blob: e4613a3c362dae8ffbb1e701b4dca51bb858625a [file] [log] [blame]
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001/* Copyright (C) 2009 Red Hat, Inc.
2 * Copyright (C) 2006 Rusty Russell IBM Corporation
3 *
4 * Author: Michael S. Tsirkin <mst@redhat.com>
5 *
6 * Inspiration, some code, and most witty comments come from
Rob Landley61516582011-05-06 09:27:36 -07007 * Documentation/virtual/lguest/lguest.c, by Rusty Russell
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00008 *
9 * This work is licensed under the terms of the GNU GPL, version 2.
10 *
11 * Generic code for virtio server in host kernel.
12 */
13
14#include <linux/eventfd.h>
15#include <linux/vhost.h>
Asias He35596b22013-08-19 09:23:19 +080016#include <linux/uio.h>
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +000017#include <linux/mm.h>
Michael S. Tsirkin64e1c802010-10-06 15:34:45 +020018#include <linux/mmu_context.h>
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +000019#include <linux/miscdevice.h>
20#include <linux/mutex.h>
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +000021#include <linux/poll.h>
22#include <linux/file.h>
23#include <linux/highmem.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090024#include <linux/slab.h>
Igor Mammedov4de72552015-07-01 11:07:09 +020025#include <linux/vmalloc.h>
Tejun Heoc23f34452010-06-02 20:40:00 +020026#include <linux/kthread.h>
Michael S. Tsirkin9e3d1952010-07-27 22:56:50 +030027#include <linux/cgroup.h>
Asias He6ac1afb2013-05-06 16:38:21 +080028#include <linux/module.h>
Igor Mammedovbcfeaca2015-06-16 18:33:35 +020029#include <linux/sort.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010030#include <linux/sched/mm.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010031#include <linux/sched/signal.h>
Jason Wanga9709d62016-06-23 02:04:31 -040032#include <linux/interval_tree_generic.h>
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +000033
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +000034#include "vhost.h"
35
Igor Mammedovc9ce42f2015-07-02 15:08:11 +020036static ushort max_mem_regions = 64;
37module_param(max_mem_regions, ushort, 0444);
38MODULE_PARM_DESC(max_mem_regions,
39 "Maximum number of memory regions in memory map. (default: 64)");
Jason Wang6b1e6cc2016-06-23 02:04:32 -040040static int max_iotlb_entries = 2048;
41module_param(max_iotlb_entries, int, 0444);
42MODULE_PARM_DESC(max_iotlb_entries,
43 "Maximum number of iotlb entries. (default: 2048)");
Igor Mammedovc9ce42f2015-07-02 15:08:11 +020044
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +000045enum {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +000046 VHOST_MEMORY_F_LOG = 0x1,
47};
48
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +030049#define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num])
50#define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num])
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +030051
Jason Wanga9709d62016-06-23 02:04:31 -040052INTERVAL_TREE_DEFINE(struct vhost_umem_node,
53 rb, __u64, __subtree_last,
Michael S. Tsirkin2f952c02016-12-06 05:57:54 +020054 START, LAST, static inline, vhost_umem_interval_tree);
Jason Wanga9709d62016-06-23 02:04:31 -040055
Greg Kurz2751c982015-04-24 14:27:24 +020056#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
Greg Kurzc5072032016-02-16 15:59:34 +010057static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
Greg Kurz2751c982015-04-24 14:27:24 +020058{
59 vq->user_be = !virtio_legacy_is_little_endian();
60}
61
Greg Kurzc5072032016-02-16 15:59:34 +010062static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq)
63{
64 vq->user_be = true;
65}
66
67static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq)
68{
69 vq->user_be = false;
70}
71
Greg Kurz2751c982015-04-24 14:27:24 +020072static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
73{
74 struct vhost_vring_state s;
75
76 if (vq->private_data)
77 return -EBUSY;
78
79 if (copy_from_user(&s, argp, sizeof(s)))
80 return -EFAULT;
81
82 if (s.num != VHOST_VRING_LITTLE_ENDIAN &&
83 s.num != VHOST_VRING_BIG_ENDIAN)
84 return -EINVAL;
85
Greg Kurzc5072032016-02-16 15:59:34 +010086 if (s.num == VHOST_VRING_BIG_ENDIAN)
87 vhost_enable_cross_endian_big(vq);
88 else
89 vhost_enable_cross_endian_little(vq);
Greg Kurz2751c982015-04-24 14:27:24 +020090
91 return 0;
92}
93
94static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
95 int __user *argp)
96{
97 struct vhost_vring_state s = {
98 .index = idx,
99 .num = vq->user_be
100 };
101
102 if (copy_to_user(argp, &s, sizeof(s)))
103 return -EFAULT;
104
105 return 0;
106}
107
108static void vhost_init_is_le(struct vhost_virtqueue *vq)
109{
110 /* Note for legacy virtio: user_be is initialized at reset time
111 * according to the host endianness. If userspace does not set an
112 * explicit endianness, the default behavior is native endian, as
113 * expected by legacy virtio.
114 */
115 vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be;
116}
117#else
Greg Kurzc5072032016-02-16 15:59:34 +0100118static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
Greg Kurz2751c982015-04-24 14:27:24 +0200119{
120}
121
122static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
123{
124 return -ENOIOCTLCMD;
125}
126
127static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
128 int __user *argp)
129{
130 return -ENOIOCTLCMD;
131}
132
133static void vhost_init_is_le(struct vhost_virtqueue *vq)
134{
Halil Pasiccda8bba2017-01-30 11:09:36 +0100135 vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1)
136 || virtio_legacy_is_little_endian();
Greg Kurz2751c982015-04-24 14:27:24 +0200137}
138#endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */
139
Greg Kurzc5072032016-02-16 15:59:34 +0100140static void vhost_reset_is_le(struct vhost_virtqueue *vq)
141{
Halil Pasiccda8bba2017-01-30 11:09:36 +0100142 vhost_init_is_le(vq);
Greg Kurzc5072032016-02-16 15:59:34 +0100143}
144
Jason Wang7235acd2016-04-25 22:14:32 -0400145struct vhost_flush_struct {
146 struct vhost_work work;
147 struct completion wait_event;
148};
149
150static void vhost_flush_work(struct vhost_work *work)
151{
152 struct vhost_flush_struct *s;
153
154 s = container_of(work, struct vhost_flush_struct, work);
155 complete(&s->wait_event);
156}
157
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000158static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh,
159 poll_table *pt)
160{
161 struct vhost_poll *poll;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000162
Krishna Kumard47effe2011-03-01 17:06:37 +0530163 poll = container_of(pt, struct vhost_poll, table);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000164 poll->wqh = wqh;
165 add_wait_queue(wqh, &poll->wait);
166}
167
Ingo Molnarac6424b2017-06-20 12:06:13 +0200168static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000169 void *key)
170{
Tejun Heoc23f34452010-06-02 20:40:00 +0200171 struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
172
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000173 if (!((unsigned long)key & poll->mask))
174 return 0;
175
Tejun Heoc23f34452010-06-02 20:40:00 +0200176 vhost_poll_queue(poll);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000177 return 0;
178}
179
Stefan Hajnoczi163049a2012-07-21 06:55:37 +0000180void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000181{
Jason Wang04b96e52016-04-25 22:14:33 -0400182 clear_bit(VHOST_WORK_QUEUED, &work->flags);
Tejun Heoc23f34452010-06-02 20:40:00 +0200183 work->fn = fn;
184 init_waitqueue_head(&work->done);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000185}
Asias He6ac1afb2013-05-06 16:38:21 +0800186EXPORT_SYMBOL_GPL(vhost_work_init);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000187
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300188/* Init poll structure */
189void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
190 unsigned long mask, struct vhost_dev *dev)
191{
192 init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup);
193 init_poll_funcptr(&poll->table, vhost_poll_func);
194 poll->mask = mask;
195 poll->dev = dev;
Jason Wang2b8b3282013-01-28 01:05:18 +0000196 poll->wqh = NULL;
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300197
198 vhost_work_init(&poll->work, fn);
199}
Asias He6ac1afb2013-05-06 16:38:21 +0800200EXPORT_SYMBOL_GPL(vhost_poll_init);
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300201
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000202/* Start polling a file. We add ourselves to file's wait queue. The caller must
203 * keep a reference to a file until after vhost_poll_stop is called. */
Jason Wang2b8b3282013-01-28 01:05:18 +0000204int vhost_poll_start(struct vhost_poll *poll, struct file *file)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000205{
206 unsigned long mask;
Jason Wang2b8b3282013-01-28 01:05:18 +0000207 int ret = 0;
Krishna Kumard47effe2011-03-01 17:06:37 +0530208
Jason Wang70181d512013-04-10 20:50:48 +0000209 if (poll->wqh)
210 return 0;
211
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000212 mask = file->f_op->poll(file, &poll->table);
213 if (mask)
214 vhost_poll_wakeup(&poll->wait, 0, 0, (void *)mask);
Jason Wang2b8b3282013-01-28 01:05:18 +0000215 if (mask & POLLERR) {
216 if (poll->wqh)
217 remove_wait_queue(poll->wqh, &poll->wait);
218 ret = -EINVAL;
219 }
220
221 return ret;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000222}
Asias He6ac1afb2013-05-06 16:38:21 +0800223EXPORT_SYMBOL_GPL(vhost_poll_start);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000224
225/* Stop polling a file. After this function returns, it becomes safe to drop the
226 * file reference. You must also flush afterwards. */
227void vhost_poll_stop(struct vhost_poll *poll)
228{
Jason Wang2b8b3282013-01-28 01:05:18 +0000229 if (poll->wqh) {
230 remove_wait_queue(poll->wqh, &poll->wait);
231 poll->wqh = NULL;
232 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000233}
Asias He6ac1afb2013-05-06 16:38:21 +0800234EXPORT_SYMBOL_GPL(vhost_poll_stop);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000235
Asias He6ac1afb2013-05-06 16:38:21 +0800236void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000237{
Jason Wang7235acd2016-04-25 22:14:32 -0400238 struct vhost_flush_struct flush;
Tejun Heoc23f34452010-06-02 20:40:00 +0200239
Jason Wang7235acd2016-04-25 22:14:32 -0400240 if (dev->worker) {
241 init_completion(&flush.wait_event);
242 vhost_work_init(&flush.work, vhost_flush_work);
243
244 vhost_work_queue(dev, &flush.work);
245 wait_for_completion(&flush.wait_event);
246 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000247}
Asias He6ac1afb2013-05-06 16:38:21 +0800248EXPORT_SYMBOL_GPL(vhost_work_flush);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000249
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300250/* Flush any work that has been scheduled. When calling this, don't hold any
251 * locks that are also used by the callback. */
252void vhost_poll_flush(struct vhost_poll *poll)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000253{
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300254 vhost_work_flush(poll->dev, &poll->work);
255}
Asias He6ac1afb2013-05-06 16:38:21 +0800256EXPORT_SYMBOL_GPL(vhost_poll_flush);
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300257
Stefan Hajnoczi163049a2012-07-21 06:55:37 +0000258void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300259{
Jason Wang04b96e52016-04-25 22:14:33 -0400260 if (!dev->worker)
261 return;
Tejun Heoc23f34452010-06-02 20:40:00 +0200262
Jason Wang04b96e52016-04-25 22:14:33 -0400263 if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
264 /* We can only add the work to the list after we're
265 * sure it was not in the list.
Peng Tao635abf02016-12-07 17:52:19 +0800266 * test_and_set_bit() implies a memory barrier.
Jason Wang04b96e52016-04-25 22:14:33 -0400267 */
Jason Wang04b96e52016-04-25 22:14:33 -0400268 llist_add(&work->node, &dev->work_list);
Tejun Heoc23f34452010-06-02 20:40:00 +0200269 wake_up_process(dev->worker);
270 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000271}
Asias He6ac1afb2013-05-06 16:38:21 +0800272EXPORT_SYMBOL_GPL(vhost_work_queue);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000273
Jason Wang526d3e72016-03-04 06:24:51 -0500274/* A lockless hint for busy polling code to exit the loop */
275bool vhost_has_work(struct vhost_dev *dev)
276{
Jason Wang04b96e52016-04-25 22:14:33 -0400277 return !llist_empty(&dev->work_list);
Jason Wang526d3e72016-03-04 06:24:51 -0500278}
279EXPORT_SYMBOL_GPL(vhost_has_work);
280
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300281void vhost_poll_queue(struct vhost_poll *poll)
282{
283 vhost_work_queue(poll->dev, &poll->work);
284}
Asias He6ac1afb2013-05-06 16:38:21 +0800285EXPORT_SYMBOL_GPL(vhost_poll_queue);
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300286
Jason Wangf8894912017-02-28 17:56:02 +0800287static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq)
288{
289 int j;
290
291 for (j = 0; j < VHOST_NUM_ADDRS; j++)
292 vq->meta_iotlb[j] = NULL;
293}
294
295static void vhost_vq_meta_reset(struct vhost_dev *d)
296{
297 int i;
298
299 for (i = 0; i < d->nvqs; ++i)
300 __vhost_vq_meta_reset(d->vqs[i]);
301}
302
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000303static void vhost_vq_reset(struct vhost_dev *dev,
304 struct vhost_virtqueue *vq)
305{
306 vq->num = 1;
307 vq->desc = NULL;
308 vq->avail = NULL;
309 vq->used = NULL;
310 vq->last_avail_idx = 0;
Jason Wang809ecb92016-12-12 14:46:49 +0800311 vq->last_used_event = 0;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000312 vq->avail_idx = 0;
313 vq->last_used_idx = 0;
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300314 vq->signalled_used = 0;
315 vq->signalled_used_valid = false;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000316 vq->used_flags = 0;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000317 vq->log_used = false;
318 vq->log_addr = -1ull;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000319 vq->private_data = NULL;
Michael S. Tsirkinea16c512014-06-05 15:20:23 +0300320 vq->acked_features = 0;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000321 vq->log_base = NULL;
322 vq->error_ctx = NULL;
323 vq->error = NULL;
324 vq->kick = NULL;
325 vq->call_ctx = NULL;
326 vq->call = NULL;
Michael S. Tsirkin73a99f02010-02-23 11:23:45 +0200327 vq->log_ctx = NULL;
Greg Kurzc5072032016-02-16 15:59:34 +0100328 vhost_reset_is_le(vq);
329 vhost_disable_cross_endian(vq);
Jason Wang03088132016-03-04 06:24:53 -0500330 vq->busyloop_timeout = 0;
Jason Wanga9709d62016-06-23 02:04:31 -0400331 vq->umem = NULL;
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400332 vq->iotlb = NULL;
Jason Wangf8894912017-02-28 17:56:02 +0800333 __vhost_vq_meta_reset(vq);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000334}
335
Tejun Heoc23f34452010-06-02 20:40:00 +0200336static int vhost_worker(void *data)
337{
338 struct vhost_dev *dev = data;
Jason Wang04b96e52016-04-25 22:14:33 -0400339 struct vhost_work *work, *work_next;
340 struct llist_node *node;
Jens Freimannd7ffde32012-06-26 00:59:58 +0000341 mm_segment_t oldfs = get_fs();
Tejun Heoc23f34452010-06-02 20:40:00 +0200342
Jens Freimannd7ffde32012-06-26 00:59:58 +0000343 set_fs(USER_DS);
Michael S. Tsirkin64e1c802010-10-06 15:34:45 +0200344 use_mm(dev->mm);
345
Tejun Heoc23f34452010-06-02 20:40:00 +0200346 for (;;) {
347 /* mb paired w/ kthread_stop */
348 set_current_state(TASK_INTERRUPTIBLE);
349
Tejun Heoc23f34452010-06-02 20:40:00 +0200350 if (kthread_should_stop()) {
Tejun Heoc23f34452010-06-02 20:40:00 +0200351 __set_current_state(TASK_RUNNING);
Michael S. Tsirkin64e1c802010-10-06 15:34:45 +0200352 break;
Tejun Heoc23f34452010-06-02 20:40:00 +0200353 }
Tejun Heoc23f34452010-06-02 20:40:00 +0200354
Jason Wang04b96e52016-04-25 22:14:33 -0400355 node = llist_del_all(&dev->work_list);
356 if (!node)
357 schedule();
358
359 node = llist_reverse_order(node);
360 /* make sure flag is seen after deletion */
361 smp_wmb();
362 llist_for_each_entry_safe(work, work_next, node, node) {
363 clear_bit(VHOST_WORK_QUEUED, &work->flags);
Tejun Heoc23f34452010-06-02 20:40:00 +0200364 __set_current_state(TASK_RUNNING);
365 work->fn(work);
Nadav Har'Eld550dda2012-02-27 15:07:29 +0200366 if (need_resched())
367 schedule();
Jason Wang04b96e52016-04-25 22:14:33 -0400368 }
Tejun Heoc23f34452010-06-02 20:40:00 +0200369 }
Michael S. Tsirkin64e1c802010-10-06 15:34:45 +0200370 unuse_mm(dev->mm);
Jens Freimannd7ffde32012-06-26 00:59:58 +0000371 set_fs(oldfs);
Michael S. Tsirkin64e1c802010-10-06 15:34:45 +0200372 return 0;
Tejun Heoc23f34452010-06-02 20:40:00 +0200373}
374
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000375static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
376{
377 kfree(vq->indirect);
378 vq->indirect = NULL;
379 kfree(vq->log);
380 vq->log = NULL;
381 kfree(vq->heads);
382 vq->heads = NULL;
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000383}
384
Jason Wange0e9b402010-09-14 23:53:05 +0800385/* Helper to allocate iovec buffers for all vqs. */
386static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
387{
Asias He6d5e6aa2013-05-06 16:38:23 +0800388 struct vhost_virtqueue *vq;
Jason Wange0e9b402010-09-14 23:53:05 +0800389 int i;
Krishna Kumard47effe2011-03-01 17:06:37 +0530390
Jason Wange0e9b402010-09-14 23:53:05 +0800391 for (i = 0; i < dev->nvqs; ++i) {
Asias He6d5e6aa2013-05-06 16:38:23 +0800392 vq = dev->vqs[i];
393 vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV,
394 GFP_KERNEL);
395 vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL);
396 vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL);
397 if (!vq->indirect || !vq->log || !vq->heads)
Jason Wange0e9b402010-09-14 23:53:05 +0800398 goto err_nomem;
399 }
400 return 0;
Krishna Kumard47effe2011-03-01 17:06:37 +0530401
Jason Wange0e9b402010-09-14 23:53:05 +0800402err_nomem:
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000403 for (; i >= 0; --i)
Asias He3ab2e422013-04-27 11:16:48 +0800404 vhost_vq_free_iovecs(dev->vqs[i]);
Jason Wange0e9b402010-09-14 23:53:05 +0800405 return -ENOMEM;
406}
407
408static void vhost_dev_free_iovecs(struct vhost_dev *dev)
409{
410 int i;
Krishna Kumard47effe2011-03-01 17:06:37 +0530411
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000412 for (i = 0; i < dev->nvqs; ++i)
Asias He3ab2e422013-04-27 11:16:48 +0800413 vhost_vq_free_iovecs(dev->vqs[i]);
Jason Wange0e9b402010-09-14 23:53:05 +0800414}
415
Zhi Yong Wu59566b6e2013-12-07 04:13:03 +0800416void vhost_dev_init(struct vhost_dev *dev,
Asias He3ab2e422013-04-27 11:16:48 +0800417 struct vhost_virtqueue **vqs, int nvqs)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000418{
Asias He6d5e6aa2013-05-06 16:38:23 +0800419 struct vhost_virtqueue *vq;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000420 int i;
Tejun Heoc23f34452010-06-02 20:40:00 +0200421
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000422 dev->vqs = vqs;
423 dev->nvqs = nvqs;
424 mutex_init(&dev->mutex);
425 dev->log_ctx = NULL;
426 dev->log_file = NULL;
Jason Wanga9709d62016-06-23 02:04:31 -0400427 dev->umem = NULL;
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400428 dev->iotlb = NULL;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000429 dev->mm = NULL;
Tejun Heoc23f34452010-06-02 20:40:00 +0200430 dev->worker = NULL;
Jason Wang04b96e52016-04-25 22:14:33 -0400431 init_llist_head(&dev->work_list);
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400432 init_waitqueue_head(&dev->wait);
433 INIT_LIST_HEAD(&dev->read_list);
434 INIT_LIST_HEAD(&dev->pending_list);
435 spin_lock_init(&dev->iotlb_lock);
Jason Wang04b96e52016-04-25 22:14:33 -0400436
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000437
438 for (i = 0; i < dev->nvqs; ++i) {
Asias He6d5e6aa2013-05-06 16:38:23 +0800439 vq = dev->vqs[i];
440 vq->log = NULL;
441 vq->indirect = NULL;
442 vq->heads = NULL;
443 vq->dev = dev;
444 mutex_init(&vq->mutex);
445 vhost_vq_reset(dev, vq);
446 if (vq->handle_kick)
447 vhost_poll_init(&vq->poll, vq->handle_kick,
448 POLLIN, dev);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000449 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000450}
Asias He6ac1afb2013-05-06 16:38:21 +0800451EXPORT_SYMBOL_GPL(vhost_dev_init);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000452
453/* Caller should have device mutex */
454long vhost_dev_check_owner(struct vhost_dev *dev)
455{
456 /* Are you the owner? If not, I don't think you mean to do that */
457 return dev->mm == current->mm ? 0 : -EPERM;
458}
Asias He6ac1afb2013-05-06 16:38:21 +0800459EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000460
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300461struct vhost_attach_cgroups_struct {
Krishna Kumard47effe2011-03-01 17:06:37 +0530462 struct vhost_work work;
463 struct task_struct *owner;
464 int ret;
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300465};
466
467static void vhost_attach_cgroups_work(struct vhost_work *work)
468{
Krishna Kumard47effe2011-03-01 17:06:37 +0530469 struct vhost_attach_cgroups_struct *s;
470
471 s = container_of(work, struct vhost_attach_cgroups_struct, work);
472 s->ret = cgroup_attach_task_all(s->owner, current);
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300473}
474
475static int vhost_attach_cgroups(struct vhost_dev *dev)
476{
Krishna Kumard47effe2011-03-01 17:06:37 +0530477 struct vhost_attach_cgroups_struct attach;
478
479 attach.owner = current;
480 vhost_work_init(&attach.work, vhost_attach_cgroups_work);
481 vhost_work_queue(dev, &attach.work);
482 vhost_work_flush(dev, &attach.work);
483 return attach.ret;
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300484}
485
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000486/* Caller should have device mutex */
Michael S. Tsirkin05c05352013-06-06 15:20:39 +0300487bool vhost_dev_has_owner(struct vhost_dev *dev)
488{
489 return dev->mm;
490}
Asias He6ac1afb2013-05-06 16:38:21 +0800491EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
Michael S. Tsirkin05c05352013-06-06 15:20:39 +0300492
493/* Caller should have device mutex */
Asias He54db63c2013-05-06 11:15:59 +0800494long vhost_dev_set_owner(struct vhost_dev *dev)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000495{
Tejun Heoc23f34452010-06-02 20:40:00 +0200496 struct task_struct *worker;
497 int err;
Krishna Kumard47effe2011-03-01 17:06:37 +0530498
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000499 /* Is there an owner already? */
Michael S. Tsirkin05c05352013-06-06 15:20:39 +0300500 if (vhost_dev_has_owner(dev)) {
Tejun Heoc23f34452010-06-02 20:40:00 +0200501 err = -EBUSY;
502 goto err_mm;
503 }
Krishna Kumard47effe2011-03-01 17:06:37 +0530504
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000505 /* No owner, become one */
506 dev->mm = get_task_mm(current);
Tejun Heoc23f34452010-06-02 20:40:00 +0200507 worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
508 if (IS_ERR(worker)) {
509 err = PTR_ERR(worker);
510 goto err_worker;
511 }
512
513 dev->worker = worker;
Michael S. Tsirkin87d6a412010-09-02 14:05:30 +0300514 wake_up_process(worker); /* avoid contributing to loadavg */
515
516 err = vhost_attach_cgroups(dev);
Michael S. Tsirkin9e3d1952010-07-27 22:56:50 +0300517 if (err)
518 goto err_cgroup;
Tejun Heoc23f34452010-06-02 20:40:00 +0200519
Jason Wange0e9b402010-09-14 23:53:05 +0800520 err = vhost_dev_alloc_iovecs(dev);
521 if (err)
522 goto err_cgroup;
523
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000524 return 0;
Michael S. Tsirkin9e3d1952010-07-27 22:56:50 +0300525err_cgroup:
526 kthread_stop(worker);
Michael S. Tsirkin615cc222010-09-02 14:16:36 +0300527 dev->worker = NULL;
Tejun Heoc23f34452010-06-02 20:40:00 +0200528err_worker:
529 if (dev->mm)
530 mmput(dev->mm);
531 dev->mm = NULL;
532err_mm:
533 return err;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000534}
Asias He6ac1afb2013-05-06 16:38:21 +0800535EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000536
Jason Wanga9709d62016-06-23 02:04:31 -0400537struct vhost_umem *vhost_dev_reset_owner_prepare(void)
538{
Michal Hocko6c5ab652017-05-08 15:57:15 -0700539 return kvzalloc(sizeof(struct vhost_umem), GFP_KERNEL);
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +0300540}
Asias He6ac1afb2013-05-06 16:38:21 +0800541EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000542
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +0300543/* Caller should have device mutex */
Jason Wanga9709d62016-06-23 02:04:31 -0400544void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem)
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +0300545{
Michael S. Tsirkin47283be2014-06-05 15:20:27 +0300546 int i;
547
Michael S. Tsirkinea5d4042011-11-27 19:05:58 +0200548 vhost_dev_cleanup(dev, true);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000549
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +0300550 /* Restore memory to default empty mapping. */
Jason Wanga9709d62016-06-23 02:04:31 -0400551 INIT_LIST_HEAD(&umem->umem_list);
552 dev->umem = umem;
Michael S. Tsirkin47283be2014-06-05 15:20:27 +0300553 /* We don't need VQ locks below since vhost_dev_cleanup makes sure
554 * VQs aren't running.
555 */
556 for (i = 0; i < dev->nvqs; ++i)
Jason Wanga9709d62016-06-23 02:04:31 -0400557 dev->vqs[i]->umem = umem;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000558}
Asias He6ac1afb2013-05-06 16:38:21 +0800559EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000560
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000561void vhost_dev_stop(struct vhost_dev *dev)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000562{
563 int i;
Krishna Kumard47effe2011-03-01 17:06:37 +0530564
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000565 for (i = 0; i < dev->nvqs; ++i) {
Asias He3ab2e422013-04-27 11:16:48 +0800566 if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) {
567 vhost_poll_stop(&dev->vqs[i]->poll);
568 vhost_poll_flush(&dev->vqs[i]->poll);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000569 }
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000570 }
571}
Asias He6ac1afb2013-05-06 16:38:21 +0800572EXPORT_SYMBOL_GPL(vhost_dev_stop);
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000573
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400574static void vhost_umem_free(struct vhost_umem *umem,
575 struct vhost_umem_node *node)
576{
577 vhost_umem_interval_tree_remove(node, &umem->umem_tree);
578 list_del(&node->link);
579 kfree(node);
580 umem->numem--;
581}
582
Jason Wanga9709d62016-06-23 02:04:31 -0400583static void vhost_umem_clean(struct vhost_umem *umem)
584{
585 struct vhost_umem_node *node, *tmp;
586
587 if (!umem)
588 return;
589
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400590 list_for_each_entry_safe(node, tmp, &umem->umem_list, link)
591 vhost_umem_free(umem, node);
592
Jason Wanga9709d62016-06-23 02:04:31 -0400593 kvfree(umem);
594}
595
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400596static void vhost_clear_msg(struct vhost_dev *dev)
597{
598 struct vhost_msg_node *node, *n;
599
600 spin_lock(&dev->iotlb_lock);
601
602 list_for_each_entry_safe(node, n, &dev->read_list, node) {
603 list_del(&node->node);
604 kfree(node);
605 }
606
607 list_for_each_entry_safe(node, n, &dev->pending_list, node) {
608 list_del(&node->node);
609 kfree(node);
610 }
611
612 spin_unlock(&dev->iotlb_lock);
613}
614
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000615/* Caller should have device mutex if and only if locked is set */
616void vhost_dev_cleanup(struct vhost_dev *dev, bool locked)
617{
618 int i;
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000619
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000620 for (i = 0; i < dev->nvqs; ++i) {
Asias He3ab2e422013-04-27 11:16:48 +0800621 if (dev->vqs[i]->error_ctx)
622 eventfd_ctx_put(dev->vqs[i]->error_ctx);
623 if (dev->vqs[i]->error)
624 fput(dev->vqs[i]->error);
625 if (dev->vqs[i]->kick)
626 fput(dev->vqs[i]->kick);
627 if (dev->vqs[i]->call_ctx)
628 eventfd_ctx_put(dev->vqs[i]->call_ctx);
629 if (dev->vqs[i]->call)
630 fput(dev->vqs[i]->call);
631 vhost_vq_reset(dev, dev->vqs[i]);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000632 }
Jason Wange0e9b402010-09-14 23:53:05 +0800633 vhost_dev_free_iovecs(dev);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000634 if (dev->log_ctx)
635 eventfd_ctx_put(dev->log_ctx);
636 dev->log_ctx = NULL;
637 if (dev->log_file)
638 fput(dev->log_file);
639 dev->log_file = NULL;
640 /* No one will access memory at this point */
Jason Wanga9709d62016-06-23 02:04:31 -0400641 vhost_umem_clean(dev->umem);
642 dev->umem = NULL;
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400643 vhost_umem_clean(dev->iotlb);
644 dev->iotlb = NULL;
645 vhost_clear_msg(dev);
646 wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM);
Jason Wang04b96e52016-04-25 22:14:33 -0400647 WARN_ON(!llist_empty(&dev->work_list));
Eric Dumazet78b620c2010-08-31 02:05:57 +0000648 if (dev->worker) {
649 kthread_stop(dev->worker);
650 dev->worker = NULL;
651 }
Michael S. Tsirkin533a19b2010-10-06 15:34:38 +0200652 if (dev->mm)
653 mmput(dev->mm);
654 dev->mm = NULL;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000655}
Asias He6ac1afb2013-05-06 16:38:21 +0800656EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000657
658static int log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
659{
660 u64 a = addr / VHOST_PAGE_SIZE / 8;
Krishna Kumard47effe2011-03-01 17:06:37 +0530661
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000662 /* Make sure 64 bit math will not overflow. */
663 if (a > ULONG_MAX - (unsigned long)log_base ||
664 a + (unsigned long)log_base > ULONG_MAX)
Dan Carpenter6d97e552010-10-11 19:24:19 +0200665 return 0;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000666
667 return access_ok(VERIFY_WRITE, log_base + a,
668 (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
669}
670
Michael S. Tsirkinec33d032016-08-01 23:20:53 +0300671static bool vhost_overflow(u64 uaddr, u64 size)
672{
673 /* Make sure 64 bit math will not overflow. */
674 return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size;
675}
676
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000677/* Caller should have vq mutex and device mutex. */
Jason Wanga9709d62016-06-23 02:04:31 -0400678static int vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem,
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000679 int log_all)
680{
Jason Wanga9709d62016-06-23 02:04:31 -0400681 struct vhost_umem_node *node;
Jeff Dike179b2842010-04-07 09:59:10 -0400682
Jason Wanga9709d62016-06-23 02:04:31 -0400683 if (!umem)
Michael S. Tsirkinf8322fb2010-05-27 12:28:03 +0300684 return 0;
Jeff Dike179b2842010-04-07 09:59:10 -0400685
Jason Wanga9709d62016-06-23 02:04:31 -0400686 list_for_each_entry(node, &umem->umem_list, link) {
687 unsigned long a = node->userspace_addr;
688
Michael S. Tsirkinec33d032016-08-01 23:20:53 +0300689 if (vhost_overflow(node->userspace_addr, node->size))
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000690 return 0;
Michael S. Tsirkinec33d032016-08-01 23:20:53 +0300691
692
693 if (!access_ok(VERIFY_WRITE, (void __user *)a,
Jason Wanga9709d62016-06-23 02:04:31 -0400694 node->size))
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000695 return 0;
696 else if (log_all && !log_access_ok(log_base,
Jason Wanga9709d62016-06-23 02:04:31 -0400697 node->start,
698 node->size))
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000699 return 0;
700 }
701 return 1;
702}
703
Jason Wangf8894912017-02-28 17:56:02 +0800704static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
705 u64 addr, unsigned int size,
706 int type)
707{
708 const struct vhost_umem_node *node = vq->meta_iotlb[type];
709
710 if (!node)
711 return NULL;
712
713 return (void *)(uintptr_t)(node->userspace_addr + addr - node->start);
714}
715
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000716/* Can we switch to this memory table? */
717/* Caller should have device mutex but not vq mutex */
Jason Wanga9709d62016-06-23 02:04:31 -0400718static int memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem,
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000719 int log_all)
720{
721 int i;
Krishna Kumard47effe2011-03-01 17:06:37 +0530722
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000723 for (i = 0; i < d->nvqs; ++i) {
724 int ok;
Michael S. Tsirkinea16c512014-06-05 15:20:23 +0300725 bool log;
726
Asias He3ab2e422013-04-27 11:16:48 +0800727 mutex_lock(&d->vqs[i]->mutex);
Michael S. Tsirkinea16c512014-06-05 15:20:23 +0300728 log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000729 /* If ring is inactive, will check when it's enabled. */
Asias He3ab2e422013-04-27 11:16:48 +0800730 if (d->vqs[i]->private_data)
Jason Wanga9709d62016-06-23 02:04:31 -0400731 ok = vq_memory_access_ok(d->vqs[i]->log_base,
732 umem, log);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000733 else
734 ok = 1;
Asias He3ab2e422013-04-27 11:16:48 +0800735 mutex_unlock(&d->vqs[i]->mutex);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +0000736 if (!ok)
737 return 0;
738 }
739 return 1;
740}
741
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400742static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
743 struct iovec iov[], int iov_size, int access);
Jason Wangbfe2bc52016-06-23 02:04:30 -0400744
Michael S. Tsirkin72952cc2016-12-06 06:01:41 +0200745static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
Jason Wangbfe2bc52016-06-23 02:04:30 -0400746 const void *from, unsigned size)
747{
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400748 int ret;
Jason Wangbfe2bc52016-06-23 02:04:30 -0400749
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400750 if (!vq->iotlb)
751 return __copy_to_user(to, from, size);
752 else {
753 /* This function should be called after iotlb
754 * prefetch, which means we're sure that all vq
755 * could be access through iotlb. So -EAGAIN should
756 * not happen in this case.
757 */
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400758 struct iov_iter t;
Jason Wangf8894912017-02-28 17:56:02 +0800759 void __user *uaddr = vhost_vq_meta_fetch(vq,
760 (u64)(uintptr_t)to, size,
761 VHOST_ADDR_DESC);
762
763 if (uaddr)
764 return __copy_to_user(uaddr, from, size);
765
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400766 ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
767 ARRAY_SIZE(vq->iotlb_iov),
768 VHOST_ACCESS_WO);
769 if (ret < 0)
770 goto out;
771 iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size);
772 ret = copy_to_iter(from, size, &t);
773 if (ret == size)
774 ret = 0;
775 }
776out:
777 return ret;
778}
Jason Wangbfe2bc52016-06-23 02:04:30 -0400779
780static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
Michael S. Tsirkin72952cc2016-12-06 06:01:41 +0200781 void __user *from, unsigned size)
Jason Wangbfe2bc52016-06-23 02:04:30 -0400782{
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400783 int ret;
784
785 if (!vq->iotlb)
786 return __copy_from_user(to, from, size);
787 else {
788 /* This function should be called after iotlb
789 * prefetch, which means we're sure that vq
790 * could be access through iotlb. So -EAGAIN should
791 * not happen in this case.
792 */
Jason Wangf8894912017-02-28 17:56:02 +0800793 void __user *uaddr = vhost_vq_meta_fetch(vq,
794 (u64)(uintptr_t)from, size,
795 VHOST_ADDR_DESC);
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400796 struct iov_iter f;
Jason Wangf8894912017-02-28 17:56:02 +0800797
798 if (uaddr)
799 return __copy_from_user(to, uaddr, size);
800
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400801 ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
802 ARRAY_SIZE(vq->iotlb_iov),
803 VHOST_ACCESS_RO);
804 if (ret < 0) {
805 vq_err(vq, "IOTLB translation failure: uaddr "
806 "%p size 0x%llx\n", from,
807 (unsigned long long) size);
808 goto out;
809 }
810 iov_iter_init(&f, READ, vq->iotlb_iov, ret, size);
811 ret = copy_from_iter(to, size, &f);
812 if (ret == size)
813 ret = 0;
814 }
815
816out:
817 return ret;
818}
819
Jason Wangf8894912017-02-28 17:56:02 +0800820static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
821 void __user *addr, unsigned int size,
822 int type)
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400823{
824 int ret;
825
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400826 ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
827 ARRAY_SIZE(vq->iotlb_iov),
828 VHOST_ACCESS_RO);
829 if (ret < 0) {
830 vq_err(vq, "IOTLB translation failure: uaddr "
831 "%p size 0x%llx\n", addr,
832 (unsigned long long) size);
833 return NULL;
834 }
835
836 if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
837 vq_err(vq, "Non atomic userspace memory access: uaddr "
838 "%p size 0x%llx\n", addr,
839 (unsigned long long) size);
840 return NULL;
841 }
842
843 return vq->iotlb_iov[0].iov_base;
844}
845
Jason Wangf8894912017-02-28 17:56:02 +0800846/* This function should be called after iotlb
847 * prefetch, which means we're sure that vq
848 * could be access through iotlb. So -EAGAIN should
849 * not happen in this case.
850 */
851static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
852 void *addr, unsigned int size,
853 int type)
854{
855 void __user *uaddr = vhost_vq_meta_fetch(vq,
856 (u64)(uintptr_t)addr, size, type);
857 if (uaddr)
858 return uaddr;
859
860 return __vhost_get_user_slow(vq, addr, size, type);
861}
862
863#define vhost_put_user(vq, x, ptr) \
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400864({ \
865 int ret = -EFAULT; \
866 if (!vq->iotlb) { \
867 ret = __put_user(x, ptr); \
868 } else { \
869 __typeof__(ptr) to = \
Jason Wangf8894912017-02-28 17:56:02 +0800870 (__typeof__(ptr)) __vhost_get_user(vq, ptr, \
871 sizeof(*ptr), VHOST_ADDR_USED); \
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400872 if (to != NULL) \
873 ret = __put_user(x, to); \
874 else \
875 ret = -EFAULT; \
876 } \
877 ret; \
878})
879
Jason Wangf8894912017-02-28 17:56:02 +0800880#define vhost_get_user(vq, x, ptr, type) \
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400881({ \
882 int ret; \
883 if (!vq->iotlb) { \
884 ret = __get_user(x, ptr); \
885 } else { \
886 __typeof__(ptr) from = \
Jason Wangf8894912017-02-28 17:56:02 +0800887 (__typeof__(ptr)) __vhost_get_user(vq, ptr, \
888 sizeof(*ptr), \
889 type); \
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400890 if (from != NULL) \
891 ret = __get_user(x, from); \
892 else \
893 ret = -EFAULT; \
894 } \
895 ret; \
896})
897
Jason Wangf8894912017-02-28 17:56:02 +0800898#define vhost_get_avail(vq, x, ptr) \
899 vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
900
901#define vhost_get_used(vq, x, ptr) \
902 vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
903
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400904static void vhost_dev_lock_vqs(struct vhost_dev *d)
905{
906 int i = 0;
907 for (i = 0; i < d->nvqs; ++i)
908 mutex_lock(&d->vqs[i]->mutex);
909}
910
911static void vhost_dev_unlock_vqs(struct vhost_dev *d)
912{
913 int i = 0;
914 for (i = 0; i < d->nvqs; ++i)
915 mutex_unlock(&d->vqs[i]->mutex);
916}
917
918static int vhost_new_umem_range(struct vhost_umem *umem,
919 u64 start, u64 size, u64 end,
920 u64 userspace_addr, int perm)
921{
922 struct vhost_umem_node *tmp, *node = kmalloc(sizeof(*node), GFP_ATOMIC);
923
924 if (!node)
925 return -ENOMEM;
926
927 if (umem->numem == max_iotlb_entries) {
928 tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link);
929 vhost_umem_free(umem, tmp);
930 }
931
932 node->start = start;
933 node->size = size;
934 node->last = end;
935 node->userspace_addr = userspace_addr;
936 node->perm = perm;
937 INIT_LIST_HEAD(&node->link);
938 list_add_tail(&node->link, &umem->umem_list);
939 vhost_umem_interval_tree_insert(node, &umem->umem_tree);
940 umem->numem++;
941
942 return 0;
943}
944
945static void vhost_del_umem_range(struct vhost_umem *umem,
946 u64 start, u64 end)
947{
948 struct vhost_umem_node *node;
949
950 while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
951 start, end)))
952 vhost_umem_free(umem, node);
953}
954
955static void vhost_iotlb_notify_vq(struct vhost_dev *d,
956 struct vhost_iotlb_msg *msg)
957{
958 struct vhost_msg_node *node, *n;
959
960 spin_lock(&d->iotlb_lock);
961
962 list_for_each_entry_safe(node, n, &d->pending_list, node) {
963 struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
964 if (msg->iova <= vq_msg->iova &&
965 msg->iova + msg->size - 1 > vq_msg->iova &&
966 vq_msg->type == VHOST_IOTLB_MISS) {
967 vhost_poll_queue(&node->vq->poll);
968 list_del(&node->node);
969 kfree(node);
970 }
971 }
972
973 spin_unlock(&d->iotlb_lock);
974}
975
976static int umem_access_ok(u64 uaddr, u64 size, int access)
977{
978 unsigned long a = uaddr;
979
Michael S. Tsirkinec33d032016-08-01 23:20:53 +0300980 /* Make sure 64 bit math will not overflow. */
981 if (vhost_overflow(uaddr, size))
982 return -EFAULT;
983
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400984 if ((access & VHOST_ACCESS_RO) &&
985 !access_ok(VERIFY_READ, (void __user *)a, size))
986 return -EFAULT;
987 if ((access & VHOST_ACCESS_WO) &&
988 !access_ok(VERIFY_WRITE, (void __user *)a, size))
989 return -EFAULT;
990 return 0;
991}
992
Michael S. Tsirkin72952cc2016-12-06 06:01:41 +0200993static int vhost_process_iotlb_msg(struct vhost_dev *dev,
994 struct vhost_iotlb_msg *msg)
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400995{
996 int ret = 0;
997
998 vhost_dev_lock_vqs(dev);
999 switch (msg->type) {
1000 case VHOST_IOTLB_UPDATE:
1001 if (!dev->iotlb) {
1002 ret = -EFAULT;
1003 break;
1004 }
1005 if (umem_access_ok(msg->uaddr, msg->size, msg->perm)) {
1006 ret = -EFAULT;
1007 break;
1008 }
Jason Wangf8894912017-02-28 17:56:02 +08001009 vhost_vq_meta_reset(dev);
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001010 if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size,
1011 msg->iova + msg->size - 1,
1012 msg->uaddr, msg->perm)) {
1013 ret = -ENOMEM;
1014 break;
1015 }
1016 vhost_iotlb_notify_vq(dev, msg);
1017 break;
1018 case VHOST_IOTLB_INVALIDATE:
Jason Wangf8894912017-02-28 17:56:02 +08001019 vhost_vq_meta_reset(dev);
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001020 vhost_del_umem_range(dev->iotlb, msg->iova,
1021 msg->iova + msg->size - 1);
1022 break;
1023 default:
1024 ret = -EINVAL;
1025 break;
1026 }
1027
1028 vhost_dev_unlock_vqs(dev);
1029 return ret;
1030}
1031ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1032 struct iov_iter *from)
1033{
1034 struct vhost_msg_node node;
1035 unsigned size = sizeof(struct vhost_msg);
1036 size_t ret;
1037 int err;
1038
1039 if (iov_iter_count(from) < size)
1040 return 0;
1041 ret = copy_from_iter(&node.msg, size, from);
1042 if (ret != size)
1043 goto done;
1044
1045 switch (node.msg.type) {
1046 case VHOST_IOTLB_MSG:
1047 err = vhost_process_iotlb_msg(dev, &node.msg.iotlb);
1048 if (err)
1049 ret = err;
1050 break;
1051 default:
1052 ret = -EINVAL;
1053 break;
1054 }
1055
1056done:
1057 return ret;
1058}
1059EXPORT_SYMBOL(vhost_chr_write_iter);
1060
1061unsigned int vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1062 poll_table *wait)
1063{
1064 unsigned int mask = 0;
1065
1066 poll_wait(file, &dev->wait, wait);
1067
1068 if (!list_empty(&dev->read_list))
1069 mask |= POLLIN | POLLRDNORM;
1070
1071 return mask;
1072}
1073EXPORT_SYMBOL(vhost_chr_poll);
1074
1075ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1076 int noblock)
1077{
1078 DEFINE_WAIT(wait);
1079 struct vhost_msg_node *node;
1080 ssize_t ret = 0;
1081 unsigned size = sizeof(struct vhost_msg);
1082
1083 if (iov_iter_count(to) < size)
1084 return 0;
1085
1086 while (1) {
1087 if (!noblock)
1088 prepare_to_wait(&dev->wait, &wait,
1089 TASK_INTERRUPTIBLE);
1090
1091 node = vhost_dequeue_msg(dev, &dev->read_list);
1092 if (node)
1093 break;
1094 if (noblock) {
1095 ret = -EAGAIN;
1096 break;
1097 }
1098 if (signal_pending(current)) {
1099 ret = -ERESTARTSYS;
1100 break;
1101 }
1102 if (!dev->iotlb) {
1103 ret = -EBADFD;
1104 break;
1105 }
1106
1107 schedule();
1108 }
1109
1110 if (!noblock)
1111 finish_wait(&dev->wait, &wait);
1112
1113 if (node) {
1114 ret = copy_to_iter(&node->msg, size, to);
1115
1116 if (ret != size || node->msg.type != VHOST_IOTLB_MISS) {
1117 kfree(node);
1118 return ret;
1119 }
1120
1121 vhost_enqueue_msg(dev, &dev->pending_list, node);
1122 }
1123
1124 return ret;
1125}
1126EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1127
1128static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1129{
1130 struct vhost_dev *dev = vq->dev;
1131 struct vhost_msg_node *node;
1132 struct vhost_iotlb_msg *msg;
1133
1134 node = vhost_new_msg(vq, VHOST_IOTLB_MISS);
1135 if (!node)
1136 return -ENOMEM;
1137
1138 msg = &node->msg.iotlb;
1139 msg->type = VHOST_IOTLB_MISS;
1140 msg->iova = iova;
1141 msg->perm = access;
1142
1143 vhost_enqueue_msg(dev, &dev->read_list, node);
1144
1145 return 0;
Jason Wangbfe2bc52016-06-23 02:04:30 -04001146}
1147
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001148static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001149 struct vring_desc __user *desc,
1150 struct vring_avail __user *avail,
1151 struct vring_used __user *used)
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001152
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001153{
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001154 size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001155
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001156 return access_ok(VERIFY_READ, desc, num * sizeof *desc) &&
1157 access_ok(VERIFY_READ, avail,
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03001158 sizeof *avail + num * sizeof *avail->ring + s) &&
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001159 access_ok(VERIFY_WRITE, used,
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03001160 sizeof *used + num * sizeof *used->ring + s);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001161}
1162
Jason Wangf8894912017-02-28 17:56:02 +08001163static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1164 const struct vhost_umem_node *node,
1165 int type)
1166{
1167 int access = (type == VHOST_ADDR_USED) ?
1168 VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1169
1170 if (likely(node->perm & access))
1171 vq->meta_iotlb[type] = node;
1172}
1173
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001174static int iotlb_access_ok(struct vhost_virtqueue *vq,
Jason Wangf8894912017-02-28 17:56:02 +08001175 int access, u64 addr, u64 len, int type)
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001176{
1177 const struct vhost_umem_node *node;
1178 struct vhost_umem *umem = vq->iotlb;
Jason Wangf8894912017-02-28 17:56:02 +08001179 u64 s = 0, size, orig_addr = addr;
1180
1181 if (vhost_vq_meta_fetch(vq, addr, len, type))
1182 return true;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001183
1184 while (len > s) {
1185 node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1186 addr,
1187 addr + len - 1);
1188 if (node == NULL || node->start > addr) {
1189 vhost_iotlb_miss(vq, addr, access);
1190 return false;
1191 } else if (!(node->perm & access)) {
1192 /* Report the possible access violation by
1193 * request another translation from userspace.
1194 */
1195 return false;
1196 }
1197
1198 size = node->size - addr + node->start;
Jason Wangf8894912017-02-28 17:56:02 +08001199
1200 if (orig_addr == addr && size >= len)
1201 vhost_vq_meta_update(vq, node, type);
1202
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001203 s += size;
1204 addr += size;
1205 }
1206
1207 return true;
1208}
1209
1210int vq_iotlb_prefetch(struct vhost_virtqueue *vq)
1211{
1212 size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1213 unsigned int num = vq->num;
1214
1215 if (!vq->iotlb)
1216 return 1;
1217
1218 return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc,
Jason Wangf8894912017-02-28 17:56:02 +08001219 num * sizeof(*vq->desc), VHOST_ADDR_DESC) &&
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001220 iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail,
1221 sizeof *vq->avail +
Jason Wangf8894912017-02-28 17:56:02 +08001222 num * sizeof(*vq->avail->ring) + s,
1223 VHOST_ADDR_AVAIL) &&
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001224 iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used,
1225 sizeof *vq->used +
Jason Wangf8894912017-02-28 17:56:02 +08001226 num * sizeof(*vq->used->ring) + s,
1227 VHOST_ADDR_USED);
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001228}
1229EXPORT_SYMBOL_GPL(vq_iotlb_prefetch);
1230
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001231/* Can we log writes? */
1232/* Caller should have device mutex but not vq mutex */
1233int vhost_log_access_ok(struct vhost_dev *dev)
1234{
Jason Wanga9709d62016-06-23 02:04:31 -04001235 return memory_access_ok(dev, dev->umem, 1);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001236}
Asias He6ac1afb2013-05-06 16:38:21 +08001237EXPORT_SYMBOL_GPL(vhost_log_access_ok);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001238
1239/* Verify access for write logging. */
1240/* Caller should have vq mutex and device mutex */
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001241static int vq_log_access_ok(struct vhost_virtqueue *vq,
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03001242 void __user *log_base)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001243{
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001244 size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
Arnd Bergmann28457ee2010-03-09 19:24:45 +01001245
Jason Wanga9709d62016-06-23 02:04:31 -04001246 return vq_memory_access_ok(log_base, vq->umem,
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001247 vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001248 (!vq->log_used || log_access_ok(log_base, vq->log_addr,
1249 sizeof *vq->used +
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03001250 vq->num * sizeof *vq->used->ring + s));
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001251}
1252
1253/* Can we start vq? */
1254/* Caller should have vq mutex and device mutex */
1255int vhost_vq_access_ok(struct vhost_virtqueue *vq)
1256{
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001257 if (vq->iotlb) {
1258 /* When device IOTLB was used, the access validation
1259 * will be validated during prefetching.
1260 */
1261 return 1;
1262 }
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001263 return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used) &&
1264 vq_log_access_ok(vq, vq->log_base);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001265}
Asias He6ac1afb2013-05-06 16:38:21 +08001266EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001267
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001268static struct vhost_umem *vhost_umem_alloc(void)
1269{
Michal Hocko6c5ab652017-05-08 15:57:15 -07001270 struct vhost_umem *umem = kvzalloc(sizeof(*umem), GFP_KERNEL);
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001271
1272 if (!umem)
1273 return NULL;
1274
1275 umem->umem_tree = RB_ROOT;
1276 umem->numem = 0;
1277 INIT_LIST_HEAD(&umem->umem_list);
1278
1279 return umem;
1280}
1281
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001282static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1283{
Jason Wanga9709d62016-06-23 02:04:31 -04001284 struct vhost_memory mem, *newmem;
1285 struct vhost_memory_region *region;
Jason Wanga9709d62016-06-23 02:04:31 -04001286 struct vhost_umem *newumem, *oldumem;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001287 unsigned long size = offsetof(struct vhost_memory, regions);
Michael S. Tsirkin98f9ca02014-05-28 17:07:02 +03001288 int i;
Krishna Kumard47effe2011-03-01 17:06:37 +05301289
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001290 if (copy_from_user(&mem, m, size))
1291 return -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001292 if (mem.padding)
1293 return -EOPNOTSUPP;
Igor Mammedovc9ce42f2015-07-02 15:08:11 +02001294 if (mem.nregions > max_mem_regions)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001295 return -E2BIG;
Michal Hocko6c5ab652017-05-08 15:57:15 -07001296 newmem = kvzalloc(size + mem.nregions * sizeof(*m->regions), GFP_KERNEL);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001297 if (!newmem)
1298 return -ENOMEM;
1299
1300 memcpy(newmem, &mem, size);
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001301 if (copy_from_user(newmem->regions, m->regions,
1302 mem.nregions * sizeof *m->regions)) {
Igor Mammedovbcfeaca2015-06-16 18:33:35 +02001303 kvfree(newmem);
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001304 return -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001305 }
1306
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001307 newumem = vhost_umem_alloc();
Jason Wanga9709d62016-06-23 02:04:31 -04001308 if (!newumem) {
Igor Mammedov4de72552015-07-01 11:07:09 +02001309 kvfree(newmem);
Jason Wanga9709d62016-06-23 02:04:31 -04001310 return -ENOMEM;
Takuya Yoshikawaa02c3782010-05-27 19:03:56 +09001311 }
Jason Wanga9709d62016-06-23 02:04:31 -04001312
Jason Wanga9709d62016-06-23 02:04:31 -04001313 for (region = newmem->regions;
1314 region < newmem->regions + mem.nregions;
1315 region++) {
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001316 if (vhost_new_umem_range(newumem,
1317 region->guest_phys_addr,
1318 region->memory_size,
1319 region->guest_phys_addr +
1320 region->memory_size - 1,
1321 region->userspace_addr,
1322 VHOST_ACCESS_RW))
Jason Wanga9709d62016-06-23 02:04:31 -04001323 goto err;
Jason Wanga9709d62016-06-23 02:04:31 -04001324 }
1325
1326 if (!memory_access_ok(d, newumem, 0))
1327 goto err;
1328
1329 oldumem = d->umem;
1330 d->umem = newumem;
Michael S. Tsirkin98f9ca02014-05-28 17:07:02 +03001331
Michael S. Tsirkin47283be2014-06-05 15:20:27 +03001332 /* All memory accesses are done under some VQ mutex. */
Michael S. Tsirkin98f9ca02014-05-28 17:07:02 +03001333 for (i = 0; i < d->nvqs; ++i) {
1334 mutex_lock(&d->vqs[i]->mutex);
Jason Wanga9709d62016-06-23 02:04:31 -04001335 d->vqs[i]->umem = newumem;
Michael S. Tsirkin98f9ca02014-05-28 17:07:02 +03001336 mutex_unlock(&d->vqs[i]->mutex);
1337 }
Jason Wanga9709d62016-06-23 02:04:31 -04001338
1339 kvfree(newmem);
1340 vhost_umem_clean(oldumem);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001341 return 0;
Jason Wanga9709d62016-06-23 02:04:31 -04001342
1343err:
1344 vhost_umem_clean(newumem);
1345 kvfree(newmem);
1346 return -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001347}
1348
Michael S. Tsirkin935cdee2012-12-06 14:03:34 +02001349long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001350{
Al Virocecb46f2012-08-27 14:21:39 -04001351 struct file *eventfp, *filep = NULL;
1352 bool pollstart = false, pollstop = false;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001353 struct eventfd_ctx *ctx = NULL;
1354 u32 __user *idxp = argp;
1355 struct vhost_virtqueue *vq;
1356 struct vhost_vring_state s;
1357 struct vhost_vring_file f;
1358 struct vhost_vring_addr a;
1359 u32 idx;
1360 long r;
1361
1362 r = get_user(idx, idxp);
1363 if (r < 0)
1364 return r;
Krishna Kumar0f3d9a12010-05-25 11:10:36 +05301365 if (idx >= d->nvqs)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001366 return -ENOBUFS;
1367
Asias He3ab2e422013-04-27 11:16:48 +08001368 vq = d->vqs[idx];
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001369
1370 mutex_lock(&vq->mutex);
1371
1372 switch (ioctl) {
1373 case VHOST_SET_VRING_NUM:
1374 /* Resizing ring with an active backend?
1375 * You don't want to do that. */
1376 if (vq->private_data) {
1377 r = -EBUSY;
1378 break;
1379 }
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001380 if (copy_from_user(&s, argp, sizeof s)) {
1381 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001382 break;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001383 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001384 if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) {
1385 r = -EINVAL;
1386 break;
1387 }
1388 vq->num = s.num;
1389 break;
1390 case VHOST_SET_VRING_BASE:
1391 /* Moving base with an active backend?
1392 * You don't want to do that. */
1393 if (vq->private_data) {
1394 r = -EBUSY;
1395 break;
1396 }
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001397 if (copy_from_user(&s, argp, sizeof s)) {
1398 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001399 break;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001400 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001401 if (s.num > 0xffff) {
1402 r = -EINVAL;
1403 break;
1404 }
Jason Wang809ecb92016-12-12 14:46:49 +08001405 vq->last_avail_idx = vq->last_used_event = s.num;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001406 /* Forget the cached index value. */
1407 vq->avail_idx = vq->last_avail_idx;
1408 break;
1409 case VHOST_GET_VRING_BASE:
1410 s.index = idx;
1411 s.num = vq->last_avail_idx;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001412 if (copy_to_user(argp, &s, sizeof s))
1413 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001414 break;
1415 case VHOST_SET_VRING_ADDR:
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001416 if (copy_from_user(&a, argp, sizeof a)) {
1417 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001418 break;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001419 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001420 if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) {
1421 r = -EOPNOTSUPP;
1422 break;
1423 }
1424 /* For 32bit, verify that the top 32bits of the user
1425 data are set to zero. */
1426 if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
1427 (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
1428 (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) {
1429 r = -EFAULT;
1430 break;
1431 }
Michael S. Tsirkin5d9a07b2014-12-21 01:00:23 +02001432
1433 /* Make sure it's safe to cast pointers to vring types. */
1434 BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
1435 BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
1436 if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
1437 (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
Michael S. Tsirkind5424832015-11-16 16:57:08 +02001438 (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001439 r = -EINVAL;
1440 break;
1441 }
1442
1443 /* We only verify access here if backend is configured.
1444 * If it is not, we don't as size might not have been setup.
1445 * We will verify when backend is configured. */
1446 if (vq->private_data) {
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001447 if (!vq_access_ok(vq, vq->num,
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001448 (void __user *)(unsigned long)a.desc_user_addr,
1449 (void __user *)(unsigned long)a.avail_user_addr,
1450 (void __user *)(unsigned long)a.used_user_addr)) {
1451 r = -EINVAL;
1452 break;
1453 }
1454
1455 /* Also validate log access for used ring if enabled. */
1456 if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) &&
1457 !log_access_ok(vq->log_base, a.log_guest_addr,
1458 sizeof *vq->used +
1459 vq->num * sizeof *vq->used->ring)) {
1460 r = -EINVAL;
1461 break;
1462 }
1463 }
1464
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001465 vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
1466 vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
1467 vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
1468 vq->log_addr = a.log_guest_addr;
1469 vq->used = (void __user *)(unsigned long)a.used_user_addr;
1470 break;
1471 case VHOST_SET_VRING_KICK:
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001472 if (copy_from_user(&f, argp, sizeof f)) {
1473 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001474 break;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001475 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001476 eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
Michael S. Tsirkin535297a2010-03-17 16:06:11 +02001477 if (IS_ERR(eventfp)) {
1478 r = PTR_ERR(eventfp);
1479 break;
1480 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001481 if (eventfp != vq->kick) {
Al Virocecb46f2012-08-27 14:21:39 -04001482 pollstop = (filep = vq->kick) != NULL;
1483 pollstart = (vq->kick = eventfp) != NULL;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001484 } else
1485 filep = eventfp;
1486 break;
1487 case VHOST_SET_VRING_CALL:
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001488 if (copy_from_user(&f, argp, sizeof f)) {
1489 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001490 break;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001491 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001492 eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
Michael S. Tsirkin535297a2010-03-17 16:06:11 +02001493 if (IS_ERR(eventfp)) {
1494 r = PTR_ERR(eventfp);
1495 break;
1496 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001497 if (eventfp != vq->call) {
1498 filep = vq->call;
1499 ctx = vq->call_ctx;
1500 vq->call = eventfp;
1501 vq->call_ctx = eventfp ?
1502 eventfd_ctx_fileget(eventfp) : NULL;
1503 } else
1504 filep = eventfp;
1505 break;
1506 case VHOST_SET_VRING_ERR:
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001507 if (copy_from_user(&f, argp, sizeof f)) {
1508 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001509 break;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001510 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001511 eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
Michael S. Tsirkin535297a2010-03-17 16:06:11 +02001512 if (IS_ERR(eventfp)) {
1513 r = PTR_ERR(eventfp);
1514 break;
1515 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001516 if (eventfp != vq->error) {
1517 filep = vq->error;
1518 vq->error = eventfp;
1519 ctx = vq->error_ctx;
1520 vq->error_ctx = eventfp ?
1521 eventfd_ctx_fileget(eventfp) : NULL;
1522 } else
1523 filep = eventfp;
1524 break;
Greg Kurz2751c982015-04-24 14:27:24 +02001525 case VHOST_SET_VRING_ENDIAN:
1526 r = vhost_set_vring_endian(vq, argp);
1527 break;
1528 case VHOST_GET_VRING_ENDIAN:
1529 r = vhost_get_vring_endian(vq, idx, argp);
1530 break;
Jason Wang03088132016-03-04 06:24:53 -05001531 case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
1532 if (copy_from_user(&s, argp, sizeof(s))) {
1533 r = -EFAULT;
1534 break;
1535 }
1536 vq->busyloop_timeout = s.num;
1537 break;
1538 case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
1539 s.index = idx;
1540 s.num = vq->busyloop_timeout;
1541 if (copy_to_user(argp, &s, sizeof(s)))
1542 r = -EFAULT;
1543 break;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001544 default:
1545 r = -ENOIOCTLCMD;
1546 }
1547
1548 if (pollstop && vq->handle_kick)
1549 vhost_poll_stop(&vq->poll);
1550
1551 if (ctx)
1552 eventfd_ctx_put(ctx);
1553 if (filep)
1554 fput(filep);
1555
1556 if (pollstart && vq->handle_kick)
Jason Wang2b8b3282013-01-28 01:05:18 +00001557 r = vhost_poll_start(&vq->poll, vq->kick);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001558
1559 mutex_unlock(&vq->mutex);
1560
1561 if (pollstop && vq->handle_kick)
1562 vhost_poll_flush(&vq->poll);
1563 return r;
1564}
Asias He6ac1afb2013-05-06 16:38:21 +08001565EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001566
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001567int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled)
1568{
1569 struct vhost_umem *niotlb, *oiotlb;
1570 int i;
1571
1572 niotlb = vhost_umem_alloc();
1573 if (!niotlb)
1574 return -ENOMEM;
1575
1576 oiotlb = d->iotlb;
1577 d->iotlb = niotlb;
1578
1579 for (i = 0; i < d->nvqs; ++i) {
1580 mutex_lock(&d->vqs[i]->mutex);
1581 d->vqs[i]->iotlb = niotlb;
1582 mutex_unlock(&d->vqs[i]->mutex);
1583 }
1584
1585 vhost_umem_clean(oiotlb);
1586
1587 return 0;
1588}
1589EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
1590
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001591/* Caller must have device mutex */
Michael S. Tsirkin935cdee2012-12-06 14:03:34 +02001592long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001593{
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001594 struct file *eventfp, *filep = NULL;
1595 struct eventfd_ctx *ctx = NULL;
1596 u64 p;
1597 long r;
1598 int i, fd;
1599
1600 /* If you are not the owner, you can become one */
1601 if (ioctl == VHOST_SET_OWNER) {
1602 r = vhost_dev_set_owner(d);
1603 goto done;
1604 }
1605
1606 /* You must be the owner to do anything else */
1607 r = vhost_dev_check_owner(d);
1608 if (r)
1609 goto done;
1610
1611 switch (ioctl) {
1612 case VHOST_SET_MEM_TABLE:
1613 r = vhost_set_memory(d, argp);
1614 break;
1615 case VHOST_SET_LOG_BASE:
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001616 if (copy_from_user(&p, argp, sizeof p)) {
1617 r = -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001618 break;
Takuya Yoshikawa7ad9c9d2010-05-27 18:58:03 +09001619 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001620 if ((u64)(unsigned long)p != p) {
1621 r = -EFAULT;
1622 break;
1623 }
1624 for (i = 0; i < d->nvqs; ++i) {
1625 struct vhost_virtqueue *vq;
1626 void __user *base = (void __user *)(unsigned long)p;
Asias He3ab2e422013-04-27 11:16:48 +08001627 vq = d->vqs[i];
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001628 mutex_lock(&vq->mutex);
1629 /* If ring is inactive, will check when it's enabled. */
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001630 if (vq->private_data && !vq_log_access_ok(vq, base))
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001631 r = -EFAULT;
1632 else
1633 vq->log_base = base;
1634 mutex_unlock(&vq->mutex);
1635 }
1636 break;
1637 case VHOST_SET_LOG_FD:
1638 r = get_user(fd, (int __user *)argp);
1639 if (r < 0)
1640 break;
1641 eventfp = fd == -1 ? NULL : eventfd_fget(fd);
1642 if (IS_ERR(eventfp)) {
1643 r = PTR_ERR(eventfp);
1644 break;
1645 }
1646 if (eventfp != d->log_file) {
1647 filep = d->log_file;
Marc-André Lureau7932c0b2015-07-17 15:32:03 +02001648 d->log_file = eventfp;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001649 ctx = d->log_ctx;
1650 d->log_ctx = eventfp ?
1651 eventfd_ctx_fileget(eventfp) : NULL;
1652 } else
1653 filep = eventfp;
1654 for (i = 0; i < d->nvqs; ++i) {
Asias He3ab2e422013-04-27 11:16:48 +08001655 mutex_lock(&d->vqs[i]->mutex);
1656 d->vqs[i]->log_ctx = d->log_ctx;
1657 mutex_unlock(&d->vqs[i]->mutex);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001658 }
1659 if (ctx)
1660 eventfd_ctx_put(ctx);
1661 if (filep)
1662 fput(filep);
1663 break;
1664 default:
Michael S. Tsirkin935cdee2012-12-06 14:03:34 +02001665 r = -ENOIOCTLCMD;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001666 break;
1667 }
1668done:
1669 return r;
1670}
Asias He6ac1afb2013-05-06 16:38:21 +08001671EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001672
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001673/* TODO: This is really inefficient. We need something like get_user()
1674 * (instruction directly accesses the data, with an exception table entry
1675 * returning -EFAULT). See Documentation/x86/exception-tables.txt.
1676 */
1677static int set_bit_to_user(int nr, void __user *addr)
1678{
1679 unsigned long log = (unsigned long)addr;
1680 struct page *page;
1681 void *base;
1682 int bit = nr + (log % PAGE_SIZE) * 8;
1683 int r;
Krishna Kumard47effe2011-03-01 17:06:37 +05301684
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001685 r = get_user_pages_fast(log, 1, 1, &page);
Michael S. Tsirkind6db3f52010-02-23 11:25:23 +02001686 if (r < 0)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001687 return r;
Michael S. Tsirkind6db3f52010-02-23 11:25:23 +02001688 BUG_ON(r != 1);
Cong Wangc6daa7f2011-11-25 23:14:26 +08001689 base = kmap_atomic(page);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001690 set_bit(bit, base);
Cong Wangc6daa7f2011-11-25 23:14:26 +08001691 kunmap_atomic(base);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001692 set_page_dirty_lock(page);
1693 put_page(page);
1694 return 0;
1695}
1696
1697static int log_write(void __user *log_base,
1698 u64 write_address, u64 write_length)
1699{
Michael S. Tsirkin28831ee2010-11-29 10:22:10 +02001700 u64 write_page = write_address / VHOST_PAGE_SIZE;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001701 int r;
Krishna Kumard47effe2011-03-01 17:06:37 +05301702
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001703 if (!write_length)
1704 return 0;
Michael S. Tsirkin3bf9be42010-11-29 10:19:07 +02001705 write_length += write_address % VHOST_PAGE_SIZE;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001706 for (;;) {
1707 u64 base = (u64)(unsigned long)log_base;
Michael S. Tsirkin28831ee2010-11-29 10:22:10 +02001708 u64 log = base + write_page / 8;
1709 int bit = write_page % 8;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001710 if ((u64)(unsigned long)log != log)
1711 return -EFAULT;
1712 r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
1713 if (r < 0)
1714 return r;
1715 if (write_length <= VHOST_PAGE_SIZE)
1716 break;
1717 write_length -= VHOST_PAGE_SIZE;
Michael S. Tsirkin28831ee2010-11-29 10:22:10 +02001718 write_page += 1;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001719 }
1720 return r;
1721}
1722
1723int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
1724 unsigned int log_num, u64 len)
1725{
1726 int i, r;
1727
1728 /* Make sure data written is seen before log. */
Michael S. Tsirkin56593382010-02-01 07:21:02 +00001729 smp_wmb();
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001730 for (i = 0; i < log_num; ++i) {
1731 u64 l = min(log[i].len, len);
1732 r = log_write(vq->log_base, log[i].addr, l);
1733 if (r < 0)
1734 return r;
1735 len -= l;
Michael S. Tsirkin5786aee2010-09-22 12:31:53 +02001736 if (!len) {
1737 if (vq->log_ctx)
1738 eventfd_signal(vq->log_ctx, 1);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001739 return 0;
Michael S. Tsirkin5786aee2010-09-22 12:31:53 +02001740 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001741 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001742 /* Length written exceeds what we have stored. This is a bug. */
1743 BUG();
1744 return 0;
1745}
Asias He6ac1afb2013-05-06 16:38:21 +08001746EXPORT_SYMBOL_GPL(vhost_log_write);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001747
Jason Wang2723fea2011-06-21 18:04:38 +08001748static int vhost_update_used_flags(struct vhost_virtqueue *vq)
1749{
1750 void __user *used;
Jason Wangbfe2bc52016-06-23 02:04:30 -04001751 if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
1752 &vq->used->flags) < 0)
Jason Wang2723fea2011-06-21 18:04:38 +08001753 return -EFAULT;
1754 if (unlikely(vq->log_used)) {
1755 /* Make sure the flag is seen before log. */
1756 smp_wmb();
1757 /* Log used flag write. */
1758 used = &vq->used->flags;
1759 log_write(vq->log_base, vq->log_addr +
1760 (used - (void __user *)vq->used),
1761 sizeof vq->used->flags);
1762 if (vq->log_ctx)
1763 eventfd_signal(vq->log_ctx, 1);
1764 }
1765 return 0;
1766}
1767
1768static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event)
1769{
Jason Wangbfe2bc52016-06-23 02:04:30 -04001770 if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
1771 vhost_avail_event(vq)))
Jason Wang2723fea2011-06-21 18:04:38 +08001772 return -EFAULT;
1773 if (unlikely(vq->log_used)) {
1774 void __user *used;
1775 /* Make sure the event is seen before log. */
1776 smp_wmb();
1777 /* Log avail event write */
1778 used = vhost_avail_event(vq);
1779 log_write(vq->log_base, vq->log_addr +
1780 (used - (void __user *)vq->used),
1781 sizeof *vhost_avail_event(vq));
1782 if (vq->log_ctx)
1783 eventfd_signal(vq->log_ctx, 1);
1784 }
1785 return 0;
1786}
1787
Greg Kurz80f7d032016-02-16 15:59:44 +01001788int vhost_vq_init_access(struct vhost_virtqueue *vq)
Jason Wang2723fea2011-06-21 18:04:38 +08001789{
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001790 __virtio16 last_used_idx;
Jason Wang2723fea2011-06-21 18:04:38 +08001791 int r;
Greg Kurze1f33be2016-02-16 15:54:28 +01001792 bool is_le = vq->is_le;
1793
Halil Pasiccda8bba2017-01-30 11:09:36 +01001794 if (!vq->private_data)
Jason Wang2723fea2011-06-21 18:04:38 +08001795 return 0;
Greg Kurz2751c982015-04-24 14:27:24 +02001796
1797 vhost_init_is_le(vq);
Jason Wang2723fea2011-06-21 18:04:38 +08001798
1799 r = vhost_update_used_flags(vq);
1800 if (r)
Greg Kurze1f33be2016-02-16 15:54:28 +01001801 goto err;
Jason Wang2723fea2011-06-21 18:04:38 +08001802 vq->signalled_used_valid = false;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001803 if (!vq->iotlb &&
1804 !access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) {
Greg Kurze1f33be2016-02-16 15:54:28 +01001805 r = -EFAULT;
1806 goto err;
1807 }
Jason Wangf8894912017-02-28 17:56:02 +08001808 r = vhost_get_used(vq, last_used_idx, &vq->used->idx);
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001809 if (r) {
1810 vq_err(vq, "Can't access used idx at %p\n",
1811 &vq->used->idx);
Greg Kurze1f33be2016-02-16 15:54:28 +01001812 goto err;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001813 }
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001814 vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
Michael S. Tsirkin64f7f052014-12-01 17:39:39 +02001815 return 0;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001816
Greg Kurze1f33be2016-02-16 15:54:28 +01001817err:
1818 vq->is_le = is_le;
1819 return r;
Jason Wang2723fea2011-06-21 18:04:38 +08001820}
Greg Kurz80f7d032016-02-16 15:59:44 +01001821EXPORT_SYMBOL_GPL(vhost_vq_init_access);
Jason Wang2723fea2011-06-21 18:04:38 +08001822
Michael S. Tsirkin47283be2014-06-05 15:20:27 +03001823static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001824 struct iovec iov[], int iov_size, int access)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001825{
Jason Wanga9709d62016-06-23 02:04:31 -04001826 const struct vhost_umem_node *node;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001827 struct vhost_dev *dev = vq->dev;
1828 struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001829 struct iovec *_iov;
1830 u64 s = 0;
1831 int ret = 0;
1832
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001833 while ((u64)len > s) {
1834 u64 size;
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03001835 if (unlikely(ret >= iov_size)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001836 ret = -ENOBUFS;
1837 break;
1838 }
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001839
Jason Wanga9709d62016-06-23 02:04:31 -04001840 node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1841 addr, addr + len - 1);
1842 if (node == NULL || node->start > addr) {
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001843 if (umem != dev->iotlb) {
1844 ret = -EFAULT;
1845 break;
1846 }
1847 ret = -EAGAIN;
1848 break;
1849 } else if (!(node->perm & access)) {
1850 ret = -EPERM;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001851 break;
1852 }
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001853
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001854 _iov = iov + ret;
Jason Wanga9709d62016-06-23 02:04:31 -04001855 size = node->size - addr + node->start;
Michael S. Tsirkinbd971202012-11-26 05:57:27 +00001856 _iov->iov_len = min((u64)len - s, size);
Christoph Hellwiga8d37822010-04-13 14:11:25 -04001857 _iov->iov_base = (void __user *)(unsigned long)
Jason Wanga9709d62016-06-23 02:04:31 -04001858 (node->userspace_addr + addr - node->start);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001859 s += size;
1860 addr += size;
1861 ++ret;
1862 }
1863
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001864 if (ret == -EAGAIN)
1865 vhost_iotlb_miss(vq, addr, access);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001866 return ret;
1867}
1868
1869/* Each buffer in the virtqueues is actually a chain of descriptors. This
1870 * function returns the next descriptor in the chain,
1871 * or -1U if we're at the end. */
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001872static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001873{
1874 unsigned int next;
1875
1876 /* If this descriptor says it doesn't chain, we're done. */
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001877 if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001878 return -1U;
1879
1880 /* Check they're not leading us off end of descriptors. */
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001881 next = vhost16_to_cpu(vq, desc->next);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001882 /* Make sure compiler knows to grab that: we don't want it changing! */
1883 /* We will use the result as an index in an array, so most
1884 * architectures only need a compiler barrier here. */
1885 read_barrier_depends();
1886
1887 return next;
1888}
1889
Michael S. Tsirkin47283be2014-06-05 15:20:27 +03001890static int get_indirect(struct vhost_virtqueue *vq,
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03001891 struct iovec iov[], unsigned int iov_size,
1892 unsigned int *out_num, unsigned int *in_num,
1893 struct vhost_log *log, unsigned int *log_num,
1894 struct vring_desc *indirect)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001895{
1896 struct vring_desc desc;
1897 unsigned int i = 0, count, found = 0;
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001898 u32 len = vhost32_to_cpu(vq, indirect->len);
Al Viroaad9a1c2014-12-10 14:49:01 -05001899 struct iov_iter from;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001900 int ret, access;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001901
1902 /* Sanity check */
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001903 if (unlikely(len % sizeof desc)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001904 vq_err(vq, "Invalid length in indirect descriptor: "
1905 "len 0x%llx not multiple of 0x%zx\n",
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001906 (unsigned long long)len,
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001907 sizeof desc);
1908 return -EINVAL;
1909 }
1910
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001911 ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001912 UIO_MAXIOV, VHOST_ACCESS_RO);
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03001913 if (unlikely(ret < 0)) {
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001914 if (ret != -EAGAIN)
1915 vq_err(vq, "Translation failure %d in indirect.\n", ret);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001916 return ret;
1917 }
Al Viroaad9a1c2014-12-10 14:49:01 -05001918 iov_iter_init(&from, READ, vq->indirect, ret, len);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001919
1920 /* We will use the result as an address to read from, so most
1921 * architectures only need a compiler barrier here. */
1922 read_barrier_depends();
1923
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001924 count = len / sizeof desc;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001925 /* Buffers are chained via a 16 bit next field, so
1926 * we can have at most 2^16 of these. */
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03001927 if (unlikely(count > USHRT_MAX + 1)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001928 vq_err(vq, "Indirect buffer length too big: %d\n",
1929 indirect->len);
1930 return -E2BIG;
1931 }
1932
1933 do {
1934 unsigned iov_count = *in_num + *out_num;
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03001935 if (unlikely(++found > count)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001936 vq_err(vq, "Loop detected: last one at %u "
1937 "indirect size %u\n",
1938 i, count);
1939 return -EINVAL;
1940 }
Al Virocbbd26b2016-11-01 22:09:04 -04001941 if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001942 vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001943 i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001944 return -EINVAL;
1945 }
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001946 if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001947 vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001948 i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001949 return -EINVAL;
1950 }
1951
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001952 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
1953 access = VHOST_ACCESS_WO;
1954 else
1955 access = VHOST_ACCESS_RO;
1956
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001957 ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
1958 vhost32_to_cpu(vq, desc.len), iov + iov_count,
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001959 iov_size - iov_count, access);
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03001960 if (unlikely(ret < 0)) {
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001961 if (ret != -EAGAIN)
1962 vq_err(vq, "Translation failure %d indirect idx %d\n",
1963 ret, i);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001964 return ret;
1965 }
1966 /* If this is an input descriptor, increment that count. */
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001967 if (access == VHOST_ACCESS_WO) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001968 *in_num += ret;
1969 if (unlikely(log)) {
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001970 log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
1971 log[*log_num].len = vhost32_to_cpu(vq, desc.len);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001972 ++*log_num;
1973 }
1974 } else {
1975 /* If it's an output descriptor, they're all supposed
1976 * to come before any input descriptors. */
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03001977 if (unlikely(*in_num)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001978 vq_err(vq, "Indirect descriptor "
1979 "has out after in: idx %d\n", i);
1980 return -EINVAL;
1981 }
1982 *out_num += ret;
1983 }
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03001984 } while ((i = next_desc(vq, &desc)) != -1);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00001985 return 0;
1986}
1987
1988/* This looks in the virtqueue and for the first available buffer, and converts
1989 * it to an iovec for convenient access. Since descriptors consist of some
1990 * number of output then some number of input descriptors, it's actually two
1991 * iovecs, but we pack them into one and note how many of each there were.
1992 *
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03001993 * This function returns the descriptor number found, or vq->num (which is
1994 * never a valid descriptor number) if none was found. A negative code is
1995 * returned on error. */
Michael S. Tsirkin47283be2014-06-05 15:20:27 +03001996int vhost_get_vq_desc(struct vhost_virtqueue *vq,
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03001997 struct iovec iov[], unsigned int iov_size,
1998 unsigned int *out_num, unsigned int *in_num,
1999 struct vhost_log *log, unsigned int *log_num)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002000{
2001 struct vring_desc desc;
2002 unsigned int i, head, found = 0;
2003 u16 last_avail_idx;
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002004 __virtio16 avail_idx;
2005 __virtio16 ring_head;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04002006 int ret, access;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002007
2008 /* Check it isn't doing very strange things with descriptor numbers. */
2009 last_avail_idx = vq->last_avail_idx;
Jason Wange3b56cd2017-02-07 15:49:50 +08002010
2011 if (vq->avail_idx == vq->last_avail_idx) {
Jason Wangf8894912017-02-28 17:56:02 +08002012 if (unlikely(vhost_get_avail(vq, avail_idx, &vq->avail->idx))) {
Jason Wange3b56cd2017-02-07 15:49:50 +08002013 vq_err(vq, "Failed to access avail idx at %p\n",
2014 &vq->avail->idx);
2015 return -EFAULT;
2016 }
2017 vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2018
2019 if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) {
2020 vq_err(vq, "Guest moved used index from %u to %u",
2021 last_avail_idx, vq->avail_idx);
2022 return -EFAULT;
2023 }
2024
2025 /* If there's nothing new since last we looked, return
2026 * invalid.
2027 */
2028 if (vq->avail_idx == last_avail_idx)
2029 return vq->num;
2030
2031 /* Only get avail ring entries after they have been
2032 * exposed by guest.
2033 */
2034 smp_rmb();
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002035 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002036
2037 /* Grab the next descriptor number they're advertising, and increment
2038 * the index we've seen. */
Jason Wangf8894912017-02-28 17:56:02 +08002039 if (unlikely(vhost_get_avail(vq, ring_head,
Jason Wangbfe2bc52016-06-23 02:04:30 -04002040 &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002041 vq_err(vq, "Failed to read head: idx %d address %p\n",
2042 last_avail_idx,
2043 &vq->avail->ring[last_avail_idx % vq->num]);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002044 return -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002045 }
2046
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002047 head = vhost16_to_cpu(vq, ring_head);
2048
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002049 /* If their number is silly, that's an error. */
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03002050 if (unlikely(head >= vq->num)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002051 vq_err(vq, "Guest says index %u > %u is available",
2052 head, vq->num);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002053 return -EINVAL;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002054 }
2055
2056 /* When we start there are none of either input nor output. */
2057 *out_num = *in_num = 0;
2058 if (unlikely(log))
2059 *log_num = 0;
2060
2061 i = head;
2062 do {
2063 unsigned iov_count = *in_num + *out_num;
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03002064 if (unlikely(i >= vq->num)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002065 vq_err(vq, "Desc index is %u > %u, head = %u",
2066 i, vq->num, head);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002067 return -EINVAL;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002068 }
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03002069 if (unlikely(++found > vq->num)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002070 vq_err(vq, "Loop detected: last one at %u "
2071 "vq size %u head %u\n",
2072 i, vq->num, head);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002073 return -EINVAL;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002074 }
Jason Wangbfe2bc52016-06-23 02:04:30 -04002075 ret = vhost_copy_from_user(vq, &desc, vq->desc + i,
2076 sizeof desc);
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03002077 if (unlikely(ret)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002078 vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2079 i, vq->desc + i);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002080 return -EFAULT;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002081 }
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002082 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
Michael S. Tsirkin47283be2014-06-05 15:20:27 +03002083 ret = get_indirect(vq, iov, iov_size,
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002084 out_num, in_num,
2085 log, log_num, &desc);
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03002086 if (unlikely(ret < 0)) {
Jason Wang6b1e6cc2016-06-23 02:04:32 -04002087 if (ret != -EAGAIN)
2088 vq_err(vq, "Failure detected "
2089 "in indirect descriptor at idx %d\n", i);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002090 return ret;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002091 }
2092 continue;
2093 }
2094
Jason Wang6b1e6cc2016-06-23 02:04:32 -04002095 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2096 access = VHOST_ACCESS_WO;
2097 else
2098 access = VHOST_ACCESS_RO;
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002099 ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2100 vhost32_to_cpu(vq, desc.len), iov + iov_count,
Jason Wang6b1e6cc2016-06-23 02:04:32 -04002101 iov_size - iov_count, access);
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03002102 if (unlikely(ret < 0)) {
Jason Wang6b1e6cc2016-06-23 02:04:32 -04002103 if (ret != -EAGAIN)
2104 vq_err(vq, "Translation failure %d descriptor idx %d\n",
2105 ret, i);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002106 return ret;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002107 }
Jason Wang6b1e6cc2016-06-23 02:04:32 -04002108 if (access == VHOST_ACCESS_WO) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002109 /* If this is an input descriptor,
2110 * increment that count. */
2111 *in_num += ret;
2112 if (unlikely(log)) {
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002113 log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2114 log[*log_num].len = vhost32_to_cpu(vq, desc.len);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002115 ++*log_num;
2116 }
2117 } else {
2118 /* If it's an output descriptor, they're all supposed
2119 * to come before any input descriptors. */
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +03002120 if (unlikely(*in_num)) {
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002121 vq_err(vq, "Descriptor has out after in: "
2122 "idx %d\n", i);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +03002123 return -EINVAL;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002124 }
2125 *out_num += ret;
2126 }
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002127 } while ((i = next_desc(vq, &desc)) != -1);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002128
2129 /* On success, increment avail index. */
2130 vq->last_avail_idx++;
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002131
2132 /* Assume notifications from guest are disabled at this point,
2133 * if they aren't we would need to update avail_event index. */
2134 BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002135 return head;
2136}
Asias He6ac1afb2013-05-06 16:38:21 +08002137EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002138
2139/* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
David Stevens8dd014a2010-07-27 18:52:21 +03002140void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002141{
David Stevens8dd014a2010-07-27 18:52:21 +03002142 vq->last_avail_idx -= n;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002143}
Asias He6ac1afb2013-05-06 16:38:21 +08002144EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002145
2146/* After we've used one of their buffers, we tell them about it. We'll then
2147 * want to notify the guest, using eventfd. */
2148int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2149{
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002150 struct vring_used_elem heads = {
2151 cpu_to_vhost32(vq, head),
2152 cpu_to_vhost32(vq, len)
2153 };
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002154
Jason Wangc49e4e52013-09-02 16:40:58 +08002155 return vhost_add_used_n(vq, &heads, 1);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002156}
Asias He6ac1afb2013-05-06 16:38:21 +08002157EXPORT_SYMBOL_GPL(vhost_add_used);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002158
David Stevens8dd014a2010-07-27 18:52:21 +03002159static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2160 struct vring_used_elem *heads,
2161 unsigned count)
2162{
2163 struct vring_used_elem __user *used;
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002164 u16 old, new;
David Stevens8dd014a2010-07-27 18:52:21 +03002165 int start;
2166
Michael S. Tsirkin5fba13b2015-11-29 13:34:44 +02002167 start = vq->last_used_idx & (vq->num - 1);
David Stevens8dd014a2010-07-27 18:52:21 +03002168 used = vq->used->ring + start;
Jason Wangc49e4e52013-09-02 16:40:58 +08002169 if (count == 1) {
Jason Wangbfe2bc52016-06-23 02:04:30 -04002170 if (vhost_put_user(vq, heads[0].id, &used->id)) {
Jason Wangc49e4e52013-09-02 16:40:58 +08002171 vq_err(vq, "Failed to write used id");
2172 return -EFAULT;
2173 }
Jason Wangbfe2bc52016-06-23 02:04:30 -04002174 if (vhost_put_user(vq, heads[0].len, &used->len)) {
Jason Wangc49e4e52013-09-02 16:40:58 +08002175 vq_err(vq, "Failed to write used len");
2176 return -EFAULT;
2177 }
Jason Wangbfe2bc52016-06-23 02:04:30 -04002178 } else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) {
David Stevens8dd014a2010-07-27 18:52:21 +03002179 vq_err(vq, "Failed to write used");
2180 return -EFAULT;
2181 }
2182 if (unlikely(vq->log_used)) {
2183 /* Make sure data is seen before log. */
2184 smp_wmb();
2185 /* Log used ring entry write. */
2186 log_write(vq->log_base,
2187 vq->log_addr +
2188 ((void __user *)used - (void __user *)vq->used),
2189 count * sizeof *used);
2190 }
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002191 old = vq->last_used_idx;
2192 new = (vq->last_used_idx += count);
2193 /* If the driver never bothers to signal in a very long while,
2194 * used index might wrap around. If that happens, invalidate
2195 * signalled_used index we stored. TODO: make sure driver
2196 * signals at least once in 2^16 and remove this. */
2197 if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2198 vq->signalled_used_valid = false;
David Stevens8dd014a2010-07-27 18:52:21 +03002199 return 0;
2200}
2201
2202/* After we've used one of their buffers, we tell them about it. We'll then
2203 * want to notify the guest, using eventfd. */
2204int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2205 unsigned count)
2206{
2207 int start, n, r;
2208
Michael S. Tsirkin5fba13b2015-11-29 13:34:44 +02002209 start = vq->last_used_idx & (vq->num - 1);
David Stevens8dd014a2010-07-27 18:52:21 +03002210 n = vq->num - start;
2211 if (n < count) {
2212 r = __vhost_add_used_n(vq, heads, n);
2213 if (r < 0)
2214 return r;
2215 heads += n;
2216 count -= n;
2217 }
2218 r = __vhost_add_used_n(vq, heads, count);
2219
2220 /* Make sure buffer is written before we update index. */
2221 smp_wmb();
Jason Wangbfe2bc52016-06-23 02:04:30 -04002222 if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
2223 &vq->used->idx)) {
David Stevens8dd014a2010-07-27 18:52:21 +03002224 vq_err(vq, "Failed to increment used idx");
2225 return -EFAULT;
2226 }
2227 if (unlikely(vq->log_used)) {
2228 /* Log used index update. */
2229 log_write(vq->log_base,
2230 vq->log_addr + offsetof(struct vring_used, idx),
2231 sizeof vq->used->idx);
2232 if (vq->log_ctx)
2233 eventfd_signal(vq->log_ctx, 1);
2234 }
2235 return r;
2236}
Asias He6ac1afb2013-05-06 16:38:21 +08002237EXPORT_SYMBOL_GPL(vhost_add_used_n);
David Stevens8dd014a2010-07-27 18:52:21 +03002238
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002239static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002240{
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002241 __u16 old, new;
2242 __virtio16 event;
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002243 bool v;
Michael S. Tsirkin0d499352010-05-11 19:44:17 +03002244
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03002245 if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002246 unlikely(vq->avail_idx == vq->last_avail_idx))
2247 return true;
2248
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03002249 if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002250 __virtio16 flags;
Jason Wang809ecb92016-12-12 14:46:49 +08002251 /* Flush out used index updates. This is paired
2252 * with the barrier that the Guest executes when enabling
2253 * interrupts. */
2254 smp_mb();
Jason Wangf8894912017-02-28 17:56:02 +08002255 if (vhost_get_avail(vq, flags, &vq->avail->flags)) {
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002256 vq_err(vq, "Failed to get flags");
2257 return true;
2258 }
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002259 return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002260 }
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002261 old = vq->signalled_used;
2262 v = vq->signalled_used_valid;
2263 new = vq->signalled_used = vq->last_used_idx;
2264 vq->signalled_used_valid = true;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002265
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002266 if (unlikely(!v))
2267 return true;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002268
Jason Wang809ecb92016-12-12 14:46:49 +08002269 /* We're sure if the following conditions are met, there's no
2270 * need to notify guest:
2271 * 1) cached used event is ahead of new
2272 * 2) old to new updating does not cross cached used event. */
2273 if (vring_need_event(vq->last_used_event, new + vq->num, new) &&
2274 !vring_need_event(vq->last_used_event, new, old))
2275 return false;
2276
2277 /* Flush out used index updates. This is paired
2278 * with the barrier that the Guest executes when enabling
2279 * interrupts. */
2280 smp_mb();
2281
Jason Wangf8894912017-02-28 17:56:02 +08002282 if (vhost_get_avail(vq, event, vhost_used_event(vq))) {
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002283 vq_err(vq, "Failed to get used event idx");
2284 return true;
2285 }
Jason Wang809ecb92016-12-12 14:46:49 +08002286 vq->last_used_event = vhost16_to_cpu(vq, event);
2287
2288 return vring_need_event(vq->last_used_event, new, old);
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002289}
2290
2291/* This actually signals the guest, using eventfd. */
2292void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2293{
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002294 /* Signal the Guest tell them we used something up. */
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002295 if (vq->call_ctx && vhost_notify(dev, vq))
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002296 eventfd_signal(vq->call_ctx, 1);
2297}
Asias He6ac1afb2013-05-06 16:38:21 +08002298EXPORT_SYMBOL_GPL(vhost_signal);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002299
2300/* And here's the combo meal deal. Supersize me! */
2301void vhost_add_used_and_signal(struct vhost_dev *dev,
2302 struct vhost_virtqueue *vq,
2303 unsigned int head, int len)
2304{
2305 vhost_add_used(vq, head, len);
2306 vhost_signal(dev, vq);
2307}
Asias He6ac1afb2013-05-06 16:38:21 +08002308EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002309
David Stevens8dd014a2010-07-27 18:52:21 +03002310/* multi-buffer version of vhost_add_used_and_signal */
2311void vhost_add_used_and_signal_n(struct vhost_dev *dev,
2312 struct vhost_virtqueue *vq,
2313 struct vring_used_elem *heads, unsigned count)
2314{
2315 vhost_add_used_n(vq, heads, count);
2316 vhost_signal(dev, vq);
2317}
Asias He6ac1afb2013-05-06 16:38:21 +08002318EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
David Stevens8dd014a2010-07-27 18:52:21 +03002319
Jason Wangd4a60602016-03-04 06:24:52 -05002320/* return true if we're sure that avaiable ring is empty */
2321bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2322{
2323 __virtio16 avail_idx;
2324 int r;
2325
Jason Wang275bf962017-01-18 15:02:01 +08002326 if (vq->avail_idx != vq->last_avail_idx)
Jason Wangd4a60602016-03-04 06:24:52 -05002327 return false;
2328
Linus Torvalds54d79892017-03-02 13:53:13 -08002329 r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
Jason Wang275bf962017-01-18 15:02:01 +08002330 if (unlikely(r))
2331 return false;
2332 vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2333
2334 return vq->avail_idx == vq->last_avail_idx;
Jason Wangd4a60602016-03-04 06:24:52 -05002335}
2336EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
2337
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002338/* OK, now we need to know about added descriptors. */
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002339bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002340{
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002341 __virtio16 avail_idx;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002342 int r;
Krishna Kumard47effe2011-03-01 17:06:37 +05302343
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002344 if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
2345 return false;
2346 vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03002347 if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
Jason Wang2723fea2011-06-21 18:04:38 +08002348 r = vhost_update_used_flags(vq);
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002349 if (r) {
2350 vq_err(vq, "Failed to enable notification at %p: %d\n",
2351 &vq->used->flags, r);
2352 return false;
2353 }
2354 } else {
Jason Wang2723fea2011-06-21 18:04:38 +08002355 r = vhost_update_avail_event(vq, vq->avail_idx);
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002356 if (r) {
2357 vq_err(vq, "Failed to update avail event index at %p: %d\n",
2358 vhost_avail_event(vq), r);
2359 return false;
2360 }
2361 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002362 /* They could have slipped one in as we were doing that: make
2363 * sure it's written, then check again. */
Michael S. Tsirkin56593382010-02-01 07:21:02 +00002364 smp_mb();
Jason Wangf8894912017-02-28 17:56:02 +08002365 r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002366 if (r) {
2367 vq_err(vq, "Failed to check avail idx at %p: %d\n",
2368 &vq->avail->idx, r);
2369 return false;
2370 }
2371
Michael S. Tsirkin3b1bbe82014-10-24 14:04:47 +03002372 return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx;
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002373}
Asias He6ac1afb2013-05-06 16:38:21 +08002374EXPORT_SYMBOL_GPL(vhost_enable_notify);
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002375
2376/* We don't need to be notified again. */
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002377void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002378{
2379 int r;
Krishna Kumard47effe2011-03-01 17:06:37 +05302380
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002381 if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
2382 return;
2383 vq->used_flags |= VRING_USED_F_NO_NOTIFY;
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03002384 if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
Jason Wang2723fea2011-06-21 18:04:38 +08002385 r = vhost_update_used_flags(vq);
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +03002386 if (r)
2387 vq_err(vq, "Failed to enable notification at %p: %d\n",
2388 &vq->used->flags, r);
2389 }
Michael S. Tsirkin3a4d5c92010-01-14 06:17:27 +00002390}
Asias He6ac1afb2013-05-06 16:38:21 +08002391EXPORT_SYMBOL_GPL(vhost_disable_notify);
2392
Jason Wang6b1e6cc2016-06-23 02:04:32 -04002393/* Create a new message. */
2394struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
2395{
2396 struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
2397 if (!node)
2398 return NULL;
2399 node->vq = vq;
2400 node->msg.type = type;
2401 return node;
2402}
2403EXPORT_SYMBOL_GPL(vhost_new_msg);
2404
2405void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
2406 struct vhost_msg_node *node)
2407{
2408 spin_lock(&dev->iotlb_lock);
2409 list_add_tail(&node->node, head);
2410 spin_unlock(&dev->iotlb_lock);
2411
2412 wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM);
2413}
2414EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
2415
2416struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
2417 struct list_head *head)
2418{
2419 struct vhost_msg_node *node = NULL;
2420
2421 spin_lock(&dev->iotlb_lock);
2422 if (!list_empty(head)) {
2423 node = list_first_entry(head, struct vhost_msg_node,
2424 node);
2425 list_del(&node->node);
2426 }
2427 spin_unlock(&dev->iotlb_lock);
2428
2429 return node;
2430}
2431EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
2432
2433
Asias He6ac1afb2013-05-06 16:38:21 +08002434static int __init vhost_init(void)
2435{
2436 return 0;
2437}
2438
2439static void __exit vhost_exit(void)
2440{
2441}
2442
2443module_init(vhost_init);
2444module_exit(vhost_exit);
2445
2446MODULE_VERSION("0.0.1");
2447MODULE_LICENSE("GPL v2");
2448MODULE_AUTHOR("Michael S. Tsirkin");
2449MODULE_DESCRIPTION("Host kernel accelerator for virtio");