]> Git Repo - linux.git/blob - arch/um/drivers/virtio_uml.c
nfsd: grant read delegations to clients holding writes
[linux.git] / arch / um / drivers / virtio_uml.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Virtio vhost-user driver
4  *
5  * Copyright(c) 2019 Intel Corporation
6  *
7  * This driver allows virtio devices to be used over a vhost-user socket.
8  *
9  * Guest devices can be instantiated by kernel module or command line
10  * parameters. One device will be created for each parameter. Syntax:
11  *
12  *              virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]
13  * where:
14  *              <socket>        := vhost-user socket path to connect
15  *              <virtio_id>     := virtio device id (as in virtio_ids.h)
16  *              <platform_id>   := (optional) platform device id
17  *
18  * example:
19  *              virtio_uml.device=/var/uml.socket:1
20  *
21  * Based on Virtio MMIO driver by Pawel Moll, copyright 2011-2014, ARM Ltd.
22  */
23 #include <linux/module.h>
24 #include <linux/platform_device.h>
25 #include <linux/slab.h>
26 #include <linux/virtio.h>
27 #include <linux/virtio_config.h>
28 #include <linux/virtio_ring.h>
29 #include <linux/time-internal.h>
30 #include <shared/as-layout.h>
31 #include <irq_kern.h>
32 #include <init.h>
33 #include <os.h>
34 #include "vhost_user.h"
35
36 #define MAX_SUPPORTED_QUEUE_SIZE        256
37
38 #define to_virtio_uml_device(_vdev) \
39         container_of(_vdev, struct virtio_uml_device, vdev)
40
41 struct virtio_uml_platform_data {
42         u32 virtio_device_id;
43         const char *socket_path;
44         struct work_struct conn_broken_wk;
45         struct platform_device *pdev;
46 };
47
48 struct virtio_uml_device {
49         struct virtio_device vdev;
50         struct platform_device *pdev;
51
52         spinlock_t sock_lock;
53         int sock, req_fd, irq;
54         u64 features;
55         u64 protocol_features;
56         u8 status;
57         u8 registered:1;
58         u8 suspended:1;
59
60         u8 config_changed_irq:1;
61         uint64_t vq_irq_vq_map;
62 };
63
64 struct virtio_uml_vq_info {
65         int kick_fd, call_fd;
66         char name[32];
67         bool suspended;
68 };
69
70 extern unsigned long long physmem_size, highmem;
71
72 #define vu_err(vu_dev, ...)     dev_err(&(vu_dev)->pdev->dev, ##__VA_ARGS__)
73
74 /* Vhost-user protocol */
75
76 static int full_sendmsg_fds(int fd, const void *buf, unsigned int len,
77                             const int *fds, unsigned int fds_num)
78 {
79         int rc;
80
81         do {
82                 rc = os_sendmsg_fds(fd, buf, len, fds, fds_num);
83                 if (rc > 0) {
84                         buf += rc;
85                         len -= rc;
86                         fds = NULL;
87                         fds_num = 0;
88                 }
89         } while (len && (rc >= 0 || rc == -EINTR));
90
91         if (rc < 0)
92                 return rc;
93         return 0;
94 }
95
96 static int full_read(int fd, void *buf, int len, bool abortable)
97 {
98         int rc;
99
100         if (!len)
101                 return 0;
102
103         do {
104                 rc = os_read_file(fd, buf, len);
105                 if (rc > 0) {
106                         buf += rc;
107                         len -= rc;
108                 }
109         } while (len && (rc > 0 || rc == -EINTR || (!abortable && rc == -EAGAIN)));
110
111         if (rc < 0)
112                 return rc;
113         if (rc == 0)
114                 return -ECONNRESET;
115         return 0;
116 }
117
118 static int vhost_user_recv_header(int fd, struct vhost_user_msg *msg)
119 {
120         return full_read(fd, msg, sizeof(msg->header), true);
121 }
122
123 static int vhost_user_recv(struct virtio_uml_device *vu_dev,
124                            int fd, struct vhost_user_msg *msg,
125                            size_t max_payload_size, bool wait)
126 {
127         size_t size;
128         int rc;
129
130         /*
131          * In virtio time-travel mode, we're handling all the vhost-user
132          * FDs by polling them whenever appropriate. However, we may get
133          * into a situation where we're sending out an interrupt message
134          * to a device (e.g. a net device) and need to handle a simulation
135          * time message while doing so, e.g. one that tells us to update
136          * our idea of how long we can run without scheduling.
137          *
138          * Thus, we need to not just read() from the given fd, but need
139          * to also handle messages for the simulation time - this function
140          * does that for us while waiting for the given fd to be readable.
141          */
142         if (wait)
143                 time_travel_wait_readable(fd);
144
145         rc = vhost_user_recv_header(fd, msg);
146
147         if (rc == -ECONNRESET && vu_dev->registered) {
148                 struct virtio_uml_platform_data *pdata;
149
150                 pdata = vu_dev->pdev->dev.platform_data;
151
152                 virtio_break_device(&vu_dev->vdev);
153                 schedule_work(&pdata->conn_broken_wk);
154         }
155         if (rc)
156                 return rc;
157         size = msg->header.size;
158         if (size > max_payload_size)
159                 return -EPROTO;
160         return full_read(fd, &msg->payload, size, false);
161 }
162
163 static int vhost_user_recv_resp(struct virtio_uml_device *vu_dev,
164                                 struct vhost_user_msg *msg,
165                                 size_t max_payload_size)
166 {
167         int rc = vhost_user_recv(vu_dev, vu_dev->sock, msg,
168                                  max_payload_size, true);
169
170         if (rc)
171                 return rc;
172
173         if (msg->header.flags != (VHOST_USER_FLAG_REPLY | VHOST_USER_VERSION))
174                 return -EPROTO;
175
176         return 0;
177 }
178
179 static int vhost_user_recv_u64(struct virtio_uml_device *vu_dev,
180                                u64 *value)
181 {
182         struct vhost_user_msg msg;
183         int rc = vhost_user_recv_resp(vu_dev, &msg,
184                                       sizeof(msg.payload.integer));
185
186         if (rc)
187                 return rc;
188         if (msg.header.size != sizeof(msg.payload.integer))
189                 return -EPROTO;
190         *value = msg.payload.integer;
191         return 0;
192 }
193
194 static int vhost_user_recv_req(struct virtio_uml_device *vu_dev,
195                                struct vhost_user_msg *msg,
196                                size_t max_payload_size)
197 {
198         int rc = vhost_user_recv(vu_dev, vu_dev->req_fd, msg,
199                                  max_payload_size, false);
200
201         if (rc)
202                 return rc;
203
204         if ((msg->header.flags & ~VHOST_USER_FLAG_NEED_REPLY) !=
205                         VHOST_USER_VERSION)
206                 return -EPROTO;
207
208         return 0;
209 }
210
211 static int vhost_user_send(struct virtio_uml_device *vu_dev,
212                            bool need_response, struct vhost_user_msg *msg,
213                            int *fds, size_t num_fds)
214 {
215         size_t size = sizeof(msg->header) + msg->header.size;
216         unsigned long flags;
217         bool request_ack;
218         int rc;
219
220         msg->header.flags |= VHOST_USER_VERSION;
221
222         /*
223          * The need_response flag indicates that we already need a response,
224          * e.g. to read the features. In these cases, don't request an ACK as
225          * it is meaningless. Also request an ACK only if supported.
226          */
227         request_ack = !need_response;
228         if (!(vu_dev->protocol_features &
229                         BIT_ULL(VHOST_USER_PROTOCOL_F_REPLY_ACK)))
230                 request_ack = false;
231
232         if (request_ack)
233                 msg->header.flags |= VHOST_USER_FLAG_NEED_REPLY;
234
235         spin_lock_irqsave(&vu_dev->sock_lock, flags);
236         rc = full_sendmsg_fds(vu_dev->sock, msg, size, fds, num_fds);
237         if (rc < 0)
238                 goto out;
239
240         if (request_ack) {
241                 uint64_t status;
242
243                 rc = vhost_user_recv_u64(vu_dev, &status);
244                 if (rc)
245                         goto out;
246
247                 if (status) {
248                         vu_err(vu_dev, "slave reports error: %llu\n", status);
249                         rc = -EIO;
250                         goto out;
251                 }
252         }
253
254 out:
255         spin_unlock_irqrestore(&vu_dev->sock_lock, flags);
256         return rc;
257 }
258
259 static int vhost_user_send_no_payload(struct virtio_uml_device *vu_dev,
260                                       bool need_response, u32 request)
261 {
262         struct vhost_user_msg msg = {
263                 .header.request = request,
264         };
265
266         return vhost_user_send(vu_dev, need_response, &msg, NULL, 0);
267 }
268
269 static int vhost_user_send_no_payload_fd(struct virtio_uml_device *vu_dev,
270                                          u32 request, int fd)
271 {
272         struct vhost_user_msg msg = {
273                 .header.request = request,
274         };
275
276         return vhost_user_send(vu_dev, false, &msg, &fd, 1);
277 }
278
279 static int vhost_user_send_u64(struct virtio_uml_device *vu_dev,
280                                u32 request, u64 value)
281 {
282         struct vhost_user_msg msg = {
283                 .header.request = request,
284                 .header.size = sizeof(msg.payload.integer),
285                 .payload.integer = value,
286         };
287
288         return vhost_user_send(vu_dev, false, &msg, NULL, 0);
289 }
290
291 static int vhost_user_set_owner(struct virtio_uml_device *vu_dev)
292 {
293         return vhost_user_send_no_payload(vu_dev, false, VHOST_USER_SET_OWNER);
294 }
295
296 static int vhost_user_get_features(struct virtio_uml_device *vu_dev,
297                                    u64 *features)
298 {
299         int rc = vhost_user_send_no_payload(vu_dev, true,
300                                             VHOST_USER_GET_FEATURES);
301
302         if (rc)
303                 return rc;
304         return vhost_user_recv_u64(vu_dev, features);
305 }
306
307 static int vhost_user_set_features(struct virtio_uml_device *vu_dev,
308                                    u64 features)
309 {
310         return vhost_user_send_u64(vu_dev, VHOST_USER_SET_FEATURES, features);
311 }
312
313 static int vhost_user_get_protocol_features(struct virtio_uml_device *vu_dev,
314                                             u64 *protocol_features)
315 {
316         int rc = vhost_user_send_no_payload(vu_dev, true,
317                         VHOST_USER_GET_PROTOCOL_FEATURES);
318
319         if (rc)
320                 return rc;
321         return vhost_user_recv_u64(vu_dev, protocol_features);
322 }
323
324 static int vhost_user_set_protocol_features(struct virtio_uml_device *vu_dev,
325                                             u64 protocol_features)
326 {
327         return vhost_user_send_u64(vu_dev, VHOST_USER_SET_PROTOCOL_FEATURES,
328                                    protocol_features);
329 }
330
331 static void vhost_user_reply(struct virtio_uml_device *vu_dev,
332                              struct vhost_user_msg *msg, int response)
333 {
334         struct vhost_user_msg reply = {
335                 .payload.integer = response,
336         };
337         size_t size = sizeof(reply.header) + sizeof(reply.payload.integer);
338         int rc;
339
340         reply.header = msg->header;
341         reply.header.flags &= ~VHOST_USER_FLAG_NEED_REPLY;
342         reply.header.flags |= VHOST_USER_FLAG_REPLY;
343         reply.header.size = sizeof(reply.payload.integer);
344
345         rc = full_sendmsg_fds(vu_dev->req_fd, &reply, size, NULL, 0);
346
347         if (rc)
348                 vu_err(vu_dev,
349                        "sending reply to slave request failed: %d (size %zu)\n",
350                        rc, size);
351 }
352
353 static irqreturn_t vu_req_read_message(struct virtio_uml_device *vu_dev,
354                                        struct time_travel_event *ev)
355 {
356         struct virtqueue *vq;
357         int response = 1;
358         struct {
359                 struct vhost_user_msg msg;
360                 u8 extra_payload[512];
361         } msg;
362         int rc;
363
364         rc = vhost_user_recv_req(vu_dev, &msg.msg,
365                                  sizeof(msg.msg.payload) +
366                                  sizeof(msg.extra_payload));
367
368         if (rc)
369                 return IRQ_NONE;
370
371         switch (msg.msg.header.request) {
372         case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG:
373                 vu_dev->config_changed_irq = true;
374                 response = 0;
375                 break;
376         case VHOST_USER_SLAVE_VRING_CALL:
377                 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
378                         if (vq->index == msg.msg.payload.vring_state.index) {
379                                 response = 0;
380                                 vu_dev->vq_irq_vq_map |= BIT_ULL(vq->index);
381                                 break;
382                         }
383                 }
384                 break;
385         case VHOST_USER_SLAVE_IOTLB_MSG:
386                 /* not supported - VIRTIO_F_ACCESS_PLATFORM */
387         case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
388                 /* not supported - VHOST_USER_PROTOCOL_F_HOST_NOTIFIER */
389         default:
390                 vu_err(vu_dev, "unexpected slave request %d\n",
391                        msg.msg.header.request);
392         }
393
394         if (ev && !vu_dev->suspended)
395                 time_travel_add_irq_event(ev);
396
397         if (msg.msg.header.flags & VHOST_USER_FLAG_NEED_REPLY)
398                 vhost_user_reply(vu_dev, &msg.msg, response);
399
400         return IRQ_HANDLED;
401 }
402
403 static irqreturn_t vu_req_interrupt(int irq, void *data)
404 {
405         struct virtio_uml_device *vu_dev = data;
406         irqreturn_t ret = IRQ_HANDLED;
407
408         if (!um_irq_timetravel_handler_used())
409                 ret = vu_req_read_message(vu_dev, NULL);
410
411         if (vu_dev->vq_irq_vq_map) {
412                 struct virtqueue *vq;
413
414                 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
415                         if (vu_dev->vq_irq_vq_map & BIT_ULL(vq->index))
416                                 vring_interrupt(0 /* ignored */, vq);
417                 }
418                 vu_dev->vq_irq_vq_map = 0;
419         } else if (vu_dev->config_changed_irq) {
420                 virtio_config_changed(&vu_dev->vdev);
421                 vu_dev->config_changed_irq = false;
422         }
423
424         return ret;
425 }
426
427 static void vu_req_interrupt_comm_handler(int irq, int fd, void *data,
428                                           struct time_travel_event *ev)
429 {
430         vu_req_read_message(data, ev);
431 }
432
433 static int vhost_user_init_slave_req(struct virtio_uml_device *vu_dev)
434 {
435         int rc, req_fds[2];
436
437         /* Use a pipe for slave req fd, SIGIO is not supported for eventfd */
438         rc = os_pipe(req_fds, true, true);
439         if (rc < 0)
440                 return rc;
441         vu_dev->req_fd = req_fds[0];
442
443         rc = um_request_irq_tt(UM_IRQ_ALLOC, vu_dev->req_fd, IRQ_READ,
444                                vu_req_interrupt, IRQF_SHARED,
445                                vu_dev->pdev->name, vu_dev,
446                                vu_req_interrupt_comm_handler);
447         if (rc < 0)
448                 goto err_close;
449
450         vu_dev->irq = rc;
451
452         rc = vhost_user_send_no_payload_fd(vu_dev, VHOST_USER_SET_SLAVE_REQ_FD,
453                                            req_fds[1]);
454         if (rc)
455                 goto err_free_irq;
456
457         goto out;
458
459 err_free_irq:
460         um_free_irq(vu_dev->irq, vu_dev);
461 err_close:
462         os_close_file(req_fds[0]);
463 out:
464         /* Close unused write end of request fds */
465         os_close_file(req_fds[1]);
466         return rc;
467 }
468
469 static int vhost_user_init(struct virtio_uml_device *vu_dev)
470 {
471         int rc = vhost_user_set_owner(vu_dev);
472
473         if (rc)
474                 return rc;
475         rc = vhost_user_get_features(vu_dev, &vu_dev->features);
476         if (rc)
477                 return rc;
478
479         if (vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)) {
480                 rc = vhost_user_get_protocol_features(vu_dev,
481                                 &vu_dev->protocol_features);
482                 if (rc)
483                         return rc;
484                 vu_dev->protocol_features &= VHOST_USER_SUPPORTED_PROTOCOL_F;
485                 rc = vhost_user_set_protocol_features(vu_dev,
486                                 vu_dev->protocol_features);
487                 if (rc)
488                         return rc;
489         }
490
491         if (vu_dev->protocol_features &
492                         BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
493                 rc = vhost_user_init_slave_req(vu_dev);
494                 if (rc)
495                         return rc;
496         }
497
498         return 0;
499 }
500
501 static void vhost_user_get_config(struct virtio_uml_device *vu_dev,
502                                   u32 offset, void *buf, u32 len)
503 {
504         u32 cfg_size = offset + len;
505         struct vhost_user_msg *msg;
506         size_t payload_size = sizeof(msg->payload.config) + cfg_size;
507         size_t msg_size = sizeof(msg->header) + payload_size;
508         int rc;
509
510         if (!(vu_dev->protocol_features &
511               BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
512                 return;
513
514         msg = kzalloc(msg_size, GFP_KERNEL);
515         if (!msg)
516                 return;
517         msg->header.request = VHOST_USER_GET_CONFIG;
518         msg->header.size = payload_size;
519         msg->payload.config.offset = 0;
520         msg->payload.config.size = cfg_size;
521
522         rc = vhost_user_send(vu_dev, true, msg, NULL, 0);
523         if (rc) {
524                 vu_err(vu_dev, "sending VHOST_USER_GET_CONFIG failed: %d\n",
525                        rc);
526                 goto free;
527         }
528
529         rc = vhost_user_recv_resp(vu_dev, msg, msg_size);
530         if (rc) {
531                 vu_err(vu_dev,
532                        "receiving VHOST_USER_GET_CONFIG response failed: %d\n",
533                        rc);
534                 goto free;
535         }
536
537         if (msg->header.size != payload_size ||
538             msg->payload.config.size != cfg_size) {
539                 rc = -EPROTO;
540                 vu_err(vu_dev,
541                        "Invalid VHOST_USER_GET_CONFIG sizes (payload %d expected %zu, config %u expected %u)\n",
542                        msg->header.size, payload_size,
543                        msg->payload.config.size, cfg_size);
544                 goto free;
545         }
546         memcpy(buf, msg->payload.config.payload + offset, len);
547
548 free:
549         kfree(msg);
550 }
551
552 static void vhost_user_set_config(struct virtio_uml_device *vu_dev,
553                                   u32 offset, const void *buf, u32 len)
554 {
555         struct vhost_user_msg *msg;
556         size_t payload_size = sizeof(msg->payload.config) + len;
557         size_t msg_size = sizeof(msg->header) + payload_size;
558         int rc;
559
560         if (!(vu_dev->protocol_features &
561               BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
562                 return;
563
564         msg = kzalloc(msg_size, GFP_KERNEL);
565         if (!msg)
566                 return;
567         msg->header.request = VHOST_USER_SET_CONFIG;
568         msg->header.size = payload_size;
569         msg->payload.config.offset = offset;
570         msg->payload.config.size = len;
571         memcpy(msg->payload.config.payload, buf, len);
572
573         rc = vhost_user_send(vu_dev, false, msg, NULL, 0);
574         if (rc)
575                 vu_err(vu_dev, "sending VHOST_USER_SET_CONFIG failed: %d\n",
576                        rc);
577
578         kfree(msg);
579 }
580
581 static int vhost_user_init_mem_region(u64 addr, u64 size, int *fd_out,
582                                       struct vhost_user_mem_region *region_out)
583 {
584         unsigned long long mem_offset;
585         int rc = phys_mapping(addr, &mem_offset);
586
587         if (WARN(rc < 0, "phys_mapping of 0x%llx returned %d\n", addr, rc))
588                 return -EFAULT;
589         *fd_out = rc;
590         region_out->guest_addr = addr;
591         region_out->user_addr = addr;
592         region_out->size = size;
593         region_out->mmap_offset = mem_offset;
594
595         /* Ensure mapping is valid for the entire region */
596         rc = phys_mapping(addr + size - 1, &mem_offset);
597         if (WARN(rc != *fd_out, "phys_mapping of 0x%llx failed: %d != %d\n",
598                  addr + size - 1, rc, *fd_out))
599                 return -EFAULT;
600         return 0;
601 }
602
603 static int vhost_user_set_mem_table(struct virtio_uml_device *vu_dev)
604 {
605         struct vhost_user_msg msg = {
606                 .header.request = VHOST_USER_SET_MEM_TABLE,
607                 .header.size = sizeof(msg.payload.mem_regions),
608                 .payload.mem_regions.num = 1,
609         };
610         unsigned long reserved = uml_reserved - uml_physmem;
611         int fds[2];
612         int rc;
613
614         /*
615          * This is a bit tricky, see also the comment with setup_physmem().
616          *
617          * Essentially, setup_physmem() uses a file to mmap() our physmem,
618          * but the code and data we *already* have is omitted. To us, this
619          * is no difference, since they both become part of our address
620          * space and memory consumption. To somebody looking in from the
621          * outside, however, it is different because the part of our memory
622          * consumption that's already part of the binary (code/data) is not
623          * mapped from the file, so it's not visible to another mmap from
624          * the file descriptor.
625          *
626          * Thus, don't advertise this space to the vhost-user slave. This
627          * means that the slave will likely abort or similar when we give
628          * it an address from the hidden range, since it's not marked as
629          * a valid address, but at least that way we detect the issue and
630          * don't just have the slave read an all-zeroes buffer from the
631          * shared memory file, or write something there that we can never
632          * see (depending on the direction of the virtqueue traffic.)
633          *
634          * Since we usually don't want to use .text for virtio buffers,
635          * this effectively means that you cannot use
636          *  1) global variables, which are in the .bss and not in the shm
637          *     file-backed memory
638          *  2) the stack in some processes, depending on where they have
639          *     their stack (or maybe only no interrupt stack?)
640          *
641          * The stack is already not typically valid for DMA, so this isn't
642          * much of a restriction, but global variables might be encountered.
643          *
644          * It might be possible to fix it by copying around the data that's
645          * between bss_start and where we map the file now, but it's not
646          * something that you typically encounter with virtio drivers, so
647          * it didn't seem worthwhile.
648          */
649         rc = vhost_user_init_mem_region(reserved, physmem_size - reserved,
650                                         &fds[0],
651                                         &msg.payload.mem_regions.regions[0]);
652
653         if (rc < 0)
654                 return rc;
655         if (highmem) {
656                 msg.payload.mem_regions.num++;
657                 rc = vhost_user_init_mem_region(__pa(end_iomem), highmem,
658                                 &fds[1], &msg.payload.mem_regions.regions[1]);
659                 if (rc < 0)
660                         return rc;
661         }
662
663         return vhost_user_send(vu_dev, false, &msg, fds,
664                                msg.payload.mem_regions.num);
665 }
666
667 static int vhost_user_set_vring_state(struct virtio_uml_device *vu_dev,
668                                       u32 request, u32 index, u32 num)
669 {
670         struct vhost_user_msg msg = {
671                 .header.request = request,
672                 .header.size = sizeof(msg.payload.vring_state),
673                 .payload.vring_state.index = index,
674                 .payload.vring_state.num = num,
675         };
676
677         return vhost_user_send(vu_dev, false, &msg, NULL, 0);
678 }
679
680 static int vhost_user_set_vring_num(struct virtio_uml_device *vu_dev,
681                                     u32 index, u32 num)
682 {
683         return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_NUM,
684                                           index, num);
685 }
686
687 static int vhost_user_set_vring_base(struct virtio_uml_device *vu_dev,
688                                      u32 index, u32 offset)
689 {
690         return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_BASE,
691                                           index, offset);
692 }
693
694 static int vhost_user_set_vring_addr(struct virtio_uml_device *vu_dev,
695                                      u32 index, u64 desc, u64 used, u64 avail,
696                                      u64 log)
697 {
698         struct vhost_user_msg msg = {
699                 .header.request = VHOST_USER_SET_VRING_ADDR,
700                 .header.size = sizeof(msg.payload.vring_addr),
701                 .payload.vring_addr.index = index,
702                 .payload.vring_addr.desc = desc,
703                 .payload.vring_addr.used = used,
704                 .payload.vring_addr.avail = avail,
705                 .payload.vring_addr.log = log,
706         };
707
708         return vhost_user_send(vu_dev, false, &msg, NULL, 0);
709 }
710
711 static int vhost_user_set_vring_fd(struct virtio_uml_device *vu_dev,
712                                    u32 request, int index, int fd)
713 {
714         struct vhost_user_msg msg = {
715                 .header.request = request,
716                 .header.size = sizeof(msg.payload.integer),
717                 .payload.integer = index,
718         };
719
720         if (index & ~VHOST_USER_VRING_INDEX_MASK)
721                 return -EINVAL;
722         if (fd < 0) {
723                 msg.payload.integer |= VHOST_USER_VRING_POLL_MASK;
724                 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
725         }
726         return vhost_user_send(vu_dev, false, &msg, &fd, 1);
727 }
728
729 static int vhost_user_set_vring_call(struct virtio_uml_device *vu_dev,
730                                      int index, int fd)
731 {
732         return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_CALL,
733                                        index, fd);
734 }
735
736 static int vhost_user_set_vring_kick(struct virtio_uml_device *vu_dev,
737                                      int index, int fd)
738 {
739         return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_KICK,
740                                        index, fd);
741 }
742
743 static int vhost_user_set_vring_enable(struct virtio_uml_device *vu_dev,
744                                        u32 index, bool enable)
745 {
746         if (!(vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)))
747                 return 0;
748
749         return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_ENABLE,
750                                           index, enable);
751 }
752
753
754 /* Virtio interface */
755
756 static bool vu_notify(struct virtqueue *vq)
757 {
758         struct virtio_uml_vq_info *info = vq->priv;
759         const uint64_t n = 1;
760         int rc;
761
762         if (info->suspended)
763                 return true;
764
765         time_travel_propagate_time();
766
767         if (info->kick_fd < 0) {
768                 struct virtio_uml_device *vu_dev;
769
770                 vu_dev = to_virtio_uml_device(vq->vdev);
771
772                 return vhost_user_set_vring_state(vu_dev, VHOST_USER_VRING_KICK,
773                                                   vq->index, 0) == 0;
774         }
775
776         do {
777                 rc = os_write_file(info->kick_fd, &n, sizeof(n));
778         } while (rc == -EINTR);
779         return !WARN(rc != sizeof(n), "write returned %d\n", rc);
780 }
781
782 static irqreturn_t vu_interrupt(int irq, void *opaque)
783 {
784         struct virtqueue *vq = opaque;
785         struct virtio_uml_vq_info *info = vq->priv;
786         uint64_t n;
787         int rc;
788         irqreturn_t ret = IRQ_NONE;
789
790         do {
791                 rc = os_read_file(info->call_fd, &n, sizeof(n));
792                 if (rc == sizeof(n))
793                         ret |= vring_interrupt(irq, vq);
794         } while (rc == sizeof(n) || rc == -EINTR);
795         WARN(rc != -EAGAIN, "read returned %d\n", rc);
796         return ret;
797 }
798
799
800 static void vu_get(struct virtio_device *vdev, unsigned offset,
801                    void *buf, unsigned len)
802 {
803         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
804
805         vhost_user_get_config(vu_dev, offset, buf, len);
806 }
807
808 static void vu_set(struct virtio_device *vdev, unsigned offset,
809                    const void *buf, unsigned len)
810 {
811         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
812
813         vhost_user_set_config(vu_dev, offset, buf, len);
814 }
815
816 static u8 vu_get_status(struct virtio_device *vdev)
817 {
818         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
819
820         return vu_dev->status;
821 }
822
823 static void vu_set_status(struct virtio_device *vdev, u8 status)
824 {
825         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
826
827         vu_dev->status = status;
828 }
829
830 static void vu_reset(struct virtio_device *vdev)
831 {
832         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
833
834         vu_dev->status = 0;
835 }
836
837 static void vu_del_vq(struct virtqueue *vq)
838 {
839         struct virtio_uml_vq_info *info = vq->priv;
840
841         if (info->call_fd >= 0) {
842                 struct virtio_uml_device *vu_dev;
843
844                 vu_dev = to_virtio_uml_device(vq->vdev);
845
846                 um_free_irq(vu_dev->irq, vq);
847                 os_close_file(info->call_fd);
848         }
849
850         if (info->kick_fd >= 0)
851                 os_close_file(info->kick_fd);
852
853         vring_del_virtqueue(vq);
854         kfree(info);
855 }
856
857 static void vu_del_vqs(struct virtio_device *vdev)
858 {
859         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
860         struct virtqueue *vq, *n;
861         u64 features;
862
863         /* Note: reverse order as a workaround to a decoding bug in snabb */
864         list_for_each_entry_reverse(vq, &vdev->vqs, list)
865                 WARN_ON(vhost_user_set_vring_enable(vu_dev, vq->index, false));
866
867         /* Ensure previous messages have been processed */
868         WARN_ON(vhost_user_get_features(vu_dev, &features));
869
870         list_for_each_entry_safe(vq, n, &vdev->vqs, list)
871                 vu_del_vq(vq);
872 }
873
874 static int vu_setup_vq_call_fd(struct virtio_uml_device *vu_dev,
875                                struct virtqueue *vq)
876 {
877         struct virtio_uml_vq_info *info = vq->priv;
878         int call_fds[2];
879         int rc;
880
881         /* no call FD needed/desired in this case */
882         if (vu_dev->protocol_features &
883                         BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) &&
884             vu_dev->protocol_features &
885                         BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
886                 info->call_fd = -1;
887                 return 0;
888         }
889
890         /* Use a pipe for call fd, since SIGIO is not supported for eventfd */
891         rc = os_pipe(call_fds, true, true);
892         if (rc < 0)
893                 return rc;
894
895         info->call_fd = call_fds[0];
896         rc = um_request_irq(vu_dev->irq, info->call_fd, IRQ_READ,
897                             vu_interrupt, IRQF_SHARED, info->name, vq);
898         if (rc < 0)
899                 goto close_both;
900
901         rc = vhost_user_set_vring_call(vu_dev, vq->index, call_fds[1]);
902         if (rc)
903                 goto release_irq;
904
905         goto out;
906
907 release_irq:
908         um_free_irq(vu_dev->irq, vq);
909 close_both:
910         os_close_file(call_fds[0]);
911 out:
912         /* Close (unused) write end of call fds */
913         os_close_file(call_fds[1]);
914
915         return rc;
916 }
917
918 static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
919                                      unsigned index, vq_callback_t *callback,
920                                      const char *name, bool ctx)
921 {
922         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
923         struct platform_device *pdev = vu_dev->pdev;
924         struct virtio_uml_vq_info *info;
925         struct virtqueue *vq;
926         int num = MAX_SUPPORTED_QUEUE_SIZE;
927         int rc;
928
929         info = kzalloc(sizeof(*info), GFP_KERNEL);
930         if (!info) {
931                 rc = -ENOMEM;
932                 goto error_kzalloc;
933         }
934         snprintf(info->name, sizeof(info->name), "%s.%d-%s", pdev->name,
935                  pdev->id, name);
936
937         vq = vring_create_virtqueue(index, num, PAGE_SIZE, vdev, true, true,
938                                     ctx, vu_notify, callback, info->name);
939         if (!vq) {
940                 rc = -ENOMEM;
941                 goto error_create;
942         }
943         vq->priv = info;
944         num = virtqueue_get_vring_size(vq);
945
946         if (vu_dev->protocol_features &
947                         BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS)) {
948                 info->kick_fd = -1;
949         } else {
950                 rc = os_eventfd(0, 0);
951                 if (rc < 0)
952                         goto error_kick;
953                 info->kick_fd = rc;
954         }
955
956         rc = vu_setup_vq_call_fd(vu_dev, vq);
957         if (rc)
958                 goto error_call;
959
960         rc = vhost_user_set_vring_num(vu_dev, index, num);
961         if (rc)
962                 goto error_setup;
963
964         rc = vhost_user_set_vring_base(vu_dev, index, 0);
965         if (rc)
966                 goto error_setup;
967
968         rc = vhost_user_set_vring_addr(vu_dev, index,
969                                        virtqueue_get_desc_addr(vq),
970                                        virtqueue_get_used_addr(vq),
971                                        virtqueue_get_avail_addr(vq),
972                                        (u64) -1);
973         if (rc)
974                 goto error_setup;
975
976         return vq;
977
978 error_setup:
979         if (info->call_fd >= 0) {
980                 um_free_irq(vu_dev->irq, vq);
981                 os_close_file(info->call_fd);
982         }
983 error_call:
984         if (info->kick_fd >= 0)
985                 os_close_file(info->kick_fd);
986 error_kick:
987         vring_del_virtqueue(vq);
988 error_create:
989         kfree(info);
990 error_kzalloc:
991         return ERR_PTR(rc);
992 }
993
994 static int vu_find_vqs(struct virtio_device *vdev, unsigned nvqs,
995                        struct virtqueue *vqs[], vq_callback_t *callbacks[],
996                        const char * const names[], const bool *ctx,
997                        struct irq_affinity *desc)
998 {
999         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1000         int i, queue_idx = 0, rc;
1001         struct virtqueue *vq;
1002
1003         /* not supported for now */
1004         if (WARN_ON(nvqs > 64))
1005                 return -EINVAL;
1006
1007         rc = vhost_user_set_mem_table(vu_dev);
1008         if (rc)
1009                 return rc;
1010
1011         for (i = 0; i < nvqs; ++i) {
1012                 if (!names[i]) {
1013                         vqs[i] = NULL;
1014                         continue;
1015                 }
1016
1017                 vqs[i] = vu_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
1018                                      ctx ? ctx[i] : false);
1019                 if (IS_ERR(vqs[i])) {
1020                         rc = PTR_ERR(vqs[i]);
1021                         goto error_setup;
1022                 }
1023         }
1024
1025         list_for_each_entry(vq, &vdev->vqs, list) {
1026                 struct virtio_uml_vq_info *info = vq->priv;
1027
1028                 if (info->kick_fd >= 0) {
1029                         rc = vhost_user_set_vring_kick(vu_dev, vq->index,
1030                                                        info->kick_fd);
1031                         if (rc)
1032                                 goto error_setup;
1033                 }
1034
1035                 rc = vhost_user_set_vring_enable(vu_dev, vq->index, true);
1036                 if (rc)
1037                         goto error_setup;
1038         }
1039
1040         return 0;
1041
1042 error_setup:
1043         vu_del_vqs(vdev);
1044         return rc;
1045 }
1046
1047 static u64 vu_get_features(struct virtio_device *vdev)
1048 {
1049         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1050
1051         return vu_dev->features;
1052 }
1053
1054 static int vu_finalize_features(struct virtio_device *vdev)
1055 {
1056         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1057         u64 supported = vdev->features & VHOST_USER_SUPPORTED_F;
1058
1059         vring_transport_features(vdev);
1060         vu_dev->features = vdev->features | supported;
1061
1062         return vhost_user_set_features(vu_dev, vu_dev->features);
1063 }
1064
1065 static const char *vu_bus_name(struct virtio_device *vdev)
1066 {
1067         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1068
1069         return vu_dev->pdev->name;
1070 }
1071
1072 static const struct virtio_config_ops virtio_uml_config_ops = {
1073         .get = vu_get,
1074         .set = vu_set,
1075         .get_status = vu_get_status,
1076         .set_status = vu_set_status,
1077         .reset = vu_reset,
1078         .find_vqs = vu_find_vqs,
1079         .del_vqs = vu_del_vqs,
1080         .get_features = vu_get_features,
1081         .finalize_features = vu_finalize_features,
1082         .bus_name = vu_bus_name,
1083 };
1084
1085 static void virtio_uml_release_dev(struct device *d)
1086 {
1087         struct virtio_device *vdev =
1088                         container_of(d, struct virtio_device, dev);
1089         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1090
1091         /* might not have been opened due to not negotiating the feature */
1092         if (vu_dev->req_fd >= 0) {
1093                 um_free_irq(vu_dev->irq, vu_dev);
1094                 os_close_file(vu_dev->req_fd);
1095         }
1096
1097         os_close_file(vu_dev->sock);
1098         kfree(vu_dev);
1099 }
1100
1101 /* Platform device */
1102
1103 static int virtio_uml_probe(struct platform_device *pdev)
1104 {
1105         struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1106         struct virtio_uml_device *vu_dev;
1107         int rc;
1108
1109         if (!pdata)
1110                 return -EINVAL;
1111
1112         vu_dev = kzalloc(sizeof(*vu_dev), GFP_KERNEL);
1113         if (!vu_dev)
1114                 return -ENOMEM;
1115
1116         vu_dev->vdev.dev.parent = &pdev->dev;
1117         vu_dev->vdev.dev.release = virtio_uml_release_dev;
1118         vu_dev->vdev.config = &virtio_uml_config_ops;
1119         vu_dev->vdev.id.device = pdata->virtio_device_id;
1120         vu_dev->vdev.id.vendor = VIRTIO_DEV_ANY_ID;
1121         vu_dev->pdev = pdev;
1122         vu_dev->req_fd = -1;
1123
1124         do {
1125                 rc = os_connect_socket(pdata->socket_path);
1126         } while (rc == -EINTR);
1127         if (rc < 0)
1128                 return rc;
1129         vu_dev->sock = rc;
1130
1131         spin_lock_init(&vu_dev->sock_lock);
1132
1133         rc = vhost_user_init(vu_dev);
1134         if (rc)
1135                 goto error_init;
1136
1137         platform_set_drvdata(pdev, vu_dev);
1138
1139         device_set_wakeup_capable(&vu_dev->vdev.dev, true);
1140
1141         rc = register_virtio_device(&vu_dev->vdev);
1142         if (rc)
1143                 put_device(&vu_dev->vdev.dev);
1144         vu_dev->registered = 1;
1145         return rc;
1146
1147 error_init:
1148         os_close_file(vu_dev->sock);
1149         return rc;
1150 }
1151
1152 static int virtio_uml_remove(struct platform_device *pdev)
1153 {
1154         struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1155
1156         unregister_virtio_device(&vu_dev->vdev);
1157         return 0;
1158 }
1159
1160 /* Command line device list */
1161
1162 static void vu_cmdline_release_dev(struct device *d)
1163 {
1164 }
1165
1166 static struct device vu_cmdline_parent = {
1167         .init_name = "virtio-uml-cmdline",
1168         .release = vu_cmdline_release_dev,
1169 };
1170
1171 static bool vu_cmdline_parent_registered;
1172 static int vu_cmdline_id;
1173
1174 static int vu_unregister_cmdline_device(struct device *dev, void *data)
1175 {
1176         struct platform_device *pdev = to_platform_device(dev);
1177         struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1178
1179         kfree(pdata->socket_path);
1180         platform_device_unregister(pdev);
1181         return 0;
1182 }
1183
1184 static void vu_conn_broken(struct work_struct *wk)
1185 {
1186         struct virtio_uml_platform_data *pdata;
1187
1188         pdata = container_of(wk, struct virtio_uml_platform_data, conn_broken_wk);
1189         vu_unregister_cmdline_device(&pdata->pdev->dev, NULL);
1190 }
1191
1192 static int vu_cmdline_set(const char *device, const struct kernel_param *kp)
1193 {
1194         const char *ids = strchr(device, ':');
1195         unsigned int virtio_device_id;
1196         int processed, consumed, err;
1197         char *socket_path;
1198         struct virtio_uml_platform_data pdata, *ppdata;
1199         struct platform_device *pdev;
1200
1201         if (!ids || ids == device)
1202                 return -EINVAL;
1203
1204         processed = sscanf(ids, ":%u%n:%d%n",
1205                            &virtio_device_id, &consumed,
1206                            &vu_cmdline_id, &consumed);
1207
1208         if (processed < 1 || ids[consumed])
1209                 return -EINVAL;
1210
1211         if (!vu_cmdline_parent_registered) {
1212                 err = device_register(&vu_cmdline_parent);
1213                 if (err) {
1214                         pr_err("Failed to register parent device!\n");
1215                         put_device(&vu_cmdline_parent);
1216                         return err;
1217                 }
1218                 vu_cmdline_parent_registered = true;
1219         }
1220
1221         socket_path = kmemdup_nul(device, ids - device, GFP_KERNEL);
1222         if (!socket_path)
1223                 return -ENOMEM;
1224
1225         pdata.virtio_device_id = (u32) virtio_device_id;
1226         pdata.socket_path = socket_path;
1227
1228         pr_info("Registering device virtio-uml.%d id=%d at %s\n",
1229                 vu_cmdline_id, virtio_device_id, socket_path);
1230
1231         pdev = platform_device_register_data(&vu_cmdline_parent, "virtio-uml",
1232                                              vu_cmdline_id++, &pdata,
1233                                              sizeof(pdata));
1234         err = PTR_ERR_OR_ZERO(pdev);
1235         if (err)
1236                 goto free;
1237
1238         ppdata = pdev->dev.platform_data;
1239         ppdata->pdev = pdev;
1240         INIT_WORK(&ppdata->conn_broken_wk, vu_conn_broken);
1241
1242         return 0;
1243
1244 free:
1245         kfree(socket_path);
1246         return err;
1247 }
1248
1249 static int vu_cmdline_get_device(struct device *dev, void *data)
1250 {
1251         struct platform_device *pdev = to_platform_device(dev);
1252         struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1253         char *buffer = data;
1254         unsigned int len = strlen(buffer);
1255
1256         snprintf(buffer + len, PAGE_SIZE - len, "%s:%d:%d\n",
1257                  pdata->socket_path, pdata->virtio_device_id, pdev->id);
1258         return 0;
1259 }
1260
1261 static int vu_cmdline_get(char *buffer, const struct kernel_param *kp)
1262 {
1263         buffer[0] = '\0';
1264         if (vu_cmdline_parent_registered)
1265                 device_for_each_child(&vu_cmdline_parent, buffer,
1266                                       vu_cmdline_get_device);
1267         return strlen(buffer) + 1;
1268 }
1269
1270 static const struct kernel_param_ops vu_cmdline_param_ops = {
1271         .set = vu_cmdline_set,
1272         .get = vu_cmdline_get,
1273 };
1274
1275 device_param_cb(device, &vu_cmdline_param_ops, NULL, S_IRUSR);
1276 __uml_help(vu_cmdline_param_ops,
1277 "virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]\n"
1278 "    Configure a virtio device over a vhost-user socket.\n"
1279 "    See virtio_ids.h for a list of possible virtio device id values.\n"
1280 "    Optionally use a specific platform_device id.\n\n"
1281 );
1282
1283
1284 static void vu_unregister_cmdline_devices(void)
1285 {
1286         if (vu_cmdline_parent_registered) {
1287                 device_for_each_child(&vu_cmdline_parent, NULL,
1288                                       vu_unregister_cmdline_device);
1289                 device_unregister(&vu_cmdline_parent);
1290                 vu_cmdline_parent_registered = false;
1291         }
1292 }
1293
1294 /* Platform driver */
1295
1296 static const struct of_device_id virtio_uml_match[] = {
1297         { .compatible = "virtio,uml", },
1298         { }
1299 };
1300 MODULE_DEVICE_TABLE(of, virtio_uml_match);
1301
1302 static int virtio_uml_suspend(struct platform_device *pdev, pm_message_t state)
1303 {
1304         struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1305         struct virtqueue *vq;
1306
1307         virtio_device_for_each_vq((&vu_dev->vdev), vq) {
1308                 struct virtio_uml_vq_info *info = vq->priv;
1309
1310                 info->suspended = true;
1311                 vhost_user_set_vring_enable(vu_dev, vq->index, false);
1312         }
1313
1314         if (!device_may_wakeup(&vu_dev->vdev.dev)) {
1315                 vu_dev->suspended = true;
1316                 return 0;
1317         }
1318
1319         return irq_set_irq_wake(vu_dev->irq, 1);
1320 }
1321
1322 static int virtio_uml_resume(struct platform_device *pdev)
1323 {
1324         struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1325         struct virtqueue *vq;
1326
1327         virtio_device_for_each_vq((&vu_dev->vdev), vq) {
1328                 struct virtio_uml_vq_info *info = vq->priv;
1329
1330                 info->suspended = false;
1331                 vhost_user_set_vring_enable(vu_dev, vq->index, true);
1332         }
1333
1334         vu_dev->suspended = false;
1335
1336         if (!device_may_wakeup(&vu_dev->vdev.dev))
1337                 return 0;
1338
1339         return irq_set_irq_wake(vu_dev->irq, 0);
1340 }
1341
1342 static struct platform_driver virtio_uml_driver = {
1343         .probe = virtio_uml_probe,
1344         .remove = virtio_uml_remove,
1345         .driver = {
1346                 .name = "virtio-uml",
1347                 .of_match_table = virtio_uml_match,
1348         },
1349         .suspend = virtio_uml_suspend,
1350         .resume = virtio_uml_resume,
1351 };
1352
1353 static int __init virtio_uml_init(void)
1354 {
1355         return platform_driver_register(&virtio_uml_driver);
1356 }
1357
1358 static void __exit virtio_uml_exit(void)
1359 {
1360         platform_driver_unregister(&virtio_uml_driver);
1361         vu_unregister_cmdline_devices();
1362 }
1363
1364 module_init(virtio_uml_init);
1365 module_exit(virtio_uml_exit);
1366 __uml_exitcall(virtio_uml_exit);
1367
1368 MODULE_DESCRIPTION("UML driver for vhost-user virtio devices");
1369 MODULE_LICENSE("GPL");
This page took 0.10507 seconds and 4 git commands to generate.