1/*-
2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3 *
4 * Copyright (c) 2005 Topspin Communications.  All rights reserved.
5 * Copyright (c) 2005 Cisco Systems.  All rights reserved.
6 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
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.
35 */
36
37#include <sys/cdefs.h>
38__FBSDID("$FreeBSD$");
39
40#define	LINUXKPI_PARAM_PREFIX ibcore_
41
42#include <linux/mm.h>
43#include <linux/dma-mapping.h>
44#include <linux/sched.h>
45#include <linux/slab.h>
46#include <linux/wait.h>
47#include <rdma/ib_umem_odp.h>
48
49#include "uverbs.h"
50
51#include <sys/priv.h>
52
53static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty)
54{
55	struct scatterlist *sg;
56	struct page *page;
57	int i;
58
59	if (umem->nmap > 0)
60		ib_dma_unmap_sg(dev, umem->sg_head.sgl,
61				umem->nmap,
62				DMA_BIDIRECTIONAL);
63
64	for_each_sg(umem->sg_head.sgl, sg, umem->npages, i) {
65
66		page = sg_page(sg);
67		put_page(page);
68	}
69
70	sg_free_table(&umem->sg_head);
71	return;
72
73}
74
75/**
76 * ib_umem_get - Pin and DMA map userspace memory.
77 *
78 * If access flags indicate ODP memory, avoid pinning. Instead, stores
79 * the mm for future page fault handling in conjunction with MMU notifiers.
80 *
81 * @context: userspace context to pin memory for
82 * @addr: userspace virtual address to start at
83 * @size: length of region to pin
84 * @access: IB_ACCESS_xxx flags for memory being pinned
85 * @dmasync: flush in-flight DMA when the memory region is written
86 */
87struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
88			    size_t size, int access, int dmasync)
89{
90	struct ib_umem *umem;
91	struct page **page_list;
92	struct vm_area_struct **vma_list;
93	unsigned long locked;
94	unsigned long cur_base;
95	unsigned long npages;
96	int ret;
97	int i;
98	struct dma_attrs dma_attrs = { 0 };
99	struct scatterlist *sg, *sg_list_start;
100	int need_release = 0;
101	unsigned int gup_flags = FOLL_WRITE;
102
103	if (dmasync)
104		dma_attrs.flags |= DMA_ATTR_WRITE_BARRIER;
105
106	if (!size)
107		return ERR_PTR(-EINVAL);
108
109	/*
110	 * If the combination of the addr and size requested for this memory
111	 * region causes an integer overflow, return error.
112	 */
113	if (((addr + size) < addr) ||
114	    PAGE_ALIGN(addr + size) < (addr + size))
115		return ERR_PTR(-EINVAL);
116
117	if (priv_check(curthread, PRIV_VM_MLOCK) != 0)
118		return ERR_PTR(-EPERM);
119
120	umem = kzalloc(sizeof *umem, GFP_KERNEL);
121	if (!umem)
122		return ERR_PTR(-ENOMEM);
123
124	umem->context   = context;
125	umem->length    = size;
126	umem->address   = addr;
127	umem->page_size = PAGE_SIZE;
128	umem->pid       = get_pid(task_pid(current));
129	/*
130	 * We ask for writable memory if any of the following
131	 * access flags are set.  "Local write" and "remote write"
132	 * obviously require write access.  "Remote atomic" can do
133	 * things like fetch and add, which will modify memory, and
134	 * "MW bind" can change permissions by binding a window.
135	 */
136	umem->writable  = !!(access &
137		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
138		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND));
139
140	if (access & IB_ACCESS_ON_DEMAND) {
141		ret = ib_umem_odp_get(context, umem);
142		if (ret) {
143			kfree(umem);
144			return ERR_PTR(ret);
145		}
146		return umem;
147	}
148
149	umem->odp_data = NULL;
150
151	page_list = (struct page **) __get_free_page(GFP_KERNEL);
152	if (!page_list) {
153		kfree(umem);
154		return ERR_PTR(-ENOMEM);
155	}
156
157	vma_list = (struct vm_area_struct **) __get_free_page(GFP_KERNEL);
158
159	npages = ib_umem_num_pages(umem);
160
161	down_write(&current->mm->mmap_sem);
162
163	locked     = npages + current->mm->pinned_vm;
164
165	cur_base = addr & PAGE_MASK;
166
167	if (npages == 0 || npages > UINT_MAX) {
168		ret = -EINVAL;
169		goto out;
170	}
171
172	ret = sg_alloc_table(&umem->sg_head, npages, GFP_KERNEL);
173	if (ret)
174		goto out;
175
176	if (!umem->writable)
177		gup_flags |= FOLL_FORCE;
178
179	need_release = 1;
180	sg_list_start = umem->sg_head.sgl;
181
182	while (npages) {
183		ret = get_user_pages(cur_base,
184				     min_t(unsigned long, npages,
185					   PAGE_SIZE / sizeof (struct page *)),
186				     gup_flags, page_list, vma_list);
187
188		if (ret < 0)
189			goto out;
190
191		umem->npages += ret;
192		cur_base += ret * PAGE_SIZE;
193		npages   -= ret;
194
195		for_each_sg(sg_list_start, sg, ret, i) {
196			sg_set_page(sg, page_list[i], PAGE_SIZE, 0);
197		}
198
199		/* preparing for next loop */
200		sg_list_start = sg;
201	}
202
203	umem->nmap = ib_dma_map_sg_attrs(context->device,
204				  umem->sg_head.sgl,
205				  umem->npages,
206				  DMA_BIDIRECTIONAL,
207				  &dma_attrs);
208
209	if (umem->nmap <= 0) {
210		ret = -ENOMEM;
211		goto out;
212	}
213
214	ret = 0;
215
216out:
217	if (ret < 0) {
218		if (need_release)
219			__ib_umem_release(context->device, umem, 0);
220		put_pid(umem->pid);
221		kfree(umem);
222	} else
223		current->mm->pinned_vm = locked;
224
225	up_write(&current->mm->mmap_sem);
226	if (vma_list)
227		free_page((unsigned long) vma_list);
228	free_page((unsigned long) page_list);
229
230	return ret < 0 ? ERR_PTR(ret) : umem;
231}
232EXPORT_SYMBOL(ib_umem_get);
233
234static void ib_umem_account(struct work_struct *work)
235{
236	struct ib_umem *umem = container_of(work, struct ib_umem, work);
237
238	down_write(&umem->mm->mmap_sem);
239	umem->mm->pinned_vm -= umem->diff;
240	up_write(&umem->mm->mmap_sem);
241	mmput(umem->mm);
242	kfree(umem);
243}
244
245/**
246 * ib_umem_release - release memory pinned with ib_umem_get
247 * @umem: umem struct to release
248 */
249void ib_umem_release(struct ib_umem *umem)
250{
251	struct ib_ucontext *context = umem->context;
252	struct mm_struct *mm;
253	struct task_struct *task;
254	unsigned long diff;
255
256	if (umem->odp_data) {
257		ib_umem_odp_release(umem);
258		return;
259	}
260
261	__ib_umem_release(umem->context->device, umem, 1);
262
263	task = get_pid_task(umem->pid, PIDTYPE_PID);
264	put_pid(umem->pid);
265	if (!task)
266		goto out;
267	mm = get_task_mm(task);
268	put_task_struct(task);
269	if (!mm)
270		goto out;
271
272	diff = ib_umem_num_pages(umem);
273
274	/*
275	 * We may be called with the mm's mmap_sem already held.  This
276	 * can happen when a userspace munmap() is the call that drops
277	 * the last reference to our file and calls our release
278	 * method.  If there are memory regions to destroy, we'll end
279	 * up here and not be able to take the mmap_sem.  In that case
280	 * we defer the vm_locked accounting to the system workqueue.
281	 */
282	if (context->closing) {
283		if (!down_write_trylock(&mm->mmap_sem)) {
284			INIT_WORK(&umem->work, ib_umem_account);
285			umem->mm   = mm;
286			umem->diff = diff;
287
288			queue_work(ib_wq, &umem->work);
289			return;
290		}
291	} else
292		down_write(&mm->mmap_sem);
293
294	mm->pinned_vm -= diff;
295	up_write(&mm->mmap_sem);
296	mmput(mm);
297out:
298	kfree(umem);
299}
300EXPORT_SYMBOL(ib_umem_release);
301
302int ib_umem_page_count(struct ib_umem *umem)
303{
304	int shift;
305	int i;
306	int n;
307	struct scatterlist *sg;
308
309	if (umem->odp_data)
310		return ib_umem_num_pages(umem);
311
312	shift = ilog2(umem->page_size);
313
314	n = 0;
315	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, i)
316		n += sg_dma_len(sg) >> shift;
317
318	return n;
319}
320EXPORT_SYMBOL(ib_umem_page_count);
321
322/*
323 * Copy from the given ib_umem's pages to the given buffer.
324 *
325 * umem - the umem to copy from
326 * offset - offset to start copying from
327 * dst - destination buffer
328 * length - buffer length
329 *
330 * Returns 0 on success, or an error code.
331 */
332int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset,
333		      size_t length)
334{
335	size_t end = offset + length;
336	int ret;
337
338	if (offset > umem->length || length > umem->length - offset) {
339		pr_err("ib_umem_copy_from not in range. offset: %zd umem length: %zd end: %zd\n",
340		       offset, umem->length, end);
341		return -EINVAL;
342	}
343
344#ifdef __linux__
345	ret = sg_pcopy_to_buffer(umem->sg_head.sgl, umem->nmap, dst, length,
346				 offset + ib_umem_offset(umem));
347#else
348	ret = 0;
349#endif
350	if (ret < 0)
351		return ret;
352	else if (ret != length)
353		return -EINVAL;
354	else
355		return 0;
356}
357EXPORT_SYMBOL(ib_umem_copy_from);
358