(bool attach, struct vmci_handle handle)
0853 {
0854     u32 context_id = vmci_get_context_id();
0855     struct vmci_event_qp ev;
0856 
0857     ev.msg.hdr.dst = vmci_make_handle(context_id, VMCI_EVENT_HANDLER);
0858     ev.msg.hdr.src = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID,
0859                       VMCI_CONTEXT_RESOURCE_ID);
0860     ev.msg.hdr.payload_size = sizeof(ev) - sizeof(ev.msg.hdr);
0861     ev.msg.event_data.event =
0862         attach ? VMCI_EVENT_QP_PEER_ATTACH : VMCI_EVENT_QP_PEER_DETACH;
0863     ev.payload.peer_id = context_id;
0864     ev.payload.handle = handle;
0865 
0866     return vmci_event_dispatch(&ev.msg.hdr);
0867 }
0868 
0869 /*
0870  * Allocates and initializes a qp_guest_endpoint structure.
0871  * Allocates a queue_pair rid (and handle) iff the given entry has
0872  * an invalid handle.  0 through VMCI_RESERVED_RESOURCE_ID_MAX
0873  * are reserved handles.  Assumes that the QP list mutex is held
0874  * by the caller.
0875  */
0876 static struct qp_guest_endpoint *
0877 qp_guest_endpoint_create(struct vmci_handle handle,
0878              u32 peer,
0879              u32 flags,
0880              u64 produce_size,
0881              u64 consume_size,
0882              void *produce_q,
0883              void *consume_q)
0884 {
0885     int result;
0886     struct qp_guest_endpoint *entry;
0887     /* One page each for the queue headers. */
0888     const u64 num_ppns = DIV_ROUND_UP(produce_size, PAGE_SIZE) +
0889         DIV_ROUND_UP(consume_size, PAGE_SIZE) + 2;
0890 
0891     if (vmci_handle_is_invalid(handle)) {
0892         u32 context_id = vmci_get_context_id();
0893 
0894         handle = vmci_make_handle(context_id, VMCI_INVALID_ID);
0895     }
0896 
0897     entry = kzalloc(sizeof(*entry), GFP_KERNEL);
0898     if (entry) {
0899         entry->qp.peer = peer;
0900         entry->qp.flags = flags;
0901         entry->qp.produce_size = produce_size;
0902         entry->qp.consume_size = consume_size;
0903         entry->qp.ref_count = 0;
0904         entry->num_ppns = num_ppns;
0905         entry->produce_q = produce_q;
0906         entry->consume_q = consume_q;
0907         INIT_LIST_HEAD(&entry->qp.list_item);
0908 
0909         /* Add resource obj */
0910         result = vmci_resource_add(&entry->resource,
0911                        VMCI_RESOURCE_TYPE_QPAIR_GUEST,
0912                        handle);
0913         entry->qp.handle = vmci_resource_handle(&entry->resource);
0914         if ((result != VMCI_SUCCESS) ||
0915             qp_list_find(&qp_guest_endpoints, entry->qp.handle)) {
0916             pr_warn("Failed to add new resource (handle=0x%x:0x%x), error: %d",
0917                 handle.context, handle.resource, result);
0918             kfree(entry);
0919             entry = NULL;
0920         }
0921     }
0922     return entry;
0923 }
0924 
0925 /*
0926  * Frees a qp_guest_endpoint structure.
0927  */
0928 static void qp_guest_endpoint_destroy(struct qp_guest_endpoint *entry)
0929 {
0930     qp_free_ppn_set(&entry->ppn_set);
0931     qp_cleanup_queue_mutex(entry->produce_q, entry->consume_q);
0932     qp_free_queue(entry->produce_q, entry->qp.produce_size);
0933     qp_free_queue(entry->consume_q, entry->qp.consume_size);
0934     /* Unlink from resource hash table and free callback */
0935     vmci_resource_remove(&entry->resource);
0936 
0937     kfree(entry);
0938 }
0939 
0940 /*
0941  * Helper to make a queue_pairAlloc hypercall when the driver is
0942  * supporting a guest device.
0943  */
0944 static int qp_alloc_hypercall(const struct qp_guest_endpoint *entry)
0945 {
0946     struct vmci_qp_alloc_msg *alloc_msg;
0947     size_t msg_size;
0948     size_t ppn_size;
0949     int result;
0950 
0951     if (!entry || entry->num_ppns <= 2)
0952         return VMCI_ERROR_INVALID_ARGS;
0953 
0954     ppn_size = vmci_use_ppn64() ? sizeof(u64) : sizeof(u32);
0955     msg_size = sizeof(*alloc_msg) +
0956         (size_t) entry->num_ppns * ppn_size;
0957     alloc_msg = kmalloc(msg_size, GFP_KERNEL);
0958     if (!alloc_msg)
0959         return VMCI_ERROR_NO_MEM;
0960 
0961     alloc_msg->hdr.dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID,
0962                           VMCI_QUEUEPAIR_ALLOC);
0963     alloc_msg->hdr.src = VMCI_ANON_SRC_HANDLE;
0964     alloc_msg->hdr.payload_size = msg_size - VMCI_DG_HEADERSIZE;
0965     alloc_msg->handle = entry->qp.handle;
0966     alloc_msg->peer = entry->qp.peer;
0967     alloc_msg->flags = entry->qp.flags;
0968     alloc_msg->produce_size = entry->qp.produce_size;
0969     alloc_msg->consume_size = entry->qp.consume_size;
0970     alloc_msg->num_ppns = entry->num_ppns;
0971 
0972     result = qp_populate_ppn_set((u8 *)alloc_msg + sizeof(*alloc_msg),
0973                      &entry->ppn_set);
0974     if (result == VMCI_SUCCESS)
0975         result = vmci_send_datagram(&alloc_msg->hdr);
0976 
0977     kfree(alloc_msg);
0978 
0979     return result;
0980 }
0981 
0982 /*
0983  * Helper to make a queue_pairDetach hypercall when the driver is
0984  * supporting a guest device.
0985  */
0986 static int qp_detatch_hypercall(struct vmci_handle handle)
0987 {
0988     struct vmci_qp_detach_msg detach_msg;
0989 
0990     detach_msg.hdr.dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID,
0991                           VMCI_QUEUEPAIR_DETACH);
0992     detach_msg.hdr.src = VMCI_ANON_SRC_HANDLE;
0993     detach_msg.hdr.payload_size = sizeof(handle);
0994     detach_msg.handle = handle;
0995 
0996     return vmci_send_datagram(&detach_msg.hdr);
0997 }
0998 
0999 /*
1000  * Adds the given entry to the list. Assumes that the list is locked.
1001  */
1002 static void qp_list_add_entry(struct qp_list *qp_list, struct qp_entry *entry)
1003 {
1004     if (entry)
1005         list_add(&entry->list_item, &qp_list->head);
1006 }
1007 
1008 /*
1009  * Removes the given entry from the list. Assumes that the list is locked.
1010  */
1011 static void qp_list_remove_entry(struct qp_list *qp_list,
1012                  struct qp_entry *entry)
1013 {
1014     if (entry)
1015         list_del(&entry->list_item);
1016 }
1017 
1018 /*
1019  * Helper for VMCI queue_pair detach interface. Frees the physical
1020  * pages for the queue pair.
1021  */
1022 static int qp_detatch_guest_work(struct vmci_handle handle)
1023 {
1024     int result;
1025     struct qp_guest_endpoint *entry;
1026     u32 ref_count = ~0; /* To avoid compiler warning below */
1027 
1028     mutex_lock(&qp_guest_endpoints.mutex);
1029 
1030     entry = qp_guest_handle_to_entry(handle);
1031     if (!entry) {
1032         mutex_unlock(&qp_guest_endpoints.mutex);
1033         return VMCI_ERROR_NOT_FOUND;
1034     }
1035 
1036     if (entry->qp.flags & VMCI_QPFLAG_LOCAL) {
1037         result = VMCI_SUCCESS;
1038 
1039         if (entry->qp.ref_count > 1) {
1040             result = qp_notify_peer_local(false, handle);
1041             /*
1042              * We can fail to notify a local queuepair
1043              * because we can't allocate.  We still want
1044              * to release the entry if that happens, so
1045              * don't bail out yet.
1046              */
1047         }
1048     } else {
1049         result = qp_detatch_hypercall(handle);
1050         if (result < VMCI_SUCCESS) {
1051             /*
1052              * We failed to notify a non-local queuepair.
1053              * That other queuepair might still be
1054              * accessing the shared memory, so don't
1055              * release the entry yet.  It will get cleaned
1056              * up by VMCIqueue_pair_Exit() if necessary
1057              * (assuming we are going away, otherwise why
1058              * did this fail?).
1059              */
1060 
1061             mutex_unlock(&qp_guest_endpoints.mutex);
1062             return result;
1063         }
1064     }
1065 
1066     /*
1067      * If we get here then we either failed to notify a local queuepair, or
1068      * we succeeded in all cases.  Release the entry if required.
1069      */
1070 
1071     entry->qp.ref_count--;
1072     if (entry->qp.ref_count == 0)
1073         qp_list_remove_entry(&qp_guest_endpoints, &entry->qp);
1074 
1075     /* If we didn't remove the entry, this could change once we unlock. */
1076     if (entry)
1077         ref_count = entry->qp.ref_count;
1078 
1079     mutex_unlock(&qp_guest_endpoints.mutex);
1080 
1081     if (ref_count == 0)
1082         qp_guest_endpoint_destroy(entry);
1083 
1084     return result;
1085 }
1086 
1087 /*
1088  * This functions handles the actual allocation of a VMCI queue
1089  * pair guest endpoint. Allocates physical pages for the queue
1090  * pair. It makes OS dependent calls through generic wrappers.
1091  */
1092 static int qp_alloc_guest_work(struct vmci_handle *handle,
1093                    struct vmci_queue **produce_q,
1094                    u64 produce_size,
1095                    struct vmci_queue **consume_q,
1096                    u64 consume_size,
1097                    u32 peer,
1098                    u32 flags,
1099                    u32 priv_flags)
1100 {
1101     const u64 num_produce_pages =
1102         DIV_ROUND_UP(produce_size, PAGE_SIZE) + 1;
1103     const u64 num_consume_pages =
1104         DIV_ROUND_UP(consume_size, PAGE_SIZE) + 1;
1105     void *my_produce_q = NULL;
1106     void *my_consume_q = NULL;
1107     int result;
1108     struct qp_guest_endpoint *queue_pair_entry = NULL;
1109 
1110     if (priv_flags != VMCI_NO_PRIVILEGE_FLAGS)
1111         return VMCI_ERROR_NO_ACCESS;
1112 
1113     mutex_lock(&qp_guest_endpoints.mutex);
1114 
1115     queue_pair_entry = qp_guest_handle_to_entry(*handle);
1116     if (queue_pair_entry) {
1117         if (queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) {
1118             /* Local attach case. */
1119             if (queue_pair_entry->qp.ref_count > 1) {
1120                 pr_devel("Error attempting to attach more than once\n");
1121                 result = VMCI_ERROR_UNAVAILABLE;
1122                 goto error_keep_entry;
1123             }
1124 
1125             if (queue_pair_entry->qp.produce_size != consume_size ||
1126                 queue_pair_entry->qp.consume_size !=
1127                 produce_size ||
1128                 queue_pair_entry->qp.flags !=
1129                 (flags & ~VMCI_QPFLAG_ATTACH_ONLY)) {
1130                 pr_devel("Error mismatched queue pair in local attach\n");
1131                 result = VMCI_ERROR_QUEUEPAIR_MISMATCH;
1132                 goto error_keep_entry;
1133             }
1134 
1135             /*
1136              * Do a local attach.  We swap the consume and
1137              * produce queues for the attacher and deliver
1138              * an attach event.
1139              */
1140             result = qp_notify_peer_local(true, *handle);
1141             if (result < VMCI_SUCCESS)
1142                 goto error_keep_entry;
1143 
1144             my_produce_q = queue_pair_entry->consume_q;
1145             my_consume_q = queue_pair_entry->produce_q;
1146             goto out;
1147         }
1148 
1149         result = VMCI_ERROR_ALREADY_EXISTS;
1150         goto error_keep_entry;
1151     }
1152 
1153     my_produce_q = qp_alloc_queue(produce_size, flags);
1154     if (!my_produce_q) {
1155         pr_warn("Error allocating pages for produce queue\n");
1156         result = VMCI_ERROR_NO_MEM;
1157         goto error;
1158     }
1159 
1160     my_consume_q = qp_alloc_queue(consume_size, flags);
1161     if (!my_consume_q) {
1162         pr_warn("Error allocating pages for consume queue\n");
1163         result = VMCI_ERROR_NO_MEM;
1164         goto error;
1165     }
1166 
1167     queue_pair_entry = qp_guest_endpoint_create(*handle, peer, flags,
1168                             produce_size, consume_size,
1169                             my_produce_q, my_consume_q);
1170     if (!queue_pair_entry) {
1171         pr_warn("Error allocating memory in %s\n", __func__);
1172         result = VMCI_ERROR_NO_MEM;
1173         goto error;
1174     }
1175 
1176     result = qp_alloc_ppn_set(my_produce_q, num_produce_pages, my_consume_q,
1177                   num_consume_pages,
1178                   &queue_pair_entry->ppn_set);
1179     if (result < VMCI_SUCCESS) {
1180         pr_warn("qp_alloc_ppn_set failed\n");
1181         goto error;
1182     }
1183 
1184     /*
1185      * It's only necessary to notify the host if this queue pair will be
1186      * attached to from another context.
1187      */
1188     if (queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) {
1189         /* Local create case. */
1190         u32 context_id = vmci_get_context_id();
1191 
1192         /*
1193          * Enforce similar checks on local queue pairs as we
1194          * do for regular ones.  The handle's context must
1195          * match the creator or attacher context id (here they
1196          * are both the current context id) and the
1197          * attach-only flag cannot exist during create.  We
1198          * also ensure specified peer is this context or an
1199          * invalid one.
1200          */
1201         if (queue_pair_entry->qp.handle.context != context_id ||
1202             (queue_pair_entry->qp.peer != VMCI_INVALID_ID &&
1203              queue_pair_entry->qp.peer != context_id)) {
1204             result = VMCI_ERROR_NO_ACCESS;
1205             goto error;
1206         }
1207 
1208         if (queue_pair_entry->qp.flags & VMCI_QPFLAG_ATTACH_ONLY) {
1209             result = VMCI_ERROR_NOT_FOUND;
1210             goto error;
1211         }
1212     } else {
1213         result = qp_alloc_hypercall(queue_pair_entry);
1214         if (result < VMCI_SUCCESS) {
1215             pr_devel("qp_alloc_hypercall result = %d\n", result);
1216             goto error;
1217         }
1218     }
1219 
1220     qp_init_queue_mutex((struct vmci_queue *)my_produce_q,
1221                 (struct vmci_queue *)my_consume_q);
1222 
1223     qp_list_add_entry(&qp_guest_endpoints, &queue_pair_entry->qp);
1224 
1225  out:
1226     queue_pair_entry->qp.ref_count++;
1227     *handle = queue_pair_entry->qp.handle;
1228     *produce_q = (struct vmci_queue *)my_produce_q;
1229     *consume_q = (struct vmci_queue *)my_consume_q;
1230 
1231     /*
1232      * We should initialize the queue pair header pages on a local
1233      * queue pair create.  For non-local queue pairs, the
1234      * hypervisor initializes the header pages in the create step.
1235      */
1236     if ((queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) &&
1237         queue_pair_entry->qp.ref_count == 1) {
1238         vmci_q_header_init((*produce_q)->q_header, *handle);
1239         vmci_q_header_init((*consume_q)->q_header, *handle);
1240     }
1241 
1242     mutex_unlock(&qp_guest_endpoints.mutex);
1243 
1244     return VMCI_SUCCESS;
1245 
1246  error:
1247     mutex_unlock(&qp_guest_endpoints.mutex);
1248     if (queue_pair_entry) {
1249         /* The queues will be freed inside the destroy routine. */
1250         qp_guest_endpoint_destroy(queue_pair_entry);
1251     } else {
1252         qp_free_queue(my_produce_q, produce_size);
1253         qp_free_queue(my_consume_q, consume_size);
1254     }
1255     return result;
1256 
1257  error_keep_entry:
1258     /* This path should only be used when an existing entry was found. */
1259     mutex_unlock(&qp_guest_endpoints.mutex);
1260     return result;
1261 }
1262 
1263 /*
1264  * The first endpoint issuing a queue pair allocation will create the state
1265  * of the queue pair in the queue pair broker.
1266  *
1267  * If the creator is a guest, it will associate a VMX virtual address range
1268  * with the queue pair as specified by the page_store. For compatibility with
1269  * older VMX'en, that would use a separate step to set the VMX virtual
1270  * address range, the virtual address range can be registered later using
1271  * vmci_qp_broker_set_page_store. In that case, a page_store of NULL should be
1272  * used.
1273  *
1274  * If the creator is the host, a page_store of NULL should be used as well,
1275  * since the host is not able to supply a page store for the queue pair.
1276  *
1277  * For older VMX and host callers, the queue pair will be created in the
1278  * VMCIQPB_CREATED_NO_MEM state, and for current VMX callers, it will be
1279  * created in VMCOQPB_CREATED_MEM state.
1280  */
1281 static int qp_broker_create(struct vmci_handle handle,
1282                 u32 peer,
1283                 u32 flags,
1284                 u32 priv_flags,
1285                 u64 produce_size,
1286                 u64 consume_size,
1287                 struct vmci_qp_page_store *page_store,
1288                 struct vmci_ctx *context,
1289                 vmci_event_release_cb wakeup_cb,
1290                 void *client_data, struct qp_broker_entry **ent)
1291 {
1292     struct qp_broker_entry *entry = NULL;
1293     const u32 context_id = vmci_ctx_get_id(context);
1294     bool is_local = flags & VMCI_QPFLAG_LOCAL;
1295     int result;
1296     u64 guest_produce_size;
1297     u64 guest_consume_size;
1298 
1299     /* Do not create if the caller asked not to. */
1300     if (flags & VMCI_QPFLAG_ATTACH_ONLY)
1301         return VMCI_ERROR_NOT_FOUND;
1302 
1303     /*
1304      * Creator's context ID should match handle's context ID or the creator
1305      * must allow the context in handle's context ID as the "peer".
1306      */
1307     if (handle.context != context_id && handle.context != peer)
1308         return VMCI_ERROR_NO_ACCESS;
1309 
1310     if (VMCI_CONTEXT_IS_VM(context_id) && VMCI_CONTEXT_IS_VM(peer))
1311         return VMCI_ERROR_DST_UNREACHABLE;
1312 
1313     /*
1314      * Creator's context ID for local queue pairs should match the
1315      * peer, if a peer is specified.
1316      */
1317     if (is_local && peer != VMCI_INVALID_ID && context_id != peer)
1318         return VMCI_ERROR_NO_ACCESS;
1319 
1320     entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
1321     if (!entry)
1322         return VMCI_ERROR_NO_MEM;
1323 
1324     if (vmci_ctx_get_id(context) == VMCI_HOST_CONTEXT_ID && !is_local) {
1325         /*
1326          * The queue pair broker entry stores values from the guest
1327          * point of view, so a creating host side endpoint should swap
1328          * produce and consume values -- unless it is a local queue
1329          * pair, in which case no swapping is necessary, since the local
1330          * attacher will swap queues.
1331          */
1332 
1333         guest_produce_size = consume_size;
1334         guest_consume_size = produce_size;
1335     } else {
1336         guest_produce_size = produce_size;
1337         guest_consume_size = consume_size;
1338     }
1339 
1340     entry->qp.handle = handle;
1341     entry->qp.peer = peer;
1342     entry->qp.flags = flags;
1343     entry->qp.produce_size = guest_produce_size;
1344     entry->qp.consume_size = guest_consume_size;
1345     entry->qp.ref_count = 1;
1346     entry->create_id = context_id;
1347     entry->attach_id = VMCI_INVALID_ID;
1348     entry->state = VMCIQPB_NEW;
1349     entry->require_trusted_attach =
1350         !!(context->priv_flags & VMCI_PRIVILEGE_FLAG_RESTRICTED);
1351     entry->created_by_trusted =
1352         !!(priv_flags & VMCI_PRIVILEGE_FLAG_TRUSTED);
1353     entry->vmci_page_files = false;
1354     entry->wakeup_cb = wakeup_cb;
1355     entry->client_data = client_data;
1356     entry->produce_q = qp_host_alloc_queue(guest_produce_size);
1357     if (entry->produce_q == NULL) {
1358         result = VMCI_ERROR_NO_MEM;
1359         goto error;
1360     }
1361     entry->consume_q = qp_host_alloc_queue(guest_consume_size);
1362     if (entry->consume_q == NULL) {
1363         result = VMCI_ERROR_NO_MEM;
1364         goto error;
1365     }
1366 
1367     qp_init_queue_mutex(entry->produce_q, entry->consume_q);
1368 
1369     INIT_LIST_HEAD(&entry->qp.list_item);
1370 
1371     if (is_local) {
1372         u8 *tmp;
1373 
1374         entry->local_mem = kcalloc(QPE_NUM_PAGES(entry->qp),
1375                        PAGE_SIZE, GFP_KERNEL);
1376         if (entry->local_mem == NULL) {
1377             result = VMCI_ERROR_NO_MEM;
1378             goto error;
1379         }
1380         entry->state = VMCIQPB_CREATED_MEM;
1381         entry->produce_q->q_header = entry->local_mem;
1382         tmp = (u8 *)entry->local_mem + PAGE_SIZE *
1383             (DIV_ROUND_UP(entry->qp.produce_size, PAGE_SIZE) + 1);
1384         entry->consume_q->q_header = (struct vmci_queue_header *)tmp;
1385     } else if (page_store) {
1386         /*
1387          * The VMX already initialized the queue pair headers, so no
1388          * need for the kernel side to do that.
1389          */
1390         result = qp_host_register_user_memory(page_store,
1391                               entry->produce_q,
1392                               entry->consume_q);
1393         if (result < VMCI_SUCCESS)
1394             goto error;
1395 
1396         entry->state = VMCIQPB_CREATED_MEM;
1397     } else {
1398         /*
1399          * A create without a page_store may be either a host
1400          * side create (in which case we are waiting for the
1401          * guest side to supply the memory) or an old style
1402          * queue pair create (in which case we will expect a
1403          * set page store call as the next step).
1404          */
1405         entry->state = VMCIQPB_CREATED_NO_MEM;
1406     }
1407 
1408     qp_list_add_entry(&qp_broker_list, &entry->qp);
1409     if (ent != NULL)
1410         *ent = entry;
1411 
1412     /* Add to resource obj */
1413     result = vmci_resource_add(&entry->resource,
1414                    VMCI_RESOURCE_TYPE_QPAIR_HOST,
1415                    handle);
1416     if (result != VMCI_SUCCESS) {
1417         pr_warn("Failed to add new resource (handle=0x%x:0x%x), error: %d",
1418             handle.context, handle.resource, result);
1419         goto error;
1420     }
1421 
1422     entry->qp.handle = vmci_resource_handle(&entry->resource);
1423     if (is_local) {
1424         vmci_q_header_init(entry->produce_q->q_header,
1425                    entry->qp.handle);
1426         vmci_q_header_init(entry->consume_q->q_header,
1427                    entry->qp.handle);
1428     }
1429 
1430     vmci_ctx_qp_create(context, entry->qp.handle);
1431 
1432     return VMCI_SUCCESS;
1433 
1434  error:
1435     if (entry != NULL) {
1436         qp_host_free_queue(entry->produce_q, guest_produce_size);
1437         qp_host_free_queue(entry->consume_q, guest_consume_size);
1438         kfree(entry);
1439     }
1440 
1441     return result;
1442 }
1443 
1444 /*
1445  * Enqueues an event datagram to notify the peer VM attached to
1446  * the given queue pair handle about attach/detach event by the
1447  * given VM.  Returns Payload size of datagram enqueued on
1448  * success, error code otherwise.
1449  */
1450 static int qp_notify_peer(bool attach,
1451               struct vmci_handle handle,
1452               u32 my_id,
1453               u32 peer_id)
1454 {
1455     int rv;
1456     struct vmci_event_qp ev;
1457 
1458     if (vmci_handle_is_invalid(handle) || my_id == VMCI_INVALID_ID ||
1459         peer_id == VMCI_INVALID_ID)
1460         return VMCI_ERROR_INVALID_ARGS;
1461 
1462     /*
1463      * In vmci_ctx_enqueue_datagram() we enforce the upper limit on
1464      * number of pending events from the hypervisor to a given VM
1465      * otherwise a rogue VM could do an arbitrary number of attach
1466      * and detach operations causing memory pressure in the host
1467      * kernel.
1468      */
1469 
1470     ev.msg.hdr.dst = vmci_make_handle(peer_id, VMCI_EVENT_HANDLER);
1471     ev.msg.hdr.src = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID,
1472                       VMCI_CONTEXT_RESOURCE_ID);
1473     ev.msg.hdr.payload_size = sizeof(ev) - sizeof(ev.msg.hdr);
1474     ev.msg.event_data.event = attach ?
1475         VMCI_EVENT_QP_PEER_ATTACH : VMCI_EVENT_QP_PEER_DETACH;
1476     ev.payload.handle = handle;
1477     ev.payload.peer_id = my_id;
1478 
1479     rv = vmci_datagram_dispatch(VMCI_HYPERVISOR_CONTEXT_ID,
1480                     &ev.msg.hdr, false);
1481     if (rv < VMCI_SUCCESS)
1482         pr_warn("Failed to enqueue queue_pair %s event datagram for context (ID=0x%x)\n",
1483             attach ? "ATTACH" : "DETACH", peer_id);
1484 
1485     return rv;
1486 }
1487 
1488 /*
1489  * The second endpoint issuing a queue pair allocation will attach to
1490  * the queue pair registered with the queue pair broker.
1491  *
1492  * If the attacher is a guest, it will associate a VMX virtual address
1493  * range with the queue pair as specified by the page_store. At this
1494  * point, the already attach host endpoint may start using the queue
1495  * pair, and an attach event is sent to it. For compatibility with
1496  * older VMX'en, that used a separate step to set the VMX virtual
1497  * address range, the virtual address range can be registered later
1498  * using vmci_qp_broker_set_page_store. In that case, a page_store of
1499  * NULL should be used, and the attach event will be generated once
1500  * the actual page store has been set.
1501  *
1502  * If the attacher is the host, a page_store of NULL should be used as
1503  * well, since the page store information is already set by the guest.
1504  *
1505  * For new VMX and host callers, the queue pair will be moved to the
1506  * VMCIQPB_ATTACHED_MEM state, and for older VMX callers, it will be
1507  * moved to the VMCOQPB_ATTACHED_NO_MEM state.
1508  */
1509 static int qp_broker_attach(struct qp_broker_entry *entry,
1510                 u32 peer,
1511                 u32 flags,
1512                 u32 priv_flags,
1513                 u64 produce_size,
1514                 u64 consume_size,
1515                 struct vmci_qp_page_store *page_store,
1516                 struct vmci_ctx *context,
1517                 vmci_event_release_cb wakeup_cb,
1518                 void *client_data,
1519                 struct qp_broker_entry **ent)
1520 {
1521     const u32 context_id = vmci_ctx_get_id(context);
1522     bool is_local = flags & VMCI_QPFLAG_LOCAL;
1523     int result;
1524 
1525     if (entry->state != VMCIQPB_CREATED_NO_MEM &&
1526         entry->state != VMCIQPB_CREATED_MEM)
1527         return VMCI_ERROR_UNAVAILABLE;
1528 
1529     if (is_local) {
1530         if (!(entry->qp.flags & VMCI_QPFLAG_LOCAL) ||
1531             context_id != entry->create_id) {
1532             return VMCI_ERROR_INVALID_ARGS;
1533         }
1534     } else if (context_id == entry->create_id ||
1535            context_id == entry->attach_id) {
1536         return VMCI_ERROR_ALREADY_EXISTS;
1537     }
1538 
1539     if (VMCI_CONTEXT_IS_VM(context_id) &&
1540         VMCI_CONTEXT_IS_VM(entry->create_id))
1541         return VMCI_ERROR_DST_UNREACHABLE;
1542 
1543     /*
1544      * If we are attaching from a restricted context then the queuepair
1545      * must have been created by a trusted endpoint.
1546      */
1547     if ((context->priv_flags & VMCI_PRIVILEGE_FLAG_RESTRICTED) &&
1548         !entry->created_by_trusted)
1549         return VMCI_ERROR_NO_ACCESS;
1550 
1551     /*
1552      * If we are attaching to a queuepair that was created by a restricted
1553      * context then we must be trusted.
1554      */
1555     if (entry->require_trusted_attach &&
1556         (!(priv_flags & VMCI_PRIVILEGE_FLAG_TRUSTED)))
1557         return VMCI_ERROR_NO_ACCESS;
1558 
1559     /*
1560      * If the creator specifies VMCI_INVALID_ID in "peer" field, access
1561      * control check is not performed.
1562      */
1563     if (entry->qp.peer != VMCI_INVALID_ID && entry->qp.peer != context_id)
1564         return VMCI_ERROR_NO_ACCESS;
1565 
1566     if (entry->create_id == VMCI_HOST_CONTEXT_ID) {
1567         /*
1568          * Do not attach if the caller doesn't support Host Queue Pairs
1569          * and a host created this queue pair.
1570          */
1571 
1572         if (!vmci_ctx_supports_host_qp(context))
1573             return VMCI_ERROR_INVALID_RESOURCE;
1574 
1575     } else if (context_id == VMCI_HOST_CONTEXT_ID) {
1576         struct vmci_ctx *create_context;
1577         bool supports_host_qp;
1578 
1579         /*
1580          * Do not attach a host to a user created queue pair if that
1581          * user doesn't support host queue pair end points.
1582          */
1583 
1584         create_context = vmci_ctx_get(entry->create_id);
1585         supports_host_qp = vmci_ctx_supports_host_qp(create_context);
1586         vmci_ctx_put(create_context);
1587 
1588         if (!supports_host_qp)
1589             return VMCI_ERROR_INVALID_RESOURCE;
1590     }
1591 
1592     if ((entry->qp.flags & ~VMCI_QP_ASYMM) != (flags & ~VMCI_QP_ASYMM_PEER))
1593         return VMCI_ERROR_QUEUEPAIR_MISMATCH;
1594 
1595     if (context_id != VMCI_HOST_CONTEXT_ID) {
1596         /*
1597          * The queue pair broker entry stores values from the guest
1598          * point of view, so an attaching guest should match the values
1599          * stored in the entry.
1600          */
1601 
1602         if (entry->qp.produce_size != produce_size ||
1603             entry->qp.consume_size != consume_size) {
1604             return VMCI_ERROR_QUEUEPAIR_MISMATCH;
1605         }
1606     } else if (entry->qp.produce_size != consume_size ||
1607            entry->qp.consume_size != produce_size) {
1608         return VMCI_ERROR_QUEUEPAIR_MISMATCH;
1609     }
1610 
1611     if (context_id != VMCI_HOST_CONTEXT_ID) {
1612         /*
1613          * If a guest attached to a queue pair, it will supply
1614          * the backing memory.  If this is a pre NOVMVM vmx,
1615          * the backing memory will be supplied by calling
1616          * vmci_qp_broker_set_page_store() following the
1617          * return of the vmci_qp_broker_alloc() call. If it is
1618          * a vmx of version NOVMVM or later, the page store
1619          * must be supplied as part of the
1620          * vmci_qp_broker_alloc call.  Under all circumstances
1621          * must the initially created queue pair not have any
1622          * memory associated with it already.
1623          */
1624 
1625         if (entry->state != VMCIQPB_CREATED_NO_MEM)
1626             return VMCI_ERROR_INVALID_ARGS;
1627 
1628         if (page_store != NULL) {
1629             /*
1630              * Patch up host state to point to guest
1631              * supplied memory. The VMX already
1632              * initialized the queue pair headers, so no
1633              * need for the kernel side to do that.
1634              */
1635 
1636             result = qp_host_register_user_memory(page_store,
1637                                   entry->produce_q,
1638                                   entry->consume_q);
1639             if (result < VMCI_SUCCESS)
1640                 return result;
1641 
1642             entry->state = VMCIQPB_ATTACHED_MEM;
1643         } else {
1644             entry->state = VMCIQPB_ATTACHED_NO_MEM;
1645         }
1646     } else if (entry->state == VMCIQPB_CREATED_NO_MEM) {
1647         /*
1648          * The host side is attempting to attach to a queue
1649          * pair that doesn't have any memory associated with
1650          * it. This must be a pre NOVMVM vmx that hasn't set
1651          * the page store information yet, or a quiesced VM.
1652          */
1653 
1654         return VMCI_ERROR_UNAVAILABLE;
1655     } else {
1656         /* The host side has successfully attached to a queue pair. */
1657         entry->state = VMCIQPB_ATTACHED_MEM;
1658     }
1659 
1660     if (entry->state == VMCIQPB_ATTACHED_MEM) {
1661         result =
1662             qp_notify_peer(true, entry->qp.handle, context_id,
1663                    entry->create_id);
1664         if (result < VMCI_SUCCESS)
1665             pr_warn("Failed to notify peer (ID=0x%x) of attach to queue pair (handle=0x%x:0x%x)\n",
1666                 entry->create_id, entry->qp.handle.context,
1667                 entry->qp.handle.resource);
1668     }
1669 
1670     entry->attach_id = context_id;
1671     entry->qp.ref_count++;
1672     if (wakeup_cb) {
1673         entry->wakeup_cb = wakeup_cb;
1674         entry->client_data = client_data;
1675     }
1676 
1677     /*
1678      * When attaching to local queue pairs, the context already has
1679      * an entry tracking the queue pair, so don't add another one.
1680      */
1681     if (!is_local)
1682         vmci_ctx_qp_create(context, entry->qp.handle);
1683 
1684     if (ent != NULL)
1685         *ent = entry;
1686 
1687     return VMCI_SUCCESS;
1688 }
1689 
1690 /*
1691  * queue_pair_Alloc for use when setting up queue pair endpoints
1692  * on the host.
1693  */
1694 static int qp_broker_alloc(struct vmci_handle handle,
1695                u32 peer,
1696                u32 flags,
1697                u32 priv_flags,
1698                u64 produce_size,
1699                u64 consume_size,
1700                struct vmci_qp_page_store *page_store,
1701                struct vmci_ctx *context,
1702                vmci_event_release_cb wakeup_cb,
1703                void *client_data,
1704                struct qp_broker_entry **ent,
1705                bool *swap)
1706 {
1707     const u32 context_id = vmci_ctx_get_id(context);
1708     bool create;
1709     struct qp_broker_entry *entry = NULL;
1710     bool is_local = flags & VMCI_QPFLAG_LOCAL;
1711     int result;
1712 
1713     if (vmci_handle_is_invalid(handle) ||
1714         (flags & ~VMCI_QP_ALL_FLAGS) || is_local ||
1715         !(produce_size || consume_size) ||
1716         !context || context_id == VMCI_INVALID_ID ||
1717         handle.context == VMCI_INVALID_ID) {
1718         return VMCI_ERROR_INVALID_ARGS;
1719     }
1720 
1721     if (page_store && !VMCI_QP_PAGESTORE_IS_WELLFORMED(page_store))
1722         return VMCI_ERROR_INVALID_ARGS;
1723 
1724     /*
1725      * In the initial argument check, we ensure that non-vmkernel hosts
1726      * are not allowed to create local queue pairs.
1727      */
1728 
1729     mutex_lock(&qp_broker_list.mutex);
1730 
1731     if (!is_local && vmci_ctx_qp_exists(context, handle)) {
1732         pr_devel("Context (ID=0x%x) already attached to queue pair (handle=0x%x:0x%x)\n",
1733              context_id, handle.context, handle.resource);
1734         mutex_unlock(&qp_broker_list.mutex);
1735         return VMCI_ERROR_ALREADY_EXISTS;
1736     }
1737 
1738     if (handle.resource != VMCI_INVALID_ID)
1739         entry = qp_broker_handle_to_entry(handle);
1740 
1741     if (!entry) {
1742         create = true;
1743         result =
1744             qp_broker_create(handle, peer, flags, priv_flags,
1745                      produce_size, consume_size, page_store,
1746                      context, wakeup_cb, client_data, ent);
1747     } else {
1748         create = false;
1749         result =
1750             qp_broker_attach(entry, peer, flags, priv_flags,
1751                      produce_size, consume_size, page_store,
1752                      context, wakeup_cb, client_data, ent);
1753     }
1754 
1755     mutex_unlock(&qp_broker_list.mutex);
1756 
1757     if (swap)
1758         *swap = (context_id == VMCI_HOST_CONTEXT_ID) &&
1759             !(create && is_local);
1760 
1761     return result;
1762 }
1763 
1764 /*
1765  * This function implements the kernel API for allocating a queue
1766  * pair.
1767  */
1768 static int qp_alloc_host_work(struct vmci_handle *handle,
1769                   struct vmci_queue **produce_q,
1770                   u64 produce_size,
1771                   struct vmci_queue **consume_q,
1772                   u64 consume_size,
1773                   u32 peer,
1774                   u32 flags,
1775                   u32 priv_flags,
1776                   vmci_event_release_cb wakeup_cb,
1777                   void *client_data)
1778 {
1779     struct vmci_handle new_handle;
1780     struct vmci_ctx *context;
1781     struct qp_broker_entry *entry;
1782     int result;
1783     bool swap;
1784 
1785     if (vmci_handle_is_invalid(*handle)) {
1786         new_handle = vmci_make_handle(
1787             VMCI_HOST_CONTEXT_ID, VMCI_INVALID_ID);
1788     } else
1789         new_handle = *handle;
1790 
1791     context = vmci_ctx_get(VMCI_HOST_CONTEXT_ID);
1792     entry = NULL;
1793     result =
1794         qp_broker_alloc(new_handle, peer, flags, priv_flags,
1795                 produce_size, consume_size, NULL, context,
1796                 wakeup_cb, client_data, &entry, &swap);
1797     if (result == VMCI_SUCCESS) {
1798         if (swap) {
1799             /*
1800              * If this is a local queue pair, the attacher
1801              * will swap around produce and consume
1802              * queues.
1803              */
1804 
1805             *produce_q = entry->consume_q;
1806             *consume_q = entry->produce_q;
1807         } else {
1808             *produce_q = entry->produce_q;
1809             *consume_q = entry->consume_q;
1810         }
1811 
1812         *handle = vmci_resource_handle(&entry->resource);
1813     } else {
1814         *handle = VMCI_INVALID_HANDLE;
1815         pr_devel("queue pair broker failed to alloc (result=%d)\n",
1816              result);
1817     }
1818     vmci_ctx_put(context);
1819     return result;
1820 }
1821 
1822 /*
1823  * Allocates a VMCI queue_pair. Only checks validity of input
1824  * arguments. The real work is done in the host or guest
1825  * specific function.
1826  */
1827 int vmci_qp_alloc(struct vmci_handle *handle,
1828           struct vmci_queue **produce_q,
1829           u64 produce_size,
1830           struct vmci_queue **consume_q,
1831           u64 consume_size,
1832           u32 peer,
1833           u32 flags,
1834           u32 priv_flags,
1835           bool guest_endpoint,
1836           vmci_event_release_cb wakeup_cb,
1837           void *client_data)
1838 {
1839     if (!handle || !produce_q || !consume_q ||
1840         (!produce_size && !consume_size) || (flags & ~VMCI_QP_ALL_FLAGS))
1841         return VMCI_ERROR_INVALID_ARGS;
1842 
1843     if (guest_endpoint) {
1844         return qp_alloc_guest_work(handle, produce_q,
1845                        produce_size, consume_q,
1846                        consume_size, peer,
1847                        flags, priv_flags);
1848     } else {
1849         return qp_alloc_host_work(handle, produce_q,
1850                       produce_size, consume_q,
1851                       consume_size, peer, flags,
1852                       priv_flags, wakeup_cb, client_data);
1853     }
1854 }
1855 
1856 /*
1857  * This function implements the host kernel API for detaching from
1858  * a queue pair.
1859  */
1860 static int qp_detatch_host_work(struct vmci_handle handle)
1861 {
1862     int result;
1863     struct vmci_ctx *context;
1864 
1865     context = vmci_ctx_get(VMCI_HOST_CONTEXT_ID);
1866 
1867     result = vmci_qp_broker_detach(handle, context);
1868 
1869     vmci_ctx_put(context);
1870     return result;
1871 }
1872 
1873 /*
1874  * Detaches from a VMCI queue_pair. Only checks validity of input argument.
1875  * Real work is done in the host or guest specific function.
1876  */
1877 static int qp_detatch(struct vmci_handle handle, bool guest_endpoint)
1878 {
1879     if (vmci_handle_is_invalid(handle))
1880         return VMCI_ERROR_INVALID_ARGS;
1881 
1882     if (guest_endpoint)
1883         return qp_detatch_guest_work(handle);
1884     else
1885         return qp_detatch_host_work(handle);
1886 }
1887 
1888 /*
1889  * Returns the entry from the head of the list. Assumes that the list is
1890  * locked.
1891  */
1892 static struct qp_entry *qp_list_get_head(struct qp_list *qp_list)
1893 {
1894     if (!list_empty(&qp_list->head)) {
1895         struct qp_entry *entry =
1896             list_first_entry(&qp_list->head, struct qp_entry,
1897                      list_item);
1898         return entry;
1899     }
1900 
1901     return NULL;
1902 }
1903 
1904 void vmci_qp_broker_exit(void)
1905 {
1906     struct qp_entry *entry;
1907     struct qp_broker_entry *be;
1908 
1909     mutex_lock(&qp_broker_list.mutex);
1910 
1911     while ((entry = qp_list_get_head(&qp_broker_list))) {
1912         be = (struct qp_broker_entry *)entry;
1913 
1914         qp_list_remove_entry(&qp_broker_list, entry);
1915         kfree(be);
1916     }
1917 
1918     mutex_unlock(&qp_broker_list.mutex);
1919 }
1920 
1921 /*
1922  * Requests that a queue pair be allocated with the VMCI queue
1923  * pair broker. Allocates a queue pair entry if one does not
1924  * exist. Attaches to one if it exists, and retrieves the page
1925  * files backing that queue_pair.  Assumes that the queue pair
1926  * broker lock is held.
1927  */
1928 int vmci_qp_broker_alloc(struct vmci_handle handle,
1929              u32 peer,
1930              u32 flags,
1931              u32 priv_flags,
1932              u64 produce_size,
1933              u64 consume_size,
1934              struct vmci_qp_page_store *page_store,
1935              struct vmci_ctx *context)
1936 {
1937     if (!QP_SIZES_ARE_VALID(produce_size, consume_size))
1938         return VMCI_ERROR_NO_RESOURCES;
1939 
1940     return qp_broker_alloc(handle, peer, flags, priv_flags,
1941                    produce_size, consume_size,
1942                    page_store, context, NULL, NULL, NULL, NULL);
1943 }
1944 
1945 /*
1946  * VMX'en with versions lower than VMCI_VERSION_NOVMVM use a separate
1947  * step to add the UVAs of the VMX mapping of the queue pair. This function
1948  * provides backwards compatibility with such VMX'en, and takes care of
1949  * registering the page store for a queue pair previously allocated by the
1950  * VMX during create or attach. This function will move the queue pair state
1951  * to either from VMCIQBP_CREATED_NO_MEM to VMCIQBP_CREATED_MEM or
1952  * VMCIQBP_ATTACHED_NO_MEM to VMCIQBP_ATTACHED_MEM. If moving to the
1953  * attached state with memory, the queue pair is ready to be used by the
1954  * host peer, and an attached event will be generated.
1955  *
1956  * Assumes that the queue pair broker lock is held.
1957  *
1958  * This function is only used by the hosted platform, since there is no
1959  * issue with backwards compatibility for vmkernel.
1960  */
1961 int vmci_qp_broker_set_page_store(struct vmci_handle handle,
1962                   u64 produce_uva,
1963                   u64 consume_uva,
1964                   struct vmci_ctx *context)
1965 {
1966     struct qp_broker_entry *entry;
1967     int result;
1968     const u32 context_id = vmci_ctx_get_id(context);
1969 
1970     if (vmci_handle_is_invalid(handle) || !context ||
1971         context_id == VMCI_INVALID_ID)
1972         return VMCI_ERROR_INVALID_ARGS;
1973 
1974     /*
1975      * We only support guest to host queue pairs, so the VMX must
1976      * supply UVAs for the mapped page files.
1977      */
1978 
1979     if (produce_uva == 0 || consume_uva == 0)
1980         return VMCI_ERROR_INVALID_ARGS;
1981 
1982     mutex_lock(&qp_broker_list.mutex);
1983 
1984     if (!vmci_ctx_qp_exists(context, handle)) {
1985         pr_warn("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n",
1986             context_id, handle.context, handle.resource);
1987         result = VMCI_ERROR_NOT_FOUND;
1988         goto out;
1989     }
1990 
1991     entry = qp_broker_handle_to_entry(handle);
1992     if (!entry) {
1993         result = VMCI_ERROR_NOT_FOUND;
1994         goto out;
1995     }
1996 
1997     /*
1998      * If I'm the owner then I can set the page store.
1999      *
2000      * Or, if a host created the queue_pair and I'm the attached peer
2001      * then I can set the page store.
2002      */
2003     if (entry->create_id != context_id &&
2004         (entry->create_id != VMCI_HOST_CONTEXT_ID ||
2005          entry->attach_id != context_id)) {
2006         result = VMCI_ERROR_QUEUEPAIR_NOTOWNER;
2007         goto out;
2008     }
2009 
2010     if (entry->state != VMCIQPB_CREATED_NO_MEM &&
2011         entry->state != VMCIQPB_ATTACHED_NO_MEM) {
2012         result = VMCI_ERROR_UNAVAILABLE;
2013         goto out;
2014     }
2015 
2016     result = qp_host_get_user_memory(produce_uva, consume_uva,
2017                      entry->produce_q, entry->consume_q);
2018     if (result < VMCI_SUCCESS)
2019         goto out;
2020 
2021     result = qp_host_map_queues(entry->produce_q, entry->consume_q);
2022     if (result < VMCI_SUCCESS) {
2023         qp_host_unregister_user_memory(entry->produce_q,
2024                            entry->consume_q);
2025         goto out;
2026     }
2027 
2028     if (entry->state == VMCIQPB_CREATED_NO_MEM)
2029         entry->state = VMCIQPB_CREATED_MEM;
2030     else
2031         entry->state = VMCIQPB_ATTACHED_MEM;
2032 
2033     entry->vmci_page_files = true;
2034 
2035     if (entry->state == VMCIQPB_ATTACHED_MEM) {
2036         result =
2037             qp_notify_peer(true, handle, context_id, entry->create_id);
2038         if (result < VMCI_SUCCESS) {
2039             pr_warn("Failed to notify peer (ID=0x%x) of attach to queue pair (handle=0x%x:0x%x)\n",
2040                 entry->create_id, entry->qp.handle.context,
2041                 entry->qp.handle.resource);
2042         }
2043     }
2044 
2045     result = VMCI_SUCCESS;
2046  out:
2047     mutex_unlock(&qp_broker_list.mutex);
2048     return result;
2049 }
2050 
2051 /*
2052  * Resets saved queue headers for the given QP broker
2053  * entry. Should be used when guest memory becomes available
2054  * again, or the guest detaches.
2055  */
2056 static void qp_reset_saved_headers(struct qp_broker_entry *entry)
2057 {
2058     entry->produce_q->saved_header = NULL;
2059     entry->consume_q->saved_header = NULL;
2060 }
2061 
2062 /*
2063  * The main entry point for detaching from a queue pair registered with the
2064  * queue pair broker. If more than one endpoint is attached to the queue
2065  * pair, the first endpoint will mainly decrement a reference count and
2066  * generate a notification to its peer. The last endpoint will clean up
2067  * the queue pair state registered with the broker.
2068  *
2069  * When a guest endpoint detaches, it will unmap and unregister the guest
2070  * memory backing the queue pair. If the host is still attached, it will
2071  * no longer be able to access the queue pair content.
2072  *
2073  * If the queue pair is already in a state where there is no memory
2074  * registered for the queue pair (any *_NO_MEM state), it will transition to
2075  * the VMCIQPB_SHUTDOWN_NO_MEM state. This will also happen, if a guest
2076  * endpoint is the first of two endpoints to detach. If the host endpoint is
2077  * the first out of two to detach, the queue pair will move to the
2078  * VMCIQPB_SHUTDOWN_MEM state.
2079  */
2080 int vmci_qp_broker_detach(struct vmci_handle handle, struct vmci_ctx *context)
2081 {
2082     struct qp_broker_entry *entry;
2083     const u32 context_id = vmci_ctx_get_id(context);
2084     u32 peer_id;
2085     bool is_local = false;
2086     int result;
2087 
2088     if (vmci_handle_is_invalid(handle) || !context ||
2089         context_id == VMCI_INVALID_ID) {
2090         return VMCI_ERROR_INVALID_ARGS;
2091     }
2092 
2093     mutex_lock(&qp_broker_list.mutex);
2094 
2095     if (!vmci_ctx_qp_exists(context, handle)) {
2096         pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n",
2097              context_id, handle.context, handle.resource);
2098         result = VMCI_ERROR_NOT_FOUND;
2099         goto out;
2100     }
2101 
2102     entry = qp_broker_handle_to_entry(handle);
2103     if (!entry) {
2104         pr_devel("Context (ID=0x%x) reports being attached to queue pair(handle=0x%x:0x%x) that isn't present in broker\n",
2105              context_id, handle.context, handle.resource);
2106         result = VMCI_ERROR_NOT_FOUND;
2107         goto out;
2108     }
2109 
2110     if (context_id != entry->create_id && context_id != entry->attach_id) {
2111         result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED;
2112         goto out;
2113     }
2114 
2115     if (context_id == entry->create_id) {
2116         peer_id = entry->attach_id;
2117         entry->create_id = VMCI_INVALID_ID;
2118     } else {
2119         peer_id = entry->create_id;
2120         entry->attach_id = VMCI_INVALID_ID;
2121     }
2122     entry->qp.ref_count--;
2123 
2124     is_local = entry->qp.flags & VMCI_QPFLAG_LOCAL;
2125 
2126     if (context_id != VMCI_HOST_CONTEXT_ID) {
2127         bool headers_mapped;
2128 
2129         /*
2130          * Pre NOVMVM vmx'en may detach from a queue pair
2131          * before setting the page store, and in that case
2132          * there is no user memory to detach from. Also, more
2133          * recent VMX'en may detach from a queue pair in the
2134          * quiesced state.
2135          */
2136 
2137         qp_acquire_queue_mutex(entry->produce_q);
2138         headers_mapped = entry->produce_q->q_header ||
2139             entry->consume_q->q_header;
2140         if (QPBROKERSTATE_HAS_MEM(entry)) {
2141             result =
2142                 qp_host_unmap_queues(INVALID_VMCI_GUEST_MEM_ID,
2143                          entry->produce_q,
2144                          entry->consume_q);
2145             if (result < VMCI_SUCCESS)
2146                 pr_warn("Failed to unmap queue headers for queue pair (handle=0x%x:0x%x,result=%d)\n",
2147                     handle.context, handle.resource,
2148                     result);
2149 
2150             qp_host_unregister_user_memory(entry->produce_q,
2151                                entry->consume_q);
2152 
2153         }
2154 
2155         if (!headers_mapped)
2156             qp_reset_saved_headers(entry);
2157 
2158         qp_release_queue_mutex(entry->produce_q);
2159 
2160         if (!headers_mapped && entry->wakeup_cb)
2161             entry->wakeup_cb(entry->client_data);
2162 
2163     } else {
2164         if (entry->wakeup_cb) {
2165             entry->wakeup_cb = NULL;
2166             entry->client_data = NULL;
2167         }
2168     }
2169 
2170     if (entry->qp.ref_count == 0) {
2171         qp_list_remove_entry(&qp_broker_list, &entry->qp);
2172 
2173         if (is_local)
2174             kfree(entry->local_mem);
2175 
2176         qp_cleanup_queue_mutex(entry->produce_q, entry->consume_q);
2177         qp_host_free_queue(entry->produce_q, entry->qp.produce_size);
2178         qp_host_free_queue(entry->consume_q, entry->qp.consume_size);
2179         /* Unlink from resource hash table and free callback */
2180         vmci_resource_remove(&entry->resource);
2181 
2182         kfree(entry);
2183 
2184         vmci_ctx_qp_destroy(context, handle);
2185     } else {
2186         qp_notify_peer(false, handle, context_id, peer_id);
2187         if (context_id == VMCI_HOST_CONTEXT_ID &&
2188             QPBROKERSTATE_HAS_MEM(entry)) {
2189             entry->state = VMCIQPB_SHUTDOWN_MEM;
2190         } else {
2191             entry->state = VMCIQPB_SHUTDOWN_NO_MEM;
2192         }
2193 
2194         if (!is_local)
2195             vmci_ctx_qp_destroy(context, handle);
2196 
2197     }
2198     result = VMCI_SUCCESS;
2199  out:
2200     mutex_unlock(&qp_broker_list.mutex);
2201     return result;
2202 }
2203 
2204 /*
2205  * Establishes the necessary mappings for a queue pair given a
2206  * reference to the queue pair guest memory. This is usually
2207  * called when a guest is unquiesced and the VMX is allowed to
2208  * map guest memory once again.
2209  */
2210 int vmci_qp_broker_map(struct vmci_handle handle,
2211                struct vmci_ctx *context,
2212                u64 guest_mem)
2213 {
2214     struct qp_broker_entry *entry;
2215     const u32 context_id = vmci_ctx_get_id(context);
2216     int result;
2217 
2218     if (vmci_handle_is_invalid(handle) || !context ||
2219         context_id == VMCI_INVALID_ID)
2220         return VMCI_ERROR_INVALID_ARGS;
2221 
2222     mutex_lock(&qp_broker_list.mutex);
2223 
2224     if (!vmci_ctx_qp_exists(context, handle)) {
2225         pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n",
2226              context_id, handle.context, handle.resource);
2227         result = VMCI_ERROR_NOT_FOUND;
2228         goto out;
2229     }
2230 
2231     entry = qp_broker_handle_to_entry(handle);
2232     if (!entry) {
2233         pr_devel("Context (ID=0x%x) reports being attached to queue pair (handle=0x%x:0x%x) that isn't present in broker\n",
2234              context_id, handle.context, handle.resource);
2235         result = VMCI_ERROR_NOT_FOUND;
2236         goto out;
2237     }
2238 
2239     if (context_id != entry->create_id && context_id != entry->attach_id) {
2240         result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED;
2241         goto out;
2242     }
2243 
2244     result = VMCI_SUCCESS;
2245 
2246     if (context_id != VMCI_HOST_CONTEXT_ID &&
2247         !QPBROKERSTATE_HAS_MEM(entry)) {
2248         struct vmci_qp_page_store page_store;
2249 
2250         page_store.pages = guest_mem;
2251         page_store.len = QPE_NUM_PAGES(entry->qp);
2252 
2253         qp_acquire_queue_mutex(entry->produce_q);
2254         qp_reset_saved_headers(entry);
2255         result =
2256             qp_host_register_user_memory(&page_store,
2257                          entry->produce_q,
2258                          entry->consume_q);
2259         qp_release_queue_mutex(entry->produce_q);
2260         if (result == VMCI_SUCCESS) {
2261             /* Move state from *_NO_MEM to *_MEM */
2262 
2263             entry->state++;
2264 
2265             if (entry->wakeup_cb)
2266                 entry->wakeup_cb(entry->client_data);
2267         }
2268     }
2269 
2270  out:
2271     mutex_unlock(&qp_broker_list.mutex);
2272     return result;
2273 }
2274 
2275 /*
2276  * Saves a snapshot of the queue headers for the given QP broker
2277  * entry. Should be used when guest memory is unmapped.
2278  * Results:
2279  * VMCI_SUCCESS on success, appropriate error code if guest memory
2280  * can't be accessed..
2281  */
2282 static int qp_save_headers(struct qp_broker_entry *entry)
2283 {
2284     int result;
2285 
2286     if (entry->produce_q->saved_header != NULL &&
2287         entry->consume_q->saved_header != NULL) {
2288         /*
2289          *  If the headers have already been saved, we don't need to do
2290          *  it again, and we don't want to map in the headers
2291          *  unnecessarily.
2292          */
2293 
2294         return VMCI_SUCCESS;
2295     }
2296 
2297     if (NULL == entry->produce_q->q_header ||
2298         NULL == entry->consume_q->q_header) {
2299         result = qp_host_map_queues(entry->produce_q, entry->consume_q);
2300         if (result < VMCI_SUCCESS)
2301             return result;
2302     }
2303 
2304     memcpy(&entry->saved_produce_q, entry->produce_q->q_header,
2305            sizeof(entry->saved_produce_q));
2306     entry->produce_q->saved_header = &entry->saved_produce_q;
2307     memcpy(&entry->saved_consume_q, entry->consume_q->q_header,
2308            sizeof(entry->saved_consume_q));
2309     entry->consume_q->saved_header = &entry->saved_consume_q;
2310 
2311     return VMCI_SUCCESS;
2312 }
2313 
2314 /*
2315  * Removes all references to the guest memory of a given queue pair, and
2316  * will move the queue pair from state *_MEM to *_NO_MEM. It is usually
2317  * called when a VM is being quiesced where access to guest memory should
2318  * avoided.
2319  */
2320 int vmci_qp_broker_unmap(struct vmci_handle handle,
2321              struct vmci_ctx *context,
2322              u32 gid)
2323 {
2324     struct qp_broker_entry *entry;
2325     const u32 context_id = vmci_ctx_get_id(context);
2326     int result;
2327 
2328     if (vmci_handle_is_invalid(handle) || !context ||
2329         context_id == VMCI_INVALID_ID)
2330         return VMCI_ERROR_INVALID_ARGS;
2331 
2332     mutex_lock(&qp_broker_list.mutex);
2333 
2334     if (!vmci_ctx_qp_exists(context, handle)) {
2335         pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n",
2336              context_id, handle.context, handle.resource);
2337         result = VMCI_ERROR_NOT_FOUND;
2338         goto out;
2339     }
2340 
2341     entry = qp_broker_handle_to_entry(handle);
2342     if (!entry) {
2343         pr_devel("Context (ID=0x%x) reports being attached to queue pair (handle=0x%x:0x%x) that isn't present in broker\n",
2344              context_id, handle.context, handle.resource);
2345         result = VMCI_ERROR_NOT_FOUND;
2346         goto out;
2347     }
2348 
2349     if (context_id != entry->create_id && context_id != entry->attach_id) {
2350         result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED;
2351         goto out;
2352     }
2353 
2354     if (context_id != VMCI_HOST_CONTEXT_ID &&
2355         QPBROKERSTATE_HAS_MEM(entry)) {
2356         qp_acquire_queue_mutex(entry->produce_q);
2357         result = qp_save_headers(entry);
2358         if (result < VMCI_SUCCESS)
2359             pr_warn("Failed to save queue headers for queue pair (handle=0x%x:0x%x,result=%d)\n",
2360                 handle.context, handle.resource, result);
2361 
2362         qp_host_unmap_queues(gid, entry->produce_q, entry->consume_q);
2363 
2364         /*
2365          * On hosted, when we unmap queue pairs, the VMX will also
2366          * unmap the guest memory, so we invalidate the previously
2367          * registered memory. If the queue pair is mapped again at a
2368          * later point in time, we will need to reregister the user
2369          * memory with a possibly new user VA.
2370          */
2371         qp_host_unregister_user_memory(entry->produce_q,
2372                            entry->consume_q);
2373 
2374         /*
2375          * Move state from *_MEM to *_NO_MEM.
2376          */
2377         entry->state--;
2378 
2379         qp_release_queue_mutex(entry->produce_q);
2380     }
2381 
2382     result = VMCI_SUCCESS;
2383 
2384  out:
2385     mutex_unlock(&qp_broker_list.mutex);
2386     return result;
2387 }
2388 
2389 /*
2390  * Destroys all guest queue pair endpoints. If active guest queue
2391  * pairs still exist, hypercalls to attempt detach from these
2392  * queue pairs will be made. Any failure to detach is silently
2393  * ignored.
2394  */
2395 void vmci_qp_guest_endpoints_exit(void)
2396 {
2397     struct qp_entry *entry;
2398     struct qp_guest_endpoint *ep;
2399 
2400     mutex_lock(&qp_guest_endpoints.mutex);
2401 
2402     while ((entry = qp_list_get_head(&qp_guest_endpoints))) {
2403         ep = (struct qp_guest_endpoint *)entry;
2404 
2405         /* Don't make a hypercall for local queue_pairs. */
2406         if (!(entry->flags & VMCI_QPFLAG_LOCAL))
2407             qp_detatch_hypercall(entry->handle);
2408 
2409         /* We cannot fail the exit, so let's reset ref_count. */
2410         entry->ref_count = 0;
2411         qp_list_remove_entry(&qp_guest_endpoints, entry);
2412 
2413         qp_guest_endpoint_destroy(ep);
2414     }
2415 
2416     mutex_unlock(&qp_guest_endpoints.mutex);
2417 }
2418 
2419 /*
2420  * Helper routine that will lock the queue pair before subsequent
2421  * operations.
2422  * Note: Non-blocking on the host side is currently only implemented in ESX.
2423  * Since non-blocking isn't yet implemented on the host personality we
2424  * have no reason to acquire a spin lock.  So to avoid the use of an
2425  * unnecessary lock only acquire the mutex if we can block.
2426  */
2427 static void qp_lock(const struct vmci_qp *qpair)
2428 {
2429     qp_acquire_queue_mutex(qpair->produce_q);
2430 }
2431 
2432 /*
2433  * Helper routine that unlocks the queue pair after calling
2434  * qp_lock.
2435  */
2436 static void qp_unlock(const struct vmci_qp *qpair)
2437 {
2438     qp_release_queue_mutex(qpair->produce_q);
2439 }
2440 
2441 /*
2442  * The queue headers may not be mapped at all times. If a queue is
2443  * currently not mapped, it will be attempted to do so.
2444  */
2445 static int qp_map_queue_headers(struct vmci_queue *produce_q,
2446                 struct vmci_queue *consume_q)
2447 {
2448     int result;
2449 
2450     if (NULL == produce_q->q_header || NULL == consume_q->q_header) {
2451         result = qp_host_map_queues(produce_q, consume_q);
2452         if (result < VMCI_SUCCESS)
2453             return (produce_q->saved_header &&
2454                 consume_q->saved_header) ?
2455                 VMCI_ERROR_QUEUEPAIR_NOT_READY :
2456                 VMCI_ERROR_QUEUEPAIR_NOTATTACHED;
2457     }
2458 
2459     return VMCI_SUCCESS;
2460 }
2461 
2462 /*
2463  * Helper routine that will retrieve the produce and consume
2464  * headers of a given queue pair. If the guest memory of the
2465  * queue pair is currently not available, the saved queue headers
2466  * will be returned, if these are available.
2467  */
2468 static int qp_get_queue_headers(const struct vmci_qp *qpair,
2469                 struct vmci_queue_header **produce_q_header,
2470                 struct vmci_queue_header **consume_q_header)
2471 {
2472     int result;
2473 
2474     result = qp_map_queue_headers(qpair->produce_q, qpair->consume_q);
2475     if (result == VMCI_SUCCESS) {
2476         *produce_q_header = qpair->produce_q->q_header;
2477         *consume_q_header = qpair->consume_q->q_header;
2478     } else if (qpair->produce_q->saved_header &&
2479            qpair->consume_q->saved_header) {
2480         *produce_q_header = qpair->produce_q->saved_header;
2481         *consume_q_header = qpair->consume_q->saved_header;
2482         result = VMCI_SUCCESS;
2483     }
2484 
2485     return result;
2486 }
2487 
2488 /*
2489  * Callback from VMCI queue pair broker indicating that a queue
2490  * pair that was previously not ready, now either is ready or
2491  * gone forever.
2492  */
2493 static int qp_wakeup_cb(void *client_data)
2494 {
2495     struct vmci_qp *qpair = (struct vmci_qp *)client_data;
2496 
2497     qp_lock(qpair);
2498     while (qpair->blocked > 0) {
2499         qpair->blocked--;
2500         qpair->generation++;
2501         wake_up(&qpair->event);
2502     }
2503     qp_unlock(qpair);
2504 
2505     return VMCI_SUCCESS;
2506 }
2507 
2508 /*
2509  * Makes the calling thread wait for the queue pair to become
2510  * ready for host side access.  Returns true when thread is
2511  * woken up after queue pair state change, false otherwise.
2512  */
2513 static bool qp_wait_for_ready_queue(struct vmci_qp *qpair)
2514 {
2515     unsigned int generation;
2516 
2517     qpair->blocked++;
2518     generation = qpair->generation;
2519     qp_unlock(qpair);
2520     wait_event(qpair->event, generation != qpair->generation);
2521     qp_lock(qpair);
2522 
2523     return true;
2524 }
2525 
2526 /*
2527  * Enqueues a given buffer to the produce queue using the provided
2528  * function. As many bytes as possible (space available in the queue)
2529  * are enqueued.  Assumes the queue->mutex has been acquired.  Returns
2530  * VMCI_ERROR_QUEUEPAIR_NOSPACE if no space was available to enqueue
2531  * data, VMCI_ERROR_INVALID_SIZE, if any queue pointer is outside the
2532  * queue (as defined by the queue size), VMCI_ERROR_INVALID_ARGS, if
2533  * an error occured when accessing the buffer,
2534  * VMCI_ERROR_QUEUEPAIR_NOTATTACHED, if the queue pair pages aren't
2535  * available.  Otherwise, the number of bytes written to the queue is
2536  * returned.  Updates the tail pointer of the produce queue.
2537  */
2538 static ssize_t qp_enqueue_locked(struct vmci_queue *produce_q,
2539                  struct vmci_queue *consume_q,
2540                  const u64 produce_q_size,
2541                  struct iov_iter *from)
2542 {
2543     s64 free_space;
2544     u64 tail;
2545     size_t buf_size = iov_iter_count(from);
2546     size_t written;
2547     ssize_t result;
2548 
2549     result = qp_map_queue_headers(produce_q, consume_q);
2550     if (unlikely(result != VMCI_SUCCESS))
2551         return result;
2552 
2553     free_space = vmci_q_header_free_space(produce_q->q_header,
2554                           consume_q->q_header,
2555                           produce_q_size);
2556     if (free_space == 0)
2557         return VMCI_ERROR_QUEUEPAIR_NOSPACE;
2558 
2559     if (free_space < VMCI_SUCCESS)
2560         return (ssize_t) free_space;
2561 
2562     written = (size_t) (free_space > buf_size ? buf_size : free_space);
2563     tail = vmci_q_header_producer_tail(produce_q->q_header);
2564     if (likely(tail + written < produce_q_size)) {
2565         result = qp_memcpy_to_queue_iter(produce_q, tail, from, written);
2566     } else {
2567         /* Tail pointer wraps around. */
2568 
2569         const size_t tmp = (size_t) (produce_q_size - tail);
2570 
2571         result = qp_memcpy_to_queue_iter(produce_q, tail, from, tmp);
2572         if (result >= VMCI_SUCCESS)
2573             result = qp_memcpy_to_queue_iter(produce_q, 0, from,
2574                          written - tmp);
2575     }
2576 
2577     if (result < VMCI_SUCCESS)
2578         return result;
2579 
2580     /*
2581      * This virt_wmb() ensures that data written to the queue
2582      * is observable before the new producer_tail is.
2583      */
2584     virt_wmb();
2585 
2586     vmci_q_header_add_producer_tail(produce_q->q_header, written,
2587                     produce_q_size);
2588     return written;
2589 }
2590 
2591 /*
2592  * Dequeues data (if available) from the given consume queue. Writes data
2593  * to the user provided buffer using the provided function.
2594  * Assumes the queue->mutex has been acquired.
2595  * Results:
2596  * VMCI_ERROR_QUEUEPAIR_NODATA if no data was available to dequeue.
2597  * VMCI_ERROR_INVALID_SIZE, if any queue pointer is outside the queue
2598  * (as defined by the queue size).
2599  * VMCI_ERROR_INVALID_ARGS, if an error occured when accessing the buffer.
2600  * Otherwise the number of bytes dequeued is returned.
2601  * Side effects:
2602  * Updates the head pointer of the consume queue.
2603  */
2604 static ssize_t qp_dequeue_locked(struct vmci_queue *produce_q,
2605                  struct vmci_queue *consume_q,
2606                  const u64 consume_q_size,
2607                  struct iov_iter *to,
2608                  bool update_consumer)
2609 {
2610     size_t buf_size = iov_iter_count(to);
2611     s64 buf_ready;
2612     u64 head;
2613     size_t read;
2614     ssize_t result;
2615 
2616     result = qp_map_queue_headers(produce_q, consume_q);
2617     if (unlikely(result != VMCI_SUCCESS))
2618         return result;
2619 
2620     buf_ready = vmci_q_header_buf_ready(consume_q->q_header,
2621                         produce_q->q_header,
2622                         consume_q_size);
2623     if (buf_ready == 0)
2624         return VMCI_ERROR_QUEUEPAIR_NODATA;
2625 
2626     if (buf_ready < VMCI_SUCCESS)
2627         return (ssize_t) buf_ready;
2628 
2629     /*
2630      * This virt_rmb() ensures that data from the queue will be read
2631      * after we have determined how much is ready to be consumed.
2632      */
2633     virt_rmb();
2634 
2635     read = (size_t) (buf_ready > buf_size ? buf_size : buf_ready);
2636     head = vmci_q_header_consumer_head(produce_q->q_header);
2637     if (likely(head + read < consume_q_size)) {
2638         result = qp_memcpy_from_queue_iter(to, consume_q, head, read);
2639     } else {
2640         /* Head pointer wraps around. */
2641 
2642         const size_t tmp = (size_t) (consume_q_size - head);
2643 
2644         result = qp_memcpy_from_queue_iter(to, consume_q, head, tmp);
2645         if (result >= VMCI_SUCCESS)
2646             result = qp_memcpy_from_queue_iter(to, consume_q, 0,
2647                            read - tmp);
2648 
2649     }
2650 
2651     if (result < VMCI_SUCCESS)
2652         return result;
2653 
2654     if (update_consumer)
2655         vmci_q_header_add_consumer_head(produce_q->q_header,
2656                         read, consume_q_size);
2657 
2658     return read;
2659 }
2660 
2661 /*
2662  * vmci_qpair_alloc() - Allocates a queue pair.
2663  * @qpair:      Pointer for the new vmci_qp struct.
2664  * @handle:     Handle to track the resource.
2665  * @produce_qsize:      Desired size of the producer queue.
2666  * @consume_qsize:      Desired size of the consumer queue.
2667  * @peer:       ContextID of the peer.
2668  * @flags:      VMCI flags.
2669  * @priv_flags: VMCI priviledge flags.
2670  *
2671  * This is the client interface for allocating the memory for a
2672  * vmci_qp structure and then attaching to the underlying
2673  * queue.  If an error occurs allocating the memory for the
2674  * vmci_qp structure no attempt is made to attach.  If an
2675  * error occurs attaching, then the structure is freed.
2676  */
2677 int vmci_qpair_alloc(struct vmci_qp **qpair,
2678              struct vmci_handle *handle,
2679              u64 produce_qsize,
2680              u64 consume_qsize,
2681              u32 peer,
2682              u32 flags,
2683              u32 priv_flags)
2684 {
2685     struct vmci_qp *my_qpair;
2686     int retval;
2687     struct vmci_handle src = VMCI_INVALID_HANDLE;
2688     struct vmci_handle dst = vmci_make_handle(peer, VMCI_INVALID_ID);
2689     enum vmci_route route;
2690     vmci_event_release_cb wakeup_cb;
2691     void *client_data;
2692 
2693     /*
2694      * Restrict the size of a queuepair.  The device already
2695      * enforces a limit on the total amount of memory that can be
2696      * allocated to queuepairs for a guest.  However, we try to
2697      * allocate this memory before we make the queuepair
2698      * allocation hypercall.  On Linux, we allocate each page
2699      * separately, which means rather than fail, the guest will
2700      * thrash while it tries to allocate, and will become
2701      * increasingly unresponsive to the point where it appears to
2702      * be hung.  So we place a limit on the size of an individual
2703      * queuepair here, and leave the device to enforce the
2704      * restriction on total queuepair memory.  (Note that this
2705      * doesn't prevent all cases; a user with only this much
2706      * physical memory could still get into trouble.)  The error
2707      * used by the device is NO_RESOURCES, so use that here too.
2708      */
2709 
2710     if (!QP_SIZES_ARE_VALID(produce_qsize, consume_qsize))
2711         return VMCI_ERROR_NO_RESOURCES;
2712 
2713     retval = vmci_route(&src, &dst, false, &route);
2714     if (retval < VMCI_SUCCESS)
2715         route = vmci_guest_code_active() ?
2716             VMCI_ROUTE_AS_GUEST : VMCI_ROUTE_AS_HOST;
2717 
2718     if (flags & (VMCI_QPFLAG_NONBLOCK | VMCI_QPFLAG_PINNED)) {
2719         pr_devel("NONBLOCK OR PINNED set");
2720         return VMCI_ERROR_INVALID_ARGS;
2721     }
2722 
2723     my_qpair = kzalloc(sizeof(*my_qpair), GFP_KERNEL);
2724     if (!my_qpair)
2725         return VMCI_ERROR_NO_MEM;
2726 
2727     my_qpair->produce_q_size = produce_qsize;
2728     my_qpair->consume_q_size = consume_qsize;
2729     my_qpair->peer = peer;
2730     my_qpair->flags = flags;
2731     my_qpair->priv_flags = priv_flags;
2732 
2733     wakeup_cb = NULL;
2734     client_data = NULL;
2735 
2736     if (VMCI_ROUTE_AS_HOST == route) {
2737         my_qpair->guest_endpoint = false;
2738         if (!(flags & VMCI_QPFLAG_LOCAL)) {
2739             my_qpair->blocked = 0;
2740             my_qpair->generation = 0;
2741             init_waitqueue_head(&my_qpair->event);
2742             wakeup_cb = qp_wakeup_cb;
2743             client_data = (void *)my_qpair;
2744         }
2745     } else {
2746         my_qpair->guest_endpoint = true;
2747     }
2748 
2749     retval = vmci_qp_alloc(handle,
2750                    &my_qpair->produce_q,
2751                    my_qpair->produce_q_size,
2752                    &my_qpair->consume_q,
2753                    my_qpair->consume_q_size,
2754                    my_qpair->peer,
2755                    my_qpair->flags,
2756                    my_qpair->priv_flags,
2757                    my_qpair->guest_endpoint,
2758                    wakeup_cb, client_data);
2759 
2760     if (retval < VMCI_SUCCESS) {
2761         kfree(my_qpair);
2762         return retval;
2763     }
2764 
2765     *qpair = my_qpair;
2766     my_qpair->handle = *handle;
2767 
2768     return retval;
2769 }
2770 EXPORT_SYMBOL_GPL(vmci_qpair_alloc);
2771 
2772 /*
2773  * vmci_qpair_detach() - Detatches the client from a queue pair.
2774  * @qpair:      Reference of a pointer to the qpair struct.
2775  *
2776  * This is the client interface for detaching from a VMCIQPair.
2777  * Note that this routine will free the memory allocated for the
2778  * vmci_qp structure too.
2779  */
2780 int vmci_qpair_detach(struct vmci_qp **qpair)
2781 {
2782     int result;
2783     struct vmci_qp *old_qpair;
2784 
2785     if (!qpair || !(*qpair))
2786         return VMCI_ERROR_INVALID_ARGS;
2787 
2788     old_qpair = *qpair;
2789     result = qp_detatch(old_qpair->handle, old_qpair->guest_endpoint);
2790 
2791     /*
2792      * The guest can fail to detach for a number of reasons, and
2793      * if it does so, it will cleanup the entry (if there is one).
2794      * The host can fail too, but it won't cleanup the entry
2795      * immediately, it will do that later when the context is
2796      * freed.  Either way, we need to release the qpair struct
2797      * here; there isn't much the caller can do, and we don't want
2798      * to leak.
2799      */
2800 
2801     memset(old_qpair, 0, sizeof(*old_qpair));
2802     old_qpair->handle = VMCI_INVALID_HANDLE;
2803     old_qpair->peer = VMCI_INVALID_ID;
2804     kfree(old_qpair);
2805     *qpair = NULL;
2806 
2807     return result;
2808 }
2809 EXPORT_SYMBOL_GPL(vmci_qpair_detach);
2810 
2811 /*
2812  * vmci_qpair_get_produce_indexes() - Retrieves the indexes of the producer.
2813  * @qpair:      Pointer to the queue pair struct.
2814  * @producer_tail:      Reference used for storing producer tail index.
2815  * @consumer_head:      Reference used for storing the consumer head index.
2816  *
2817  * This is the client interface for getting the current indexes of the
2818  * QPair from the point of the view of the caller as the producer.
2819  */
2820 int vmci_qpair_get_produce_indexes(const struct vmci_qp *qpair,
2821                    u64 *producer_tail,
2822                    u64 *consumer_head)
2823 {
2824     struct vmci_queue_header *produce_q_header;
2825     struct vmci_queue_header *consume_q_header;
2826     int result;
2827 
2828     if (!qpair)
2829         return VMCI_ERROR_INVALID_ARGS;
2830 
2831     qp_lock(qpair);
2832     result =
2833         qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header);
2834     if (result == VMCI_SUCCESS)
2835         vmci_q_header_get_pointers(produce_q_header, consume_q_header,
2836                        producer_tail, consumer_head);
2837     qp_unlock(qpair);
2838 
2839     if (result == VMCI_SUCCESS &&
2840         ((producer_tail && *producer_tail >= qpair->produce_q_size) ||
2841          (consumer_head && *consumer_head >= qpair->produce_q_size)))
2842         return VMCI_ERROR_INVALID_SIZE;
2843 
2844     return result;
2845 }
2846 EXPORT_SYMBOL_GPL(vmci_qpair_get_produce_indexes);
2847 
2848 /*
2849  * vmci_qpair_get_consume_indexes() - Retrieves the indexes of the consumer.
2850  * @qpair:      Pointer to the queue pair struct.
2851  * @consumer_tail:      Reference used for storing consumer tail index.
2852  * @producer_head:      Reference used for storing the producer head index.
2853  *
2854  * This is the client interface for getting the current indexes of the
2855  * QPair from the point of the view of the caller as the consumer.
2856  */
2857 int vmci_qpair_get_consume_indexes(const struct vmci_qp *qpair,
2858                    u64 *consumer_tail,
2859                    u64 *producer_head)
2860 {
2861     struct vmci_queue_header *produce_q_header;
2862     struct vmci_queue_header *consume_q_header;
2863     int result;
2864 
2865     if (!qpair)
2866         return VMCI_ERROR_INVALID_ARGS;
2867 
2868     qp_lock(qpair);
2869     result =
2870         qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header);
2871     if (result == VMCI_SUCCESS)
2872         vmci_q_header_get_pointers(consume_q_header, produce_q_header,
2873                        consumer_tail, producer_head);
2874     qp_unlock(qpair);
2875 
2876     if (result == VMCI_SUCCESS &&
2877         ((consumer_tail && *consumer_tail >= qpair->consume_q_size) ||
2878          (producer_head && *producer_head >= qpair->consume_q_size)))
2879         return VMCI_ERROR_INVALID_SIZE;
2880 
2881     return result;
2882 }
2883 EXPORT_SYMBOL_GPL(vmci_qpair_get_consume_indexes);
2884 
2885 /*
2886  * vmci_qpair_produce_free_space() - Retrieves free space in producer queue.
2887  * @qpair:      Pointer to the queue pair struct.
2888  *
2889  * This is the client interface for getting the amount of free
2890  * space in the QPair from the point of the view of the caller as
2891  * the producer which is the common case.  Returns < 0 if err, else
2892  * available bytes into which data can be enqueued if > 0.
2893  */
2894 s64 vmci_qpair_produce_free_space(const struct vmci_qp *qpair)
2895 {
2896     struct vmci_queue_header *produce_q_header;
2897     struct vmci_queue_header *consume_q_header;
2898     s64 result;
2899 
2900     if (!qpair)
2901         return VMCI_ERROR_INVALID_ARGS;
2902 
2903     qp_lock(qpair);
2904     result =
2905         qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header);
2906     if (result == VMCI_SUCCESS)
2907         result = vmci_q_header_free_space(produce_q_header,
2908                           consume_q_header,
2909                           qpair->produce_q_size);
2910     else
2911         result = 0;
2912 
2913     qp_unlock(qpair);
2914 
2915     return result;
2916 }
2917 EXPORT_SYMBOL_GPL(vmci_qpair_produce_free_space);
2918 
2919 /*
2920  * vmci_qpair_consume_free_space() - Retrieves free space in consumer queue.
2921  * @qpair:      Pointer to the queue pair struct.
2922  *
2923  * This is the client interface for getting the amount of free
2924  * space in the QPair from the point of the view of the caller as
2925  * the consumer which is not the common case.  Returns < 0 if err, else
2926  * available bytes into which data can be enqueued if > 0.
2927  */
2928 s64 vmci_qpair_consume_free_space(const struct vmci_qp *qpair)
2929 {
2930     struct vmci_queue_header *produce_q_header;
2931     struct vmci_queue_header *consume_q_header;
2932     s64 result;
2933 
2934     if (!qpair)
2935         return VMCI_ERROR_INVALID_ARGS;
2936 
2937     qp_lock(qpair);
2938     result =
2939         qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header);
2940     if (result == VMCI_SUCCESS)
2941         result = vmci_q_header_free_space(consume_q_header,
2942                           produce_q_header,
2943                           qpair->consume_q_size);
2944     else
2945         result = 0;
2946 
2947     qp_unlock(qpair);
2948 
2949     return result;
2950 }
2951 EXPORT_SYMBOL_GPL(vmci_qpair_consume_free_space);
2952 
2953 /*
2954  * vmci_qpair_produce_buf_ready() - Gets bytes ready to read from
2955  * producer queue.
2956  * @qpair:      Pointer to the queue pair struct.
2957  *
2958  * This is the client interface for getting the amount of
2959  * enqueued data in the QPair from the point of the view of the
2960  * caller as the producer which is not the common case.  Returns < 0 if err,
2961  * else available bytes that may be read.
2962  */
2963 s64 vmci_qpair_produce_buf_ready(const struct vmci_qp *qpair)
2964 {
2965     struct vmci_queue_header *produce_q_header;
2966     struct vmci_queue_header *consume_q_header;
2967     s64 result;
2968 
2969     if (!qpair)
2970         return VMCI_ERROR_INVALID_ARGS;
2971 
2972     qp_lock(qpair);
2973     result =
2974         qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header);
2975     if (result == VMCI_SUCCESS)
2976         result = vmci_q_header_buf_ready(produce_q_header,
2977                          consume_q_header,
2978                          qpair->produce_q_size);
2979     else
2980         result = 0;
2981 
2982     qp_unlock(qpair);
2983 
2984     return result;
2985 }
2986 EXPORT_SYMBOL_GPL(vmci_qpair_produce_buf_ready);
2987 
2988 /*
2989  * vmci_qpair_consume_buf_ready() - Gets bytes ready to read from
2990  * consumer queue.
2991  * @qpair:      Pointer to the queue pair struct.
2992  *
2993  * This is the client interface for getting the amount of
2994  * enqueued data in the QPair from the point of the view of the
2995  * caller as the consumer which is the normal case.  Returns < 0 if err,
2996  * else available bytes that may be read.
2997  */
2998 s64 vmci_qpair_consume_buf_ready(const struct vmci_qp *qpair)
2999 {
3000     struct vmci_queue_header *produce_q_header;
3001     struct vmci_queue_header *consume_q_header;
3002     s64 result;
3003 
3004     if (!qpair)
3005         return VMCI_ERROR_INVALID_ARGS;
3006 
3007     qp_lock(qpair);
3008     result =
3009         qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header);
3010     if (result == VMCI_SUCCESS)
3011         result = vmci_q_header_buf_ready(consume_q_header,
3012                          produce_q_header,
3013                          qpair->consume_q_size);
3014     else
3015         result = 0;
3016 
3017     qp_unlock(qpair);
3018 
3019     return result;
3020 }
3021 EXPORT_SYMBOL_GPL(vmci_qpair_consume_buf_ready);
3022 
3023 /*
3024  * vmci_qpair_enqueue() - Throw data on the queue.
3025  * @qpair:      Pointer to the queue pair struct.
3026  * @buf:        Pointer to buffer containing data
3027  * @buf_size:   Length of buffer.
3028  * @buf_type:   Buffer type (Unused).
3029  *
3030  * This is the client interface for enqueueing data into the queue.
3031  * Returns number of bytes enqueued or < 0 on error.
3032  */
3033 ssize_t vmci_qpair_enqueue(struct vmci_qp *qpair,
3034                const void *buf,
3035                size_t buf_size,
3036                int buf_type)
3037 {
3038     ssize_t result;
3039     struct iov_iter from;
3040     struct kvec v = {.iov_base = (void *)buf, .iov_len = buf_size};
3041 
3042     if (!qpair || !buf)
3043         return VMCI_ERROR_INVALID_ARGS;
3044 
3045     iov_iter_kvec(&from, WRITE, &v, 1, buf_size);
3046 
3047     qp_lock(qpair);
3048 
3049     do {
3050         result = qp_enqueue_locked(qpair->produce_q,
3051                        qpair->consume_q,
3052                        qpair->produce_q_size,
3053                        &from);
3054 
3055         if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY &&
3056             !qp_wait_for_ready_queue(qpair))
3057             result = VMCI_ERROR_WOULD_BLOCK;
3058 
3059     } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY);
3060 
3061     qp_unlock(qpair);
3062 
3063     return result;
3064 }
3065 EXPORT_SYMBOL_GPL(vmci_qpair_enqueue);
3066 
3067 /*
3068  * vmci_qpair_dequeue() - Get data from the queue.
3069  * @qpair:      Pointer to the queue pair struct.
3070  * @buf:        Pointer to buffer for the data
3071  * @buf_size:   Length of buffer.
3072  * @buf_type:   Buffer type (Unused).
3073  *
3074  * This is the client interface for dequeueing data from the queue.
3075  * Returns number of bytes dequeued or < 0 on error.
3076  */
3077 ssize_t vmci_qpair_dequeue(struct vmci_qp *qpair,
3078                void *buf,
3079                size_t buf_size,
3080                int buf_type)
3081 {
3082     ssize_t result;
3083     struct iov_iter to;
3084     struct kvec v = {.iov_base = buf, .iov_len = buf_size};
3085 
3086     if (!qpair || !buf)
3087         return VMCI_ERROR_INVALID_ARGS;
3088 
3089     iov_iter_kvec(&to, READ, &v, 1, buf_size);
3090 
3091     qp_lock(qpair);
3092 
3093     do {
3094         result = qp_dequeue_locked(qpair->produce_q,
3095                        qpair->consume_q,
3096                        qpair->consume_q_size,
3097                        &to, true);
3098 
3099         if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY &&
3100             !qp_wait_for_ready_queue(qpair))
3101             result = VMCI_ERROR_WOULD_BLOCK;
3102 
3103     } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY);
3104 
3105     qp_unlock(qpair);
3106 
3107     return result;
3108 }
3109 EXPORT_SYMBOL_GPL(vmci_qpair_dequeue);
3110 
3111 /*
3112  * vmci_qpair_peek() - Peek at the data in the queue.
3113  * @qpair:      Pointer to the queue pair struct.
3114  * @buf:        Pointer to buffer for the data
3115  * @buf_size:   Length of buffer.
3116  * @buf_type:   Buffer type (Unused on Linux).
3117  *
3118  * This is the client interface for peeking into a queue.  (I.e.,
3119  * copy data from the queue without updating the head pointer.)
3120  * Returns number of bytes dequeued or < 0 on error.
3121  */
3122 ssize_t vmci_qpair_peek(struct vmci_qp *qpair,
3123             void *buf,
3124             size_t buf_size,
3125             int buf_type)
3126 {
3127     struct iov_iter to;
3128     struct kvec v = {.iov_base = buf, .iov_len = buf_size};
3129     ssize_t result;
3130 
3131     if (!qpair || !buf)
3132         return VMCI_ERROR_INVALID_ARGS;
3133 
3134     iov_iter_kvec(&to, READ, &v, 1, buf_size);
3135 
3136     qp_lock(qpair);
3137 
3138     do {
3139         result = qp_dequeue_locked(qpair->produce_q,
3140                        qpair->consume_q,
3141                        qpair->consume_q_size,
3142                        &to, false);
3143 
3144         if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY &&
3145             !qp_wait_for_ready_queue(qpair))
3146             result = VMCI_ERROR_WOULD_BLOCK;
3147 
3148     } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY);
3149 
3150     qp_unlock(qpair);
3151 
3152     return result;
3153 }
3154 EXPORT_SYMBOL_GPL(vmci_qpair_peek);
3155 
3156 /*
3157  * vmci_qpair_enquev() - Throw data on the queue using iov.
3158  * @qpair:      Pointer to the queue pair struct.
3159  * @iov:        Pointer to buffer containing data
3160  * @iov_size:   Length of buffer.
3161  * @buf_type:   Buffer type (Unused).
3162  *
3163  * This is the client interface for enqueueing data into the queue.
3164  * This function uses IO vectors to handle the work. Returns number
3165  * of bytes enqueued or < 0 on error.
3166  */
3167 ssize_t vmci_qpair_enquev(struct vmci_qp *qpair,
3168               struct msghdr *msg,
3169               size_t iov_size,
3170               int buf_type)
3171 {
3172     ssize_t result;
3173 
3174     if (!qpair)
3175         return VMCI_ERROR_INVALID_ARGS;
3176 
3177     qp_lock(qpair);
3178 
3179     do {
3180         result = qp_enqueue_locked(qpair->produce_q,
3181                        qpair->consume_q,
3182                        qpair->produce_q_size,
3183                        &msg->msg_iter);
3184 
3185         if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY &&
3186             !qp_wait_for_ready_queue(qpair))
3187             result = VMCI_ERROR_WOULD_BLOCK;
3188 
3189     } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY);
3190 
3191     qp_unlock(qpair);
3192 
3193     return result;
3194 }
3195 EXPORT_SYMBOL_GPL(vmci_qpair_enquev);
3196 
3197 /*
3198  * vmci_qpair_dequev() - Get data from the queue using iov.
3199  * @qpair:      Pointer to the queue pair struct.
3200  * @iov:        Pointer to buffer for the data
3201  * @iov_size:   Length of buffer.
3202  * @buf_type:   Buffer type (Unused).
3203  *
3204  * This is the client interface for dequeueing data from the queue.
3205  * This function uses IO vectors to handle the work. Returns number
3206  * of bytes dequeued or < 0 on error.
3207  */
3208 ssize_t vmci_qpair_dequev(struct vmci_qp *qpair,
3209               struct msghdr *msg,
3210               size_t iov_size,
3211               int buf_type)
3212 {
3213     ssize_t result;
3214 
3215     if (!qpair)
3216         return VMCI_ERROR_INVALID_ARGS;
3217 
3218     qp_lock(qpair);
3219 
3220     do {
3221         result = qp_dequeue_locked(qpair->produce_q,
3222                        qpair->consume_q,
3223                        qpair->consume_q_size,
3224                        &msg->msg_iter, true);
3225 
3226         if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY &&
3227             !qp_wait_for_ready_queue(qpair))
3228             result = VMCI_ERROR_WOULD_BLOCK;
3229 
3230     } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY);
3231 
3232     qp_unlock(qpair);
3233 
3234     return result;
3235 }
3236 EXPORT_SYMBOL_GPL(vmci_qpair_dequev);
3237 
3238 /*
3239  * vmci_qpair_peekv() - Peek at the data in the queue using iov.
3240  * @qpair:      Pointer to the queue pair struct.
3241  * @iov:        Pointer to buffer for the data
3242  * @iov_size:   Length of buffer.
3243  * @buf_type:   Buffer type (Unused on Linux).
3244  *
3245  * This is the client interface for peeking into a queue.  (I.e.,
3246  * copy data from the queue without updating the head pointer.)
3247  * This function uses IO vectors to handle the work. Returns number
3248  * of bytes peeked or < 0 on error.
3249  */
3250 ssize_t vmci_qpair_peekv(struct vmci_qp *qpair,
3251              struct msghdr *msg,
3252              size_t iov_size,
3253              int buf_type)
3254 {
3255     ssize_t result;
3256 
3257     if (!qpair)
3258         return VMCI_ERROR_INVALID_ARGS;
3259 
3260     qp_lock(qpair);
3261 
3262     do {
3263         result = qp_dequeue_locked(qpair->produce_q,
3264                        qpair->consume_q,
3265                        qpair->consume_q_size,
3266                        &msg->msg_iter, false);
3267 
3268         if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY &&
3269             !qp_wait_for_ready_queue(qpair))
3270             result = VMCI_ERROR_WOULD_BLOCK;
3271 
3272     } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY);
3273 
3274     qp_unlock(qpair);
3275     return result;
3276 }
3277 EXPORT_SYMBOL_GPL(vmci_qpair_peekv);