1685a6bf8SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 206164d2bSGeorge Zhang /* 306164d2bSGeorge Zhang * VMware VMCI Driver 406164d2bSGeorge Zhang * 506164d2bSGeorge Zhang * Copyright (C) 2012 VMware, Inc. All rights reserved. 606164d2bSGeorge Zhang */ 706164d2bSGeorge Zhang 806164d2bSGeorge Zhang #include <linux/vmw_vmci_defs.h> 906164d2bSGeorge Zhang #include <linux/vmw_vmci_api.h> 1042281d20SAndy King #include <linux/highmem.h> 1106164d2bSGeorge Zhang #include <linux/kernel.h> 1242281d20SAndy King #include <linux/mm.h> 1306164d2bSGeorge Zhang #include <linux/module.h> 1406164d2bSGeorge Zhang #include <linux/mutex.h> 1542281d20SAndy King #include <linux/pagemap.h> 166d6dfb4fSAndy King #include <linux/pci.h> 1742281d20SAndy King #include <linux/sched.h> 1842281d20SAndy King #include <linux/slab.h> 19d2f83e90SRusty Russell #include <linux/uio.h> 2006164d2bSGeorge Zhang #include <linux/wait.h> 21f6dcf8e7SDavid Rientjes #include <linux/vmalloc.h> 22d838df2eSAl Viro #include <linux/skbuff.h> 2306164d2bSGeorge Zhang 2406164d2bSGeorge Zhang #include "vmci_handle_array.h" 2506164d2bSGeorge Zhang #include "vmci_queue_pair.h" 2606164d2bSGeorge Zhang #include "vmci_datagram.h" 2706164d2bSGeorge Zhang #include "vmci_resource.h" 2806164d2bSGeorge Zhang #include "vmci_context.h" 2906164d2bSGeorge Zhang #include "vmci_driver.h" 3006164d2bSGeorge Zhang #include "vmci_event.h" 3106164d2bSGeorge Zhang #include "vmci_route.h" 3206164d2bSGeorge Zhang 3306164d2bSGeorge Zhang /* 3406164d2bSGeorge Zhang * In the following, we will distinguish between two kinds of VMX processes - 3506164d2bSGeorge Zhang * the ones with versions lower than VMCI_VERSION_NOVMVM that use specialized 3606164d2bSGeorge Zhang * VMCI page files in the VMX and supporting VM to VM communication and the 3706164d2bSGeorge Zhang * newer ones that use the guest memory directly. We will in the following 3806164d2bSGeorge Zhang * refer to the older VMX versions as old-style VMX'en, and the newer ones as 3906164d2bSGeorge Zhang * new-style VMX'en. 4006164d2bSGeorge Zhang * 4106164d2bSGeorge Zhang * The state transition datagram is as follows (the VMCIQPB_ prefix has been 4206164d2bSGeorge Zhang * removed for readability) - see below for more details on the transtions: 4306164d2bSGeorge Zhang * 4406164d2bSGeorge Zhang * -------------- NEW ------------- 4506164d2bSGeorge Zhang * | | 4606164d2bSGeorge Zhang * \_/ \_/ 4706164d2bSGeorge Zhang * CREATED_NO_MEM <-----------------> CREATED_MEM 4806164d2bSGeorge Zhang * | | | 4906164d2bSGeorge Zhang * | o-----------------------o | 5006164d2bSGeorge Zhang * | | | 5106164d2bSGeorge Zhang * \_/ \_/ \_/ 5206164d2bSGeorge Zhang * ATTACHED_NO_MEM <----------------> ATTACHED_MEM 5306164d2bSGeorge Zhang * | | | 5406164d2bSGeorge Zhang * | o----------------------o | 5506164d2bSGeorge Zhang * | | | 5606164d2bSGeorge Zhang * \_/ \_/ \_/ 5706164d2bSGeorge Zhang * SHUTDOWN_NO_MEM <----------------> SHUTDOWN_MEM 5806164d2bSGeorge Zhang * | | 5906164d2bSGeorge Zhang * | | 6006164d2bSGeorge Zhang * -------------> gone <------------- 6106164d2bSGeorge Zhang * 6206164d2bSGeorge Zhang * In more detail. When a VMCI queue pair is first created, it will be in the 6306164d2bSGeorge Zhang * VMCIQPB_NEW state. It will then move into one of the following states: 6406164d2bSGeorge Zhang * 6506164d2bSGeorge Zhang * - VMCIQPB_CREATED_NO_MEM: this state indicates that either: 6606164d2bSGeorge Zhang * 6706164d2bSGeorge Zhang * - the created was performed by a host endpoint, in which case there is 6806164d2bSGeorge Zhang * no backing memory yet. 6906164d2bSGeorge Zhang * 7006164d2bSGeorge Zhang * - the create was initiated by an old-style VMX, that uses 7106164d2bSGeorge Zhang * vmci_qp_broker_set_page_store to specify the UVAs of the queue pair at 7206164d2bSGeorge Zhang * a later point in time. This state can be distinguished from the one 7306164d2bSGeorge Zhang * above by the context ID of the creator. A host side is not allowed to 7406164d2bSGeorge Zhang * attach until the page store has been set. 7506164d2bSGeorge Zhang * 7606164d2bSGeorge Zhang * - VMCIQPB_CREATED_MEM: this state is the result when the queue pair 7706164d2bSGeorge Zhang * is created by a VMX using the queue pair device backend that 7806164d2bSGeorge Zhang * sets the UVAs of the queue pair immediately and stores the 7906164d2bSGeorge Zhang * information for later attachers. At this point, it is ready for 8006164d2bSGeorge Zhang * the host side to attach to it. 8106164d2bSGeorge Zhang * 8206164d2bSGeorge Zhang * Once the queue pair is in one of the created states (with the exception of 8306164d2bSGeorge Zhang * the case mentioned for older VMX'en above), it is possible to attach to the 8406164d2bSGeorge Zhang * queue pair. Again we have two new states possible: 8506164d2bSGeorge Zhang * 8606164d2bSGeorge Zhang * - VMCIQPB_ATTACHED_MEM: this state can be reached through the following 8706164d2bSGeorge Zhang * paths: 8806164d2bSGeorge Zhang * 8906164d2bSGeorge Zhang * - from VMCIQPB_CREATED_NO_MEM when a new-style VMX allocates a queue 9006164d2bSGeorge Zhang * pair, and attaches to a queue pair previously created by the host side. 9106164d2bSGeorge Zhang * 9206164d2bSGeorge Zhang * - from VMCIQPB_CREATED_MEM when the host side attaches to a queue pair 9306164d2bSGeorge Zhang * already created by a guest. 9406164d2bSGeorge Zhang * 9506164d2bSGeorge Zhang * - from VMCIQPB_ATTACHED_NO_MEM, when an old-style VMX calls 9606164d2bSGeorge Zhang * vmci_qp_broker_set_page_store (see below). 9706164d2bSGeorge Zhang * 9806164d2bSGeorge Zhang * - VMCIQPB_ATTACHED_NO_MEM: If the queue pair already was in the 9906164d2bSGeorge Zhang * VMCIQPB_CREATED_NO_MEM due to a host side create, an old-style VMX will 10006164d2bSGeorge Zhang * bring the queue pair into this state. Once vmci_qp_broker_set_page_store 10106164d2bSGeorge Zhang * is called to register the user memory, the VMCIQPB_ATTACH_MEM state 10206164d2bSGeorge Zhang * will be entered. 10306164d2bSGeorge Zhang * 10406164d2bSGeorge Zhang * From the attached queue pair, the queue pair can enter the shutdown states 10506164d2bSGeorge Zhang * when either side of the queue pair detaches. If the guest side detaches 10606164d2bSGeorge Zhang * first, the queue pair will enter the VMCIQPB_SHUTDOWN_NO_MEM state, where 10706164d2bSGeorge Zhang * the content of the queue pair will no longer be available. If the host 10806164d2bSGeorge Zhang * side detaches first, the queue pair will either enter the 10906164d2bSGeorge Zhang * VMCIQPB_SHUTDOWN_MEM, if the guest memory is currently mapped, or 11006164d2bSGeorge Zhang * VMCIQPB_SHUTDOWN_NO_MEM, if the guest memory is not mapped 11106164d2bSGeorge Zhang * (e.g., the host detaches while a guest is stunned). 11206164d2bSGeorge Zhang * 11306164d2bSGeorge Zhang * New-style VMX'en will also unmap guest memory, if the guest is 11406164d2bSGeorge Zhang * quiesced, e.g., during a snapshot operation. In that case, the guest 11506164d2bSGeorge Zhang * memory will no longer be available, and the queue pair will transition from 11606164d2bSGeorge Zhang * *_MEM state to a *_NO_MEM state. The VMX may later map the memory once more, 11706164d2bSGeorge Zhang * in which case the queue pair will transition from the *_NO_MEM state at that 11806164d2bSGeorge Zhang * point back to the *_MEM state. Note that the *_NO_MEM state may have changed, 11906164d2bSGeorge Zhang * since the peer may have either attached or detached in the meantime. The 12006164d2bSGeorge Zhang * values are laid out such that ++ on a state will move from a *_NO_MEM to a 12106164d2bSGeorge Zhang * *_MEM state, and vice versa. 12206164d2bSGeorge Zhang */ 12306164d2bSGeorge Zhang 12406164d2bSGeorge Zhang /* The Kernel specific component of the struct vmci_queue structure. */ 12506164d2bSGeorge Zhang struct vmci_queue_kern_if { 12606164d2bSGeorge Zhang struct mutex __mutex; /* Protects the queue. */ 12706164d2bSGeorge Zhang struct mutex *mutex; /* Shared by producer and consumer queues. */ 1286d6dfb4fSAndy King size_t num_pages; /* Number of pages incl. header. */ 1296d6dfb4fSAndy King bool host; /* Host or guest? */ 1306d6dfb4fSAndy King union { 1316d6dfb4fSAndy King struct { 1326d6dfb4fSAndy King dma_addr_t *pas; 1336d6dfb4fSAndy King void **vas; 1346d6dfb4fSAndy King } g; /* Used by the guest. */ 1356d6dfb4fSAndy King struct { 1366d6dfb4fSAndy King struct page **page; 1376d6dfb4fSAndy King struct page **header_page; 1386d6dfb4fSAndy King } h; /* Used by the host. */ 1396d6dfb4fSAndy King } u; 14006164d2bSGeorge Zhang }; 14106164d2bSGeorge Zhang 14206164d2bSGeorge Zhang /* 14306164d2bSGeorge Zhang * This structure is opaque to the clients. 14406164d2bSGeorge Zhang */ 14506164d2bSGeorge Zhang struct vmci_qp { 14606164d2bSGeorge Zhang struct vmci_handle handle; 14706164d2bSGeorge Zhang struct vmci_queue *produce_q; 14806164d2bSGeorge Zhang struct vmci_queue *consume_q; 14906164d2bSGeorge Zhang u64 produce_q_size; 15006164d2bSGeorge Zhang u64 consume_q_size; 15106164d2bSGeorge Zhang u32 peer; 15206164d2bSGeorge Zhang u32 flags; 15306164d2bSGeorge Zhang u32 priv_flags; 15406164d2bSGeorge Zhang bool guest_endpoint; 15506164d2bSGeorge Zhang unsigned int blocked; 15606164d2bSGeorge Zhang unsigned int generation; 15706164d2bSGeorge Zhang wait_queue_head_t event; 15806164d2bSGeorge Zhang }; 15906164d2bSGeorge Zhang 16006164d2bSGeorge Zhang enum qp_broker_state { 16106164d2bSGeorge Zhang VMCIQPB_NEW, 16206164d2bSGeorge Zhang VMCIQPB_CREATED_NO_MEM, 16306164d2bSGeorge Zhang VMCIQPB_CREATED_MEM, 16406164d2bSGeorge Zhang VMCIQPB_ATTACHED_NO_MEM, 16506164d2bSGeorge Zhang VMCIQPB_ATTACHED_MEM, 16606164d2bSGeorge Zhang VMCIQPB_SHUTDOWN_NO_MEM, 16706164d2bSGeorge Zhang VMCIQPB_SHUTDOWN_MEM, 16806164d2bSGeorge Zhang VMCIQPB_GONE 16906164d2bSGeorge Zhang }; 17006164d2bSGeorge Zhang 17106164d2bSGeorge Zhang #define QPBROKERSTATE_HAS_MEM(_qpb) (_qpb->state == VMCIQPB_CREATED_MEM || \ 17206164d2bSGeorge Zhang _qpb->state == VMCIQPB_ATTACHED_MEM || \ 17306164d2bSGeorge Zhang _qpb->state == VMCIQPB_SHUTDOWN_MEM) 17406164d2bSGeorge Zhang 17506164d2bSGeorge Zhang /* 17606164d2bSGeorge Zhang * In the queue pair broker, we always use the guest point of view for 17706164d2bSGeorge Zhang * the produce and consume queue values and references, e.g., the 17806164d2bSGeorge Zhang * produce queue size stored is the guests produce queue size. The 17906164d2bSGeorge Zhang * host endpoint will need to swap these around. The only exception is 18006164d2bSGeorge Zhang * the local queue pairs on the host, in which case the host endpoint 18106164d2bSGeorge Zhang * that creates the queue pair will have the right orientation, and 18206164d2bSGeorge Zhang * the attaching host endpoint will need to swap. 18306164d2bSGeorge Zhang */ 18406164d2bSGeorge Zhang struct qp_entry { 18506164d2bSGeorge Zhang struct list_head list_item; 18606164d2bSGeorge Zhang struct vmci_handle handle; 18706164d2bSGeorge Zhang u32 peer; 18806164d2bSGeorge Zhang u32 flags; 18906164d2bSGeorge Zhang u64 produce_size; 19006164d2bSGeorge Zhang u64 consume_size; 19106164d2bSGeorge Zhang u32 ref_count; 19206164d2bSGeorge Zhang }; 19306164d2bSGeorge Zhang 19406164d2bSGeorge Zhang struct qp_broker_entry { 19506164d2bSGeorge Zhang struct vmci_resource resource; 19606164d2bSGeorge Zhang struct qp_entry qp; 19706164d2bSGeorge Zhang u32 create_id; 19806164d2bSGeorge Zhang u32 attach_id; 19906164d2bSGeorge Zhang enum qp_broker_state state; 20006164d2bSGeorge Zhang bool require_trusted_attach; 20106164d2bSGeorge Zhang bool created_by_trusted; 20206164d2bSGeorge Zhang bool vmci_page_files; /* Created by VMX using VMCI page files */ 20306164d2bSGeorge Zhang struct vmci_queue *produce_q; 20406164d2bSGeorge Zhang struct vmci_queue *consume_q; 20506164d2bSGeorge Zhang struct vmci_queue_header saved_produce_q; 20606164d2bSGeorge Zhang struct vmci_queue_header saved_consume_q; 20706164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb; 20806164d2bSGeorge Zhang void *client_data; 20906164d2bSGeorge Zhang void *local_mem; /* Kernel memory for local queue pair */ 21006164d2bSGeorge Zhang }; 21106164d2bSGeorge Zhang 21206164d2bSGeorge Zhang struct qp_guest_endpoint { 21306164d2bSGeorge Zhang struct vmci_resource resource; 21406164d2bSGeorge Zhang struct qp_entry qp; 21506164d2bSGeorge Zhang u64 num_ppns; 21606164d2bSGeorge Zhang void *produce_q; 21706164d2bSGeorge Zhang void *consume_q; 218e6389a13SDmitry Torokhov struct ppn_set ppn_set; 21906164d2bSGeorge Zhang }; 22006164d2bSGeorge Zhang 22106164d2bSGeorge Zhang struct qp_list { 22206164d2bSGeorge Zhang struct list_head head; 22306164d2bSGeorge Zhang struct mutex mutex; /* Protect queue list. */ 22406164d2bSGeorge Zhang }; 22506164d2bSGeorge Zhang 22606164d2bSGeorge Zhang static struct qp_list qp_broker_list = { 22706164d2bSGeorge Zhang .head = LIST_HEAD_INIT(qp_broker_list.head), 22806164d2bSGeorge Zhang .mutex = __MUTEX_INITIALIZER(qp_broker_list.mutex), 22906164d2bSGeorge Zhang }; 23006164d2bSGeorge Zhang 23106164d2bSGeorge Zhang static struct qp_list qp_guest_endpoints = { 23206164d2bSGeorge Zhang .head = LIST_HEAD_INIT(qp_guest_endpoints.head), 23306164d2bSGeorge Zhang .mutex = __MUTEX_INITIALIZER(qp_guest_endpoints.mutex), 23406164d2bSGeorge Zhang }; 23506164d2bSGeorge Zhang 23606164d2bSGeorge Zhang #define INVALID_VMCI_GUEST_MEM_ID 0 23706164d2bSGeorge Zhang #define QPE_NUM_PAGES(_QPE) ((u32) \ 23842281d20SAndy King (DIV_ROUND_UP(_QPE.produce_size, PAGE_SIZE) + \ 23942281d20SAndy King DIV_ROUND_UP(_QPE.consume_size, PAGE_SIZE) + 2)) 24006164d2bSGeorge Zhang 24106164d2bSGeorge Zhang 24206164d2bSGeorge Zhang /* 24306164d2bSGeorge Zhang * Frees kernel VA space for a given queue and its queue header, and 24406164d2bSGeorge Zhang * frees physical data pages. 24506164d2bSGeorge Zhang */ 24606164d2bSGeorge Zhang static void qp_free_queue(void *q, u64 size) 24706164d2bSGeorge Zhang { 24806164d2bSGeorge Zhang struct vmci_queue *queue = q; 24906164d2bSGeorge Zhang 25006164d2bSGeorge Zhang if (queue) { 2516d6dfb4fSAndy King u64 i; 25206164d2bSGeorge Zhang 2536d6dfb4fSAndy King /* Given size does not include header, so add in a page here. */ 2546d6dfb4fSAndy King for (i = 0; i < DIV_ROUND_UP(size, PAGE_SIZE) + 1; i++) { 2556d6dfb4fSAndy King dma_free_coherent(&vmci_pdev->dev, PAGE_SIZE, 2566d6dfb4fSAndy King queue->kernel_if->u.g.vas[i], 2576d6dfb4fSAndy King queue->kernel_if->u.g.pas[i]); 2586d6dfb4fSAndy King } 25906164d2bSGeorge Zhang 2606d6dfb4fSAndy King vfree(queue); 26106164d2bSGeorge Zhang } 26206164d2bSGeorge Zhang } 26306164d2bSGeorge Zhang 26406164d2bSGeorge Zhang /* 2656d6dfb4fSAndy King * Allocates kernel queue pages of specified size with IOMMU mappings, 2666d6dfb4fSAndy King * plus space for the queue structure/kernel interface and the queue 2676d6dfb4fSAndy King * header. 26806164d2bSGeorge Zhang */ 26906164d2bSGeorge Zhang static void *qp_alloc_queue(u64 size, u32 flags) 27006164d2bSGeorge Zhang { 27106164d2bSGeorge Zhang u64 i; 27206164d2bSGeorge Zhang struct vmci_queue *queue; 273aa6467f1SJorgen Hansen size_t pas_size; 274aa6467f1SJorgen Hansen size_t vas_size; 275aa6467f1SJorgen Hansen size_t queue_size = sizeof(*queue) + sizeof(*queue->kernel_if); 276146180c0SDan Carpenter u64 num_pages; 277aa6467f1SJorgen Hansen 278146180c0SDan Carpenter if (size > SIZE_MAX - PAGE_SIZE) 279146180c0SDan Carpenter return NULL; 280146180c0SDan Carpenter num_pages = DIV_ROUND_UP(size, PAGE_SIZE) + 1; 281aa6467f1SJorgen Hansen if (num_pages > 282aa6467f1SJorgen Hansen (SIZE_MAX - queue_size) / 283aa6467f1SJorgen Hansen (sizeof(*queue->kernel_if->u.g.pas) + 284aa6467f1SJorgen Hansen sizeof(*queue->kernel_if->u.g.vas))) 285aa6467f1SJorgen Hansen return NULL; 286aa6467f1SJorgen Hansen 287aa6467f1SJorgen Hansen pas_size = num_pages * sizeof(*queue->kernel_if->u.g.pas); 288aa6467f1SJorgen Hansen vas_size = num_pages * sizeof(*queue->kernel_if->u.g.vas); 289aa6467f1SJorgen Hansen queue_size += pas_size + vas_size; 29006164d2bSGeorge Zhang 2916d6dfb4fSAndy King queue = vmalloc(queue_size); 2926d6dfb4fSAndy King if (!queue) 29306164d2bSGeorge Zhang return NULL; 29406164d2bSGeorge Zhang 2956d6dfb4fSAndy King queue->q_header = NULL; 29606164d2bSGeorge Zhang queue->saved_header = NULL; 29706164d2bSGeorge Zhang queue->kernel_if = (struct vmci_queue_kern_if *)(queue + 1); 2986d6dfb4fSAndy King queue->kernel_if->mutex = NULL; 2996d6dfb4fSAndy King queue->kernel_if->num_pages = num_pages; 3006d6dfb4fSAndy King queue->kernel_if->u.g.pas = (dma_addr_t *)(queue->kernel_if + 1); 3016d6dfb4fSAndy King queue->kernel_if->u.g.vas = 3026d6dfb4fSAndy King (void **)((u8 *)queue->kernel_if->u.g.pas + pas_size); 30306164d2bSGeorge Zhang queue->kernel_if->host = false; 30406164d2bSGeorge Zhang 3056d6dfb4fSAndy King for (i = 0; i < num_pages; i++) { 3066d6dfb4fSAndy King queue->kernel_if->u.g.vas[i] = 3076d6dfb4fSAndy King dma_alloc_coherent(&vmci_pdev->dev, PAGE_SIZE, 3086d6dfb4fSAndy King &queue->kernel_if->u.g.pas[i], 3096d6dfb4fSAndy King GFP_KERNEL); 3106d6dfb4fSAndy King if (!queue->kernel_if->u.g.vas[i]) { 3116d6dfb4fSAndy King /* Size excl. the header. */ 31206164d2bSGeorge Zhang qp_free_queue(queue, i * PAGE_SIZE); 31306164d2bSGeorge Zhang return NULL; 31406164d2bSGeorge Zhang } 3156d6dfb4fSAndy King } 3166d6dfb4fSAndy King 3176d6dfb4fSAndy King /* Queue header is the first page. */ 3186d6dfb4fSAndy King queue->q_header = queue->kernel_if->u.g.vas[0]; 3196d6dfb4fSAndy King 3206d6dfb4fSAndy King return queue; 3216d6dfb4fSAndy King } 32206164d2bSGeorge Zhang 32306164d2bSGeorge Zhang /* 32406164d2bSGeorge Zhang * Copies from a given buffer or iovector to a VMCI Queue. Uses 32506164d2bSGeorge Zhang * kmap()/kunmap() to dynamically map/unmap required portions of the queue 32606164d2bSGeorge Zhang * by traversing the offset -> page translation structure for the queue. 32706164d2bSGeorge Zhang * Assumes that offset + size does not wrap around in the queue. 32806164d2bSGeorge Zhang */ 329d1038084SAl Viro static int qp_memcpy_to_queue_iter(struct vmci_queue *queue, 33006164d2bSGeorge Zhang u64 queue_offset, 331d1038084SAl Viro struct iov_iter *from, 332d1038084SAl Viro size_t size) 33306164d2bSGeorge Zhang { 33406164d2bSGeorge Zhang struct vmci_queue_kern_if *kernel_if = queue->kernel_if; 33506164d2bSGeorge Zhang size_t bytes_copied = 0; 33606164d2bSGeorge Zhang 33706164d2bSGeorge Zhang while (bytes_copied < size) { 3386d6dfb4fSAndy King const u64 page_index = 3396d6dfb4fSAndy King (queue_offset + bytes_copied) / PAGE_SIZE; 3406d6dfb4fSAndy King const size_t page_offset = 34106164d2bSGeorge Zhang (queue_offset + bytes_copied) & (PAGE_SIZE - 1); 34206164d2bSGeorge Zhang void *va; 34306164d2bSGeorge Zhang size_t to_copy; 34406164d2bSGeorge Zhang 3456d6dfb4fSAndy King if (kernel_if->host) 3466d6dfb4fSAndy King va = kmap(kernel_if->u.h.page[page_index]); 3476d6dfb4fSAndy King else 3486d6dfb4fSAndy King va = kernel_if->u.g.vas[page_index + 1]; 3496d6dfb4fSAndy King /* Skip header. */ 35006164d2bSGeorge Zhang 35106164d2bSGeorge Zhang if (size - bytes_copied > PAGE_SIZE - page_offset) 35206164d2bSGeorge Zhang /* Enough payload to fill up from this page. */ 35306164d2bSGeorge Zhang to_copy = PAGE_SIZE - page_offset; 35406164d2bSGeorge Zhang else 35506164d2bSGeorge Zhang to_copy = size - bytes_copied; 35606164d2bSGeorge Zhang 357d1038084SAl Viro if (!copy_from_iter_full((u8 *)va + page_offset, to_copy, 358d1038084SAl Viro from)) { 3596d6dfb4fSAndy King if (kernel_if->host) 3606d6dfb4fSAndy King kunmap(kernel_if->u.h.page[page_index]); 36106164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 36206164d2bSGeorge Zhang } 36306164d2bSGeorge Zhang bytes_copied += to_copy; 3646d6dfb4fSAndy King if (kernel_if->host) 3656d6dfb4fSAndy King kunmap(kernel_if->u.h.page[page_index]); 36606164d2bSGeorge Zhang } 36706164d2bSGeorge Zhang 36806164d2bSGeorge Zhang return VMCI_SUCCESS; 36906164d2bSGeorge Zhang } 37006164d2bSGeorge Zhang 37106164d2bSGeorge Zhang /* 37206164d2bSGeorge Zhang * Copies to a given buffer or iovector from a VMCI Queue. Uses 37306164d2bSGeorge Zhang * kmap()/kunmap() to dynamically map/unmap required portions of the queue 37406164d2bSGeorge Zhang * by traversing the offset -> page translation structure for the queue. 37506164d2bSGeorge Zhang * Assumes that offset + size does not wrap around in the queue. 37606164d2bSGeorge Zhang */ 37753f58d8eSAl Viro static int qp_memcpy_from_queue_iter(struct iov_iter *to, 37806164d2bSGeorge Zhang const struct vmci_queue *queue, 37953f58d8eSAl Viro u64 queue_offset, size_t size) 38006164d2bSGeorge Zhang { 38106164d2bSGeorge Zhang struct vmci_queue_kern_if *kernel_if = queue->kernel_if; 38206164d2bSGeorge Zhang size_t bytes_copied = 0; 38306164d2bSGeorge Zhang 38406164d2bSGeorge Zhang while (bytes_copied < size) { 3856d6dfb4fSAndy King const u64 page_index = 3866d6dfb4fSAndy King (queue_offset + bytes_copied) / PAGE_SIZE; 3876d6dfb4fSAndy King const size_t page_offset = 38806164d2bSGeorge Zhang (queue_offset + bytes_copied) & (PAGE_SIZE - 1); 38906164d2bSGeorge Zhang void *va; 39006164d2bSGeorge Zhang size_t to_copy; 39153f58d8eSAl Viro int err; 39206164d2bSGeorge Zhang 3936d6dfb4fSAndy King if (kernel_if->host) 3946d6dfb4fSAndy King va = kmap(kernel_if->u.h.page[page_index]); 3956d6dfb4fSAndy King else 3966d6dfb4fSAndy King va = kernel_if->u.g.vas[page_index + 1]; 3976d6dfb4fSAndy King /* Skip header. */ 39806164d2bSGeorge Zhang 39906164d2bSGeorge Zhang if (size - bytes_copied > PAGE_SIZE - page_offset) 40006164d2bSGeorge Zhang /* Enough payload to fill up this page. */ 40106164d2bSGeorge Zhang to_copy = PAGE_SIZE - page_offset; 40206164d2bSGeorge Zhang else 40306164d2bSGeorge Zhang to_copy = size - bytes_copied; 40406164d2bSGeorge Zhang 405ce3d6e7dSAl Viro err = copy_to_iter((u8 *)va + page_offset, to_copy, to); 406ce3d6e7dSAl Viro if (err != to_copy) { 4076d6dfb4fSAndy King if (kernel_if->host) 4086d6dfb4fSAndy King kunmap(kernel_if->u.h.page[page_index]); 40906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 41006164d2bSGeorge Zhang } 41106164d2bSGeorge Zhang bytes_copied += to_copy; 4126d6dfb4fSAndy King if (kernel_if->host) 4136d6dfb4fSAndy King kunmap(kernel_if->u.h.page[page_index]); 41406164d2bSGeorge Zhang } 41506164d2bSGeorge Zhang 41606164d2bSGeorge Zhang return VMCI_SUCCESS; 41706164d2bSGeorge Zhang } 41806164d2bSGeorge Zhang 41906164d2bSGeorge Zhang /* 42006164d2bSGeorge Zhang * Allocates two list of PPNs --- one for the pages in the produce queue, 42106164d2bSGeorge Zhang * and the other for the pages in the consume queue. Intializes the list 42206164d2bSGeorge Zhang * of PPNs with the page frame numbers of the KVA for the two queues (and 42306164d2bSGeorge Zhang * the queue headers). 42406164d2bSGeorge Zhang */ 42506164d2bSGeorge Zhang static int qp_alloc_ppn_set(void *prod_q, 42606164d2bSGeorge Zhang u64 num_produce_pages, 42706164d2bSGeorge Zhang void *cons_q, 428e6389a13SDmitry Torokhov u64 num_consume_pages, struct ppn_set *ppn_set) 42906164d2bSGeorge Zhang { 430f2db7361SVishnu DASA u64 *produce_ppns; 431f2db7361SVishnu DASA u64 *consume_ppns; 43206164d2bSGeorge Zhang struct vmci_queue *produce_q = prod_q; 43306164d2bSGeorge Zhang struct vmci_queue *consume_q = cons_q; 43406164d2bSGeorge Zhang u64 i; 43506164d2bSGeorge Zhang 43606164d2bSGeorge Zhang if (!produce_q || !num_produce_pages || !consume_q || 43706164d2bSGeorge Zhang !num_consume_pages || !ppn_set) 43806164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 43906164d2bSGeorge Zhang 44006164d2bSGeorge Zhang if (ppn_set->initialized) 44106164d2bSGeorge Zhang return VMCI_ERROR_ALREADY_EXISTS; 44206164d2bSGeorge Zhang 44306164d2bSGeorge Zhang produce_ppns = 4446da2ec56SKees Cook kmalloc_array(num_produce_pages, sizeof(*produce_ppns), 4456da2ec56SKees Cook GFP_KERNEL); 44606164d2bSGeorge Zhang if (!produce_ppns) 44706164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 44806164d2bSGeorge Zhang 44906164d2bSGeorge Zhang consume_ppns = 4506da2ec56SKees Cook kmalloc_array(num_consume_pages, sizeof(*consume_ppns), 4516da2ec56SKees Cook GFP_KERNEL); 45206164d2bSGeorge Zhang if (!consume_ppns) { 45306164d2bSGeorge Zhang kfree(produce_ppns); 45406164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 45506164d2bSGeorge Zhang } 45606164d2bSGeorge Zhang 457f2db7361SVishnu DASA for (i = 0; i < num_produce_pages; i++) 45806164d2bSGeorge Zhang produce_ppns[i] = 4596d6dfb4fSAndy King produce_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT; 46006164d2bSGeorge Zhang 461f2db7361SVishnu DASA for (i = 0; i < num_consume_pages; i++) 46206164d2bSGeorge Zhang consume_ppns[i] = 4636d6dfb4fSAndy King consume_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT; 46406164d2bSGeorge Zhang 46506164d2bSGeorge Zhang ppn_set->num_produce_pages = num_produce_pages; 46606164d2bSGeorge Zhang ppn_set->num_consume_pages = num_consume_pages; 46706164d2bSGeorge Zhang ppn_set->produce_ppns = produce_ppns; 46806164d2bSGeorge Zhang ppn_set->consume_ppns = consume_ppns; 46906164d2bSGeorge Zhang ppn_set->initialized = true; 47006164d2bSGeorge Zhang return VMCI_SUCCESS; 47106164d2bSGeorge Zhang } 47206164d2bSGeorge Zhang 47306164d2bSGeorge Zhang /* 47406164d2bSGeorge Zhang * Frees the two list of PPNs for a queue pair. 47506164d2bSGeorge Zhang */ 476e6389a13SDmitry Torokhov static void qp_free_ppn_set(struct ppn_set *ppn_set) 47706164d2bSGeorge Zhang { 47806164d2bSGeorge Zhang if (ppn_set->initialized) { 47906164d2bSGeorge Zhang /* Do not call these functions on NULL inputs. */ 48006164d2bSGeorge Zhang kfree(ppn_set->produce_ppns); 48106164d2bSGeorge Zhang kfree(ppn_set->consume_ppns); 48206164d2bSGeorge Zhang } 48306164d2bSGeorge Zhang memset(ppn_set, 0, sizeof(*ppn_set)); 48406164d2bSGeorge Zhang } 48506164d2bSGeorge Zhang 48606164d2bSGeorge Zhang /* 48706164d2bSGeorge Zhang * Populates the list of PPNs in the hypercall structure with the PPNS 48806164d2bSGeorge Zhang * of the produce queue and the consume queue. 48906164d2bSGeorge Zhang */ 490e6389a13SDmitry Torokhov static int qp_populate_ppn_set(u8 *call_buf, const struct ppn_set *ppn_set) 49106164d2bSGeorge Zhang { 492f2db7361SVishnu DASA if (vmci_use_ppn64()) { 49306164d2bSGeorge Zhang memcpy(call_buf, ppn_set->produce_ppns, 494f2db7361SVishnu DASA ppn_set->num_produce_pages * 495f2db7361SVishnu DASA sizeof(*ppn_set->produce_ppns)); 49606164d2bSGeorge Zhang memcpy(call_buf + 497f2db7361SVishnu DASA ppn_set->num_produce_pages * 498f2db7361SVishnu DASA sizeof(*ppn_set->produce_ppns), 49906164d2bSGeorge Zhang ppn_set->consume_ppns, 500f2db7361SVishnu DASA ppn_set->num_consume_pages * 501f2db7361SVishnu DASA sizeof(*ppn_set->consume_ppns)); 502f2db7361SVishnu DASA } else { 503f2db7361SVishnu DASA int i; 504f2db7361SVishnu DASA u32 *ppns = (u32 *) call_buf; 505f2db7361SVishnu DASA 506f2db7361SVishnu DASA for (i = 0; i < ppn_set->num_produce_pages; i++) 507f2db7361SVishnu DASA ppns[i] = (u32) ppn_set->produce_ppns[i]; 508f2db7361SVishnu DASA 509f2db7361SVishnu DASA ppns = &ppns[ppn_set->num_produce_pages]; 510f2db7361SVishnu DASA 511f2db7361SVishnu DASA for (i = 0; i < ppn_set->num_consume_pages; i++) 512f2db7361SVishnu DASA ppns[i] = (u32) ppn_set->consume_ppns[i]; 513f2db7361SVishnu DASA } 51406164d2bSGeorge Zhang 51506164d2bSGeorge Zhang return VMCI_SUCCESS; 51606164d2bSGeorge Zhang } 51706164d2bSGeorge Zhang 51806164d2bSGeorge Zhang /* 51906164d2bSGeorge Zhang * Allocates kernel VA space of specified size plus space for the queue 52006164d2bSGeorge Zhang * and kernel interface. This is different from the guest queue allocator, 52106164d2bSGeorge Zhang * because we do not allocate our own queue header/data pages here but 52206164d2bSGeorge Zhang * share those of the guest. 52306164d2bSGeorge Zhang */ 52406164d2bSGeorge Zhang static struct vmci_queue *qp_host_alloc_queue(u64 size) 52506164d2bSGeorge Zhang { 52606164d2bSGeorge Zhang struct vmci_queue *queue; 527aa6467f1SJorgen Hansen size_t queue_page_size; 528146180c0SDan Carpenter u64 num_pages; 52906164d2bSGeorge Zhang const size_t queue_size = sizeof(*queue) + sizeof(*(queue->kernel_if)); 530aa6467f1SJorgen Hansen 531146180c0SDan Carpenter if (size > SIZE_MAX - PAGE_SIZE) 532146180c0SDan Carpenter return NULL; 533146180c0SDan Carpenter num_pages = DIV_ROUND_UP(size, PAGE_SIZE) + 1; 534aa6467f1SJorgen Hansen if (num_pages > (SIZE_MAX - queue_size) / 535aa6467f1SJorgen Hansen sizeof(*queue->kernel_if->u.h.page)) 536aa6467f1SJorgen Hansen return NULL; 537aa6467f1SJorgen Hansen 538aa6467f1SJorgen Hansen queue_page_size = num_pages * sizeof(*queue->kernel_if->u.h.page); 53906164d2bSGeorge Zhang 54006164d2bSGeorge Zhang queue = kzalloc(queue_size + queue_page_size, GFP_KERNEL); 54106164d2bSGeorge Zhang if (queue) { 54206164d2bSGeorge Zhang queue->q_header = NULL; 54306164d2bSGeorge Zhang queue->saved_header = NULL; 5446d6dfb4fSAndy King queue->kernel_if = (struct vmci_queue_kern_if *)(queue + 1); 54506164d2bSGeorge Zhang queue->kernel_if->host = true; 54606164d2bSGeorge Zhang queue->kernel_if->mutex = NULL; 54706164d2bSGeorge Zhang queue->kernel_if->num_pages = num_pages; 5486d6dfb4fSAndy King queue->kernel_if->u.h.header_page = 54906164d2bSGeorge Zhang (struct page **)((u8 *)queue + queue_size); 5506d6dfb4fSAndy King queue->kernel_if->u.h.page = 5516d6dfb4fSAndy King &queue->kernel_if->u.h.header_page[1]; 55206164d2bSGeorge Zhang } 55306164d2bSGeorge Zhang 55406164d2bSGeorge Zhang return queue; 55506164d2bSGeorge Zhang } 55606164d2bSGeorge Zhang 55706164d2bSGeorge Zhang /* 55806164d2bSGeorge Zhang * Frees kernel memory for a given queue (header plus translation 55906164d2bSGeorge Zhang * structure). 56006164d2bSGeorge Zhang */ 56106164d2bSGeorge Zhang static void qp_host_free_queue(struct vmci_queue *queue, u64 queue_size) 56206164d2bSGeorge Zhang { 56306164d2bSGeorge Zhang kfree(queue); 56406164d2bSGeorge Zhang } 56506164d2bSGeorge Zhang 56606164d2bSGeorge Zhang /* 56706164d2bSGeorge Zhang * Initialize the mutex for the pair of queues. This mutex is used to 56806164d2bSGeorge Zhang * protect the q_header and the buffer from changing out from under any 56906164d2bSGeorge Zhang * users of either queue. Of course, it's only any good if the mutexes 57006164d2bSGeorge Zhang * are actually acquired. Queue structure must lie on non-paged memory 57106164d2bSGeorge Zhang * or we cannot guarantee access to the mutex. 57206164d2bSGeorge Zhang */ 57306164d2bSGeorge Zhang static void qp_init_queue_mutex(struct vmci_queue *produce_q, 57406164d2bSGeorge Zhang struct vmci_queue *consume_q) 57506164d2bSGeorge Zhang { 57606164d2bSGeorge Zhang /* 57706164d2bSGeorge Zhang * Only the host queue has shared state - the guest queues do not 57806164d2bSGeorge Zhang * need to synchronize access using a queue mutex. 57906164d2bSGeorge Zhang */ 58006164d2bSGeorge Zhang 58106164d2bSGeorge Zhang if (produce_q->kernel_if->host) { 58206164d2bSGeorge Zhang produce_q->kernel_if->mutex = &produce_q->kernel_if->__mutex; 58306164d2bSGeorge Zhang consume_q->kernel_if->mutex = &produce_q->kernel_if->__mutex; 58406164d2bSGeorge Zhang mutex_init(produce_q->kernel_if->mutex); 58506164d2bSGeorge Zhang } 58606164d2bSGeorge Zhang } 58706164d2bSGeorge Zhang 58806164d2bSGeorge Zhang /* 58906164d2bSGeorge Zhang * Cleans up the mutex for the pair of queues. 59006164d2bSGeorge Zhang */ 59106164d2bSGeorge Zhang static void qp_cleanup_queue_mutex(struct vmci_queue *produce_q, 59206164d2bSGeorge Zhang struct vmci_queue *consume_q) 59306164d2bSGeorge Zhang { 59406164d2bSGeorge Zhang if (produce_q->kernel_if->host) { 59506164d2bSGeorge Zhang produce_q->kernel_if->mutex = NULL; 59606164d2bSGeorge Zhang consume_q->kernel_if->mutex = NULL; 59706164d2bSGeorge Zhang } 59806164d2bSGeorge Zhang } 59906164d2bSGeorge Zhang 60006164d2bSGeorge Zhang /* 60106164d2bSGeorge Zhang * Acquire the mutex for the queue. Note that the produce_q and 60206164d2bSGeorge Zhang * the consume_q share a mutex. So, only one of the two need to 60306164d2bSGeorge Zhang * be passed in to this routine. Either will work just fine. 60406164d2bSGeorge Zhang */ 60506164d2bSGeorge Zhang static void qp_acquire_queue_mutex(struct vmci_queue *queue) 60606164d2bSGeorge Zhang { 60706164d2bSGeorge Zhang if (queue->kernel_if->host) 60806164d2bSGeorge Zhang mutex_lock(queue->kernel_if->mutex); 60906164d2bSGeorge Zhang } 61006164d2bSGeorge Zhang 61106164d2bSGeorge Zhang /* 61206164d2bSGeorge Zhang * Release the mutex for the queue. Note that the produce_q and 61306164d2bSGeorge Zhang * the consume_q share a mutex. So, only one of the two need to 61406164d2bSGeorge Zhang * be passed in to this routine. Either will work just fine. 61506164d2bSGeorge Zhang */ 61606164d2bSGeorge Zhang static void qp_release_queue_mutex(struct vmci_queue *queue) 61706164d2bSGeorge Zhang { 61806164d2bSGeorge Zhang if (queue->kernel_if->host) 61906164d2bSGeorge Zhang mutex_unlock(queue->kernel_if->mutex); 62006164d2bSGeorge Zhang } 62106164d2bSGeorge Zhang 62206164d2bSGeorge Zhang /* 62306164d2bSGeorge Zhang * Helper function to release pages in the PageStoreAttachInfo 62406164d2bSGeorge Zhang * previously obtained using get_user_pages. 62506164d2bSGeorge Zhang */ 62606164d2bSGeorge Zhang static void qp_release_pages(struct page **pages, 62706164d2bSGeorge Zhang u64 num_pages, bool dirty) 62806164d2bSGeorge Zhang { 62906164d2bSGeorge Zhang int i; 63006164d2bSGeorge Zhang 63106164d2bSGeorge Zhang for (i = 0; i < num_pages; i++) { 63206164d2bSGeorge Zhang if (dirty) 63306164d2bSGeorge Zhang set_page_dirty(pages[i]); 63406164d2bSGeorge Zhang 63509cbfeafSKirill A. Shutemov put_page(pages[i]); 63606164d2bSGeorge Zhang pages[i] = NULL; 63706164d2bSGeorge Zhang } 63806164d2bSGeorge Zhang } 63906164d2bSGeorge Zhang 64006164d2bSGeorge Zhang /* 64106164d2bSGeorge Zhang * Lock the user pages referenced by the {produce,consume}Buffer 64206164d2bSGeorge Zhang * struct into memory and populate the {produce,consume}Pages 64306164d2bSGeorge Zhang * arrays in the attach structure with them. 64406164d2bSGeorge Zhang */ 64506164d2bSGeorge Zhang static int qp_host_get_user_memory(u64 produce_uva, 64606164d2bSGeorge Zhang u64 consume_uva, 64706164d2bSGeorge Zhang struct vmci_queue *produce_q, 64806164d2bSGeorge Zhang struct vmci_queue *consume_q) 64906164d2bSGeorge Zhang { 65006164d2bSGeorge Zhang int retval; 65106164d2bSGeorge Zhang int err = VMCI_SUCCESS; 65206164d2bSGeorge Zhang 653240ddd49SJan Kara retval = get_user_pages_fast((uintptr_t) produce_uva, 65473b0140bSIra Weiny produce_q->kernel_if->num_pages, 65573b0140bSIra Weiny FOLL_WRITE, 656240ddd49SJan Kara produce_q->kernel_if->u.h.header_page); 6577fb2fd4eSDan Carpenter if (retval < (int)produce_q->kernel_if->num_pages) { 658bf136121SDavidlohr Bueso pr_debug("get_user_pages_fast(produce) failed (retval=%d)", 659bf136121SDavidlohr Bueso retval); 6606d6dfb4fSAndy King qp_release_pages(produce_q->kernel_if->u.h.header_page, 6616d6dfb4fSAndy King retval, false); 66206164d2bSGeorge Zhang err = VMCI_ERROR_NO_MEM; 66306164d2bSGeorge Zhang goto out; 66406164d2bSGeorge Zhang } 66506164d2bSGeorge Zhang 666240ddd49SJan Kara retval = get_user_pages_fast((uintptr_t) consume_uva, 66773b0140bSIra Weiny consume_q->kernel_if->num_pages, 66873b0140bSIra Weiny FOLL_WRITE, 669240ddd49SJan Kara consume_q->kernel_if->u.h.header_page); 6707fb2fd4eSDan Carpenter if (retval < (int)consume_q->kernel_if->num_pages) { 671bf136121SDavidlohr Bueso pr_debug("get_user_pages_fast(consume) failed (retval=%d)", 672bf136121SDavidlohr Bueso retval); 6736d6dfb4fSAndy King qp_release_pages(consume_q->kernel_if->u.h.header_page, 6746d6dfb4fSAndy King retval, false); 6756d6dfb4fSAndy King qp_release_pages(produce_q->kernel_if->u.h.header_page, 67606164d2bSGeorge Zhang produce_q->kernel_if->num_pages, false); 67706164d2bSGeorge Zhang err = VMCI_ERROR_NO_MEM; 67806164d2bSGeorge Zhang } 67906164d2bSGeorge Zhang 68006164d2bSGeorge Zhang out: 68106164d2bSGeorge Zhang return err; 68206164d2bSGeorge Zhang } 68306164d2bSGeorge Zhang 68406164d2bSGeorge Zhang /* 68506164d2bSGeorge Zhang * Registers the specification of the user pages used for backing a queue 68606164d2bSGeorge Zhang * pair. Enough information to map in pages is stored in the OS specific 68706164d2bSGeorge Zhang * part of the struct vmci_queue structure. 68806164d2bSGeorge Zhang */ 68906164d2bSGeorge Zhang static int qp_host_register_user_memory(struct vmci_qp_page_store *page_store, 69006164d2bSGeorge Zhang struct vmci_queue *produce_q, 69106164d2bSGeorge Zhang struct vmci_queue *consume_q) 69206164d2bSGeorge Zhang { 69306164d2bSGeorge Zhang u64 produce_uva; 69406164d2bSGeorge Zhang u64 consume_uva; 69506164d2bSGeorge Zhang 69606164d2bSGeorge Zhang /* 69706164d2bSGeorge Zhang * The new style and the old style mapping only differs in 69806164d2bSGeorge Zhang * that we either get a single or two UVAs, so we split the 69906164d2bSGeorge Zhang * single UVA range at the appropriate spot. 70006164d2bSGeorge Zhang */ 70106164d2bSGeorge Zhang produce_uva = page_store->pages; 70206164d2bSGeorge Zhang consume_uva = page_store->pages + 70306164d2bSGeorge Zhang produce_q->kernel_if->num_pages * PAGE_SIZE; 70406164d2bSGeorge Zhang return qp_host_get_user_memory(produce_uva, consume_uva, produce_q, 70506164d2bSGeorge Zhang consume_q); 70606164d2bSGeorge Zhang } 70706164d2bSGeorge Zhang 70806164d2bSGeorge Zhang /* 70906164d2bSGeorge Zhang * Releases and removes the references to user pages stored in the attach 71006164d2bSGeorge Zhang * struct. Pages are released from the page cache and may become 71106164d2bSGeorge Zhang * swappable again. 71206164d2bSGeorge Zhang */ 71306164d2bSGeorge Zhang static void qp_host_unregister_user_memory(struct vmci_queue *produce_q, 71406164d2bSGeorge Zhang struct vmci_queue *consume_q) 71506164d2bSGeorge Zhang { 7166d6dfb4fSAndy King qp_release_pages(produce_q->kernel_if->u.h.header_page, 71706164d2bSGeorge Zhang produce_q->kernel_if->num_pages, true); 7186d6dfb4fSAndy King memset(produce_q->kernel_if->u.h.header_page, 0, 7196d6dfb4fSAndy King sizeof(*produce_q->kernel_if->u.h.header_page) * 72006164d2bSGeorge Zhang produce_q->kernel_if->num_pages); 7216d6dfb4fSAndy King qp_release_pages(consume_q->kernel_if->u.h.header_page, 72206164d2bSGeorge Zhang consume_q->kernel_if->num_pages, true); 7236d6dfb4fSAndy King memset(consume_q->kernel_if->u.h.header_page, 0, 7246d6dfb4fSAndy King sizeof(*consume_q->kernel_if->u.h.header_page) * 72506164d2bSGeorge Zhang consume_q->kernel_if->num_pages); 72606164d2bSGeorge Zhang } 72706164d2bSGeorge Zhang 72806164d2bSGeorge Zhang /* 72906164d2bSGeorge Zhang * Once qp_host_register_user_memory has been performed on a 73006164d2bSGeorge Zhang * queue, the queue pair headers can be mapped into the 73106164d2bSGeorge Zhang * kernel. Once mapped, they must be unmapped with 73206164d2bSGeorge Zhang * qp_host_unmap_queues prior to calling 73306164d2bSGeorge Zhang * qp_host_unregister_user_memory. 73406164d2bSGeorge Zhang * Pages are pinned. 73506164d2bSGeorge Zhang */ 73606164d2bSGeorge Zhang static int qp_host_map_queues(struct vmci_queue *produce_q, 73706164d2bSGeorge Zhang struct vmci_queue *consume_q) 73806164d2bSGeorge Zhang { 73906164d2bSGeorge Zhang int result; 74006164d2bSGeorge Zhang 74106164d2bSGeorge Zhang if (!produce_q->q_header || !consume_q->q_header) { 74206164d2bSGeorge Zhang struct page *headers[2]; 74306164d2bSGeorge Zhang 74406164d2bSGeorge Zhang if (produce_q->q_header != consume_q->q_header) 74506164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 74606164d2bSGeorge Zhang 7476d6dfb4fSAndy King if (produce_q->kernel_if->u.h.header_page == NULL || 7486d6dfb4fSAndy King *produce_q->kernel_if->u.h.header_page == NULL) 74906164d2bSGeorge Zhang return VMCI_ERROR_UNAVAILABLE; 75006164d2bSGeorge Zhang 7516d6dfb4fSAndy King headers[0] = *produce_q->kernel_if->u.h.header_page; 7526d6dfb4fSAndy King headers[1] = *consume_q->kernel_if->u.h.header_page; 75306164d2bSGeorge Zhang 75406164d2bSGeorge Zhang produce_q->q_header = vmap(headers, 2, VM_MAP, PAGE_KERNEL); 75506164d2bSGeorge Zhang if (produce_q->q_header != NULL) { 75606164d2bSGeorge Zhang consume_q->q_header = 75706164d2bSGeorge Zhang (struct vmci_queue_header *)((u8 *) 75806164d2bSGeorge Zhang produce_q->q_header + 75906164d2bSGeorge Zhang PAGE_SIZE); 76006164d2bSGeorge Zhang result = VMCI_SUCCESS; 76106164d2bSGeorge Zhang } else { 76206164d2bSGeorge Zhang pr_warn("vmap failed\n"); 76306164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 76406164d2bSGeorge Zhang } 76506164d2bSGeorge Zhang } else { 76606164d2bSGeorge Zhang result = VMCI_SUCCESS; 76706164d2bSGeorge Zhang } 76806164d2bSGeorge Zhang 76906164d2bSGeorge Zhang return result; 77006164d2bSGeorge Zhang } 77106164d2bSGeorge Zhang 77206164d2bSGeorge Zhang /* 77306164d2bSGeorge Zhang * Unmaps previously mapped queue pair headers from the kernel. 77406164d2bSGeorge Zhang * Pages are unpinned. 77506164d2bSGeorge Zhang */ 77606164d2bSGeorge Zhang static int qp_host_unmap_queues(u32 gid, 77706164d2bSGeorge Zhang struct vmci_queue *produce_q, 77806164d2bSGeorge Zhang struct vmci_queue *consume_q) 77906164d2bSGeorge Zhang { 78006164d2bSGeorge Zhang if (produce_q->q_header) { 78106164d2bSGeorge Zhang if (produce_q->q_header < consume_q->q_header) 78206164d2bSGeorge Zhang vunmap(produce_q->q_header); 78306164d2bSGeorge Zhang else 78406164d2bSGeorge Zhang vunmap(consume_q->q_header); 78506164d2bSGeorge Zhang 78606164d2bSGeorge Zhang produce_q->q_header = NULL; 78706164d2bSGeorge Zhang consume_q->q_header = NULL; 78806164d2bSGeorge Zhang } 78906164d2bSGeorge Zhang 79006164d2bSGeorge Zhang return VMCI_SUCCESS; 79106164d2bSGeorge Zhang } 79206164d2bSGeorge Zhang 79306164d2bSGeorge Zhang /* 79406164d2bSGeorge Zhang * Finds the entry in the list corresponding to a given handle. Assumes 79506164d2bSGeorge Zhang * that the list is locked. 79606164d2bSGeorge Zhang */ 79706164d2bSGeorge Zhang static struct qp_entry *qp_list_find(struct qp_list *qp_list, 79806164d2bSGeorge Zhang struct vmci_handle handle) 79906164d2bSGeorge Zhang { 80006164d2bSGeorge Zhang struct qp_entry *entry; 80106164d2bSGeorge Zhang 80206164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle)) 80306164d2bSGeorge Zhang return NULL; 80406164d2bSGeorge Zhang 80506164d2bSGeorge Zhang list_for_each_entry(entry, &qp_list->head, list_item) { 80606164d2bSGeorge Zhang if (vmci_handle_is_equal(entry->handle, handle)) 80706164d2bSGeorge Zhang return entry; 80806164d2bSGeorge Zhang } 80906164d2bSGeorge Zhang 81006164d2bSGeorge Zhang return NULL; 81106164d2bSGeorge Zhang } 81206164d2bSGeorge Zhang 81306164d2bSGeorge Zhang /* 81406164d2bSGeorge Zhang * Finds the entry in the list corresponding to a given handle. 81506164d2bSGeorge Zhang */ 81606164d2bSGeorge Zhang static struct qp_guest_endpoint * 81706164d2bSGeorge Zhang qp_guest_handle_to_entry(struct vmci_handle handle) 81806164d2bSGeorge Zhang { 81906164d2bSGeorge Zhang struct qp_guest_endpoint *entry; 82006164d2bSGeorge Zhang struct qp_entry *qp = qp_list_find(&qp_guest_endpoints, handle); 82106164d2bSGeorge Zhang 82206164d2bSGeorge Zhang entry = qp ? container_of( 82306164d2bSGeorge Zhang qp, struct qp_guest_endpoint, qp) : NULL; 82406164d2bSGeorge Zhang return entry; 82506164d2bSGeorge Zhang } 82606164d2bSGeorge Zhang 82706164d2bSGeorge Zhang /* 82806164d2bSGeorge Zhang * Finds the entry in the list corresponding to a given handle. 82906164d2bSGeorge Zhang */ 83006164d2bSGeorge Zhang static struct qp_broker_entry * 83106164d2bSGeorge Zhang qp_broker_handle_to_entry(struct vmci_handle handle) 83206164d2bSGeorge Zhang { 83306164d2bSGeorge Zhang struct qp_broker_entry *entry; 83406164d2bSGeorge Zhang struct qp_entry *qp = qp_list_find(&qp_broker_list, handle); 83506164d2bSGeorge Zhang 83606164d2bSGeorge Zhang entry = qp ? container_of( 83706164d2bSGeorge Zhang qp, struct qp_broker_entry, qp) : NULL; 83806164d2bSGeorge Zhang return entry; 83906164d2bSGeorge Zhang } 84006164d2bSGeorge Zhang 84106164d2bSGeorge Zhang /* 84206164d2bSGeorge Zhang * Dispatches a queue pair event message directly into the local event 84306164d2bSGeorge Zhang * queue. 84406164d2bSGeorge Zhang */ 84506164d2bSGeorge Zhang static int qp_notify_peer_local(bool attach, struct vmci_handle handle) 84606164d2bSGeorge Zhang { 84706164d2bSGeorge Zhang u32 context_id = vmci_get_context_id(); 84806164d2bSGeorge Zhang struct vmci_event_qp ev; 84906164d2bSGeorge Zhang 85006164d2bSGeorge Zhang ev.msg.hdr.dst = vmci_make_handle(context_id, VMCI_EVENT_HANDLER); 85106164d2bSGeorge Zhang ev.msg.hdr.src = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 85206164d2bSGeorge Zhang VMCI_CONTEXT_RESOURCE_ID); 85306164d2bSGeorge Zhang ev.msg.hdr.payload_size = sizeof(ev) - sizeof(ev.msg.hdr); 85406164d2bSGeorge Zhang ev.msg.event_data.event = 85506164d2bSGeorge Zhang attach ? VMCI_EVENT_QP_PEER_ATTACH : VMCI_EVENT_QP_PEER_DETACH; 85606164d2bSGeorge Zhang ev.payload.peer_id = context_id; 85706164d2bSGeorge Zhang ev.payload.handle = handle; 85806164d2bSGeorge Zhang 85906164d2bSGeorge Zhang return vmci_event_dispatch(&ev.msg.hdr); 86006164d2bSGeorge Zhang } 86106164d2bSGeorge Zhang 86206164d2bSGeorge Zhang /* 86306164d2bSGeorge Zhang * Allocates and initializes a qp_guest_endpoint structure. 86406164d2bSGeorge Zhang * Allocates a queue_pair rid (and handle) iff the given entry has 86506164d2bSGeorge Zhang * an invalid handle. 0 through VMCI_RESERVED_RESOURCE_ID_MAX 86606164d2bSGeorge Zhang * are reserved handles. Assumes that the QP list mutex is held 86706164d2bSGeorge Zhang * by the caller. 86806164d2bSGeorge Zhang */ 86906164d2bSGeorge Zhang static struct qp_guest_endpoint * 87006164d2bSGeorge Zhang qp_guest_endpoint_create(struct vmci_handle handle, 87106164d2bSGeorge Zhang u32 peer, 87206164d2bSGeorge Zhang u32 flags, 87306164d2bSGeorge Zhang u64 produce_size, 87406164d2bSGeorge Zhang u64 consume_size, 87506164d2bSGeorge Zhang void *produce_q, 87606164d2bSGeorge Zhang void *consume_q) 87706164d2bSGeorge Zhang { 87806164d2bSGeorge Zhang int result; 87906164d2bSGeorge Zhang struct qp_guest_endpoint *entry; 88006164d2bSGeorge Zhang /* One page each for the queue headers. */ 88142281d20SAndy King const u64 num_ppns = DIV_ROUND_UP(produce_size, PAGE_SIZE) + 88242281d20SAndy King DIV_ROUND_UP(consume_size, PAGE_SIZE) + 2; 88306164d2bSGeorge Zhang 88406164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle)) { 88506164d2bSGeorge Zhang u32 context_id = vmci_get_context_id(); 88606164d2bSGeorge Zhang 88706164d2bSGeorge Zhang handle = vmci_make_handle(context_id, VMCI_INVALID_ID); 88806164d2bSGeorge Zhang } 88906164d2bSGeorge Zhang 89006164d2bSGeorge Zhang entry = kzalloc(sizeof(*entry), GFP_KERNEL); 89106164d2bSGeorge Zhang if (entry) { 89206164d2bSGeorge Zhang entry->qp.peer = peer; 89306164d2bSGeorge Zhang entry->qp.flags = flags; 89406164d2bSGeorge Zhang entry->qp.produce_size = produce_size; 89506164d2bSGeorge Zhang entry->qp.consume_size = consume_size; 89606164d2bSGeorge Zhang entry->qp.ref_count = 0; 89706164d2bSGeorge Zhang entry->num_ppns = num_ppns; 89806164d2bSGeorge Zhang entry->produce_q = produce_q; 89906164d2bSGeorge Zhang entry->consume_q = consume_q; 90006164d2bSGeorge Zhang INIT_LIST_HEAD(&entry->qp.list_item); 90106164d2bSGeorge Zhang 90206164d2bSGeorge Zhang /* Add resource obj */ 90306164d2bSGeorge Zhang result = vmci_resource_add(&entry->resource, 90406164d2bSGeorge Zhang VMCI_RESOURCE_TYPE_QPAIR_GUEST, 90506164d2bSGeorge Zhang handle); 90606164d2bSGeorge Zhang entry->qp.handle = vmci_resource_handle(&entry->resource); 90706164d2bSGeorge Zhang if ((result != VMCI_SUCCESS) || 90806164d2bSGeorge Zhang qp_list_find(&qp_guest_endpoints, entry->qp.handle)) { 90906164d2bSGeorge Zhang pr_warn("Failed to add new resource (handle=0x%x:0x%x), error: %d", 91006164d2bSGeorge Zhang handle.context, handle.resource, result); 91106164d2bSGeorge Zhang kfree(entry); 91206164d2bSGeorge Zhang entry = NULL; 91306164d2bSGeorge Zhang } 91406164d2bSGeorge Zhang } 91506164d2bSGeorge Zhang return entry; 91606164d2bSGeorge Zhang } 91706164d2bSGeorge Zhang 91806164d2bSGeorge Zhang /* 91906164d2bSGeorge Zhang * Frees a qp_guest_endpoint structure. 92006164d2bSGeorge Zhang */ 92106164d2bSGeorge Zhang static void qp_guest_endpoint_destroy(struct qp_guest_endpoint *entry) 92206164d2bSGeorge Zhang { 92306164d2bSGeorge Zhang qp_free_ppn_set(&entry->ppn_set); 92406164d2bSGeorge Zhang qp_cleanup_queue_mutex(entry->produce_q, entry->consume_q); 92506164d2bSGeorge Zhang qp_free_queue(entry->produce_q, entry->qp.produce_size); 92606164d2bSGeorge Zhang qp_free_queue(entry->consume_q, entry->qp.consume_size); 92706164d2bSGeorge Zhang /* Unlink from resource hash table and free callback */ 92806164d2bSGeorge Zhang vmci_resource_remove(&entry->resource); 92906164d2bSGeorge Zhang 93006164d2bSGeorge Zhang kfree(entry); 93106164d2bSGeorge Zhang } 93206164d2bSGeorge Zhang 93306164d2bSGeorge Zhang /* 93406164d2bSGeorge Zhang * Helper to make a queue_pairAlloc hypercall when the driver is 93506164d2bSGeorge Zhang * supporting a guest device. 93606164d2bSGeorge Zhang */ 93706164d2bSGeorge Zhang static int qp_alloc_hypercall(const struct qp_guest_endpoint *entry) 93806164d2bSGeorge Zhang { 93906164d2bSGeorge Zhang struct vmci_qp_alloc_msg *alloc_msg; 94006164d2bSGeorge Zhang size_t msg_size; 941f2db7361SVishnu DASA size_t ppn_size; 94206164d2bSGeorge Zhang int result; 94306164d2bSGeorge Zhang 94406164d2bSGeorge Zhang if (!entry || entry->num_ppns <= 2) 94506164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 94606164d2bSGeorge Zhang 947f2db7361SVishnu DASA ppn_size = vmci_use_ppn64() ? sizeof(u64) : sizeof(u32); 94806164d2bSGeorge Zhang msg_size = sizeof(*alloc_msg) + 949f2db7361SVishnu DASA (size_t) entry->num_ppns * ppn_size; 95006164d2bSGeorge Zhang alloc_msg = kmalloc(msg_size, GFP_KERNEL); 95106164d2bSGeorge Zhang if (!alloc_msg) 95206164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 95306164d2bSGeorge Zhang 95406164d2bSGeorge Zhang alloc_msg->hdr.dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 95506164d2bSGeorge Zhang VMCI_QUEUEPAIR_ALLOC); 95606164d2bSGeorge Zhang alloc_msg->hdr.src = VMCI_ANON_SRC_HANDLE; 95706164d2bSGeorge Zhang alloc_msg->hdr.payload_size = msg_size - VMCI_DG_HEADERSIZE; 95806164d2bSGeorge Zhang alloc_msg->handle = entry->qp.handle; 95906164d2bSGeorge Zhang alloc_msg->peer = entry->qp.peer; 96006164d2bSGeorge Zhang alloc_msg->flags = entry->qp.flags; 96106164d2bSGeorge Zhang alloc_msg->produce_size = entry->qp.produce_size; 96206164d2bSGeorge Zhang alloc_msg->consume_size = entry->qp.consume_size; 96306164d2bSGeorge Zhang alloc_msg->num_ppns = entry->num_ppns; 96406164d2bSGeorge Zhang 96506164d2bSGeorge Zhang result = qp_populate_ppn_set((u8 *)alloc_msg + sizeof(*alloc_msg), 96606164d2bSGeorge Zhang &entry->ppn_set); 96706164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 96806164d2bSGeorge Zhang result = vmci_send_datagram(&alloc_msg->hdr); 96906164d2bSGeorge Zhang 97006164d2bSGeorge Zhang kfree(alloc_msg); 97106164d2bSGeorge Zhang 97206164d2bSGeorge Zhang return result; 97306164d2bSGeorge Zhang } 97406164d2bSGeorge Zhang 97506164d2bSGeorge Zhang /* 97606164d2bSGeorge Zhang * Helper to make a queue_pairDetach hypercall when the driver is 97706164d2bSGeorge Zhang * supporting a guest device. 97806164d2bSGeorge Zhang */ 97906164d2bSGeorge Zhang static int qp_detatch_hypercall(struct vmci_handle handle) 98006164d2bSGeorge Zhang { 98106164d2bSGeorge Zhang struct vmci_qp_detach_msg detach_msg; 98206164d2bSGeorge Zhang 98306164d2bSGeorge Zhang detach_msg.hdr.dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 98406164d2bSGeorge Zhang VMCI_QUEUEPAIR_DETACH); 98506164d2bSGeorge Zhang detach_msg.hdr.src = VMCI_ANON_SRC_HANDLE; 98606164d2bSGeorge Zhang detach_msg.hdr.payload_size = sizeof(handle); 98706164d2bSGeorge Zhang detach_msg.handle = handle; 98806164d2bSGeorge Zhang 98906164d2bSGeorge Zhang return vmci_send_datagram(&detach_msg.hdr); 99006164d2bSGeorge Zhang } 99106164d2bSGeorge Zhang 99206164d2bSGeorge Zhang /* 99306164d2bSGeorge Zhang * Adds the given entry to the list. Assumes that the list is locked. 99406164d2bSGeorge Zhang */ 99506164d2bSGeorge Zhang static void qp_list_add_entry(struct qp_list *qp_list, struct qp_entry *entry) 99606164d2bSGeorge Zhang { 99706164d2bSGeorge Zhang if (entry) 99806164d2bSGeorge Zhang list_add(&entry->list_item, &qp_list->head); 99906164d2bSGeorge Zhang } 100006164d2bSGeorge Zhang 100106164d2bSGeorge Zhang /* 100206164d2bSGeorge Zhang * Removes the given entry from the list. Assumes that the list is locked. 100306164d2bSGeorge Zhang */ 100406164d2bSGeorge Zhang static void qp_list_remove_entry(struct qp_list *qp_list, 100506164d2bSGeorge Zhang struct qp_entry *entry) 100606164d2bSGeorge Zhang { 100706164d2bSGeorge Zhang if (entry) 100806164d2bSGeorge Zhang list_del(&entry->list_item); 100906164d2bSGeorge Zhang } 101006164d2bSGeorge Zhang 101106164d2bSGeorge Zhang /* 101206164d2bSGeorge Zhang * Helper for VMCI queue_pair detach interface. Frees the physical 101306164d2bSGeorge Zhang * pages for the queue pair. 101406164d2bSGeorge Zhang */ 101506164d2bSGeorge Zhang static int qp_detatch_guest_work(struct vmci_handle handle) 101606164d2bSGeorge Zhang { 101706164d2bSGeorge Zhang int result; 101806164d2bSGeorge Zhang struct qp_guest_endpoint *entry; 101906164d2bSGeorge Zhang u32 ref_count = ~0; /* To avoid compiler warning below */ 102006164d2bSGeorge Zhang 102106164d2bSGeorge Zhang mutex_lock(&qp_guest_endpoints.mutex); 102206164d2bSGeorge Zhang 102306164d2bSGeorge Zhang entry = qp_guest_handle_to_entry(handle); 102406164d2bSGeorge Zhang if (!entry) { 102506164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 102606164d2bSGeorge Zhang return VMCI_ERROR_NOT_FOUND; 102706164d2bSGeorge Zhang } 102806164d2bSGeorge Zhang 102906164d2bSGeorge Zhang if (entry->qp.flags & VMCI_QPFLAG_LOCAL) { 103006164d2bSGeorge Zhang result = VMCI_SUCCESS; 103106164d2bSGeorge Zhang 103206164d2bSGeorge Zhang if (entry->qp.ref_count > 1) { 103306164d2bSGeorge Zhang result = qp_notify_peer_local(false, handle); 103406164d2bSGeorge Zhang /* 103506164d2bSGeorge Zhang * We can fail to notify a local queuepair 103606164d2bSGeorge Zhang * because we can't allocate. We still want 103706164d2bSGeorge Zhang * to release the entry if that happens, so 103806164d2bSGeorge Zhang * don't bail out yet. 103906164d2bSGeorge Zhang */ 104006164d2bSGeorge Zhang } 104106164d2bSGeorge Zhang } else { 104206164d2bSGeorge Zhang result = qp_detatch_hypercall(handle); 104306164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 104406164d2bSGeorge Zhang /* 104506164d2bSGeorge Zhang * We failed to notify a non-local queuepair. 104606164d2bSGeorge Zhang * That other queuepair might still be 104706164d2bSGeorge Zhang * accessing the shared memory, so don't 104806164d2bSGeorge Zhang * release the entry yet. It will get cleaned 104906164d2bSGeorge Zhang * up by VMCIqueue_pair_Exit() if necessary 105006164d2bSGeorge Zhang * (assuming we are going away, otherwise why 105106164d2bSGeorge Zhang * did this fail?). 105206164d2bSGeorge Zhang */ 105306164d2bSGeorge Zhang 105406164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 105506164d2bSGeorge Zhang return result; 105606164d2bSGeorge Zhang } 105706164d2bSGeorge Zhang } 105806164d2bSGeorge Zhang 105906164d2bSGeorge Zhang /* 106006164d2bSGeorge Zhang * If we get here then we either failed to notify a local queuepair, or 106106164d2bSGeorge Zhang * we succeeded in all cases. Release the entry if required. 106206164d2bSGeorge Zhang */ 106306164d2bSGeorge Zhang 106406164d2bSGeorge Zhang entry->qp.ref_count--; 106506164d2bSGeorge Zhang if (entry->qp.ref_count == 0) 106606164d2bSGeorge Zhang qp_list_remove_entry(&qp_guest_endpoints, &entry->qp); 106706164d2bSGeorge Zhang 106806164d2bSGeorge Zhang /* If we didn't remove the entry, this could change once we unlock. */ 106906164d2bSGeorge Zhang if (entry) 107006164d2bSGeorge Zhang ref_count = entry->qp.ref_count; 107106164d2bSGeorge Zhang 107206164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 107306164d2bSGeorge Zhang 107406164d2bSGeorge Zhang if (ref_count == 0) 107506164d2bSGeorge Zhang qp_guest_endpoint_destroy(entry); 107606164d2bSGeorge Zhang 107706164d2bSGeorge Zhang return result; 107806164d2bSGeorge Zhang } 107906164d2bSGeorge Zhang 108006164d2bSGeorge Zhang /* 108106164d2bSGeorge Zhang * This functions handles the actual allocation of a VMCI queue 108206164d2bSGeorge Zhang * pair guest endpoint. Allocates physical pages for the queue 108306164d2bSGeorge Zhang * pair. It makes OS dependent calls through generic wrappers. 108406164d2bSGeorge Zhang */ 108506164d2bSGeorge Zhang static int qp_alloc_guest_work(struct vmci_handle *handle, 108606164d2bSGeorge Zhang struct vmci_queue **produce_q, 108706164d2bSGeorge Zhang u64 produce_size, 108806164d2bSGeorge Zhang struct vmci_queue **consume_q, 108906164d2bSGeorge Zhang u64 consume_size, 109006164d2bSGeorge Zhang u32 peer, 109106164d2bSGeorge Zhang u32 flags, 109206164d2bSGeorge Zhang u32 priv_flags) 109306164d2bSGeorge Zhang { 109406164d2bSGeorge Zhang const u64 num_produce_pages = 109542281d20SAndy King DIV_ROUND_UP(produce_size, PAGE_SIZE) + 1; 109606164d2bSGeorge Zhang const u64 num_consume_pages = 109742281d20SAndy King DIV_ROUND_UP(consume_size, PAGE_SIZE) + 1; 109806164d2bSGeorge Zhang void *my_produce_q = NULL; 109906164d2bSGeorge Zhang void *my_consume_q = NULL; 110006164d2bSGeorge Zhang int result; 110106164d2bSGeorge Zhang struct qp_guest_endpoint *queue_pair_entry = NULL; 110206164d2bSGeorge Zhang 110306164d2bSGeorge Zhang if (priv_flags != VMCI_NO_PRIVILEGE_FLAGS) 110406164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 110506164d2bSGeorge Zhang 110606164d2bSGeorge Zhang mutex_lock(&qp_guest_endpoints.mutex); 110706164d2bSGeorge Zhang 110806164d2bSGeorge Zhang queue_pair_entry = qp_guest_handle_to_entry(*handle); 110906164d2bSGeorge Zhang if (queue_pair_entry) { 111006164d2bSGeorge Zhang if (queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) { 111106164d2bSGeorge Zhang /* Local attach case. */ 111206164d2bSGeorge Zhang if (queue_pair_entry->qp.ref_count > 1) { 111306164d2bSGeorge Zhang pr_devel("Error attempting to attach more than once\n"); 111406164d2bSGeorge Zhang result = VMCI_ERROR_UNAVAILABLE; 111506164d2bSGeorge Zhang goto error_keep_entry; 111606164d2bSGeorge Zhang } 111706164d2bSGeorge Zhang 111806164d2bSGeorge Zhang if (queue_pair_entry->qp.produce_size != consume_size || 111906164d2bSGeorge Zhang queue_pair_entry->qp.consume_size != 112006164d2bSGeorge Zhang produce_size || 112106164d2bSGeorge Zhang queue_pair_entry->qp.flags != 112206164d2bSGeorge Zhang (flags & ~VMCI_QPFLAG_ATTACH_ONLY)) { 112306164d2bSGeorge Zhang pr_devel("Error mismatched queue pair in local attach\n"); 112406164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_MISMATCH; 112506164d2bSGeorge Zhang goto error_keep_entry; 112606164d2bSGeorge Zhang } 112706164d2bSGeorge Zhang 112806164d2bSGeorge Zhang /* 112906164d2bSGeorge Zhang * Do a local attach. We swap the consume and 113006164d2bSGeorge Zhang * produce queues for the attacher and deliver 113106164d2bSGeorge Zhang * an attach event. 113206164d2bSGeorge Zhang */ 113306164d2bSGeorge Zhang result = qp_notify_peer_local(true, *handle); 113406164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 113506164d2bSGeorge Zhang goto error_keep_entry; 113606164d2bSGeorge Zhang 113706164d2bSGeorge Zhang my_produce_q = queue_pair_entry->consume_q; 113806164d2bSGeorge Zhang my_consume_q = queue_pair_entry->produce_q; 113906164d2bSGeorge Zhang goto out; 114006164d2bSGeorge Zhang } 114106164d2bSGeorge Zhang 114206164d2bSGeorge Zhang result = VMCI_ERROR_ALREADY_EXISTS; 114306164d2bSGeorge Zhang goto error_keep_entry; 114406164d2bSGeorge Zhang } 114506164d2bSGeorge Zhang 114606164d2bSGeorge Zhang my_produce_q = qp_alloc_queue(produce_size, flags); 114706164d2bSGeorge Zhang if (!my_produce_q) { 114806164d2bSGeorge Zhang pr_warn("Error allocating pages for produce queue\n"); 114906164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 115006164d2bSGeorge Zhang goto error; 115106164d2bSGeorge Zhang } 115206164d2bSGeorge Zhang 115306164d2bSGeorge Zhang my_consume_q = qp_alloc_queue(consume_size, flags); 115406164d2bSGeorge Zhang if (!my_consume_q) { 115506164d2bSGeorge Zhang pr_warn("Error allocating pages for consume queue\n"); 115606164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 115706164d2bSGeorge Zhang goto error; 115806164d2bSGeorge Zhang } 115906164d2bSGeorge Zhang 116006164d2bSGeorge Zhang queue_pair_entry = qp_guest_endpoint_create(*handle, peer, flags, 116106164d2bSGeorge Zhang produce_size, consume_size, 116206164d2bSGeorge Zhang my_produce_q, my_consume_q); 116306164d2bSGeorge Zhang if (!queue_pair_entry) { 116406164d2bSGeorge Zhang pr_warn("Error allocating memory in %s\n", __func__); 116506164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 116606164d2bSGeorge Zhang goto error; 116706164d2bSGeorge Zhang } 116806164d2bSGeorge Zhang 116906164d2bSGeorge Zhang result = qp_alloc_ppn_set(my_produce_q, num_produce_pages, my_consume_q, 117006164d2bSGeorge Zhang num_consume_pages, 117106164d2bSGeorge Zhang &queue_pair_entry->ppn_set); 117206164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 117306164d2bSGeorge Zhang pr_warn("qp_alloc_ppn_set failed\n"); 117406164d2bSGeorge Zhang goto error; 117506164d2bSGeorge Zhang } 117606164d2bSGeorge Zhang 117706164d2bSGeorge Zhang /* 117806164d2bSGeorge Zhang * It's only necessary to notify the host if this queue pair will be 117906164d2bSGeorge Zhang * attached to from another context. 118006164d2bSGeorge Zhang */ 118106164d2bSGeorge Zhang if (queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) { 118206164d2bSGeorge Zhang /* Local create case. */ 118306164d2bSGeorge Zhang u32 context_id = vmci_get_context_id(); 118406164d2bSGeorge Zhang 118506164d2bSGeorge Zhang /* 118606164d2bSGeorge Zhang * Enforce similar checks on local queue pairs as we 118706164d2bSGeorge Zhang * do for regular ones. The handle's context must 118806164d2bSGeorge Zhang * match the creator or attacher context id (here they 118906164d2bSGeorge Zhang * are both the current context id) and the 119006164d2bSGeorge Zhang * attach-only flag cannot exist during create. We 119106164d2bSGeorge Zhang * also ensure specified peer is this context or an 119206164d2bSGeorge Zhang * invalid one. 119306164d2bSGeorge Zhang */ 119406164d2bSGeorge Zhang if (queue_pair_entry->qp.handle.context != context_id || 119506164d2bSGeorge Zhang (queue_pair_entry->qp.peer != VMCI_INVALID_ID && 119606164d2bSGeorge Zhang queue_pair_entry->qp.peer != context_id)) { 119706164d2bSGeorge Zhang result = VMCI_ERROR_NO_ACCESS; 119806164d2bSGeorge Zhang goto error; 119906164d2bSGeorge Zhang } 120006164d2bSGeorge Zhang 120106164d2bSGeorge Zhang if (queue_pair_entry->qp.flags & VMCI_QPFLAG_ATTACH_ONLY) { 120206164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 120306164d2bSGeorge Zhang goto error; 120406164d2bSGeorge Zhang } 120506164d2bSGeorge Zhang } else { 120606164d2bSGeorge Zhang result = qp_alloc_hypercall(queue_pair_entry); 120706164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 120806164d2bSGeorge Zhang pr_warn("qp_alloc_hypercall result = %d\n", result); 120906164d2bSGeorge Zhang goto error; 121006164d2bSGeorge Zhang } 121106164d2bSGeorge Zhang } 121206164d2bSGeorge Zhang 121306164d2bSGeorge Zhang qp_init_queue_mutex((struct vmci_queue *)my_produce_q, 121406164d2bSGeorge Zhang (struct vmci_queue *)my_consume_q); 121506164d2bSGeorge Zhang 121606164d2bSGeorge Zhang qp_list_add_entry(&qp_guest_endpoints, &queue_pair_entry->qp); 121706164d2bSGeorge Zhang 121806164d2bSGeorge Zhang out: 121906164d2bSGeorge Zhang queue_pair_entry->qp.ref_count++; 122006164d2bSGeorge Zhang *handle = queue_pair_entry->qp.handle; 122106164d2bSGeorge Zhang *produce_q = (struct vmci_queue *)my_produce_q; 122206164d2bSGeorge Zhang *consume_q = (struct vmci_queue *)my_consume_q; 122306164d2bSGeorge Zhang 122406164d2bSGeorge Zhang /* 122506164d2bSGeorge Zhang * We should initialize the queue pair header pages on a local 122606164d2bSGeorge Zhang * queue pair create. For non-local queue pairs, the 122706164d2bSGeorge Zhang * hypervisor initializes the header pages in the create step. 122806164d2bSGeorge Zhang */ 122906164d2bSGeorge Zhang if ((queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) && 123006164d2bSGeorge Zhang queue_pair_entry->qp.ref_count == 1) { 123106164d2bSGeorge Zhang vmci_q_header_init((*produce_q)->q_header, *handle); 123206164d2bSGeorge Zhang vmci_q_header_init((*consume_q)->q_header, *handle); 123306164d2bSGeorge Zhang } 123406164d2bSGeorge Zhang 123506164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 123606164d2bSGeorge Zhang 123706164d2bSGeorge Zhang return VMCI_SUCCESS; 123806164d2bSGeorge Zhang 123906164d2bSGeorge Zhang error: 124006164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 124106164d2bSGeorge Zhang if (queue_pair_entry) { 124206164d2bSGeorge Zhang /* The queues will be freed inside the destroy routine. */ 124306164d2bSGeorge Zhang qp_guest_endpoint_destroy(queue_pair_entry); 124406164d2bSGeorge Zhang } else { 124506164d2bSGeorge Zhang qp_free_queue(my_produce_q, produce_size); 124606164d2bSGeorge Zhang qp_free_queue(my_consume_q, consume_size); 124706164d2bSGeorge Zhang } 124806164d2bSGeorge Zhang return result; 124906164d2bSGeorge Zhang 125006164d2bSGeorge Zhang error_keep_entry: 125106164d2bSGeorge Zhang /* This path should only be used when an existing entry was found. */ 125206164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 125306164d2bSGeorge Zhang return result; 125406164d2bSGeorge Zhang } 125506164d2bSGeorge Zhang 125606164d2bSGeorge Zhang /* 125706164d2bSGeorge Zhang * The first endpoint issuing a queue pair allocation will create the state 125806164d2bSGeorge Zhang * of the queue pair in the queue pair broker. 125906164d2bSGeorge Zhang * 126006164d2bSGeorge Zhang * If the creator is a guest, it will associate a VMX virtual address range 126106164d2bSGeorge Zhang * with the queue pair as specified by the page_store. For compatibility with 126206164d2bSGeorge Zhang * older VMX'en, that would use a separate step to set the VMX virtual 126306164d2bSGeorge Zhang * address range, the virtual address range can be registered later using 126406164d2bSGeorge Zhang * vmci_qp_broker_set_page_store. In that case, a page_store of NULL should be 126506164d2bSGeorge Zhang * used. 126606164d2bSGeorge Zhang * 126706164d2bSGeorge Zhang * If the creator is the host, a page_store of NULL should be used as well, 126806164d2bSGeorge Zhang * since the host is not able to supply a page store for the queue pair. 126906164d2bSGeorge Zhang * 127006164d2bSGeorge Zhang * For older VMX and host callers, the queue pair will be created in the 127106164d2bSGeorge Zhang * VMCIQPB_CREATED_NO_MEM state, and for current VMX callers, it will be 127206164d2bSGeorge Zhang * created in VMCOQPB_CREATED_MEM state. 127306164d2bSGeorge Zhang */ 127406164d2bSGeorge Zhang static int qp_broker_create(struct vmci_handle handle, 127506164d2bSGeorge Zhang u32 peer, 127606164d2bSGeorge Zhang u32 flags, 127706164d2bSGeorge Zhang u32 priv_flags, 127806164d2bSGeorge Zhang u64 produce_size, 127906164d2bSGeorge Zhang u64 consume_size, 128006164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 128106164d2bSGeorge Zhang struct vmci_ctx *context, 128206164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 128306164d2bSGeorge Zhang void *client_data, struct qp_broker_entry **ent) 128406164d2bSGeorge Zhang { 128506164d2bSGeorge Zhang struct qp_broker_entry *entry = NULL; 128606164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 128706164d2bSGeorge Zhang bool is_local = flags & VMCI_QPFLAG_LOCAL; 128806164d2bSGeorge Zhang int result; 128906164d2bSGeorge Zhang u64 guest_produce_size; 129006164d2bSGeorge Zhang u64 guest_consume_size; 129106164d2bSGeorge Zhang 129206164d2bSGeorge Zhang /* Do not create if the caller asked not to. */ 129306164d2bSGeorge Zhang if (flags & VMCI_QPFLAG_ATTACH_ONLY) 129406164d2bSGeorge Zhang return VMCI_ERROR_NOT_FOUND; 129506164d2bSGeorge Zhang 129606164d2bSGeorge Zhang /* 129706164d2bSGeorge Zhang * Creator's context ID should match handle's context ID or the creator 129806164d2bSGeorge Zhang * must allow the context in handle's context ID as the "peer". 129906164d2bSGeorge Zhang */ 130006164d2bSGeorge Zhang if (handle.context != context_id && handle.context != peer) 130106164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 130206164d2bSGeorge Zhang 130306164d2bSGeorge Zhang if (VMCI_CONTEXT_IS_VM(context_id) && VMCI_CONTEXT_IS_VM(peer)) 130406164d2bSGeorge Zhang return VMCI_ERROR_DST_UNREACHABLE; 130506164d2bSGeorge Zhang 130606164d2bSGeorge Zhang /* 130706164d2bSGeorge Zhang * Creator's context ID for local queue pairs should match the 130806164d2bSGeorge Zhang * peer, if a peer is specified. 130906164d2bSGeorge Zhang */ 131006164d2bSGeorge Zhang if (is_local && peer != VMCI_INVALID_ID && context_id != peer) 131106164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 131206164d2bSGeorge Zhang 131306164d2bSGeorge Zhang entry = kzalloc(sizeof(*entry), GFP_ATOMIC); 131406164d2bSGeorge Zhang if (!entry) 131506164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 131606164d2bSGeorge Zhang 131706164d2bSGeorge Zhang if (vmci_ctx_get_id(context) == VMCI_HOST_CONTEXT_ID && !is_local) { 131806164d2bSGeorge Zhang /* 131906164d2bSGeorge Zhang * The queue pair broker entry stores values from the guest 132006164d2bSGeorge Zhang * point of view, so a creating host side endpoint should swap 132106164d2bSGeorge Zhang * produce and consume values -- unless it is a local queue 132206164d2bSGeorge Zhang * pair, in which case no swapping is necessary, since the local 132306164d2bSGeorge Zhang * attacher will swap queues. 132406164d2bSGeorge Zhang */ 132506164d2bSGeorge Zhang 132606164d2bSGeorge Zhang guest_produce_size = consume_size; 132706164d2bSGeorge Zhang guest_consume_size = produce_size; 132806164d2bSGeorge Zhang } else { 132906164d2bSGeorge Zhang guest_produce_size = produce_size; 133006164d2bSGeorge Zhang guest_consume_size = consume_size; 133106164d2bSGeorge Zhang } 133206164d2bSGeorge Zhang 133306164d2bSGeorge Zhang entry->qp.handle = handle; 133406164d2bSGeorge Zhang entry->qp.peer = peer; 133506164d2bSGeorge Zhang entry->qp.flags = flags; 133606164d2bSGeorge Zhang entry->qp.produce_size = guest_produce_size; 133706164d2bSGeorge Zhang entry->qp.consume_size = guest_consume_size; 133806164d2bSGeorge Zhang entry->qp.ref_count = 1; 133906164d2bSGeorge Zhang entry->create_id = context_id; 134006164d2bSGeorge Zhang entry->attach_id = VMCI_INVALID_ID; 134106164d2bSGeorge Zhang entry->state = VMCIQPB_NEW; 134206164d2bSGeorge Zhang entry->require_trusted_attach = 134306164d2bSGeorge Zhang !!(context->priv_flags & VMCI_PRIVILEGE_FLAG_RESTRICTED); 134406164d2bSGeorge Zhang entry->created_by_trusted = 134506164d2bSGeorge Zhang !!(priv_flags & VMCI_PRIVILEGE_FLAG_TRUSTED); 134606164d2bSGeorge Zhang entry->vmci_page_files = false; 134706164d2bSGeorge Zhang entry->wakeup_cb = wakeup_cb; 134806164d2bSGeorge Zhang entry->client_data = client_data; 134906164d2bSGeorge Zhang entry->produce_q = qp_host_alloc_queue(guest_produce_size); 135006164d2bSGeorge Zhang if (entry->produce_q == NULL) { 135106164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 135206164d2bSGeorge Zhang goto error; 135306164d2bSGeorge Zhang } 135406164d2bSGeorge Zhang entry->consume_q = qp_host_alloc_queue(guest_consume_size); 135506164d2bSGeorge Zhang if (entry->consume_q == NULL) { 135606164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 135706164d2bSGeorge Zhang goto error; 135806164d2bSGeorge Zhang } 135906164d2bSGeorge Zhang 136006164d2bSGeorge Zhang qp_init_queue_mutex(entry->produce_q, entry->consume_q); 136106164d2bSGeorge Zhang 136206164d2bSGeorge Zhang INIT_LIST_HEAD(&entry->qp.list_item); 136306164d2bSGeorge Zhang 136406164d2bSGeorge Zhang if (is_local) { 136506164d2bSGeorge Zhang u8 *tmp; 136606164d2bSGeorge Zhang 136706164d2bSGeorge Zhang entry->local_mem = kcalloc(QPE_NUM_PAGES(entry->qp), 136806164d2bSGeorge Zhang PAGE_SIZE, GFP_KERNEL); 136906164d2bSGeorge Zhang if (entry->local_mem == NULL) { 137006164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 137106164d2bSGeorge Zhang goto error; 137206164d2bSGeorge Zhang } 137306164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_MEM; 137406164d2bSGeorge Zhang entry->produce_q->q_header = entry->local_mem; 137506164d2bSGeorge Zhang tmp = (u8 *)entry->local_mem + PAGE_SIZE * 137642281d20SAndy King (DIV_ROUND_UP(entry->qp.produce_size, PAGE_SIZE) + 1); 137706164d2bSGeorge Zhang entry->consume_q->q_header = (struct vmci_queue_header *)tmp; 137806164d2bSGeorge Zhang } else if (page_store) { 137906164d2bSGeorge Zhang /* 138006164d2bSGeorge Zhang * The VMX already initialized the queue pair headers, so no 138106164d2bSGeorge Zhang * need for the kernel side to do that. 138206164d2bSGeorge Zhang */ 138306164d2bSGeorge Zhang result = qp_host_register_user_memory(page_store, 138406164d2bSGeorge Zhang entry->produce_q, 138506164d2bSGeorge Zhang entry->consume_q); 138606164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 138706164d2bSGeorge Zhang goto error; 138806164d2bSGeorge Zhang 138906164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_MEM; 139006164d2bSGeorge Zhang } else { 139106164d2bSGeorge Zhang /* 139206164d2bSGeorge Zhang * A create without a page_store may be either a host 139306164d2bSGeorge Zhang * side create (in which case we are waiting for the 139406164d2bSGeorge Zhang * guest side to supply the memory) or an old style 139506164d2bSGeorge Zhang * queue pair create (in which case we will expect a 139606164d2bSGeorge Zhang * set page store call as the next step). 139706164d2bSGeorge Zhang */ 139806164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_NO_MEM; 139906164d2bSGeorge Zhang } 140006164d2bSGeorge Zhang 140106164d2bSGeorge Zhang qp_list_add_entry(&qp_broker_list, &entry->qp); 140206164d2bSGeorge Zhang if (ent != NULL) 140306164d2bSGeorge Zhang *ent = entry; 140406164d2bSGeorge Zhang 140506164d2bSGeorge Zhang /* Add to resource obj */ 140606164d2bSGeorge Zhang result = vmci_resource_add(&entry->resource, 140706164d2bSGeorge Zhang VMCI_RESOURCE_TYPE_QPAIR_HOST, 140806164d2bSGeorge Zhang handle); 140906164d2bSGeorge Zhang if (result != VMCI_SUCCESS) { 141006164d2bSGeorge Zhang pr_warn("Failed to add new resource (handle=0x%x:0x%x), error: %d", 141106164d2bSGeorge Zhang handle.context, handle.resource, result); 141206164d2bSGeorge Zhang goto error; 141306164d2bSGeorge Zhang } 141406164d2bSGeorge Zhang 141506164d2bSGeorge Zhang entry->qp.handle = vmci_resource_handle(&entry->resource); 141606164d2bSGeorge Zhang if (is_local) { 141706164d2bSGeorge Zhang vmci_q_header_init(entry->produce_q->q_header, 141806164d2bSGeorge Zhang entry->qp.handle); 141906164d2bSGeorge Zhang vmci_q_header_init(entry->consume_q->q_header, 142006164d2bSGeorge Zhang entry->qp.handle); 142106164d2bSGeorge Zhang } 142206164d2bSGeorge Zhang 142306164d2bSGeorge Zhang vmci_ctx_qp_create(context, entry->qp.handle); 142406164d2bSGeorge Zhang 142506164d2bSGeorge Zhang return VMCI_SUCCESS; 142606164d2bSGeorge Zhang 142706164d2bSGeorge Zhang error: 142806164d2bSGeorge Zhang if (entry != NULL) { 142906164d2bSGeorge Zhang qp_host_free_queue(entry->produce_q, guest_produce_size); 143006164d2bSGeorge Zhang qp_host_free_queue(entry->consume_q, guest_consume_size); 143106164d2bSGeorge Zhang kfree(entry); 143206164d2bSGeorge Zhang } 143306164d2bSGeorge Zhang 143406164d2bSGeorge Zhang return result; 143506164d2bSGeorge Zhang } 143606164d2bSGeorge Zhang 143706164d2bSGeorge Zhang /* 143806164d2bSGeorge Zhang * Enqueues an event datagram to notify the peer VM attached to 143906164d2bSGeorge Zhang * the given queue pair handle about attach/detach event by the 144006164d2bSGeorge Zhang * given VM. Returns Payload size of datagram enqueued on 144106164d2bSGeorge Zhang * success, error code otherwise. 144206164d2bSGeorge Zhang */ 144306164d2bSGeorge Zhang static int qp_notify_peer(bool attach, 144406164d2bSGeorge Zhang struct vmci_handle handle, 144506164d2bSGeorge Zhang u32 my_id, 144606164d2bSGeorge Zhang u32 peer_id) 144706164d2bSGeorge Zhang { 144806164d2bSGeorge Zhang int rv; 144906164d2bSGeorge Zhang struct vmci_event_qp ev; 145006164d2bSGeorge Zhang 145106164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || my_id == VMCI_INVALID_ID || 145206164d2bSGeorge Zhang peer_id == VMCI_INVALID_ID) 145306164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 145406164d2bSGeorge Zhang 145506164d2bSGeorge Zhang /* 145606164d2bSGeorge Zhang * In vmci_ctx_enqueue_datagram() we enforce the upper limit on 145706164d2bSGeorge Zhang * number of pending events from the hypervisor to a given VM 145806164d2bSGeorge Zhang * otherwise a rogue VM could do an arbitrary number of attach 145906164d2bSGeorge Zhang * and detach operations causing memory pressure in the host 146006164d2bSGeorge Zhang * kernel. 146106164d2bSGeorge Zhang */ 146206164d2bSGeorge Zhang 146306164d2bSGeorge Zhang ev.msg.hdr.dst = vmci_make_handle(peer_id, VMCI_EVENT_HANDLER); 146406164d2bSGeorge Zhang ev.msg.hdr.src = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 146506164d2bSGeorge Zhang VMCI_CONTEXT_RESOURCE_ID); 146606164d2bSGeorge Zhang ev.msg.hdr.payload_size = sizeof(ev) - sizeof(ev.msg.hdr); 146706164d2bSGeorge Zhang ev.msg.event_data.event = attach ? 146806164d2bSGeorge Zhang VMCI_EVENT_QP_PEER_ATTACH : VMCI_EVENT_QP_PEER_DETACH; 146906164d2bSGeorge Zhang ev.payload.handle = handle; 147006164d2bSGeorge Zhang ev.payload.peer_id = my_id; 147106164d2bSGeorge Zhang 147206164d2bSGeorge Zhang rv = vmci_datagram_dispatch(VMCI_HYPERVISOR_CONTEXT_ID, 147306164d2bSGeorge Zhang &ev.msg.hdr, false); 147406164d2bSGeorge Zhang if (rv < VMCI_SUCCESS) 147506164d2bSGeorge Zhang pr_warn("Failed to enqueue queue_pair %s event datagram for context (ID=0x%x)\n", 147606164d2bSGeorge Zhang attach ? "ATTACH" : "DETACH", peer_id); 147706164d2bSGeorge Zhang 147806164d2bSGeorge Zhang return rv; 147906164d2bSGeorge Zhang } 148006164d2bSGeorge Zhang 148106164d2bSGeorge Zhang /* 148206164d2bSGeorge Zhang * The second endpoint issuing a queue pair allocation will attach to 148306164d2bSGeorge Zhang * the queue pair registered with the queue pair broker. 148406164d2bSGeorge Zhang * 148506164d2bSGeorge Zhang * If the attacher is a guest, it will associate a VMX virtual address 148606164d2bSGeorge Zhang * range with the queue pair as specified by the page_store. At this 148706164d2bSGeorge Zhang * point, the already attach host endpoint may start using the queue 148806164d2bSGeorge Zhang * pair, and an attach event is sent to it. For compatibility with 148906164d2bSGeorge Zhang * older VMX'en, that used a separate step to set the VMX virtual 149006164d2bSGeorge Zhang * address range, the virtual address range can be registered later 149106164d2bSGeorge Zhang * using vmci_qp_broker_set_page_store. In that case, a page_store of 149206164d2bSGeorge Zhang * NULL should be used, and the attach event will be generated once 149306164d2bSGeorge Zhang * the actual page store has been set. 149406164d2bSGeorge Zhang * 149506164d2bSGeorge Zhang * If the attacher is the host, a page_store of NULL should be used as 149606164d2bSGeorge Zhang * well, since the page store information is already set by the guest. 149706164d2bSGeorge Zhang * 149806164d2bSGeorge Zhang * For new VMX and host callers, the queue pair will be moved to the 149906164d2bSGeorge Zhang * VMCIQPB_ATTACHED_MEM state, and for older VMX callers, it will be 150006164d2bSGeorge Zhang * moved to the VMCOQPB_ATTACHED_NO_MEM state. 150106164d2bSGeorge Zhang */ 150206164d2bSGeorge Zhang static int qp_broker_attach(struct qp_broker_entry *entry, 150306164d2bSGeorge Zhang u32 peer, 150406164d2bSGeorge Zhang u32 flags, 150506164d2bSGeorge Zhang u32 priv_flags, 150606164d2bSGeorge Zhang u64 produce_size, 150706164d2bSGeorge Zhang u64 consume_size, 150806164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 150906164d2bSGeorge Zhang struct vmci_ctx *context, 151006164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 151106164d2bSGeorge Zhang void *client_data, 151206164d2bSGeorge Zhang struct qp_broker_entry **ent) 151306164d2bSGeorge Zhang { 151406164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 151506164d2bSGeorge Zhang bool is_local = flags & VMCI_QPFLAG_LOCAL; 151606164d2bSGeorge Zhang int result; 151706164d2bSGeorge Zhang 151806164d2bSGeorge Zhang if (entry->state != VMCIQPB_CREATED_NO_MEM && 151906164d2bSGeorge Zhang entry->state != VMCIQPB_CREATED_MEM) 152006164d2bSGeorge Zhang return VMCI_ERROR_UNAVAILABLE; 152106164d2bSGeorge Zhang 152206164d2bSGeorge Zhang if (is_local) { 152306164d2bSGeorge Zhang if (!(entry->qp.flags & VMCI_QPFLAG_LOCAL) || 152406164d2bSGeorge Zhang context_id != entry->create_id) { 152506164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 152606164d2bSGeorge Zhang } 152706164d2bSGeorge Zhang } else if (context_id == entry->create_id || 152806164d2bSGeorge Zhang context_id == entry->attach_id) { 152906164d2bSGeorge Zhang return VMCI_ERROR_ALREADY_EXISTS; 153006164d2bSGeorge Zhang } 153106164d2bSGeorge Zhang 153206164d2bSGeorge Zhang if (VMCI_CONTEXT_IS_VM(context_id) && 153306164d2bSGeorge Zhang VMCI_CONTEXT_IS_VM(entry->create_id)) 153406164d2bSGeorge Zhang return VMCI_ERROR_DST_UNREACHABLE; 153506164d2bSGeorge Zhang 153606164d2bSGeorge Zhang /* 153706164d2bSGeorge Zhang * If we are attaching from a restricted context then the queuepair 153806164d2bSGeorge Zhang * must have been created by a trusted endpoint. 153906164d2bSGeorge Zhang */ 154006164d2bSGeorge Zhang if ((context->priv_flags & VMCI_PRIVILEGE_FLAG_RESTRICTED) && 154106164d2bSGeorge Zhang !entry->created_by_trusted) 154206164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 154306164d2bSGeorge Zhang 154406164d2bSGeorge Zhang /* 154506164d2bSGeorge Zhang * If we are attaching to a queuepair that was created by a restricted 154606164d2bSGeorge Zhang * context then we must be trusted. 154706164d2bSGeorge Zhang */ 154806164d2bSGeorge Zhang if (entry->require_trusted_attach && 154906164d2bSGeorge Zhang (!(priv_flags & VMCI_PRIVILEGE_FLAG_TRUSTED))) 155006164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 155106164d2bSGeorge Zhang 155206164d2bSGeorge Zhang /* 155306164d2bSGeorge Zhang * If the creator specifies VMCI_INVALID_ID in "peer" field, access 155406164d2bSGeorge Zhang * control check is not performed. 155506164d2bSGeorge Zhang */ 155606164d2bSGeorge Zhang if (entry->qp.peer != VMCI_INVALID_ID && entry->qp.peer != context_id) 155706164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 155806164d2bSGeorge Zhang 155906164d2bSGeorge Zhang if (entry->create_id == VMCI_HOST_CONTEXT_ID) { 156006164d2bSGeorge Zhang /* 156106164d2bSGeorge Zhang * Do not attach if the caller doesn't support Host Queue Pairs 156206164d2bSGeorge Zhang * and a host created this queue pair. 156306164d2bSGeorge Zhang */ 156406164d2bSGeorge Zhang 156506164d2bSGeorge Zhang if (!vmci_ctx_supports_host_qp(context)) 156606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_RESOURCE; 156706164d2bSGeorge Zhang 156806164d2bSGeorge Zhang } else if (context_id == VMCI_HOST_CONTEXT_ID) { 156906164d2bSGeorge Zhang struct vmci_ctx *create_context; 157006164d2bSGeorge Zhang bool supports_host_qp; 157106164d2bSGeorge Zhang 157206164d2bSGeorge Zhang /* 157306164d2bSGeorge Zhang * Do not attach a host to a user created queue pair if that 157406164d2bSGeorge Zhang * user doesn't support host queue pair end points. 157506164d2bSGeorge Zhang */ 157606164d2bSGeorge Zhang 157706164d2bSGeorge Zhang create_context = vmci_ctx_get(entry->create_id); 157806164d2bSGeorge Zhang supports_host_qp = vmci_ctx_supports_host_qp(create_context); 157906164d2bSGeorge Zhang vmci_ctx_put(create_context); 158006164d2bSGeorge Zhang 158106164d2bSGeorge Zhang if (!supports_host_qp) 158206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_RESOURCE; 158306164d2bSGeorge Zhang } 158406164d2bSGeorge Zhang 158506164d2bSGeorge Zhang if ((entry->qp.flags & ~VMCI_QP_ASYMM) != (flags & ~VMCI_QP_ASYMM_PEER)) 158606164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 158706164d2bSGeorge Zhang 158806164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 158906164d2bSGeorge Zhang /* 159006164d2bSGeorge Zhang * The queue pair broker entry stores values from the guest 159106164d2bSGeorge Zhang * point of view, so an attaching guest should match the values 159206164d2bSGeorge Zhang * stored in the entry. 159306164d2bSGeorge Zhang */ 159406164d2bSGeorge Zhang 159506164d2bSGeorge Zhang if (entry->qp.produce_size != produce_size || 159606164d2bSGeorge Zhang entry->qp.consume_size != consume_size) { 159706164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 159806164d2bSGeorge Zhang } 159906164d2bSGeorge Zhang } else if (entry->qp.produce_size != consume_size || 160006164d2bSGeorge Zhang entry->qp.consume_size != produce_size) { 160106164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 160206164d2bSGeorge Zhang } 160306164d2bSGeorge Zhang 160406164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 160506164d2bSGeorge Zhang /* 160606164d2bSGeorge Zhang * If a guest attached to a queue pair, it will supply 160706164d2bSGeorge Zhang * the backing memory. If this is a pre NOVMVM vmx, 160806164d2bSGeorge Zhang * the backing memory will be supplied by calling 160906164d2bSGeorge Zhang * vmci_qp_broker_set_page_store() following the 161006164d2bSGeorge Zhang * return of the vmci_qp_broker_alloc() call. If it is 161106164d2bSGeorge Zhang * a vmx of version NOVMVM or later, the page store 161206164d2bSGeorge Zhang * must be supplied as part of the 161306164d2bSGeorge Zhang * vmci_qp_broker_alloc call. Under all circumstances 161406164d2bSGeorge Zhang * must the initially created queue pair not have any 161506164d2bSGeorge Zhang * memory associated with it already. 161606164d2bSGeorge Zhang */ 161706164d2bSGeorge Zhang 161806164d2bSGeorge Zhang if (entry->state != VMCIQPB_CREATED_NO_MEM) 161906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 162006164d2bSGeorge Zhang 162106164d2bSGeorge Zhang if (page_store != NULL) { 162206164d2bSGeorge Zhang /* 162306164d2bSGeorge Zhang * Patch up host state to point to guest 162406164d2bSGeorge Zhang * supplied memory. The VMX already 162506164d2bSGeorge Zhang * initialized the queue pair headers, so no 162606164d2bSGeorge Zhang * need for the kernel side to do that. 162706164d2bSGeorge Zhang */ 162806164d2bSGeorge Zhang 162906164d2bSGeorge Zhang result = qp_host_register_user_memory(page_store, 163006164d2bSGeorge Zhang entry->produce_q, 163106164d2bSGeorge Zhang entry->consume_q); 163206164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 163306164d2bSGeorge Zhang return result; 163406164d2bSGeorge Zhang 163506164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_MEM; 163606164d2bSGeorge Zhang } else { 163706164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_NO_MEM; 163806164d2bSGeorge Zhang } 163906164d2bSGeorge Zhang } else if (entry->state == VMCIQPB_CREATED_NO_MEM) { 164006164d2bSGeorge Zhang /* 164106164d2bSGeorge Zhang * The host side is attempting to attach to a queue 164206164d2bSGeorge Zhang * pair that doesn't have any memory associated with 164306164d2bSGeorge Zhang * it. This must be a pre NOVMVM vmx that hasn't set 164406164d2bSGeorge Zhang * the page store information yet, or a quiesced VM. 164506164d2bSGeorge Zhang */ 164606164d2bSGeorge Zhang 164706164d2bSGeorge Zhang return VMCI_ERROR_UNAVAILABLE; 164806164d2bSGeorge Zhang } else { 164906164d2bSGeorge Zhang /* The host side has successfully attached to a queue pair. */ 165006164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_MEM; 165106164d2bSGeorge Zhang } 165206164d2bSGeorge Zhang 165306164d2bSGeorge Zhang if (entry->state == VMCIQPB_ATTACHED_MEM) { 165406164d2bSGeorge Zhang result = 165506164d2bSGeorge Zhang qp_notify_peer(true, entry->qp.handle, context_id, 165606164d2bSGeorge Zhang entry->create_id); 165706164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 165806164d2bSGeorge Zhang pr_warn("Failed to notify peer (ID=0x%x) of attach to queue pair (handle=0x%x:0x%x)\n", 165906164d2bSGeorge Zhang entry->create_id, entry->qp.handle.context, 166006164d2bSGeorge Zhang entry->qp.handle.resource); 166106164d2bSGeorge Zhang } 166206164d2bSGeorge Zhang 166306164d2bSGeorge Zhang entry->attach_id = context_id; 166406164d2bSGeorge Zhang entry->qp.ref_count++; 166506164d2bSGeorge Zhang if (wakeup_cb) { 166606164d2bSGeorge Zhang entry->wakeup_cb = wakeup_cb; 166706164d2bSGeorge Zhang entry->client_data = client_data; 166806164d2bSGeorge Zhang } 166906164d2bSGeorge Zhang 167006164d2bSGeorge Zhang /* 167106164d2bSGeorge Zhang * When attaching to local queue pairs, the context already has 167206164d2bSGeorge Zhang * an entry tracking the queue pair, so don't add another one. 167306164d2bSGeorge Zhang */ 167406164d2bSGeorge Zhang if (!is_local) 167506164d2bSGeorge Zhang vmci_ctx_qp_create(context, entry->qp.handle); 167606164d2bSGeorge Zhang 167706164d2bSGeorge Zhang if (ent != NULL) 167806164d2bSGeorge Zhang *ent = entry; 167906164d2bSGeorge Zhang 168006164d2bSGeorge Zhang return VMCI_SUCCESS; 168106164d2bSGeorge Zhang } 168206164d2bSGeorge Zhang 168306164d2bSGeorge Zhang /* 168406164d2bSGeorge Zhang * queue_pair_Alloc for use when setting up queue pair endpoints 168506164d2bSGeorge Zhang * on the host. 168606164d2bSGeorge Zhang */ 168706164d2bSGeorge Zhang static int qp_broker_alloc(struct vmci_handle handle, 168806164d2bSGeorge Zhang u32 peer, 168906164d2bSGeorge Zhang u32 flags, 169006164d2bSGeorge Zhang u32 priv_flags, 169106164d2bSGeorge Zhang u64 produce_size, 169206164d2bSGeorge Zhang u64 consume_size, 169306164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 169406164d2bSGeorge Zhang struct vmci_ctx *context, 169506164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 169606164d2bSGeorge Zhang void *client_data, 169706164d2bSGeorge Zhang struct qp_broker_entry **ent, 169806164d2bSGeorge Zhang bool *swap) 169906164d2bSGeorge Zhang { 170006164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 170106164d2bSGeorge Zhang bool create; 170206164d2bSGeorge Zhang struct qp_broker_entry *entry = NULL; 170306164d2bSGeorge Zhang bool is_local = flags & VMCI_QPFLAG_LOCAL; 170406164d2bSGeorge Zhang int result; 170506164d2bSGeorge Zhang 170606164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || 170706164d2bSGeorge Zhang (flags & ~VMCI_QP_ALL_FLAGS) || is_local || 170806164d2bSGeorge Zhang !(produce_size || consume_size) || 170906164d2bSGeorge Zhang !context || context_id == VMCI_INVALID_ID || 171006164d2bSGeorge Zhang handle.context == VMCI_INVALID_ID) { 171106164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 171206164d2bSGeorge Zhang } 171306164d2bSGeorge Zhang 171406164d2bSGeorge Zhang if (page_store && !VMCI_QP_PAGESTORE_IS_WELLFORMED(page_store)) 171506164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 171606164d2bSGeorge Zhang 171706164d2bSGeorge Zhang /* 171806164d2bSGeorge Zhang * In the initial argument check, we ensure that non-vmkernel hosts 171906164d2bSGeorge Zhang * are not allowed to create local queue pairs. 172006164d2bSGeorge Zhang */ 172106164d2bSGeorge Zhang 172206164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 172306164d2bSGeorge Zhang 172406164d2bSGeorge Zhang if (!is_local && vmci_ctx_qp_exists(context, handle)) { 172506164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) already attached to queue pair (handle=0x%x:0x%x)\n", 172606164d2bSGeorge Zhang context_id, handle.context, handle.resource); 172706164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 172806164d2bSGeorge Zhang return VMCI_ERROR_ALREADY_EXISTS; 172906164d2bSGeorge Zhang } 173006164d2bSGeorge Zhang 173106164d2bSGeorge Zhang if (handle.resource != VMCI_INVALID_ID) 173206164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 173306164d2bSGeorge Zhang 173406164d2bSGeorge Zhang if (!entry) { 173506164d2bSGeorge Zhang create = true; 173606164d2bSGeorge Zhang result = 173706164d2bSGeorge Zhang qp_broker_create(handle, peer, flags, priv_flags, 173806164d2bSGeorge Zhang produce_size, consume_size, page_store, 173906164d2bSGeorge Zhang context, wakeup_cb, client_data, ent); 174006164d2bSGeorge Zhang } else { 174106164d2bSGeorge Zhang create = false; 174206164d2bSGeorge Zhang result = 174306164d2bSGeorge Zhang qp_broker_attach(entry, peer, flags, priv_flags, 174406164d2bSGeorge Zhang produce_size, consume_size, page_store, 174506164d2bSGeorge Zhang context, wakeup_cb, client_data, ent); 174606164d2bSGeorge Zhang } 174706164d2bSGeorge Zhang 174806164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 174906164d2bSGeorge Zhang 175006164d2bSGeorge Zhang if (swap) 175106164d2bSGeorge Zhang *swap = (context_id == VMCI_HOST_CONTEXT_ID) && 175206164d2bSGeorge Zhang !(create && is_local); 175306164d2bSGeorge Zhang 175406164d2bSGeorge Zhang return result; 175506164d2bSGeorge Zhang } 175606164d2bSGeorge Zhang 175706164d2bSGeorge Zhang /* 175806164d2bSGeorge Zhang * This function implements the kernel API for allocating a queue 175906164d2bSGeorge Zhang * pair. 176006164d2bSGeorge Zhang */ 176106164d2bSGeorge Zhang static int qp_alloc_host_work(struct vmci_handle *handle, 176206164d2bSGeorge Zhang struct vmci_queue **produce_q, 176306164d2bSGeorge Zhang u64 produce_size, 176406164d2bSGeorge Zhang struct vmci_queue **consume_q, 176506164d2bSGeorge Zhang u64 consume_size, 176606164d2bSGeorge Zhang u32 peer, 176706164d2bSGeorge Zhang u32 flags, 176806164d2bSGeorge Zhang u32 priv_flags, 176906164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 177006164d2bSGeorge Zhang void *client_data) 177106164d2bSGeorge Zhang { 177206164d2bSGeorge Zhang struct vmci_handle new_handle; 177306164d2bSGeorge Zhang struct vmci_ctx *context; 177406164d2bSGeorge Zhang struct qp_broker_entry *entry; 177506164d2bSGeorge Zhang int result; 177606164d2bSGeorge Zhang bool swap; 177706164d2bSGeorge Zhang 177806164d2bSGeorge Zhang if (vmci_handle_is_invalid(*handle)) { 177906164d2bSGeorge Zhang new_handle = vmci_make_handle( 178006164d2bSGeorge Zhang VMCI_HOST_CONTEXT_ID, VMCI_INVALID_ID); 178106164d2bSGeorge Zhang } else 178206164d2bSGeorge Zhang new_handle = *handle; 178306164d2bSGeorge Zhang 178406164d2bSGeorge Zhang context = vmci_ctx_get(VMCI_HOST_CONTEXT_ID); 178506164d2bSGeorge Zhang entry = NULL; 178606164d2bSGeorge Zhang result = 178706164d2bSGeorge Zhang qp_broker_alloc(new_handle, peer, flags, priv_flags, 178806164d2bSGeorge Zhang produce_size, consume_size, NULL, context, 178906164d2bSGeorge Zhang wakeup_cb, client_data, &entry, &swap); 179006164d2bSGeorge Zhang if (result == VMCI_SUCCESS) { 179106164d2bSGeorge Zhang if (swap) { 179206164d2bSGeorge Zhang /* 179306164d2bSGeorge Zhang * If this is a local queue pair, the attacher 179406164d2bSGeorge Zhang * will swap around produce and consume 179506164d2bSGeorge Zhang * queues. 179606164d2bSGeorge Zhang */ 179706164d2bSGeorge Zhang 179806164d2bSGeorge Zhang *produce_q = entry->consume_q; 179906164d2bSGeorge Zhang *consume_q = entry->produce_q; 180006164d2bSGeorge Zhang } else { 180106164d2bSGeorge Zhang *produce_q = entry->produce_q; 180206164d2bSGeorge Zhang *consume_q = entry->consume_q; 180306164d2bSGeorge Zhang } 180406164d2bSGeorge Zhang 180506164d2bSGeorge Zhang *handle = vmci_resource_handle(&entry->resource); 180606164d2bSGeorge Zhang } else { 180706164d2bSGeorge Zhang *handle = VMCI_INVALID_HANDLE; 180806164d2bSGeorge Zhang pr_devel("queue pair broker failed to alloc (result=%d)\n", 180906164d2bSGeorge Zhang result); 181006164d2bSGeorge Zhang } 181106164d2bSGeorge Zhang vmci_ctx_put(context); 181206164d2bSGeorge Zhang return result; 181306164d2bSGeorge Zhang } 181406164d2bSGeorge Zhang 181506164d2bSGeorge Zhang /* 181606164d2bSGeorge Zhang * Allocates a VMCI queue_pair. Only checks validity of input 181706164d2bSGeorge Zhang * arguments. The real work is done in the host or guest 181806164d2bSGeorge Zhang * specific function. 181906164d2bSGeorge Zhang */ 182006164d2bSGeorge Zhang int vmci_qp_alloc(struct vmci_handle *handle, 182106164d2bSGeorge Zhang struct vmci_queue **produce_q, 182206164d2bSGeorge Zhang u64 produce_size, 182306164d2bSGeorge Zhang struct vmci_queue **consume_q, 182406164d2bSGeorge Zhang u64 consume_size, 182506164d2bSGeorge Zhang u32 peer, 182606164d2bSGeorge Zhang u32 flags, 182706164d2bSGeorge Zhang u32 priv_flags, 182806164d2bSGeorge Zhang bool guest_endpoint, 182906164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 183006164d2bSGeorge Zhang void *client_data) 183106164d2bSGeorge Zhang { 183206164d2bSGeorge Zhang if (!handle || !produce_q || !consume_q || 183306164d2bSGeorge Zhang (!produce_size && !consume_size) || (flags & ~VMCI_QP_ALL_FLAGS)) 183406164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 183506164d2bSGeorge Zhang 183606164d2bSGeorge Zhang if (guest_endpoint) { 183706164d2bSGeorge Zhang return qp_alloc_guest_work(handle, produce_q, 183806164d2bSGeorge Zhang produce_size, consume_q, 183906164d2bSGeorge Zhang consume_size, peer, 184006164d2bSGeorge Zhang flags, priv_flags); 184106164d2bSGeorge Zhang } else { 184206164d2bSGeorge Zhang return qp_alloc_host_work(handle, produce_q, 184306164d2bSGeorge Zhang produce_size, consume_q, 184406164d2bSGeorge Zhang consume_size, peer, flags, 184506164d2bSGeorge Zhang priv_flags, wakeup_cb, client_data); 184606164d2bSGeorge Zhang } 184706164d2bSGeorge Zhang } 184806164d2bSGeorge Zhang 184906164d2bSGeorge Zhang /* 185006164d2bSGeorge Zhang * This function implements the host kernel API for detaching from 185106164d2bSGeorge Zhang * a queue pair. 185206164d2bSGeorge Zhang */ 185306164d2bSGeorge Zhang static int qp_detatch_host_work(struct vmci_handle handle) 185406164d2bSGeorge Zhang { 185506164d2bSGeorge Zhang int result; 185606164d2bSGeorge Zhang struct vmci_ctx *context; 185706164d2bSGeorge Zhang 185806164d2bSGeorge Zhang context = vmci_ctx_get(VMCI_HOST_CONTEXT_ID); 185906164d2bSGeorge Zhang 186006164d2bSGeorge Zhang result = vmci_qp_broker_detach(handle, context); 186106164d2bSGeorge Zhang 186206164d2bSGeorge Zhang vmci_ctx_put(context); 186306164d2bSGeorge Zhang return result; 186406164d2bSGeorge Zhang } 186506164d2bSGeorge Zhang 186606164d2bSGeorge Zhang /* 186706164d2bSGeorge Zhang * Detaches from a VMCI queue_pair. Only checks validity of input argument. 186806164d2bSGeorge Zhang * Real work is done in the host or guest specific function. 186906164d2bSGeorge Zhang */ 187006164d2bSGeorge Zhang static int qp_detatch(struct vmci_handle handle, bool guest_endpoint) 187106164d2bSGeorge Zhang { 187206164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle)) 187306164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 187406164d2bSGeorge Zhang 187506164d2bSGeorge Zhang if (guest_endpoint) 187606164d2bSGeorge Zhang return qp_detatch_guest_work(handle); 187706164d2bSGeorge Zhang else 187806164d2bSGeorge Zhang return qp_detatch_host_work(handle); 187906164d2bSGeorge Zhang } 188006164d2bSGeorge Zhang 188106164d2bSGeorge Zhang /* 188206164d2bSGeorge Zhang * Returns the entry from the head of the list. Assumes that the list is 188306164d2bSGeorge Zhang * locked. 188406164d2bSGeorge Zhang */ 188506164d2bSGeorge Zhang static struct qp_entry *qp_list_get_head(struct qp_list *qp_list) 188606164d2bSGeorge Zhang { 188706164d2bSGeorge Zhang if (!list_empty(&qp_list->head)) { 188806164d2bSGeorge Zhang struct qp_entry *entry = 188906164d2bSGeorge Zhang list_first_entry(&qp_list->head, struct qp_entry, 189006164d2bSGeorge Zhang list_item); 189106164d2bSGeorge Zhang return entry; 189206164d2bSGeorge Zhang } 189306164d2bSGeorge Zhang 189406164d2bSGeorge Zhang return NULL; 189506164d2bSGeorge Zhang } 189606164d2bSGeorge Zhang 189706164d2bSGeorge Zhang void vmci_qp_broker_exit(void) 189806164d2bSGeorge Zhang { 189906164d2bSGeorge Zhang struct qp_entry *entry; 190006164d2bSGeorge Zhang struct qp_broker_entry *be; 190106164d2bSGeorge Zhang 190206164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 190306164d2bSGeorge Zhang 190406164d2bSGeorge Zhang while ((entry = qp_list_get_head(&qp_broker_list))) { 190506164d2bSGeorge Zhang be = (struct qp_broker_entry *)entry; 190606164d2bSGeorge Zhang 190706164d2bSGeorge Zhang qp_list_remove_entry(&qp_broker_list, entry); 190806164d2bSGeorge Zhang kfree(be); 190906164d2bSGeorge Zhang } 191006164d2bSGeorge Zhang 191106164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 191206164d2bSGeorge Zhang } 191306164d2bSGeorge Zhang 191406164d2bSGeorge Zhang /* 191506164d2bSGeorge Zhang * Requests that a queue pair be allocated with the VMCI queue 191606164d2bSGeorge Zhang * pair broker. Allocates a queue pair entry if one does not 191706164d2bSGeorge Zhang * exist. Attaches to one if it exists, and retrieves the page 191806164d2bSGeorge Zhang * files backing that queue_pair. Assumes that the queue pair 191906164d2bSGeorge Zhang * broker lock is held. 192006164d2bSGeorge Zhang */ 192106164d2bSGeorge Zhang int vmci_qp_broker_alloc(struct vmci_handle handle, 192206164d2bSGeorge Zhang u32 peer, 192306164d2bSGeorge Zhang u32 flags, 192406164d2bSGeorge Zhang u32 priv_flags, 192506164d2bSGeorge Zhang u64 produce_size, 192606164d2bSGeorge Zhang u64 consume_size, 192706164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 192806164d2bSGeorge Zhang struct vmci_ctx *context) 192906164d2bSGeorge Zhang { 193006164d2bSGeorge Zhang return qp_broker_alloc(handle, peer, flags, priv_flags, 193106164d2bSGeorge Zhang produce_size, consume_size, 193206164d2bSGeorge Zhang page_store, context, NULL, NULL, NULL, NULL); 193306164d2bSGeorge Zhang } 193406164d2bSGeorge Zhang 193506164d2bSGeorge Zhang /* 193606164d2bSGeorge Zhang * VMX'en with versions lower than VMCI_VERSION_NOVMVM use a separate 193706164d2bSGeorge Zhang * step to add the UVAs of the VMX mapping of the queue pair. This function 193806164d2bSGeorge Zhang * provides backwards compatibility with such VMX'en, and takes care of 193906164d2bSGeorge Zhang * registering the page store for a queue pair previously allocated by the 194006164d2bSGeorge Zhang * VMX during create or attach. This function will move the queue pair state 194106164d2bSGeorge Zhang * to either from VMCIQBP_CREATED_NO_MEM to VMCIQBP_CREATED_MEM or 194206164d2bSGeorge Zhang * VMCIQBP_ATTACHED_NO_MEM to VMCIQBP_ATTACHED_MEM. If moving to the 194306164d2bSGeorge Zhang * attached state with memory, the queue pair is ready to be used by the 194406164d2bSGeorge Zhang * host peer, and an attached event will be generated. 194506164d2bSGeorge Zhang * 194606164d2bSGeorge Zhang * Assumes that the queue pair broker lock is held. 194706164d2bSGeorge Zhang * 194806164d2bSGeorge Zhang * This function is only used by the hosted platform, since there is no 194906164d2bSGeorge Zhang * issue with backwards compatibility for vmkernel. 195006164d2bSGeorge Zhang */ 195106164d2bSGeorge Zhang int vmci_qp_broker_set_page_store(struct vmci_handle handle, 195206164d2bSGeorge Zhang u64 produce_uva, 195306164d2bSGeorge Zhang u64 consume_uva, 195406164d2bSGeorge Zhang struct vmci_ctx *context) 195506164d2bSGeorge Zhang { 195606164d2bSGeorge Zhang struct qp_broker_entry *entry; 195706164d2bSGeorge Zhang int result; 195806164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 195906164d2bSGeorge Zhang 196006164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 196106164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) 196206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 196306164d2bSGeorge Zhang 196406164d2bSGeorge Zhang /* 196506164d2bSGeorge Zhang * We only support guest to host queue pairs, so the VMX must 196606164d2bSGeorge Zhang * supply UVAs for the mapped page files. 196706164d2bSGeorge Zhang */ 196806164d2bSGeorge Zhang 196906164d2bSGeorge Zhang if (produce_uva == 0 || consume_uva == 0) 197006164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 197106164d2bSGeorge Zhang 197206164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 197306164d2bSGeorge Zhang 197406164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 197506164d2bSGeorge Zhang pr_warn("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 197606164d2bSGeorge Zhang context_id, handle.context, handle.resource); 197706164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 197806164d2bSGeorge Zhang goto out; 197906164d2bSGeorge Zhang } 198006164d2bSGeorge Zhang 198106164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 198206164d2bSGeorge Zhang if (!entry) { 198306164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 198406164d2bSGeorge Zhang goto out; 198506164d2bSGeorge Zhang } 198606164d2bSGeorge Zhang 198706164d2bSGeorge Zhang /* 198806164d2bSGeorge Zhang * If I'm the owner then I can set the page store. 198906164d2bSGeorge Zhang * 199006164d2bSGeorge Zhang * Or, if a host created the queue_pair and I'm the attached peer 199106164d2bSGeorge Zhang * then I can set the page store. 199206164d2bSGeorge Zhang */ 199306164d2bSGeorge Zhang if (entry->create_id != context_id && 199406164d2bSGeorge Zhang (entry->create_id != VMCI_HOST_CONTEXT_ID || 199506164d2bSGeorge Zhang entry->attach_id != context_id)) { 199606164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTOWNER; 199706164d2bSGeorge Zhang goto out; 199806164d2bSGeorge Zhang } 199906164d2bSGeorge Zhang 200006164d2bSGeorge Zhang if (entry->state != VMCIQPB_CREATED_NO_MEM && 200106164d2bSGeorge Zhang entry->state != VMCIQPB_ATTACHED_NO_MEM) { 200206164d2bSGeorge Zhang result = VMCI_ERROR_UNAVAILABLE; 200306164d2bSGeorge Zhang goto out; 200406164d2bSGeorge Zhang } 200506164d2bSGeorge Zhang 200606164d2bSGeorge Zhang result = qp_host_get_user_memory(produce_uva, consume_uva, 200706164d2bSGeorge Zhang entry->produce_q, entry->consume_q); 200806164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 200906164d2bSGeorge Zhang goto out; 201006164d2bSGeorge Zhang 201106164d2bSGeorge Zhang result = qp_host_map_queues(entry->produce_q, entry->consume_q); 201206164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 201306164d2bSGeorge Zhang qp_host_unregister_user_memory(entry->produce_q, 201406164d2bSGeorge Zhang entry->consume_q); 201506164d2bSGeorge Zhang goto out; 201606164d2bSGeorge Zhang } 201706164d2bSGeorge Zhang 201806164d2bSGeorge Zhang if (entry->state == VMCIQPB_CREATED_NO_MEM) 201906164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_MEM; 202006164d2bSGeorge Zhang else 202106164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_MEM; 202206164d2bSGeorge Zhang 202306164d2bSGeorge Zhang entry->vmci_page_files = true; 202406164d2bSGeorge Zhang 202506164d2bSGeorge Zhang if (entry->state == VMCIQPB_ATTACHED_MEM) { 202606164d2bSGeorge Zhang result = 202706164d2bSGeorge Zhang qp_notify_peer(true, handle, context_id, entry->create_id); 202806164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 202906164d2bSGeorge Zhang pr_warn("Failed to notify peer (ID=0x%x) of attach to queue pair (handle=0x%x:0x%x)\n", 203006164d2bSGeorge Zhang entry->create_id, entry->qp.handle.context, 203106164d2bSGeorge Zhang entry->qp.handle.resource); 203206164d2bSGeorge Zhang } 203306164d2bSGeorge Zhang } 203406164d2bSGeorge Zhang 203506164d2bSGeorge Zhang result = VMCI_SUCCESS; 203606164d2bSGeorge Zhang out: 203706164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 203806164d2bSGeorge Zhang return result; 203906164d2bSGeorge Zhang } 204006164d2bSGeorge Zhang 204106164d2bSGeorge Zhang /* 204206164d2bSGeorge Zhang * Resets saved queue headers for the given QP broker 204306164d2bSGeorge Zhang * entry. Should be used when guest memory becomes available 204406164d2bSGeorge Zhang * again, or the guest detaches. 204506164d2bSGeorge Zhang */ 204606164d2bSGeorge Zhang static void qp_reset_saved_headers(struct qp_broker_entry *entry) 204706164d2bSGeorge Zhang { 204806164d2bSGeorge Zhang entry->produce_q->saved_header = NULL; 204906164d2bSGeorge Zhang entry->consume_q->saved_header = NULL; 205006164d2bSGeorge Zhang } 205106164d2bSGeorge Zhang 205206164d2bSGeorge Zhang /* 205306164d2bSGeorge Zhang * The main entry point for detaching from a queue pair registered with the 205406164d2bSGeorge Zhang * queue pair broker. If more than one endpoint is attached to the queue 205506164d2bSGeorge Zhang * pair, the first endpoint will mainly decrement a reference count and 205606164d2bSGeorge Zhang * generate a notification to its peer. The last endpoint will clean up 205706164d2bSGeorge Zhang * the queue pair state registered with the broker. 205806164d2bSGeorge Zhang * 205906164d2bSGeorge Zhang * When a guest endpoint detaches, it will unmap and unregister the guest 206006164d2bSGeorge Zhang * memory backing the queue pair. If the host is still attached, it will 206106164d2bSGeorge Zhang * no longer be able to access the queue pair content. 206206164d2bSGeorge Zhang * 206306164d2bSGeorge Zhang * If the queue pair is already in a state where there is no memory 206406164d2bSGeorge Zhang * registered for the queue pair (any *_NO_MEM state), it will transition to 206506164d2bSGeorge Zhang * the VMCIQPB_SHUTDOWN_NO_MEM state. This will also happen, if a guest 206606164d2bSGeorge Zhang * endpoint is the first of two endpoints to detach. If the host endpoint is 206706164d2bSGeorge Zhang * the first out of two to detach, the queue pair will move to the 206806164d2bSGeorge Zhang * VMCIQPB_SHUTDOWN_MEM state. 206906164d2bSGeorge Zhang */ 207006164d2bSGeorge Zhang int vmci_qp_broker_detach(struct vmci_handle handle, struct vmci_ctx *context) 207106164d2bSGeorge Zhang { 207206164d2bSGeorge Zhang struct qp_broker_entry *entry; 207306164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 207406164d2bSGeorge Zhang u32 peer_id; 207506164d2bSGeorge Zhang bool is_local = false; 207606164d2bSGeorge Zhang int result; 207706164d2bSGeorge Zhang 207806164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 207906164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) { 208006164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 208106164d2bSGeorge Zhang } 208206164d2bSGeorge Zhang 208306164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 208406164d2bSGeorge Zhang 208506164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 208606164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 208706164d2bSGeorge Zhang context_id, handle.context, handle.resource); 208806164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 208906164d2bSGeorge Zhang goto out; 209006164d2bSGeorge Zhang } 209106164d2bSGeorge Zhang 209206164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 209306164d2bSGeorge Zhang if (!entry) { 209406164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) reports being attached to queue pair(handle=0x%x:0x%x) that isn't present in broker\n", 209506164d2bSGeorge Zhang context_id, handle.context, handle.resource); 209606164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 209706164d2bSGeorge Zhang goto out; 209806164d2bSGeorge Zhang } 209906164d2bSGeorge Zhang 210006164d2bSGeorge Zhang if (context_id != entry->create_id && context_id != entry->attach_id) { 210106164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 210206164d2bSGeorge Zhang goto out; 210306164d2bSGeorge Zhang } 210406164d2bSGeorge Zhang 210506164d2bSGeorge Zhang if (context_id == entry->create_id) { 210606164d2bSGeorge Zhang peer_id = entry->attach_id; 210706164d2bSGeorge Zhang entry->create_id = VMCI_INVALID_ID; 210806164d2bSGeorge Zhang } else { 210906164d2bSGeorge Zhang peer_id = entry->create_id; 211006164d2bSGeorge Zhang entry->attach_id = VMCI_INVALID_ID; 211106164d2bSGeorge Zhang } 211206164d2bSGeorge Zhang entry->qp.ref_count--; 211306164d2bSGeorge Zhang 211406164d2bSGeorge Zhang is_local = entry->qp.flags & VMCI_QPFLAG_LOCAL; 211506164d2bSGeorge Zhang 211606164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 211706164d2bSGeorge Zhang bool headers_mapped; 211806164d2bSGeorge Zhang 211906164d2bSGeorge Zhang /* 212006164d2bSGeorge Zhang * Pre NOVMVM vmx'en may detach from a queue pair 212106164d2bSGeorge Zhang * before setting the page store, and in that case 212206164d2bSGeorge Zhang * there is no user memory to detach from. Also, more 212306164d2bSGeorge Zhang * recent VMX'en may detach from a queue pair in the 212406164d2bSGeorge Zhang * quiesced state. 212506164d2bSGeorge Zhang */ 212606164d2bSGeorge Zhang 212706164d2bSGeorge Zhang qp_acquire_queue_mutex(entry->produce_q); 212806164d2bSGeorge Zhang headers_mapped = entry->produce_q->q_header || 212906164d2bSGeorge Zhang entry->consume_q->q_header; 213006164d2bSGeorge Zhang if (QPBROKERSTATE_HAS_MEM(entry)) { 213106164d2bSGeorge Zhang result = 213206164d2bSGeorge Zhang qp_host_unmap_queues(INVALID_VMCI_GUEST_MEM_ID, 213306164d2bSGeorge Zhang entry->produce_q, 213406164d2bSGeorge Zhang entry->consume_q); 213506164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 213606164d2bSGeorge Zhang pr_warn("Failed to unmap queue headers for queue pair (handle=0x%x:0x%x,result=%d)\n", 213706164d2bSGeorge Zhang handle.context, handle.resource, 213806164d2bSGeorge Zhang result); 213906164d2bSGeorge Zhang 214006164d2bSGeorge Zhang qp_host_unregister_user_memory(entry->produce_q, 2141234b7f8dSGustavo A. R. Silva entry->consume_q); 214206164d2bSGeorge Zhang 214306164d2bSGeorge Zhang } 214406164d2bSGeorge Zhang 214506164d2bSGeorge Zhang if (!headers_mapped) 214606164d2bSGeorge Zhang qp_reset_saved_headers(entry); 214706164d2bSGeorge Zhang 214806164d2bSGeorge Zhang qp_release_queue_mutex(entry->produce_q); 214906164d2bSGeorge Zhang 215006164d2bSGeorge Zhang if (!headers_mapped && entry->wakeup_cb) 215106164d2bSGeorge Zhang entry->wakeup_cb(entry->client_data); 215206164d2bSGeorge Zhang 215306164d2bSGeorge Zhang } else { 215406164d2bSGeorge Zhang if (entry->wakeup_cb) { 215506164d2bSGeorge Zhang entry->wakeup_cb = NULL; 215606164d2bSGeorge Zhang entry->client_data = NULL; 215706164d2bSGeorge Zhang } 215806164d2bSGeorge Zhang } 215906164d2bSGeorge Zhang 216006164d2bSGeorge Zhang if (entry->qp.ref_count == 0) { 216106164d2bSGeorge Zhang qp_list_remove_entry(&qp_broker_list, &entry->qp); 216206164d2bSGeorge Zhang 216306164d2bSGeorge Zhang if (is_local) 216406164d2bSGeorge Zhang kfree(entry->local_mem); 216506164d2bSGeorge Zhang 216606164d2bSGeorge Zhang qp_cleanup_queue_mutex(entry->produce_q, entry->consume_q); 216706164d2bSGeorge Zhang qp_host_free_queue(entry->produce_q, entry->qp.produce_size); 216806164d2bSGeorge Zhang qp_host_free_queue(entry->consume_q, entry->qp.consume_size); 216906164d2bSGeorge Zhang /* Unlink from resource hash table and free callback */ 217006164d2bSGeorge Zhang vmci_resource_remove(&entry->resource); 217106164d2bSGeorge Zhang 217206164d2bSGeorge Zhang kfree(entry); 217306164d2bSGeorge Zhang 217406164d2bSGeorge Zhang vmci_ctx_qp_destroy(context, handle); 217506164d2bSGeorge Zhang } else { 217606164d2bSGeorge Zhang qp_notify_peer(false, handle, context_id, peer_id); 217706164d2bSGeorge Zhang if (context_id == VMCI_HOST_CONTEXT_ID && 217806164d2bSGeorge Zhang QPBROKERSTATE_HAS_MEM(entry)) { 217906164d2bSGeorge Zhang entry->state = VMCIQPB_SHUTDOWN_MEM; 218006164d2bSGeorge Zhang } else { 218106164d2bSGeorge Zhang entry->state = VMCIQPB_SHUTDOWN_NO_MEM; 218206164d2bSGeorge Zhang } 218306164d2bSGeorge Zhang 218406164d2bSGeorge Zhang if (!is_local) 218506164d2bSGeorge Zhang vmci_ctx_qp_destroy(context, handle); 218606164d2bSGeorge Zhang 218706164d2bSGeorge Zhang } 218806164d2bSGeorge Zhang result = VMCI_SUCCESS; 218906164d2bSGeorge Zhang out: 219006164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 219106164d2bSGeorge Zhang return result; 219206164d2bSGeorge Zhang } 219306164d2bSGeorge Zhang 219406164d2bSGeorge Zhang /* 219506164d2bSGeorge Zhang * Establishes the necessary mappings for a queue pair given a 219606164d2bSGeorge Zhang * reference to the queue pair guest memory. This is usually 219706164d2bSGeorge Zhang * called when a guest is unquiesced and the VMX is allowed to 219806164d2bSGeorge Zhang * map guest memory once again. 219906164d2bSGeorge Zhang */ 220006164d2bSGeorge Zhang int vmci_qp_broker_map(struct vmci_handle handle, 220106164d2bSGeorge Zhang struct vmci_ctx *context, 220206164d2bSGeorge Zhang u64 guest_mem) 220306164d2bSGeorge Zhang { 220406164d2bSGeorge Zhang struct qp_broker_entry *entry; 220506164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 220606164d2bSGeorge Zhang int result; 220706164d2bSGeorge Zhang 220806164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 220906164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) 221006164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 221106164d2bSGeorge Zhang 221206164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 221306164d2bSGeorge Zhang 221406164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 221506164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 221606164d2bSGeorge Zhang context_id, handle.context, handle.resource); 221706164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 221806164d2bSGeorge Zhang goto out; 221906164d2bSGeorge Zhang } 222006164d2bSGeorge Zhang 222106164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 222206164d2bSGeorge Zhang if (!entry) { 222306164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) reports being attached to queue pair (handle=0x%x:0x%x) that isn't present in broker\n", 222406164d2bSGeorge Zhang context_id, handle.context, handle.resource); 222506164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 222606164d2bSGeorge Zhang goto out; 222706164d2bSGeorge Zhang } 222806164d2bSGeorge Zhang 222906164d2bSGeorge Zhang if (context_id != entry->create_id && context_id != entry->attach_id) { 223006164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 223106164d2bSGeorge Zhang goto out; 223206164d2bSGeorge Zhang } 223306164d2bSGeorge Zhang 223406164d2bSGeorge Zhang result = VMCI_SUCCESS; 223506164d2bSGeorge Zhang 223606164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 223706164d2bSGeorge Zhang struct vmci_qp_page_store page_store; 223806164d2bSGeorge Zhang 223906164d2bSGeorge Zhang page_store.pages = guest_mem; 224006164d2bSGeorge Zhang page_store.len = QPE_NUM_PAGES(entry->qp); 224106164d2bSGeorge Zhang 224206164d2bSGeorge Zhang qp_acquire_queue_mutex(entry->produce_q); 224306164d2bSGeorge Zhang qp_reset_saved_headers(entry); 224406164d2bSGeorge Zhang result = 224506164d2bSGeorge Zhang qp_host_register_user_memory(&page_store, 224606164d2bSGeorge Zhang entry->produce_q, 224706164d2bSGeorge Zhang entry->consume_q); 224806164d2bSGeorge Zhang qp_release_queue_mutex(entry->produce_q); 224906164d2bSGeorge Zhang if (result == VMCI_SUCCESS) { 225006164d2bSGeorge Zhang /* Move state from *_NO_MEM to *_MEM */ 225106164d2bSGeorge Zhang 225206164d2bSGeorge Zhang entry->state++; 225306164d2bSGeorge Zhang 225406164d2bSGeorge Zhang if (entry->wakeup_cb) 225506164d2bSGeorge Zhang entry->wakeup_cb(entry->client_data); 225606164d2bSGeorge Zhang } 225706164d2bSGeorge Zhang } 225806164d2bSGeorge Zhang 225906164d2bSGeorge Zhang out: 226006164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 226106164d2bSGeorge Zhang return result; 226206164d2bSGeorge Zhang } 226306164d2bSGeorge Zhang 226406164d2bSGeorge Zhang /* 226506164d2bSGeorge Zhang * Saves a snapshot of the queue headers for the given QP broker 226606164d2bSGeorge Zhang * entry. Should be used when guest memory is unmapped. 226706164d2bSGeorge Zhang * Results: 226806164d2bSGeorge Zhang * VMCI_SUCCESS on success, appropriate error code if guest memory 226906164d2bSGeorge Zhang * can't be accessed.. 227006164d2bSGeorge Zhang */ 227106164d2bSGeorge Zhang static int qp_save_headers(struct qp_broker_entry *entry) 227206164d2bSGeorge Zhang { 227306164d2bSGeorge Zhang int result; 227406164d2bSGeorge Zhang 227506164d2bSGeorge Zhang if (entry->produce_q->saved_header != NULL && 227606164d2bSGeorge Zhang entry->consume_q->saved_header != NULL) { 227706164d2bSGeorge Zhang /* 227806164d2bSGeorge Zhang * If the headers have already been saved, we don't need to do 227906164d2bSGeorge Zhang * it again, and we don't want to map in the headers 228006164d2bSGeorge Zhang * unnecessarily. 228106164d2bSGeorge Zhang */ 228206164d2bSGeorge Zhang 228306164d2bSGeorge Zhang return VMCI_SUCCESS; 228406164d2bSGeorge Zhang } 228506164d2bSGeorge Zhang 228606164d2bSGeorge Zhang if (NULL == entry->produce_q->q_header || 228706164d2bSGeorge Zhang NULL == entry->consume_q->q_header) { 228806164d2bSGeorge Zhang result = qp_host_map_queues(entry->produce_q, entry->consume_q); 228906164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 229006164d2bSGeorge Zhang return result; 229106164d2bSGeorge Zhang } 229206164d2bSGeorge Zhang 229306164d2bSGeorge Zhang memcpy(&entry->saved_produce_q, entry->produce_q->q_header, 229406164d2bSGeorge Zhang sizeof(entry->saved_produce_q)); 229506164d2bSGeorge Zhang entry->produce_q->saved_header = &entry->saved_produce_q; 229606164d2bSGeorge Zhang memcpy(&entry->saved_consume_q, entry->consume_q->q_header, 229706164d2bSGeorge Zhang sizeof(entry->saved_consume_q)); 229806164d2bSGeorge Zhang entry->consume_q->saved_header = &entry->saved_consume_q; 229906164d2bSGeorge Zhang 230006164d2bSGeorge Zhang return VMCI_SUCCESS; 230106164d2bSGeorge Zhang } 230206164d2bSGeorge Zhang 230306164d2bSGeorge Zhang /* 230406164d2bSGeorge Zhang * Removes all references to the guest memory of a given queue pair, and 230506164d2bSGeorge Zhang * will move the queue pair from state *_MEM to *_NO_MEM. It is usually 230606164d2bSGeorge Zhang * called when a VM is being quiesced where access to guest memory should 230706164d2bSGeorge Zhang * avoided. 230806164d2bSGeorge Zhang */ 230906164d2bSGeorge Zhang int vmci_qp_broker_unmap(struct vmci_handle handle, 231006164d2bSGeorge Zhang struct vmci_ctx *context, 231106164d2bSGeorge Zhang u32 gid) 231206164d2bSGeorge Zhang { 231306164d2bSGeorge Zhang struct qp_broker_entry *entry; 231406164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 231506164d2bSGeorge Zhang int result; 231606164d2bSGeorge Zhang 231706164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 231806164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) 231906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 232006164d2bSGeorge Zhang 232106164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 232206164d2bSGeorge Zhang 232306164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 232406164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 232506164d2bSGeorge Zhang context_id, handle.context, handle.resource); 232606164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 232706164d2bSGeorge Zhang goto out; 232806164d2bSGeorge Zhang } 232906164d2bSGeorge Zhang 233006164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 233106164d2bSGeorge Zhang if (!entry) { 233206164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) reports being attached to queue pair (handle=0x%x:0x%x) that isn't present in broker\n", 233306164d2bSGeorge Zhang context_id, handle.context, handle.resource); 233406164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 233506164d2bSGeorge Zhang goto out; 233606164d2bSGeorge Zhang } 233706164d2bSGeorge Zhang 233806164d2bSGeorge Zhang if (context_id != entry->create_id && context_id != entry->attach_id) { 233906164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 234006164d2bSGeorge Zhang goto out; 234106164d2bSGeorge Zhang } 234206164d2bSGeorge Zhang 234306164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 234406164d2bSGeorge Zhang qp_acquire_queue_mutex(entry->produce_q); 234506164d2bSGeorge Zhang result = qp_save_headers(entry); 234606164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 234706164d2bSGeorge Zhang pr_warn("Failed to save queue headers for queue pair (handle=0x%x:0x%x,result=%d)\n", 234806164d2bSGeorge Zhang handle.context, handle.resource, result); 234906164d2bSGeorge Zhang 235006164d2bSGeorge Zhang qp_host_unmap_queues(gid, entry->produce_q, entry->consume_q); 235106164d2bSGeorge Zhang 235206164d2bSGeorge Zhang /* 235306164d2bSGeorge Zhang * On hosted, when we unmap queue pairs, the VMX will also 235406164d2bSGeorge Zhang * unmap the guest memory, so we invalidate the previously 235506164d2bSGeorge Zhang * registered memory. If the queue pair is mapped again at a 235606164d2bSGeorge Zhang * later point in time, we will need to reregister the user 235706164d2bSGeorge Zhang * memory with a possibly new user VA. 235806164d2bSGeorge Zhang */ 235906164d2bSGeorge Zhang qp_host_unregister_user_memory(entry->produce_q, 236006164d2bSGeorge Zhang entry->consume_q); 236106164d2bSGeorge Zhang 236206164d2bSGeorge Zhang /* 236306164d2bSGeorge Zhang * Move state from *_MEM to *_NO_MEM. 236406164d2bSGeorge Zhang */ 236506164d2bSGeorge Zhang entry->state--; 236606164d2bSGeorge Zhang 236706164d2bSGeorge Zhang qp_release_queue_mutex(entry->produce_q); 236806164d2bSGeorge Zhang } 236906164d2bSGeorge Zhang 237006164d2bSGeorge Zhang result = VMCI_SUCCESS; 237106164d2bSGeorge Zhang 237206164d2bSGeorge Zhang out: 237306164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 237406164d2bSGeorge Zhang return result; 237506164d2bSGeorge Zhang } 237606164d2bSGeorge Zhang 237706164d2bSGeorge Zhang /* 237806164d2bSGeorge Zhang * Destroys all guest queue pair endpoints. If active guest queue 237906164d2bSGeorge Zhang * pairs still exist, hypercalls to attempt detach from these 238006164d2bSGeorge Zhang * queue pairs will be made. Any failure to detach is silently 238106164d2bSGeorge Zhang * ignored. 238206164d2bSGeorge Zhang */ 238306164d2bSGeorge Zhang void vmci_qp_guest_endpoints_exit(void) 238406164d2bSGeorge Zhang { 238506164d2bSGeorge Zhang struct qp_entry *entry; 238606164d2bSGeorge Zhang struct qp_guest_endpoint *ep; 238706164d2bSGeorge Zhang 238806164d2bSGeorge Zhang mutex_lock(&qp_guest_endpoints.mutex); 238906164d2bSGeorge Zhang 239006164d2bSGeorge Zhang while ((entry = qp_list_get_head(&qp_guest_endpoints))) { 239106164d2bSGeorge Zhang ep = (struct qp_guest_endpoint *)entry; 239206164d2bSGeorge Zhang 239306164d2bSGeorge Zhang /* Don't make a hypercall for local queue_pairs. */ 239406164d2bSGeorge Zhang if (!(entry->flags & VMCI_QPFLAG_LOCAL)) 239506164d2bSGeorge Zhang qp_detatch_hypercall(entry->handle); 239606164d2bSGeorge Zhang 239706164d2bSGeorge Zhang /* We cannot fail the exit, so let's reset ref_count. */ 239806164d2bSGeorge Zhang entry->ref_count = 0; 239906164d2bSGeorge Zhang qp_list_remove_entry(&qp_guest_endpoints, entry); 240006164d2bSGeorge Zhang 240106164d2bSGeorge Zhang qp_guest_endpoint_destroy(ep); 240206164d2bSGeorge Zhang } 240306164d2bSGeorge Zhang 240406164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 240506164d2bSGeorge Zhang } 240606164d2bSGeorge Zhang 240706164d2bSGeorge Zhang /* 240806164d2bSGeorge Zhang * Helper routine that will lock the queue pair before subsequent 240906164d2bSGeorge Zhang * operations. 241006164d2bSGeorge Zhang * Note: Non-blocking on the host side is currently only implemented in ESX. 241106164d2bSGeorge Zhang * Since non-blocking isn't yet implemented on the host personality we 241206164d2bSGeorge Zhang * have no reason to acquire a spin lock. So to avoid the use of an 241306164d2bSGeorge Zhang * unnecessary lock only acquire the mutex if we can block. 241406164d2bSGeorge Zhang */ 241506164d2bSGeorge Zhang static void qp_lock(const struct vmci_qp *qpair) 241606164d2bSGeorge Zhang { 241706164d2bSGeorge Zhang qp_acquire_queue_mutex(qpair->produce_q); 241806164d2bSGeorge Zhang } 241906164d2bSGeorge Zhang 242006164d2bSGeorge Zhang /* 242106164d2bSGeorge Zhang * Helper routine that unlocks the queue pair after calling 242245412befSAndy King * qp_lock. 242306164d2bSGeorge Zhang */ 242406164d2bSGeorge Zhang static void qp_unlock(const struct vmci_qp *qpair) 242506164d2bSGeorge Zhang { 242606164d2bSGeorge Zhang qp_release_queue_mutex(qpair->produce_q); 242706164d2bSGeorge Zhang } 242806164d2bSGeorge Zhang 242906164d2bSGeorge Zhang /* 243006164d2bSGeorge Zhang * The queue headers may not be mapped at all times. If a queue is 243106164d2bSGeorge Zhang * currently not mapped, it will be attempted to do so. 243206164d2bSGeorge Zhang */ 243306164d2bSGeorge Zhang static int qp_map_queue_headers(struct vmci_queue *produce_q, 243445412befSAndy King struct vmci_queue *consume_q) 243506164d2bSGeorge Zhang { 243606164d2bSGeorge Zhang int result; 243706164d2bSGeorge Zhang 243806164d2bSGeorge Zhang if (NULL == produce_q->q_header || NULL == consume_q->q_header) { 243906164d2bSGeorge Zhang result = qp_host_map_queues(produce_q, consume_q); 244006164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 244106164d2bSGeorge Zhang return (produce_q->saved_header && 244206164d2bSGeorge Zhang consume_q->saved_header) ? 244306164d2bSGeorge Zhang VMCI_ERROR_QUEUEPAIR_NOT_READY : 244406164d2bSGeorge Zhang VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 244506164d2bSGeorge Zhang } 244606164d2bSGeorge Zhang 244706164d2bSGeorge Zhang return VMCI_SUCCESS; 244806164d2bSGeorge Zhang } 244906164d2bSGeorge Zhang 245006164d2bSGeorge Zhang /* 245106164d2bSGeorge Zhang * Helper routine that will retrieve the produce and consume 245206164d2bSGeorge Zhang * headers of a given queue pair. If the guest memory of the 245306164d2bSGeorge Zhang * queue pair is currently not available, the saved queue headers 245406164d2bSGeorge Zhang * will be returned, if these are available. 245506164d2bSGeorge Zhang */ 245606164d2bSGeorge Zhang static int qp_get_queue_headers(const struct vmci_qp *qpair, 245706164d2bSGeorge Zhang struct vmci_queue_header **produce_q_header, 245806164d2bSGeorge Zhang struct vmci_queue_header **consume_q_header) 245906164d2bSGeorge Zhang { 246006164d2bSGeorge Zhang int result; 246106164d2bSGeorge Zhang 246245412befSAndy King result = qp_map_queue_headers(qpair->produce_q, qpair->consume_q); 246306164d2bSGeorge Zhang if (result == VMCI_SUCCESS) { 246406164d2bSGeorge Zhang *produce_q_header = qpair->produce_q->q_header; 246506164d2bSGeorge Zhang *consume_q_header = qpair->consume_q->q_header; 246606164d2bSGeorge Zhang } else if (qpair->produce_q->saved_header && 246706164d2bSGeorge Zhang qpair->consume_q->saved_header) { 246806164d2bSGeorge Zhang *produce_q_header = qpair->produce_q->saved_header; 246906164d2bSGeorge Zhang *consume_q_header = qpair->consume_q->saved_header; 247006164d2bSGeorge Zhang result = VMCI_SUCCESS; 247106164d2bSGeorge Zhang } 247206164d2bSGeorge Zhang 247306164d2bSGeorge Zhang return result; 247406164d2bSGeorge Zhang } 247506164d2bSGeorge Zhang 247606164d2bSGeorge Zhang /* 247706164d2bSGeorge Zhang * Callback from VMCI queue pair broker indicating that a queue 247806164d2bSGeorge Zhang * pair that was previously not ready, now either is ready or 247906164d2bSGeorge Zhang * gone forever. 248006164d2bSGeorge Zhang */ 248106164d2bSGeorge Zhang static int qp_wakeup_cb(void *client_data) 248206164d2bSGeorge Zhang { 248306164d2bSGeorge Zhang struct vmci_qp *qpair = (struct vmci_qp *)client_data; 248406164d2bSGeorge Zhang 248506164d2bSGeorge Zhang qp_lock(qpair); 248606164d2bSGeorge Zhang while (qpair->blocked > 0) { 248706164d2bSGeorge Zhang qpair->blocked--; 248806164d2bSGeorge Zhang qpair->generation++; 248906164d2bSGeorge Zhang wake_up(&qpair->event); 249006164d2bSGeorge Zhang } 249106164d2bSGeorge Zhang qp_unlock(qpair); 249206164d2bSGeorge Zhang 249306164d2bSGeorge Zhang return VMCI_SUCCESS; 249406164d2bSGeorge Zhang } 249506164d2bSGeorge Zhang 249606164d2bSGeorge Zhang /* 249706164d2bSGeorge Zhang * Makes the calling thread wait for the queue pair to become 249806164d2bSGeorge Zhang * ready for host side access. Returns true when thread is 249906164d2bSGeorge Zhang * woken up after queue pair state change, false otherwise. 250006164d2bSGeorge Zhang */ 250106164d2bSGeorge Zhang static bool qp_wait_for_ready_queue(struct vmci_qp *qpair) 250206164d2bSGeorge Zhang { 250306164d2bSGeorge Zhang unsigned int generation; 250406164d2bSGeorge Zhang 250506164d2bSGeorge Zhang qpair->blocked++; 250606164d2bSGeorge Zhang generation = qpair->generation; 250706164d2bSGeorge Zhang qp_unlock(qpair); 250806164d2bSGeorge Zhang wait_event(qpair->event, generation != qpair->generation); 250906164d2bSGeorge Zhang qp_lock(qpair); 251006164d2bSGeorge Zhang 251106164d2bSGeorge Zhang return true; 251206164d2bSGeorge Zhang } 251306164d2bSGeorge Zhang 251406164d2bSGeorge Zhang /* 251506164d2bSGeorge Zhang * Enqueues a given buffer to the produce queue using the provided 251606164d2bSGeorge Zhang * function. As many bytes as possible (space available in the queue) 251706164d2bSGeorge Zhang * are enqueued. Assumes the queue->mutex has been acquired. Returns 251806164d2bSGeorge Zhang * VMCI_ERROR_QUEUEPAIR_NOSPACE if no space was available to enqueue 251906164d2bSGeorge Zhang * data, VMCI_ERROR_INVALID_SIZE, if any queue pointer is outside the 252006164d2bSGeorge Zhang * queue (as defined by the queue size), VMCI_ERROR_INVALID_ARGS, if 252106164d2bSGeorge Zhang * an error occured when accessing the buffer, 252206164d2bSGeorge Zhang * VMCI_ERROR_QUEUEPAIR_NOTATTACHED, if the queue pair pages aren't 252306164d2bSGeorge Zhang * available. Otherwise, the number of bytes written to the queue is 252406164d2bSGeorge Zhang * returned. Updates the tail pointer of the produce queue. 252506164d2bSGeorge Zhang */ 252606164d2bSGeorge Zhang static ssize_t qp_enqueue_locked(struct vmci_queue *produce_q, 252706164d2bSGeorge Zhang struct vmci_queue *consume_q, 252806164d2bSGeorge Zhang const u64 produce_q_size, 2529d1038084SAl Viro struct iov_iter *from) 253006164d2bSGeorge Zhang { 253106164d2bSGeorge Zhang s64 free_space; 253206164d2bSGeorge Zhang u64 tail; 2533d1038084SAl Viro size_t buf_size = iov_iter_count(from); 253406164d2bSGeorge Zhang size_t written; 253506164d2bSGeorge Zhang ssize_t result; 253606164d2bSGeorge Zhang 253745412befSAndy King result = qp_map_queue_headers(produce_q, consume_q); 253806164d2bSGeorge Zhang if (unlikely(result != VMCI_SUCCESS)) 253906164d2bSGeorge Zhang return result; 254006164d2bSGeorge Zhang 254106164d2bSGeorge Zhang free_space = vmci_q_header_free_space(produce_q->q_header, 254206164d2bSGeorge Zhang consume_q->q_header, 254306164d2bSGeorge Zhang produce_q_size); 254406164d2bSGeorge Zhang if (free_space == 0) 254506164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_NOSPACE; 254606164d2bSGeorge Zhang 254706164d2bSGeorge Zhang if (free_space < VMCI_SUCCESS) 254806164d2bSGeorge Zhang return (ssize_t) free_space; 254906164d2bSGeorge Zhang 255006164d2bSGeorge Zhang written = (size_t) (free_space > buf_size ? buf_size : free_space); 255106164d2bSGeorge Zhang tail = vmci_q_header_producer_tail(produce_q->q_header); 255206164d2bSGeorge Zhang if (likely(tail + written < produce_q_size)) { 2553d1038084SAl Viro result = qp_memcpy_to_queue_iter(produce_q, tail, from, written); 255406164d2bSGeorge Zhang } else { 255506164d2bSGeorge Zhang /* Tail pointer wraps around. */ 255606164d2bSGeorge Zhang 255706164d2bSGeorge Zhang const size_t tmp = (size_t) (produce_q_size - tail); 255806164d2bSGeorge Zhang 2559d1038084SAl Viro result = qp_memcpy_to_queue_iter(produce_q, tail, from, tmp); 256006164d2bSGeorge Zhang if (result >= VMCI_SUCCESS) 2561d1038084SAl Viro result = qp_memcpy_to_queue_iter(produce_q, 0, from, 256206164d2bSGeorge Zhang written - tmp); 256306164d2bSGeorge Zhang } 256406164d2bSGeorge Zhang 256506164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 256606164d2bSGeorge Zhang return result; 256706164d2bSGeorge Zhang 256806164d2bSGeorge Zhang vmci_q_header_add_producer_tail(produce_q->q_header, written, 256906164d2bSGeorge Zhang produce_q_size); 257006164d2bSGeorge Zhang return written; 257106164d2bSGeorge Zhang } 257206164d2bSGeorge Zhang 257306164d2bSGeorge Zhang /* 257406164d2bSGeorge Zhang * Dequeues data (if available) from the given consume queue. Writes data 257506164d2bSGeorge Zhang * to the user provided buffer using the provided function. 257606164d2bSGeorge Zhang * Assumes the queue->mutex has been acquired. 257706164d2bSGeorge Zhang * Results: 257806164d2bSGeorge Zhang * VMCI_ERROR_QUEUEPAIR_NODATA if no data was available to dequeue. 257906164d2bSGeorge Zhang * VMCI_ERROR_INVALID_SIZE, if any queue pointer is outside the queue 258006164d2bSGeorge Zhang * (as defined by the queue size). 258106164d2bSGeorge Zhang * VMCI_ERROR_INVALID_ARGS, if an error occured when accessing the buffer. 258206164d2bSGeorge Zhang * Otherwise the number of bytes dequeued is returned. 258306164d2bSGeorge Zhang * Side effects: 258406164d2bSGeorge Zhang * Updates the head pointer of the consume queue. 258506164d2bSGeorge Zhang */ 258606164d2bSGeorge Zhang static ssize_t qp_dequeue_locked(struct vmci_queue *produce_q, 258706164d2bSGeorge Zhang struct vmci_queue *consume_q, 258806164d2bSGeorge Zhang const u64 consume_q_size, 258953f58d8eSAl Viro struct iov_iter *to, 259045412befSAndy King bool update_consumer) 259106164d2bSGeorge Zhang { 259253f58d8eSAl Viro size_t buf_size = iov_iter_count(to); 259306164d2bSGeorge Zhang s64 buf_ready; 259406164d2bSGeorge Zhang u64 head; 259506164d2bSGeorge Zhang size_t read; 259606164d2bSGeorge Zhang ssize_t result; 259706164d2bSGeorge Zhang 259845412befSAndy King result = qp_map_queue_headers(produce_q, consume_q); 259906164d2bSGeorge Zhang if (unlikely(result != VMCI_SUCCESS)) 260006164d2bSGeorge Zhang return result; 260106164d2bSGeorge Zhang 260206164d2bSGeorge Zhang buf_ready = vmci_q_header_buf_ready(consume_q->q_header, 260306164d2bSGeorge Zhang produce_q->q_header, 260406164d2bSGeorge Zhang consume_q_size); 260506164d2bSGeorge Zhang if (buf_ready == 0) 260606164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_NODATA; 260706164d2bSGeorge Zhang 260806164d2bSGeorge Zhang if (buf_ready < VMCI_SUCCESS) 260906164d2bSGeorge Zhang return (ssize_t) buf_ready; 261006164d2bSGeorge Zhang 261106164d2bSGeorge Zhang read = (size_t) (buf_ready > buf_size ? buf_size : buf_ready); 261206164d2bSGeorge Zhang head = vmci_q_header_consumer_head(produce_q->q_header); 261306164d2bSGeorge Zhang if (likely(head + read < consume_q_size)) { 261453f58d8eSAl Viro result = qp_memcpy_from_queue_iter(to, consume_q, head, read); 261506164d2bSGeorge Zhang } else { 261606164d2bSGeorge Zhang /* Head pointer wraps around. */ 261706164d2bSGeorge Zhang 261806164d2bSGeorge Zhang const size_t tmp = (size_t) (consume_q_size - head); 261906164d2bSGeorge Zhang 262053f58d8eSAl Viro result = qp_memcpy_from_queue_iter(to, consume_q, head, tmp); 262106164d2bSGeorge Zhang if (result >= VMCI_SUCCESS) 262253f58d8eSAl Viro result = qp_memcpy_from_queue_iter(to, consume_q, 0, 262306164d2bSGeorge Zhang read - tmp); 262406164d2bSGeorge Zhang 262506164d2bSGeorge Zhang } 262606164d2bSGeorge Zhang 262706164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 262806164d2bSGeorge Zhang return result; 262906164d2bSGeorge Zhang 263006164d2bSGeorge Zhang if (update_consumer) 263106164d2bSGeorge Zhang vmci_q_header_add_consumer_head(produce_q->q_header, 263206164d2bSGeorge Zhang read, consume_q_size); 263306164d2bSGeorge Zhang 263406164d2bSGeorge Zhang return read; 263506164d2bSGeorge Zhang } 263606164d2bSGeorge Zhang 263706164d2bSGeorge Zhang /* 263806164d2bSGeorge Zhang * vmci_qpair_alloc() - Allocates a queue pair. 263906164d2bSGeorge Zhang * @qpair: Pointer for the new vmci_qp struct. 264006164d2bSGeorge Zhang * @handle: Handle to track the resource. 264106164d2bSGeorge Zhang * @produce_qsize: Desired size of the producer queue. 264206164d2bSGeorge Zhang * @consume_qsize: Desired size of the consumer queue. 264306164d2bSGeorge Zhang * @peer: ContextID of the peer. 264406164d2bSGeorge Zhang * @flags: VMCI flags. 264506164d2bSGeorge Zhang * @priv_flags: VMCI priviledge flags. 264606164d2bSGeorge Zhang * 264706164d2bSGeorge Zhang * This is the client interface for allocating the memory for a 264806164d2bSGeorge Zhang * vmci_qp structure and then attaching to the underlying 264906164d2bSGeorge Zhang * queue. If an error occurs allocating the memory for the 265006164d2bSGeorge Zhang * vmci_qp structure no attempt is made to attach. If an 265106164d2bSGeorge Zhang * error occurs attaching, then the structure is freed. 265206164d2bSGeorge Zhang */ 265306164d2bSGeorge Zhang int vmci_qpair_alloc(struct vmci_qp **qpair, 265406164d2bSGeorge Zhang struct vmci_handle *handle, 265506164d2bSGeorge Zhang u64 produce_qsize, 265606164d2bSGeorge Zhang u64 consume_qsize, 265706164d2bSGeorge Zhang u32 peer, 265806164d2bSGeorge Zhang u32 flags, 265906164d2bSGeorge Zhang u32 priv_flags) 266006164d2bSGeorge Zhang { 266106164d2bSGeorge Zhang struct vmci_qp *my_qpair; 266206164d2bSGeorge Zhang int retval; 266306164d2bSGeorge Zhang struct vmci_handle src = VMCI_INVALID_HANDLE; 266406164d2bSGeorge Zhang struct vmci_handle dst = vmci_make_handle(peer, VMCI_INVALID_ID); 266506164d2bSGeorge Zhang enum vmci_route route; 266606164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb; 266706164d2bSGeorge Zhang void *client_data; 266806164d2bSGeorge Zhang 266906164d2bSGeorge Zhang /* 267006164d2bSGeorge Zhang * Restrict the size of a queuepair. The device already 267106164d2bSGeorge Zhang * enforces a limit on the total amount of memory that can be 267206164d2bSGeorge Zhang * allocated to queuepairs for a guest. However, we try to 267306164d2bSGeorge Zhang * allocate this memory before we make the queuepair 267406164d2bSGeorge Zhang * allocation hypercall. On Linux, we allocate each page 267506164d2bSGeorge Zhang * separately, which means rather than fail, the guest will 267606164d2bSGeorge Zhang * thrash while it tries to allocate, and will become 267706164d2bSGeorge Zhang * increasingly unresponsive to the point where it appears to 267806164d2bSGeorge Zhang * be hung. So we place a limit on the size of an individual 267906164d2bSGeorge Zhang * queuepair here, and leave the device to enforce the 268006164d2bSGeorge Zhang * restriction on total queuepair memory. (Note that this 268106164d2bSGeorge Zhang * doesn't prevent all cases; a user with only this much 268206164d2bSGeorge Zhang * physical memory could still get into trouble.) The error 268306164d2bSGeorge Zhang * used by the device is NO_RESOURCES, so use that here too. 268406164d2bSGeorge Zhang */ 268506164d2bSGeorge Zhang 268606164d2bSGeorge Zhang if (produce_qsize + consume_qsize < max(produce_qsize, consume_qsize) || 268706164d2bSGeorge Zhang produce_qsize + consume_qsize > VMCI_MAX_GUEST_QP_MEMORY) 268806164d2bSGeorge Zhang return VMCI_ERROR_NO_RESOURCES; 268906164d2bSGeorge Zhang 269006164d2bSGeorge Zhang retval = vmci_route(&src, &dst, false, &route); 269106164d2bSGeorge Zhang if (retval < VMCI_SUCCESS) 269206164d2bSGeorge Zhang route = vmci_guest_code_active() ? 269306164d2bSGeorge Zhang VMCI_ROUTE_AS_GUEST : VMCI_ROUTE_AS_HOST; 269406164d2bSGeorge Zhang 269545412befSAndy King if (flags & (VMCI_QPFLAG_NONBLOCK | VMCI_QPFLAG_PINNED)) { 269645412befSAndy King pr_devel("NONBLOCK OR PINNED set"); 269706164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 269806164d2bSGeorge Zhang } 269906164d2bSGeorge Zhang 270006164d2bSGeorge Zhang my_qpair = kzalloc(sizeof(*my_qpair), GFP_KERNEL); 270106164d2bSGeorge Zhang if (!my_qpair) 270206164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 270306164d2bSGeorge Zhang 270406164d2bSGeorge Zhang my_qpair->produce_q_size = produce_qsize; 270506164d2bSGeorge Zhang my_qpair->consume_q_size = consume_qsize; 270606164d2bSGeorge Zhang my_qpair->peer = peer; 270706164d2bSGeorge Zhang my_qpair->flags = flags; 270806164d2bSGeorge Zhang my_qpair->priv_flags = priv_flags; 270906164d2bSGeorge Zhang 271006164d2bSGeorge Zhang wakeup_cb = NULL; 271106164d2bSGeorge Zhang client_data = NULL; 271206164d2bSGeorge Zhang 271306164d2bSGeorge Zhang if (VMCI_ROUTE_AS_HOST == route) { 271406164d2bSGeorge Zhang my_qpair->guest_endpoint = false; 271506164d2bSGeorge Zhang if (!(flags & VMCI_QPFLAG_LOCAL)) { 271606164d2bSGeorge Zhang my_qpair->blocked = 0; 271706164d2bSGeorge Zhang my_qpair->generation = 0; 271806164d2bSGeorge Zhang init_waitqueue_head(&my_qpair->event); 271906164d2bSGeorge Zhang wakeup_cb = qp_wakeup_cb; 272006164d2bSGeorge Zhang client_data = (void *)my_qpair; 272106164d2bSGeorge Zhang } 272206164d2bSGeorge Zhang } else { 272306164d2bSGeorge Zhang my_qpair->guest_endpoint = true; 272406164d2bSGeorge Zhang } 272506164d2bSGeorge Zhang 272606164d2bSGeorge Zhang retval = vmci_qp_alloc(handle, 272706164d2bSGeorge Zhang &my_qpair->produce_q, 272806164d2bSGeorge Zhang my_qpair->produce_q_size, 272906164d2bSGeorge Zhang &my_qpair->consume_q, 273006164d2bSGeorge Zhang my_qpair->consume_q_size, 273106164d2bSGeorge Zhang my_qpair->peer, 273206164d2bSGeorge Zhang my_qpair->flags, 273306164d2bSGeorge Zhang my_qpair->priv_flags, 273406164d2bSGeorge Zhang my_qpair->guest_endpoint, 273506164d2bSGeorge Zhang wakeup_cb, client_data); 273606164d2bSGeorge Zhang 273706164d2bSGeorge Zhang if (retval < VMCI_SUCCESS) { 273806164d2bSGeorge Zhang kfree(my_qpair); 273906164d2bSGeorge Zhang return retval; 274006164d2bSGeorge Zhang } 274106164d2bSGeorge Zhang 274206164d2bSGeorge Zhang *qpair = my_qpair; 274306164d2bSGeorge Zhang my_qpair->handle = *handle; 274406164d2bSGeorge Zhang 274506164d2bSGeorge Zhang return retval; 274606164d2bSGeorge Zhang } 274706164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_alloc); 274806164d2bSGeorge Zhang 274906164d2bSGeorge Zhang /* 275006164d2bSGeorge Zhang * vmci_qpair_detach() - Detatches the client from a queue pair. 275106164d2bSGeorge Zhang * @qpair: Reference of a pointer to the qpair struct. 275206164d2bSGeorge Zhang * 275306164d2bSGeorge Zhang * This is the client interface for detaching from a VMCIQPair. 275406164d2bSGeorge Zhang * Note that this routine will free the memory allocated for the 275506164d2bSGeorge Zhang * vmci_qp structure too. 275606164d2bSGeorge Zhang */ 275706164d2bSGeorge Zhang int vmci_qpair_detach(struct vmci_qp **qpair) 275806164d2bSGeorge Zhang { 275906164d2bSGeorge Zhang int result; 276006164d2bSGeorge Zhang struct vmci_qp *old_qpair; 276106164d2bSGeorge Zhang 276206164d2bSGeorge Zhang if (!qpair || !(*qpair)) 276306164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 276406164d2bSGeorge Zhang 276506164d2bSGeorge Zhang old_qpair = *qpair; 276606164d2bSGeorge Zhang result = qp_detatch(old_qpair->handle, old_qpair->guest_endpoint); 276706164d2bSGeorge Zhang 276806164d2bSGeorge Zhang /* 276906164d2bSGeorge Zhang * The guest can fail to detach for a number of reasons, and 277006164d2bSGeorge Zhang * if it does so, it will cleanup the entry (if there is one). 277106164d2bSGeorge Zhang * The host can fail too, but it won't cleanup the entry 277206164d2bSGeorge Zhang * immediately, it will do that later when the context is 277306164d2bSGeorge Zhang * freed. Either way, we need to release the qpair struct 277406164d2bSGeorge Zhang * here; there isn't much the caller can do, and we don't want 277506164d2bSGeorge Zhang * to leak. 277606164d2bSGeorge Zhang */ 277706164d2bSGeorge Zhang 277806164d2bSGeorge Zhang memset(old_qpair, 0, sizeof(*old_qpair)); 277906164d2bSGeorge Zhang old_qpair->handle = VMCI_INVALID_HANDLE; 278006164d2bSGeorge Zhang old_qpair->peer = VMCI_INVALID_ID; 278106164d2bSGeorge Zhang kfree(old_qpair); 278206164d2bSGeorge Zhang *qpair = NULL; 278306164d2bSGeorge Zhang 278406164d2bSGeorge Zhang return result; 278506164d2bSGeorge Zhang } 278606164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_detach); 278706164d2bSGeorge Zhang 278806164d2bSGeorge Zhang /* 278906164d2bSGeorge Zhang * vmci_qpair_get_produce_indexes() - Retrieves the indexes of the producer. 279006164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 279106164d2bSGeorge Zhang * @producer_tail: Reference used for storing producer tail index. 279206164d2bSGeorge Zhang * @consumer_head: Reference used for storing the consumer head index. 279306164d2bSGeorge Zhang * 279406164d2bSGeorge Zhang * This is the client interface for getting the current indexes of the 279506164d2bSGeorge Zhang * QPair from the point of the view of the caller as the producer. 279606164d2bSGeorge Zhang */ 279706164d2bSGeorge Zhang int vmci_qpair_get_produce_indexes(const struct vmci_qp *qpair, 279806164d2bSGeorge Zhang u64 *producer_tail, 279906164d2bSGeorge Zhang u64 *consumer_head) 280006164d2bSGeorge Zhang { 280106164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 280206164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 280306164d2bSGeorge Zhang int result; 280406164d2bSGeorge Zhang 280506164d2bSGeorge Zhang if (!qpair) 280606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 280706164d2bSGeorge Zhang 280806164d2bSGeorge Zhang qp_lock(qpair); 280906164d2bSGeorge Zhang result = 281006164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 281106164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 281206164d2bSGeorge Zhang vmci_q_header_get_pointers(produce_q_header, consume_q_header, 281306164d2bSGeorge Zhang producer_tail, consumer_head); 281406164d2bSGeorge Zhang qp_unlock(qpair); 281506164d2bSGeorge Zhang 281606164d2bSGeorge Zhang if (result == VMCI_SUCCESS && 281706164d2bSGeorge Zhang ((producer_tail && *producer_tail >= qpair->produce_q_size) || 281806164d2bSGeorge Zhang (consumer_head && *consumer_head >= qpair->produce_q_size))) 281906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_SIZE; 282006164d2bSGeorge Zhang 282106164d2bSGeorge Zhang return result; 282206164d2bSGeorge Zhang } 282306164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_get_produce_indexes); 282406164d2bSGeorge Zhang 282506164d2bSGeorge Zhang /* 28263f8b6fb7SMasahiro Yamada * vmci_qpair_get_consume_indexes() - Retrieves the indexes of the consumer. 282706164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 282806164d2bSGeorge Zhang * @consumer_tail: Reference used for storing consumer tail index. 282906164d2bSGeorge Zhang * @producer_head: Reference used for storing the producer head index. 283006164d2bSGeorge Zhang * 283106164d2bSGeorge Zhang * This is the client interface for getting the current indexes of the 283206164d2bSGeorge Zhang * QPair from the point of the view of the caller as the consumer. 283306164d2bSGeorge Zhang */ 283406164d2bSGeorge Zhang int vmci_qpair_get_consume_indexes(const struct vmci_qp *qpair, 283506164d2bSGeorge Zhang u64 *consumer_tail, 283606164d2bSGeorge Zhang u64 *producer_head) 283706164d2bSGeorge Zhang { 283806164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 283906164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 284006164d2bSGeorge Zhang int result; 284106164d2bSGeorge Zhang 284206164d2bSGeorge Zhang if (!qpair) 284306164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 284406164d2bSGeorge Zhang 284506164d2bSGeorge Zhang qp_lock(qpair); 284606164d2bSGeorge Zhang result = 284706164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 284806164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 284906164d2bSGeorge Zhang vmci_q_header_get_pointers(consume_q_header, produce_q_header, 285006164d2bSGeorge Zhang consumer_tail, producer_head); 285106164d2bSGeorge Zhang qp_unlock(qpair); 285206164d2bSGeorge Zhang 285306164d2bSGeorge Zhang if (result == VMCI_SUCCESS && 285406164d2bSGeorge Zhang ((consumer_tail && *consumer_tail >= qpair->consume_q_size) || 285506164d2bSGeorge Zhang (producer_head && *producer_head >= qpair->consume_q_size))) 285606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_SIZE; 285706164d2bSGeorge Zhang 285806164d2bSGeorge Zhang return result; 285906164d2bSGeorge Zhang } 286006164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_get_consume_indexes); 286106164d2bSGeorge Zhang 286206164d2bSGeorge Zhang /* 286306164d2bSGeorge Zhang * vmci_qpair_produce_free_space() - Retrieves free space in producer queue. 286406164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 286506164d2bSGeorge Zhang * 286606164d2bSGeorge Zhang * This is the client interface for getting the amount of free 286706164d2bSGeorge Zhang * space in the QPair from the point of the view of the caller as 286806164d2bSGeorge Zhang * the producer which is the common case. Returns < 0 if err, else 286906164d2bSGeorge Zhang * available bytes into which data can be enqueued if > 0. 287006164d2bSGeorge Zhang */ 287106164d2bSGeorge Zhang s64 vmci_qpair_produce_free_space(const struct vmci_qp *qpair) 287206164d2bSGeorge Zhang { 287306164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 287406164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 287506164d2bSGeorge Zhang s64 result; 287606164d2bSGeorge Zhang 287706164d2bSGeorge Zhang if (!qpair) 287806164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 287906164d2bSGeorge Zhang 288006164d2bSGeorge Zhang qp_lock(qpair); 288106164d2bSGeorge Zhang result = 288206164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 288306164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 288406164d2bSGeorge Zhang result = vmci_q_header_free_space(produce_q_header, 288506164d2bSGeorge Zhang consume_q_header, 288606164d2bSGeorge Zhang qpair->produce_q_size); 288706164d2bSGeorge Zhang else 288806164d2bSGeorge Zhang result = 0; 288906164d2bSGeorge Zhang 289006164d2bSGeorge Zhang qp_unlock(qpair); 289106164d2bSGeorge Zhang 289206164d2bSGeorge Zhang return result; 289306164d2bSGeorge Zhang } 289406164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_produce_free_space); 289506164d2bSGeorge Zhang 289606164d2bSGeorge Zhang /* 289706164d2bSGeorge Zhang * vmci_qpair_consume_free_space() - Retrieves free space in consumer queue. 289806164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 289906164d2bSGeorge Zhang * 290006164d2bSGeorge Zhang * This is the client interface for getting the amount of free 290106164d2bSGeorge Zhang * space in the QPair from the point of the view of the caller as 290206164d2bSGeorge Zhang * the consumer which is not the common case. Returns < 0 if err, else 290306164d2bSGeorge Zhang * available bytes into which data can be enqueued if > 0. 290406164d2bSGeorge Zhang */ 290506164d2bSGeorge Zhang s64 vmci_qpair_consume_free_space(const struct vmci_qp *qpair) 290606164d2bSGeorge Zhang { 290706164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 290806164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 290906164d2bSGeorge Zhang s64 result; 291006164d2bSGeorge Zhang 291106164d2bSGeorge Zhang if (!qpair) 291206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 291306164d2bSGeorge Zhang 291406164d2bSGeorge Zhang qp_lock(qpair); 291506164d2bSGeorge Zhang result = 291606164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 291706164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 291806164d2bSGeorge Zhang result = vmci_q_header_free_space(consume_q_header, 291906164d2bSGeorge Zhang produce_q_header, 292006164d2bSGeorge Zhang qpair->consume_q_size); 292106164d2bSGeorge Zhang else 292206164d2bSGeorge Zhang result = 0; 292306164d2bSGeorge Zhang 292406164d2bSGeorge Zhang qp_unlock(qpair); 292506164d2bSGeorge Zhang 292606164d2bSGeorge Zhang return result; 292706164d2bSGeorge Zhang } 292806164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_consume_free_space); 292906164d2bSGeorge Zhang 293006164d2bSGeorge Zhang /* 293106164d2bSGeorge Zhang * vmci_qpair_produce_buf_ready() - Gets bytes ready to read from 293206164d2bSGeorge Zhang * producer queue. 293306164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 293406164d2bSGeorge Zhang * 293506164d2bSGeorge Zhang * This is the client interface for getting the amount of 293606164d2bSGeorge Zhang * enqueued data in the QPair from the point of the view of the 293706164d2bSGeorge Zhang * caller as the producer which is not the common case. Returns < 0 if err, 293806164d2bSGeorge Zhang * else available bytes that may be read. 293906164d2bSGeorge Zhang */ 294006164d2bSGeorge Zhang s64 vmci_qpair_produce_buf_ready(const struct vmci_qp *qpair) 294106164d2bSGeorge Zhang { 294206164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 294306164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 294406164d2bSGeorge Zhang s64 result; 294506164d2bSGeorge Zhang 294606164d2bSGeorge Zhang if (!qpair) 294706164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 294806164d2bSGeorge Zhang 294906164d2bSGeorge Zhang qp_lock(qpair); 295006164d2bSGeorge Zhang result = 295106164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 295206164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 295306164d2bSGeorge Zhang result = vmci_q_header_buf_ready(produce_q_header, 295406164d2bSGeorge Zhang consume_q_header, 295506164d2bSGeorge Zhang qpair->produce_q_size); 295606164d2bSGeorge Zhang else 295706164d2bSGeorge Zhang result = 0; 295806164d2bSGeorge Zhang 295906164d2bSGeorge Zhang qp_unlock(qpair); 296006164d2bSGeorge Zhang 296106164d2bSGeorge Zhang return result; 296206164d2bSGeorge Zhang } 296306164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_produce_buf_ready); 296406164d2bSGeorge Zhang 296506164d2bSGeorge Zhang /* 296606164d2bSGeorge Zhang * vmci_qpair_consume_buf_ready() - Gets bytes ready to read from 296706164d2bSGeorge Zhang * consumer queue. 296806164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 296906164d2bSGeorge Zhang * 297006164d2bSGeorge Zhang * This is the client interface for getting the amount of 297106164d2bSGeorge Zhang * enqueued data in the QPair from the point of the view of the 297206164d2bSGeorge Zhang * caller as the consumer which is the normal case. Returns < 0 if err, 297306164d2bSGeorge Zhang * else available bytes that may be read. 297406164d2bSGeorge Zhang */ 297506164d2bSGeorge Zhang s64 vmci_qpair_consume_buf_ready(const struct vmci_qp *qpair) 297606164d2bSGeorge Zhang { 297706164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 297806164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 297906164d2bSGeorge Zhang s64 result; 298006164d2bSGeorge Zhang 298106164d2bSGeorge Zhang if (!qpair) 298206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 298306164d2bSGeorge Zhang 298406164d2bSGeorge Zhang qp_lock(qpair); 298506164d2bSGeorge Zhang result = 298606164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 298706164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 298806164d2bSGeorge Zhang result = vmci_q_header_buf_ready(consume_q_header, 298906164d2bSGeorge Zhang produce_q_header, 299006164d2bSGeorge Zhang qpair->consume_q_size); 299106164d2bSGeorge Zhang else 299206164d2bSGeorge Zhang result = 0; 299306164d2bSGeorge Zhang 299406164d2bSGeorge Zhang qp_unlock(qpair); 299506164d2bSGeorge Zhang 299606164d2bSGeorge Zhang return result; 299706164d2bSGeorge Zhang } 299806164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_consume_buf_ready); 299906164d2bSGeorge Zhang 300006164d2bSGeorge Zhang /* 300106164d2bSGeorge Zhang * vmci_qpair_enqueue() - Throw data on the queue. 300206164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 300306164d2bSGeorge Zhang * @buf: Pointer to buffer containing data 300406164d2bSGeorge Zhang * @buf_size: Length of buffer. 300506164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 300606164d2bSGeorge Zhang * 300706164d2bSGeorge Zhang * This is the client interface for enqueueing data into the queue. 300806164d2bSGeorge Zhang * Returns number of bytes enqueued or < 0 on error. 300906164d2bSGeorge Zhang */ 301006164d2bSGeorge Zhang ssize_t vmci_qpair_enqueue(struct vmci_qp *qpair, 301106164d2bSGeorge Zhang const void *buf, 301206164d2bSGeorge Zhang size_t buf_size, 301306164d2bSGeorge Zhang int buf_type) 301406164d2bSGeorge Zhang { 301506164d2bSGeorge Zhang ssize_t result; 3016d1038084SAl Viro struct iov_iter from; 3017d1038084SAl Viro struct kvec v = {.iov_base = (void *)buf, .iov_len = buf_size}; 301806164d2bSGeorge Zhang 301906164d2bSGeorge Zhang if (!qpair || !buf) 302006164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 302106164d2bSGeorge Zhang 3022aa563d7bSDavid Howells iov_iter_kvec(&from, WRITE, &v, 1, buf_size); 3023d1038084SAl Viro 302406164d2bSGeorge Zhang qp_lock(qpair); 302506164d2bSGeorge Zhang 302606164d2bSGeorge Zhang do { 302706164d2bSGeorge Zhang result = qp_enqueue_locked(qpair->produce_q, 302806164d2bSGeorge Zhang qpair->consume_q, 302906164d2bSGeorge Zhang qpair->produce_q_size, 3030d1038084SAl Viro &from); 303106164d2bSGeorge Zhang 303206164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 303306164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 303406164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 303506164d2bSGeorge Zhang 303606164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 303706164d2bSGeorge Zhang 303806164d2bSGeorge Zhang qp_unlock(qpair); 303906164d2bSGeorge Zhang 304006164d2bSGeorge Zhang return result; 304106164d2bSGeorge Zhang } 304206164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_enqueue); 304306164d2bSGeorge Zhang 304406164d2bSGeorge Zhang /* 304506164d2bSGeorge Zhang * vmci_qpair_dequeue() - Get data from the queue. 304606164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 304706164d2bSGeorge Zhang * @buf: Pointer to buffer for the data 304806164d2bSGeorge Zhang * @buf_size: Length of buffer. 304906164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 305006164d2bSGeorge Zhang * 305106164d2bSGeorge Zhang * This is the client interface for dequeueing data from the queue. 305206164d2bSGeorge Zhang * Returns number of bytes dequeued or < 0 on error. 305306164d2bSGeorge Zhang */ 305406164d2bSGeorge Zhang ssize_t vmci_qpair_dequeue(struct vmci_qp *qpair, 305506164d2bSGeorge Zhang void *buf, 305606164d2bSGeorge Zhang size_t buf_size, 305706164d2bSGeorge Zhang int buf_type) 305806164d2bSGeorge Zhang { 305906164d2bSGeorge Zhang ssize_t result; 3060ce3d6e7dSAl Viro struct iov_iter to; 3061ce3d6e7dSAl Viro struct kvec v = {.iov_base = buf, .iov_len = buf_size}; 306206164d2bSGeorge Zhang 306306164d2bSGeorge Zhang if (!qpair || !buf) 306406164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 306506164d2bSGeorge Zhang 3066aa563d7bSDavid Howells iov_iter_kvec(&to, READ, &v, 1, buf_size); 3067ce3d6e7dSAl Viro 306806164d2bSGeorge Zhang qp_lock(qpair); 306906164d2bSGeorge Zhang 307006164d2bSGeorge Zhang do { 307106164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 307206164d2bSGeorge Zhang qpair->consume_q, 307306164d2bSGeorge Zhang qpair->consume_q_size, 307453f58d8eSAl Viro &to, true); 307506164d2bSGeorge Zhang 307606164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 307706164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 307806164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 307906164d2bSGeorge Zhang 308006164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 308106164d2bSGeorge Zhang 308206164d2bSGeorge Zhang qp_unlock(qpair); 308306164d2bSGeorge Zhang 308406164d2bSGeorge Zhang return result; 308506164d2bSGeorge Zhang } 308606164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_dequeue); 308706164d2bSGeorge Zhang 308806164d2bSGeorge Zhang /* 308906164d2bSGeorge Zhang * vmci_qpair_peek() - Peek at the data in the queue. 309006164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 309106164d2bSGeorge Zhang * @buf: Pointer to buffer for the data 309206164d2bSGeorge Zhang * @buf_size: Length of buffer. 309306164d2bSGeorge Zhang * @buf_type: Buffer type (Unused on Linux). 309406164d2bSGeorge Zhang * 309506164d2bSGeorge Zhang * This is the client interface for peeking into a queue. (I.e., 309606164d2bSGeorge Zhang * copy data from the queue without updating the head pointer.) 309706164d2bSGeorge Zhang * Returns number of bytes dequeued or < 0 on error. 309806164d2bSGeorge Zhang */ 309906164d2bSGeorge Zhang ssize_t vmci_qpair_peek(struct vmci_qp *qpair, 310006164d2bSGeorge Zhang void *buf, 310106164d2bSGeorge Zhang size_t buf_size, 310206164d2bSGeorge Zhang int buf_type) 310306164d2bSGeorge Zhang { 3104ce3d6e7dSAl Viro struct iov_iter to; 3105ce3d6e7dSAl Viro struct kvec v = {.iov_base = buf, .iov_len = buf_size}; 310606164d2bSGeorge Zhang ssize_t result; 310706164d2bSGeorge Zhang 310806164d2bSGeorge Zhang if (!qpair || !buf) 310906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 311006164d2bSGeorge Zhang 3111aa563d7bSDavid Howells iov_iter_kvec(&to, READ, &v, 1, buf_size); 3112ce3d6e7dSAl Viro 311306164d2bSGeorge Zhang qp_lock(qpair); 311406164d2bSGeorge Zhang 311506164d2bSGeorge Zhang do { 311606164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 311706164d2bSGeorge Zhang qpair->consume_q, 311806164d2bSGeorge Zhang qpair->consume_q_size, 311953f58d8eSAl Viro &to, false); 312006164d2bSGeorge Zhang 312106164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 312206164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 312306164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 312406164d2bSGeorge Zhang 312506164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 312606164d2bSGeorge Zhang 312706164d2bSGeorge Zhang qp_unlock(qpair); 312806164d2bSGeorge Zhang 312906164d2bSGeorge Zhang return result; 313006164d2bSGeorge Zhang } 313106164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_peek); 313206164d2bSGeorge Zhang 313306164d2bSGeorge Zhang /* 313406164d2bSGeorge Zhang * vmci_qpair_enquev() - Throw data on the queue using iov. 313506164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 313606164d2bSGeorge Zhang * @iov: Pointer to buffer containing data 313706164d2bSGeorge Zhang * @iov_size: Length of buffer. 313806164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 313906164d2bSGeorge Zhang * 314006164d2bSGeorge Zhang * This is the client interface for enqueueing data into the queue. 314106164d2bSGeorge Zhang * This function uses IO vectors to handle the work. Returns number 314206164d2bSGeorge Zhang * of bytes enqueued or < 0 on error. 314306164d2bSGeorge Zhang */ 314406164d2bSGeorge Zhang ssize_t vmci_qpair_enquev(struct vmci_qp *qpair, 31454c946d9cSAl Viro struct msghdr *msg, 314606164d2bSGeorge Zhang size_t iov_size, 314706164d2bSGeorge Zhang int buf_type) 314806164d2bSGeorge Zhang { 314906164d2bSGeorge Zhang ssize_t result; 315006164d2bSGeorge Zhang 31514c946d9cSAl Viro if (!qpair) 315206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 315306164d2bSGeorge Zhang 315406164d2bSGeorge Zhang qp_lock(qpair); 315506164d2bSGeorge Zhang 315606164d2bSGeorge Zhang do { 315706164d2bSGeorge Zhang result = qp_enqueue_locked(qpair->produce_q, 315806164d2bSGeorge Zhang qpair->consume_q, 315906164d2bSGeorge Zhang qpair->produce_q_size, 3160d1038084SAl Viro &msg->msg_iter); 316106164d2bSGeorge Zhang 316206164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 316306164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 316406164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 316506164d2bSGeorge Zhang 316606164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 316706164d2bSGeorge Zhang 316806164d2bSGeorge Zhang qp_unlock(qpair); 316906164d2bSGeorge Zhang 317006164d2bSGeorge Zhang return result; 317106164d2bSGeorge Zhang } 317206164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_enquev); 317306164d2bSGeorge Zhang 317406164d2bSGeorge Zhang /* 317506164d2bSGeorge Zhang * vmci_qpair_dequev() - Get data from the queue using iov. 317606164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 317706164d2bSGeorge Zhang * @iov: Pointer to buffer for the data 317806164d2bSGeorge Zhang * @iov_size: Length of buffer. 317906164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 318006164d2bSGeorge Zhang * 318106164d2bSGeorge Zhang * This is the client interface for dequeueing data from the queue. 318206164d2bSGeorge Zhang * This function uses IO vectors to handle the work. Returns number 318306164d2bSGeorge Zhang * of bytes dequeued or < 0 on error. 318406164d2bSGeorge Zhang */ 318506164d2bSGeorge Zhang ssize_t vmci_qpair_dequev(struct vmci_qp *qpair, 3186d838df2eSAl Viro struct msghdr *msg, 318706164d2bSGeorge Zhang size_t iov_size, 318806164d2bSGeorge Zhang int buf_type) 318906164d2bSGeorge Zhang { 319006164d2bSGeorge Zhang ssize_t result; 319106164d2bSGeorge Zhang 3192d838df2eSAl Viro if (!qpair) 319306164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 319406164d2bSGeorge Zhang 319532b083a3SAndy King qp_lock(qpair); 319632b083a3SAndy King 319706164d2bSGeorge Zhang do { 319806164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 319906164d2bSGeorge Zhang qpair->consume_q, 320006164d2bSGeorge Zhang qpair->consume_q_size, 320153f58d8eSAl Viro &msg->msg_iter, true); 320206164d2bSGeorge Zhang 320306164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 320406164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 320506164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 320606164d2bSGeorge Zhang 320706164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 320806164d2bSGeorge Zhang 320906164d2bSGeorge Zhang qp_unlock(qpair); 321006164d2bSGeorge Zhang 321106164d2bSGeorge Zhang return result; 321206164d2bSGeorge Zhang } 321306164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_dequev); 321406164d2bSGeorge Zhang 321506164d2bSGeorge Zhang /* 321606164d2bSGeorge Zhang * vmci_qpair_peekv() - Peek at the data in the queue using iov. 321706164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 321806164d2bSGeorge Zhang * @iov: Pointer to buffer for the data 321906164d2bSGeorge Zhang * @iov_size: Length of buffer. 322006164d2bSGeorge Zhang * @buf_type: Buffer type (Unused on Linux). 322106164d2bSGeorge Zhang * 322206164d2bSGeorge Zhang * This is the client interface for peeking into a queue. (I.e., 322306164d2bSGeorge Zhang * copy data from the queue without updating the head pointer.) 322406164d2bSGeorge Zhang * This function uses IO vectors to handle the work. Returns number 322506164d2bSGeorge Zhang * of bytes peeked or < 0 on error. 322606164d2bSGeorge Zhang */ 322706164d2bSGeorge Zhang ssize_t vmci_qpair_peekv(struct vmci_qp *qpair, 3228d838df2eSAl Viro struct msghdr *msg, 322906164d2bSGeorge Zhang size_t iov_size, 323006164d2bSGeorge Zhang int buf_type) 323106164d2bSGeorge Zhang { 323206164d2bSGeorge Zhang ssize_t result; 323306164d2bSGeorge Zhang 3234d838df2eSAl Viro if (!qpair) 323506164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 323606164d2bSGeorge Zhang 323706164d2bSGeorge Zhang qp_lock(qpair); 323806164d2bSGeorge Zhang 323906164d2bSGeorge Zhang do { 324006164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 324106164d2bSGeorge Zhang qpair->consume_q, 324206164d2bSGeorge Zhang qpair->consume_q_size, 324353f58d8eSAl Viro &msg->msg_iter, false); 324406164d2bSGeorge Zhang 324506164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 324606164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 324706164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 324806164d2bSGeorge Zhang 324906164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 325006164d2bSGeorge Zhang 325106164d2bSGeorge Zhang qp_unlock(qpair); 325206164d2bSGeorge Zhang return result; 325306164d2bSGeorge Zhang } 325406164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_peekv); 3255