]>
Commit | Line | Data |
---|---|---|
bc38a6ab RD |
1 | /* |
2 | * Copyright (c) 2005 Topspin Communications. All rights reserved. | |
33b9b3ee | 3 | * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. |
2a1d9b7f RD |
4 | * Copyright (c) 2005 Mellanox Technologies. All rights reserved. |
5 | * Copyright (c) 2005 Voltaire, Inc. All rights reserved. | |
67cdb40c | 6 | * Copyright (c) 2005 PathScale, Inc. All rights reserved. |
bc38a6ab RD |
7 | * |
8 | * This software is available to you under a choice of one of two | |
9 | * licenses. You may choose to be licensed under the terms of the GNU | |
10 | * General Public License (GPL) Version 2, available from the file | |
11 | * COPYING in the main directory of this source tree, or the | |
12 | * OpenIB.org BSD license below: | |
13 | * | |
14 | * Redistribution and use in source and binary forms, with or | |
15 | * without modification, are permitted provided that the following | |
16 | * conditions are met: | |
17 | * | |
18 | * - Redistributions of source code must retain the above | |
19 | * copyright notice, this list of conditions and the following | |
20 | * disclaimer. | |
21 | * | |
22 | * - Redistributions in binary form must reproduce the above | |
23 | * copyright notice, this list of conditions and the following | |
24 | * disclaimer in the documentation and/or other materials | |
25 | * provided with the distribution. | |
26 | * | |
27 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
28 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
29 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
30 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
31 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
32 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
33 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
34 | * SOFTWARE. | |
bc38a6ab RD |
35 | */ |
36 | ||
37 | #include <linux/module.h> | |
38 | #include <linux/init.h> | |
39 | #include <linux/device.h> | |
40 | #include <linux/err.h> | |
41 | #include <linux/fs.h> | |
42 | #include <linux/poll.h> | |
a99bbaf5 | 43 | #include <linux/sched.h> |
bc38a6ab | 44 | #include <linux/file.h> |
70a30e16 | 45 | #include <linux/cdev.h> |
a265e558 | 46 | #include <linux/anon_inodes.h> |
5a0e3ad6 | 47 | #include <linux/slab.h> |
bc38a6ab | 48 | |
7c0f6ba6 | 49 | #include <linux/uaccess.h> |
bc38a6ab | 50 | |
e6bd18f5 JG |
51 | #include <rdma/ib.h> |
52 | ||
bc38a6ab | 53 | #include "uverbs.h" |
43579b5f | 54 | #include "core_priv.h" |
fd3c7904 | 55 | #include "rdma_core.h" |
bc38a6ab RD |
56 | |
57 | MODULE_AUTHOR("Roland Dreier"); | |
58 | MODULE_DESCRIPTION("InfiniBand userspace verbs access"); | |
59 | MODULE_LICENSE("Dual BSD/GPL"); | |
60 | ||
bc38a6ab RD |
61 | enum { |
62 | IB_UVERBS_MAJOR = 231, | |
63 | IB_UVERBS_BASE_MINOR = 192, | |
64 | IB_UVERBS_MAX_DEVICES = 32 | |
65 | }; | |
66 | ||
67 | #define IB_UVERBS_BASE_DEV MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR) | |
68 | ||
70a30e16 RD |
69 | static struct class *uverbs_class; |
70 | ||
6276e08a | 71 | static DEFINE_SPINLOCK(map_lock); |
bc38a6ab RD |
72 | static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES); |
73 | ||
74 | static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file, | |
057aec0d | 75 | struct ib_device *ib_dev, |
bc38a6ab RD |
76 | const char __user *buf, int in_len, |
77 | int out_len) = { | |
9afed76d AC |
78 | [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context, |
79 | [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device, | |
80 | [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port, | |
81 | [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd, | |
82 | [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd, | |
83 | [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr, | |
7e6edb9b | 84 | [IB_USER_VERBS_CMD_REREG_MR] = ib_uverbs_rereg_mr, |
9afed76d | 85 | [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr, |
6b52a12b SM |
86 | [IB_USER_VERBS_CMD_ALLOC_MW] = ib_uverbs_alloc_mw, |
87 | [IB_USER_VERBS_CMD_DEALLOC_MW] = ib_uverbs_dealloc_mw, | |
6b73597e | 88 | [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel, |
9afed76d AC |
89 | [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq, |
90 | [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq, | |
91 | [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq, | |
92 | [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq, | |
93 | [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq, | |
94 | [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp, | |
95 | [IB_USER_VERBS_CMD_QUERY_QP] = ib_uverbs_query_qp, | |
96 | [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp, | |
97 | [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp, | |
98 | [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send, | |
99 | [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv, | |
100 | [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv, | |
101 | [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah, | |
102 | [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah, | |
103 | [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast, | |
104 | [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast, | |
105 | [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq, | |
106 | [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq, | |
107 | [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq, | |
108 | [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq, | |
53d0bd1e SH |
109 | [IB_USER_VERBS_CMD_OPEN_XRCD] = ib_uverbs_open_xrcd, |
110 | [IB_USER_VERBS_CMD_CLOSE_XRCD] = ib_uverbs_close_xrcd, | |
42849b26 | 111 | [IB_USER_VERBS_CMD_CREATE_XSRQ] = ib_uverbs_create_xsrq, |
436f2ad0 | 112 | [IB_USER_VERBS_CMD_OPEN_QP] = ib_uverbs_open_qp, |
f21519b2 YD |
113 | }; |
114 | ||
f21519b2 | 115 | static int (*uverbs_ex_cmd_table[])(struct ib_uverbs_file *file, |
057aec0d | 116 | struct ib_device *ib_dev, |
f21519b2 YD |
117 | struct ib_udata *ucore, |
118 | struct ib_udata *uhw) = { | |
119 | [IB_USER_VERBS_EX_CMD_CREATE_FLOW] = ib_uverbs_ex_create_flow, | |
5a77abf9 | 120 | [IB_USER_VERBS_EX_CMD_DESTROY_FLOW] = ib_uverbs_ex_destroy_flow, |
02d1aa7a | 121 | [IB_USER_VERBS_EX_CMD_QUERY_DEVICE] = ib_uverbs_ex_query_device, |
565197dd | 122 | [IB_USER_VERBS_EX_CMD_CREATE_CQ] = ib_uverbs_ex_create_cq, |
6d8a7497 | 123 | [IB_USER_VERBS_EX_CMD_CREATE_QP] = ib_uverbs_ex_create_qp, |
f213c052 YH |
124 | [IB_USER_VERBS_EX_CMD_CREATE_WQ] = ib_uverbs_ex_create_wq, |
125 | [IB_USER_VERBS_EX_CMD_MODIFY_WQ] = ib_uverbs_ex_modify_wq, | |
126 | [IB_USER_VERBS_EX_CMD_DESTROY_WQ] = ib_uverbs_ex_destroy_wq, | |
de019a94 YH |
127 | [IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL] = ib_uverbs_ex_create_rwq_ind_table, |
128 | [IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL] = ib_uverbs_ex_destroy_rwq_ind_table, | |
189aba99 | 129 | [IB_USER_VERBS_EX_CMD_MODIFY_QP] = ib_uverbs_ex_modify_qp, |
bc38a6ab RD |
130 | }; |
131 | ||
bc38a6ab | 132 | static void ib_uverbs_add_one(struct ib_device *device); |
7c1eb45a | 133 | static void ib_uverbs_remove_one(struct ib_device *device, void *client_data); |
bc38a6ab | 134 | |
feb7c1e3 CH |
135 | int uverbs_dealloc_mw(struct ib_mw *mw) |
136 | { | |
137 | struct ib_pd *pd = mw->pd; | |
138 | int ret; | |
139 | ||
140 | ret = mw->device->dealloc_mw(mw); | |
141 | if (!ret) | |
142 | atomic_dec(&pd->usecnt); | |
143 | return ret; | |
144 | } | |
145 | ||
35d4a0b6 | 146 | static void ib_uverbs_release_dev(struct kobject *kobj) |
70a30e16 RD |
147 | { |
148 | struct ib_uverbs_device *dev = | |
35d4a0b6 | 149 | container_of(kobj, struct ib_uverbs_device, kobj); |
70a30e16 | 150 | |
036b1063 | 151 | cleanup_srcu_struct(&dev->disassociate_srcu); |
35d4a0b6 | 152 | kfree(dev); |
70a30e16 RD |
153 | } |
154 | ||
35d4a0b6 YH |
155 | static struct kobj_type ib_uverbs_dev_ktype = { |
156 | .release = ib_uverbs_release_dev, | |
157 | }; | |
158 | ||
1e7710f3 | 159 | static void ib_uverbs_release_async_event_file(struct kref *ref) |
04d29b0e | 160 | { |
1e7710f3 MB |
161 | struct ib_uverbs_async_event_file *file = |
162 | container_of(ref, struct ib_uverbs_async_event_file, ref); | |
04d29b0e RD |
163 | |
164 | kfree(file); | |
165 | } | |
166 | ||
70a30e16 | 167 | void ib_uverbs_release_ucq(struct ib_uverbs_file *file, |
1e7710f3 | 168 | struct ib_uverbs_completion_event_file *ev_file, |
70a30e16 RD |
169 | struct ib_ucq_object *uobj) |
170 | { | |
171 | struct ib_uverbs_event *evt, *tmp; | |
172 | ||
173 | if (ev_file) { | |
db1b5ddd | 174 | spin_lock_irq(&ev_file->ev_queue.lock); |
70a30e16 RD |
175 | list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) { |
176 | list_del(&evt->list); | |
177 | kfree(evt); | |
178 | } | |
db1b5ddd | 179 | spin_unlock_irq(&ev_file->ev_queue.lock); |
70a30e16 | 180 | |
1e7710f3 | 181 | uverbs_uobject_put(&ev_file->uobj_file.uobj); |
70a30e16 RD |
182 | } |
183 | ||
db1b5ddd | 184 | spin_lock_irq(&file->async_file->ev_queue.lock); |
70a30e16 RD |
185 | list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) { |
186 | list_del(&evt->list); | |
187 | kfree(evt); | |
188 | } | |
db1b5ddd | 189 | spin_unlock_irq(&file->async_file->ev_queue.lock); |
70a30e16 RD |
190 | } |
191 | ||
192 | void ib_uverbs_release_uevent(struct ib_uverbs_file *file, | |
193 | struct ib_uevent_object *uobj) | |
194 | { | |
195 | struct ib_uverbs_event *evt, *tmp; | |
196 | ||
db1b5ddd | 197 | spin_lock_irq(&file->async_file->ev_queue.lock); |
70a30e16 RD |
198 | list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) { |
199 | list_del(&evt->list); | |
200 | kfree(evt); | |
201 | } | |
db1b5ddd | 202 | spin_unlock_irq(&file->async_file->ev_queue.lock); |
70a30e16 RD |
203 | } |
204 | ||
6be60aed MB |
205 | void ib_uverbs_detach_umcast(struct ib_qp *qp, |
206 | struct ib_uqp_object *uobj) | |
f4e40156 JM |
207 | { |
208 | struct ib_uverbs_mcast_entry *mcast, *tmp; | |
209 | ||
210 | list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) { | |
211 | ib_detach_mcast(qp, &mcast->gid, mcast->lid); | |
212 | list_del(&mcast->list); | |
213 | kfree(mcast); | |
214 | } | |
215 | } | |
216 | ||
70a30e16 | 217 | static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file, |
fd3c7904 MB |
218 | struct ib_ucontext *context, |
219 | bool device_removed) | |
bc38a6ab | 220 | { |
f7c6a7b5 | 221 | context->closing = 1; |
fd3c7904 | 222 | uverbs_cleanup_ucontext(context, device_removed); |
8ada2c1c SR |
223 | put_pid(context->tgid); |
224 | ||
43579b5f PP |
225 | ib_rdmacg_uncharge(&context->cg_obj, context->device, |
226 | RDMACG_RESOURCE_HCA_HANDLE); | |
227 | ||
bc38a6ab RD |
228 | return context->device->dealloc_ucontext(context); |
229 | } | |
230 | ||
35d4a0b6 YH |
231 | static void ib_uverbs_comp_dev(struct ib_uverbs_device *dev) |
232 | { | |
233 | complete(&dev->comp); | |
234 | } | |
235 | ||
cf8966b3 | 236 | void ib_uverbs_release_file(struct kref *ref) |
bc38a6ab RD |
237 | { |
238 | struct ib_uverbs_file *file = | |
239 | container_of(ref, struct ib_uverbs_file, ref); | |
036b1063 YH |
240 | struct ib_device *ib_dev; |
241 | int srcu_key; | |
242 | ||
243 | srcu_key = srcu_read_lock(&file->device->disassociate_srcu); | |
244 | ib_dev = srcu_dereference(file->device->ib_dev, | |
245 | &file->device->disassociate_srcu); | |
246 | if (ib_dev && !ib_dev->disassociate_ucontext) | |
247 | module_put(ib_dev->owner); | |
248 | srcu_read_unlock(&file->device->disassociate_srcu, srcu_key); | |
bc38a6ab | 249 | |
35d4a0b6 YH |
250 | if (atomic_dec_and_test(&file->device->refcount)) |
251 | ib_uverbs_comp_dev(file->device); | |
70a30e16 | 252 | |
efdd6f53 | 253 | kobject_put(&file->device->kobj); |
bc38a6ab RD |
254 | kfree(file); |
255 | } | |
256 | ||
db1b5ddd | 257 | static ssize_t ib_uverbs_event_read(struct ib_uverbs_event_queue *ev_queue, |
1e7710f3 MB |
258 | struct ib_uverbs_file *uverbs_file, |
259 | struct file *filp, char __user *buf, | |
260 | size_t count, loff_t *pos, | |
e0fcc611 | 261 | size_t eventsz) |
bc38a6ab | 262 | { |
63aaf647 | 263 | struct ib_uverbs_event *event; |
bc38a6ab RD |
264 | int ret = 0; |
265 | ||
db1b5ddd | 266 | spin_lock_irq(&ev_queue->lock); |
bc38a6ab | 267 | |
db1b5ddd MB |
268 | while (list_empty(&ev_queue->event_list)) { |
269 | spin_unlock_irq(&ev_queue->lock); | |
bc38a6ab RD |
270 | |
271 | if (filp->f_flags & O_NONBLOCK) | |
272 | return -EAGAIN; | |
273 | ||
db1b5ddd MB |
274 | if (wait_event_interruptible(ev_queue->poll_wait, |
275 | (!list_empty(&ev_queue->event_list) || | |
036b1063 YH |
276 | /* The barriers built into wait_event_interruptible() |
277 | * and wake_up() guarentee this will see the null set | |
278 | * without using RCU | |
279 | */ | |
1e7710f3 | 280 | !uverbs_file->device->ib_dev))) |
bc38a6ab RD |
281 | return -ERESTARTSYS; |
282 | ||
036b1063 | 283 | /* If device was disassociated and no event exists set an error */ |
db1b5ddd | 284 | if (list_empty(&ev_queue->event_list) && |
1e7710f3 | 285 | !uverbs_file->device->ib_dev) |
036b1063 YH |
286 | return -EIO; |
287 | ||
db1b5ddd | 288 | spin_lock_irq(&ev_queue->lock); |
bc38a6ab RD |
289 | } |
290 | ||
db1b5ddd | 291 | event = list_entry(ev_queue->event_list.next, struct ib_uverbs_event, list); |
63aaf647 | 292 | |
bc38a6ab RD |
293 | if (eventsz > count) { |
294 | ret = -EINVAL; | |
295 | event = NULL; | |
63aaf647 | 296 | } else { |
db1b5ddd | 297 | list_del(ev_queue->event_list.next); |
63aaf647 RD |
298 | if (event->counter) { |
299 | ++(*event->counter); | |
300 | list_del(&event->obj_list); | |
301 | } | |
302 | } | |
bc38a6ab | 303 | |
db1b5ddd | 304 | spin_unlock_irq(&ev_queue->lock); |
bc38a6ab RD |
305 | |
306 | if (event) { | |
307 | if (copy_to_user(buf, event, eventsz)) | |
308 | ret = -EFAULT; | |
309 | else | |
310 | ret = eventsz; | |
311 | } | |
312 | ||
313 | kfree(event); | |
314 | ||
315 | return ret; | |
316 | } | |
317 | ||
1e7710f3 MB |
318 | static ssize_t ib_uverbs_async_event_read(struct file *filp, char __user *buf, |
319 | size_t count, loff_t *pos) | |
320 | { | |
321 | struct ib_uverbs_async_event_file *file = filp->private_data; | |
322 | ||
db1b5ddd | 323 | return ib_uverbs_event_read(&file->ev_queue, file->uverbs_file, filp, |
e0fcc611 MB |
324 | buf, count, pos, |
325 | sizeof(struct ib_uverbs_async_event_desc)); | |
1e7710f3 MB |
326 | } |
327 | ||
328 | static ssize_t ib_uverbs_comp_event_read(struct file *filp, char __user *buf, | |
329 | size_t count, loff_t *pos) | |
330 | { | |
331 | struct ib_uverbs_completion_event_file *comp_ev_file = | |
332 | filp->private_data; | |
333 | ||
db1b5ddd | 334 | return ib_uverbs_event_read(&comp_ev_file->ev_queue, |
1e7710f3 | 335 | comp_ev_file->uobj_file.ufile, filp, |
e0fcc611 MB |
336 | buf, count, pos, |
337 | sizeof(struct ib_uverbs_comp_event_desc)); | |
1e7710f3 MB |
338 | } |
339 | ||
db1b5ddd | 340 | static unsigned int ib_uverbs_event_poll(struct ib_uverbs_event_queue *ev_queue, |
1e7710f3 | 341 | struct file *filp, |
bc38a6ab RD |
342 | struct poll_table_struct *wait) |
343 | { | |
344 | unsigned int pollflags = 0; | |
bc38a6ab | 345 | |
db1b5ddd | 346 | poll_wait(filp, &ev_queue->poll_wait, wait); |
bc38a6ab | 347 | |
db1b5ddd MB |
348 | spin_lock_irq(&ev_queue->lock); |
349 | if (!list_empty(&ev_queue->event_list)) | |
bc38a6ab | 350 | pollflags = POLLIN | POLLRDNORM; |
db1b5ddd | 351 | spin_unlock_irq(&ev_queue->lock); |
bc38a6ab RD |
352 | |
353 | return pollflags; | |
354 | } | |
355 | ||
1e7710f3 MB |
356 | static unsigned int ib_uverbs_async_event_poll(struct file *filp, |
357 | struct poll_table_struct *wait) | |
358 | { | |
359 | return ib_uverbs_event_poll(filp->private_data, filp, wait); | |
360 | } | |
361 | ||
362 | static unsigned int ib_uverbs_comp_event_poll(struct file *filp, | |
363 | struct poll_table_struct *wait) | |
364 | { | |
365 | struct ib_uverbs_completion_event_file *comp_ev_file = | |
366 | filp->private_data; | |
367 | ||
db1b5ddd | 368 | return ib_uverbs_event_poll(&comp_ev_file->ev_queue, filp, wait); |
1e7710f3 MB |
369 | } |
370 | ||
371 | static int ib_uverbs_async_event_fasync(int fd, struct file *filp, int on) | |
abdf119b | 372 | { |
db1b5ddd | 373 | struct ib_uverbs_event_queue *ev_queue = filp->private_data; |
abdf119b | 374 | |
db1b5ddd | 375 | return fasync_helper(fd, filp, on, &ev_queue->async_queue); |
abdf119b GN |
376 | } |
377 | ||
1e7710f3 | 378 | static int ib_uverbs_comp_event_fasync(int fd, struct file *filp, int on) |
bc38a6ab | 379 | { |
1e7710f3 MB |
380 | struct ib_uverbs_completion_event_file *comp_ev_file = |
381 | filp->private_data; | |
382 | ||
db1b5ddd | 383 | return fasync_helper(fd, filp, on, &comp_ev_file->ev_queue.async_queue); |
1e7710f3 MB |
384 | } |
385 | ||
386 | static int ib_uverbs_async_event_close(struct inode *inode, struct file *filp) | |
387 | { | |
388 | struct ib_uverbs_async_event_file *file = filp->private_data; | |
389 | struct ib_uverbs_file *uverbs_file = file->uverbs_file; | |
6b73597e | 390 | struct ib_uverbs_event *entry, *tmp; |
036b1063 | 391 | int closed_already = 0; |
6b73597e | 392 | |
1e7710f3 | 393 | mutex_lock(&uverbs_file->device->lists_mutex); |
db1b5ddd MB |
394 | spin_lock_irq(&file->ev_queue.lock); |
395 | closed_already = file->ev_queue.is_closed; | |
396 | file->ev_queue.is_closed = 1; | |
397 | list_for_each_entry_safe(entry, tmp, &file->ev_queue.event_list, list) { | |
6b73597e RD |
398 | if (entry->counter) |
399 | list_del(&entry->obj_list); | |
400 | kfree(entry); | |
401 | } | |
db1b5ddd | 402 | spin_unlock_irq(&file->ev_queue.lock); |
036b1063 YH |
403 | if (!closed_already) { |
404 | list_del(&file->list); | |
1e7710f3 MB |
405 | ib_unregister_event_handler(&uverbs_file->event_handler); |
406 | } | |
407 | mutex_unlock(&uverbs_file->device->lists_mutex); | |
408 | ||
409 | kref_put(&uverbs_file->ref, ib_uverbs_release_file); | |
410 | kref_put(&file->ref, ib_uverbs_release_async_event_file); | |
411 | ||
412 | return 0; | |
413 | } | |
414 | ||
415 | static int ib_uverbs_comp_event_close(struct inode *inode, struct file *filp) | |
416 | { | |
417 | struct ib_uverbs_completion_event_file *file = filp->private_data; | |
418 | struct ib_uverbs_event *entry, *tmp; | |
419 | ||
db1b5ddd MB |
420 | spin_lock_irq(&file->ev_queue.lock); |
421 | list_for_each_entry_safe(entry, tmp, &file->ev_queue.event_list, list) { | |
1e7710f3 MB |
422 | if (entry->counter) |
423 | list_del(&entry->obj_list); | |
424 | kfree(entry); | |
036b1063 | 425 | } |
db1b5ddd | 426 | spin_unlock_irq(&file->ev_queue.lock); |
bc38a6ab | 427 | |
1e7710f3 | 428 | uverbs_close_fd(filp); |
bc38a6ab RD |
429 | |
430 | return 0; | |
431 | } | |
432 | ||
1e7710f3 | 433 | const struct file_operations uverbs_event_fops = { |
6b73597e | 434 | .owner = THIS_MODULE, |
1e7710f3 MB |
435 | .read = ib_uverbs_comp_event_read, |
436 | .poll = ib_uverbs_comp_event_poll, | |
437 | .release = ib_uverbs_comp_event_close, | |
438 | .fasync = ib_uverbs_comp_event_fasync, | |
439 | .llseek = no_llseek, | |
440 | }; | |
441 | ||
442 | static const struct file_operations uverbs_async_event_fops = { | |
443 | .owner = THIS_MODULE, | |
444 | .read = ib_uverbs_async_event_read, | |
445 | .poll = ib_uverbs_async_event_poll, | |
446 | .release = ib_uverbs_async_event_close, | |
447 | .fasync = ib_uverbs_async_event_fasync, | |
bc1db9af | 448 | .llseek = no_llseek, |
bc38a6ab RD |
449 | }; |
450 | ||
451 | void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context) | |
452 | { | |
db1b5ddd | 453 | struct ib_uverbs_event_queue *ev_queue = cq_context; |
6b73597e RD |
454 | struct ib_ucq_object *uobj; |
455 | struct ib_uverbs_event *entry; | |
456 | unsigned long flags; | |
457 | ||
db1b5ddd | 458 | if (!ev_queue) |
6b73597e RD |
459 | return; |
460 | ||
db1b5ddd MB |
461 | spin_lock_irqsave(&ev_queue->lock, flags); |
462 | if (ev_queue->is_closed) { | |
463 | spin_unlock_irqrestore(&ev_queue->lock, flags); | |
6b73597e RD |
464 | return; |
465 | } | |
bc38a6ab RD |
466 | |
467 | entry = kmalloc(sizeof *entry, GFP_ATOMIC); | |
305a7e87 | 468 | if (!entry) { |
db1b5ddd | 469 | spin_unlock_irqrestore(&ev_queue->lock, flags); |
bc38a6ab | 470 | return; |
305a7e87 | 471 | } |
bc38a6ab | 472 | |
63aaf647 RD |
473 | uobj = container_of(cq->uobject, struct ib_ucq_object, uobject); |
474 | ||
475 | entry->desc.comp.cq_handle = cq->uobject->user_handle; | |
476 | entry->counter = &uobj->comp_events_reported; | |
bc38a6ab | 477 | |
db1b5ddd | 478 | list_add_tail(&entry->list, &ev_queue->event_list); |
63aaf647 | 479 | list_add_tail(&entry->obj_list, &uobj->comp_list); |
db1b5ddd | 480 | spin_unlock_irqrestore(&ev_queue->lock, flags); |
bc38a6ab | 481 | |
db1b5ddd MB |
482 | wake_up_interruptible(&ev_queue->poll_wait); |
483 | kill_fasync(&ev_queue->async_queue, SIGIO, POLL_IN); | |
bc38a6ab RD |
484 | } |
485 | ||
486 | static void ib_uverbs_async_handler(struct ib_uverbs_file *file, | |
63aaf647 RD |
487 | __u64 element, __u64 event, |
488 | struct list_head *obj_list, | |
489 | u32 *counter) | |
bc38a6ab | 490 | { |
63aaf647 | 491 | struct ib_uverbs_event *entry; |
bc38a6ab RD |
492 | unsigned long flags; |
493 | ||
db1b5ddd MB |
494 | spin_lock_irqsave(&file->async_file->ev_queue.lock, flags); |
495 | if (file->async_file->ev_queue.is_closed) { | |
496 | spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags); | |
6b73597e RD |
497 | return; |
498 | } | |
499 | ||
bc38a6ab | 500 | entry = kmalloc(sizeof *entry, GFP_ATOMIC); |
305a7e87 | 501 | if (!entry) { |
db1b5ddd | 502 | spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags); |
bc38a6ab | 503 | return; |
305a7e87 | 504 | } |
bc38a6ab | 505 | |
63aaf647 RD |
506 | entry->desc.async.element = element; |
507 | entry->desc.async.event_type = event; | |
377b5134 | 508 | entry->desc.async.reserved = 0; |
63aaf647 | 509 | entry->counter = counter; |
bc38a6ab | 510 | |
db1b5ddd | 511 | list_add_tail(&entry->list, &file->async_file->ev_queue.event_list); |
63aaf647 RD |
512 | if (obj_list) |
513 | list_add_tail(&entry->obj_list, obj_list); | |
db1b5ddd | 514 | spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags); |
bc38a6ab | 515 | |
db1b5ddd MB |
516 | wake_up_interruptible(&file->async_file->ev_queue.poll_wait); |
517 | kill_fasync(&file->async_file->ev_queue.async_queue, SIGIO, POLL_IN); | |
bc38a6ab RD |
518 | } |
519 | ||
520 | void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr) | |
521 | { | |
7162a3e0 RD |
522 | struct ib_ucq_object *uobj = container_of(event->element.cq->uobject, |
523 | struct ib_ucq_object, uobject); | |
63aaf647 | 524 | |
7162a3e0 | 525 | ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle, |
63aaf647 RD |
526 | event->event, &uobj->async_list, |
527 | &uobj->async_events_reported); | |
bc38a6ab RD |
528 | } |
529 | ||
530 | void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr) | |
531 | { | |
63aaf647 RD |
532 | struct ib_uevent_object *uobj; |
533 | ||
a040f95d | 534 | /* for XRC target qp's, check that qp is live */ |
fd3c7904 | 535 | if (!event->element.qp->uobject) |
a040f95d JM |
536 | return; |
537 | ||
63aaf647 RD |
538 | uobj = container_of(event->element.qp->uobject, |
539 | struct ib_uevent_object, uobject); | |
540 | ||
541 | ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle, | |
542 | event->event, &uobj->event_list, | |
543 | &uobj->events_reported); | |
bc38a6ab RD |
544 | } |
545 | ||
f213c052 YH |
546 | void ib_uverbs_wq_event_handler(struct ib_event *event, void *context_ptr) |
547 | { | |
548 | struct ib_uevent_object *uobj = container_of(event->element.wq->uobject, | |
549 | struct ib_uevent_object, uobject); | |
550 | ||
551 | ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle, | |
552 | event->event, &uobj->event_list, | |
553 | &uobj->events_reported); | |
554 | } | |
555 | ||
f520ba5a RD |
556 | void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr) |
557 | { | |
63aaf647 RD |
558 | struct ib_uevent_object *uobj; |
559 | ||
560 | uobj = container_of(event->element.srq->uobject, | |
561 | struct ib_uevent_object, uobject); | |
562 | ||
563 | ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle, | |
564 | event->event, &uobj->event_list, | |
565 | &uobj->events_reported); | |
f520ba5a RD |
566 | } |
567 | ||
6b73597e RD |
568 | void ib_uverbs_event_handler(struct ib_event_handler *handler, |
569 | struct ib_event *event) | |
bc38a6ab RD |
570 | { |
571 | struct ib_uverbs_file *file = | |
572 | container_of(handler, struct ib_uverbs_file, event_handler); | |
573 | ||
63aaf647 RD |
574 | ib_uverbs_async_handler(file, event->element.port_num, event->event, |
575 | NULL, NULL); | |
bc38a6ab RD |
576 | } |
577 | ||
03c40442 YH |
578 | void ib_uverbs_free_async_event_file(struct ib_uverbs_file *file) |
579 | { | |
1e7710f3 | 580 | kref_put(&file->async_file->ref, ib_uverbs_release_async_event_file); |
03c40442 YH |
581 | file->async_file = NULL; |
582 | } | |
583 | ||
db1b5ddd | 584 | void ib_uverbs_init_event_queue(struct ib_uverbs_event_queue *ev_queue) |
bc38a6ab | 585 | { |
db1b5ddd MB |
586 | spin_lock_init(&ev_queue->lock); |
587 | INIT_LIST_HEAD(&ev_queue->event_list); | |
588 | init_waitqueue_head(&ev_queue->poll_wait); | |
589 | ev_queue->is_closed = 0; | |
590 | ev_queue->async_queue = NULL; | |
1e7710f3 MB |
591 | } |
592 | ||
593 | struct file *ib_uverbs_alloc_async_event_file(struct ib_uverbs_file *uverbs_file, | |
594 | struct ib_device *ib_dev) | |
595 | { | |
596 | struct ib_uverbs_async_event_file *ev_file; | |
bc38a6ab | 597 | struct file *filp; |
03c40442 | 598 | int ret; |
bc38a6ab | 599 | |
03c40442 | 600 | ev_file = kzalloc(sizeof(*ev_file), GFP_KERNEL); |
6b73597e RD |
601 | if (!ev_file) |
602 | return ERR_PTR(-ENOMEM); | |
603 | ||
db1b5ddd | 604 | ib_uverbs_init_event_queue(&ev_file->ev_queue); |
6b73597e | 605 | ev_file->uverbs_file = uverbs_file; |
03c40442 | 606 | kref_get(&ev_file->uverbs_file->ref); |
1e7710f3 MB |
607 | kref_init(&ev_file->ref); |
608 | filp = anon_inode_getfile("[infinibandevent]", &uverbs_async_event_fops, | |
a265e558 | 609 | ev_file, O_RDONLY); |
b1e4594b | 610 | if (IS_ERR(filp)) |
03c40442 YH |
611 | goto err_put_refs; |
612 | ||
036b1063 YH |
613 | mutex_lock(&uverbs_file->device->lists_mutex); |
614 | list_add_tail(&ev_file->list, | |
615 | &uverbs_file->device->uverbs_events_file_list); | |
616 | mutex_unlock(&uverbs_file->device->lists_mutex); | |
617 | ||
1e7710f3 MB |
618 | WARN_ON(uverbs_file->async_file); |
619 | uverbs_file->async_file = ev_file; | |
620 | kref_get(&uverbs_file->async_file->ref); | |
621 | INIT_IB_EVENT_HANDLER(&uverbs_file->event_handler, | |
622 | ib_dev, | |
623 | ib_uverbs_event_handler); | |
624 | ret = ib_register_event_handler(&uverbs_file->event_handler); | |
625 | if (ret) | |
626 | goto err_put_file; | |
627 | ||
628 | /* At that point async file stuff was fully set */ | |
03c40442 YH |
629 | |
630 | return filp; | |
631 | ||
632 | err_put_file: | |
633 | fput(filp); | |
1e7710f3 MB |
634 | kref_put(&uverbs_file->async_file->ref, |
635 | ib_uverbs_release_async_event_file); | |
03c40442 YH |
636 | uverbs_file->async_file = NULL; |
637 | return ERR_PTR(ret); | |
a7dab9e8 | 638 | |
03c40442 YH |
639 | err_put_refs: |
640 | kref_put(&ev_file->uverbs_file->ref, ib_uverbs_release_file); | |
1e7710f3 | 641 | kref_put(&ev_file->ref, ib_uverbs_release_async_event_file); |
6b73597e | 642 | return filp; |
6b73597e RD |
643 | } |
644 | ||
2dbd5186 EC |
645 | static int verify_command_mask(struct ib_device *ib_dev, __u32 command) |
646 | { | |
647 | u64 mask; | |
648 | ||
649 | if (command <= IB_USER_VERBS_CMD_OPEN_QP) | |
650 | mask = ib_dev->uverbs_cmd_mask; | |
651 | else | |
652 | mask = ib_dev->uverbs_ex_cmd_mask; | |
653 | ||
654 | if (mask & ((u64)1 << command)) | |
655 | return 0; | |
656 | ||
657 | return -1; | |
658 | } | |
659 | ||
bc38a6ab RD |
660 | static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf, |
661 | size_t count, loff_t *pos) | |
662 | { | |
663 | struct ib_uverbs_file *file = filp->private_data; | |
036b1063 | 664 | struct ib_device *ib_dev; |
bc38a6ab | 665 | struct ib_uverbs_cmd_hdr hdr; |
74a0b0a5 | 666 | __u32 command; |
f21519b2 | 667 | __u32 flags; |
036b1063 YH |
668 | int srcu_key; |
669 | ssize_t ret; | |
057aec0d | 670 | |
f73a1dbc LR |
671 | if (!ib_safe_file_access(filp)) { |
672 | pr_err_once("uverbs_write: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n", | |
673 | task_tgid_vnr(current), current->comm); | |
e6bd18f5 | 674 | return -EACCES; |
f73a1dbc | 675 | } |
e6bd18f5 | 676 | |
bc38a6ab RD |
677 | if (count < sizeof hdr) |
678 | return -EINVAL; | |
679 | ||
680 | if (copy_from_user(&hdr, buf, sizeof hdr)) | |
681 | return -EFAULT; | |
682 | ||
036b1063 YH |
683 | srcu_key = srcu_read_lock(&file->device->disassociate_srcu); |
684 | ib_dev = srcu_dereference(file->device->ib_dev, | |
685 | &file->device->disassociate_srcu); | |
686 | if (!ib_dev) { | |
687 | ret = -EIO; | |
688 | goto out; | |
689 | } | |
690 | ||
74a0b0a5 EC |
691 | if (hdr.command & ~(__u32)(IB_USER_VERBS_CMD_FLAGS_MASK | |
692 | IB_USER_VERBS_CMD_COMMAND_MASK)) { | |
693 | ret = -EINVAL; | |
694 | goto out; | |
695 | } | |
696 | ||
697 | command = hdr.command & IB_USER_VERBS_CMD_COMMAND_MASK; | |
2dbd5186 EC |
698 | if (verify_command_mask(ib_dev, command)) { |
699 | ret = -EOPNOTSUPP; | |
700 | goto out; | |
701 | } | |
74a0b0a5 | 702 | |
eaebc7d2 EC |
703 | if (!file->ucontext && |
704 | command != IB_USER_VERBS_CMD_GET_CONTEXT) { | |
705 | ret = -EINVAL; | |
706 | goto out; | |
707 | } | |
708 | ||
f21519b2 YD |
709 | flags = (hdr.command & |
710 | IB_USER_VERBS_CMD_FLAGS_MASK) >> IB_USER_VERBS_CMD_FLAGS_SHIFT; | |
bc38a6ab | 711 | |
f21519b2 | 712 | if (!flags) { |
f21519b2 | 713 | if (command >= ARRAY_SIZE(uverbs_cmd_table) || |
036b1063 YH |
714 | !uverbs_cmd_table[command]) { |
715 | ret = -EINVAL; | |
716 | goto out; | |
717 | } | |
400dbc96 | 718 | |
036b1063 YH |
719 | if (hdr.in_words * 4 != count) { |
720 | ret = -EINVAL; | |
721 | goto out; | |
722 | } | |
400dbc96 | 723 | |
036b1063 | 724 | ret = uverbs_cmd_table[command](file, ib_dev, |
f21519b2 YD |
725 | buf + sizeof(hdr), |
726 | hdr.in_words * 4, | |
727 | hdr.out_words * 4); | |
728 | ||
f21519b2 | 729 | } else if (flags == IB_USER_VERBS_CMD_FLAG_EXTENDED) { |
f21519b2 YD |
730 | struct ib_uverbs_ex_cmd_hdr ex_hdr; |
731 | struct ib_udata ucore; | |
732 | struct ib_udata uhw; | |
f21519b2 YD |
733 | size_t written_count = count; |
734 | ||
f21519b2 | 735 | if (command >= ARRAY_SIZE(uverbs_ex_cmd_table) || |
036b1063 YH |
736 | !uverbs_ex_cmd_table[command]) { |
737 | ret = -ENOSYS; | |
738 | goto out; | |
739 | } | |
f21519b2 | 740 | |
036b1063 YH |
741 | if (!file->ucontext) { |
742 | ret = -EINVAL; | |
743 | goto out; | |
744 | } | |
f21519b2 | 745 | |
036b1063 YH |
746 | if (count < (sizeof(hdr) + sizeof(ex_hdr))) { |
747 | ret = -EINVAL; | |
748 | goto out; | |
749 | } | |
f21519b2 | 750 | |
036b1063 YH |
751 | if (copy_from_user(&ex_hdr, buf + sizeof(hdr), sizeof(ex_hdr))) { |
752 | ret = -EFAULT; | |
753 | goto out; | |
754 | } | |
f21519b2 YD |
755 | |
756 | count -= sizeof(hdr) + sizeof(ex_hdr); | |
757 | buf += sizeof(hdr) + sizeof(ex_hdr); | |
758 | ||
036b1063 YH |
759 | if ((hdr.in_words + ex_hdr.provider_in_words) * 8 != count) { |
760 | ret = -EINVAL; | |
761 | goto out; | |
762 | } | |
f21519b2 | 763 | |
036b1063 YH |
764 | if (ex_hdr.cmd_hdr_reserved) { |
765 | ret = -EINVAL; | |
766 | goto out; | |
767 | } | |
7efb1b19 | 768 | |
f21519b2 | 769 | if (ex_hdr.response) { |
036b1063 YH |
770 | if (!hdr.out_words && !ex_hdr.provider_out_words) { |
771 | ret = -EINVAL; | |
772 | goto out; | |
773 | } | |
6cc3df84 YD |
774 | |
775 | if (!access_ok(VERIFY_WRITE, | |
776 | (void __user *) (unsigned long) ex_hdr.response, | |
036b1063 YH |
777 | (hdr.out_words + ex_hdr.provider_out_words) * 8)) { |
778 | ret = -EFAULT; | |
779 | goto out; | |
780 | } | |
f21519b2 | 781 | } else { |
036b1063 YH |
782 | if (hdr.out_words || ex_hdr.provider_out_words) { |
783 | ret = -EINVAL; | |
784 | goto out; | |
785 | } | |
f21519b2 YD |
786 | } |
787 | ||
a96e4e2f RD |
788 | INIT_UDATA_BUF_OR_NULL(&ucore, buf, (unsigned long) ex_hdr.response, |
789 | hdr.in_words * 8, hdr.out_words * 8); | |
790 | ||
791 | INIT_UDATA_BUF_OR_NULL(&uhw, | |
792 | buf + ucore.inlen, | |
793 | (unsigned long) ex_hdr.response + ucore.outlen, | |
794 | ex_hdr.provider_in_words * 8, | |
795 | ex_hdr.provider_out_words * 8); | |
f21519b2 | 796 | |
036b1063 | 797 | ret = uverbs_ex_cmd_table[command](file, |
057aec0d | 798 | ib_dev, |
f21519b2 YD |
799 | &ucore, |
800 | &uhw); | |
036b1063 YH |
801 | if (!ret) |
802 | ret = written_count; | |
803 | } else { | |
804 | ret = -ENOSYS; | |
400dbc96 | 805 | } |
f21519b2 | 806 | |
036b1063 YH |
807 | out: |
808 | srcu_read_unlock(&file->device->disassociate_srcu, srcu_key); | |
809 | return ret; | |
bc38a6ab RD |
810 | } |
811 | ||
812 | static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma) | |
813 | { | |
814 | struct ib_uverbs_file *file = filp->private_data; | |
036b1063 YH |
815 | struct ib_device *ib_dev; |
816 | int ret = 0; | |
817 | int srcu_key; | |
bc38a6ab | 818 | |
036b1063 YH |
819 | srcu_key = srcu_read_lock(&file->device->disassociate_srcu); |
820 | ib_dev = srcu_dereference(file->device->ib_dev, | |
821 | &file->device->disassociate_srcu); | |
822 | if (!ib_dev) { | |
823 | ret = -EIO; | |
824 | goto out; | |
825 | } | |
826 | ||
827 | if (!file->ucontext) | |
828 | ret = -ENODEV; | |
bc38a6ab | 829 | else |
036b1063 YH |
830 | ret = ib_dev->mmap(file->ucontext, vma); |
831 | out: | |
832 | srcu_read_unlock(&file->device->disassociate_srcu, srcu_key); | |
833 | return ret; | |
bc38a6ab RD |
834 | } |
835 | ||
5b2d281a RD |
836 | /* |
837 | * ib_uverbs_open() does not need the BKL: | |
838 | * | |
2a72f212 | 839 | * - the ib_uverbs_device structures are properly reference counted and |
5b2d281a RD |
840 | * everything else is purely local to the file being created, so |
841 | * races against other open calls are not a problem; | |
842 | * - there is no ioctl method to race against; | |
2a72f212 AC |
843 | * - the open method will either immediately run -ENXIO, or all |
844 | * required initialization will be done. | |
5b2d281a | 845 | */ |
bc38a6ab RD |
846 | static int ib_uverbs_open(struct inode *inode, struct file *filp) |
847 | { | |
70a30e16 | 848 | struct ib_uverbs_device *dev; |
bc38a6ab | 849 | struct ib_uverbs_file *file; |
036b1063 | 850 | struct ib_device *ib_dev; |
70a30e16 | 851 | int ret; |
036b1063 YH |
852 | int module_dependent; |
853 | int srcu_key; | |
bc38a6ab | 854 | |
2a72f212 | 855 | dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev); |
35d4a0b6 | 856 | if (!atomic_inc_not_zero(&dev->refcount)) |
70a30e16 RD |
857 | return -ENXIO; |
858 | ||
036b1063 YH |
859 | srcu_key = srcu_read_lock(&dev->disassociate_srcu); |
860 | mutex_lock(&dev->lists_mutex); | |
861 | ib_dev = srcu_dereference(dev->ib_dev, | |
862 | &dev->disassociate_srcu); | |
863 | if (!ib_dev) { | |
864 | ret = -EIO; | |
70a30e16 RD |
865 | goto err; |
866 | } | |
bc38a6ab | 867 | |
036b1063 YH |
868 | /* In case IB device supports disassociate ucontext, there is no hard |
869 | * dependency between uverbs device and its low level device. | |
870 | */ | |
871 | module_dependent = !(ib_dev->disassociate_ucontext); | |
872 | ||
873 | if (module_dependent) { | |
874 | if (!try_module_get(ib_dev->owner)) { | |
875 | ret = -ENODEV; | |
876 | goto err; | |
877 | } | |
878 | } | |
879 | ||
880 | file = kzalloc(sizeof(*file), GFP_KERNEL); | |
63c47c28 | 881 | if (!file) { |
70a30e16 | 882 | ret = -ENOMEM; |
036b1063 YH |
883 | if (module_dependent) |
884 | goto err_module; | |
885 | ||
886 | goto err; | |
63c47c28 | 887 | } |
bc38a6ab | 888 | |
70a30e16 | 889 | file->device = dev; |
771addf6 MB |
890 | spin_lock_init(&file->idr_lock); |
891 | idr_init(&file->idr); | |
70a30e16 RD |
892 | file->ucontext = NULL; |
893 | file->async_file = NULL; | |
bc38a6ab | 894 | kref_init(&file->ref); |
95ed644f | 895 | mutex_init(&file->mutex); |
d1e09f30 | 896 | mutex_init(&file->cleanup_mutex); |
bc38a6ab | 897 | |
bc38a6ab | 898 | filp->private_data = file; |
35d4a0b6 | 899 | kobject_get(&dev->kobj); |
036b1063 YH |
900 | list_add_tail(&file->list, &dev->uverbs_file_list); |
901 | mutex_unlock(&dev->lists_mutex); | |
902 | srcu_read_unlock(&dev->disassociate_srcu, srcu_key); | |
bc38a6ab | 903 | |
bc1db9af | 904 | return nonseekable_open(inode, filp); |
70a30e16 RD |
905 | |
906 | err_module: | |
036b1063 | 907 | module_put(ib_dev->owner); |
70a30e16 RD |
908 | |
909 | err: | |
036b1063 YH |
910 | mutex_unlock(&dev->lists_mutex); |
911 | srcu_read_unlock(&dev->disassociate_srcu, srcu_key); | |
35d4a0b6 YH |
912 | if (atomic_dec_and_test(&dev->refcount)) |
913 | ib_uverbs_comp_dev(dev); | |
914 | ||
70a30e16 | 915 | return ret; |
bc38a6ab RD |
916 | } |
917 | ||
918 | static int ib_uverbs_close(struct inode *inode, struct file *filp) | |
919 | { | |
920 | struct ib_uverbs_file *file = filp->private_data; | |
d1e09f30 JG |
921 | |
922 | mutex_lock(&file->cleanup_mutex); | |
923 | if (file->ucontext) { | |
fd3c7904 | 924 | ib_uverbs_cleanup_ucontext(file, file->ucontext, false); |
d1e09f30 JG |
925 | file->ucontext = NULL; |
926 | } | |
927 | mutex_unlock(&file->cleanup_mutex); | |
771addf6 | 928 | idr_destroy(&file->idr); |
036b1063 YH |
929 | |
930 | mutex_lock(&file->device->lists_mutex); | |
036b1063 YH |
931 | if (!file->is_closed) { |
932 | list_del(&file->list); | |
933 | file->is_closed = 1; | |
934 | } | |
935 | mutex_unlock(&file->device->lists_mutex); | |
70a30e16 RD |
936 | |
937 | if (file->async_file) | |
1e7710f3 MB |
938 | kref_put(&file->async_file->ref, |
939 | ib_uverbs_release_async_event_file); | |
bc38a6ab | 940 | |
bc38a6ab RD |
941 | kref_put(&file->ref, ib_uverbs_release_file); |
942 | ||
943 | return 0; | |
944 | } | |
945 | ||
2b8693c0 | 946 | static const struct file_operations uverbs_fops = { |
9afed76d AC |
947 | .owner = THIS_MODULE, |
948 | .write = ib_uverbs_write, | |
949 | .open = ib_uverbs_open, | |
bc1db9af RD |
950 | .release = ib_uverbs_close, |
951 | .llseek = no_llseek, | |
bc38a6ab RD |
952 | }; |
953 | ||
2b8693c0 | 954 | static const struct file_operations uverbs_mmap_fops = { |
9afed76d AC |
955 | .owner = THIS_MODULE, |
956 | .write = ib_uverbs_write, | |
bc38a6ab | 957 | .mmap = ib_uverbs_mmap, |
9afed76d | 958 | .open = ib_uverbs_open, |
bc1db9af RD |
959 | .release = ib_uverbs_close, |
960 | .llseek = no_llseek, | |
bc38a6ab RD |
961 | }; |
962 | ||
963 | static struct ib_client uverbs_client = { | |
964 | .name = "uverbs", | |
965 | .add = ib_uverbs_add_one, | |
966 | .remove = ib_uverbs_remove_one | |
967 | }; | |
968 | ||
f4e91eb4 TJ |
969 | static ssize_t show_ibdev(struct device *device, struct device_attribute *attr, |
970 | char *buf) | |
bc38a6ab | 971 | { |
036b1063 YH |
972 | int ret = -ENODEV; |
973 | int srcu_key; | |
f4e91eb4 | 974 | struct ib_uverbs_device *dev = dev_get_drvdata(device); |
036b1063 | 975 | struct ib_device *ib_dev; |
70a30e16 RD |
976 | |
977 | if (!dev) | |
978 | return -ENODEV; | |
bc38a6ab | 979 | |
036b1063 YH |
980 | srcu_key = srcu_read_lock(&dev->disassociate_srcu); |
981 | ib_dev = srcu_dereference(dev->ib_dev, &dev->disassociate_srcu); | |
982 | if (ib_dev) | |
983 | ret = sprintf(buf, "%s\n", ib_dev->name); | |
984 | srcu_read_unlock(&dev->disassociate_srcu, srcu_key); | |
985 | ||
986 | return ret; | |
bc38a6ab | 987 | } |
f4e91eb4 | 988 | static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); |
bc38a6ab | 989 | |
f4e91eb4 TJ |
990 | static ssize_t show_dev_abi_version(struct device *device, |
991 | struct device_attribute *attr, char *buf) | |
274c0891 | 992 | { |
f4e91eb4 | 993 | struct ib_uverbs_device *dev = dev_get_drvdata(device); |
036b1063 YH |
994 | int ret = -ENODEV; |
995 | int srcu_key; | |
996 | struct ib_device *ib_dev; | |
70a30e16 RD |
997 | |
998 | if (!dev) | |
999 | return -ENODEV; | |
036b1063 YH |
1000 | srcu_key = srcu_read_lock(&dev->disassociate_srcu); |
1001 | ib_dev = srcu_dereference(dev->ib_dev, &dev->disassociate_srcu); | |
1002 | if (ib_dev) | |
1003 | ret = sprintf(buf, "%d\n", ib_dev->uverbs_abi_ver); | |
1004 | srcu_read_unlock(&dev->disassociate_srcu, srcu_key); | |
274c0891 | 1005 | |
036b1063 | 1006 | return ret; |
274c0891 | 1007 | } |
f4e91eb4 | 1008 | static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL); |
274c0891 | 1009 | |
0933e2d9 AK |
1010 | static CLASS_ATTR_STRING(abi_version, S_IRUGO, |
1011 | __stringify(IB_USER_VERBS_ABI_VERSION)); | |
bc38a6ab | 1012 | |
6d6a0e71 AC |
1013 | static dev_t overflow_maj; |
1014 | static DECLARE_BITMAP(overflow_map, IB_UVERBS_MAX_DEVICES); | |
1015 | ||
1016 | /* | |
1017 | * If we have more than IB_UVERBS_MAX_DEVICES, dynamically overflow by | |
1018 | * requesting a new major number and doubling the number of max devices we | |
1019 | * support. It's stupid, but simple. | |
1020 | */ | |
1021 | static int find_overflow_devnum(void) | |
1022 | { | |
1023 | int ret; | |
1024 | ||
1025 | if (!overflow_maj) { | |
1026 | ret = alloc_chrdev_region(&overflow_maj, 0, IB_UVERBS_MAX_DEVICES, | |
1027 | "infiniband_verbs"); | |
1028 | if (ret) { | |
aba25a3e | 1029 | pr_err("user_verbs: couldn't register dynamic device number\n"); |
6d6a0e71 AC |
1030 | return ret; |
1031 | } | |
1032 | } | |
1033 | ||
1034 | ret = find_first_zero_bit(overflow_map, IB_UVERBS_MAX_DEVICES); | |
1035 | if (ret >= IB_UVERBS_MAX_DEVICES) | |
1036 | return -1; | |
1037 | ||
1038 | return ret; | |
1039 | } | |
1040 | ||
bc38a6ab RD |
1041 | static void ib_uverbs_add_one(struct ib_device *device) |
1042 | { | |
38707980 | 1043 | int devnum; |
ddbd6883 | 1044 | dev_t base; |
bc38a6ab | 1045 | struct ib_uverbs_device *uverbs_dev; |
036b1063 | 1046 | int ret; |
bc38a6ab RD |
1047 | |
1048 | if (!device->alloc_ucontext) | |
1049 | return; | |
1050 | ||
de6eb66b | 1051 | uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL); |
bc38a6ab RD |
1052 | if (!uverbs_dev) |
1053 | return; | |
1054 | ||
036b1063 YH |
1055 | ret = init_srcu_struct(&uverbs_dev->disassociate_srcu); |
1056 | if (ret) { | |
1057 | kfree(uverbs_dev); | |
1058 | return; | |
1059 | } | |
1060 | ||
35d4a0b6 | 1061 | atomic_set(&uverbs_dev->refcount, 1); |
fd60ae40 | 1062 | init_completion(&uverbs_dev->comp); |
53d0bd1e SH |
1063 | uverbs_dev->xrcd_tree = RB_ROOT; |
1064 | mutex_init(&uverbs_dev->xrcd_tree_mutex); | |
35d4a0b6 | 1065 | kobject_init(&uverbs_dev->kobj, &ib_uverbs_dev_ktype); |
036b1063 YH |
1066 | mutex_init(&uverbs_dev->lists_mutex); |
1067 | INIT_LIST_HEAD(&uverbs_dev->uverbs_file_list); | |
1068 | INIT_LIST_HEAD(&uverbs_dev->uverbs_events_file_list); | |
70a30e16 | 1069 | |
bc38a6ab | 1070 | spin_lock(&map_lock); |
38707980 AC |
1071 | devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES); |
1072 | if (devnum >= IB_UVERBS_MAX_DEVICES) { | |
bc38a6ab | 1073 | spin_unlock(&map_lock); |
6d6a0e71 AC |
1074 | devnum = find_overflow_devnum(); |
1075 | if (devnum < 0) | |
1076 | goto err; | |
1077 | ||
1078 | spin_lock(&map_lock); | |
1079 | uverbs_dev->devnum = devnum + IB_UVERBS_MAX_DEVICES; | |
1080 | base = devnum + overflow_maj; | |
1081 | set_bit(devnum, overflow_map); | |
1082 | } else { | |
1083 | uverbs_dev->devnum = devnum; | |
1084 | base = devnum + IB_UVERBS_BASE_DEV; | |
1085 | set_bit(devnum, dev_map); | |
bc38a6ab | 1086 | } |
bc38a6ab RD |
1087 | spin_unlock(&map_lock); |
1088 | ||
036b1063 | 1089 | rcu_assign_pointer(uverbs_dev->ib_dev, device); |
f4fd0b22 | 1090 | uverbs_dev->num_comp_vectors = device->num_comp_vectors; |
bc38a6ab | 1091 | |
055422dd AC |
1092 | cdev_init(&uverbs_dev->cdev, NULL); |
1093 | uverbs_dev->cdev.owner = THIS_MODULE; | |
1094 | uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops; | |
98508715 | 1095 | cdev_set_parent(&uverbs_dev->cdev, &uverbs_dev->kobj); |
055422dd | 1096 | kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum); |
ddbd6883 | 1097 | if (cdev_add(&uverbs_dev->cdev, base, 1)) |
70a30e16 | 1098 | goto err_cdev; |
bc38a6ab | 1099 | |
1e35a088 | 1100 | uverbs_dev->dev = device_create(uverbs_class, device->dev.parent, |
055422dd | 1101 | uverbs_dev->cdev.dev, uverbs_dev, |
91bd418f | 1102 | "uverbs%d", uverbs_dev->devnum); |
f4e91eb4 | 1103 | if (IS_ERR(uverbs_dev->dev)) |
bc38a6ab RD |
1104 | goto err_cdev; |
1105 | ||
f4e91eb4 | 1106 | if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev)) |
bc38a6ab | 1107 | goto err_class; |
f4e91eb4 | 1108 | if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version)) |
274c0891 | 1109 | goto err_class; |
bc38a6ab RD |
1110 | |
1111 | ib_set_client_data(device, &uverbs_client, uverbs_dev); | |
1112 | ||
1113 | return; | |
1114 | ||
1115 | err_class: | |
055422dd | 1116 | device_destroy(uverbs_class, uverbs_dev->cdev.dev); |
bc38a6ab RD |
1117 | |
1118 | err_cdev: | |
055422dd | 1119 | cdev_del(&uverbs_dev->cdev); |
6d6a0e71 AC |
1120 | if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES) |
1121 | clear_bit(devnum, dev_map); | |
1122 | else | |
1123 | clear_bit(devnum, overflow_map); | |
bc38a6ab RD |
1124 | |
1125 | err: | |
35d4a0b6 YH |
1126 | if (atomic_dec_and_test(&uverbs_dev->refcount)) |
1127 | ib_uverbs_comp_dev(uverbs_dev); | |
fd60ae40 | 1128 | wait_for_completion(&uverbs_dev->comp); |
35d4a0b6 | 1129 | kobject_put(&uverbs_dev->kobj); |
bc38a6ab RD |
1130 | return; |
1131 | } | |
1132 | ||
036b1063 YH |
1133 | static void ib_uverbs_free_hw_resources(struct ib_uverbs_device *uverbs_dev, |
1134 | struct ib_device *ib_dev) | |
1135 | { | |
1136 | struct ib_uverbs_file *file; | |
1e7710f3 | 1137 | struct ib_uverbs_async_event_file *event_file; |
036b1063 YH |
1138 | struct ib_event event; |
1139 | ||
1140 | /* Pending running commands to terminate */ | |
1141 | synchronize_srcu(&uverbs_dev->disassociate_srcu); | |
1142 | event.event = IB_EVENT_DEVICE_FATAL; | |
1143 | event.element.port_num = 0; | |
1144 | event.device = ib_dev; | |
1145 | ||
1146 | mutex_lock(&uverbs_dev->lists_mutex); | |
1147 | while (!list_empty(&uverbs_dev->uverbs_file_list)) { | |
1148 | struct ib_ucontext *ucontext; | |
036b1063 YH |
1149 | file = list_first_entry(&uverbs_dev->uverbs_file_list, |
1150 | struct ib_uverbs_file, list); | |
1151 | file->is_closed = 1; | |
036b1063 | 1152 | list_del(&file->list); |
036b1063 YH |
1153 | kref_get(&file->ref); |
1154 | mutex_unlock(&uverbs_dev->lists_mutex); | |
d1e09f30 | 1155 | |
d1e09f30 JG |
1156 | |
1157 | mutex_lock(&file->cleanup_mutex); | |
1158 | ucontext = file->ucontext; | |
1159 | file->ucontext = NULL; | |
1160 | mutex_unlock(&file->cleanup_mutex); | |
1161 | ||
1162 | /* At this point ib_uverbs_close cannot be running | |
1163 | * ib_uverbs_cleanup_ucontext | |
1164 | */ | |
036b1063 | 1165 | if (ucontext) { |
d1e09f30 JG |
1166 | /* We must release the mutex before going ahead and |
1167 | * calling disassociate_ucontext. disassociate_ucontext | |
1168 | * might end up indirectly calling uverbs_close, | |
1169 | * for example due to freeing the resources | |
1170 | * (e.g mmput). | |
1171 | */ | |
870201f9 | 1172 | ib_uverbs_event_handler(&file->event_handler, &event); |
036b1063 | 1173 | ib_dev->disassociate_ucontext(ucontext); |
cf8966b3 | 1174 | mutex_lock(&file->cleanup_mutex); |
fd3c7904 | 1175 | ib_uverbs_cleanup_ucontext(file, ucontext, true); |
cf8966b3 | 1176 | mutex_unlock(&file->cleanup_mutex); |
036b1063 YH |
1177 | } |
1178 | ||
1179 | mutex_lock(&uverbs_dev->lists_mutex); | |
1180 | kref_put(&file->ref, ib_uverbs_release_file); | |
1181 | } | |
1182 | ||
1183 | while (!list_empty(&uverbs_dev->uverbs_events_file_list)) { | |
1184 | event_file = list_first_entry(&uverbs_dev-> | |
1185 | uverbs_events_file_list, | |
1e7710f3 | 1186 | struct ib_uverbs_async_event_file, |
036b1063 | 1187 | list); |
db1b5ddd MB |
1188 | spin_lock_irq(&event_file->ev_queue.lock); |
1189 | event_file->ev_queue.is_closed = 1; | |
1190 | spin_unlock_irq(&event_file->ev_queue.lock); | |
036b1063 YH |
1191 | |
1192 | list_del(&event_file->list); | |
1e7710f3 MB |
1193 | ib_unregister_event_handler( |
1194 | &event_file->uverbs_file->event_handler); | |
1195 | event_file->uverbs_file->event_handler.device = | |
1196 | NULL; | |
036b1063 | 1197 | |
db1b5ddd MB |
1198 | wake_up_interruptible(&event_file->ev_queue.poll_wait); |
1199 | kill_fasync(&event_file->ev_queue.async_queue, SIGIO, POLL_IN); | |
036b1063 YH |
1200 | } |
1201 | mutex_unlock(&uverbs_dev->lists_mutex); | |
1202 | } | |
1203 | ||
7c1eb45a | 1204 | static void ib_uverbs_remove_one(struct ib_device *device, void *client_data) |
bc38a6ab | 1205 | { |
7c1eb45a | 1206 | struct ib_uverbs_device *uverbs_dev = client_data; |
036b1063 | 1207 | int wait_clients = 1; |
bc38a6ab RD |
1208 | |
1209 | if (!uverbs_dev) | |
1210 | return; | |
1211 | ||
f4e91eb4 | 1212 | dev_set_drvdata(uverbs_dev->dev, NULL); |
055422dd AC |
1213 | device_destroy(uverbs_class, uverbs_dev->cdev.dev); |
1214 | cdev_del(&uverbs_dev->cdev); | |
70a30e16 | 1215 | |
6d6a0e71 AC |
1216 | if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES) |
1217 | clear_bit(uverbs_dev->devnum, dev_map); | |
1218 | else | |
1219 | clear_bit(uverbs_dev->devnum - IB_UVERBS_MAX_DEVICES, overflow_map); | |
fd60ae40 | 1220 | |
036b1063 YH |
1221 | if (device->disassociate_ucontext) { |
1222 | /* We disassociate HW resources and immediately return. | |
1223 | * Userspace will see a EIO errno for all future access. | |
1224 | * Upon returning, ib_device may be freed internally and is not | |
1225 | * valid any more. | |
1226 | * uverbs_device is still available until all clients close | |
1227 | * their files, then the uverbs device ref count will be zero | |
1228 | * and its resources will be freed. | |
1229 | * Note: At this point no more files can be opened since the | |
1230 | * cdev was deleted, however active clients can still issue | |
1231 | * commands and close their open files. | |
1232 | */ | |
1233 | rcu_assign_pointer(uverbs_dev->ib_dev, NULL); | |
1234 | ib_uverbs_free_hw_resources(uverbs_dev, device); | |
1235 | wait_clients = 0; | |
1236 | } | |
1237 | ||
35d4a0b6 YH |
1238 | if (atomic_dec_and_test(&uverbs_dev->refcount)) |
1239 | ib_uverbs_comp_dev(uverbs_dev); | |
036b1063 YH |
1240 | if (wait_clients) |
1241 | wait_for_completion(&uverbs_dev->comp); | |
35d4a0b6 | 1242 | kobject_put(&uverbs_dev->kobj); |
bc38a6ab RD |
1243 | } |
1244 | ||
2c9ede55 | 1245 | static char *uverbs_devnode(struct device *dev, umode_t *mode) |
71c29bd5 | 1246 | { |
b2bc4782 GR |
1247 | if (mode) |
1248 | *mode = 0666; | |
71c29bd5 RD |
1249 | return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev)); |
1250 | } | |
1251 | ||
bc38a6ab RD |
1252 | static int __init ib_uverbs_init(void) |
1253 | { | |
1254 | int ret; | |
1255 | ||
bc38a6ab RD |
1256 | ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES, |
1257 | "infiniband_verbs"); | |
1258 | if (ret) { | |
aba25a3e | 1259 | pr_err("user_verbs: couldn't register device number\n"); |
bc38a6ab RD |
1260 | goto out; |
1261 | } | |
1262 | ||
70a30e16 RD |
1263 | uverbs_class = class_create(THIS_MODULE, "infiniband_verbs"); |
1264 | if (IS_ERR(uverbs_class)) { | |
1265 | ret = PTR_ERR(uverbs_class); | |
aba25a3e | 1266 | pr_err("user_verbs: couldn't create class infiniband_verbs\n"); |
bc38a6ab RD |
1267 | goto out_chrdev; |
1268 | } | |
1269 | ||
71c29bd5 RD |
1270 | uverbs_class->devnode = uverbs_devnode; |
1271 | ||
0933e2d9 | 1272 | ret = class_create_file(uverbs_class, &class_attr_abi_version.attr); |
bc38a6ab | 1273 | if (ret) { |
aba25a3e | 1274 | pr_err("user_verbs: couldn't create abi_version attribute\n"); |
bc38a6ab RD |
1275 | goto out_class; |
1276 | } | |
1277 | ||
bc38a6ab RD |
1278 | ret = ib_register_client(&uverbs_client); |
1279 | if (ret) { | |
aba25a3e | 1280 | pr_err("user_verbs: couldn't register client\n"); |
a265e558 | 1281 | goto out_class; |
bc38a6ab RD |
1282 | } |
1283 | ||
1284 | return 0; | |
1285 | ||
bc38a6ab | 1286 | out_class: |
70a30e16 | 1287 | class_destroy(uverbs_class); |
bc38a6ab RD |
1288 | |
1289 | out_chrdev: | |
1290 | unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES); | |
1291 | ||
1292 | out: | |
1293 | return ret; | |
1294 | } | |
1295 | ||
1296 | static void __exit ib_uverbs_cleanup(void) | |
1297 | { | |
1298 | ib_unregister_client(&uverbs_client); | |
70a30e16 | 1299 | class_destroy(uverbs_class); |
bc38a6ab | 1300 | unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES); |
6d6a0e71 AC |
1301 | if (overflow_maj) |
1302 | unregister_chrdev_region(overflow_maj, IB_UVERBS_MAX_DEVICES); | |
bc38a6ab RD |
1303 | } |
1304 | ||
1305 | module_init(ib_uverbs_init); | |
1306 | module_exit(ib_uverbs_cleanup); |